A Longitudinal Approach To Measuring Income Mobility Among 4shnv3v8xp

Open access Dissertation DOI:10.14264/UQL.2015.
665
A longitudinal approach to measuring income mobility among Filipino households

— Source link
Arturo Martinez
Published on: 28 May 2015
Topics: Income distribution, Household income, Total personal income, Income inequality metrics and
Distribution (economics)
Related papers:
Rising Income Inequality in China : A Race to the Top
Global income inequality by the numbers : in history and now --an overview--
Income Mobility and Income Inequality in Rural China
Income Inequality, Mobilityy and Economic Insecurity in Australia
A New Data Set Measuring Income Inequality
Share this paper:

View more about this paper here: https://typeset.io/papers/a-longitudinal-approach-to-measuring-income-mobility-among-
4shnv3v8xp
A Longitudinal Approach to Measuring
Income Mobility among Filipino Households
Arturo M. Martinez Jr.
Master of Science in Statistics
Bachelor of Science in Statistics
A thesis submitted for the degree of Doctor of Philosophy at

The University of Queensland in 2015
Institute for Social Science Research
1
Abstract
This study takes a longitudinal approach to analysing household income distribution in the
Philippines over the past decade using data from the redesigned Philippine Family Income and
Expenditure Survey and Labour Force Survey. Using existing and newly-developed analytical
tools, the study examined the income trajectories or income mobility of a sample of households
tracked in 2003, 2006, and 2009 and found income mobility within this time period to be
significant. However, income poverty rates and inequality levels over the past decade remained
stagnant because significant positive and negative mobility existed across space and over time,
thereby suggesting that the country’s economic growth has created both winners and losers. In
addition, positive changes in socio-economic capital were offset by reductions in economic
returns to capital, highlighting that investments in socio-economic capital development should
be complemented with effective management of economic returns to capital. Furthermore,
there are some indications that households with lower income experienced slightly faster
income growth. However, transitory income fluctuations contribute significantly suggesting
that convergence of income of the poor and non-poor may be in part due to random variations.
This result indicates that in order to achieve sustainable and inclusive growth, it is also
important to provide economic risk-management tools for those who periodically move into
and out of economic hardships.
2
3
Declaration by author
This thesis is composed of my original work, and contains no material previously published
or written by another person except where due reference has been made in the text. I have
clearly stated the contribution by others to jointly-authored works that I have included in my
thesis.
I have clearly stated the contribution of others to my thesis as a whole, including statistical
assistance, survey design, data analysis, significant technical procedures, professional
editorial advice, and any other original research work used or reported in my thesis. The
content of my thesis is the result of work I have carried out since the commencement of my
research higher degree candidature and does not include a substantial part of work that has
been submitted to qualify for the award of any other degree or diploma in any university or
other tertiary institution. I have clearly stated which parts of my thesis, if any, have been
submitted to qualify for another award.
I acknowledge that an electronic copy of my thesis must be lodged with the University
Library and, subject to the General Award Rules of The University of Queensland,
immediately made available for research and study in accordance with the Copyright Act
1968.
I acknowledge that copyright of all material contained in my thesis resides with the copyright
holder(s) of that material. Where appropriate I have obtained copyright permission from the
copyright holder to reproduce material in this thesis.
4
Publications during candidature
Journal articles
Martinez, A., Western, M., Haynes, M., Tomaszewski, W. & Macarayan, E. (2014). Multiple
Job Holding and Income Mobility in Indonesia. Research in Social Stratification and Mobility,
37, pp. 91-104. http://dx.doi.org/10.1016/j.rssm.2013.09.008
Martinez, A., Lucio, E. & Villaruel, M. (2014). Evaluating Weight-Reallocated Estimators for
Small Area Estimation of Poverty. Electronic Journal of Applied Statistical Analysis, pp. 417-
431. 10.1285/i20705948v7n2p417
Martinez, A., Western, M., Haynes, M. & Tomaszewski, W. (2014). Is there income mobility
in the Philippines? Asian-Pacific Economic Literature, 28(1), pp.96-115.
http://dx.doi.org/10.1111/apel.12047
Martinez, A., Western, M., Haynes, M., & Tomaszewski, W. (2013). Measuring Income
Mobility using Pseudo-Panel Data. The Philippine Statistician, 62(2), pp. 71-99.
Maligalig, M. & Martinez, A. (2013). Developing a Master Sample Design for Household
Surveys in Developing Countries: A Case Study in Bangladesh. Survey Methods: Insights from
the Field. http://dx.doi.org/10.13094/SMIF-2013-00009
Martinez, A. (2012). Small Area Estimation with a Multivariate Spatial Temporal Model. The
Philippine Statistician, 61(2), pp. 1-17.
Conference Papers
Haynes, M. & Martinez, A. (2014). Modelling the relationships between household residential
mobility and childbearing over the life course in Australia. Paper presented at the International
Conference of the Society for Longitudinal and Life Course Studies, Amsterdam, Netherlands.
Later published as Life Course Centre Working Paper No. 2015-01.
Martinez, A., Perales, F., Western, M., Haynes, M., & Tomaszewski, W. (2014). A
Comparative Study of the Determinants of Multidimensional Poverty Dynamics in Australia
and United Kingdom. Paper presented at the Summer School on Advanced Poverty Research,
University of Bremen, Germany.
Mobility using Pseudo-Panel Data. Paper presented at the Australian Statistics Conference,
Sydney, Australia.
Martinez, A., Western, M., Haynes, M., & Tomaszewski, W. (2013). Methodological
Developments in Measuring Intra-generational Income Mobility using Repeated Cross-
Sectional Data. Paper presented at the International Conference of Research Committee (RC)
28: New Horizons in Research on Stratification, Mobility and Inequality, Brisbane, Australia.
Martinez, A., Western, M., Haynes, M., & Tomaszewski, W. (2012). Moving to Higher Quality
Employment: An Exploratory Study of Multiple Job Holders in Indonesia. Paper presented at
the Symposium on Asian Perspective on Social Stratification and Inequality, Tohoku
University, Sendai, Japan.
5
Working Papers / Reports
Albert, J., Dumagan, J., & Martinez, A. (2015). Inequalities in Income, Labour, and Education:
The Challenge of Inclusive Growth. Discussion Paper No. 2015-01. Manila: Philippine
Institute for Development Studies.
Albert, J., & Martinez, A. (2015). Is Poverty Really Decreasing, And If Not, Why Not? To be
published as a Policy Note. Manila: Philippine Institute for Development Studies.
Albert, J. & Martinez, A. (2015). Are poverty and inequality changing? Rappler. Available at
http://www.rappler.com/thought-leaders/84833-poverty-inequality-data
Baffour, B., Haynes, M., Western, M., Pennay, D., Misson, S. & Martinez, A. (2014).
Weighting strategies for combining data from dual-frame telephone surveys: emerging
evidence from Australia. Under review in Journal of Statistics and Survey Methodology.
Martinez, A. (2015). Analytical Consideration when Measuring Income Mobility. Life Course
Centre Working Paper No. 2014-4.
Martinez, A. & Perales, F. (2014). The Dynamics of Multidimensional Poverty in Australia.

Life Course Centre Working Paper No. 2014-8. Under review in Research in Social
Stratification and Mobility.
Martinez, A., Lucio, E., Western, M., Tomaszewski, W. & Haynes, M. (2014). Measuring pro-
poor growth in the Philippines using Pseudo-Panel Data. Under review in Human Welfare.
Martinez, A., Haynes, M., Western, M. & Tomaszewski, W. (2014). How long do poor
Filipinos stay in poverty? Under review in Journal of Asia Pacific Economy.
Martinez, A., Western, M., Haynes, M. & Tomaszewski, W. (2014). How Income
Segmentation Affects Income Mobility: Evidence from Panel Data in the Philippines. Under
review in Asia & the Pacific Policy Studies.
Martinez, A., Western, M., Tomaszewski, W. & Haynes, M. (2014). What drives income
distribution dynamics in the Philippines? Under review in Social Science Research.
Povey, J., Martinez, A. & Baffour, B. (2014). Cambodia household food safety survey, Final
report. Phase II, sanitary and phytosanitary standards management systems project, Cambodia.
Grant 0224-CAM. Prepared for the Asian Development Bank. University of Queensland,
Institute for Social Science Research and UNIQUEST.
Povey, J., Kapelle, N. & Martinez, A. (2014). Queensland school business managers’ wellbeing
survey. Final report. Prepared for School Business Managers’ Association
Queensland. University of Queensland, Institute for Social Science Research.
Publications included in this thesis
Martinez, A., Lucio, E., Western, M., Tomaszewski, W. & Haynes, M. (2014). Measuring pro-
poor growth in the Philippines using Pseudo-Panel Data. Under review in Human Welfare.
– substantial discussion is included as part of Chapter 2.
6
Contributor Statement of contribution
Arturo Martinez Jr. Statistical analysis of data
Wrote the paper
Eduardo Lucio Edited paper
Mark Western Edited paper
Michele Haynes Edited paper
Wojtek Tomaszewski Edited paper
Martinez, A., Western, M., Haynes, M. & Tomaszewski, W. (2014). Is there income mobility
in the Philippines? Asian-Pacific Economic Literature, 28(1), pp.96-115.
http://dx.doi.org/10.1111/apel.12047 – substantial discussion is included as part of Chapter 4.
Wrote the paper
Martinez, A., Haynes, M., Western, M., & Tomaszewski, W. (2014). How long do poor
Filipinos stay in poverty? Under review in Journal of Asia Pacific Economy – substantial
discussion is included as part of Chapter 5.
Wrote the paper
Martinez, A., Western, M., Tomaszewski, W. & Haynes, M. (2014). How Income
Segmentation Affects Income Mobility: Evidence from Panel Data in the Philippines. Revised
version has been resubmitted for review in Asia & the Pacific Policy Studies – substantial
Wrote the paper
7
Martinez, A., Western, M., Tomaszewski, W. & Haynes, M. (2014). What drives income
distribution dynamics in the Philippines? Under review in Social Science Research –
substantial discussion is included as part of Chapter 7.
Wrote the paper
Martinez, A., Western, M., Haynes, M., Tomaszewski, W. & Macarayan, E. (2014). Multiple
Job Holding and Income Mobility in Indonesia. Research in Social Stratification and Mobility,
37, pp. 91-104. http://dx.doi.org/10.1016/j.rssm.2013.09.008 – discussion of Theoretical
Framework is included as part of Chapter 8.
Wrote the paper
Erlyn Macarayan Edited paper
Mobility using Pseudo-Panel Data. The Philippine Statistician, 62(2), pp. 71-99. – substantial
Wrote the paper
8
Contributions by others to the thesis
This thesis has benefitted from substantial inputs from Professor Mark Western, Professor
Michele Haynes, Dr. Wojtek Tomaszewski who helped me in the conception and design of
the project and interpretation of research data and the editorial work of Dr. Francisco Perales,
Mr. Eduardo Lucio, Ms. Erlyn Macarayan, Ms. Louise Marquart and Ms. Donna Lampa on
selected chapters.
Statement of parts of the thesis submitted to qualify for the award of another degree
None.
9
Acknowledgements
I have written this thesis while a full-time student at the Institute for Social Science
Research, from 2012 to 2015, with financial support from The University of Queensland
through the International Postgraduate Research Scholarship and Centennial Awards.
I am infinitely grateful to my supervisors, Professor Mark Western, Professor Michele
Haynes and Dr. Wojtek Tomaszewski who shared their expertise on the subject matter and
dedicated their valuable time in providing thorough comments on how to improve the thesis. I
have also benefitted from discussions from colleagues within ISSR and other experts from
various international development agencies. In particular, the study also benefitted from fruitful
discussion with Dr. Francisco Perales, Dr. Jose Ramon Albert, Dr. Hai-Anh Dang, Dr. Dalisay
Maligalig and Professor Gary Fields. I am also thankful for the very insightful comments
provided by the external referees who reviewed this study. Nevertheless, all errors in this study
are my responsibility.
I would also like to acknowledge my family and friends who have been my main source
of inspiration. Finally, I thank God for all the wonderful blessings over the years.
10
Keywords
income mobility, poverty, income inequality, pseudo-panel, economic growth, social
statistics, developing countries
Australian and New Zealand Standard Research Classifications (ANZSRC)

ANZSRC code: 140202, Economic Development and Growth, 40%
ANZSRC code: 160807, Sociological Methodology and Research Methods, 30%
ANZSRC code: 010401, Applied Statistics, 30%
Fields of Research (FoR) Classification

FoR code: 1402, Applied Economics, 40%
FoR code: 1608, Sociology, 30%
FoR code: 0104, Statistics, 30%
11
Contents
Chapter 1 Analytical Tools for the Analysis of Income Mobility ...................................... 25

1.1 Introduction................................................................................................................ 25
1.2 What is Income Mobility and Why it is Important? .................................................. 28
1.3 How Do We Measure Income Mobility?................................................................... 30
1.3.1 What Characterizes Mobility of Incomes? .......................................................... 32
1.3.2 Analytical Tools for Measuring Income Mobility .............................................. 34
1.4 Incorporating Income Mobility in the Analysis of Poverty, Inequality and Pro-Poor
Growth ....................................................................................................................... 36
1.4.1 Measuring Intertemporal Poverty ....................................................................... 37
1.4.2 Income Mobility-Adjusted Pro-Poor Growth Assessment ................................. 44
1.4.3 Measuring Dynamics of Income Inequality........................................................ 47
1.4.4 Identifying Correlates of Income Mobility ......................................................... 48
1.5 Analytical Considerations When Examining Income Mobility................................. 50
1.5.1 Correcting for Measurement Error and Data Contamination ............................ 50
1.5.2 Decomposing Income into Permanent and Transitory Components ................. 52
1.6 Summary .................................................................................................................. 54
Chapter 2 Has Economic Growth been Pro-Poor in the Philippines? .............................. 56
2.1 Introduction ............................................................................................................... 56
2.2 History of Economic Growth in the Philippines Over the Past Three Decades ........ 58
2.3 Poverty, Inequality and Pro-poor Growth Patterns in the Philippines ...................... 60
2.3.1 Cross-Sectional Perspective ............................................................................... 60
2.3.2 Longitudinal Perspective .................................................................................... 64
2.4 Summary .................................................................................................................... 69
Chapter 3 Family Income and Expenditure Survey and Labour Force Survey .............. 75
3.1 Introduction ................................................................................................................ 75
3.2 Survey Content and Administration ........................................................................... 75
3.3 Observation Period ..................................................................................................... 76
3.4 Income Measure ......................................................................................................... 77
3.5 Sampling Design and Survey Weight Adjustments for Non-Coverage Bias ............ 79
3.6 Poverty Lines ............................................................................................................ 84
3.7 Summary ................................................................................................................... 84
Chapter 4 Is there Income Mobility in the Philippines? .................................................... 87
4.1 Introduction ............................................................................................................... 87
4.2 Is there income mobility in the Philippines? ............................................................. 89
4.2.1 Is there relative income mobility? ..................................................................... 89
4.2.2 Is there absolute income mobility? .................................................................... 89
4.2.3 Is there equalizing mobility?.............................................................................. 92
4.3 Discussion .................................................................................................................. 99
4.4 Summary .................................................................................................................. 101
Chapter 5 How Long Do the Poor Stay in Poverty? ......................................................... 107
5.1 Introduction.............................................................................................................. 107
5.2 Intertemporal Poverty in the Philippines ................................................................. 108
5.3 Methodology ............................................................................................................ 112
5.4 Empirical Results ..................................................................................................... 113
5.4.1 Intertemporal Poverty in the Philippines ......................................................... 113
5.4.2 Where are the Persistently and Transiently Poor? ........................................... 121
5.4.3 Who are the Persistently and Transiently Poor? .............................................. 129
5.5. Summary ................................................................................................................ 137
12
Chapter 6 Who are Income Mobile? .................................................................................. 146
6.1 Introduction ............................................................................................................ 146
6.2 Different Patterns of Income Mobility ................................................................... 147
6.3. Methods ................................................................................................................... 149
6.3.1 Classifying Households According to Income Mobility Trajectories .............. 149
6.3.2 Measures of Socio-economic Advantage ......................................................... 151
6.3.3 Other Correlates of Income Mobility ............................................................... 153
6.3.4 Statistical Models of Income Mobility ............................................................. 153
6.4.1 Trends in Income Inequality and Polarization .................................................. 154
6.4.2 Income Mobility and Inequality ....................................................................... 156
6.4.3 Testing Convergence, Divergence and Symmetry of Income Mobility ........... 158
6.4.4 Estimated Statistical Models ............................................................................ 163
6.5 Summary and Discussion ........................................................................................ 165
Chapter 7 What Drives Income Distribution Dynamics in the Philippines? ................. 172
7.1 Introduction.............................................................................................................. 172
7.2 Concepts and Methods ............................................................................................. 174
7.2.1 Drivers of Income Distribution Dynamics........................................................ 174
7.2.2 Estimating the Contribution of SECs and SERs to the Evolution of the Income
Distribution ....................................................................................................... 175
7.2.3 Constructing Indices of Socio-Economic Capital ............................................ 179
7.3 Results...................................................................................................................... 180
7.3.1 Drivers of Household Income Distribution Dynamics in the Philippines ....... 180
7.3.2 Robustness Checks ........................................................................................... 185
7.3.3 Potential Limitations of the Accounting Exercise ............................................ 186
7.4 Summary and Discussion ........................................................................................ 187
Chapter 8 Multiple Jobholding and Socio-Economic Mobility in the Philippines ........ 196
8.1 Introduction.............................................................................................................. 196
8.2. Theoretical Model for Multiple Jobholding and Income Mobility .......................... 197
8.2.1 Determinants of Multiple Jobholding .............................................................. 197
8.2.2 Socio-Economic Mobility and Multiple Job Holding...................................... 201
8.3. Data and Implementation of Concepts ................................................................... 204
8.3.1 Merged FIES-LFS ............................................................................................ 204
8.3.2 Measuring Socio-Economic Mobility .............................................................. 205
8.3.3 Distinguishing between Constrained and Non-Constrained Pluriactivity ........ 206
8.4.1 Background on the Philippines’s Labour Market Over the Past Decade ......... 207
8.4.2 Discussion of Empirical Results ....................................................................... 211
8.5 Conclusion and Policy Implications ........................................................................ 216
Chapter 9 Evaluating the Feasibility of Using Pseudo-Panel Data to Measure Income
Mobility ................................................................................................................................. 219
9.1 Introduction.............................................................................................................. 219
9.2 Developments in Mobility Estimation using Repeated Cross-Sectional Data ........ 220
9.2.1 What is Pseudo-Panel Estimation? .................................................................... 220
9.2.2 Estimation of Income Mobility using Pseudo-Panel Data ................................ 224
9.2.3 Extending Pseudo-Panel Methods to Measure Broad Class of Income Mobility
Measures ........................................................................................................... 228
9.3 Data .......................................................................................................................... 230
9.4 Discussion of Empirical Results .............................................................................. 231
9.5 Summary and Future Directions .............................................................................. 235
13
Chapter 10 Summary and Conclusion .............................................................................. .243
10.1 Introduction………………………………………………………………………..243
10.2 Motivation and Research Goals...………………………………………………....243
10.3 Main Findings………………………...…………………………………………...244
10.4 Broad Policy Implications………………………………………………...……….245
10.5 Limitations and Future Directions………………………………………………….246
References…………………………………………………………………………….……250
14
List of Tables and Figures
Table 1.1 Differences Between Income Mobility Perspectives ............................................... 34

Table 1.2 Sample Income Transition Matrix ........................................................................... 35
Table 1.3 Formula of Different Income Mobility Indices ....................................................... 38
Table 2.1 Socio-Economic Indicators for Selected Southeast Asian Countries ...................... 57
Table 2.2 Distribution of Household Monthly Income Per Capita in the Philippines, 1985-
2009 ......................................................................................................................... 62
Table 3.1 Regional Price Differences ...................................................................................... 78
Table 3.2 Sample Size.............................................................................................................. 80
Table 3.3 Comparison of Full Cross-Sectional and Longitudinal Subsample ......................... 81
Table 3.4 Features of Longitudinal Subsample Using Attrition-Adjusted Weights ................ 81
Table 3.5 Average Household Income Per Capita by Region ................................................. 83
Table 4.1 Summary of Relative Income Mobility Measures ................................................... 89
Table 4.2 (Absolute) Income Transition Matrix, 2003-2009 ................................................... 90
Table 4.3 Summary of Income Mobility Measures ................................................................. 92
Table 4.4 Inequality-Reducing Effect of Income Mobility ..................................................... 93
Table 4.5 Comparison of Philippines’ Income Inequality ....................................................... 94
Table 4.6 Regression Estimates of the Relation between Changes ......................................... 96
Table 4.7 Decomposition of Change in Inequality into ........................................................... 97
Table 4.8 (Absolute) Income Transition Matrix, 2003-2006 ................................................. 100
Table 4.9 (Absolute) Income Transition Matrix, 2006-2009 ................................................. 100
Table 5.1 Cross-Sectional Measures of Poverty in the Philippines, 2003-2009.................... 115
Table 5.2 Poverty Transition Matrix, US$1.25/Day Poverty Line ........................................ 116
Table 5.3 Poverty Transition Matrix, US$2/Day Poverty Line ............................................. 116
Table 5.4 Poverty Transition Matrix, Half of Median Poverty Line ..................................... 116
Table 5.5 Poverty Transition Matrix, Government Poverty Line .......................................... 117
Table 5.6 Population Share by Intertemporal Poverty Status ................................................ 120
Table 5.7 Regression Coefficients of Multinomial Logistic Models for Intertemporal Poverty
in the Philippines ................................................................................................... 131
Table 6.1 Decomposition of Inequality by Income Clusters ................................................. 157
Table 6.2 Decomposition of Inequality by Income Source ................................................... 157
Table 6.3 Distribution of Income Growth Rates (%) ............................................................. 159
Table 6.4 Distribution of Income Trajectories (%) ................................................................ 159
Table 6.5 Distribution of Income Trajectories (%), by Segments of Initial Income ............. 159
Table 6.6 Distribution of Income Trajectories (%), by Segments of Permanent Income ...... 162
Table 6.7 Distribution of Income Mobility by Household Characteristics (%) ..................... 162
Table 6.8 Regression Coefficients of Multinomial Logistic Models ..................................... 166
Table 7.1 Trade-off between Socio-Economic Capital.......................................................... 184
Table 8.1 Definition of Formal and Informal Employment ................................................... 206
Table 8.2 Trends in Key Labour Market Indicators, 2003-2012 ........................................... 208
Table 8.3 Distribution of Workers (%), by Production Sector of Main Job .......................... 209
Table 8.4 Distribution of Workers (%), by Main Occupation ............................................... 210
Table 8.5 Distribution of Workers (%), by Status of Main Employment .............................. 210
Table 8.6 Average Daily Basic Pay of Wage and Salary Workers ........................................ 211
Table 8.7 Distribution of Employment Status (%), 2003-2009 ............................................. 211
Table 8.8 Distribution of Multiple Job Holders (%), by ........................................................ 213
Table 8.9 Regression Coefficients of Logistic and Multinomial Logistic Models on the
Propensity to Take Multiple Jobs .......................................................................... 215
Table 8.10 Regression Coefficients of Economic Mobility Models...................................... 217
15
Table 9.1 Estimates of Poverty Dynamics in the Philippines, 2003-2006.......................... 237
Table 9.4 Estimates of Income Mobility in the Philippines, 2003-2006 ............................ 238
Figure 1.1 Relationship Between Different Income Mobility Indices…...............................39

Figure 1.2 Sample Growth Incidence Curve………………………………………………..45
Figure 2.1 Comparison of Estimates of Growth between Survey and National Accounts .... 65
Figure 2.2 Growth Incidence Curves in the Philippines, 1985-2009 ..................................... 65
Figure 2.3 Income Re-ranking Scenarios Considered for Constructing Pseudo-Panel Data
Sets ........................................................................................................................ 67
Figure 2.4 Poverty and Inequality Dynamics in the Philippines, 1985-2009 ........................ 67
Figure 2.5 Individual Income Growth in the Philippines, 1985-2009 ................................... 68
Figure 3.1 Regional Poverty Map of the Philippines ............................................................. 85
Figure 4.1 Income Mobility Curve, 2003-2009 ..................................................................... 92
Figure 4.2 Change in the Logarithm of Income between 2003 and 2009,............................. 94
Figure 4.3 Range of Values of the Change in the Logarithm of Income between 2003 and
2009, by Percentile of Average Income ................................................................ 96
Figure 5.1 Headcount Poverty Curve, 2009......................................................................... 114
Figure 5.2 Intertemporal Poverty Estimates using the JR Approach ................................... 117
Figure 5.3 Intertemporal Poverty Estimates using the DAG Approach .............................. 118
Figure 5.4 Map of Persistent and Transient (Headcount) Poverty....................................... 124
Figure 5.5 Regional Intertemporal Poverty Estimates (JR Approach) ................................ 125
Figure 5.6 Regional Intertemporal Poverty Estimates (DAG Approach) ............................ 126
Figure 5.7 Regional Intertemporal Poverty Estimates (Spells Approach) ........................... 127
Figure 6.1 Joint Distribution of Income Trajectories, 2003-2006 and 2006-2009 .............. 150
Figure 6.2 Different Types of Income Trajectories, 2003-2009 .......................................... 151
Figure 6.3 Income Inequality in the Philippines, 2003-2009............................................... 156
Figure 6.4 Anonymous and Non-Anonymous Growth Incidence Curves, 2003-2009........ 158
Figure 7.1 Temporal Changes in the Levels of Socio-Economic Capital............................ 180
Figure 7.2 Temporal Changes in the Socio-Economic Returns to Capital .......................... 181
Figure 7.3 Estimates Contribution of Different Factors on Poverty and Inequality ............ 183
Figure 8.1 Constrained and Non-constrained Pluriactivity.................................................. 200
16
List of Appendices
A2.1 Constructing Synthetic Pseudo-Panel Data of Income ...................................... 71

Figure A2.1 Correlation of Income Rank using Actual Panel Data in the Philippines, 2006
2009……………….……………………………………………………..……74
Table A3.1 Descriptive Statistics for Household Expenditure Per Capita…………………86
A4.1 Mobility of Household Incomes ....................................................................... 104
Figure A4.1 Mobility Curve Using Household Income Data…………………………..….105
Table A4.1 (Absolute) Household Income Transition Matrix, 2003-2009………………..105
Table A4.2 Selected Indicators of Mobility, 2003-2009…………………………………..106
Table A5.1 Intertemporal Poverty Headcount Rate using the Components Approach, by
Household Characteristics………………………………………………...…..140
Table A5.2 Intertemporal Poverty Headcount Rate using the Spells Approach, by
Household Characteristics……………………………………...……………..141
Table A5.3 Intertemporal Poverty Headcount Rate using the JR Approach, by Region….142
Table A5.4 Standard Errors of Intertemporal Poverty Headcount Rate using the Components
Approach……………………………………………………………………...143
Table A5.5 Standard Errors of Intertemporal Poverty Headcount Rate using the Spells
Approach……………………………………………………………………...144
Table A5.6 Standard Errors of Intertemporal Poverty Headcount Rate using the JR
Approach, by Region…………………………………………………………145
Table A6.1 Regression Coefficients of Multinomial Logistic Models for Income Growth
Trajectories in the Philippines………………………………………………..169
Table A7.1 Descriptive Statistics of Variables Included in SEC Indices………………....191
Table A7.2 Estimated Contribution of Different Factors on Changes in Poverty and
Inequality in the Philippines………………………………………………….193
Table A7.3 Estimated Contribution of Different Factors on Changes in Poverty and
Inequality, by Region…………………………………………………………194
Table A9.1 Standard Error of Estimates of Poverty Dynamics, 2003-2006……………....240
Table A9.4 Standard Error of Estimates of Income Mobility, 2003-2006…………….......241
17
List of Abbreviations used in the thesis
ADB – Asian Development Bank

APIS – Annual Poverty Indicators Survey
ARMM – Autonomous Region of Muslim Mindanao
ASEAN – Association of Southeast Asian Nations
BLES – (The Philippine) Bureau of Labour and Employment Statistics
CALABARZON – Cavite, Laguna, Batangas, Rizal and Quezon
CAR – Cordillera Administrative Region
CCT – Conditional Cash Transfer
CEDAW – Committee on the Elimination of Discrimination Against Women
DASP – Distribution Analysis Software Package
FGT – Foster, Greer and Thorbecke
FIES –Family Income and Expenditures Survey
GDP – Gross Domestic Product
GIC – Growth Incidence Curve
GMM – Generalized Method of Moments
GRC – Gradin, del Rio and Canto
HDI – Human Development Index
ILO – International Labour Organization
JR – Jalan and Ravallion
KILM- Key Indicators of the Labour market
LFS – Labour Force Survey
MDG – Millennium Development Goals
MIMAROPA – Mindoro, Marinduque, Romblon, Palawan
NCR – National Capital Region
NSCB – (The Philippine) National Statistical Coordination Board
NSO – (The Philippine) National Statistics Office
OECD – Organisation for Economic Co-operation and Development
PAG-ASA – Philippine Atmospheric Geophysical and Astronomical Services Administration
UNDP – United Nations Development Programme
UNESCAP – United Nations Economic and Social Commission for Asia and the Pacific
UNICEF – United Nations International Children’s Emergency Fund
UNSD – United Nations Statistical Division
WB – (The) World Bank
WHO – World Health Organization
WDI – World Development Indicators
18
Introduction
In response to the challenge of ensuring that socio-economic opportunities are evenly
distributed, the United Nations (UN) along with its approximately two hundred member states
have committed to reducing various forms of socio-economic deprivation and exclusion,
particularly in developing countries. In this context, a list of targets between 1990 and 2015
were identified. The targets span multiple dimensions of socio-economic development such as
income, health, education, employment and gender equality. These constitute the Millennium
Development Goals (MDG).
Eradicating extreme hunger and poverty is the first of the eight MDGs that UN member
states have committed to pursue by 2015. Under this goal, there are three specific targets: to
halve the proportion of people living under US$1.25 a day, to achieve full and productive
employment and decent work for all- including women and young people, and to halve the
proportion of people who suffer from hunger (UN 2014). Although substantial gains have been
made towards this goal — for instance in 2010, the global poverty rate has already fallen to
less than half of the proportion of poor estimated in 1990 — much effort is still needed to
improve the lives of the 1.2 billion people who remain extremely poor. As 2015 marks the
deadline set for achieving the MDGs, it is important to evaluate where we stand and identify
what else can be done for those who remain in the shackles of poverty. This serves as the main
motivation of this research. In particular, the study focuses on income as a measure of socio-
economic well-being.
The Philippines is an important case study for examining how income distribution
evolves over time because of the seemingly weak impact of economic growth in reducing
poverty and inequality. In general, a country’s economic performance is usually measured in
terms of its gross domestic product (GDP). From 2009 to 2012, the country’s GDP per capita
has grown rapidly, expanding by 4.1% annually which is among the highest growth trajectories
in Asia (WDI 2014). Since 2000, GDP per capita grew at an annual rate of 3.1 (WDI 2014).
Given the large contribution of the household sector to GDP, it is intuitive to think that the
country’s observed economic growth would have an impact on the mean and the variance of
its household income distribution. However, a cursory look at conventional socio-economic
indicators seems to suggest that the benefits of economic expansion have not had a significant
impact on the income distribution with headcount poverty and inequality rates remaining high
and barely moving (WDI 2014). There are several possible reasons why this could be the case.
First, it is possible that the observed economic growth is accruing to the richest households
only while the income of the remaining households is fixed. To some extent, the propensity of
19
the richest households to be under-represented in household surveys may partially explain why
economic growth does not seem to have a significant impact on the mean and variance of the
household income distribution. Second, there is a growing gap between the coverage of
national accounts and household surveys. For example, rents of homeowners and the value of
financial services are imputed in the national accounts but not in household surveys (WB 2010).
Third, it is possible to attribute this observation to the limitations of conventional indicators
like average household income, cross-sectional poverty and inequality which simply quantify
changes in the marginal distribution of income. Previous studies suggest that such an approach
does not give a complete appraisal of the dynamics of the country’s development process
(Reyes, Tabuga, Mina, Asis & Datu 2011; Bayudan-Dacuycuy & Lim 2013). The objective of
this study is to provide a benchmark for probing beyond indicators of marginal distributional
income changes by adopting a longitudinal approach in examining income mobility patterns.
This approach is intended to provide a better means of understanding the dynamics of living
standards in the Philippines.
Although several studies have analysed growth, poverty and inequality in the Philippines
(e.g., Balisacan & Fujisaki 1998; Balisacan & Pernia 2002; Pernia 2003; Schelzig 2005; Aldaba
2009), most have only provided static snapshots of poverty and inequality in the country. While
these indicators are useful for studying how income distribution evolves over time, they are
unable to identify whether those who are initially poor are able to get out of poverty and in
what ways the persistently poor differ from the transiently poor. To address this shortcoming,
I use conventional and newly developed statistical methods that are designed to take into
account the temporal dimension of the income distribution using longitudinal data rather than
a series of cross-sectional data. The use of a portfolio of statistical techniques allows us to
examine the robustness of results, which in turn enhances the reliability of inferences for
drawing policy implications.
The importance of providing a longitudinal perspective in income distributional analysis
can be illustrated with the following simple example. Consider two scenarios for a two-
individual society consisting of one income-poor and one non-poor individual. In the first
scenario, incomes of both individuals did not change between the initial and final time periods.
In the second scenario, the individuals switched incomes. Although there is a complete reversal
of income ranks in the second scenario, the cross-sectional estimates will portray stagnant
poverty distribution in both scenarios since this approach does not distinguish between
persistent and transient forms of disadvantage. From a policy perspective, it is important to
distinguish between these scenarios because they could require different policies. For instance,
20
the first scenario calls for a policy that would address persistent forms of disadvantage while
the second scenario calls for a policy that would minimize socio-economic vulnerabilities.
In general, socio-economic mobility attempts to measure how people’s living standards
change across space and over time. It describes the ability of people to partake in the socio-
economic opportunities created by economic growth. In addition, it also determines which
disparities in economic opportunities at a specific point in time can be used as a measure of
welfare inequality over the long term (Fields 2008). If any person can work their way up the
social hierarchy, then high inequality may be less consequential for a country’s long-term
growth trajectory. By exploiting longitudinal data and adopting richer methodologies, we can
broaden our understanding of socio-economic mobility in the Philippines and eventually
contribute to a more evidence-based policy-making.
The thesis is divided into three parts. The first part consists of two chapters that provide
the building blocks needed for examining income mobility in the Philippines. In Chapter 1, I
address the question ‘how is income mobility measured?’. Here, I discuss two important
analytical considerations. First, I review several definitions of income mobility and explain
how each concept is operationalized using empirical data. The discussion highlights that a good
understanding of income mobility would require taking a multidimensional perspective.
Secondly, I discuss why high levels of income mobility should not always represent a desirable
outcome as conventionally perceived due to the inflationary impact of measurement errors and
transient income fluctuation on mobility estimates. In Chapter 2, I provide a background of the
socio-economic history of the Philippines over the past thirty years. The discussion is centred
on answering the question ‘has economic growth been pro-poor in the Philippines?’. A
simple simulation experiment showed that shifting from a cross-sectional to a longitudinal
perspective could portray the economic development process in the Philippines in a different
light than is conventionally perceived. This identifies the need to analyse income mobility
patterns using actual panel data.
The second part of the study discusses the results from the empirical investigation of
income mobility patterns in the Philippines. In Chapter 3, I describe the longitudinal data from
the Philippine Family Income and Expenditure Survey and Labour Force Survey which serve
as the main data sources for the succeeding analyses. The period of time covered in this study
coincides with the decade that precedes the rapid economic growth episode currently
experienced by the country. I split the analysis into two main periods with both similarities and
differences in economic conditions. In Chapter 4, I ask the question ‘is there income mobility
in the Philippines?’. As pointed out earlier, it is tempting to think that the income distribution
21
in the country is stagnant based on cross-sectional trends of poverty and inequality. Contrary
to this perception, I find that significant amounts of positive and negative mobility exist, but
they tend to offset each other and result in small changes in the income distribution at the
aggregated level. There is also some evidence that transitory fluctuations contribute
significantly to the observed level of income mobility. After establishing that there is
considerable amount of mobility in the country, I focus the discussion on mobility at the low-
income range in Chapter 5. In particular, I ask the question ‘how long do people stay in
poverty?’. The results suggest that poverty in the Philippines can be considered as mostly
chronic or persistent in nature. Thus, the finding that the initially poor experienced slightly
better income mobility outcomes may not be enough to eliminate poverty soon. Despite faster
income growth, it may take generations for the poor to exit poverty if they are trapped in vicious
cycles of socio-economic deprivation. Nevertheless, the results presented in Chapter 5 also
suggest the relative importance of transient poverty increases as the poverty line decreases or
the poverty measure under consideration becomes more sensitive to the illfare of the poorest
of the poor. This result highlights the importance of examining the robustness of poverty
estimates to measurement parameters and estimation methodology. After examining poverty
dynamics, Chapter 6 discusses the relationship between inequality and income mobility where
I ask the question ‘who are the income-mobile?’. The statistical analyses lead to mixed
findings. While there is evidence that initially disadvantaged households experienced slightly
better income mobility outcomes than initially high-income households, the differences in their
income trajectories taper off when a proxy measure of permanent income rather than initial
income is used to group households into different income groups. The results also suggest that
advantaged households had the most erratic income fluctuations, experiencing the highest
income gains (losses) in 2003-2006 and highest income losses (gains) in 2006-2009.
The third part of this thesis briefly discusses policy implications on and future directions
for income mobility studies in the Philippines from both substantive and technical viewpoints.
After providing a longitudinal perspective of poverty and income inequality and given the fact
that a good understanding of the factors that contribute to income mobility is important in
identifying appropriate policy intervention programs, I ask the question ‘what drives income
distribution dynamics in the Philippines?’ in Chapter 7. To address this question, I use an
exact accounting procedure in measuring the contribution of socio-economic capital, socio-
economic returns on capital and shocks on the observed changes in poverty and income
inequality. The results indicate that the higher levels of ownership of assets and higher
economic returns to formal and non-agricultural employment have contributed to lower poverty
22
while human capital and access to basic services remain stagnant. However, the results also
portray the impact of changes in socio-economic capital and changes in economic returns to
capital as offsetting forces that contribute to slow poverty and inequality reduction despite the
rapid economic growth that the Philippines has experienced since the beginning of the 21st
century. Furthermore, the findings in this chapter also point to the importance of accounting
for socio-economic shocks when outlining poverty-alleviation strategies. Given the importance
of employment in inducing income mobility and the increasing prevalence of non-traditional
employment, Chapter 8 investigates whether non-traditional employment can induce upward
mobility using multiple job holding as a case study by asking the question, ‘is multiple job
holding correlated with income mobility?’. Unlike in industrialized countries where
pluriactivity is used to enhance one’s skills to be able to move into better occupations, I find
that multiple job holding in the Philippines is mainly used as a coping mechanism, a stark
indicator of the high prevalence of underemployment and limited leverage against risks of
income shortfall. This suggests that the emergence of non-standard employment arrangements
which are not usually covered by labour policies can push workers trapped in precarious jobs
to poverty. On the other hand, Chapter 9 takes a step further in investigating the future of
income mobility literature in the Philippines (and other developing countries) with regard to
data availability. To reconcile the need for providing a more dynamic perspective of the
evolution of income distribution with the lack of panel data in developing countries, I ask the
question ‘how can we measure income mobility when there is no (or lack of) panel data?’.
I evaluate the performance of several pseudo-panel estimation techniques in measuring a wide
array of income mobility indicators and find that methods with more flexible income model
specifications perform better than those with highly parameterized models. More importantly,
these flexible pseudo-panel procedures produced estimates of poverty dynamics and
movement-based indices which are consistent with the estimates computed from the actual
panel data. Nevertheless, further improvements are warranted to be able to develop a more
satisfactory estimation procedure for indices measuring temporal dependence and the
inequality-reducing effect of income mobility.
In summary, despite the faster economic growth rates that the country has experienced
over the past decade, one of the main recurring findings of this study is the presence of a strong
offsetting effect of income losses and income gains across space, over time and between socio-
economic capital, socio-economic returns to capital and shocks. These offsetting forces have
contributed to the seemingly stagnant income distribution in the country. In addition, I also
find that the amount of mobility is significantly lower when permanent rather than current
23
income is used. Overall, these results call for the need to identify a more aggressive mix of
policy recommendations that will facilitate inclusive economic growth, minimize socio-
economic vulnerabilities, and reduce poverty and long-term inequality in the Philippines more
sustainably.
This research can be considered as a benchmark for future income mobility analysis in
the Philippines. It surpasses conventional studies which have sketched poverty and inequality
in the country as a one-time event and ignored the persistence and recurrence of such states
over time. The analyses presented here strike a balance between improving (household) income
distribution measurement theory and enhancing the accessibility of distributional statistics to
policymaking. The topics covered in this study are quite broad because as a benchmark study,
one of the main objectives of this research is to demonstrate the use of a wide range of analytical
tools that enrich income distributional analysis by exploiting the longitudinal feature of
household panel data. Although broad policy implications of the findings are discussed, very
detailed policy recommendations are reserved for future studies. One of the potential
shortcomings of this study is the lack of detailed analysis of the spatial disparities in income
mobility. This is mainly driven by data limitations as the panel data used in this study can
provide reliable estimates at the national and broad-regional levels only. Future research that
will further explain this topic can consider using small area estimation techniques in order to
provide more disaggregated estimates of income mobility. In addition, since the study covers
three time periods only, the use of more sophisticated analytical tools (e.g., multi-level models)
that require more time points is very limited. Nevertheless, while more research is warranted to
address these limitations, I hope this thesis helps to shift the attention of policymakers away
from static and towards more dynamic measures of well-being to better understand how this
analytical approach will affect the direction of policies and programs on inclusive growth.
24
Chapter 1 Analytical Tools for the Analysis of Income Mobility
1.1 Introduction
The concept of well-being refers to human welfare in a wide range of aspects such as
income, health, education, work, family, and other things that are important to people’s lives.
Implicitly related to this concept is the term socio-economic deprivation which refers to the
lack of capability for valuable functionings that will allow one to tap opportunities needed to
improve his/her living standards (Sen 1999). The measurement of well-being is an essential
component of policy-making. For instance, a society’s progress or socio-economic
development is often gauged by how much people’s well-being or living standards have
improved and by how much socio-economic deprivation has been reduced over time.
Measuring well-being also allows socio-economic planners to identify policy priorities that
will put the needs of the people first and will address the challenges that societies face ahead
(Organisation for Economic Co-operation and Development (OECD) 2013). A good example
is the implementation of the MDGs which show a clear, global commitment to end the multiple
dimensions of poverty and disadvantage.
There are different ways of measuring a country’s overall well-being. Initially, traditional
metrics have centred on the monetary aspects of well-being. Until 1970s, the use of GDP and
other income-based measures flourished.1 Later on, there had been emphasis in understanding
how income is distributed throughout the population and this led to the emergence of various
measures of income poverty (Foster, Greer & Thorbecke (FGT) 1984) and income inequality
(Shorrocks 1982). However, narrow conceptualisations of disadvantage that are solely based
on income ignore the fact that people have different capabilities to convert income into
resources that improve living standards (Callander, Schofield & Shrestha 2011). For instance,
people may have sufficient money to purchase food, but face difficulties in accessing markets
that sell food because they live in remote areas. Conventional income-based measures of
poverty would classify these individuals as non-poor. A better way to conceptualize
disadvantage is in terms of the lack of capabilities, freedom or resources to participate in
mainstream society (Nussbaum & Sen 1993; Sen 1999; United Nations Development
Programme (UNDP) 2008). This implies a shift from conceiving disadvantage in terms of ‘the
1
Here, income is used as a general term to encompass different monetary measures of well-being. In Chapter 8,
I use income in a more specific context to refer to the amount of money or its equivalent that accrue to an individual
or group of individuals as a result of an economic transaction such as rendering labour, sale of goods or services,
returns from investments. This definition distinguishes income from consumption expenditure. In other chapters,
household expenditure is referred to as income.
25
means of living’ people dispose of to the ‘opportunities’ they are given to choose the life that
they want to live (McLachlan, Gilfillan & Gordon 2013). To shift the attention of policymakers
away from conventional monetary-based measures of well-being and socio-economic
deprivation in the 1980s, researchers started exploring more holistic measures of human
development in the 1980s (McGillivray 2006). In 1990, the UN published its first Human
Development Report which also launched the Human Development Index (HDI), a simple
yardstick that expresses well-being in terms of income, health and education (UNDP 1990).
Despite its limitations, HDI is still widely used to complement traditional income-based
indicators in measuring socio-economic development among developing countries (UNDP
2013). On the other hand, more advances have been made in terms of developing holistic
measures of well-being in industrialized countries due to its more advanced data collection
systems. For instance, the Better Life Index (BLI) proposed by OECD covers eleven
dimensions of well-being which include housing, income, jobs, community, education,
environment, governance, health, life satisfaction, safety and work-life balance (OECD 2011).2
Although contemporary leading poverty researchers are unambiguous in declaring that
disadvantage goes beyond income deprivation, with the debate progressively moving into the
multi-faceted nature of socio-economic disadvantage, income remains a very important
resource and a gatekeeper to participation in socio-economic transactions (Harding &
Szukalska 2000). Furthermore, there is still much to learn about how to maximize the
information provided by income data to better understand societal progress, particularly in
developing countries where there are limited non-pecuniary socio-economic indicators. For
instance, estimates of income poverty and inequality are conventionally presented as static
cross-sectional aggregates which do not provide any information as to how persistent is poverty
for different groups of people. Ignoring how people’s incomes move over time may lead to
inappropriate policy interventions. To illustrate how solely relying on these static indicators
masks important features of a country’s development process, consider a hypothetical
population consisting of ten people. Of these ten people, four were rich and six were poor in
the initial time period. Suppose in the subsequent period, all initially rich became poor and all
initially poor became rich. Furthermore, suppose the incomes of the initially poor became
2
Ravallion (2011) argues that the indices such as HDI and the Multidimensional Poverty Index (MPI) recently
proposed by researchers from Oxford Poverty and Human Development Initiative (OPHI) (Alkire and Santos
2013), suffer from issues about marginal rates of substitution. In particular, according to Ravallion (2011), it is
difficult to find an economic theory that can justify how multiple indicators should be weighted to form a singular
index because there is no consistent valuation across different dimensions of well-being. Rather than a single
multidimensional index, Ravallion (2011) thinks that developing a reliable set of multiple indicators of poverty
and deprivation should be given more priority.
26
higher than the initial incomes of the initially rich. This scenario would lead to a twenty-
percentage point reduction in poverty rate and an increase in inequality. Whether this turn of
events is good or bad is subject to value judgment. On the good side, the development process
has allowed the initially poor to catch up with the rest of the population. However, the complete
reversal of incomes may also portray a very unstable income distribution. In general, while
changes in the cross-sectional estimates of poverty and inequality are useful in providing a
general picture of a developing country’s socio-economic progress, they do not provide a
complete appraisal of the temporal dynamics of the development process.
The study of income mobility is an emerging research paradigm in income distributional
analysis which capitalizes on the increasing availability of panel data (Fields 2011). Unlike
static cross-sectional measures which are based on changes in the marginal distribution of
income, income mobility is measured based on the joint temporal distribution of income. A
good understanding of income mobility trends is important for evidence-based policy planning
because different income mobility regimes call for a different mix of policies. For instance,
large increases in cross-sectional estimates of inequality may merit less concern when they are
accompanied by high income mobility rates since it means that the poor can eventually get out
of poverty. Analogously, a significant increase in cross-sectional poverty and inequality may
be problematic when it is accompanied by low levels of income mobility. Nevertheless, high
income mobility rates may not always be desirable. As pointed out in the example earlier,
complete reversal of incomes wherein the richest swaps income with the poorest, second richest
swaps income with the second poorest and so on, will still give the same cross-sectional
estimates of poverty and inequality but at the same time will portray a highly unstable income
distribution.
In summary, the main point that the discussions presented in this chapter seeks to convey
is that while the analysis of income mobility provides a broader picture of the income
distribution dynamics than conventional static socio-economic indicators, there are also
additional technical considerations that have to be carefully taken into account. In general, the
accompanying methodological decisions may lead to variations in research findings which in
turn could cause confusion among non-technical audience. Policy responses to findings of
income distributional analysis may lead to sub-optimal intervention programs when the
estimates are not well explained to policymakers and other key stakeholders. In contrast, when
the robustness of the results to measurement parameters are carefully examined, income
mobility research can empower users and stakeholders by providing them with a better
understanding of the impact of methodological decisions on research findings.
27
As a springboard for the analyses presented in the succeeding chapters, this chapter
reviews the state of the art of examining income mobility by presenting the analytical tools that
are used throughout this study. The rest of the discussion highlights that while examining
income mobility patterns enriches our understanding of the evolution of poverty, inequality
and income distribution, in general, there are additional technical considerations that have to
be taken into account. A good understanding of these substantive and methodological issues
will help us communicate the results of income distributional studies better so that it will feed
into public discussion. This chapter addresses the following questions:
(i) Why it is important to probe beyond cross-sectional trends of poverty, inequality and
economic growth when examining a country’s income distribution?
(ii) How do we measure the different dimensions of income mobility?
(iii) What are the important considerations when examining income mobility?
1.2 What is Income Mobility and Why it is Important?

Income mobility can be likened to a concept of a ladder where the ladder represents the
income distribution.3 Some individuals climb up while others slide down. People also move
from one step to another at different rates. Researchers have offered several interrelated
reasons explaining the relevance of examining the patterns at which people are moving along
the income distribution. First, income mobility is often regarded as a corrective tool for the
negative impacts of high or increasing inequality. Friedman (1962 p.1971) articulated this
hypothesis using an example, “Consider two societies that have the same distribution of annual
income. In one, there is great mobility and change so that the position of particular families in
the income hierarchy varies widely from year to year. In the other, there is great rigidity so
that each family stays in the same position year after year. Clearly, in any meaningful sense,
the second would be the more unequal society.” To elucidate this idea further, it is helpful to
first distinguish the two types of inequality: inequality of outcomes and inequality of
opportunities.4 Consider a simple scenario wherein a new job needs to be filled. Applying and
being considered for this job represents an “opportunity” while being hired or denied represents
an “outcome.” In this example, there is inequality of opportunity when job applicants with the
3 The concept of income mobility originated from the notion of social mobility. Broadly defined, social mobility refers to the
shifting of individuals in social space (Sorokin 1927, 1959). Defining the segments of this social space is based on the
stratification of individuals’ well-being. There are several ways of doing this. Sociologists use occupations as the basis for
social stratification while economists use income as their yardstick.
4 The relationship of income mobility with inequality of outcomes and opportunities is more commonly discussed in an
intergenerational context. Income mobility, when measured in terms of correlation between children’s and parental income, is
consistent with the meritocratic idea that an individual’s well-being should depend on their own abilities and efforts rather
than their parents’ (Jenkins 2011). Nevertheless, this can also be extended in an intragenerational context.
28
same level of skills are discriminated on the basis of factors that do not have anything to do
with their expected job performance and for which they do not have control of. Examples of
these factors are sex, ethnicity, religion, among others. On the other hand, if selection is based
solely on skills wherein the most skilled has the highest probability of getting the job, then any
variation in employment outcomes would represent inequality of outcomes. More generally,
when new opportunities in the form of wealth, income, socio-economic services, among others
are created in the course of economic growth but the chance to access these new opportunities
is mediated by factors that are beyond a person’s control, we can say that there is inequality of
opportunities. Otherwise, traditional view of inequality suggest that if outcomes reflect the
level of effort, then inequality of outcomes will just be a result of variations in effort (Roemer
1998; Bardhan, Bowles & Gintis 2000).5 Following this argument, increasing or high level of
cross-sectional inequality need not have negative normative implications and should only be
considered as a “distributional problem” if it is predominantly characterized by inequality of
opportunities (Atkinson, Bourguignon and Morrison 1992).
By providing a wider perspective of how the distribution of income evolves over time,
the examination of income mobility offers a way to distinguish inequality of opportunities from
inequality of outcomes (Shorrocks 1978; Fields 2010). For instance, a scenario wherein cross-
sectional inequality is increasing at the same time that the initially poor continuously
experience low income mobility prospects may portray inequality of opportunities. On the
other hand, a scenario wherein both initially poor and non-poor are enjoying high levels of
income mobility may portray inequality of outcomes (Gottschalk 1997; Van Kerm 2006).
Several studies provide empirical evidence of why inequality of outcomes could be of less
concern than inequality of opportunities in the long-run. Using World Values Survey data for
thirty OECD countries, Fisher (2009) finds that living in a perceived socio-economically
mobile society uplifts individual life satisfaction. On the other hand, Tocqueville ([1856] 1986)
surmised that when upward income mobility of others is unaccompanied by one’s own, the
initially poor who did not experience income mobility may resort to political actions (e.g., hold
political rallies) which they think would help them become upwardly mobile like their non-
poor counterparts (Hirschman & Rothschild 1973).
Different income mobility regimes call for a different mix of policies. A good
understanding of the income mobility patterns enables policy planning and more efficient
5 The works of Roemer, Bardhan, Bowles & Gintis provide excellent discussion about traditional and contemporary views of
inequality. One of their main arguments is that it makes more sense to identify which mechanism of transmission of advantage
is justified and which is unfair instead of pursuing an abstract objective of disconnecting initial levels of advantage and
economic outcomes.
29
allocation of efforts and resources.6 In theory, there are several possible types of income
mobility regimes; zero mobility and perfect mobility are the two extreme cases. A society is
said to have zero mobility if the income of every unit in the population is held fixed and there
is perfect mobility if the conditional distribution of income destinations is the same for every
income origin. In empirical application, the true income mobility regime often falls in between
these two extreme cases. When initially poor people experience low income mobility, they
become trapped in poverty and this could lead to the perception that there is no merit to work
hard because the chance of getting out of poverty is limited. Without any mobility-enhancing
intervention, such case could lead to a vicious cycle of poverty. This represents a significant
waste of human resources (Asian Development Bank (ADB) 2012). Thus, finding an
appropriate redistributive policy (e.g., conditional cash transfer) might be the way to counter
this problem. Analogously, a scenario where high income range people experience low income
mobility, portrays a system where socio-economic advantage accumulates over time. To be
able to make the high income state more permeable to people from lower income segments,
creation of additional high quality jobs and provision of trainings to meet the skill-requirement
of these new jobs might be the way to move forward. In these two examples, low income
mobility is portrayed as a problem that has to be addressed. However, the opposite case, i.e.,
having high income mobility, is not necessarily a desirable outcome. For instance, a high level
of income mobility that is mainly driven by large fluctuations in transitory income could
represent a very unstable economic system (Jarvis & Jenkins 1998). Analogously, low levels
of income mobility may also be regarded as a good indicator if it represents a mature economy
that has already achieved long-run equilibrium.
To be able to identify the ideal level of income mobility, it is first important to understand
how mobility is conceptualized. The next section enumerates the different definitions of
income mobility and discusses how various conceptualisations can lead to qualitatively
different income mobility trends.
1.3 How Do We Measure Income Mobility?

Technically, income mobility can be regarded as a transformation between two income
vectors over a period of time. However, there is a long standing debate as to which features of
the vector transformation characterise low or high mobility (Maasoumi 1998; Fields & Ok
6 It is often considered that static analysis of development is more suitable for treating symptoms of socio-economic
disadvantage while a more dynamic analysis allows us to tease out causal relations among different factors and thus plan a
more effective intervention of escaping disadvantage.
30
1999a; Fields 2008, 2009). The debate can be partially attributed to the different beliefs on
which characteristic aspects of income are important to be examined over time. In addition to
the different perspectives on what characterizes mobility, Ferreira et al. (2013) identified two
other basic considerations that one needs to navigate to be able to measure income mobility:
(i) concept of income and (ii) the time gap between the two income vectors. The first
consideration entails choosing a specific type of income for which mobility should be
measured. Hence, it answers the basic question mobility of what. As pointed out earlier, the
term “income” is used throughout this study as a general term that encompasses different
monetary measures of well-being and there are several types of income that could be of interest
for mobility research. For instance, some researchers prefer to use earnings which includes
regular cash income plus other income received from transfers while other researchers argue
that consumption expenditure provides more insights into a person’s economic well-being than
the information provided by earnings. Chapter 3 discusses this issue further. The second
consideration entails choosing the length of observation period for which income mobility is
to be measured. It answers the question how far apart in time the two income vectors are from
one another (Ferreira et al. 2013). In this context, there are two modes of measuring income
mobility – intra-generational and intergenerational. Intra-generational income mobility
corresponds to the historical income trajectories of the same set of people while
intergenerational mobility refers to the income history of people from the same lineage across
generations (i.e., parents and children) (Ferreira et al. 2013). The focus throughout the study is
intra-generational mobility.7
The following section provides a general overview of some of the commonly used
conceptual definitions of income mobility. In this review, I adopt the taxonomy proposed by
Fields (2008) who defined income mobility in three perspectives: mobility as movement,
mobility as origin independence and mobility as equalizer of long term income.8 While the
measures under these perspectives may differ in terms of functional forms, the choice of
income mobility concept to be examined goes beyond the differences in formula and should be
tailored to specific research questions that a study wants to answer (Ferreira et al. 2013).
7 Although it is worthwhile to examine both intra-generational and intergenerational mobility as they capture short to medium-
term and long-term changes in income distribution, respectively, the analyses presented in the succeeding chapters correspond
to intra-generational mobility because information about parental income is not available.
8 It can be argued that these three income mobility perspectives are not mutually exclusive. For instance, Jenkins (2011)
categorized the concepts of income mobility into four broad groups: mobility as positional change, mobility as individual
income growth, mobility as reduction of longer-term inequality and mobility as income risk. In his discussion, the mobility as
origin-independence perspective proposed by Fields (2008) is subsumed under mobility as positional change.
31
For notation, the succeeding discussion assumes that the target population consists of N
individual units whose incomes are observed for two different time points; the term Yit is used
to denote the ith individual’s income at time t.
1.3.1 What Characterizes Mobility of Incomes?
The first main perspective views mobility in terms of income movements. There are four
sub-concepts within this perspective. First, income mobility may be gauged in terms of gross
movements or what is commonly referred to as income flux. Mobility measures following this
concept can be expressed as functions of the absolute income differences denoted by |Yi2 – Yi1|.
Second, mobility measures based on net income movements can be expressed as functions of
actual differences in income levels denoted by (Yi2 – Yi1). Unlike the first concept, the second
concept distinguishes positive (or upward) from negative (or downward) income mobility and
hence, other researchers also refer to the income flux and net income movement as non-
directional and directional mobility, respectively. The third concept under the movement
perspective views mobility in terms of changes in income shares denoted by ∑𝑌𝑌𝑖1 - 𝑌𝑖2
∑ 𝑌𝑖2
while
𝑖1
the fourth concept views mobility in terms of the changes in income ranks denoted by
Rank(Yi2) – Rank(Yi1). In summary, the first two concepts measure changes in income levels
(absolute mobility) while the last two concepts examine changes in an individual’s income in
relation to the incomes of everyone else in the society (relative mobility). Hence, the last two
concepts of mobility depend not only on whether an individual’s income changed over time
but also on how the change alters his/her income share or income rank (Jenkins 2011).9
The second perspective views mobility in terms of the extent to which an individual’s
income in the past influences his/her current income (Lillard & Willis 1978). This is referred
to as origin independence perspective. The basic property underpinning this perspective is that
a more mobile society is one where an individual’s first period-income is less important in
predicting his/her income in the succeeding periods (Ferreira et al. 2013). In other words,
mobility is high when an individual’s income destination is weakly correlated to one’s income
origin and it is low when the correlation between the income origin and income destination is
strong.
The third main perspective views mobility as an equalizer of long-term incomes. This
perspective evaluates the extent to which long-term incomes are distributed more or less
9Under the positional movement concept, it is not possible for all individuals to be uniformly upwardly (or downwardly)
mobile (Jenkins 2011).
32
equally over time. In general, an individual’s income at any given time will differ from his/her
average income taken over several successive time periods. Using temporal average income
will smoothen the longitudinal variability in each individual’s income as well as the variability
across individuals.10 The equalizer of long-term income perspective characterizes mobility in
terms of the speed at which inequality is reduced as the observation period is lengthened
(Shorrocks 1978, Jenkins 2011).11 Under this perspective, mobility is high when the inequality
in longitudinally-averaged income is less than the inequality at any particular point in time
(Ferreira et al. 2013).
To impart meaning to the discussions above, consider the example provided in Table 1.1
which shows four income mobility scenarios labelled as S1, S2, S3 and S4, based on a
hypothetical population consisting of four persons labelled as A, B, C and D. Each number
corresponds to the amount of income units that each individual holds at a specific time point.
First, we can use Scenario S1 to illustrate the difference between gross and net income changes.
Under S1, the incomes of persons A and B increased by one unit each while that of persons C
and D dropped by one unit. Summing up the absolute income differences across the four
individuals is one of the many ways of measuring total income flux. In this example, the sum
is four income units. On the other hand, summing up the individual income differences without
taking the absolute value gives us a value of zero. Thus, whereas the first concept yields a non-
zero estimate of mobility, the second concept leads to null mobility. The second scenario can
be used to illustrate the difference between measuring income mobility in terms of changes in
income levels and changes in either income share or income ranks. Under S2, the final incomes
of persons A, B, C and D are twice their respective initial incomes. Summing up the income
differences across all persons would give a non-zero estimate of mobility whereas summing up
the changes in either income shares or income ranks gives a mobility estimate of zero. The
third scenario illustrates the difference between movement-based and origin independence-
based perspectives of income mobility. Under S3, person A swaps income with person D and
person B swaps income with person C. In this scenario, the correlation between the initial and
final income vectors is -1. Since the initial income vector perfectly predicts the values of the
final income vector, we can say that there is no mobility based on the origin independence
perspective. Similarly, since the observed increases in the income of the initially two poorest
persons offset the observed income declines of the remaining two persons, we can also say that
there is no directional income mobility under this scenario. However, since all persons
10 The longitudinally averaged income can be used to approximate permanent or long-term income.
11 The speed of inequality reduction depends on the chosen inequality measure (Schluter and Trede 2003).
33
observed change in both the actual income levels and income ranks, Scenario S3 portrays a
mobile society in terms of non-directional income mobility and positional movement. Lastly,
the fourth scenario illustrates the difference between mobility based on origin independence
and mobility as equalizer of long-term income perspectives. The relatively high yet negative
correlation (i.e., -0.8) between the two income vectors implies that there is low income mobility
based on the origin independence perspective. However, since inequality in average income
(4.5, 3, 4.5, 3) is lower compared to either the initial or final income vector, we can say that
there is mobility based on the concept of equalizer of long-term income.
This section has discussed the differences between various concepts of income mobility
that are commonly used in the literature. The concepts explained here are used intensively
throughout the thesis, particularly when I describe the income mobility patterns in Chapter 4.
The next section enumerates several analytical tools that can be used to compute these concepts
using empirical data.
Table 1.1 Differences Between Income Mobility Perspectives

Initial income Final income
Scenario
vector vector
A BC D A BC D
S1 (1, 2, 3, 4) (2, 3, 2, 3)
S2 (1, 2, 3, 4) (2, 4, 6, 8)
S3 (1, 2, 3, 4) (4, 3, 2, 1)
S4 (1, 2, 3, 4) (8, 4, 6, 2)
1.3.2 Analytical Tools for Measuring Income Mobility
An income transition matrix is one of the most popular analytical tools used to summarize
income mobility under the movement perspective. In particular, a transition matrix with
𝑃(𝑏𝑗−1 ≤𝑌𝑖𝑡−1 <𝑏𝑗 , 𝑏𝑘−1 ≤𝑌𝑖𝑡 <𝑏𝑘 )
boundaries between R income classes consists of each element 𝑝𝑘|𝑗 = 𝑃(𝑏𝑗−1 ≤𝑌𝑖𝑡−1 <𝑏𝑗 )
which is equal to the conditional probability that individual i moves to class k of income at time
t given he/she was in class j at t-1, where 0 = b0< b1< … < bR-1< bR (Table 1.2). When income
mobility is low, one would expect that the diagonal elements of the transition matrix would be
close to one. On the other hand, when income mobility is high, the diagonal elements of the
transition matrix are likely to take equal values. Depending on one’s research objective, the
income classes can be specified in a number of ways. For example, it can be divided into R =
2 classes where the classes correspond to the state of being poor or non-poor. This is useful for
measuring movements into and out of poverty. More generally, incomes could be divided into
34
different income brackets according to absolute income cut-off points or income quantiles to
measure absolute and relative mobility, respectively.
To be able to compare two transition matrices, it is useful to summarize the mobility
implied from each transition matrix into a single scalar value. A simple approach is to count
the number of classes for which an individual moved during an observation period. For
example, if an individual moved from class 1 to class 3, then that individual is given a value of
2 because he/she moved by two classes. Equation (1.1) is computed by taking the average
number of classes moved across all individuals. The immobility ratio is another simple
summary measure that can be computed by counting the number of individuals who remained
in the same class and dividing the count by the total number of individuals in the target
population (1.2). However, one of the disadvantages of using an income transition matrix to
provide a compact summary of the income mobility process is that it ignores the mobility
experienced by individuals whose incomes changed yet remained in the same income class.
Table 1.2 Sample Income Transition Matrix

Poorest 2nd 3rd 4th Richest
Quintile
Quintile Quintile Quintile Quintile Quintile
Poorest Quintile 0.5 0.25 0.2 0.05 0
2nd Quintile 0.1 0.4 0.35 0.1 0.05
3rd Quintile 0.05 0.2 0.45 0.2 0.1
4th Quintile 0 0.05 0.35 0.45 0.15
Richest Quintile 0 0.05 0.15 0.25 0.55
∑𝑖 | ∑𝑅 𝑅
𝑟=1 𝑟∗𝐼{𝑏𝑟−1 ≤𝑌𝑖𝑡−1 <𝑏𝑟 }−∑𝑟=1 𝑟∗𝐼{𝑏𝑟−1 ≤𝑌𝑖𝑡 <𝑏𝑟 } |
𝑎𝑣𝑒 # 𝑠𝑡𝑎𝑡𝑒𝑠 𝑚𝑜𝑣𝑒𝑑 = (1.1)
𝑁
𝑐𝑎𝑟𝑑𝑖𝑛𝑎𝑙𝑖𝑡𝑦(𝐼{𝑏𝑟−1 ≤𝑌𝑖𝑡−1 <𝑏𝑟 ,𝑏𝑟−1 ≤𝑌𝑖𝑡 <𝑏𝑟 })
𝑖𝑚𝑚𝑜𝑏𝑖𝑙𝑖𝑡𝑦 𝑟𝑎𝑡𝑖𝑜 = (1.2)
𝑁
In addition to using a transition matrix, there are a number of alternative ways of

measuring income mobility. For instance, the class of indices proposed by Fields and Ok
(1999b) can be used to measure the first two concepts of income mobility under the movement
perspective. On the other hand, the average rank and King’s (1983) indices are used to measure
the third mobility concept under the movement perspective while Hart’s (1976) index can be
used to measure mobility based on the origin independence perspective. Furthermore, some
examples of mobility indices that can be used when taking the equalizer of long-term income
35
perspective include the measures proposed by Shorrocks (1978), Chakravarty, Dutta &
Weywark (CDW) (1985) and Fields (2010). Table 1.3 provides the computational formula for
each of these mobility indices.
While the list of concepts and empirical measures provided in this section is not
exhaustive, these are the most commonly used tools in mobility studies.12 It is important to
note that the differences in mobility estimates based on varying concepts and analytical tools
are not necessarily of limited practical interest because each concept corresponds to inherently
distinct notions of what mobility is (Ferreira et al. 2013). Figure 1.1 illustrates this point. Using
simulated data, I compare the relationship among different income mobility indices. While
there is a general linear (either positive or negative) relationship among the indices considered,
the strength of the correlation depends on the income mobility concept being measured. For
instance, indices that do not differentiate between downward and upward income mobility
(e.g., Fields-Ok’s, King’s, average rank jump, Hart’s and Shorrocks’ indices) are strongly
correlated with each other but exhibit more variability when compared to equalization indices
and poverty dynamics.
In providing a descriptive summary of the income mobility patterns in Chapter 4,
measures related to the Fields-Ok’s, average rank jump, Hart’s, Shorrocks’ indices and
transition matrices are used intensively. This mix of indicators is chosen to provide an optimal
insight to the different perspectives of mobility without having to use a very long-list of
overlapping indicators.
1.4 Incorporating Income Mobility in the Analysis of Poverty, Inequality and Pro-
Poor Growth
A number of studies have emphasized the usefulness of capturing the temporal dynamics
of poverty, inequality and pro-poorness of growth for policy-targeting (Grimm 2007; Van
Kerm 2009; Bourguignon 2011; Palmisano & Peragine 2014; Palmisano & Van de gaer 2013).
Conventional measures of poverty, inequality and pro-poorness of growth are focused on
understanding the pure cross-sectional effect of economic growth and ignores the identity of
income recipients (Palmisano & Peragine 2014). In technical parlance, this is termed as the
temporal anonymity property. However, people change their income positions over time.
Today’s poor (rich) is not necessarily the same set of people who were poor (rich) yesterday.
Thus, if the objective is to understand how growth impacts the living conditions of the initially
advantaged or disadvantaged, then it is important to capture income mobility. This section
12 Fields (2008) provides a more comprehensive review of the various income mobility concepts used in the literature.
36
extends the discussion provided in Section 1.3 by reviewing the different ways of taking
income mobility into account when measuring poverty, inequality and pro-poorness of growth.
As in the previous section, the discussion highlights the differences between various
methodologies to demonstrate the sensitivity of resulting estimates to the choice of
measurement techniques.13
1.4.1 Measuring Intertemporal Poverty

Since the work of Sen (1976), many of the poverty estimation studies have centred on
tracking snapshots of poverty by examining the trends of its incidence, depth and severity over
time using repeated cross-sectional data collected at different time points. However, these
poverty measures are not very useful in distinguishing a society where most of the poor have
been trapped in socio-economic dearth for a long time from a society where many of the poor
have been experiencing transient downturn in fortunes (Bane & Ellwood 1986). Given that a
longitudinal perspective provides a broader picture of poverty and the panel data required to
provide this broader picture is gradually becoming more available, several advances in
measuring intertemporal poverty has been made over the years (Jalan & Ravallion 1998; Yaqub
2000; Foster 2009; Gradin, del Rio & Canto (GRC) 2012). Some of the ideal properties of an
intertemporal poverty measure include continuity, focus, monotonicity, scale invariance,
duration sensitivity and transfer axioms (GRC 2012).14
This section reviews several estimation tools for measuring persistent and transient
poverty using two analytical frameworks – components approach and spells approach. The
notations used in this section are slightly different from the notations used in Section 1.3. Here,
Yit is used to denote the ith person’s income at time t, normalized by the poverty line z. In other
words, Yit in this section is equal to the Yit used in the previous section divided by the poverty
line.
13 For an introductory review of poverty and inequality measurement, readers are referred to Foster et al. (2013).
14 According to GRC (2012), the continuity axiom states that any increase in household income at any time period when the
household is non-poor should not change the value of the intertemporal poverty measure. The monotonicity axiom states that
any decrease in household income during episode of poverty should increase the value of the intertemporal poverty measure.
The scale invariance axiom states that rescaling the income and poverty line by the same factor should not change the value
of the intertemporal poverty measure. The focus axiom states that any increase in the income of the non-poor should not affect
the intertemporal poverty measure. The duration sensitivity axiom states that an intertemporal poverty measure should be able
to differentiate the impact of shorter or longer poverty spells.
37
Table 1.3 Formula of Different Income Mobility Indices
Concept Index Formula
𝑁
𝑜𝑚𝑒𝑖𝑡
Fields-Ok | ( )|
𝑜𝑚𝑒𝑖𝑡−
𝑖
𝑁
| 𝑖 𝑜𝑚𝑒𝑖𝑡|
King ( )
𝑜𝑚𝑒𝑖
𝑖
𝑁
Income Movement Average Rank Jump | 𝑟𝑎 𝑜𝑚𝑒𝑖𝑡 𝑟𝑎 ( 𝑜𝑚𝑒𝑖𝑡− )|

𝑖
𝑁
Poverty Persistence { 𝑜𝑚𝑒𝑖𝑡 ,} 𝑒𝑟𝑒 {𝑖 𝑜𝑚𝑒𝑖𝑡− }

𝑖 𝑃
𝑁
Poverty Inflow { 𝑜𝑚𝑒𝑖𝑡 ,} 𝑒𝑟𝑒 {𝑖 𝑜𝑚𝑒𝑖𝑡− }

𝑖
Time dependence Hart 𝑜𝑟𝑟𝑒𝑙( 𝑜𝑚𝑒𝑖𝑡 , 𝑜𝑚𝑒𝑖𝑡− )
(∑ 𝑜𝑚𝑒𝑖𝑡)
Shorrocks ∑𝑡 𝑡 ( 𝑜𝑚𝑒𝑖𝑡)
Mobility as Equalizer of (∑ 𝑜𝑚𝑒𝑖𝑡)

Fields
Income ( 𝑜𝑚𝑒𝑖𝑡− )
Chakravarty, Dutta and 𝐼(𝐼𝑛𝑐 )

1
Weywark (CDW) 𝐼(𝐼𝑛𝑐 𝑖𝑡−1 )
Source: Fields (2008 and 2010)
Note: In the notations above, I{condition} is an indicator function which takes a value of 1 if the condition is satisfied and 0 otherwise. On the other hand,
I( ) is an inequality index. Z is used to denote the poverty line.
38
Figure 1.1 Relationship between Different Income Mobility Indices
Source: Author’s computations using simulated data and the Stata tools for income mobility analysis developed by Van Kerm (2002).
39
Measuring Intertemporal Poverty using the Components Approach
The components approach can be traced back to the permanent income hypothesis
proposed by Friedman which states that over a person’s lifecycle, each has his/her own
permanent income stream but it can have short-term transitory fluctuations from time to time
(Friedman 1957). The main interest of the components approach lies in disentangling the
contribution of the short-term and long-term components to a person’s income intertemporal
poverty status.
Jalan & Ravallion (1998) Approach
Jalan & Ravallion (JR) (1998) proposed using the longitudinal average as an estimate of
∑𝑇
𝑡=1 𝑌𝑖𝑡
permanent income. In particular, suppose 𝑌̅𝑖 = , 𝑔𝑖𝑡 = 𝑚𝑎𝑥( , 𝑌𝑖𝑡 ) and 𝑔̅𝑖 = 𝑚𝑎𝑥( , 𝑌̅𝑖 ). The
measures of total poverty and poverty persistence are given by (1.3) and (1.4), respectively. On
the other hand, transient poverty is estimated by subtracting poverty persistence from total
poverty (1.5).
𝑡 𝑡𝑎𝑙 ∑𝑁 𝑇 𝛼
𝑖=1 ∑𝑡=1 𝑔𝑖𝑡
𝛼 = (1.3)
𝑁
𝑟𝑠𝑖𝑠𝑡 ∑𝑁 ̅𝑖𝛼
𝑖=1 𝑔
𝛼 = (1.4)
𝑁
𝑡𝑟𝑎𝑛𝑠𝑖 𝑛𝑡 𝑡 𝑡𝑎𝑙 𝑟𝑠𝑖𝑠𝑡
𝛼 = 𝛼 𝛼 (1.5)
These measures are identified by the parameter α and hence, they are analogous to the class of
poverty indices proposed by Foster, Greer & Thorbecke (FGT) (1984). When α = 0, they
measure the incidence of intertemporal poverty, depth when α = 1 and severity when α = 2. If
all units have longitudinal average incomes that exceed the poverty line z, but income in some
time periods fall below z, Pαpersist will be equal to 0 while Pαtransient will be equal to Pα . On the
other hand, if all units have longitudinal average income falling below z, both PαTotal and Pαpersist
will take a value of 1 and consequently, Pαtransient will be 0.
One of the main caveats of the JR approach is that the values of the resulting poverty
measures do not necessarily range between zero and one because the Pαtotal is not always
guaranteed to be higher than Pαpersist or Pαtransient .
Duclos, Araar & Giles (2010) Approach

Duclos, Araar & Giles (DAG) (2010) also proposed using longitudinal average income
as an estimate of permanent income. However, unlike the JR approach, the DAG approach
differentiates a poor person whose income consistently fell below the poverty line throughout
the observation period from a person with the same longitudinal average income but
40
experienced both poverty and non-poverty. By doing so, the DAG approach takes into account
a person’s risk-aversion to unexpected income fluctuations. This is consistent with the notion
that a person’s (economic) disutility tends to increase as the variation in their income stream
increases which also leads to higher transient poverty (Gottschalk 1982). The DAG approach
entails computing the variability-adjusted poverty status of each person (1.6). The parameter α
≥ 1 represents a person’s level of risk aversion to income variations wherein a value of unity
means that the person is risk-neutral. On the other hand, the risk-premium that person i would
be willing to pay to be able to remove the variability in its poverty gap status is given by 𝛼 (𝑔𝑖 )
(1.7). Under the DAG approach, transient poverty is defined as the total cost that will be
incurred due to variability in poverty gaps over time while total poverty is the sum of the
average poverty gap in the population, the cost of inequality in equally-distributed poverty gaps
among individuals and transient poverty.15
1
1 ∑𝑇 𝛼 𝛼
𝑖=1 𝑔𝑖𝑡
𝑖𝛼 (𝑔𝑖 ) 𝛼 =( ) (1.6)
𝛼 (𝑔𝑖 ) = 𝑖𝛼 (𝑔𝑖 ) 𝑖 (𝑔𝑖 ) (1.7)

1
𝑡 𝑡𝑎𝑙 ∑𝑁 𝑇
𝑖=1 ∑𝑡=1 𝑔𝑖𝑡 ∑𝑁 𝑇 𝛼 𝛼 ∑𝑁 ∑𝑇
𝑖=1 ∑𝑡=1 𝑔𝑖𝑡
𝛼
𝑖=1 𝑡=1 𝑔𝑖𝑡 ∑𝑁
𝑖=1 𝛾𝛼 (𝑔𝑖 )
𝛼 = + ( ) - + (1.8)
𝑁 𝑁 𝑁 𝑁
𝑡𝑟𝑎𝑛𝑠𝑖 𝑛𝑡 ∑𝑁
𝑖=1 𝛾𝛼 (𝑔𝑖 )
𝛼 = (1.9)
𝑁
𝑟𝑠𝑖𝑠𝑡 𝑛𝑡 𝑡 𝑡𝑎𝑙 𝑡𝑟𝑎𝑛𝑠𝑖 𝑛𝑡
𝛼 = 𝛼 𝛼 (1.10)
Although the JR and DAG approaches both fall under the components framework of
measuring poverty dynamics, the two approaches have remarkable differences. For instance,
while the income movements above the poverty line influences a person’s persistent poverty
status under the JR approach, the same cannot be said about the DAG approach because it
censors income movements at the poverty line. In other words, under the JR approach, a person
experiencing numerous episodes of poverty may still be considered not persistently poor if
his/her income for at least one time period is high enough to make the longitudinally-averaged
income exceed the poverty line. On the other hand, since only poverty gaps are considered
when using the DAG approach, high incomes for few time periods cannot compensate for the
numerous episodes that a person spent in poverty. Another important difference between the
two approaches lies on how transient poverty is conceptualized. In the JR approach, transient
poverty is simply the difference between total poverty and poverty persistence whereas under
15 When a household is risk neutral (α = 1), transient poverty will be equal to 0.
41
the DAG approach, transient poverty is intimately linked with the level of risk aversion a person
has towards income fluctuations.
Measuring Intertemporal Poverty using the Spells Approach

Unlike the components approach, the spells approach treats a person’s poverty status in
each time period, independently. This approach is consistent with the arguments presented by
Jappelli (1990) stating that it is not safe to assume that individuals can “borrow” income from
different time periods when there are variations in the liquidity constraints over time.
Conventional Spells Approach

The conventional spells approach entails counting the number of time periods when the
observed income of each person fell below the poverty line. Then, a specific frequency
threshold 𝜏 ≤ T is used to distinguish transient from persistent poverty (1.13). For example, in
the study of Gaiha & Deolalikar (1993) covering nine years of data, the authors defined
persistent poverty as those whose income fell below the poverty line for at least five years. On
the other hand, in a study covering three survey years, Reyes et al. (2011) defined persistent
poverty as those who experienced income shortfall for at least two years.
𝑟𝑠𝑖𝑠𝑡 ∑𝑁
𝑖=1 𝑉𝑖
= (1.11)
𝑁
𝑡𝑟𝑎𝑛𝑠𝑖 𝑛𝑡 ∑𝑁
𝑖=1( −𝑉𝑖 )
= (1.12)
𝑁
where
𝑖𝑓 ∑𝑡 (𝑌𝑖𝑡 < ) 𝜏
𝑉𝑖 = { (1.13)
0, 𝑜𝑡 𝑒𝑟 𝑖𝑠𝑒
The intertemporal poverty measures presented in (1.11) and (1.12) have several
limitations. First, they only estimate the number of persistently and transiently poor and do not
provide any information about the depth and severity of intertemporal poverty. These measures
are not also absolutely sensitive to the duration of poverty. For example, a persistently poor
person who stays in poverty for an additional year because of lower income, will not reflect an
increase in Ppersist. Similarly, a transiently poor person who stays in poverty for an additional
year will not increase Ptransient as long as the time spent in poverty does not exceed 𝜏.
42
Foster (2009) Approach
To address the limitations of the conventional spells-based measures of poverty
dynamics, Foster (2009) introduced a class of intertemporal poverty measures that are sensitive
to poverty duration and can be used to estimate incidence (1.14), depth (1.15) and severity of
intertemporal poverty (1.16). The proposed measures are provided below:
𝐾0 = 𝐻𝐷 (1.14)
𝐾 = 𝐻𝐷𝐺 (1.15)
𝐾2 = 𝐻𝐷𝐺𝑆 (1.16)
where H is the proportion of the persistently poor persons (i.e., H = Ppersist), D is the average
duration that persistently poor persons spent in poverty, G is the average proportional income
shortfall of persistently poor persons and S is the average squared proportional income shortfall
of persistently poor persons. From this, it is straightforward to estimate transient poverty. First,
we estimate 𝐾(𝑌, , 𝜏 = 0) using the same formula where all poverty spells are accounted for.
Transient poverty is then estimated by subtracting poverty persistence from total poverty.
Although the Foster approach takes into account the number of episodes spent in poverty,
it does not take into consideration whether some poverty episodes occurred consecutively.
Hence, it fails to consider that continuous episodes of poverty can be more harmful than the
same number of periods spent in poverty but spread in between several episodes of non-poverty
(Bane & Ellwood 1986; Jappelli 1990).16
Gradin, del Rio and Canto (GRC) (2012) Approach

Gradin, del Rio & Canto (2012) noted that many of the techniques discussed above fail
to satisfy several properties of an ideal intertemporal poverty measure. For instance, the
conventional spells approach and Foster approach violate the poverty duration sensitivity
property because these approaches only take into account the number of poverty episodes
within the observation period but not the duration spent in consecutive episodes of poverty.
On the other hand, components-based measures such as the JR approach violate the
intertemporal focus, poverty spell duration sensitivity and regressive transfer axioms because
periods of high income compensate for periods of low income. To address this issue, GRC
proposed a class of intertemporal poverty measures which circumvent these limitations.
16 This is consistent with the state dependence hypothesis about poverty which states that the longer a household stays in
poverty, the lower the chance of escaping it.
43
𝛾 𝛽
𝑝𝑖𝐺𝑅 (𝑌𝑖 ; ) = ∑𝑡 𝑔𝑖𝑡 𝑖𝑡 (1.17)
such that
𝑧𝑖 −𝑌𝑖𝑡 𝛾
𝛾 ( ) 𝑖𝑓 𝑌𝑖𝑡 < 𝑡
𝑔𝑖𝑡 ={ 𝑧𝑖 (1.18)
0 𝑜𝑡 𝑒𝑟 𝑖𝑠𝑒
𝑠 𝛽
𝛽
𝑖𝑡 = ( 𝑖𝑡 ) (1.19)
∑𝑁
𝑖 𝑝𝑖𝐺𝑅 (𝑌𝑖 ; ) 𝑖𝑓 𝛼 0
𝑁
𝐺𝑅 (𝑌; ) = { 𝑞 (1.20)
𝑖𝑓 𝛼 = 0
𝑁
The term γ is analogous to the parameter used in the conventional poverty measures
proposed by FGT (1984) which indicates index’s sensitivity to the depth of poverty. The
parameter β indicates the level of sensitivity of the poverty measure to the duration of the
poverty spell. In particular, higher values of β provide more penalty to longer episodes of
poverty.
In summary, the discussion in this section suggests that there are various ways of
capturing the temporal dynamics of poverty. In Chapter 5, I examine the robustness of the
intertemporal poverty trends by estimating poverty using all the approaches presented in this
section.
1.4.2 Income Mobility-Adjusted Pro-Poor Growth Assessment

Economic growth is said to be pro-poor when it allows the disadvantaged to catch-up
with the rest of the population by providing an impetus for them to reach their full economic
potential. Hence, an economic growth that is pro-poor allows a country to maximize its
available human resources. Given these benefits, it is not surprising to note that the pro-poor
growth literature has flourished over the years and now, it offers a portfolio of methods for
measuring pro-poor growth (Kakwani and Pernia 2000; Kakwani & Son 2004, 2008; Ravallion
& Chen 2003; Araar et al. 2009; Duclos 2009; Essama-Nssah & Lambert 2009). However, like
conventional static measures of poverty, most of the existing pro-poor growth measures are
based on changes in the marginal distribution of income. As mentioned earlier, this is what is
referred to as temporal anonymity. Income distributional measures that satisfy temporal
anonymity property ignore any reordering of incomes among individual units. For example, if
the individual units swap income with each other, any income distributional measure satisfying
temporal anonymity should have the same value. Previous studies suggest that the level of pro-
poorness of growth is usually lower under the temporal anonymity assumption than when the
44
joint distribution of income is used (Fields & Puerta 2010). Furthermore, departing from the
temporal anonymity principle when assessing pro-poorness of growth provides a way of
evaluating the extent up to which the benefits of economic growth reach the initially poor and
those who just entered poverty. Thus, some researchers argue that pro-poorness of growth is
better measured using the joint income distribution. To elucidate this, consider the simple
example provided by Grimm (2007) wherein the population is divided equally into two income
groups – the poor and the non-poor. Suppose that between two time periods, all initially poor
persons observed fast income growth such that by the end of the observation period, their
incomes are equal to the initial income levels of the initially non-poor. At the same time, the
income of the initially non-poor contracted to levels that are even lower than the initial income
of the initially poor. Following the conventional pro-poor growth assessment tools which
satisfy the temporal anonymity principle, one would conclude that the growth pattern has been
anti-poor both in absolute and relative terms. In particular, since poverty gap increased, one
could conclude that growth has not been pro-poor in absolute terms. Furthermore, since the
bottom half of the population observed negative income growth while the average income of
the other half remained constant, then growth will be judged as anti-poor in relative terms.
However, if one evaluates the observed growth pattern on the basis of individuals’ income
group membership at the initial time period, this leads to the assumption that the growth process
has been pro-poor.
Figure 1.2 Sample Growth Incidence Curve
Source: Author’s computations using simulated data.

Note: In this figure, the GIC intersects the x-axis at approximately Pi = 40. If the poverty line is set at or lower
than Y1(40), then growth is said to be absolutely pro-poor. Otherwise, it is not absolutely pro-poor. On the
other hand, growth is said to be relatively pro-poor if the GIC of the poor uniformly lies above the green line.
45
The growth incidence curve (GIC) proposed by Ravallion & Chen (RC) (2003) is one
of the most commonly used analytical tool for examining pro-poorness of growth. For
notation, Yt(pi) is used to denote the income of the individual whose income rank at time t is pi
and z denotes the poverty line. The GIC plots the income growth (y-axis) at each percentile of
the income distribution (x-axis) within the time period under consideration. Growth is
considered to be “absolutely pro-poor” when the GIC lies above the x-axis up to the maximum
value of p wherein Y1(pi) < z17 while it is “not absolutely pro-poor” if the curve lies below the
x-axis for the same range of p. However, when the curve switches sign before that maximum
value of p wherein Y1(pi) < z, then GIC fails to provide an unambiguous pro-poor growth
assessment. On the other hand, under the relative definition, growth is considered pro-poor
when the GIC lies above the horizontal line corresponding to g or the growth in mean income
up to maximum value of p wherein Y1(pi) < z. Again, GIC fails to provide a conclusive
assessment when the curve switches signs before that maximum value of p wherein Y1(pi) < z.
In general, unless the economic growth is uniformly distributed within the income distribution
(i.e. income changes by the same amount at each percentile), a pro-poor growth pattern would
exhibit a downward sloping GIC. Figure 1.2 illustrates these points.
When the GIC fails to yield an unambiguous pro-poor growth assessment, RC (2003)
proposed to use the concept of rate of pro-poor growth (RPPG). The authors defined RPPG as
the mean growth rate of the poor. Technically, RPPG is the area under the GIC up to the
maximum value of p wherein Y(pi) < z. To assess the pro-poorness of growth, the
corresponding evaluation function can be expressed as the difference between the observed
growth in mean incomes g and (21).18 Thus, the evaluation function can be expressed as
𝐹𝐺𝑇(0)
∑ =1 𝑔( 𝑖 )
𝑊𝑅𝑃𝑃𝐺 (𝑌 , 𝑌2 , 𝑔, ) = 𝑔 𝑖
(1.21)
𝐹𝐺 (0)
where FGT(0) is the headcount poverty index. Following the absolute definition, growth is
said to be absolutely pro-poor when RPPG or the second term in WRPPG is positive. On the
other hand, growth is considered to be relatively pro-poor when RPPG is greater than g or when
WRPPG is negative.
17 z denotes the poverty line. If the GIC lies above the x-axis for all p, then there is first-order poverty dominance (Atkinson
1987; Foster and Shorrocks 1988).
18 It can be shown that the RPPG is equal to the ratio of the change in Watts poverty index to the headcount poverty rate. The
Watts poverty index is computed as ∑𝑁 ( 𝑌𝑖 ) {𝑌𝑖 < }. As will be seen later, the Watts index is typically used to
𝑁 𝑖
estimate the average exit time from poverty. It is also one of the few poverty measures that satisfy the focus, monotonicity,
transfer and decomposability properties (Foster et al. 2013).
46
Noticeably, the GIC approach proposed by RC (2003) is only based on the changes in
the marginal distribution of income over time and does not take into account individual income
movements. Grimm (2007) adjusted the GIC approach when the assumption of temporal
anonymity is removed and the joint distribution of income is used to account for a person’s
income mobility. In particular, Grimm (2007) coined the term “individual growth incidence
curve” (IGIC) and defined the individual rate of pro-poor growth index (IRPPG) as the area
under the IGIC up to the maximum value of p(Y1) wherein Y(p(Y1)i) < z. Implicitly, this
includes income growth rates of non-poor individuals at the final time period as long as they
were classified as poor at the initial time period. The pro-poor growth evaluation function can
be expressed as
𝐹𝐺𝑇 (0)
∑ (𝑌 1) =1 𝑔( (𝑌1 )𝑖 )
1 𝑖
𝑊𝐼𝑅𝑃𝑃𝐺 (𝑌 , 𝑌2 , 𝑔, ) = 𝑔 (1.22)
𝐹𝐺 1 (0)
where
g(p(Y1)i) = (𝑌(𝑝(𝑌 )𝑖 )2 𝑌(𝑝(𝑌 )𝑖 ) )/𝑌(𝑝(𝑌 )𝑖 ) (1.23)
Under Grimm’s (2007) approach, growth is said to be absolutely pro-poor when either the
IRPPG or the second term in WIRPPG is positive. On the other hand, growth is considered to be
relatively pro-poor when IRPPG is greater than g or when WIRPPG is negative. The only
difference of (1.22) from that of RC (2003) is that income growth rates are computed with
respect to the income quantile in which a person belonged to during the initial time period.
1.4.3 Measuring Dynamics of Income Inequality

As pointed out in Section 1.3, one of the income mobility perspectives is based on the
extent to which income dynamics contributes to a more equitable distribution of long-run
income. This perspective is appealing because it directly links income mobility with inequality.
For socio-economic researchers, it is important to examine whether an increase in income
mobility can contribute to transitory variations in income so that permanent income inequality
would be less than observed income inequality (Jarvis & Jenkins 1998). In other words, high
income inequalities in fluid societies might be less problematic because the distribution of
lifetime income would be generally even through income mobility (Krugman 1992). The
indices proposed by Fields (2010), CDW (1985) and Shorrocks (1978) follow this approach.
Jenkins & Van Kerm (2006) decomposed the changes in cross-sectional income
inequality, measured using the generalized Gini coefficient, as a function of two components:
a measure of pro-poorness of growth and changes in individual incomes (redistribution or re-
ranking), as shown in (24).
47
∆𝐺(𝑠) = 𝑅𝑒𝑑𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜 (𝑣) 𝑟𝑜𝑝𝑜𝑜𝑟(𝑣) (1.24)
where
𝑌
𝑅𝑒𝑑𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜 (𝑣) = ∬[ (Ф ; 𝑣) (Ф2 ; 𝑣)] ( ̅2 ) 𝑓(𝑌 ; 𝑌2 )𝑑𝑌 𝑑𝑌2 (1.25)
𝑌2
𝑌 𝑌1
𝑟𝑜𝑝𝑜𝑜𝑟(𝑣) = ∬ (Ф ; 𝑣) ( ̅2 𝑌̅1
) 𝑓(𝑌 ; 𝑌2 )𝑑𝑌 𝑑𝑌2 (1.26)
𝑌 2
and w() is the weight of an individual which is a decreasing function of the individual’s rank
in the income distribution as identified by the parameter v, f(Y1, Y2) is the joint probability
density distribution of Y1 and Y2, Фt is the cumulative density distribution of Yt and 𝑌̅𝑡 is the
mean of the income distribution {Yit}.
As the definition of pro-poor growth implies, the first component measures how much
income growth benefits those on lower incomes relative to those on higher incomes while the
second component measures how much re-ranking in income positions is associated with the
income growth (Jenkins & Van Kerm 2009). Decomposing the temporal changes in inequality
into pro-poor and redistribution components is particularly useful for comparing groups in
which inequality is moving at the same pace (Jenkins & Van Kerm 2006).
I use Jenkins & Van Kerm’s (2006) approach of decomposing changes in inequality in
Chapter 4 to briefly examine whether there is equalizing mobility in the Philippines.
1.4.4 Identifying Correlates of Income Mobility

When identifying the factors that drive income mobility in the succeeding chapters, the
proximate determinants of mobility will be broadly grouped in this thesis into (i) (geographic)
location, (ii) household composition, (iii) education, (iv) employment, (v) access to (basic)
services and (vi) physical assets.19 Several studies have highlighted that specific locational
endowments may prompt concentration of skills which in turn expands better economic
opportunities while locational disadvantage such as remoteness tends to increase the cost of
carrying out economic activities which in turn has an adverse impact on income mobility
prospects (Lobao, Hooks & Tickamyer 2007; Aslam & Corrado 2012). Urban-rural disparities
in various income and non-income measures of well-being have also been well-documented in
19 There are other factors that can influence the household income distribution based on the existing literature. For example,
health is directly correlated with productivity which in turn, is directly correlated with economic well-being (Baker 2004). In
addition, social networks can also be used to access essential resources such as education, healthcare and other utilities more
easily (Acock & Hurlbert 2011, Jain & Sonnen 2011). However, this study does not include these types of variables as they
are not available in the data used here.
48
many empirical studies (WB 2013). For instance Fields et al. (2003) find that poverty is
predominantly a rural phenomenon in Indonesia, South Africa, Spain and Venezuela. In the
Philippines, socio-economic development landscape has a very distinctive spatial feature
wherein people living in the National Capital Region (NCR) and its neighbouring provinces
have significantly lower poverty rates compared to those living in central and southern
Philippines (Balisacan 1994, 1997; Balisacan & Hill 2003; Barrios & Landagan 2004; Schelzig
2005). In general, geography can act as either a gateway to better living standards especially
when a specific location is endowed with rich natural resources or to economic challenges
when a location is too remote and has very limited access to various social services. In addition
to geography, one of the recurring findings in the development literature is related to human
production theory. In particular, a number of studies provide empirical support that increases
in household size tend to be correlated with downward income mobility because the additional
resources needed to sustain a larger household is not usually compensated by higher income
flows (Fields et al. 2003). Furthermore, consistent with human capital theory, higher education
often leads to higher productivity and therefore, upward economic mobility prospects (Morgan,
Grusky & Fields 2006; Hout & DiPrete 2006; Greenstone et al. 2013). There are a number of
case studies that have shown that higher educational attainment (of the household head) reduces
the household’s vulnerability to poverty (Azam & Imai 2009). The Philippines is one of the
countries which have a high regard for education, and this perspective is deeply rooted in its
culture. For many poor Filipino households, education is considered one of the most important
legacies that parents can impart to their children to be able to move away from socio-economic
deprivation (Maligalig et al. 2014). Furthermore, several studies find a strong link between
poverty cycles and employment transitions. For instance, Dartanto & Nurkholis (2013) find
that a transition from formal to informal employment can push a non-poor household into
poverty in Indonesia. In the Philippines, ADB (2010) find that middle class households with
more members working as own-account and casual workers have higher risk of falling into
poverty. Analogously, access to (basic) services and assets are also found to be significant
correlates of well-being (WB 2004). For instance, access to high-quality healthcare services
helps workers avoid employment interruptions due to sickness which in turn, allows them to
continue translating their labour into financial capital (Schelzig 2005). Similarly, many forms
of physical assets (e.g., land) and technological innovations are also useful tools for extracting
more wealth (Carter 2000; Schelzig 2005; Moser 2006). In particular, ownership of land
minimizes the risk of long poverty spells (Adam and Jane 1995) while household losing land
have higher risk of experiencing downward mobility (Justino & Verwimp 2013).
49
The explanatory variables that will be included in the statistical models in Chapters 5, 6
and 7 fall under the categories described in this section.
1.5 Analytical Considerations When Examining Income Mobility

This section highlights the importance of being cautious when interpreting mobility
estimates at face value due to other confounding factors like measurement errors and transitory
income fluctuations (Lillard and Willis 1978; Solon 2001; Fields et al. 2003; Antman &
McKenzie 2007; Krebs, Krishna & Maloney 2012). For example, an income mobility pattern
that portrays low-income individuals to be catching up with high-income individuals may be
artificially driven by the downward bias in the reported incomes of individuals at the top of the
income hierarchy. On the other hand, it is also possible that a significant portion of the observed
income mobility is driven by transitory income shocks. Depending on the objective of the
study, it may be important to carefully examine the impact of these factors on the income
mobility estimates.
1.5.1 Correcting for Measurement Error and Data Contamination

The discussion in the previous sections assumes that income is measured accurately.
However, previous studies suggest that income, particularly those derived from household
surveys, are prone to measurement errors (Duncan & Hill 1985; Bound & Kruger 1991;
Gottschalk and Huynh 2010; Glewwe 2012). If measurement errors are non-negligible, the
results of income distributional analysis may produce biased results and thus, lead to
misleading conclusions and policy implications. This section briefly examines the issue of
measurement errors in the context of income mobility estimation.
What are the common sources of measurement errors? The most basic forms of data
contamination may arise from randomly misreporting income or data encoding mistakes.
Conventionally, these errors are assumed to average out (i.e., zero-mean) and are uncorrelated
with the true income. This type of error is commonly referred to as classical measurement
errors. On the other hand, non-classical measurement errors either have non-zero means or are
correlated with the true income. Either way, both types of measurement errors contribute to
additional noise in observed incomes and findings from previous studies suggest that they can
lead to biased estimates of cross-sectional poverty and inequality (van Praag, Hagenaars & van
Eck 1983; Chesher & Schluter 2002; Jenkins 2011).20 However, there are only few studies that
20 Ravallion (1994) and Chesher & Schluter (2002) proposed adjustment procedures to correct the bias induced by
measurement errors on cross-sectional estimates of poverty and inequality.
50
have investigated the extent of measurement error bias on income mobility estimates. When
income is measured with error, the true historical income profile is unobserved. Thus, the
estimated mobility of observed income will reflect the changes in the joint distribution of the
true income and measurement errors.
Measurement errors tend to make income less correlated over time. Hence, when income
mobility is viewed in terms of origin-independence perspective, measurement errors can lead
to over-estimation of mobility. This is supported by the findings of Glewwe (2012) who
estimated that about 15% to 42% of observed mobility in Viet Nam in the 1990s can be
considered as upward bias due to measurement errors present in the survey data used. Krebs et
al. (2013) also noted a non-negligible upward bias on income mobility estimates in Mexico due
to measurement errors. Overall, although there are limited studies which examine the impact
of measurement errors on income mobility estimated in terms of other mobility perspectives,
it is safe to assume that the effect are not necessarily trivial (Boheim & Jenkins 2006).
There are several ways to minimize the bias induced by measurement errors. For
instance, when the main income data source is a household survey, the survey records may be
matched with reports by the same respondents from administrative data that are assumed to be
error-free. In turn, this supplementary data can be used to derive appropriate income adjustment
factors. In particular, some studies use tax data records to correct for the potential bias present
in the data on observed income.21 On the other hand, some researchers minimize measurement
error bias by restricting the sample to population groups that are less likely to misreport income.
For example, Gottschalk & Huynh (2010) proposed excluding those who are self-employed or
those whom a large portion of income is imputed from the analysis. Some researchers trim the
data by removing the bottom and top 1% while others take out all outlying income values
before estimating mobility. Furthermore, others rely on more sophisticated statistical modeling
techniques. One example is the use of a latent class of Markov models to gauge the impact of
measurement errors on income transition matrices (Breen & Moisio 2004; Worts, McDonough
& Sacker 2010). Given sufficient length of longitudinal data, the main idea behind this
approach is to assume that the true transition probabilities are stable over time and this can be
estimated from the Markov model. The resulting residuals are then treated as measurement
errors. The use of instrumental variables is an alternative tool that can be used to address the
bias caused by measurement error and it is particularly useful when the concept of origin-
21
However, this approach may not be feasible for many developing countries where such type of administrative
tax data is usually inaccessible if not unavailable.
51
independence is being used. The idea behind instrumentation is to use a proxy variable that is
highly correlated with the outcome of interest but is uncorrelated with the measurement error.
Arellano & Bond (1991) recommended the use of the income lagged two periods while Holtz-
Eakin, Newey & Rosen (1988) proposed using income lagged three or more periods as
instruments. In a more general setting and/or in the absence of reliable instruments, one can
conjecture different forms of measurement error and create synthetic data of measurement
errors by drawing from its assumed distribution. In turn, this can be incorporated to the
observed income and estimate different income mobility indicators. This approach is useful for
identifying bounds for income mobility estimates.22
In this study, I approximate the impact of measurement error using a simple simulation
exercise and find that the income mobility estimates presented here may be higher than the
actual magnitude of income mobility. Nevertheless, the bias could be minimal depending on
the form of the measurement error.23
1.5.2 Decomposing Income into Permanent and Transitory Components

As suggested earlier, a person’s observed income can be decomposed into permanent and
transitory components (Friedman 1957). Permanent income refers to a person’s income over a
long horizon while transitory income refers to income received from unanticipated sources.
Although transitory income can be either positive or negative, it is expected to average out in
the long-run.
In general, the relevance of decomposing income mobility due to the dynamics in
permanent and transitory income, for policy planning is manifold. First, understanding whether
the observed mobility is a result of movements in either permanent or transitory income is
important for outlining poverty reduction programs. Without distinguishing poverty
persistence from transient poverty, policy planners may not be able to properly target intended
program recipients which could lead to the transiently poor receiving disproportionately more
assistance than the persistently poor or in some cases, the persistently poor people receive
assistance that is only enough to protect them from temporary economic risks (Jenkins 2011).
In other words, programs with unclear targeting mechanisms are at risk of being poorly-
22 The same approach was adopted by Khor & Pencavel (2008) when estimating income mobility in China using Chinese
Household Income Project (CHIP) data. In particular, the authors find that when the mean of the simulated error is
approximately 10% of the mean of measured income, the average quintile move in urban China increases by approximately
4% while immobility ratio increases by 5%.
23 Assuming that there is classical measurement error which inflates the variance of the observed (log) income by 5%, the
results of a simple simulation experiment that I carried out suggest that income mobility in the Philippines can be overestimated
by 15%. On the other hand, a positive auto-correlation between the measurement errors can offset the bias-increasing effect
of measurement errors on mobility estimates. However, it is difficult to infer the form of measurement errors.
52
implemented and prone to undercoverage and leakage (Dutrey 2007; Ravallion 2009). In
general, poverty reduction programs are cost-inefficient when it is not clear what type of
poverty is being addressed and who the intended recipients are. Nevertheless, while it makes
sense to spend more effort to alleviate the living conditions of the persistently poor, it is also
useful to institutionalize risk-coping mechanisms that will enhance the socio-economic security
of the transiently poor (Deaton 1991). Furthermore, it is also important to examine the
dynamics in the transitory component of income as the cumulative effects of temporary income
fluctuations among the poor could eventually lead to poverty persistence, especially when there
are irreversible asset losses (Baulch & Hoddinott 2000; Hoddinott 2006). More generally,
disentangling the contribution of permanent and transitory components on income mobility
allows us to understand the incentive and security aspects of income mobility. For instance, the
prospect of upward or downward mobility in the long-run, provides incentives for individuals
to be engaged in productive economic activities. This is referred to as the incentive aspect of
income mobility and it is primarily concerned with the dynamics of permanent incomes
(Jenkins 2011). On the other hand, the security aspect of income mobility is contextualized
within the assumption in economic theory that individuals are risk averse (Kaufmann 1970 as
cited in Sinn 1981). In other words, for an average person, the ability to predict future income
is important when planning consumption behaviour. Thus, people become more concerned on
the arrangement of expenditures when income streams are fluctuating due to mobility of
transitory income (Fachinger & Himmelreicher 2012). Second, decomposing income mobility
into its permanent and transitory components also allows us to distinguish income mobility as
a desirable outcome from income mobility as an indicator of instability (Friedman & Kuznets
1954).24 In particular, income mobility is desirable when the growth in a person’s permanent
income is negatively correlated with his initial level of income as such type of growth pattern
allows the poor to catch-up with the rich (Benabou & Ok 2011). In this context, income
mobility makes the distribution of opportunities more equitable in the long-run (Atkinson et al.
1992). On the other hand, income mobility may be perceived as an indicator of socio-economic
insecurity when it is mostly driven by fluctuations in short-term income (Jarvis & Jenkins 1998;
Creedy & Wilhelm 2002; Allanson 2008; Rohdes, Tang & Rao 2013).25
24 Analogously, cross-sectional estimates of inequality may also be decomposed into inequality due to disparities in permanent
income and inequality due to varying income fluctuations. Up to some extent, income inequalities arising from disparities in
permanent income may be associated with inequality of opportunities.
25 Jenkins (2011) argue that fluctuation in transitory income is not a perfect indicator of economic instability. This is because
income fluctuations may arise from voluntary choices made by an individual. For example, if an individual voluntarily decides
to work shorter hours, the resulting income fluctuations will not necessarily imply insecurity.
53
There is a wide range of literature discussing the different methodologies for
decomposing income into its permanent and transitory components. One of the most commonly
used approach is to estimate variance-components models with varying complexity and
conditioned on various covariates (Dickens 2000; Geweke & Keane 2000; Moffitt &
Gottschalk 2002; Zandvakili 2002; Ramos 2003; Gustavsson 2007). While the econometric
literature offer several estimation methods to fit the variance component models using
longitudinal data, there are several issues regarding the use of parametric models. One of the
potential issues of this method is that it requires panel data of adequate length to be able to
estimate the model parameters consistently.26 However, while panel data is regularly collected
in many industrialized countries, it is not collected frequently in developing countries. On the
average, developing countries with nationwide longitudinal data on income have three to four
waves only. In such cases, a simpler alternative approach is warranted. A candidate measure
of permanent income is the individual’s or household’s income averaged over a specific time
period after adjusting for inflation. This approach averages out the measurement error over
time. This is also the approach that I use to derive a proxy measure of permanent income in the
succeeding chapters.
1.6 Summary
This chapter presented the building blocks for examining income mobility, many of
which are heavily used in the succeeding chapters. First, it reviewed the different definitions
of income mobility. The existing literature offers three broad perspectives of what mobility
means: as movements, as origin independence and as equalizer of long-term incomes.
Additionally, there are different indicators for measuring the various concepts of income
mobility. These measures do not necessarily produce quantitatively and qualitatively similar
results. In many cases, a satisfactory measure of one particular concept may be a poor measure
of another. Thus, when the objective is to provide a more holistic picture of a country’s
underlying income mobility process, it is important to examine it under different perspectives
due to its multidimensional nature. However, calculating too many income mobility measures
may also result in a confusing array of numbers. Instead of providing a more comprehensive
view of the income mobility process, this may just obscure the big picture. To strike a balance
between these two considerations, a possible approach is to focus on a limited number of
26 Even if long panel data is available, there are other issues in using variance components model. For instance, Shin & Solon
(2009) argued that parametric models are in some sense, “arbitrary mechanical constructs” such that the estimates can be
sensitive to variations in how the underlying model is specified. In addition, choosing which parametric model specification
is the most appropriate in different contexts is not an easy task.
54
indicators capturing different aspects of income mobility. Thus, in examining the mobility
patterns throughout the study, I choose an optimal mix of indicators to cover as many
dimensions of income mobility as possible within the context of the research question under
consideration.
As previous studies have shown that income data are prone to measurement errors, the
chapter briefly discussed the impact of measurement errors on income mobility estimates.
There are several ways to minimize the bias due to measurement errors. One is to use external
data to measure the degree of measurement bias and derive appropriate adjustment factors;
another is to rely on finding suitable instruments to estimate income mobility parameters
consistently. In the absence of auxiliary information, a general approach that can be adopted is
to simulate measurement errors using distributional assumptions. Such simulation studies can
help constructing bounds for the proportion of the observed income mobility that can be
attributed to measurement errors.
Third, the chapter also emphasized that the normative assumption of more income
mobility being always a desirable outcome should be examined with caution. An income
mobility regime that is mainly driven by fluctuations in the transitory component of income
may represent socio-economic insecurity. For policy planning, it is important to determine
whether the observed mobility is a result of changes in permanent or transitory income. In this
context, the chapter reviewed several econometric methods to decompose observed income
into its permanent and transitory components. Much of the proposed procedures rely on using
parametric models. If the available panel data is not of sufficient length to allow (consistent)
estimation of the parameters of these models, one can adopt simpler techniques such as
longitudinal-averaging of individual incomes to approximate the permanent component.
In summary, the main point that the discussions presented in this chapter seeks to convey
is that while the analysis of income mobility provides a broader picture of the income
distribution dynamics than conventional static socio-economic indicators, there are also
additional technical considerations that have to be carefully taken into account. In general, the
accompanying methodological decisions may lead to variations in research findings which in
turn could cause confusion among non-technical audience. Policy responses to findings of
income distributional analysis may lead to sub-optimal intervention programs when the
estimates are not well explained to policymakers and other key stakeholders. In contrast, when
the robustness of the results to measurement parameters are carefully examined, income
mobility research can empower users and stakeholders by providing them with a better
understanding of the impact of methodological decisions on research findings.
55
Chapter 2 Has Economic Growth been Pro-Poor in the Philippines?
2.1 Introduction
The notion of an Asian Century has attracted the interest of the global community
towards the region. As billions of Asians are expected to enjoy living standards similar to those
in Europe today, few decades from now, the Asian Century is a term used to refer to the
forecasted domination of the region in terms of the world’s socio-economic and political
landscape (ADB 2011a). This is not surprising considering that, over the past five decades,
Asia has witnessed rapid economic growth, significant poverty reduction and improved living
standards. If these trends continue, economists forecast that the region will account for half of
the global output, trade and investment by 2050 (ADB 2007). Signs of the advent of the Asian
Century are quite apparent today. With the sluggish economy currently experienced by Western
countries, the engine of growth of the global economy is now being driven by China and India,
Asia’s powerhouse economies (Eichengreen, Gupta & Kumar 2010; Santos-Paulino & Wan
2010). Nevertheless, other developing Asian countries like the Philippines are also showing
signs of growth momentum. Since 2009, the country’s GDP per capita has been increasing at
annual rate of 4.1%. In the first quarter of 2013, the GDP grew by 7.8% surpassing China’s
7.7% (CEIC 2014).27 Due to this solid performance, the country has been dubbed one of the
emerging Asian Tigers (Coclanis 2013). However, a cursory review of the country’s historical
growth performance reveals that its economic development path has been characterized by
several boom and bust cycles in the past (Aldaba 2009; Schelzig 2005). For many years, the
Philippines had been labelled as the Sick Man of Asia due to its lagging economic performance
and low growth elasticity of poverty when compared to its Asian neighbours (Kind 2000)
(Table 2.1). Thus, the rapid economic growth regime currently experienced by the country is a
critical juncture. If complemented by an appropriate mix of socio-economic policies, economic
experts believe that this episode could serve as a window of opportunity for the country to
accelerate its development and be freed from shackles of poverty and economic stagnation
(WB 2013). Otherwise, when institutions continue to operate in the favour of a privileged few,
the current strong growth episode could end up as just another part of its perennial boom-bust
cycle (Bird & Hill 2009). The first step to be able to outline growth-sustaining policies is to
understand the underlying factors that drive the economic prospects of the Philippines.
Although there are several metrics that can be used to evaluate the economic prospects of the
27This also surpassed the growth rates of other key Asian countries such as Indonesia (6%), Thailand (5.3%) and Viet Nam
(4.9%) (WDI 2014).
56
country, this study reviews the evolution of the household income distribution in the
Philippines. Since an economic growth that reduces social exclusion and minimizes the gap
between the poor and the rich contributes to a more sustainable development (Aldaba 2009;
Canlas, Khan & Zhuang 2009), analysing the historical development path of the household
income distribution allows us to evaluate the sustainability of the Philippines economic growth.
The main objective of this chapter is to review the trends in economic growth, poverty
and inequality in the Philippines over the past thirty years leading to the current rapid economic
regime experienced by the country. This is done by drawing from findings in the previous
literature and by using data from WB’s PovcalNet. PovcalNet is an online computational tool
that contains grouped income distribution data that allows users to reproduce comparable
estimates of average income, poverty and inequality that WB researchers use (WB 2012). The
last part of the chapter briefly demonstrates how our perception of the evolution of the income
distribution may change when income mobility is taken into account. To do this, I simulate
synthetic unit-level income from PovcalNet’s grouped distribution data for the Philippines for
the years 1985 to 2009. Then, I create pseudo-panels by assuming different income reranking
scenarios. The discussion of the results of the simulation experiment sets the tone for the
detailed discussion of income mobility in the succeeding chapters. In particular, this chapter is
outlined to answer the following questions:
i. How did the household income distribution in the Philippines evolve over the past
thirty years?
ii. Have the poor Filipinos benefited from economic growth more than the non-poor?
Table 2.1 Socio-Economic Indicators for Selected Southeast Asian Countries

US$2/day Inequality
GDP per capita
Poverty Rate (Gini, %)
country Period
initial final initial final initial final
year year %growth year year year year
Indonesia 1984-2011 646.10 1650.52 3.47 88.40 43.33 30.50 38.14
Lao PDR 1992-2008 271.73 561.52 4.54 84.80 66.00 30.40 36.70
Malaysia 1984-2009 2713.11 5984.92 3.16 12.30 2.27 48.60 46.20
Philippines 1985-2009 907.09 1325.90 1.58 61.90 41.50 41.00 43.00
Thailand 1981-2010 915.45 3163.90 4.28 44.10 4.05 45.20 39.40
Viet Nam 1993-2008 317.12 775.76 5.96 85.70 43.40 35.70 35.60
Source: WDI (2014)
57
2.2 Brief History of Economic Growth in the Philippines Over the Past Three Decades
The Republic of the Philippines is an archipelago consisting of 7,107 islands located in

the Pacific Ring of Fire within the Southeast Asian region.28 The country is divided into three
main geographical divisions: Luzon in the north, Visayas in the center and Mindanao in the
south. Within each division, the country is further divided into 17 regions and 81 administrative
provinces. In terms of population structure, the Philippines is the seventh most populated
country in Asia with a population size of about 98.4 million (in 2013) and an average annual
population growth of 1.7% (WDI 2014). The country has a relatively young population with a
median age of 23.4 years; where 33.4% of the country’s population are under 15 years of age
and 6.8% are aged 60 years or over (NSO 2013). The average life expectancy (at birth) in the
Philippines is approximately 65.2 years for men and 72.1 years for women (UNESCAP 2013).
The Philippines is a rapidly urbanizing country. For instance, compared to 1980s where only
62.5% of the population lived in rural areas, current estimates suggest that about 49.1% of the
population are now living in urban areas (UNESCAP 2013). In terms of employment structure,
the Philippines’s labour force mainly relies on services sector as its main source of employment
where about 58.9% of the employed population work in services, 32.1% in agriculture while
the rest are employed in industry (ADB 2013a).
In terms of the recent economic history, the 1980s have been regarded as the “lost
decade” in the Philippines. Much of the unsatisfactory growth performance happened in the
early part of the 1980s as the weakening demand for exports was compounded by difficulties
in tapping funds from the international market and series of domestic political struggles. These
events partially triggered the government during that period to declare a moratorium on its
foreign debt servicing (ADB 2007). The economy revived its optimism as a new democratic
government was established in 1986. In particular, from 1985 to 1988, the GDP per capita and
household expenditure per capita, were growing at approximately the same pace, about 2 to
2.5% per year. However, this recovery was short-lived as political power struggle and natural
calamities plagued the country. In 1990, a major earthquake hit the northern and central parts
of the country while a volcano erupted in 1991. The severity of these natural disasters impeded
growth. From 1988 to 1991, the GDP per capita grew by only 0.3% per year. Despite this,
survey-based estimates suggest that household expenditure per capita continued to grow at a
28 The Pacific Ring of Fire is an area within the Pacific ocean where there is a number of plate tectonic activities (e.g.,
earthquakes and volcanic eruptions). It is estimated that about nine in ten earthquakes occur in the Pacific Ring of Fire (Park
2007).
58
modest pace of 2.5% per year during the same period. There are two competing reasons for
this apparent inconsistency. First, it is possible that the household sector is growing much more
rapidly than the growth of enterprise or government sectors (WB 2001). On the other hand, it
is also possible that there are measurement biases in both the national accounts and household
surveys.29 Nevertheless, the slow poverty reduction from 1988 to 1991 seems to be more
congruent to the slow GDP growth experienced during this period.
The Philippines experienced a severe electric power crisis from early 1990s until 1994.
During this period, the GDP per capita dropped by 0.08% per year while household expenditure
increased by only 1.04% per year. The country’s economy rebounded only after the electric
power crisis was addressed in 1994. From this year until 1997, GDP per capita rose by 2.9%
per year while household expenditure per capita grew more rapidly at an annual rate of 7.6%.
Overall, the gains that transpired from 1985 to 1997 have been largely concentrated on
the first and last three years (WB 2001). Growth was impeded again when the Asian financial
crisis struck in 1997. This was further aggravated by the severe drought that struck the country
in 1998 which affected a significant number of poor agriculture-based households. Between
1997 and 2000, estimates of GDP and average household expenditure barely changed. In the
first part of 2000, income from semi-conductors, one of the country’s main exports, dropped
significantly when the global economy weakened due partly to the dot-com bubble burst and
speculations that the change of millennium will have a severe effect on technological products
(Aldaba 2009). The country also experienced another political crisis when the president was
ousted from office due to corruption-related allegations (Canlas et al. 2009). Nevertheless, the
country’s GDP per capita still expanded by 1.7% annually from 2000 to 2003 (WDI 2014).
From 2003 to 2006, growth in the GDP per capita is relatively high based on historical standard.
However, it slowed down until 2009 due to the global price hikes in oil and food in 2007 and
financial crisis in 2008 (Canlas et al. 2009). On the average, GDP per capita improved by 2.2%
while household expenditure per capita increased by 1.5% per year from 2006 to 2009 (WDI
29The WB (2010) report offers several reasons to explain this divergence. First, growth in GDP may be driven by flow of
investments, the benefits of which usually accrue to the upper tail of the income distribution (WB 2010). Since very affluent
households are not adequately represented in household surveys and are prone to report lower income and consumption, it is
possible that the mean income and expenditure derived from household surveys are underestimated (Deaton & Dupriez 2011).
However, some may argue that the poorest of the poor are also likely to be under-represented in household surveys. When
housing units are used as ultimate sampling units, those who usually live in makeshift housing and in remote areas may not
have chance of being in the sample. Consequently, this will contribute to the overestimation of mean income. Another possible
reason is that there are consumption items included in national accounts but not in household surveys (WB 2010). The list
includes: imputed rents to homeowners, indirectly imputed financial services, and consumption by non-profit institutions
serving households (Deaton 2005). The exclusion of these items in household surveys may contribute to the divergence of
national accounts and survey-based estimates.
59
2014). Since 2009, the Philippines has posted remarkable economic gains. In 2012, the
country’s economy grew by 6.6% while estimates for the first quarter of 2013 place economic
growth at around 7.8% (WDI 2014). With these recent developments, economic analysts are
revising their growth forecast upwards to reflect the growth momentum transpiring in the
Philippines (WB 2014). In addition, for the first half of 2013, various international credit rating
agencies have also upgraded the country’s investment grade (ADB 2013b).
2.3 Poverty, Inequality and Pro-poor Growth Patterns in the Philippines

2.3.1 Cross-Sectional Perspective
The Philippines has experienced stagnant economic growth until the late 1990s but since
then, the country’s economy has grown faster. This section reviews how economic growth has
affected the household income distribution by examining poverty, inequality and pro-poor
growth patterns in the Philippines. There are several studies that have done similar analysis
(Balisacan & Fujisaki 1998; Balisacan & Pernia 2002; Schelzig 2005; Aldaba 2009). However,
these studies used varying methodologies and looked at different time periods. Thus, for
purposes of comparability across years when reviewing the country’s poverty, inequality and
pro-poor growth trends, I use the Povcalnet’s database in the succeeding discussion.
Comparisons are drawn from cross-sectional estimates for the period 1985 to 2009.
Figure 2.1 Comparison of Estimates of Growth from

Survey and National Accounts
Source: WDI (2014) and Povcalnet (2014)
60
As pointed out earlier, survey-based estimates and those that are derived from national
accounts do not necessarily produce qualitatively similar results. Figure 2.1 illustrates this
point. From 1985 to 1997, the Philippines’ GDP per capita and average household expenditure
per capita generally moved in the same direction. Thereafter, GDP increased while average
household expenditure went down. Some researchers argue that the GDP growth in the
Philippines might have been overestimated in the recent years (Medalla and Jandoc 2008) while
others contend that the estimates from national accounts compiled by government are
statistically sound (Virola 2010).
Table 2.2 summarizes the trends in economic growth, poverty and inequality in the
Philippines. Over the past three decades, the country’s household expenditure per capita
increased by 150% or approximately 1.7% annually between the periods 1985 and 2009. Prior
to the peak of the 1997 Asian financial crisis, household expenditure experienced steady
expansion. In particular, the household expenditure per capita grew by an average of 3.5% per
year from 1985 to 1997 but it barely moved after the 1997 crisis. Despite the growth during
this period, poverty reduction had been fairly slow. In particular, headcount poverty rate
dropped by only 0.87 percentage points annually between periods 1985 and 2009 (from 62%
in 1985 to 42% in 2009).
In the context of graduation from poverty, estimates of the Watts index predicts that if real
income of all Filipinos in 1985 increased by 2% per year, then the average exit time from the
US$2-poverty threshold was about 18 years.30 However, due to the stagnating poverty rates
after the Asian financial crisis, the average exit time from the US$2-poverty threshold dropped
only to 9 years almost three decades later.
Table 2.2 also provides estimates of conventional measures of inequality such as the Gini
coefficient and generalized entropy (GE) indices. The results suggest that regardless of the
index under investigation, the inequality in the Philippines is persistently high. At the same
time, there seems to be no significant change in the level of income inequality in the Philippines
between 1984 and 2009. This can be partly attributed to the rise in income inequality from
1984 to 1997 being compensated by the decline in inequality between 1997 and 2009. The
fastest increase in inequality transpired during the country’s highest income growth period,
during the years 1994 to 1997.
30 This is computed by dividing by the value of the Watts index by the expected income growth rate. In this case, the Watts
index value in 1985 is 36. Dividing this by an assumed annual income growth rate of 2% will yield 18 years. This means that
in 1985, poverty was expected to be eradicated by 2003. However, in 2009, the Watts index suggests that the average poverty
exit time is still 9 years.
61
Table 2.2 Distribution of Household Monthly Income Per Capita in the Philippines, 1985-2009
(in 2005 PPP US$)
Statistics 1985 1988 1991 1994 1997 2000 2003 2006 2009
GDP per capita (US$) 907.09 964.00 972.65 970.39 1,057.42 1,060.55 1,116.06 1,241.55 1,325.90
Mean 825.84 899.76 970.56 1001.4 1258.56 1237.92 1218.12 1187.88 1243.8
Lower Bound 809.4 882.6 950.04 981.24 1229.16 1209.36 1192.8 1163.76 1218.96
Upper Bound 842.28 916.92 991.08 1021.56 1287.96 1266.48 1243.44 1212 1268.64
Headcount poverty Index 0.62 0.57 0.56 0.53 0.44 0.45 0.44 0.46 0.42
Lower Bound 0.62 0.57 0.55 0.52 0.43 0.44 0.43 0.45 0.41
Upper Bound 0.63 0.58 0.57 0.54 0.45 0.46 0.45 0.46 0.42
Poverty Gap 0.25 0.22 0.22 0.2 0.16 0.16 0.16 0.16 0.14
Lower Bound 0.25 0.21 0.21 0.2 0.15 0.16 0.16 0.16 0.13
Upper Bound 0.25 0.22 0.22 0.21 0.16 0.17 0.16 0.17 0.14
Squared Poverty Gap 0.13 0.11 0.11 0.1 0.07 0.08 0.08 0.08 0.06
Lower Bound 0.12 0.1 0.11 0.1 0.07 0.07 0.07 0.07 0.06
Upper Bound 0.13 0.11 0.11 0.1 0.08 0.08 0.08 0.08 0.06
Watt's Index 0.36 0.31 0.31 0.29 0.22 0.23 0.22 0.23 0.18
Lower Bound 0.36 0.3 0.31 0.28 0.21 0.22 0.22 0.22 0.18
Upper Bound 0.37 0.32 0.32 0.29 0.22 0.23 0.23 0.23 0.19
Gini 0.41 0.41 0.44 0.43 0.46 0.46 0.45 0.44 0.43
Lower Bound 0.4 0.4 0.43 0.42 0.45 0.45 0.44 0.43 0.42
Upper Bound 0.42 0.41 0.45 0.44 0.47 0.47 0.45 0.45 0.44
GE(0) 0.28 0.27 0.32 0.3 0.36 0.35 0.33 0.32 0.3
Lower Bound 0.27 0.26 0.31 0.29 0.34 0.34 0.32 0.31 0.29
Upper Bound 0.29 0.28 0.33 0.32 0.37 0.37 0.34 0.33 0.31
GE(1) 0.32 0.3 0.36 0.34 0.41 0.4 0.36 0.35 0.34
Lower Bound 0.3 0.29 0.34 0.32 0.39 0.38 0.34 0.34 0.32
Upper Bound 0.33 0.32 0.38 0.35 0.43 0.42 0.38 0.37 0.35
GE(2) 0.51 0.47 0.58 0.53 0.71 0.69 0.56 0.54 0.52
Lower Bound 0.46 0.43 0.53 0.48 0.64 0.62 0.51 0.49 0.48
Upper Bound 0.56 0.52 0.64 0.57 0.78 0.76 0.61 0.59 0.56
Source: Author’s computations using simulated data from Povcalnet and WDI data.
Note: The Povcalnet data are originally expressed as monthly estimates. I multiplied them by 12 to approximate annual figures. The headcount poverty
index corresponds to the proportion of the population living below the poverty line. The poverty gap index refers to the average income shortfall of the poor
in proportion to the poverty line. The squared poverty gap is the squared income shortfall of poor in proportion to the poverty line. The Watts index is an
approximate measure of the average exit time out of poverty. All poverty indices are computed using the WB US$2/day poverty line. The Gini index is a
measure of inequality that ranges between 0 and 1 with higher values indicating greater inequality. The Generalized Entropy (GE) index measures the
redundancy or lack of randomness in the income distribution, with higher values corresponding to higher levels of inequality. The GE measures are more
sensitive to differences in the lower income brackets if the value of the parameter α is close to 0 and they are more sensitive to differences in the higher
income brackets if α is close to 1.The lower and upper bounds correspond to two standard errors below and above the point estimate of poverty and inequality.
For details, readers may refer to Foster et al. 2013.
62
Figure 2.2 shows the GICs from 1985 to 2009. The GIC plots the income growth at each
percentile. Growth is said to be absolutely pro-poor when it is above zero for all percentiles
and it is relatively pro-poor when the growth of the income of the poor is higher than the growth
in mean income (RC 2003). Following these definitions, I find mixed results about the pro-
poor growth process that transpired in the country. For instance, growth has been absolutely
pro-poor between 1985 and 2009, particularly during 1985-1988, 1991-1994, 1994-1997 and
2006-2009. However, there were periods when growth was not absolutely pro-poor. These
periods include the episodes of decreasing household income between 1997 and 2006, and even
during the modest growth periods in 1985-1988 and 1991-1994. On the other hand, it is clear
that during high growth periods in 1988-1991 and 1994-1997, the income of the rich increased
faster than the income of the poor suggesting that the respective growths during these periods
were not pro-poor. Nevertheless, there is a sign that the poor benefited more from the observed
growth than the non-poor during 2006-2009. Meanwhile for the other years, the GICs do not
provide a clear pro-poor growth assessment.
The finding that income inequality remains pervasively high and the pace of poverty
reduction slow leads socio-economists to conclude that an inclusive growth is yet to be a
sustainable feature of the Philippines’ economic development narrative even on the heels of a
rosy macroeconomic picture. Several studies concluded that policies have been ineffective in
redistributing the benefits of growth to the country’s poorest of the poor (Schelzig 2005;
Aldaba 2009). Poor Filipinos are trapped in precarious jobs due to the country’s highly
segmented labour markets (Usui 2011, 2012) and the poor’s limited capacity to invest on
education (Maligalig et al. 2014). Furthermore, many people who managed to get out of
poverty at some point are immediately pulled back to economic dearth during crises due to
their lack of access to risk management tools (Reyes & Tabuga 2012). Development experts
offer various explanation why the pattern of growing inequality at the backdrop of rapid
economic growth is also observed in many developing Asian countries. For instance, Zhuang,
Kanbur and Rhee (2014) argue that globalization and market-oriented reforms which
developing Asian countries are experiencing have inflationary impact on both growth and
inequality. Nevertheless, amidst these substantive potential explanation, several technical
issues remain. For instance, the data used for poverty, inequality and pro-poor growth
calculations are based on analyses of repeated cross-sectional data. As pointed out in Chapter
1, repeated cross-sectional data overlook one vital point that should be of concern when one is
examining the welfare of the poor. In particular, it does not give any information as to what
happened to particular individuals. Individual incomes change from year to year but repeated
cross-sectional data is unable to reveal whether particular individuals experienced income
63
mobility. One of the things that have recently captured the interest of development experts in
the Philippines is the extent to which our perception about poverty, inequality and pro-poorness
of growth in the country could change when we shift from a static to a more dynamic
perspective (Reyes et al. 2012; Bayudan-Dacuycuy & Lim 2013). The next section concludes
the review of the Philippines’ socio-economic history with a simple thought-experiment. Using
several simulated pseudo-panel data sets, I will re-examine household income distribution
trends with a more dynamic perspective.
2.3.2 Longitudinal Perspective

To account for individual income mobility when examining pro-poorness of growth, I
construct several sets of pseudo-panel data following the methodology described in Appendix
A2.1 because there is no actual panel data that cover the past three decades in the Philippines.
In a nutshell, the approach entails considering different income re-ranking scenarios and for
each scenario, computing dynamic measures of poverty and inequality as well as Grimm’s
(2007) individual rate of pro-poor growth as discussed in Chapter 1. Figure 2.3 provides a
graphical summary of the various re-ranking scenarios considered. The first scenario is an
extreme case corresponding to a static income ranking scenario wherein the income rank of
each individual in 1985 remains the same in 2009. Another extreme case is the complete
reversal of income ranks scenario wherein the individual with the lowest income in 1985 is the
same individual with the highest income in 2009, the individual with the second lowest income
in 1985 is the same individual with the second highest income in 2009, and so on. The other
scenarios considered are generated by using different values for the correlation of the income
ranks between the two time periods. Note that the neutral case corresponds to when an
individual’s income rank in 1985 is independent with his/her income rank in 2009.
64
Figure 2.2 Growth Incidence Curves in the Philippines, 1985-2009
Source: Author’s computations using simulated data from Povcalnet and DASP package of Stata.
Note: The green line represents the growth in mean income while the red line represents mean of growth rates.
65
Turning back to the estimates provided in Table 2.2, the US$2 headcount poverty rates
in the country decreased from 62% in 1985 to 42% in 2009. If the income ranks are perfectly
positively correlated, then the conventional perception about changes in cross-sectional
poverty will not differ. Looking at movements into and out of poverty, one can conclude that
41% stayed in poverty and 21% got out of poverty. On the other hand, at the extreme case that
income ranks are perfectly negatively correlated, about 5% of the population remained in
poverty, 58% got out of poverty while 37% slid down into poverty. Moreover, if the incomes
were absolutely driven by random fluctuations (i.e., income ranks in initial and final time
periods are independent), the results of my simulation suggest that about 26% of the country’s
population were poor in both periods, 37% got out of poverty while 16% fell into poverty (left
panel of Figure 2.4).
In terms of changes in inequality, it can also be observed from the right panel of Figure
2.3 that when the correlation of income ranks between the initial and final time period is close
to (positive) one, the observed changes in income inequality can be primarily attributed to low
level of pro-poor growth accompanying low income mobility. On the other hand, as the
correlation moves away from (positive) one, the observed change in income inequality over
the past thirty years becomes a portrait of offsetting forces of highly pro-poor growth and high
income mobility. Furthermore, in terms of the estimated values of the IRPPG, Figure 2.5 shows
that growth for the past three decades has been absolutely pro-poor because the average income
growth of the poor is positive in every income reranking scenario considered. However, in
relative terms, shifting from cross-sectional to longitudinal perspective paints a different
picture. Specifically, the observed growth in the Philippines allows the poor to catch-up with
the non-poor for majority of the scenarios considered except for instances when the income
ranks between the initial and final time period are strongly positively correlated (Figure 2.5).
Of course, whether a pro-poor growth on the basis of IRPPG should be considered as a
desirable outcome or not is a value judgment. Nevertheless, the results presented here hint us
on how our perception of trends in poverty, inequality and pro-poor growth based on cross-
sectional estimates will change when income mobility is accounted for. This validates the
argument that it is important to take income mobility into consideration when examining
household income distribution.
66
Figure 2.3 Income Re-ranking Scenarios Considered
for Constructing Pseudo-Panel Data Sets
Source: Author’s computations using simulated data from Povcalnet.
67
Figure 2.4 Poverty and Inequality Dynamics in the Philippines, 1985-2009
1
Poverty Outflow
Income Reranking
.8
Poverty Inflow
.9
Progressivity of growth
Poverty Persistence
.8
.7
.6
.6
.5
.4
.4
.3
.2
.2
.1
0
0
-1 -.8 -.6 -.4 -.2 0 .2 .4 .6 .8 1
correlation of income ranks -1 -.8 -.6 -.4 -.2 0 .2 .4 .6 .8 1
correlation of income ranks
Figure 2.5 Individual Income Growth in the Philippines, 1985-2009

6
ave. income growth of poor

ave. income growth of non-poor
growth in mean income
4
2
0
-2
-1 -.8 -.6 -.4 -.2 0 .2 .4 .6 .8 1

correlation of income ranks
68
2.4 Summary
Not everyone benefits from economic growth and not everyone falls behind when the
economy contracts. In other words, the consequences of economic development do not accrue
uniformly across all segments in the society. A historical analysis of economic growth pattern
provides socio-economic planners the opportunity to examine the distribution of growth. This
information is helpful when evaluating a country’s ability to sustain rapid economic growth or
identify binding constraints which hamper growth.
Using a range of analytical tools for examining pro-poor growth, this chapter reviewed
the evolution of the income distribution of the Philippines from 1985 to 2009. Data on
household expenditure per capita suggest that the economic growth observed in the country
has been translated to a modest reduction in absolute poverty. However, the data do not provide
sufficient empirical evidence to conclude that the gap between the rich and the poor is
narrowing down. In other words, the income growth experienced by the poor does not seem to
allow them to catch-up with the rich. The statistics presented in this chapter echo the findings
of previous studies about the Philippines’s dismal performance in terms of accelerating income
growth, reducing poverty and closing the gap between the rich and the poor since the 1980s.
Even during episodes of faster economic expansion, the pace of poverty reduction in the
country has been persistently slow. In fact, compared to its other Asian neighbours, studies by
Aldaba (2009), Balisacan (2001) and Schelzig (2005) conclude that the growth elasticity of
poverty in the country has been significantly lower. Whether these results directly imply the
persistence of cumulative advantage and poverty traps is worth rethinking for various reasons.
For instance, the analytical tools used to draw these conclusions fail to account for the profile
of people who moved into and out of poverty. Minimal changes in aggregate poverty rates do
not imply that there have been no movements in and out of poverty. This provides an
incomplete picture of the socio-economic development in the Philippines which in turn, makes
it difficult for policymakers to identify appropriate intervention programs. In fact, through a
simple simulation experiment, the analysis of pseudo-panel data has demonstrated that
conclusions about changes in poverty, inequality and pro-poor growth patterns can be more
dynamic than we conventionally think if mobility of individual incomes is explicitly taken into
account.
The analysis presented in this chapter provides the motivation for examining income
mobility in the succeeding chapters. If a country with a specific set of temporal changes in its
marginal distribution of income portrays a multitude of possible income mobility regimes and
69
in turn, different policy implications for making economic growth more beneficial for
everyone, then it is important to shift our attention from a static to a more dynamic income
distributional analysis. In the case of the Philippines, if the slow reduction in cross-sectional
poverty and income inequality accompanying the rapid economic growth over the past decade
has occurred in the context of high levels of income mobility, it would be indicative that upward
mobility is achievable. Although such pattern cannot discount the difficulties that the poor still
confront, it would encourage policymakers in the country to continue the existing programs
that expand opportunities for income growth and promote greater access to absolute income
mobility across the income distribution. On the other hand, if slow reduction in cross-sectional
poverty and income inequality accompanying the rapid economic growth has occurred in the
context of low income mobility, it could prompt policymakers to re-examine the effectiveness
of existing programs in breaking the vicious cycles of poverty and disadvantage in the
Philippines.
70
Appendix A2.1 Constructing Synthetic Pseudo-Panel Data of Income
The Povcalnet database is the main data source of the synthetic pseudo-panel data. It is
an online poverty analysis tool developed by the WB which contains grouped distribution data
in the form of income shares of different income quantiles aggregated for each country. The
income data is expressed as either income or consumption expenditure in 2005 purchasing
power parity (PPP) adjusted-US dollars.31 This is done by inflating (deflating) the current year
income into 2005 prices using national consumer price indices of each country and then
applying the PPP conversion factors developed by International Comparison Program. The data
available for the Philippines is based on household expenditure per capita and has periodicity
of three years starting 1985 to 2009.
Using the grouped distribution data for the Philippines, I fit Lorenz parametric models to
simulate individual-level incomes following the approach by Datt (1998). In general, a Lorenz
function can be expressed as
L(y) = g(L(p), p, θ) (A2.1)
where
y – income , µ Y– average income, f(y) – income density curve and L(p) – share to total income
of the bottom p percent of the population and θ are parameters of the Lorenz function.
Generally, one can consider different parametric forms for the Lorenz function. The choice
depends on which form will yield a valid Lorenz curve. In this study, I use the Log Normal
form. Preliminary investigations suggest that results are generally robust under different
parametric specifications.
From the grouped distribution data, one can estimate the parameter(s) θ using the grouped
distribution data (p, L(p)). In addition, all Lorenz functions can be expressed as
𝑥
𝐿(𝑝) = ∫ 𝑦𝑓(𝑦)𝑑𝑦 (A2.2)
𝜇𝑦 0
𝑥
𝑝= ∫0 𝑓(𝑦)𝑑𝑦 (A2.3)
It follows that the derivative of a Lorenz function with respect to y evaluated at a point
p0 is equal to the ratio of the income quantile at p0 to overall mean income.
𝐿′(𝑝 = 𝑝0 ) ∗ 𝜇𝑦 = 𝑦(𝑝0 ) (A2.4)
The last equation suggests that, a synthetic income quantile yp can be imputed by
multiplying the derivative of the Lorenz function (with respect to y) evaluated at p = p0 by the
average income. Where appropriate, I evaluate the derivative of L(p)GQ for 10,000 unique
31 Prior to the use of 2005 PPP, the WB estimates poverty and inequality using the 1993 PPP. Milanovic (2009) find that the
difference in poverty and inequality estimates between 1993 and 2005 PPP are not trivial.
71
values p0 that were uniformly distributed within [0,1] range to simulate the entire parametric
Lorenz-based income distribution. This produces an individual-level income dataset with
10,000 data points (i.e., “individuals”) per year. Since the simulated individual-level income
may not exactly match the underlying income distribution from which the grouped distribution
data was derived, I implemented the adjustment procedure proposed by Shorrocks & Wan
(2008) to ensure that the characteristics of the synthetic sample exactly match the actual Lorenz
coordinates used in modelling.32
In the absence of genuine panel data of income for the past three decades, I adopt a naïve
approach to be able to incorporate a longitudinal perspective in my analysis. To create a
pseudo-panel data that will allow me to implement individual pro-poor growth assessment
discussed in Chapter 1, the approach entails the following steps. First, I assume that the
population is closed to births, deaths and migration throughout the observation period.
Certainly, this assumption is somewhat unrealistic but not too far-stretched for the purpose of
demonstrating how conclusions about income distribution trends may change when income
mobility is incorporated. To construct pseudo-panel datasets, recall that from the grouped
distribution data, 10,000 individual income data points are simulated for each (survey) year.
Under the closed population assumption, each of the 10,000 points corresponds to a panel
individual. Since there is no prior information that will enable me to match the identity of the
individuals from 1985 to 2009, I consider a number of possible income re-ranking scenarios.
Each scenario corresponds to one pseudo-panel data set. Figure 2.2 provides an illustration of
the different scenarios considered. The leftmost panel on the first row of the figure depicts an
income reversal scenario wherein the correlation between the initial and final income ranks is
equal to one. In other words, the initially poorest becomes the richest, the initially second
poorest becomes the second richest and so on. The third and fourth panels on the third row of
the figure portray the scenario wherein an individual’s income ranking at the initial time period
is independent of his/her ranking at the final time period. The last panel on the last row
illustrates the scenario wherein income ranks are absolutely persistent.
What are the feasible ranges of the correlation between the income ranks?
Most of the discussion provided in Section 2.4.2 operates under the assumption that there
is no prior information about the correlation between the income ranks of individuals in the
32 Often, the Lorenz coordinates provided in Povcalnet are based from published figures. Thus, employing Shorrocks and
Wan’s algorithm implies that the simulated unit-level income data will produce descriptive statistics that are consistent with
the published figures.
72
initial and final time periods. Hence, the simulations consider different possible correlation
values between -1 and 1. As mentioned earlier, a rank correlation equal to -1 implies that there
is a complete reversal of individual income ranking while a value of +1 implies that income
ranking is completely static. In estimating poverty dynamics based from cross-sectional data,
Dang, Lanjouw, Luoto & McKenzie (DLLM) (2014) confronted the same dilemma. Instead of
choosing a point estimate for the correlation between individuals’ (real) incomes in the initial
and final time periods, the authors used 0 and 1. These correlation values were then used to
construct lower and upper bounds for different indices of poverty dynamics. Noticeably,
DLLM (2014) did not consider negative correlations. They argued that while it is possible for
some individuals to have negative correlation in incomes over time as a result of different
socio-economic mechanisms, it is unlikely that the correlation will be negative for the entire
population.
So what are the reasonable values for this correlation? As a form of validation exercise,
I use actual panel data available from the Philippines between 2006 and 2009, the details of
which are described in Chapter 3. Figure A2.1 shows the correlation between the rank of
individuals’ initial and final incomes in the Philippines. The Pearson’s correlation index gives
a value of 0.85. However, one could expect that the correlation of income ranks from 1985 to
2009 to be much lower as we believe that there would be more income mobility for a longer
observation period. To validate this hypothesis, I also use panel data with a longer observation
period for a country which is quite similar to Philippines. In this context, I refer to the Indonesia
Family Life Survey. From 1993 to 2007, the estimated correlation of income ranks is 0.41. If
the true income rank correlation from 1985 to 2009 is within the vicinity of this estimate, then
Figure 2.4 would suggest that the initially poor are catching-up in the sense that their incomes
are growing faster than that of the initially non-poor.
73
Appendix Figure A2.1 Correlation of Income Rank using
Correlation of Income
Actual Panel Ranks
Data in in Philippines,
Philippines, 2006-2009
2006-2009
8000
6000
4000
2000
0 2000 4000 6000 8000

rank in 2006
74
Chapter 3 Family Income and Expenditure Survey and Labour Force Survey
3.1 Introduction
In the previous chapters, I have briefly examined how the income distribution in the
Philippines has evolved since the 1980s. One of the interesting patterns that emerged is that
both poverty and inequality barely changed over the past decade despite moderate to rapid
economic growth. More importantly, I have also discussed the importance of probing beyond
conventional cross-sectional indicators of the income distribution to be able to better
understand the dynamics of socio-economic development. After having shown that important
features of the income distribution dynamics may be overlooked if the patterns of income
mobility are not taken into account using a simple simulation experiment based on pseudo-
panel data, the rest of this paper examines income mobility patterns in the Philippines using
actual panel data. This chapter describes the Family Income and Expenditure Survey and
Labour Force Survey which serve as the main data sources for the analyses presented in the
succeeding chapters. The following discussion centres on the content of FIES and LFS,
observation period, income measure and representativeness of survey data.
3.2 Survey Content and Administration
The Philippine National Statistics Office (NSO) conducts the FIES every three years to
collect household income distribution data. The main data collection instrument is an
approximately seventy-page questionnaire which includes detailed questions about different
sources of household income and a comprehensive list of expenditure items. Data is collected
through face-to-face interviews wherein the main respondent, typically the head of the
household, is asked to answer the survey questionnaire. In addition to household earnings and
expenditure, FIES also collects information about household characteristics such as the profile
of household head, household composition, the dwelling unit characteristics, type of assets held
and access to basic services. In particular, about 60% of the survey instrument is allotted to the
consumption module, 15% for the earnings module and the remaining 25% is allotted for
household characteristics and other information (Ericta & Fabian 2009).
The Labour Force Survey (LFS) is likewise collected by the NSO every quarter and it is
the main data source for official employment statistics in the Philippines. It has a four-page
questionnaire that collects information about the employment status of each household
member. In particular, it asks questions such as labour force status, type of employment and
75
sector of employment. The LFS also collects the basic socio-demographic characteristics such
as age, sex and educational qualification of each household member (NSO 2012).
The FIES follows a semestral recall method wherein each sampled household is visited
twice every year. The first visit is usually conducted in July of the reference year while the
second visit is conducted in January following the reference year (Ericta & Fabian 2009). The
NSO follows this scheme to capture the seasonal variations of household income and
consumption (Ericta & Fabian 2009). On the other hand, the LFS is collected every April, July,
October of the reference year and in January of the succeeding year (NSO 2012). The July and
January rounds of LFS coincide with the semestral rounds of FIES making it possible to merge
the household-level data collected from FIES with the individual-level data of LFS (NSO
2003).
Both FIES and LFS have undergone several revisions over the years (NSO 2003; Ericta
& Fabian 2009). For instance, along with other household surveys conducted by NSO, both
surveys started following the 2003 Master Sample Design for Philippines Household Surveys
in 2003.33 The 2003 Master Sample Design provides a scheme where a subsample of
households used in previous survey waves are rotated back for the succeeding waves (Ericta &
Fabian 2009).
3.3 Observation Period
The reference years for this study are 2003, 2006 and 2009. This period is quite
interesting for both substantive and technical reasons. We have seen in Chapter 2 that compared
to previous years, the Philippines experienced more rapid economic growth during this period.
From 2003 to 2009, the country posted an annual average growth of 2.9% in terms of GDP per
capita. This is about twice as fast as the country’s average income growth rate two decades
earlier (WDI 2014). In particular, the first period, 2003-2006, marks the transition from several
decades of slow economic growth to faster economic expansion. However, the higher
economic growth rates occurred in the context of a slight increase in cross-sectional poverty,
which could be indicative that the poor have benefitted less from economic growth. The second
period, 2006-2009, continues the rapid economic growth trend. Although this period coincides
with the 2008 global financial crisis, both cross-sectional headcount poverty rates and income
inequality did not change significantly during this period.
33
In repeated cross-sectional surveys, a master sample is a sample from which subsamples can be drawn for the purpose of
more than one (household) survey or more than one round of survey.
76
The study period also precedes the more rapid economic episode that is currently
experienced by the Philippines. Although the transition into a faster economic growth regime
could have paved way for more significant gains in reducing the number of poor and the gap
between the poor and the rich, estimates of poverty rates and inequality levels barely changed
during this period. The availability of panel data during this period through the redesigned FIES
and LFS provides the opportunity to examine income mobility to give a more nuanced
assessment of seemingly trivial changes in cross-sectional poverty and inequality and hence, a
more comprehensive appraisal of the country’s socio-economic development.
3.4 Income Measure
There are several monetary measures that are of interest for income distributional
analysis. Two of the most commonly used measures are income and consumption expenditure.
Haughton and Khandker (2009) identified the advantages and disadvantages of using income
or consumption expenditure. For instance, income data collected from surveys can be
compared with administrative tax data records to check the reliability of the survey data. More
importantly, examining the patterns of the distribution of different sources of income (e.g.,
employment, remittances, subsidies, etc.) is also relevant in policy analysis. In the Philippines,
household income which is the sum of income from paid employment, self-employment, assets,
transfers, remittances and other sources is used as basis for computing official poverty statistics
(NSCB 2003).Nevertheless, a number of researchers examining income distribution in the
country have increasingly favoured the use of consumption expenditure (David & Maligalig
2001; Balisacan & Pernia 2002) for various pragmatic and conceptual reasons. In particular,
advocates of consumption expenditure-based measures argue that expenditure is more closely
related to a person’s living standards because it does not only reflect the welfare level that a
person can achieve using its income but also captures one’s ability to access savings or credit
markets during episodes of low income. In addition, some argue that income data usually
suffers from downward bias when survey respondents discount the extent to which they
consume their own production, especially in developing countries that heavily depend on the
informal economy in which various income sources are hard to capture and can have erratic
fluctuations from time to time (Deaton & Zaidi 2002). In addition, consumption expenditure
flow is generally smoother than income and thus, the former is considered a better measure of
people’s long-term economic prospects (Jefferson 2012). The other benefits of using
consumption expenditure as a measure of material well-being include its ability to better reflect
price change, private and government transfers as well as insurance value of government
77
programs and credit markets (Meyer & Sullivan 2003). For further discussion, readers may
refer to the work of Meyer & Sullivan (2003) who provided well-developed arguments in
favour of using consumption expenditure rather than income when examining a country’s
income distribution.
Table 3.1 Regional Price Differences
Region Region
National Capital Region 7 - Central
100 105.4
Region (NCR) Visayas
Cordillera
Region 8 - Eastern
Administrative Region 100.3 90.3
Visayas
(CAR)
Region 9 - Zamboanga
Region 1 - Ilocos 93.4 94.1
Peninsula
Region 2 - Cagayan Region 10 - Northern
92.6 89.2
Valley Mindanao
Region 3 - Central
96.5 Region 11 - Davao 99.7
Luzon
Region 4-A - Region 12 -
95.3 87.2
CALABARZON Soccsksargen
Autonomous Region of
Region 4-B -
96 Muslim Mindanao 107.6
MIMAROPA
(ARMM)
Region 5 - Bicol 97.8 Caraga Region 88.9
Region 6 - Western
96.3
Visayas
Source: Sta. Ana and Varona (2012)
Note: 100 = reference category
Following the common practice in many developing countries (Deaton 1997), this study
mainly uses consumption expenditure, unless stated otherwise. Household expenditure derived
from FIES is the sum of expenditure on food, utilities, household operation, personal care,
taxes and miscellaneous items. From this point onwards, the term income is used loosely to
refer to this chosen monetary measure unless stated otherwise.34 This income measure has been
divided by the household size, assuming that there are no economies of scale in consumption
(Lanjouw & Ravallion 1995). This is slightly different from the approach commonly used in
industrialized countries which adopts an equivalence scale to take into account the possibility
that children generally consume less than adults. Nevertheless, the choice of expressing the
income measure in per capita terms is consistent with the usual practice in developing countries
(Deaton 2004). Within each household, all household members are given the same income
34I have also done preliminary analysis of mobility of household income per capita. However, the results are qualitatively
similar with the patterns depicted by the mobility of household expenditure per capita. To save space, I decided to focus on
expenditure. Nevertheless, Appendix 4.1 presents some results based on household income per capita.
78
value. This is consistent with the notion that households act as the main budgetary units for
which socio-economic decisions of individuals are made (WB 2014a). Unless stated otherwise,
all estimates are expressed as monthly income per capita in constant 2005 PPP US$ to account
for inflation. The income measure is further adjusted to account for differences in regional
prices using the results based from the spatial price indices estimated by Sta. Ana & Varona
(2012) (Table 3.1). In such cases, the prices in the National Capital Region is used as
benchmark.
3.5 Sampling Design and Survey Weight Adjustments for Non-Coverage Bias
The FIES and LFS follow multistage stratified sampling design (NSO 2003). The target
population of both surveys includes all households in the Philippines except institutional
households and households from least accessible barangays (LAB) or villages (NSO 2003).35
The villages (or combination of villages in some instances) were treated as the primary
sampling units (PSUs). For each geographic region, the total number of sampled households
were computed such that it would satisfy a pre-determined level of reliability.36 Hence, the
total number of sampled PSUs per region was computed by dividing the target number of
sampled households per region by the desired sample size per PSU. The PSUs were selected
using the probability proportional to size sampling scheme where size is gauged in terms of the
number of households enumerated in the 2000 Census of Population and Housing. For each
sampled PSU, housing units were selected with equal probability (NSO 2003). This resulted to
a target sample of approximately 40,000 households for each survey wave.
The target sample was grouped into four replicates. In this study, I use the households
from the fourth replicate only because these are the households that were tracked over time.
There are three potential panel data sets that can be constructed: (i) households that are
observed in both 2003 and 2006, (ii) households that are observed in both 2006 and 2009; and
households that are observed in all three periods.37Table 3.2 shows the sample size for each of
the panel data sets.
35 A village is considered an LAB if (i) there is no regular means of transportation, (ii) the cost of one-way fare from the
nearest accessible village is more than $10 to $15 based on 2003 prices or (iii) it takes more than 8 hours to reach the village.
Of the 41,942 villages in the Philippines, 350 were classified as LABs and excluded from the target population (NSO 2003).
On the other hand, institutional households refer to institutions that provide care to a group of people (e.g., health care
institutions, etc.) (UNESCAP 2009).
36 The regional sample size was computed so that the expected coefficient of variation of headcount poverty rate would not
exceed 10% in NCR and 5% in areas outside NCR (NSO 2003).

37 The full sample is designed to produce reliable estimates at the national and regional levels. On the other hand, the
longitudinal sample is expected to provide reliable estimates at the national level.
79
Table 3.2 Sample Size
2003,
2003 2003- 2006-
Regions 2006,
(replicate #4) 2006 2009
2009
Philippines 10,476 7,899 7,509 6,519
National Capital Region (NCR) 954 611 724 449
Cordillera Administrative Region
(CAR) 404 297 319 261
Region 1 - Ilocos 616 517 469 441
Region 2 - Cagayan Valley 523 421 406 372
Region 3 - Central Luzon 842 666 618 551
Region 4-A CALABARZON 1,021 745 681 604
Region 4-B MIMAROPA 450 331 297 266
Region 5 - Bicol 612 481 469 396
Region 6 - Western Visayas 736 623 558 525
Region 7 - Central Visayas 745 550 505 464
Region 8 - Eastern Visayas 560 424 381 353
Region 9 - Zamboanga Peninsula 451 338 334 299
Region 10 - Northern Mindanao 558 384 354 322
Region 11 - Davao 559 415 406 346
Region 12 - Soccsksargen 550 411 397 341
Autonomous Region of Muslim
442 351 280 257
Mindanao (ARMM)
Caraga Region 453 334 311 272
Source: Author’s computations using longitudinal subsample of FIES 2003, 2006 and 2009.
While the longitudinal subsample is expected to provide reliable estimates at the national level,
it is not free from the risk of producing biased estimates. There are several sources of bias in
this context. First, the survey does not follow households that moved out of its previous
dwelling unit. Excluding them from the analysis could lead to the well-known non-coverage
bias common in longitudinal studies. Second, bias may also arise from panel nonresponse when
households that remained in the same dwelling unit refuse to participate in the succeeding
survey waves. The consequences of these biases are major concerns in many longitudinal
studies especially when the profile of sampling units that drop out are systematically correlated
with the characteristic of being studied (Ashenfelter, Deaton & Solon 1986). To investigate
this issue, I compare the measure of central tendency and dispersion of household expenditure
per capita between the full cross-sectional sample and the longitudinal subsample. Preliminary
investigation suggests that measures of central tendency and dispersion tend to be
underestimated in the longitudinal subsample, especially when I examine households that
80
appear in all three waves (Table 3.3).38 To formally test whether the differences in the
distributions are statistically significant, I use the Kolmogorov-Smirnov test. This test confirms
that there are significant differences in the distributions of the full sample and longitudinal
subsample. In this context, I am likely to produce biased estimates of income mobility if I do
not introduce further adjustments.
Table 3.3 Comparison of Full Cross-Sectional and Longitudinal Subsample

2003 2006 2009
Mean Gini Mean Gini Mean Gini
Time period
Std Std Std Std Std Std
Error Error Error Error Error Error
Full sample
cross-sectional
1258.53 0.44 1228.05 0.441 1286.33 0.43
sample
9.43 0.002 10.88 0.003 12.79 0.002
Longitudinal
subsample
2003-2006 1158.34 0.434 1121.06 0.44
13.34 0.005 13.41 0.004
2006-2009 1197.41 0.449 1223.23 0.426
15.52 0.005 14.34 0.004
2003, 2006 and
1138.48 0.428 1132.76 0.438 1159.69 0.414
2009
28.32 0.006 28.80 0.005 25.86 0.004
Source: Author’s computations using household expenditure per capita data from the longitudinal subsample of
FIES 2003, 2006 and 2009.
Notes: For each year, the first column represents the average and its corresponding standard error while
the estimates of the Gini coefficient and its standard error are provided in the second column. The mean
is expressed in 2005 PPP US$.
Table 3.4 Features of Longitudinal Subsample Using Attrition-Adjusted Weights

2003 2006 2009
Time period
Mean Gini Mean Gini Mean Gini
Longitudinal
subsample
(Adjusted)
2003, 2006 and
1234.84 0.431 1233.27 0.445 1267.91 0.423
2009
31.30 0.006 32.57 0.006 29.21 0.005
FIES 2003, 2006 and 2009.
Notes: For each year, the first column represents the average and its corresponding standard error while
the estimates of the Gini coefficient and its standard error are provided in the second column. The mean
is expressed in 2005 PPP US$.
38 The same findings are drawn for household income per capita. Various software packages such as DASP (Araar & Duclos
2007), Stata’s income mobility tools (Van Kerm 2002), ADECOMP (Azevedo, Nguyen & Sanfelice 2012) and MCLUST
(Fraley & Raftery 2012) are used extensively in most of the computations.
81
Rubin (1987) proposed the use of weighting procedure to address attrition and non-
coverage bias. Hence, I introduce weights for non-response by estimating logistic regression
models for the probability of appearing in 2003, 2006 and 2009 waves, and specifying
consumption, age of household head, sex of household head and urbanity as controls. The
inverse of the predicted probabilities are multiplied with the existing survey weights. Table
3.4 shows the average household expenditure per capita and measure of inequality estimated
using the adjusted survey weights. Although it is noticeable that the adjusted estimates are now
more closely aligned with estimates based on the full sample, some studies suggest that the
weighting procedure only corrects for observable characteristics related to dropping out of the
sample (Fiztgerald, Gottschalk & Moffitt 1998). In other words, the longitudinal subsample
may still be systematically different from the full sample in terms of unobservable
characteristics. Nevertheless, it is important to note how this residual bias can affect the
estimates. Given that households that moved from their original dwelling were not followed, it
is possible that households with deteriorating socio-economic status and migrated between
2003 and 2006 and households with improved socio-economic status and migrated between
2006 and 2009 are systematically underrepresented. If we examine the numbers provided for
the full cross-sectional sample in Table 3.3 and the numbers provided in the longitudinal
subsample in Table 3.4, this would explain why the average income in 2003 and 2009 is lower
in the longitudinal subsample but it is higher in 2006. Furthermore, it is apparent that the year-
on-year differences in the mean and Gini coefficient are smaller using the longitudinal
subsample than using the full cross-sectional sample. This could imply that the estimates
provided in this study are likely to represent lower bounds of the actual magnitude of income
mobility in the Philippines.
Throughout the rest of this study, all analysis incorporates weights adjusted for attrition
and non-coverage. These weights are also multiplied by the household size. Hence, a household
consisting of three family members is weighted thrice as high than a single-person
household.39For convenience, I restrict the analysis to data for the 6,519 households that appear
in all waves.40
As can be inferred from Table 3.5 and Figure 3.1, regions in the northern part, particularly
the National Capital Region (NCR) have significantly higher income than the rest. Thus, I also
39Thus, the weights sum up to the total individual population.

40I decided against trimming the data (to remove the outliers) because the public use file of FIES has already undergone
various data cleaning processes (Ericta & Fabian 2009).
82
provide subnational estimates in most of the succeeding analyses to capture the spatial
variations in economic well-being within the country.
Table 3.5 Average Household Income Per Capita by Region

Panel Sample
Location 2003 2006 2009
mean std err mean std err mean std err
LUZON
National Capital
2239.77 133.50 2297.86 158.83 2195.53 114.48
Region (NCR)
Cordillera
Administrative 1157.25 111.67 1065.41 133.72 1054.19 100.43
Region (CAR)
Region 1 - Ilocos 1133.38 100.41 1232.92 106.72 1300.44 92.23
Region 2 - Cagayan
1216.74 90.40 1217.32 87.59 1295.92 100.16
Valley
Region 3 - Central
1337.95 62.13 1411.11 73.77 1403.81 74.00
Luzon
Region 4-A -
1613.97 90.42 1545.10 92.11 1503.36 80.93
CALABARZON
Region 4-B -
757.33 88.42 744.52 62.63 930.42 136.97
MIMAROPA
Region 5 - Bicol 1192.61 190.30 1041.54 149.76 1041.21 122.31
VISAYAS
Region 6 - Western
1045.55 74.98 1044.50 84.10 1137.46 80.80
Visayas
Region 7 - Central
957.23 75.94 960.16 75.94 977.80 71.32
Visayas
Region 8 - Eastern
972.25 79.49 994.93 73.74 1201.69 121.17
Visayas
MINDANAO
Region 9 -
Zamboanga 846.81 105.07 928.04 138.61 1032.59 140.25
Peninsula
Region 10 -
1039.68 102.08 1095.81 101.71 1095.94 74.97
Northern Mindanao
Region 11 - Davao 1071.99 102.23 915.09 75.76 1046.81 99.60
Region 12 -
836.42 58.55 825.25 62.58 995.79 71.98
Soccsksargen
Autonomous
Region of Muslim 576.56 52.03 516.88 30.00 567.09 34.35
Mindanao (ARMM)
Caraga Region 742.23 55.29 746.05 53.21 777.95 61.96
Source: Author’s computations using household expenditure per capita data from the longitudinal subsample of FIES 2003,
2006 and 2009.
83
3.6 Poverty Lines
In the succeeding discussion, poverty is measured using four sets of poverty lines. Three
of them are absolute poverty thresholds in the sense that the poverty status of a household does
not depend on the incomes of other households. These are the US1.25/day (US$456/year),
US$2/day (US$729.6/year) poverty line proposed by WB and the official poverty line compiled
by the NSCB. Unlike the US$1.25 and US$2/day poverty lines which take a single scalar value,
the national poverty line differs across provinces. On the other hand, the half-of-median
threshold is a relative poverty line in the sense that it implicitly depends on the distribution of
the incomes of all households. Because the half-of-median yields the smallest poverty threshold
followed by the official poverty line while the US$2-a-day produces the highest poverty
threshold, one would expect that poverty estimates will be highest based on the US$2-a-day
poverty line. Thus, the US$1.25/day and US$2/day-based estimates can be considered as lower
and upper bound of poverty, respectively.
3.7 Summary
This chapter described the features of the main survey data that are pertinent to the
measurement of income mobility. The succeeding chapters cover various topics about income
mobility patterns in the Philippines.
84
Figure 3.1 Regional Map of the Philippines
85
Appendix Table A3.1 Descriptive Statistics for Household Expenditure Per Capita
(in 2005 PPP US$)
2003 2006 2009
Household characteristic
Mean Median Std Dev Obs Mean Median Std Dev Obs Mean Median Std Dev Obs
Philippines
Urbanity
Rural 856.28 617.55 1,299.57 4,047.00 836.55 600.63 785.56 4,047.00 904.00 663.42 809.76 4,047.00
Urban 1,625.43 1,222.39 1,435.02 2,472.00 1,642.59 1,158.03 1,706.54 2,472.00 1,643.38 1,178.89 1,589.99 2,472.00
Major Island Group
NCR 2,239.77 1,631.83 1,896.19 449.00 2,297.86 1,572.89 2,497.04 449.00 2,195.53 1,624.21 1,906.35 449.00
Luzon excl. NCR 1,316.74 966.99 1,591.75 2,891.00 1,305.11 942.77 1,286.49 2,891.00 1,318.59 982.37 1,261.51 2,891.00
Visayas 996.47 696.19 972.32 1,342.00 1,001.86 662.25 1,023.92 1,342.00 1,089.44 723.04 1,156.18 1,342.00
Mindanao 879.85 587.52 860.63 1,837.00 867.17 572.20 889.19 1,837.00 950.81 633.62 1,017.51 1,837.00
Gender of household head
Female 1,819.13 1,267.04 2,620.60 951.00 1,665.71 1,200.25 1,586.66 1,080.00 1,689.69 1,268.03 1,513.39 1,273.00
Male 1,147.69 817.39 1,112.64 5,568.00 1,156.94 773.37 1,327.74 5,439.00 1,177.44 821.74 1,242.52 5,246.00
Age of household head
Hhld head's age ≤ 35 1,057.62 752.26 1,090.41 1,379.00 1,041.69 659.66 1,455.68 900.00 1,029.78 699.56 1,251.77 636.00
35 < Hhld head's age ≤ 44 1,111.98 786.26 1,711.71 1,614.00 1,043.42 726.19 1,016.28 1,618.00 1,113.45 783.60 1,075.31 1,421.00
Hhld head's age > 44 1,405.34 993.76 1,362.68 3,526.00 1,393.73 947.58 1,504.67 4,001.00 1,377.38 966.49 1,395.28 4,462.00
Education of household head
Primary school 799.02 617.40 651.25 3,253.00 776.75 581.63 668.01 3,228.00 816.58 639.14 606.71 3,181.00
Secondary school 1,336.65 1,041.70 1,045.21 2,728.00 1,314.64 991.63 1,116.67 2,734.00 1,335.08 1,024.85 1,105.46 2,757.00
College 2,930.43 2,237.76 3,343.99 538.00 3,018.15 2,257.93 2,843.42 557.00 2,990.61 2,392.24 2,524.09 581.00
Family size
Family size ≤ 3 1,906.94 1,285.16 2,666.63 1,470.00 2,058.65 1,334.69 2,518.04 1,529.00 2,103.07 1,424.34 2,327.92 1,693.00
3 < Family size ≤ 5 1,348.82 1,001.87 1,130.78 2,448.00 1,354.42 979.91 1,188.48 2,467.00 1,430.39 1,044.28 1,249.85 2,443.00
5 < Family size ≤ 7 1,045.05 759.88 922.57 1,679.00 1,023.06 726.19 946.29 1,615.00 1,064.79 770.61 928.09 1,571.00
7 < Family size ≤ 9 793.15 571.46 653.19 657.00 802.88 574.13 709.83 669.00 903.24 683.06 775.46 588.00
Family size > 9 841.58 523.23 1,169.48 265.00 761.18 523.87 928.67 239.00 718.34 583.01 480.59 224.00
Main source of Income
Agriculture 1,464.68 1,071.22 1,576.83 4,473.00 1,451.08 1,027.53 1,476.05 4,483.00 1,460.32 1,057.88 1,416.70 4,704.00
Non-Agriculture 595.30 490.58 402.41 2,046.00 625.45 480.11 808.81 2,036.00 645.69 528.90 507.97 1,815.00
Source: Author’s computations using household expenditure per capita data from the longitudinal subsample of FIES 2003, 2006 and 2009.
86
Chapter 4 Is there Income Mobility in the Philippines?
4.1 Introduction
I noted in Chapter 2 that the economic development in the Philippines has been
characterized by boom and bust cycles until the 2000s. However, starting 2000, the country
experienced a more rapid growth episode and estimates suggest that it is gaining more
momentum in the recent years (WDI 2014). Modest to rapid economic growth can be expected
to raise household income. However, this link seems to be weak in the Philippines. Household
survey estimates suggest that from 2003 to 2009, average household income per capita
increased by only 0.36% annually. Further, despite 4.1% annual GDP growth from 2009 to
2012, average household income barely moved (NSCB 2013a). Estimates from household
surveys also suggest that income inequality remained high over the past decade. For instance,
the Gini coefficient based on household income per capita hardly changed from 0.44 in 2003
to 0.43 in 2009. Nevertheless, the minimal changes in the cross-sectional estimates of poverty
and inequality do not necessarily imply that the country’s income distribution is stagnant. As
mentioned in Chapter 1, examination of income mobility could provide additional information
on the underlying mechanisms that drive the income distribution to change. Thus, the main
objective of this chapter is to provide a descriptive analysis of the income mobility that
transpired over the past decade in the Philippines. In particular, the chapter addresses the
following questions:
(i) Why is average household income not growing at the same pace as the country’s
overall economy?
(ii) What does a small change in cross-sectional inequality mean?
(iii) Are all households’ incomes static over time or are they changing at different
rates?
(iv) If there is mobility, is it characterized by genuine income movements?
To answer these questions, I use a portfolio of analytical methods, both conventional and new
in the income mobility literature to document the income mobility patterns. Acquiring this
knowledge is the first step to being able to provide inputs for policymakers in developing
policies that will foster more inclusive economic growth. Throughout the chapter, I use
household expenditure per capita as the main income measure.
87
4.2 Is there income mobility in the Philippines?
4.2.1 Is there relative income mobility?
We learned from Chapter 1 that relative mobility refers to how the income of each unit
changes in comparison with the changes observed in other units of the population. A simple
way to gauge the level of relative mobility is to estimate the proportion of the population
moving from one income quantile to another over time using an income transition matrix.
Using household expenditure per capita data derived from the FIES, the results from this
exercise show that in 2009, about 85% of the household population were found in a vingtile
different from its origin in 2003, 72% were found in a different decile; and 52% were found in
a different quintile.41 Nevertheless, income persistence was also strong. From 2003 to 2009,
about 55% of the population did not move beyond two vingtiles from 2003 to 2009.
With respect to directional mobility, I find that about 11% of the population moved one
vingtile up while 17% moved one decile up. Long-distance upward moves were also not trivial.
For instance, the proportion of the population moving in a higher quintile from 2003 to 2009
was 26.3%. Because relative mobility is based on income ranks, one could expect that
downward mobility is as frequent as upward mobility. This is because in a fixed population,
for a person to be able to move up an income rank, another person has to go down portraying
a zero-sum game. For instance, I find that about 25.7% moved into a lower quintile during the
observation period.
Another way of measuring the extent of relative mobility is to look at how income ranks
among the population units have changed over time. To what extent does current income rank
dictate one’s ranking in the future? In general, a stagnant household income distribution will
imply that people that were initially at the bottom of the income hierarchy will remain at the
bottom, while the rich will continue occupying the top spot. If we plot the ranks of the initial
and final-period incomes, a stagnant distribution will resemble a perfect unit-slope diagonal
line. The rank correlation calculated is relatively high with the Pearson correlation estimated
at 0.80. Regressing the income rank in 2009 on the income rank in 2003, I find that the slope
coefficient is close to one, which indicates income rigidity. Nevertheless, while income ranks
are persistent, there is also considerable relative mobility in the sense that a significant fraction
of the population experienced higher or lower income ranks in 2009 compared to their initial
ranks in 2003. In addition, based on the results of the simple linear regression analysis, 40% of
the variation in income ranking in 2009 cannot be explained by the income rank in 2003.
41(Income) vingtiles divide the population into twenty groups according to income, ten groups for deciles and five groups for
quintiles.
88
Table 4.1 provides a summary of the amount of relative mobility that occurred from 2003
to 2009. Overall, the results show that despite the strong persistence of income ranking, the
income dynamics that transpired during this period is also characterized by considerable
medium and long-distance movements of income ranks.
Table 4.1 Summary of Relative Income Mobility Measures

Income mobility indicator 2003-2009
Average number of vingtiles moved (non-directional) 2.77
0.03
Average number of vingtiles moved (directional) 0

0.05
Proportion of population remaining in leading 0.15
diagonals 0.005
Proportion of population moving one vingtile up 0.11

0.005
Proportion of population moving one vingtile down 0.12

0.005
Proportion of population moving two vingtiles up 0.08

0.004
Proportion of population moving two vingtiles down 0.09

0.004
Proportion of population moving at least three vingtiles 0.23
up 0.006
Proportion of population moving at least three vingtiles 0.22
down 0.006
Correlation of income ranks 0.8***
Source: Author’s computations using household expenditure per capita data from the longitudinal
subsample of FIES 2003, 2006 and 2009.
Note: The numbers in smaller font size are standard errors.
4.2.2 Is there absolute income mobility?

If all incomes increased by a constant proportional factor, the relative mobility measures
presented in the previous section would indicate that there is no mobility. This is because
relative mobility indicators only capture variations in income shares or rank orders of the
population units over time. In other words, relative mobility measures are more sensitive to the
changes in the shape of the income distribution rather than the changes in location. In contrast,
absolute mobility gauges how income levels change from one time period to another. In this
context, absolute mobility captures the growth dimension of income dynamics.
Nevertheless, there is still much mobility going on. For instance, following the approach
used in ADB (2010a), I group each household according to its income levels to determine
89
Table 4.2 (Absolute) Income Transition Matrix, 2003-2009
2009
low upper
extreme moderate middle
middle middle rich
poverty poverty income
income income
extreme
poverty 0.4942 0.3951 0.1045 0.0062 0.0000 0.0000
moderate
2003
poverty 0.2185 0.4219 0.3341 0.0243 0.0007 0.0004

low middle
income 0.0423 0.1983 0.5660 0.1860 0.0074 0.0000
middle income 0.0040 0.0231 0.2933 0.5776 0.0934 0.0085
upper middle
income 0.0000 0.0000 0.0487 0.5298 0.3644 0.0572
rich 0.0000 0.0000 0.0221 0.2665 0.4836 0.2279
Source: Author’s computations using household expenditure per capita data from the longitudinal subsample of FIES
2003, 2006 and 2009.
whether it is extremely poor, moderately poor, lower middle income, middle income, upper
middle income and rich.42
The results are summarized in the income transition matrix presented in Table 4.2. From
2003 to 2009, about 49% of the household population changed income status. Interestingly, I
find that both positive and negative mobility are prominent features of the income dynamics
observed during this period. More importantly, it appears that there is as much positive
mobility as there is negative mobility. Noticeably, five out of ten Filipinos who lived in extreme
poverty in 2003 were still in the same situation in 2009; four moved to moderate poverty and
the remaining 10% reached low middle income status. About 40% of those who lived in
moderate poverty in 2003 remained in the same income status in 2009 while more than 20%
fell to extreme poverty. In addition, those who were in either middle or upper middle-income
status in 2003 were less likely to experience income increase than those who were initially in
lower middle-income strata. Although not shown in Table 4.2, the number of units found at
the right of the off-diagonal section of the transition matrix is slightly higher than number of
units at the left off-diagonal. In particular, about 26% of the population moved up to a higher
income status while 22% moved down.
Thus far, the inferences based on (absolute) transition matrix have implicitly relied on
the following measures of positive and negative mobility denoted by MU and MD where MU
42The first group consists of incomes not exceeding US$1.25/day which is considered as an extreme form of income poverty,
the second group consists of incomes falling in between US$1.25 and US$2 (moderate poverty), third group consists of
incomes falling in between US$2 and US$4 (lower middle income), fourth group consists of incomes falling in between US$4
and US$10 (middle income), fifth group consists of incomes falling in between US$10 and US$20 (upper middle income) and
last group consists of incomes exceeding US$20/day (rich). The cut-off points are expressed as daily expenditure per capita in
2005 PPP US$.
90
represents the proportion of the population whose income went up a certain income threshold
c while MD represents the proportion of the population whose income went down with respect
to c.
𝑀𝑈 = ∑ (𝑌𝑖 ) (𝑌𝑖2 ) (4.1)
𝑛
𝑀𝐷 = ∑ (𝑌𝑖 ) (𝑌𝑖2 ) (4.2)

𝑛
Noticeably, the value of these measures depend on the pre-specified cut-off point c. In this
context, the transition matrix above treats all households with the same income status as
identical. In other words, this transition matrix and other scalar measures derived from it, fail
to capture positive and negative income movements occurring within the same income group.
Recently, Foster & Rothbaum (2012) proposed a more general approach in comparing positive
and negative income mobility rates. In particular, the authors defined MU* and MD* such that
𝑀𝑈∗ = ∫0 𝐻 ∑ (𝑌𝑖 ) (𝑌𝑖2 )𝑑 (4.3)

𝑛
𝑀𝐷∗ = ∫0 𝐻 ∑ (𝑌𝑖 ) (𝑌𝑖2 )𝑑 (4.4)

𝑛
The mobility measures MU* and MD* are integrated for all values of c, 0 ≤ c ≤ cH, where cH can
be set to be equal to the maximum income observed throughout the observation period. Hence,
these mobility measures are not sensitive to a predetermined income threshold. Such a mobility
measure provides an analytical tool for examining income mobility that incorporates the
distribution sensitivity approach of the income transition matrix but is not pegged with respect
to some predetermined income cut-off points (Foster & Rothbaum 2012). Figure 4.1
summarizes the upward and negative mobility estimates based on these mobility measures. The
y-axis of the income mobility curve represents the values of MU and MD while the x-axis
represents the different cut-off points c. We can see that for cut-off points less than the mean
income, mobility rates increase uniformly and peak around the mean income. Thereafter,
mobility rates gradually decrease. Consistent with the findings from the income transition
matrices, I find that the total amount of positive income mobility is slightly higher than negative
mobility. However, the difference between the two does not seem to be significant.
Remarkably, the pattern of positive and negative mobility is quite symmetric across different
cut-off points. The symmetry implies that for every household that experienced a positive
(absolute) income increase at any point in the income distribution, there is a household that
experienced a reduction in its (absolute) income level. Table 4.3 provides a summary of the
amount of absolute mobility that occurred from 2003 to 2009.
91
Figure 4.1 Income Mobility Curve, 2003-2009
.1 M*u = 256.34
0 M*d = 220.55
-.1
0 1200 2400 3600 4800 6000 7200 8400 9600 10800 12000
values of c
Table 4.3 Summary of Income Mobility Measures

Income mobility indicator Estimate
Average absolute change 492.68
|Income2009 - Income2003| 13.69
Average absolute percentage change 0.41
|Income2009 - Income2003|/Income2003 0.006
Average income change 33.07
(Income2009 - Income2003) 15.32
Average percentage change 0.16

(Income2009 - Income2003)/Income2003 0.008
Source: Author’s computations using household expenditure per capita
data from the longitudinal subsample of FIES 2003, 2006 and 2009.
Note: The numbers in smaller font size are standard errors.
4.2.3 Is there equalizing mobility?

The level of income inequality in the Philippines is one of the highest in Southeast Asia
(WDI 2014). Despite slight indications of a decreasing trend over the recent years, estimates
suggest that income inequality in the country has remained very high. Because the observation
that it has remained high is based on cross-sectional trends, we can only conclude that the gap
between the rich and the poor has remained wide. However, as noted from the previous section,
households are not necessarily static with respect to their position in the income distribution.
In other words, those who are poor today are not necessarily the same households who were
poor yesterday. The same is true for the middle class and the rich households. In particular, if
poorer households have higher upward mobility prospects, then it is possible that despite the
92
high levels of cross-sectional inequality, we would still observe lower inequality in the long-
run.
Examining the relationship between long-run inequality and income mobility is the main
theme of this section. Unlike the measures used in the previous section which view mobility in
terms of income movements, this section investigates the extent to which income inequality
that exists at any given time, is offset when household incomes are averaged over time and
whether there is greater or less mobility at the lower income segments relative to mobility in
higher income ranges.
Table 4.4 Inequality-Reducing Effect of Income Mobility

GE(0) GE(1) Gini GE(2)
2003-2009
Single year income 0.3081 0.3449 0.4309 0.6074
Average income 0.2773 0.3049 0.4115 0.4710
Shorrocks’ R 0.0870 0.0980 0.0390 0.2246
Source: Author’s computations using household expenditure per capita data from the
longitudinal subsample of FIES 2003, 2006 and 2009.
Note: GE(a) – is the generalized entropy index for measuring inequality. The index is
more sensitive to changes in the lower incomes for lower values of a and it is more
sensitive to changes in the upper incomes for higher values of a.
One of the long-standing views about mobility is that it is a channel for making economic
opportunities more evenly distributed. In other words, income mobility is seen as equalizer of
incomes in the long-run. I begin the empirical investigation with the computation of the
stability index proposed by Shorrocks (1978). This index directly links the concept of mobility
with income inequality by providing an estimate of the relative reduction of cross-sectional
inequality achieved through mobility of incomes. Table 4.4 presents the average inequality
based on single-period incomes, the inequality of permanent incomes computed by taking the
longitudinal average income of each household and the Shorrocks’s rigidity or stability index.43
Values of this index depend on the level of sensitivity of the underlying inequality measure to
incomes in different parts of the distribution. Depending on the inequality measure being used,
the results show that about 5% to 20% of cross-sectional inequality is reduced when household
incomes are averaged from 2003 to 2009. The relative reduction in income inequality is lowest
when inequality is measured based on the Gini coefficient and it increases as one uses an
inequality measure that is more sensitive to the changes in the lower or higher income range.
43 As shown in Chapter 1, the Shorrocks’ rigidity or stability index is equal to one minus the ratio of the inequality of
longitudinally averaged incomes to the average inequality over time.
93
However, the reduced inequality is still considerably higher than the level of inequality in other
neighbouring countries, based on current estimates (Table 4.5).
Table 4.5 Comparison of Philippines’ Income Inequality

With Other Southeast Asian Countries
Country Year Gini
Cambodia 2009 0.3603
Indonesia 2010 0.3557
Lao PDR 2008 0.3674
Philippines 2003 to 2009 0.4115
Thailand 2010 0.3937
Vietnam 2008 0.3557
Source: Author’s computations using household expenditure per
capita data from the longitudinal subsample of FIES 2003, 2006,
2009 and WDI.
Notes: The estimate for Philippines is based on longitudinally-
Averaged income while the rest are based on current year incomes.
Figure 4.2 Change in the Logarithm of Income between 2003 and 2009,
by Income Percentile
1
2003 percentile
2009 percentile
2003-2009 average percentile
.5
-.5
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100
In the previous section, we have seen that the mobility patterns observed in the country
are not homogeneous. For instance, the income mobility that occurred from 2003 to 2009 is
not uniformly positive nor negative. But how does the observed income mobility vary across
the different segments of the income distribution? Have the incomes of the initially poor
households increased faster? To answer this question, I first group the initial incomes of the
household population into percentiles. I then take the average logarithmic change between the
initial and final period-incomes of all people for each percentile, as illustrated by the downward
sloping line in Figure 4.2. The results suggest that the majority of the richest 40% of the
household population in the Philippines experienced income reduction from 2003 to 2009 and
94
those who experienced larger income increases started with lower initial incomes. In fact, if I
regress the logarithmic change from 2003 to 2009 on the percentile of income in 2003, I find
that 10 higher percentile points on the 2003 income distribution is associated with a 0.9
percentage points lower average income growth from 2003 to 2009. Whether this implies that
the poor benefitted more from the economic growth or not merits further investigation.
However, the observed income changes may either be driven by changes in permanent or
transitory income. In other words, if some of the households in 2009 experienced transitory
shocks in their income, the income fluctuations may have placed them below or above their
permanent (that is, steady-state) income. Khor & Pencavel (2008) argued that those who were
below their steady-state income in the final period were more likely to have experienced
smaller income increase between the initial and final periods, while the opposite holds for those
who were above their steady-state income in the final period. This pattern is illustrated by the
upward sloping line in Figure 4.2 wherein the average logarithmic change in income is plotted
with respect to the income percentile in 2009. In other words, income growth from 2003 to
2009 was lower for households with lower income in 2009. The results of the regression
model(s) depicted in Table 4.6 suggest that 10 higher percentile points on the 2009 income
distribution is associated with a 0.6 percentage points faster income growth from 2003 to 2009.
To address the issue, I also compute the income growth for households grouped
according to their average income throughout the observation period. Obviously, using
incomes averaged over a six-year period as a measure of permanent income is not without
question. Nevertheless, it is still helpful to use this approximate measure of steady-state income
when answering whether the observed income growth pattern benefitted the poor more than
the rich. The solid line in Figure 4.2 exhibits a slightly downward pattern for the bottom 20%
of the population. Thereafter, there is not a dominant positive or negative slope. This is
consistent with the regression estimates in Table 4.6, which suggest that 10 higher percentiles
on average income is associated with only 0.01 percentage points lower income growth from
2003 to 2009.
The results provided in Figure 4.2 are based on observed income growth averaged across
all units within each percentile. However, units within each percentile have varied income
growth experiences. Figure 4.3 plots the minimum and maximum income growth for each
percentile where the percentile is based from the average of incomes over the six years. The
typical lowest income growth is approximately 17% annual income reduction, while the typical
highest income growth is approximately 20% annual income increase. While these variations
in income growth experiences are generally independent of income position, there is slight
95
evidence suggesting that the ultra-rich tend to experience both the highest and lowest income
growth.
Table 4.6 Regression Estimates of the Relation between Changes
in Income and Percentiles of Income
2003-2009
(I) (II) (III)
Percentile of initial -0.0009***
income
Percentile of final 0.0006***
income
Percentile of averaged
income -0.0001***
0.0536*** -0.0259*** 0.0143***
Intercept
R2 0.1197 0.0586 0.0029
Figure 4.3 Range of Values of the Change in the Logarithm of Income between 2003
and 2009, by Percentile of Average Income
4
minimum
maximum
-2
-4
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100
percentile of average income
Overall, the panel data suggest that incomes of poor people are increasing slightly faster
than those who were initially non-poor. However, the advantage does not seem to be
remarkable and more effort is needed for the poor to be able to catch-up. Moreover, the
considerable proportion of the household population experiencing negative mobility tend to
offset the inequality-reducing effect of faster income growth among the poor. This observation
can be further validated when I decompose the changes in inequality levels from 2003 to 2009
into pro-poor and re-ranking components following the approach proposed by Jenkins & Van
96
Kerm (2009). In particular, suppose the change in inequality from time t to t+1 is denoted by
ΔG(s). As discussed in Chapter 1, Jenkins & Van Kerm (2009) showed that this can be
expressed as the difference between the amount of income re-ranking and pro-poor income
dynamics that transpired during the observation period. The results are presented in Table 4.7.
The first two rows correspond to the values of the S-Gini index for 2003 and 2009, respectively.
The third row corresponds to the contribution of income mobility or redistribution to the change
in inequality while the last row corresponds to the contribution of the progressivity of growth
or the extent to which growth benefits the poor. The results show that while many of the initially
poor households experienced improvements in terms of the share of income held, this is offset
by income re-ranking that contributes to wider income gaps. The consequence is the observed
slow pace of the reduction in income inequality.
Table 4.7 Decomposition of Change in Inequality into

Re-ranking and Pro-poor Components
v= 1.5 v=2 v=3 v=4
Initial S-Gini 0.298 0.428 0.551 0.612
Final S-Gini 0.294 0.422 0.54 0.598
Change -0.004 -0.007 -0.011 -0.014
R-component 0.058 0.071 0.078 0.082
P-component 0.062 0.078 0.09 0.096
Kakawani
2.38 2.98 3.433 3.67
index
As pointed out at the beginning of this section, the estimates provide weak evidence that
household incomes have increased significantly, which suggests a stagnant income
distribution. To summarize the findings in this section, I find that income persistence or income
immobility is indeed strong. For instance, the numbers presented in Table 4.2 suggest that about
half of the household population who were in extreme poverty in 2003 remained in the same
status in 2009, while some 23% of the households that were classified rich in 2003 remained
in the same position six years after. Stronger income status persistence emerge when one looks
at the transition matrix based on income quintiles. In most of these cases, the entries in the
main diagonal tend to be the largest in any row, suggesting substantial immobility.
Nevertheless, the distribution of household income in the Philippines is much more dynamic
than conventionally perceived based on the growth of average per capita household income. In
both absolute and relative terms, I find that a considerable number of initially poor people have
managed to move out of poverty while some initially rich people experienced negative income
97
movements. In addition, a considerable number of middle class households have either fallen
into poverty or have managed to move up the income ranks. Remarkably, for a given number
of people that experienced positive income movements at any point in the income distribution,
a commensurate number of people observed negative income movements. This behaviour
contributed to a slow pace of improvement in average incomes. However, income mobility is
contributing to a gradual reduction in long-run inequality. For instance, I find that those who
are in the bottom 20% experienced slightly better mobility. However, for the remainder of the
population, the observed mobility has a mean-reversion effect, which suggests that a significant
portion of the observed mobility is driven by transitory fluctuations. I turn to this issue in the
next section.
4.3 Discussion
Is the observed income mobility driven by permanent income dynamics or transitory
income fluctuations? For instance, it is possible that the global financial crisis that began in
2008 might have caused a transitory income shock, especially for the richest segment of the
population. This would explain why many rich people experienced significant income declines
in 2009. To answer this question, I replicate Table 4.2 using 2003-2006 and 2006-2009 as
reference periods, which allows us to examine income mobility before and after the global
financial crisis. The results show that income persistence was stronger before the crisis
especially in the bottom and top income tiers. Unlike the poor and rich people, middle income
people were less mobile from 2006 to 2009. It is possible that during this period, middle income
people were using their savings as buffer against the economic shocks.
From 2003 to 2006, the country’s economy measured in terms of GDP per capita
increased by an annual rate of 2.04%. Despite this, (absolute) poverty increased during this
period. Several studies point to potential contributing factors. For instance, Reyes et al. (2011)
noted that the family size of chronically poor households increased faster than their real
incomes. Consequently, the reduced income per member pushed these households into more
severe poverty. Estimates from labour force survey also reveal that labour outcomes seem to
have deteriorated during this period. In particular, labour participation rate decreased from
66.7% in 2003 to 64.2% in 2006 (WDI 2013) while the proportion of employed persons
working less than 40 hours per week rose by 1.5 percentage points based from my estimates
using the labour force survey. The share of the employed population engaged in unpaid family
employment also increased from 10% to 11.5% over the three year period. Because households
of lower economic status rely mostly on income from labour, it is not surprising to note that
98
the inferior job quality has been accompanied by increased poverty. Santos (2008) surmised
that the expanded coverage of the value tax, which began in 2005, may have also resulted in
higher poverty rates as the resulting inflation reduced the real income of the poor. Virola (2008)
identified climate-induced shocks as another factor contributing to the downward mobility
experienced by households in the low-income range.
The results presented in Tables 4.8 and 4.9 portray a stronger income persistence at the
top of the income pyramid before the financial crisis. This may be attributed to the fact that
much of the economic growth observed during this period occurred in high productivity sectors
while the growth in the agriculture sector where the poor households are concentrated remained
sluggish (Canlas et al. 2009).
From 2006 to 2009, GDP per capita increased at an annual rate of 2.19%, slightly higher
than the 2003-2006 growth rate. Relative to 2003-2006, poverty persistence was significantly
lower during this latter period. This outcome may be partially attributed to the expansion of
poverty reduction efforts by the national government. For instance, the coverage of the
Conditional Cash Transfer program expanded from covering 161 municipalities in 2008 to 277
municipalities in 2009 (Virola 2008). Like the poor, middle income households also
experienced more positive income mobility prospects. This may also be explained by the
improvement in quality of employment in various labour market indicators during this period.
For instance, vulnerable employment decreased from 44.5% in 2003 to 42.6% in 2006 (WDI
2014), while the proportion of employed persons working less than 40 hours per week
decreased from 38% in 2006 to 35.4% in 2009. Wages, particularly in the government sector,
also increased significantly during this period with the enactment of the Salary Standardization
Law.44 However, the gains observed during this period were partially offset by the food price
and global financial crises in 2008. In particular, the significant increase in food prices in 2008
pushed vulnerable households into poverty while the global financial crisis depressed the value
of assets of middle income and rich households (Yap, Reyes & Cuenca 2009). Nevertheless,
compared to other countries, the Philippines had shown more resilience to the adverse impact
of the 2008 global financial crisis (ADB 2012a)45; which may explain why low to middle
44 The Salary Standardization Law increased government employees’ salary to make it at par with their counterparts who are
doing similar jobs in the private sector (TUCP 2009).
45A report from ADB (2012a) identified several factors why the Philippines was not significantly affected by the financial
crisis relative to other countries, especially those which also have a large number of households with at least one (international)
migrant family member. First, ADB (2012a) surmised that since Filipino migrants were spread across the world, the negative
impact of the crisis was less stark than in other countries whose migrant workers were mainly concentrated on the heavily
affected countries. Second, the buoyant demand for Filipino workers from a broad range of fields such as domestic services,
health care, engineering and computer hardware and software development, lead to an increased deployment of Filipino
migrant workers even during the financial crisis.
99
income households still experienced better income mobility prospects from 2006 to 2009 than
the 2003-2006 period.

2006
low upper
middle middle rich
income income
extreme
poverty 0.6730 0.2757 0.0497 0.0016 0.0000 0.0000
moderate
2003
poverty 0.2591 0.5033 0.2207 0.0164 0.0004 0.0000

low middle
income 0.0331 0.2219 0.5840 0.1564 0.0041 0.0006
middle income 0.0035 0.0258 0.2801 0.5887 0.0923 0.0096
upper middle
income 0.0000 0.0000 0.0313 0.4975 0.3845 0.0868
rich 0.0000 0.0000 0.0000 0.2271 0.4458 0.3271
2006 and 2009.
Notes: Extreme poverty - US$1.25/day or lower; moderate poverty - US$1.25 to US$2; lower middle income – US$2 to US$4;
middle income – US$4 to US$10; upper middle income –US$10 to US$20; rich - exceeding US$20/day (rich).

2009
low upper
middle middle rich
income income
extreme
poverty 0.5581 0.3518 0.0893 0.0008 0.0000 0.0000
moderate
2006
poverty 0.1687 0.4752 0.3361 0.0197 0.0002 0.0000

low middle
income 0.0171 0.1622 0.6295 0.1845 0.0064 0.0003
middle income 0.0010 0.0170 0.2446 0.6373 0.0945 0.0057
upper middle
income 0.0000 0.0000 0.0490 0.5305 0.3564 0.0641
rich 0.0000 0.0000 0.0000 0.2704 0.5089 0.2207
2006 and 2009.
Notes: Extreme poverty - US$1.25/day or lower; moderate poverty - US$1.25 to US$2; lower middle income – US$2 to US$4;
middle income – US$4 to US$10; upper middle income –US$10 to US$20; rich - exceeding US$20/day (rich).
While household income distribution in the Philippines is much more dynamic than
conventionally perceived, income mobility has been lacklustre compared with other
developing countries. Latin American countries such as Brazil, Chile, Colombia and Costa
Rica also have high levels of income inequality. However, in the past 15 years, these countries
100
have experienced significantly more upward mobility (Ferreira et al. 2013). In contrast, income
mobility in the Philippines has been characterized by offsetting forces of upward and
downward mobility. While high economic growth contributed to upward mobility of some
households, many man-made and natural crises and the lack of inclusive growth have pushed
a large number of households into poverty. This finding is consistent with findings from
previous studies suggesting that the country has been ineffective in minimizing poverty inflows
during income shocks because of the lack of social protection services and inefficient
redistribution policies (Manasan 2009; Balisacan et al. 2010).
4.4 Summary
Recent estimates of macroeconomic indicators paint a vibrant Philippine economy. For
instance, from 2009 to 2012, real GDP per capita grew by 4.1% annually. In the first quarter
of 2013, total GDP expanded by 7.8%, faster than China (7.7%), Indonesia (6%), Thailand
(5.3%) and Viet Nam (4.9%) (NEDA 2013). This outcome is in sharp contrast to the economic
contraction that the country experienced in the 1980s and much higher than its modest growth
performance in the 1990s and first half of 2000s. However, the rapid economic growth has yet
to be manifest in the distribution of household income in the Philippines based on conventional
indicators. For instance from 2009 to 2012, average real per capita household income barely
moved (NSCB 2013b). At the same time, income inequality has remained persistently high
over the past decade (ADB 2012b). Together, these indicators portray a stagnant household
This chapter has investigated this puzzle by examining the mobility patterns of household
incomes in the Philippines. The contribution of examining household income dynamics in the
Philippines is twofold. First, studying household income mobility addresses the limitations of
the conventional indicators that focus only on the features of the marginal distributions and
how they change over time. However, marginal distributions usually fail to provide a good
representation of the dynamic features of a country’s growth process. Second, measuring how
much income mobility is present is important in gauging the impact of economic growth on
living standards.
The analyses provide a broad snapshot of the income mobility patterns in the Philippines
from 2003 to 2009. The investigation provides three main findings. First, the distribution of
household incomes in the Philippines is much more dynamic than conventionally thought. For
instance, about 72% of the population moved into a different income decile and 53% moved
101
into a different quintile during the observation period. The average annual percentage change
of income per capita is about 21%. Nevertheless, income persistence is strong. For instance,
the correlation between the ranks in 2003 and 2009 is approximately 0.8. Second, the observed
income mobility is characterized by high positive and negative mobility rates. Interestingly,
the patterns of positive mobility are quite symmetric with the patterns of negative mobility
throughout the income distribution. In other words, for every person who experienced an
income increase at any point in the income distribution, there is another individual (or a
commensurate number of individuals) who experienced an income decline. This finding means
that the income gains experienced by a significant number of Filipinos during this period of
economic of growth has been neutralized by the income reductions experienced by others. This
offsetting of positive and negative income mobility heavily contributes to the static nature of
the indicators of the income distribution at the aggregate level. Third, the empirical
investigation also reveals that a non-negligible portion of the observed mobility is driven by
fluctuations in the transitory component of income. This puts into question the sustainability
of the observed economic mobility.
The findings from this chapter serve as roadmap for the remainder of this study. First, it
is operationally useful to differentiate persistent and transient poverty to be able provide
optimal intervention programs for the varying needs of the poor. Thus, Chapter 5 examines the
duration of poverty experiences of Filipinos. Second, while I find evidence that the country’s
poor have experienced slightly better income mobility prospects, an important task for policy-
targeting is to be able examine the profile of upwardly and downwardly income mobile
individuals. Chapter 6 discusses this topic. Third, a good understanding of the factors that
contribute to income mobility is instructive for devising policies that would distribute the
benefits of economic growth more equitably. Thus, Chapter 7 identifies the proximate
determinants of income mobility in the Philippines. Fourth, as households in the Philippines
heavily rely on earnings from employment, there is a need to investigate the status of quality
of employment in the country and its role in fostering more sustainable positive mobility. An
examination of the effectiveness of existing social safety nets in minimizing economic
vulnerabilities is also warranted as the results of this chapter show that both positive and
negative mobility are common features of the country’s income dynamics. This topic is
examined in Chapter 8. Fifth, it is operationally useful to provide a longitudinal perspective
when examining the evolution of income or any other measure of living standards. In general,
while static indicators of development are useful for a quick analytical assessment of economic
progress, they hide a number of important features of the underlying development process.
102
However, to be able to provide a longitudinal perspective, panel data that track income or other
measures of living standards of the same set of individuals are needed. However, many
developing countries do not have adequate panel data because it is often costly to collect. The
pseudo-panel estimation approach discussed in Chapter 9 is a good example of a methodology
for exploiting the information available from repeated cross-sectional survey data more
optimally to answer questions that are conventionally answered by longitudinal data.
103
Appendix 4.1 Mobility of Household Incomes
In this section, I briefly examine the mobility patterns in the Philippines using household
income data. There are some advantages in using household income rather than expenditure to
measure poverty, inequality and mobility. For instance, it is arguably easier to link income with
socio-economic policies because income can be decomposed into components such as earnings
from employment and receipts from government transfers -- things that concern socio-
economic planners. Furthermore, it is also possible to measure poverty, inequality and mobility
at finer levels of disaggregation using income data collected from administrative systems.
However, using income as a welfare measure has disadvantages too. For instance, since income
is generally more sensitive to erratic fluctuations (Jefferson 2012), it is intuitive to expect more
mobility using household income data than expenditure-based estimates. Additionally, the
proportion of people who are classified as poverty-vulnerable tend to be higher using income-
based measures. Nevertheless, this conclusion is not impeccable as there are studies that found
more mobility using expenditure data (Gradin, Canto & del Rio 2008). The objective of this
section is to briefly compare the differences in mobility patterns between household
expenditure and income in the Philippines.
Appendix Figure 4.1 presents the mobility curve based on household income data. The
mobility pattern is remarkably similar when it is compared with the expenditure-based mobility
curve shown in Figure 4.2. The upward mobility curve presented in Appendix Figure 4.1 is
generally symmetric with the downward mobility curve which implies that downward mobility
offsets upward mobility. As expected, I also observe higher mobility rates using income data
as evidenced by higher values of Foster & Rothbaum’s (2012) mobility indices. This is also
confirmed from the various mobility indicators presented in Appendix Table 4.2.
Using the same income thresholds to distinguish extreme poverty, moderate poverty, low
middle income, middle income, upper middle income and rich, I replicate the absolute mobility
matrix presented in Table 4.2 using household income. The results are shown in Appendix
Table 4.1. About half of the people living in extreme poverty in 2003 still remained extremely
poor in 2009, 35% moved to moderate poverty while the rest entered middle income status.
This is approximately the same with the estimate that I calculated using expenditure data.
Interestingly, income persistence among the rich seems to be stronger when using income data.
In particular, 31% of the initially rich in 2003 remained rich in 2009 which is higher than the
23% estimated using expenditure data.
104
Appendix Figure A0.1 Mobility Curve Using Household Income Data, 2003-2009
.1
M*u = 329.31
M*d = 302.00
0
-.1
0 1200 2400 3600 4800 6000 7200 8400 9600 10800 12000
values of c
Source: Author’s computations using household income per capita data from the longitudinal subsample
of FIES 2003, 2006 and 2009.
Appendix Table A0.1 (Absolute) Household Income Transition Matrix, 2003-2009

2009
low upper
middle middle rich
income income
extreme
poverty 0.4952 0.3530 0.1406 0.01 0.0000 0.0012
moderate
2003
poverty 0.2257 0.4047 0.3147 0.0536 0.0012 0.0000

low middle
income 0.0553 0.1927 0.5288 0.2126 0.0090 0.0016
middle income 0.0052 0.0359 0.2893 0.5384 0.1156 0.0156
upper middle
income 0.0000 0.0030 0.0411 0.5056 0.3616 0.0887
rich 0.0000 0.0000 0.0366 0.3163 0.3359 0.3113
Source: Author’s computations using household income per capita data from the longitudinal subsample of FIES 2003,
2006 and 2009
105
Appendix Table A4.2. Selected Indicators of Mobility, 2003-2009
Income mobility indicator Income Expenditure
Average number of vingtiles 2.91 2.77
moved (non-directional) 0.04 0.03
Average number of vingtiles 0 0

moved (directional) 0.06 0.05
Proportion of population 0.14 0.15

remaining in leading diagonals 0.005 0.005
Proportion of population moving 0.11 0.11

one vingtile up 0.005 0.005

one vingtile down 0.005 0.005

two vingtiles up 0.004 0.004

two vingtiles down 0.004 0.004
Proportion of population moving at 0.23 0.23

least three vingtiles up 0.006 0.006
Proportion of population moving at 0.23 0.22

least three vingtiles down 0.006 0.006
Correlation of income ranks 0.77*** 0.8***

Average absolute change 672.61 492.68
|Income2009 - Income2003| 20.04 13.69
Average absolute percentage 0.49 0.41
change |Income2009 -
0.006
Income2003|/Income2003 0.01
Average income change 25.98 33.07
(Income2009 - Income2003) 22.11 15.32
Average percentage change 0.21 0.16
(Income2009 -
0.01 0.008
Income2003)/Income2003
Source: Author’s computations using household income and expenditure per capita data from the
106
Chapter 5 How Long Do the Poor Stay in Poverty?
5.1 Introduction
Official headline statistics of poverty in the Philippines are usually presented as a cross-
sectional snapshot picture of disadvantage (NSCB 2013a). These static measures describe
poverty in the country as a one-time event, and ignore the persistence and recurrence of poverty
over time. However, the poor are not all the same: some experience disadvantage as a
temporary once-off event, others experience recurring episodes of disadvantage, and others
endure longer spells of socio-economic deprivation and the associated long-lasting grip of
social exclusion. Overall, emerging from poverty is not a simple rags to riches story because
moving out of poverty today does not completely remove the risk of falling back into economic
dearth in the future. This chapter further probes the results from Chapter 4 suggesting that
households with lower incomes in the Philippines have experienced slightly better income
mobility prospects than the rest of the population but this may not be enough to set forth a
virtuous cycle of development that will permanently improve the lives of more than four
million poor Filipino households (NSCB 2013b) and put the Philippines on-track of its goal of
becoming the next Asian Tiger. This is done by examining the factors that characterize
intertemporal income poverty, i.e., duration and frequency of episodes that households spend
below the poverty line.
A household’s intertemporal poverty experience can either be persistent (chronic) or
transient. Persistent poverty could be further transmitted across generations giving way to
vicious cycles of socio-economic hardship while transient poverty is characterized by a
household’s inability of meeting its minimum basic needs from time to time. High levels of
persistent poverty are harmful to a country’s long-term growth prospects and thus, governments
and international development community alike, aim to eradicate persistent poverty. This
commitment is exemplified in the MDGs which identify eradication of extreme poverty as its
first objective (UN 2014). Nevertheless, tackling transient poverty is also important because it
can also slow down a country’s growth momentum (Jalan & Ravallion 2000). When outlining
intervention programs, it is important that policy planners differentiate between persistent and
transient poverty as the policy interventions necessary to ameliorate these different types of
disadvantage may be very different. For instance, providing social safety nets to minimize the
adverse impact of socio-economic shocks may not be an optimal strategy to reduce
disadvantage amongst people who have experienced uninterrupted, even intergenerational,
107
spells of disadvantage. Similarly, providing long-term social assistance to those who are
transiently poor may not be cost-effective. The key message is that no poverty and disadvantage
reduction programs will serve as a “one-size-fits-all” policy lever, and hence probing beyond
cross-sectional measures of poverty and examining its intertemporal patterns can guide
efficient and cost-effective policymaking decisions.
Given the importance of examining intertemporal poverty for policy-making, several
studies have attempted to incorporate a longitudinal perspective in the analysis of poverty in
the Philippines using either actual panel or pseudo-panel data (Balisacan & Pernia 2002;
Tabunda & Albert 2002; Albacea & Gironella 2003; Reyes et al. 2011; Bayudan-Dacuycuy
and Lim 2013). However, they fall short of examining the sensitivity of their estimates to
different methodologies, different types of poverty indices and different poverty line
specifications, to which I hereafter refer as measurement parameters. Examining the robustness
of intertemporal poverty estimates is important because previous poverty estimates highly
depend on the measurement parameters used (Kurosaki 2006). In addition, knowledge on the
robustness of estimates provides more nuanced insight into poverty and disadvantage patterns
which in turn, could help researchers to better communicate key policy messages.
The discussion in this chapter contributes to the existing poverty literature in the
Philippines by providing a broader examination of intertemporal poverty using a more
complete set of analytical tools. In particular, the chapter addresses the following substantive
and methodological questions:
Substantive
(i) Is poverty in the Philippines characterized by long episodes of poverty spells or
transitory movements around the poverty line?
(ii) What is the spatial distribution of intertemporal poverty in the Philippines?
(iii) Aside from geography, what are the characteristics of intertemporally poor?
Methodological
(iv) Are the observed patterns sensitive to the estimation parameters?
(v) How robust are the poverty rankings to the parameters used in the estimation
process?
5.2 Intertemporal Poverty in the Philippines
Like many developing countries, poverty reduction is explicitly articulated in the

Philippines’s development plan (Reyes et al. 2011). Over several decades, the country has been
working to reduce poverty by implementing a multitude of anti-poverty programs which aim
108
to improve the quality of life of the poor (Schelzig 2005; Aldaba 2009; Bayudan-Dacuycuy &
Lim 2013). From 1986 to 1992, the administration of then President Corazon Aquino
implemented three major intervention programs which include the Tulong sa Tao (Help for the
People), the Comprehensive Agrarian Reform Program (CARP) and the Community
Employment and Development. From 1992 to 1998, the Ramos administration implemented
the Social Reform Agenda which focused on the development of the twenty poorest provinces
and the poorest sectors. From 1998 to 2001, the Estrada Administration launched the Lingap
Para sa Mahirap (Care for the Poor) which provided assistance to the poorest families from
each local government unit. From 2006 to 2010, the Arroyo administration started the Kapit
Bisig Laban sa Kahirapan (KBLK) (Linking Arms Against Poverty Program) which offered
socio-economic services to the poor. The current Aquino administration is implementing the
Social Reform Program (SRP) which provides conditional cash transfers to families that satisfy
certain criteria. While most of these programs occupy the centrepiece of each administration’s
platform, each program has its own thrust. For instance, the CARP focused on agricultural
development, KBLK paid attention on improving delivery of social services (Schelzig 2005)
while SRP is spending effort on institution-building and promoting effective participation in
governance (NEDA 2011).
During the period that these programs were implemented, the Philippines noted a
reduction in poverty rates. For instance, I noted from Chapter 2 that the proportion of the
population living below US$2 a day dropped by approximately 20 percentage points between
1985 and 2009. However, it can also be noted that the pace of poverty reduction in the
Philippines is significantly lower than other countries’ pace and this process has been painfully
slow despite faster economic growth that the Philippines experienced during the past decade
(Aldaba 2009). There are several possible reasons that could explain why poverty is barely
declining amidst faster economic growth rates in the country. First, it is possible that most of
the poor have incomes far below the poverty line and it takes a very long time before the
benefits of economic growth trickle down to the persistently poor. Second, it is possible that
the poor don’t contribute much to economic growth because they are concentrated on low
productivity sectors due to their limited skills. Third, as suggested in Chapter 4, even if many
low income people are moving into and out of poverty transiently, the number of people falling
into poverty may offset the number of people moving out of poverty at any given time. These
scenarios call for different policy response; the first two should focus on long-term human
capital development programs because it will be hard for the persistently poor to get out of
poverty by solely relying on their efforts while the other scenario should be geared towards
109
minimizing recurring socio-economic vulnerabilities. However, Reyes (2002) identified that
one of the bottlenecks in policy planning that has contributed to the slow poverty reduction is
that the previous intervention programs were not well-targeted and tend to discount the
heterogeneous needs of the poor. That study argued that there was no clear guideline where
long-term human development efforts and short-term risk management initiatives should be
channelled. This issue could be attributed to the lack of intertemporal poverty data that can be
used to inform policies. As pointed out in Chapter 3, socio-economic planners in the
Philippines had to rely on poverty data collected from cross-sectional surveys for many years
and it was only in 2003 when the FIES, the data source of official poverty estimates, was
redesigned to collect nationally-representative panel data. With cross-sectional surveys,
researchers were unable to measure intertemporal poverty directly. Nevertheless, there were
some efforts to examine the intertemporal patterns of poverty even before FIES was redesigned
by analysing either pseudo-panel data (e.g., Balisacan and Pernia 2002; Albacea & Gironella
2003) or actual longitudinal data from one-shot surveys (e.g., Reyes 2002a; Tabunda & Albert
2002), however, the data limitations made it hard for these studies to generalize their results to
larger populations.
More recently, Reyes et al. (2011) and Bayudan-Dacuycuy & Lim (2013) and used the
panel data from the redesigned FIES to provide a more dynamic perspective of intertemporal
poverty in the Philippines. Both studies characterize the country’s poverty to be mostly
persistent in nature. However, while they have addressed the limitations of previous research
by providing direct survey estimates of persistent and transient poverty at the national-level,
they also have critical limitations. First, both studies employed confining methodology as they
solely relied on the spells approach only for measuring intertemporal poverty. As explained in
Chapter 1, the spells approach narrowly focuses on frequency of poverty episodes over time
and ignore the fact that people can “borrow” income from different time periods. Second, both
studies restrictively measured poverty in terms of incidence or head count of poor. Third, both
studies used the official poverty lines compiled by the government, the use of which remains
debatable because the official poverty thresholds are based on different regional food menus
which tend to make the resulting estimates inconsistent across regions (Bersales 2009).
This chapter contributes to the existing literature by measuring intertemporal poverty
using a more comprehensive set of analytical tools and measurement parameters. To do this, I
adopt the spells and components approaches as described in Chapter 1. In addition, I measure
intertemporal poverty in terms of incidence, depth and severity. Four different sets of poverty
lines are used to examine the robustness of estimates to poverty line specifications. Examining
110
robustness of intertemporal poverty estimates is important for both methodological and
substantive reasons. The existing literature suggests that the poverty estimates may vary
significantly depending on measurement parameters (Grootaert & Kanbur 1995; JR 1998,
McCulloch & Baulch 1999; Kurosaki 2006; Christiaesen & Shorrocks 2012). If robustness of
the parameter estimates is not guaranteed, it is possible that the poverty levels reported are
arbitrary leading to inaccurate interpretations and outcomes.46 For example, if the magnitude
of poverty reported is higher or stagnant than it actually is, the business community may be
dissuaded from investing in the country (Balisacan 1997).
There are several examples of poverty reduction programs in the Philippines that rely on
the reliability of poverty rankings. For example, only households living in selected
municipalities from the 20 poorest provinces are eligible for the national government’s
conditional cash transfer program during the first stages of its inception (Reyes & Tabuga
2012). Other government agencies such as PhilHealth, a government-owned corporation that
provides health insurance, as well as non-government agencies also use poverty rankings to
identify areas that need priority assistance (Addawe, Martinez & Perez 2007). With the use of
narrow estimation methodologies, it will be difficult to gauge whether these intervention
programs that rely on poverty rankings are implemented optimally.
Another important contribution of this study is to provide estimates of intertemporal
poverty at the sub-national level. Locating where the poor are is important for socio-economic
planning. For one, it makes the delivery of social services more efficient and cost-effective
(Kanbur 1987). It also helps exploit dynamic externalities and geographic spill-over effects of
economic growth (Ravallion & Jalan 1996). However, previous studies such as that of Reyes
et al. 2011 and Bayudan-Dacuycuy & Lim 2013 do not clarify whether the areas with the
highest levels of total poverty are also the same location as those areas with the highest levels
of persistent or transient poverty. This is important to know because policymakers may be more
concerned on channelling resources to areas with slightly lower levels of total poverty but with
more prevalent persistent poverty than areas with higher levels of total poverty but significantly
lower poverty persistence. Considering that persistent and transient poverty call for different
policy mix, this type of research will also improve the efficiency of poverty reduction programs
by being able to target appropriate interventions to where these are needed.
46Recently, ADB (2014) released a report suggesting that poverty in Asia increased from 1.6 billion in 2005 to 1.8 billion in
2010. This result contradicts a number of poverty assessments that were published by various international development
agencies which suggest that poverty in the region has declined. Ravallion (2014) surmised that this contradiction stems from
the measurement parameters used by ADB.
111
5.3 Methodology
To provide a more comprehensive picture of the intertemporal poverty in the Philippines

compared to previous studies, this chapter measures intertemporal poverty using the
methodologies proposed by Jalan & Ravallion (JR) (1998), Duclos, Araar & Giles (DAG)
(2010), Foster (2009) and Gradin, del Rio and Canto (GRC) (2012) as discussed in Chapter
1.47 The second part of this chapter differentiates the characteristics of the persistently poor,
transiently poor and non-poor by estimating multinomial logistic models. The dependent
variable measures the intertemporal poverty status of each household and takes two forms as
shown in (5.1). The first form evaluates poverty status using the components approach while
the second form uses the spells approach. As pointed out in Chapter 1, under the components
approach, a household is considered persistently poor if its longitudinally-averaged income
falls below the specified poverty line, transiently poor if its longitudinally-averaged income is
higher than the poverty line but at least one of its cross-sectional incomes fell below the poverty
line and non-poor if the household never experienced income shortfall (below the poverty line).
Under the spells approach, a household is considered persistently poor if at least two of its
cross-sectional incomes fell below the poverty line, transiently poor if only one of its cross-
sectional incomes fell below the poverty line and non-poor if the household never experienced
income shortfall. Each outcome of interest is regressed on several household characteristics
such as the human capital available to the household, assets held by the household, access to
basic services and geographic characteristics in the initial time period. These variables are
included in the model because as discussed in Chapter 1, higher levels of human capital and
assets, greater access to basic services and favourable geographic characteristics reduce
poverty risk. For instance, it is well-established in the literature that higher levels of education
expand socio-economic opportunities of poor Filipinos and thus, minimize the risk of staying
in poverty for a long time (Maligalig et al. 2014). As discussed in Chapter 1 too, access to
productive assets, basic services and spatial endowments can also be used to minimize the risk
of long poverty spells (WB 2004; Schelzig 2005; WB 2013). Changes in these factors are also
included as control variables in the regression models to capture for demographic and economic
events (5.1). This is further discussed in Chapter 6.
47 There are some limitations with regards to the measurement of poverty from the data. In particular, I do not have information
about the poverty status of the households before the beginning of the observation period. Moreover, since I am using survey
data conducted every three years, there is no way to know how people moved into and out of poverty in between the survey
years. For instance, a household that is classified as poor in two consecutive survey years may have been non-poor in between.
In other words, there is incomplete information about the duration of poverty because the data is censored. For simplicity, the
discussion assumes that the observation periods follow a continuum.
112
0, 𝑙𝑑 𝑖𝑠 𝑜 𝑝𝑜𝑜𝑟 where poverty status is gauged based on
(a) 𝑊 = { , 𝑙𝑑 𝑖𝑠 𝑝𝑒𝑟𝑠𝑖𝑠𝑡𝑒 𝑡𝑙𝑦 𝑝𝑜𝑜𝑟
2, 𝑙𝑑 𝑖𝑠 𝑡𝑟𝑎 𝑠𝑖𝑒 𝑡𝑙𝑦 𝑝𝑜𝑜𝑟 longitudinally-averaged income
0, 𝑙𝑑 𝑖𝑠 𝑜 𝑝𝑜𝑜𝑟
(b) 𝑊2 = { , 𝑙𝑑 𝑖𝑠 𝑝𝑒𝑟𝑠𝑖𝑠𝑡𝑒 𝑡𝑙𝑦 𝑝𝑜𝑜𝑟 where poverty status is gauged based on the
2, 𝑙𝑑 𝑖𝑠 𝑡𝑟𝑎 𝑠𝑖𝑒 𝑡𝑙𝑦 𝑝𝑜𝑜𝑟 number of episodes spent in poverty
+ 𝛽3 𝐴 𝑒𝑠𝑠 𝑡𝑜 𝑆𝑒𝑟𝑣𝑖 𝑒𝑠 + 𝛽4 𝐺𝑒𝑜𝑔𝑟𝑎𝑝 𝑦
𝑊𝑖 = 𝛽 𝐻𝑢𝑚𝑎 𝑎𝑝𝑖𝑡𝑎𝑙 + 𝛽2 𝐴𝑠𝑠𝑒𝑡 + 𝛽3 𝐴 𝑒𝑠𝑠 𝑡𝑜 𝑆𝑒𝑟𝑣𝑖 𝑒𝑠 + 𝛽4 𝐺𝑒𝑜𝑔𝑟𝑎𝑝 𝑦
+𝛽5 ∆ 𝐻𝑢𝑚𝑎 𝑎𝑝𝑖𝑡𝑎𝑙 + 𝛽6 ∆𝐴𝑠𝑠𝑒𝑡𝑠 + 𝛽7 ∆𝐴 𝑒𝑠𝑠 𝑡𝑜 𝑆𝑒𝑟𝑣𝑖 𝑒𝑠 + 𝜀𝑖𝑡 (5.1)
5.4 Empirical Results

5.4.1 Intertemporal Poverty in the Philippines
I begin with a presentation of cross-sectional estimates of poverty. Figure 5.1 plots the
headcount poverty rate or the proportion of poor people in y-axis against different low income
thresholds in the x-axis. I placed two vertical lines in the same figure to represent the lowest
and highest poverty line under consideration in this study. Depending on the poverty line used,
it is estimated that about 13% to 39% of the population can be considered poor in 2009 (Figure
5.1). Furthermore, Table 5.1 presents headcount poverty rate, poverty gap, severity of poverty
and Watts index using the four sets of poverty lines. Separate estimates are also provided for
urban and rural areas. The estimates suggest that the poverty gap or average income shortfall
is about 3% to 13% while the severity of poverty or average squared income shortfall is roughly
1% to 5% of the poverty line in 2009. If all household incomes per capita were increased by
2% per year, the estimates of Watts index suggest that poverty is expected to be eradicated
after 8 to 9 years from 2009.48 Table 5.1 also indicates that poverty in the Philippines has a
remarkable geographic feature. In particular, the proportion of poor in rural areas is about two
to four times the headcount poverty rate in urban areas. The magnitude of income shortfall and
the inequality among the poor are also significantly higher in rural areas than in urban areas.
Hence, it is not surprising to note that from 2009, it will take about 13 years for all households
in rural areas to exit US$2/day poverty compared to the 4 years needed for households from
urban areas to accomplish the same feat, assuming a uniform 2% annual income growth.
Although the numbers presented in Table 5.1 are useful for gauging how significant
poverty is in the Philippines, they provide limited information about poverty dynamics over
48 As explained earlier, the average exit time needed to exit poverty is computed by dividing the value of the Watts index by
the expected income growth rate. For example, the Watts index value in 2009 using the US$2/day poverty line is 16.64.
Dividing this by an assumed annual income growth rate of 2% will yield 8.32. This means that, under the assumption that all
incomes grow at a uniform rate of 2% per year, US$2 poverty rate will be eradicated after 8.32 years.
113
time. For instance, the slight increase in poverty from 2003 to 2006 and minimal reduction
between 2006 and 2009, do not readily imply limited movements into and out of poverty
(Reyes et al. 2011). This can be proved by examining poverty transition rates presented in
Tables 5.2 to 5.5. The rows in each table are divided into three main panels, from top to bottom,
poverty transition rates between 2003 and 2006; 2006 and 2009; 2003 and 2009 while the
columns present estimates for the national level as well as for urban and rural areas.
The estimated poverty transition rates show a large degree of poverty inflow and outflow
despite the minimal changes in the cross-sectional measures of poverty. For example, 75% of
the population with incomes below US$2/day in 2003 were non-poor in 2009 while the other
25% remained (US$2/day) poor. This highlights the importance of providing a longitudinal
perspective when examining poverty.
Poverty Estimates using the Components Approach

In this section, I present the estimates of persistent and transient poverty using the JR and
DAG approaches. As explained in Chapter 1, although both methodologies follow the
components approach of measuring intertemporal poverty, they conceptualize transient poverty
differently. The JR approach defines transient poverty as the residual when persistent poverty
is subtracted from total poverty whereas DAG links transient poverty in terms of a person’s
level of risk aversion.
Figure 5.1 Headcount Poverty Curve, 2009

1
US$1.25/day
US$2/day
.8
.6
.4
.2
0
0 5.5 11 16.5 22 27.5
poverty line (US$ per day)
FIES 2003, 2006 and 2009.
114
Table 5.1 Cross-Sectional Measures of Poverty in the Philippines, 2003-2009
2003 2006 2009
Poverty Measure
Phils Urban Rural Phils Urban Rural Phils Urban Rural
US$1.25/day Poverty Line
Headcount poverty rate 18.29 6.54 29.67 19.25 7.32 30.80 15.40 6.54 23.97
0.55 0.59 0.85 0.06 0.62 0.86 0.54 0.63 0.82
Poverty gap 4.47 1.19 7.65 4.74 1.53 7.85 3.15 1.17 5.07
0.17 0.14 0.29 0.18 0.17 0.29 0.14 0.14 0.24
Severity of Poverty 1.63 0.36 2.86 1.66 0.47 2.81 0.99 0.31 1.65
0.08 0.06 0.15 0.08 0.07 0.14 0.06 0.05 0.11
Watts Index 5.68 1.44 9.79 5.94 1.86 9.90 3.85 1.37 6.24
0.24 0.18 0.41 0.24 0.22 0.40 0.19 0.17 0.32
US$2/day Poverty Line

0.71 0.96 0.88 0.72 1.03 0.87 0.71 1.00 0.89
Poverty gap 14.24 6.04 22.19 15.12 7.00 22.99 12.53 6.03 18.83
0.30 0.33 0.44 0.31 0.36 0.44 0.28 0.34 0.41
0.18 0.16 0.28 0.18 0.19 0.28 0.15 0.16 0.24
Watts Index 19.62 7.81 31.05 20.72 9.10 31.99 16.64 7.72 25.28
0.46 0.46 0.70 0.47 0.51 0.70 0.41 0.47 0.61
0.5*Median Poverty Line

0.51 0.49 0.82 0.53 0.57 0.83 0.50 0.56 0.78
Poverty gap 3.73 0.93 6.44 3.51 1.05 5.90 2.66 0.95 4.31
0.16 0.12 0.27 0.15 0.14 0.25 0.13 0.12 0.22
0.07 0.05 0.13 0.06 0.05 0.11 0.05 0.04 0.10
Watts Index 4.70 1.12 8.17 4.33 1.26 7.31 3.22 1.11 5.27
0.21 0.16 0.37 0.20 0.18 0.34 0.17 0.15 0.29
Official Poverty Line
0.64 0.80 0.90 0.66 0.85 0.90 0.66 0.89 0.91
Poverty gap 7.15 2.75 11.42 7.44 3.12 11.63 7.24 3.53 10.83
0.21 0.22 0.34 0.22 0.25 0.35 0.22 0.26 0.33
0.11 0.10 0.18 0.11 0.11 0.18 0.10 0.11 0.17
Watts Index 9.21 3.38 14.85 9.51 3.86 14.99 9.14 4.34 13.79
0.30 0.29 0.49 0.31 0.33 0.49 0.30 0.34 0.46
2006 and 2009.
Note: Since these numbers are estimated from the longitudinal data, they are slightly different from the estimates presented
in Chapter 2 which used pseudo-panel data. The numbers in smaller font size are standard errors.
Figures 5.2 and 5.3 show the results. For both figures, the rows correspond to the
estimates for each of the four sets of poverty lines used in the computations while the columns
correspond to the different measures of poverty estimated at the national, urban and rural levels.
The height of the bars corresponds to the magnitude of intertemporal poverty wherein the dark-
shaded bars correspond to persistent poverty while the light-shaded bars correspond to transient
poverty.
115
There are several interesting features that can be drawn from the estimates based on the
JR procedure. First, most of the indicators show that poverty is mostly persistent in nature. For
instance, about 96% of the total observed US$2/day poverty is persistent and only 4% can be
attributed to the effect of transitory income shortfall. Interestingly, the relative importance of
persistent poverty declines with the poverty line. For example, if US$1.25/day poverty line is
used instead of US$2/day, the relative importance of headcount poverty persistence drops to
87%.
Table 5.2 Poverty Transition Matrix, US$1.25/Day Poverty Line

Philippines Urban Rural
Poor Non-Poor Poor Non-Poor Poor Non-Poor
2003 2006
Poor 67.21 32.79 54.84 45.16 69.86 30.14
Non-Poor 8.51 91.49 4.00 96.00 14.33 85.67
2006 2009
Poor 55.69 44.31 47.56 52.44 57.57 42.43
Non-Poor 5.79 94.21 3.30 96.70 9.02 90.98
2003 2009
Poor 49.42 50.58 35.68 64.32 52.36 47.64
Non-Poor 7.78 92.22 4.50 95.50 12.00 88.00
Table 5.3 Poverty Transition Matrix, US$2/Day Poverty Line

2003 2006
Poor 84.64 15.36 75.77 24.23 87.81 12.19
Non-Poor 15.56 84.44 10.96 89.04 24.02 75.98
2006 2009
Poor 76.14 23.86 66.57 33.43 79.91 20.09
Non-Poor 10.69 89.31 7.24 92.76 17.20 82.80
2003 2009
Poor 75.28 24.72 64.05 35.95 79.30 20.70
Non-Poor 14.61 85.39 10.47 89.53 22.24 77.76
Table 5.4 Poverty Transition Matrix, Half of Median Poverty Line

2003 2006
Poor 61.61 38.39 51.35 48.65 63.42 36.58
Non-Poor 7.30 92.70 3.51 96.49 12.01 87.99
116
2006 2009
Poor 55.58 44.42 52.26 47.74 56.32 43.68
Non-Poor 5.10 94.90 2.46 97.54 8.33 91.67
2003 2009
Poor 47.41 52.59 35.46 64.54 49.52 50.48
Non-Poor 6.74 93.26 3.84 96.16 10.36 89.64
Table 5.5 Poverty Transition Matrix, Government Poverty Line

2003 2006
Poor 71.81 28.19 61.66 38.34 74.77 25.23
Non-Poor 12.06 87.94 7.25 92.75 18.99 81.01
2006 2009
Poor 72.10 27.90 65.21 34.79 74.33 25.67
Non-Poor 11.26 88.74 7.00 93.00 17.36 82.64
2003 2009
Poor 67.07 32.93 57.00 43.00 70.02 29.98
Non-Poor 14.11 85.89 9.22 90.78 21.15 78.85
Figure 5.2 Intertemporal Poverty Estimates using the JR Approach

Headcount Poverty Rate Poverty Gap Poverty Severity
2.5
30
2
6
20
US$1.25
1.5
4
1
10
.5
0
Philippines Urban Rural Philippines Urban Rural Philippines Urban Rural

60
10
10 15 20
8
40
US$2
6
4
20
2
0

10 15 20 25
2
1.5
half of median
1
2
.5
5
0

10 20 30 40
4
government/official
10
3
2
5
1
0

2003, 2006 and 2009
117
Figure 5.3 Intertemporal Poverty Estimates using the
DAG Approach
alpha = 1.5 alpha = 2 alpha = 2.5
15
20
15
15
US$1.25
10
10
10
5
5
0
0
30
40
30
30
20
20
US$2
20
10
10
10
0
0
10
15
20
half of median
8
15
10
6
10
4
5
2
0
0
20
10 15 20 25
10 15 20
15
gov't/official
10
5
5
0
0
2003, 2006 and 2009.
The share of transient poverty also increases when we shift from poverty incidence to
poverty gap and poverty severity. In particular, about 10% to 29% of the poverty gap and 16%
to 38% of the poverty severity can be linked with transient poverty. Second, transient poverty
is more common in urban areas than in rural areas. In urban areas, transient poverty accounts
for approximately 11% to 29% of its total headcount poverty rate while only 2% to 10% of the
rural poverty can be attributed to transitory downturn in monetary fortunes.
As discussed in Chapter 1, the JR approach does not differentiate a household with a
relatively stable income flow from a household with highly fluctuating income as long as their
longitudinally-average incomes are equal. Hence, an interesting question to be asked is how
will these poverty estimates be affected when we take into account people’s risk aversion to
unstable income flows? The DAG approach is used to investigate this issue. In the following
analysis, I set the aversion parameter α = 1.5, 2 or 2.5, where higher values imply greater risk
aversion. These values imply that households are not risk-neutral.49 Figure 5.3 presents the
poverty estimates. The rows identify the poverty lines used while the columns represent the
intertemporal poverty estimates for different levels of risk aversion to income fluctuations at
49 A value of 1 for the risk aversion parameter α would imply that all households are risk-neutral.
118
the national, urban and rural level. Again the height of the dark-shaded and light-shaded bars
correspond to the magnitude of persistent and transient poverty, respectively. At α = 1.5 and
using US$2/day poverty line, the height of the light-shaded bar suggests that households would
be willing to increase the average poverty gap by 1 percentage point just to remove the
variability in the income shortfall observed from 2003 to 2009. If I divide the height of the
light-shaded bar by the total height of the stacked bar, I find that the estimated share of transient
poverty based on the DAG approach is roughly 5% to 7% of the total poverty when α= 1.5 and
increases to 8% to 9% when α= 2.5. In other words, as households become more risk-averse,
transient poverty becomes more common.
Poverty Estimates using the Spells Approach

Table 5.6 presents all possible poverty spells in years 2003, 2006 and 2009 with respect
to different poverty lines. The results show that about 46% to 76% of the population never
experienced poverty, 12% to 15% were poor for only one survey year, 9% to 13% were poor
for two years and 6% to 28% were consistently poor for all three years. Following the
conventional spells approach, these numbers may also be used to estimate persistent and
transient poverty. If I define poverty persistence as being poor for at least two survey years, the
estimates presented in Table 5.6 suggest that between 16% to 41% of the population were
persistently poor. Despite the differences in the methodologies, it is interesting to note that the
spells-based estimates of poverty persistence are roughly the same as the estimated proportion
of population who were persistently poor based on the JR approach.50 Both methodologies
suggest that poverty is not just once-in-a-lifetime experience wherein a significant fraction of
the population experience poverty for a long time period. However, a different story emerges
when I compare the estimates of total poverty. Following the spells approach, total poverty
would correspond to those who were poor for at least one time period. Given this definition,
total poverty is estimated to be approximately 29% to 54% of the population. These numbers
are significantly higher relative to the estimates of total poverty based on the components
approach, which placed poverty at around 18% to 41%. Consequently, the relative importance
of persistent and transient poverty also differ across these two estimation approaches. In
particular, the results based on the conventional spells approach suggest that approximately
58% to 77% of the total (headcount) poverty observed is persistent, whereas the components
50The estimated proportion of population who are persistently poor based on the JR approach is 15% using the US$1.25
poverty line and 40% using the US$2 threshold.
119
Table 5.6 Population Share by Intertemporal Poverty Status
Poverty Status $1.25 poverty line $2 poverty line
(2003, 2006, 2009) Phils Urban Rural Phils Urban Rural
7.75 1.91 13.4 28.33 12.33 43.84
P,P,P
0.39 0.35 0.67 0.65 0.79 0.90
4.55 1.67 7.33 5.97 4.1 7.78
P,P,NP
0.29 0.29 0.48 0.33 0.45 0.48
1.29 0.42 2.14 2.18 1.56 2.78
P,NP,P
0.16 0.16 0.28 0.20 0.27 0.29
4.7 2.53 6.81 4.05 3.7 4.39
P,NP,NP
0.29 0.37 0.44 0.27 0.42 0.34
2.97 1.57 4.33 4.83 4.32 5.32
NP,P,P
0.25 0.31 0.40 0.33 0.53 0.41
3.98 2.17 5.74 4.42 4.26 4.58
NP,P,NP
0.27 0.34 0.41 0.30 0.48 0.34
3.38 2.64 4.11 3.86 3.87 3.84
NP,NP,P
0.28 0.42 0.38 0.30 0.48 0.36
71.37 87.08 56.15 46.37 65.86 27.47
NP,NP,NP
0.65 0.81 0.91 0.73 1.11 0.79
Half of median poverty line Gov't poverty line
Phils Urban Rural Phils Urban Rural
5.91 1.33 10.34 15.5 6.12 26.36
P,P,P
0.35 0.28 0.61 0.54 0.60 0.83
3.58 1.07 6 4.05 1.93 6.24
P,P,NP
0.26 0.23 0.45 0.27 0.30 0.43
1.39 0.33 2.42 2.72 1.32 4.16
P,NP,P
0.16 0.13 0.28 0.23 0.26 0.37
4.52 1.95 7 5.2 3.68 6.83
P,NP,NP
0.28 0.32 0.45 0.30 0.42 0.43
2.8 1.67 3.88 5.02 3.48 6.68
NP,P,P
0.26 0.34 0.39 0.32 0.47 0.44
3.38 1.67 5.03 4.13 3.19 5.17
NP,P,NP
0.25 0.31 0.39 0.27 0.39 0.37
2.91 1.99 3.81 5.69 5 6.52
NP,NP,P
0.25 0.35 0.36 0.34 0.51 0.44
75.53 89.98 61.52 61.08 80.29 44.03
NP,NP,NP
0.62 0.73 0.90 0.72 1.02 0.88
FIES 2003, 2006 and 2009.
Note: P - Poor, NP - Non-Poor. The numbers in smaller font size are standard errors.
120
approach suggests that persistent poverty account for 84% to 96% of total (headcount)
poverty.51
As pointed out at the beginning of this chapter, the concepts of persistent and transient
should not be used interchangeably because they are measuring different aspects of socio-
economic disadvantage. Measuring persistent and transient poverty at the national-level is the
first step to understand the dynamic feature of poverty in the country. Overall, the results
presented in this section suggest that poverty in the Philippines can be considered to be mostly
persistent in nature. Nevertheless, the magnitude of transient poverty is not negligible. The
findings of the robustness checks point that the share of transient to total observed poverty
tends to increase dramatically as I increase the poverty line and/or use a poverty index that is
more sensitive to the illfare of the poorest of the poor.
5.4.2 Where are the Persistently and Transiently Poor?

This section revisits the spatial distribution of intertemporal poverty in the Philippines.
Similarly to the previous section, estimates are presented for both the components and spells
approaches. This allows us to evaluate the robustness of spatial poverty rankings across the
methodologies considered in this study. Although this section does not provide more
disaggregated estimates beyond the regional-level due to sample size restrictions, the analysis
identifies proximate areas that could be potentially useful for the targeting of poverty-reduction
programs.52
Components Approach
Figure 5.4 shows the spatial distribution of intertemporal headcount poverty while Figure
5.5 also presents the estimates for intertemporal poverty gap and poverty severity for the 17
geographic regions in the Philippines using the JR approach. The columns represent the
estimated incidence, depth and severity of poverty while the row correspond to the different
poverty lines used. One take on the results is that, when longitudinal average income data is
used, it provides evidence that is consistent with an orthodox view that most of the persistently
poor live in the regions with the lowest average per capita income. In particular, I find that
51 Had I defined persistent poverty as those who experienced poverty for two consecutive survey years, the relative importance
of persistent poverty would be approximately 50% to 73%.
52 A more informative exercise is to provide estimates at the administrative level (e.g., provinces or municipalities). However,
the small sample sizes may lead to high standard errors of intertemporal poverty estimates. In this context, there are several
small area estimation techniques that can be used (Elbers, Lanjouw & Lanjouw 2003; Martinez 2013; Martinez, Lucio &
Vilaruel 2014), etc. This is reserved for future research.
121
Regions 4B, 9, Caraga and ARMM have the highest prevalence of total poverty and poverty
persistence. Not surprisingly, these regions have the lowest longitudinally-averaged income
per capita (Appendix Table A5.3). On the other hand, NCR, the region with the highest
longitudinally-averaged income, posted the lowest total poverty and poverty persistence rate.
Other regions with low poverty persistence are also from the northern Philippines which
include Regions 3 and 4A.
Focusing on the robustness of the results, I find that the regional ranking of poverty
persistence is relatively uniform across varying measurement parameters. Nevertheless, there
are slight changes in the rankings that are worth pointing out when measurement parameters
are changed. For example, the magnitude of persistent poverty tends to be higher in Region 9
when lower poverty lines or poverty measures that are more sensitive to the ill fare of the
poorest of the poor are used. The opposite happens for Caraga wherein poverty persistence
decreases when either lower poverty lines or more inequality-sensitive measures are used.
A high level of transient poverty indicates how erratic a household income flow is over
time. Understanding how erratic income flows create poverty cycles in certain areas is
important to be able to prevent economic stagnation. Thus, like persistent poverty, it is also
instructive to examine the regional distribution of transient poverty. Interestingly, a separate
examination of transient poverty reveals a very different picture compared to what the estimates
of poverty persistence portray. Both Figures 5.4 and 5.5 highlight that it is not necessarily the
case that regions with the highest persistent poverty are also the same regions with the highest
transient poverty. Furthermore, in contrast to poverty persistence wherein the regional ranking
is quite robust under various measurement parameter specifications, the regional ranking with
respect to transient poverty heavily depends on the poverty line and type of poverty measure
used. For example, NCR ranks third among the regions with the highest proportion of
households that are transiently poor with respect to the US$2 poverty line but it ranks bottom
with respect to other measures of poverty persistence. On the other hand, ARMM posted the
lowest transient poverty with respect to US$2 poverty threshold.53
When aversion to income fluctuations is taken into account by using the DAG approach
the same regions with high and low persistent and transient poverty can be identified (Figure
5.6).
53As pointed out in Chapter 1, the JR approach does not ensure that the total poverty is always greater than or equal to poverty
persistence. This happens in the case of Region 7 and ARMM wherein the computed US$2 transient poverty rates are negative.
Appendix Table A5.3 provides adjusted estimates which ensure that all poverty estimates are non-negative.
122
Spells Approach
Figure 5.7 shows the estimates of selected indicators of intertemporal poverty following the
spells approach. In particular, the first column presents the duration-adjusted poverty
headcount ratio proposed by Foster (2009) while the second and third columns present selected
indices from the class of intertemporal poverty measures proposed by GRC (2012). As
explained in Section 1.4.1, the intertemporal poverty measures proposed by Foster (2009) and
GRC (2012) address the limitations of conventional spells-based measures that narrowly focus
on headcount poverty rate and are insensitive to poverty duration. It is clear from Figure 5.7
that Regions 7, 9, ARMM and Caraga have the highest intertemporal poverty estimates. In 7
out of 12 indicators, ARMM ranked first among the regions with the highest levels of poverty.
For instance, roughly 55% to 90% of the household population in ARMM have experienced
poverty for at least one episode from 2003 to 2009. Interestingly, except for the US$2/day
poverty line, ARMM is not in the list of four regions with the highest average duration of
poverty. Nevertheless, the average duration of poverty is still relatively high because a typical
poor household in ARMM experienced poverty in two out of the three survey years. Like
ARMM, Region 9 has widespread poverty that persists over the years. In fact, the estimates
show that Region 9 posted the highest average poverty duration wherein households spend 2.1
to 2.5 (survey) years living with income less than the poverty line. In contrast, Regions 3, 4A
and NCR have consistently shown the lowest intertemporal poverty levels.
In summary, poverty in the Philippines has a remarkable spatial feature. In general, most
of the regions with the highest levels of chronic or persistent poverty are in the southern part
of the Philippines: Regions 7, 9, ARMM and Caraga. This is in sharp contrast with the regions
that posted the lowest levels of poverty persistence which are all located up north, close to the
country’s centre of commerce. As I investigate the economic structure of each region, I find
that the areas which heavily rely in the agriculture sector based on share of agricultural
employment to total employment tend to have higher proportion of persistently poor
households. This is consistent with the findings from a number of studies that have concluded
poverty in the Philippines a predominantly rural and agricultural phenomenon (Schelzig 2005;
Aldaba 2010). There are several substantial explanations for this. First, the structure of
agriculture is predominantly small-scale and thus, those who are agriculture-dependent are
unable to take advantage of economies of scale. Second, the Philippines has significant
123
Figure 5.4 Map of Persistent and Transient (Headcount) Poverty
Persistent Poverty Transient Poverty
0.57 – 5.28 0 – 0.67

5.28 – 10.88 0.67 – 1.19
10.88 – 18.95 1.19 – 1.52
18.95 – 23.16 1.52 – 2.14
23.16 – 31.45 2.14 – 2.51
FIES 2003, 2006 and 2009.
124
gov't/official half of median US$2/day US$1.25/day
0
0
0
0
10
20
30
40
50
10
20
30
40
20
40
60
80
10
20
30
40
C A A A
ar R R R
a R MM M
A ga eg C M R MM
eg
R R R a R
R MM eg ion eg ra eg ion
e io 9 io ga io 9
R gio n n n
eg n 4 R 4- 4
io 9 C -B eg B C -B
n a a
R 4- R rag R ion R rag
e B eg a eg 9 eg a
R gio R io R ion R io
eg n eg n e 5 eg n
io 5 io 7 R gio io 7
R n1 R n1 eg n R n1
e 0 eg 0 io 7 eg 0
R gio R ion R n1 R ion
eg n e 5 eg 2 e 5
io 7 R gio R io R gio
R n1 eg n eg n eg n
eg 2 i 8 i 8 i 8
R ion R on R on R on
eg eg 1 eg 1 eg 1
io 8 io 1 io 1 io 1
R n1 R n1 R n1 R n1
eg 1 eg 2 eg 0 eg 2
R ion io io io
n n n
eg 6 6 6 6
io C C C
n R A R A R A
1 eg R eg R eg R
C i i i
R A R on R on R on
e R e e e
R gio R gio 2 R gio 2 R gio 2
eg n eg n eg n eg n
Poverty Headcount
io 2 io 1 io 1 io 1
n n n n
R 4- R 4- R 4- R 4-
eg A eg A eg A eg A
io io io io
n n n n
3 3 3 3
N N N N
C C C C
R R R R
0
5
0
5
0
0
5
10
15
10
10
20
30
10
15
C A A A
ar R R R
a R MM M R MM
A ga eg C M eg
R R R a R
R MM eg ion eg ra eg ion
eg io 9 io ga io 9
R io n n n
eg n 4 R 4- 4
io 9 C -B eg B C -B
n ar ar
R 4- R ag R ion R ag
eg B e a eg 9 e a
R io R gio R ion R gio
eg n eg n e 5 eg n
io 5 io 7 R gio io 7
R n1 R n1 eg n R n1
eg 0 eg 0 io 7 eg 0
R io R ion R n1 R ion
eg n eg 5 eg 2 eg 5
io 7
R n1 R ion R ion R ion
eg eg eg
eg 2 8 8 8
R io R ion R ion R ion
125
eg n eg 1 eg 1 eg 1
io 8 io 1 io 1 io 1
R n1 R n1 R n1 R n1
eg 1 eg 2 eg 0 eg 2
R ion io io io
n n n
eg 6 6 6 6
io C C C
Poverty Gap
n R A R A R A
1 eg R eg R eg R
C io io io
R A R n R n R n
eg R e e e
R io R gio 2 R gio 2 R gio 2
eg n eg n eg n eg n
io 2 io 1 io 1 io 1
n n n n
R 4- R 4- R 4- R 4-
eg A eg A eg A eg A
io io io io
n n n n
3 3 3 3
N N N N
C C C C
R R R R
0
2
4
6
8
0
1
2
3
4
5
0
2
4
6
C A A
0
5
10
15
ar R A R
a R
A ga
R MM
eg M R MM
eg
R R io C M R io
R MM eg n R ar eg n
eg io 9 eg a io 9
R io n io ga n
eg n 4 n 4
io 9 C -B R 4- C -B
n ar eg B ar
R 4- R ag R ion R ag
eg B eg a eg 9 eg a
R ion R ion R ion
eg eg R ion eg
Figure 5.5 Regional Intertemporal Poverty Estimates (JR Approach)
io 5 io 7 e 5 io 7
R n1 R n1 R gio R n1
eg 0 eg 0 eg n eg 0
i io 7 i
R ion R on R on
eg e 5 R n1 e 5
io 7 R gio
eg 2
R gio
R n1 eg n R io eg n
eg 2 eg n
i 8 8 i 8
R ion R on R ion R on
eg eg 1 eg 1 eg 1
io 8 io 1 io 1 io 1
R n1 R n1 R 1 n R n1
eg 1 eg 2 eg 0 eg 2
i io io io
R on n n n
eg 6 6 6 6
io C C C
n R AR R A R AR
1 eg eg R eg
R CA R on i R on i
e R ion e
eg R e
Poverty Severity
R io R gio 2 R gio 2 R gio 2

eg n eg n eg n eg n
io 2 io 1 io 1 io 1
n n n n
R 4- R 4- R 4- R 4-
eg A eg A eg A eg A
io io io io
n n n n
3 3 3 3
N N N N
C C C C
R R R R
R R R
0
5
10
15
20
0
5
10
15
20
0
10
20
30
40
0
5
10
15
20
eg A
io R egi R R egi
eg on R M eg on
C n9 io 9 R eg M io 9
R ar R n4 eg io R n4
eg ag eg -B io n 9 eg -B
io a R io n io
R n1 eg n
R egi 0 io 7 C 4-B
ar A n
eg on n R ag R RM 7
eg
io 7 A 10 e io M
n R R gi o a n
R 4 M eg n
e - C M io 7 C 10
R gi o B a a
eg n R r ag R n1 R r ag
e eg 0 e
R i on 5 R gi o a R i on R gi o a
eg 1 eg n e eg n
io 1 io 5 R gi o 5 io 5
R n1 R n1 eg n R n1
eg 2 e e
io R gi o 1 R i on 8 R gi o 1
eg n eg 1 eg n
A n8 io 8 io 1 io 8
R
R M R n1 R n1 R n1
eg M eg 2 eg 2 eg 2
R i on io io io
n n n
R egi 6 6 6 6
alpha = 1.5
eg on R CA R CA R CA
io 1 eg R eg R eg R
n i o i o i o
4- R n R n R n
A R eg 2 R eg 2 R eg 2
R CA eg io eg io eg io
eg R io n 1 io n 1 io n 1
i o n n n
R n R 4 R 4 R 4
eg 2 eg -A eg -A eg -A
io io io io
n n n n
N 3 N 3 N 3 N 3
C C C C
R R R R
R R R
0
5
10
15
20
25
0
5
10
15
20
0
10
20
30
40
0
5
10
15
20
25
eg A
io R egi R R egi
eg on R M eg on
C n9 io 9 R eg M io 9
R ar R n4 eg io R n4
eg ag e - io n 9 eg -B
io a R gi o B n io
R n1 eg n
R egi 0 io 7 C 4-B
a A n
eg on n R r ag R RM 7
eg
io 7 A 10 e io M
R R gi o a n
R n4 M eg n
eg -B C M io 7 C 10
R io a a
eg n R r ag R n1 R r ag
eg a eg 0 eg a
R i on 5 R io R i on R io
eg 1 eg n e eg n
R n1 R n1 eg n R n1
eg 2 e e
126
eg n eg 1 eg n
A n8 io 8 io 1 io 8
R
R M R n1 R n1 R n1
eg M eg 2 eg 2 eg 2
R i on io io io
n n n
R egi 6 6 6 6
alpha = 2
io 1 eg R eg R eg R
n
4- R i on R i on R i on
n n n
R i on R 4 R 4 R 4
eg 2 eg -A eg -A eg -A
io io io io
n n n n
N 3 N 3 N 3 N 3
C C C C
R R R R
R R R
0
0
5
0
0
10
20
30
10
15
20
25
10
20
30
40
10
20
30
eg A
io R egi R R egi
eg on R M eg on
C n io 9 R eg M io 9
R ar 9 n n
eg ag R 4 eg io R 4
io a eg -B io n 9 eg -B
R io n io
R n1 eg n
R egi 0 io 7 C 4-B
a A n
eg o n R r ag R RM 7
eg
io n 7 A 10 e io M
n R R gi o a n
R 4 M eg n
C M C 10
Figure 5.6 Regional Intertemporal Poverty Estimates (DAG Approach)
eg -B io 7
a a
R io R r ag R n1 R r ag
eg n eg a eg 0 e
R i on 5 R io R i on R gi o a
eg 1 eg n e eg n
R n1 R n1 eg n R n1
eg 2 e e
eg n eg 1 eg n
A n8 io 8 io 1 io 8
R
R M R n1 R n1 R n1
eg M eg 2 eg 2 eg 2
R i on io io io
n n n
R egi 6 6 6 6
alpha = 2.5
io 1 eg R eg R eg R
n i i i
4- R on R on R on
i n n n
R on R 4- R 4- R 4-
eg 2 eg A eg A eg A
io io io io
n n n n
N 3 N 3 N 3 N 3
C C C C
R R R R
R
0
0
0
1
.2
.4
.6
.8
.2
.4
.6
.8
.2
.4
.6
.8
A A
0
1
.5
1.5
eg
C io R R
ar M R M
a A n R Ca M eg M
A ga R RM 9 io
R eg eg ra n
R M io M i o ga C
e
R gi M n n R a 9
eg ra
eg o 1 R 4-
eg B i o ga
io n R Ca 0 R n
R n 9 eg ra R i on eg 4-
eg 4- i o ga eg 9 io B
io B n
R 4- R i on R n1
R n1
eg 0 eg B e eg 0
R i on R gi o 5 R i on
R i on eg 5 eg n eg 5
e io 7
R gi o 7 R i on R n1 R i on
eg n e e e
R i on 5 R gi o 7 R gi o 2 R gi o 8
eg 1 eg n eg n eg n
io 2 R i on 8 R i on 8 R i on 7
R n1 eg 1 eg 1 eg 1
eg 1 io 1 io 1 io 1
R i on R n1 R n1 R n1
eg 8 eg 2 eg 0 eg 2
R i on io
n R i on
Foster
R i on
eg 6 eg 6 6 eg 6
io R i on R CA R i on
n eg 2 eg R eg 1
1 io i o io
R CA n R n n
R egi R R C 1 R egi 2 R C 2
eg on eg A eg on eg A
io 2 io R io 1 io R
n n n
R n4 R 4- R 4- R 4-
eg -A eg A eg A eg A
io io io io
n n n n
N 3 N 3 N 3 N 3
C C C C
R R R R
0
1
2
3
R A
0
1
2
0
1
2
0
1
2
.5
1.5
.5
1.5
.5
1.5
C R A
ar
eg M
a io R Ca M R RM
A ga A n eg ra eg M
R R RM 9 io
R M eg i o ga
n n
R egi M io M R 4- R Ca 9
eg o n eg B eg ra
io n R C a 10 i o ga
R n 9 eg ra R i on
eg 9 R n
eg 4- i o ga eg 4-
io B R i on io B
R n1 R n4 e R n1
eg 0 eg -B R gi o 5 eg 0
R i on R i on eg n R i on
e eg 5 io 7 eg 5
R gi o 7 R n1
eg n R i on
e e 2 R i on
e
R i on 5 R gi o 7 R gi o R gi o 8
eg 1 eg n eg n eg n
io 2 R i on 8 R i on 8 R i on 7
eg 1
127
R n1 eg 1 io 1 eg 1
eg 1 io 1 io 1
eg 8 eg 2 eg 0 eg 2
io
R i on R i on n R i on
eg 6 eg 6 6 eg 6
io R CA
n R i on eg R R i on
1 eg 2
io i
eg 1
io
R CA n R on n
e
R gi R R C 1 R egi 2 R C 2
eg on eg A eg on eg A
io 2 io R io 1 io R
n
R n4 R n4 R 4-
eg A R n4
eg -A eg -A eg -A
io io io io
Mean poverty duration
n n n n
N 3 N 3 N 3 N 3
C C C C
R R R R
R
0
20
40
60
80
0
20
40
60
0
20
40
60
C A
0
20
40
60
80
100
eg
ar io A R RM
a R
A ga A n M eg M
R R RM 9 R Ca M io
R M eg n
e
R gi M io M
eg ra
R Ca 9
eg on n i o ga eg ra
i 1 R n4 i o ga
R on 9 R Ca 0 eg -B R n
eg 4- eg ra eg 4-
io B i o ga R i on io B
R n1 R 4 n eg 9
R n1
eg 0 eg -B R i on eg 0
e
Figure 5.7 Regional Intertemporal Poverty Estimates (Spells Approach)
R i on R i on R gi o 5 R i on
e
eg 5 eg n eg 5
R gi o 7 R i on io 7 R i on
eg n e R n1 e
R i on 5 R gi o 7 e R gi o 8
eg 1 eg n R gi o 2 eg n
eg n
io 2 R i on 8 R i on 7
R n1 eg 1 R i on 8 eg 1
eg 1 eg 1
io 1 io 1 io 1
eg 8 eg 2 eg 0 eg 2
R i on R i on io R i on
eg 6 eg 6 n eg 6
io 6
n R i on R CA R i on
1 eg 2 eg R eg 1
R CA io
n R n i o io
n
e
% Ever poor
R gi R R 1 R egi 2 R 2
eg o eg C eg on eg C
io n 2 io AR io 1 io AR
n n n
R 4-
eg A
R 4-
eg A R n4 R 4-
eg A
eg -A
io io io io
n n n n
N 3 N 3 N 3 N 3
C C C C
R R R R
exposure to bad weather, as it sits in one of the most tropical cyclone-prone areas in the world
that experience an average of ten typhoons every year (PAG-ASA 2014).54 These weather
disturbances often have debilitating effects as estimates suggest that a strong typhoon in the
Philippines can destroy more than US$ 75 million (approximately 3 billion Philippine pesos
using prevailing exchange rates) worth of agricultural crops (PAG-ASA 2014). Small scale
agricultural workers are usually the most affected because their produce over several months
can be wiped out easily by one typhoon. Third, the Philippines has weak institutions to support
agricultural expansion or buffer risks present in the agricultural sector. Previous studies
identify inadequate provision of socio-economic services such as lack of access to concrete or
paved roads exacerbates the chronic poverty in remote agricultural areas (Schelzig 2005;
Aldaba 2009). Weak property rights also contribute to why rural households are trapped in
longer poverty episodes compared to their urban counterparts. While there have been several
agrarian reforms that have been implemented in the Philippines since the 1970s which aim to
improve the welfare of the low income households in rural areas who depend on access to land
for their day-to-day living, existing systems (e.g., credit market) have not delivered the
maximal economic opportunities envisaged by the reform (Balisacan 2002).
Although more urbanized regions like Regions 3, 4A and NCR show lower levels of persistent
poverty, transient poverty is not a trivial problem in these areas. What are the causes of transient income
downturn in urban areas? One is the lack of opportunities due to population strain. As the prospects of
more economic opportunities usually attract the rural poor to migrate to urban centers, urban population
starts growing rapidly. The scarce opportunities available for the growing population can lead to a
highly unstable income flow. Second, it is also widely perceived that people living in urban areas are
more exposed to health and safety risks due to makeshift housing, poor sanitation, fire hazards and
crime (Housing and Urban Development Coordinating Council 2008; Aldaba 2009). In general, when
socio-economic shocks erode the meagre assets accumulated by households, they can be pulled back
into transient poverty.
Although the different intertemporal poverty estimation tools have produced roughly
similar lists of regions with the highest and lowest levels of poverty persistence, the rankings
are not perfectly robust across all approaches, especially with respect to transient poverty.
While it may be true that the minimal differences in the regional rankings may not have a
profound policy impact, policy research still needs to address the issue of the robustness of
poverty estimates by using more comprehensive methods. In addition, the impact may be more
severe if a similar computational exercise was undertaken at finer administrative levels where
54 PAG-ASA, the Philippines’s weather bureau, stands for Philippine Atmospheric, Geophysical and Astronomical Services.
128
resources are allocated on the basis of various indicators which includes poverty. However,
due to sample size limitations, this study did not conduct a rigorous intertemporal poverty
ranking of the provinces. This is reserved for future research.
5.4.3 Who are the Persistently and Transiently Poor?
The objective of this section is to draw our attention to the possible heterogeneity in the
intertemporal poverty experience of Filipino households. Appendix Tables A5.1 and A5.2
provide a descriptive summary of the proportion of the population classified as persistently
poor, transiently poor and non-poor, disaggregated by different household characteristics. The
results show that the intertemporal poverty is higher in households whose heads are male, with
low educational attainment and are working in the informal sector. In addition, those who are
living in southern Philippines especially those who heavily rely on entrepreneurial activities
and on agriculture sector have higher risk on experiencing more severe poverty status over
time.
Table 5.7 summarizes the results of the statistical models that I estimated to measure the
correlation of each factor with intertemporal poverty in the presence of other household
characteristics using the components and spells approach and different poverty line
specifications. In particular, it shows the regression coefficients of the multinomial logistic
models for the probability of being classified as persistently poor, transiently poor and non-
poor. These coefficients can be interpreted as multinomial log-odds, i.e., for a unit change in
the explanatory variable, the logit of the propensity to be classified as persistently poor (or
transiently poor) relative to being non-poor will change by an amount equivalent to the
regression coefficient, holding all other explanatory variables constant. 55 As pointed out in
Chapter 1, the explanatory variables are chosen based on standard human capital theories.
In general, the regression models presented in Table 5.7 confirm the importance of
standard socio-demographic characteristics like sex, age and education in explaining a
household’s intertemporal poverty status. In particular, the results show that individuals living
in female-headed households are at risk of spending more time in poverty than their
counterparts living in male-headed households. Interestingly, this result deviates from the
descriptive statistics provided in Appendix Tables A5.1 and A5.2 wherein I observe lower
poverty rates for female-headed households. However, this does not imply that women in the
country are not disadvantaged. In fact, controlling for a number of socio-demographic
55 If we take the exponential of these regression coefficients, we will get the relative risk ratios.
129
characteristics seems to erode the advantage of female-headed households over male-headed
households.56 Hence, the model estimates suggest that after controlling for other factors, living
in female-headed households in the country is still positively correlated with longer poverty
spells. Thus, it is not surprising that the Philippine government still acknowledges women as
one of the most vulnerable groups who must be provided with social protection against the risk
of falling into a vicious cycle of poverty (NCRFW 2004). On the other hand, the age of the
household head presents a typical concave relationship with income (Kearl & Pope 1983) and
consequently, a convex relationship with the length of stay in poverty. In particular, during
prime age years, individuals are in the process of climbing up the occupational ladder. With
higher income accompanying this process, these people have less risk of falling into poverty.
However, beyond a certain age threshold, individuals start to experience income deterioration.
For instance, some of those who used to work in the formal sector could only rely mostly on
pensions after retirement, representing a fraction of their previous income. With advancement
in age, some of those who work in the informal sector have less employment opportunities
because they are less capable of performing physical tasks that jobs in the informal sector entail.
Furthermore, the estimated models reiterate the importance of education in minimizing the risk
of falling into poverty. Better educated households, as proxied by educational attainment of the
household head, face less risk of long poverty spells. In this context, higher educational
attainment serves as a mechanism for expanding one’s overall social mobility prospects. This
is consistent with the human capital theory which suggests that the skills and knowledge
imparted by higher educational attainment improves an individual’s productivity, and in turn,
his/her ability to mobilize resources (Tilak 2002). While educational attainment explains a
significant portion of the differences in household income and poverty status, education
remains a development puzzle in the country. For instance, compared to other countries with
similar level of development, the Philippines has much higher gross enrolment rates in
secondary and tertiary education (WDI 2014). Despite this advantage, significant pockets of
56 The descriptive statistics confirm results from previous studies suggesting that Philippines is one of the few countries where
female-headed households do not necessarily portray a vulnerable group (Chant 1997). Whereas female-headed households in
many countries are usually characterized as a vulnerable group due to the presumed lack of ability of women to mobilize socio-
economic resources for the family, female-headed households in the Philippines are more likely to be found in middle and
high-income groups. Several reasons have been offered to explain this anomaly. First, the majority of these female heads are
widows in their senior years who have had enough time to accumulate assets (Chant 1997). Second, female-headed households
are more likely to have higher proportions of members who are already working. In fact, Ofstedal, Reidy & Knodel (2004)
estimate that about 40% of total income of female-headed households come from the contributions of children and other family
members compared to 25% for male-headed households. Third, female-headed households in the country are characterized by
successful women who have higher educational attainment than an average male household head.
130
Table 5.7 Regression Coefficients of Multinomial Logistic Models for Intertemporal Poverty in the Philippines
(Base = Non-poor)
Components Approach Spells Approach
Variable US$1.25 US$2 US$1.25 US$2
Persistent Transient Persistent Transient Persistent Transient Persistent Transient
Main Island (base = NCR)
Luzon 14.74 1.615** 1.761*** .7193** 1.738 1.835** 1.37*** .7893***
Visayas 15.94 2.246*** 2.63*** .8773*** 2.871** 2.439*** 2.161*** .918***
Mindanao 16.18 2.438*** 3.055*** 1.192*** 3.093** 2.643*** 2.645*** 1.15***
Urban -.7792*** -.3362** -.8372*** -.4097*** -.6535*** -.3525** -.752*** -.4219***
Hhld head is Male -.8212* -0.02772 -0.3164 0.1372 -.7392* 0.01908 -0.3364 0.2069
Hhld head's Age -.146*** -.07295*** -.06805** 1.29E-02 -.146*** -.0658** -.04758* 0.008981
2
Hhld head's Age (x 10000) 15.39*** 7.989*** 8.414*** 0.1004 15.47*** 7.21*** 6.263** 0.5388
Marital Status of hhld head
(base = Single)
Married -0.1719 -.7561* 0.1995 0.0541 -0.06679 -.854** 0.2322 -0.004984
Other -0.9965 -.8357** -0.5165 -0.03779 -0.8185 -.8883** -0.4676 -0.02353
Hhld head's Educational
Attainment
(base = Primary education)
Secondary education -1.168*** -.6984*** -1.059*** -.3709*** -.9229*** -.7924*** -1.007*** -.314**
College education -2.815** -1.69*** -2.782*** -1.255*** -2.813** -1.723*** -2.717*** -1.117***
Hhld type (base = Single family)
Extended family 0.07459 0.115 -0.1549 0.1655 -0.07087 0.1817 -0.1142 0.163
Two or more non-related
-11.13 -16.69 -13.45 0.7791 -9.873 -12.28 -12.91 0.8591
individuals
Proportion of hhld members who
3.247*** 1.297*** 3.364*** 1.538*** 2.72*** 1.381*** 3.119*** 1.487***
are young
Family size .742*** .4392*** .8163*** .3285*** .7093*** .428*** .7315*** .3336***
131
Table 5.8 (con’t) Regression Coefficients of Multinomial Logistic Models for Intertemporal Poverty in the Philippines
(Base = Non-poor)
Agricultural hhld .8974*** .2856* 1.176*** .4413** .8569*** .2588* 1.073*** .4618**
At least one hhld member is
-.7014** -.3591* -.7894*** -.2559* -.8061*** -.2991* -.7763*** -0.2092
working abroad
Proportion of employed hhld
0.4205 0.2748 .4709* 0.2968 .4735* 0.228 0.3439 .3691*
members
Proportion of employed members
-0.07962 -0.1215 -0.08583 0.05797 -0.1474 -0.0843 -0.04303 0.03026
with permanent job
Proportion of employed members
-.402* -0.1718 -.5577*** -.382** -.4598* -0.1229 -.4919*** -.3998**
with formal job
Hhld owns land/house -.4552** -0.2301 -.3351* -0.1405 -.4813** -0.185 -.3067* -0.1419
Type of toilet facility
(base = water-sealed)
Closed pit .6018*** 0.2427 .5974*** 0.1466 .5389*** 0.2467 .4783** 0.2026
Open pit / others .2949* 0.1142 .628*** 0.2503 .3215** 0.06999 .5258*** .3141*
Hhld had access to electricity -1.188*** -.7223*** -1.18*** -.523*** -1.068*** -.7524*** -1.123*** -.4776**
Hhld had access to water faucet -0.007578 -0.1731 -.4351*** -.3877*** -0.01936 -0.1857 -.4319*** -.3783***
Hhld owns refrigerator -1.939*** -1.089*** -1.616*** -.8847*** -2.053*** -1.008*** -1.54*** -.8438***
Hhld owns information gadget -.7699*** -.4366*** -1.051*** -.5772*** -.7733*** -.3914*** -1.066*** -.4502**
Hhld owns phone -2.778*** -.7764*** -1.671*** -.847*** -2.251*** -.7423*** -1.523*** -.8375***
Hhld owns washing machine -2.676*** -.9971*** -1.893*** -.8125*** -2.116*** -.9966*** -1.729*** -.7868***
Hhld owns transportation -1.749** -.716** -1.603*** -.4255** -1.014* -.8483** -1.191*** -.4947**
132
Table 5.9 (con’t) Regression Coefficients of Multinomial Logistic Models for Intertemporal Poverty in the Philippines
(Base = Non-poor)
Change in family size .3561*** .1942*** .3646*** .1605*** .3297*** .1928*** .3417*** .1523***
Hhld head's educational
0.4681 .4844* .5828** 0.1836 0.2238 .6087** .5648** 0.143
attainment deteriorated
Hhld head's educational
-.5328* -.4706** -.6961*** -0.2968 -.7037** -.3785* -.6523*** -0.2691
attainment improved
Hhld's main source of income
.6844*** .3169* .9244*** .4908** .8224*** 0.1815 .8916*** .4555*
shifted from non-agri to agri
Hhld's main source of income
-0.1961 0.111 -.4627* 0.06591 -0.1116 0.08199 -.3925* 0.04958
shifted from agri to non-agri
Hhld shifted from land/house
.5236** 0.2467 .354* 0.1604 .5041** 0.2397 .3573* 0.1283
ownership to non-ownership
Hhld shifted from land/house
-0.1944 -0.1136 -0.04462 0.0284 -0.2016 -0.0989 0.0085 -0.0115
non-ownership to ownership
Change in the proportion of
-0.2479 0.0131 -.5914** -0.2664 -0.2002 0.008346 -.4976** -0.2935
employed members
Intercept -16.29 -2.146** -2.41** -2.172*** -2.915* -2.553** -1.936** -2.513***
Source: Author’s computations using data from the longitudinal subsample of FIES 2003, 2006 and 2009.
Note: The dependent variable is based on household expenditure per capita.
133
poverty remain in the country. In fact, other Southeast Asian countries such as Indonesia and
Thailand with almost the same levels of secondary and tertiary enrolment rates have
significantly better poverty trends. This finding challenges the quality of the education system
in the country and the extent to which it is possible to arrest poverty in the future. Thus, it is
imperative for socio-economic planners to implement policies that will make the education
system more responsive to the needs of the poor. To raise the country’s quality of education
and be at par with other countries in Southeast Asia, the Philippines shifted from a 10-year to
12-year basic education system starting 2014. This system, commonly referred as K-12, is
widely adopted by both industrialized and developing countries (Magno 2011). The system
aims to provide competitive basic education. However, whether it will contribute to poverty
reduction or not, is a test of time. The result may also be indicative of the limited good jobs
available in the labour market. Several studies show that the Philippines confronts a dual jobs
challenge which entail expanding formal sector employment and at the same time, improving
the quality of jobs in the informal economy (WB 2013). In addition, the impact of expanding
non-traditional employment arrangements which accompany globalization on the working
poor is an issue that has to be examined as well. This topic is further discussed in Chapter 8.
Household composition is also a significant explanatory factor in inferring one’s length
of stay in poverty. Larger households have higher risks of experiencing longer poverty spells.
In particular, the odds of staying in poverty increases with the number of dependent children
in the family. This is consistent with the mounting evidence suggesting that lower fertility is
correlated with improved socio-economic outcomes. In particular, instead of allocating a
portion of its available resources for more productive economic activities, households must
reallocate its available resources for every additional member. For example, having more
dependent children in the family limits the ability of women to engage in paid employment as
they are usually expected to do childrearing (Adair et al. 2002). At the same time, having more
children in the family could also have a negative impact on household savings. For instance,
using data from the 2009 FIES, I estimate that the correlation between the number of children
and household saving is -0.11 (p-value<0.0001). Since larger households have less savings,
they are more vulnerable to unexpected income shocks arising from illness, unemployment,
among other factors that lead to reduced income flows. Nevertheless, larger household sizes
could also have a negative impact on the duration of poverty spell in the later stage of the
household’s life cycle. For instance, the estimated models suggest that the risk of staying in
poverty decreases with the proportion of household members who are working. In this context,
some argue that parents who were initially disadvantaged during their early childrearing years
134
would fare better in the future because there will be more children to contribute in mobilizing
resources for the household even after these children form their own families due to close
family ties of Filipinos (Ongsotto & Ongsotto 2002). However, several arguments have been
offered against this hypothesis. For instance, studies show that an additional child in the
household reduces the probability of other children in the family being enrolled in school
(Conley 2000) which in turn, may hinder these children from reaching their full economic
potential in the future. In other words, while they may be able to contribute to income
generation for the household, the income may be at sub-optimal levels.
The patterns of poverty dynamics in the country have a remarkable spatial feature. The
results of the estimated models suggest that those who live in rural areas particularly in the
southern part of the country have higher risks of staying in poverty. 57 On the other hand,
individuals living in urban areas where most of the economic activities are centred, experience
shorter poverty spells. Some even argue that those who experience longer-than-average poverty
spells in urban areas may be partly considered as a spill-over effect of the socio-economic
disadvantage in rural areas (Reyes 2002b; CEDAW 2009). As poor households migrate from
rural to urban areas, many of them remain poor until they get good jobs. Thus, they inflate the
number of urban poor. Hence, urban poverty reflects residual absorption of rural migrants
(Mitra 1992). In general, the prominent spatial feature of the distribution of poverty is not
unique in the Philippines. In many developing countries, significant pockets of poverty are
clustered in specific areas (Bigman & Fofack 2000; Hennigner & Snel 2002). Factors like
climate, geography, natural resources, access to urban centers and local political conditions and
economic opportunities drive the significant spatial variations in the length of poverty spells
(Ravallion & Wodon 1997). The model results also confirm that greater access to basic services
such as electricity, clean water and sanitary toilet facilities is correlated with lower
intertemporal income poverty. For instance, having access to electricity contributes positively
to higher household savings since a unit cost of lighting with electricity is generally cheaper
than using candles or oil lamp. In turn, households can then use the additional savings as a
cushion against the risk of falling into poverty in the future. On the other hand, experts agree
that access to clean water and sanitation facilities (e.g., sanitary toilets) has a multiplier effect
on many socio-economic indicators particularly, movements into and out of poverty (WHO
and UNICEF 2008). In particular, access to these facilities have a direct impact on health
outcomes. Not surprisingly, those who lack access to clean water and sanitation facilities have
57 Most of the poorest provinces are in Mindanao. In addition, the risk of persistent poverty is highest in Mindanao.
135
higher risks of contracting diseases like cholera, typhoid, infectious hepatitis and polio.
Consequently, these health shocks may contribute to income depletion of affected households,
pushing them towards poverty. On the other hand, the cost-savings incurred by households
with access to electricity, clean water and sanitation facilities may contribute to a household’s
increased propensity to start-up income-generating (micro-) entrepreneurial activities. This
result suggests the need for the government to facilitate universal access to basic services.
The estimated models also suggest that ownership of land and other productive
(disposable) assets like television, radio, telephone, washing machine, among others is
negatively correlated with the duration of income poverty spells. There are several reasons why
these variables are significantly correlated with intertemporal income poverty. For instance,
during periods of economic uncertainties, these assets may be sold to cushion the disruptions
in income flows. Some of these assets may also be used to improve access to information that
will enhance efficiency for planning their routine economic activities. For example, farmers
who own television or radio may be warned about an impending weather disturbance earlier.
More importantly, many of these assets could also be used to generate income. Furthermore,
most of these variables could be considered as indicators of material deprivation and hence, it
is difficult to infer the direction of their relationship with intertemporal poverty.
Variation in employment outcomes is one of the statistically significant correlates of
intertemporal poverty. Labour serves as one of the few assets that low income individuals have
access to. In this context, having more household members who are working would naturally
decrease the risk of staying in poverty for extended periods of time. This is confirmed by the
negative coefficient of the proportion of household members who are employed, on poverty
status as discussed earlier. But beyond the number of members working for the family, the type
of employment also matters. For instance, the estimated models suggest that households which
rely mostly on wages, particularly in non-agricultural sectors, experience shorter poverty spells
than those who rely on agricultural wage employment or earnings from self-employment. The
level of productivity in the agriculture sector is one of the reasons why this is the case. In
particular, low agricultural productivity contributes significantly to persistent poverty in many
developing countries. In addition to low productivity, frequent income fluctuations arising
from crop loss (due to weather disturbances) or sudden changes in food prices also contribute
to longer poverty durations for those who rely on agricultural wage employment. Among
agricultural workers, farmers and fishermen have the highest risk of more severe poverty spells
(NSCB 2012). Like agricultural wage employment, self-employment is also correlated with
longer poverty durations relative to those who rely on non-agricultural wage employment.
136
Although they do not comprise a homogeneous group, a significant bulk of the self-employed
in many developing countries including the Philippines are working on own-account with
significantly less income. Banerjee & Duflo (2007) argue that much of the self-employed in
developing countries are running their business because “they are still relatively poor and every
little bit helps” until they find a more stable wage job. In the case of the Philippines, this is
confirmed by the study of Hasan & Jandoc (2010) who concluded that the majority of the self-
employed in the country are not “capitalists in waiting.” In this context, it is not surprising to
note that transitions from either agricultural wage or self-employment into non-agricultural
wage employment reduces the risk of long poverty spells. Similarly, employment in the formal
sector decreases the risk of being trapped in longer episodes of poverty. Compared to an
informal job, a typical job in the formal sector is associated with higher and more stable income
flows. Formal jobs also offer wider social protection coverage. These features serve as a
cushion for unexpected income shocks which in turn, decreases the risk of falling into poverty.
Interestingly, the negative effect of formal employment is stronger in the US$2 poverty line-
based model suggesting that having a formal job is not quite common among the poorest of the
poor. Lastly, households where at least one member is working abroad tend to spend less time
in poverty. The Philippines is one of the large-labour exporting countries and previous studies
suggest that it contributes to improved macro and micro socio-economic outcomes for the
country (Ang, Sugiyarto & Jha 2009). For instance, remittance from a migrant worker abroad
eases liquidity constraint for many low income households. This allows households to
restructure their economic activities away from traditional subsistence activities and towards
more efficient economic ventures (Brown & Leeves 2008).58 In addition, remittance from
abroad has a positive impact on investment on productive assets which in turn, could lead to
lower risks of falling into poverty. Similarly, domestic remittances also contribute positively
to minimizing poverty durations. In fact, its impact on the length of poverty spells could be
stronger than remittance from abroad because low income individuals are more likely to receive
remittance from internal migration (Pernia 2008).
5.5. Summary
Reducing poverty is considered as one of the most important tasks of the developing
world. Not surprisingly, poverty monitoring remains at the heart of the economic development
58 Some argue that international remittance could also lead to negative outcomes on a household’s welfare. For instance,
Rodriguez & Tiongson (2001) find that households with migrant workers have lower labour force participation rates and
shorter work hours. The authors attribute it to the propensity of the migrant’s relatives to substitute income-generating activities
for more leisure.
137
literature aiming to advance poverty reduction efforts. In the Philippines much effort is needed
to be able to eradicate extreme poverty as according to recent estimates, about two out of five
Filipinos are still living below US$2/day (WDI 2014). Compared to its other Southeast Asian
neighbours, the country has shown lower growth-elasticity of poverty and performed dismally
on other social economic indicators for many years. This has earned the country its title Sick
Man of Asia (Ching 1993). Nevertheless, a more robust and stronger economic performance
has been noted in recent years, prompting economic analysts to upgrade their growth forecasts
for the Philippines.
The examination of household panel data in the Philippines reveals that despite faster
economic growth over the past decade, poverty remains a prominent feature of its development
landscape. This poverty has two forms: persistent and transient. For instance, while about 40%
of the population were US$2/day-poor at any given (survey) year between 2003 and 2009,
estimates suggest that about 30% were poor in all (survey) years examined while 60%
experienced. Overall, this suggests a more dynamic poverty phenomenon in the country than
what is conventionally perceived when examination is only based on trends in cross-sectional
indicators of poverty. To be able to respond to the challenge of reconciling stronger economic
growth with positive gains in poverty reduction, it is important to implement policies that can
minimize both persistent and transient forms of poverty. However, this calls for a different mix
of policies. This prompts the need to address the question: how long do the poor stay in
poverty? Consistent with the findings of Reyes et al. (2011) and Bayudan-Dacuycuy & Lim
(2013), the results presented in this chapter suggest that most poverty experiences of Filipino
households were persistent. In other words, households tended to stay in poverty for extended
periods of time. This pattern is in sharp contrast with the intertemporal poverty in industrialized
countries wherein the share of persistent poverty would normally be much smaller compared
with transient poverty. This implies that poverty reduction programs in developing countries
like the Philippines should be aimed primarily at providing long-term human capital
development. Nevertheless, the robustness analysis presented in this chapter revealed that the
relative importance of persistent and transient poverty were sensitive to the type of poverty
measure used and the poverty line specified. In particular, the relative importance of transient
poverty increased dramatically as the poverty line decreased or as the poverty measure becomes
more sensitive to the illfare of the poorest of the poor.
In conclusion, the results can be used to illuminate several important broad policy
implications. First, an insignificant change in the cross-sectional estimates of poverty from
2003 to 2009 does not imply that all poor households were systematically excluded from
138
reaping the benefits of the faster economic growth that transpired during this period. In
particular, about 15% to 20% of the households classified as poor in an initial time period
managed to escape poverty in the succeeding survey wave. Nevertheless, the risks of falling
into poverty were also not trivial. For instance, about 10% to 15% of the initially non-poor
households fell into poverty in the succeeding wave. In other words, transition from poverty to
non-poverty does not necessarily imply a permanent escape from socio-economic dearth. To
be able to understand poverty vulnerability in the Philippines better, future research may use
the concept of fuzzy logic (Lemmi & Betti 2006) or construct poverty vulnerability lines such
as those proposed by Dang and Lanjouw (2014). More importantly, the relevance of poverty
vulnerability also means that existing social protection systems should be improved to
minimize the adverse consequences of income shocks for both the poor and economically
vulnerable. Second, for about 15% to 40% of the population, poverty has been a long episode
of socio-economic deprivation. A more aggressive policy intervention is needed for these
persistently poor households. One of the first steps might be to institutionalize an effective
targeting system that will identify persistently poor households. Equally important is to ensure
that intervention programs are accessible as many of these chronically poor households are
likely situated in remote and hard-to-reach areas. Third, given that estimates could change
depending on measurement parameters, programs that rely on poverty ranking should be
examined rigorously. Overall, both persistent and transient poverty should be of concern for
the country’s socio-economic planners and there should be a balanced policy program that
supplements long-term investment on the development of human capital of the persistently
poor with provision of social safety nets that can stabilize income flows of the transiently poor.
139
Appendix Table A5.1 Intertemporal Poverty Headcount Rate using the Components Approach (%),
by Household Characteristics
US$1.25 US$2 half of median poverty line government/official
Hhld characteristics Persistently Transiently Persistently Transiently Persistently Transiently Persistently Transiently
Non-Poor Non-Poor Non-Poor Non-Poor
Poor Poor Poor Poor Poor Poor Poor Poor
Rural 25.45 18.4 56.15 58 14.53 27.47 18.8 19.68 61.52 36.74 21.72 41.54
Urban 4.9 8.02 87.08 20.5 13.64 65.86 3.4 6.62 89.98 10.16 13.37 76.46
NCR 0 1.09 98.91 4.29 11.11 84.6 0 0.89 99.11 2.4 8.66 88.94
Luzon 9.03 10.77 80.2 30.7 16.31 52.99 6.37 9.55 84.07 17.79 17.08 65.13
Visayas 21.44 18.3 60.26 52.04 12.95 35.01 15.2 19.36 65.44 29.54 19.75 50.7
Mindanao 28.64 18.67 52.7 60.13 11.93 27.94 21.89 20.04 58.07 38.77 20.34 40.89
Female-headed hhld 5.86 8.33 85.81 20.17 12.52 67.31 3.78 8.39 87.83 10.29 13.77 75.94
Male-headed hhld 16.75 14.03 69.22 42.43 14.33 43.24 12.33 13.98 73.69 25.66 18.18 56.16
Single 2.97 11.54 85.49 15.9 13.87 70.23 2.61 7.48 89.91 5.92 14.89 79.18
Married 16.73 13.53 69.74 41.89 13.92 44.19 12.33 13.72 73.95 25.42 17.8 56.78
Widowed/Separated/
6.55 11.75 81.71 25.19 15.44 59.37 4.01 10.58 85.4 12.89 16.58 70.53
Others
Primary school 26.64 19.07 54.28 59.09 14.78 26.13 20.19 19.28 60.53 38.91 22.17 38.92
Secondary school 7.21 9.95 82.84 27.62 14.97 57.41 4.63 9.79 85.58 13.29 15.91 70.8
College 0.24 1.28 98.48 2.18 5.94 91.88 0 0.66 99.34 0.2 3.25 96.55
Family size ≤ 3 4.27 10.06 85.67 21.23 17.87 60.9 3.12 8.3 88.59 8.5 18.01 73.49
3 < Family size ≤ 5 9.27 11.22 79.5 31.45 14.01 54.54 6.12 11.17 82.71 15.78 15.62 68.61
5 < Family size ≤ 7 18.44 13.73 67.83 44.36 13.28 42.37 13.89 13.87 72.24 27.76 18.18 54.06
7 < Family size ≤ 9 28.29 19.59 52.12 60.96 13.24 25.81 21.6 19.12 59.28 40.77 22.24 36.99
Family size > 9 33.65 17.26 49.09 60.46 11.45 28.1 24.7 21.31 53.99 48.43 15.71 35.86
Hhld head's age ≤ 35 23.12 14.58 62.3 50.58 11.77 37.66 18.27 15.69 66.05 33.07 18.18 48.75
35 < Hhld head's age ≤ 44 17.59 13.41 69 44.43 13.82 41.75 12.09 14.16 73.74 26.98 18.1 54.92
Hhld head's age > 44 9.76 12.53 77.71 30.61 15.51 53.88 6.89 11.39 81.72 16.57 17.01 66.43
Non-agriculture hhld 7.06 10.39 82.55 26.37 14.85 58.79 4.68 9.23 86.09 13.49 15.74 70.77
Agriculture hhld 38.36 21.37 40.27 76.19 11.99 11.82 29.42 24.45 46.12 51.96 22.83 25.21
Main source of income:
18.49 14.17 67.34 42.65 14.11 43.24 13.95 14.43 71.62 26.37 18.08 55.55
entrepreneurial income
38.39 22.16 39.45 80.55 9.1 10.35 29 26.62 44.38 54.92 21.21 23.87
Agricultural wage/salary
Main source of income: Non-
7.15 10.53 82.32 27.89 15.05 57.06 4.56 9.27 86.17 14.38 16.36 69.26
agricultural wage/salary
140
Appendix Table A5.2 Intertemporal Poverty Headcount Rate using the Spells Approach (%),
by Household Characteristics
Rural 27.2 16.65 56.15 59.72 12.81 27.47 22.64 15.85 61.52 40.98 17.48 41.54
Urban 5.57 7.34 87.08 22.3 11.83 65.86 4.41 5.61 89.98 12.23 11.31 76.46
NCR 0.24 0.85 98.91 6.27 9.14 84.6 0.24 0.65 99.11 2.79 8.27 88.94
Luzon 9.78 10.02 80.2 32.43 14.57 52.99 7.89 8.04 84.07 21.08 13.79 65.13
Visayas 23.29 16.45 60.26 53.42 11.57 35.01 19.36 15.21 65.44 33.03 16.27 50.7
Mindanao 30.61 16.7 52.7 62.23 9.83 27.94 25.5 16.43 58.07 42.53 16.58 40.89
Male-headed hhld 18.06 12.72 69.22 44.17 12.59 43.24 15.02 11.29 73.69 29.06 14.78 56.16
Single 3.74 10.78 85.49 17.47 12.3 70.23 3.74 6.35 89.91 6.84 13.97 79.18
Married 18 12.25 69.74 43.66 12.15 44.19 15.01 11.04 73.95 28.79 14.43 56.78
Widowed/Separated/
7.42 10.87 81.71 26.91 13.71 59.37 4.88 9.72 85.4 14.84 14.63 70.53
Others
Primary school 27.94 17.77 54.28 60.96 12.92 26.13 23.72 15.75 60.53 43.03 18.05 38.92
College 0.24 1.28 98.48 2.61 5.51 91.88 0 0.66 99.34 0.18 3.26 96.55
Family size ≤ 3 5.31 9.02 85.67 24.1 15 60.9 4.14 7.28 88.59 12.51 14 73.49
3 < Family size ≤ 5 10.28 10.22 79.5 33.46 12 54.54 8.07 9.23 82.71 18.42 12.97 68.61
5 < Family size ≤ 7 19.89 12.28 67.83 46.05 11.59 42.37 16.94 10.83 72.24 31.19 14.75 54.06
7 < Family size ≤ 9 29.27 18.62 52.12 61.7 12.5 25.81 25.51 15.21 59.28 44.35 18.66 36.99
Family size > 9 35.92 14.99 49.09 60.8 11.11 28.1 27.42 18.59 53.99 50.55 13.6 35.86
Hhld head's age ≤ 35 24.99 12.71 62.3 52.59 9.76 37.66 22.13 11.83 66.05 37.34 13.91 48.75
35 < Hhld head's age ≤ 44 18.51 12.49 69 46.3 11.95 41.75 15.02 11.24 73.74 29.86 15.22 54.92
Hhld head's age > 44 10.82 11.47 77.71 32.17 13.95 53.88 8.28 10 81.72 19.31 14.26 66.43
19.71 12.95 67.34 44.26 12.5 43.24 16.12 12.26 71.62 29.32 15.13 55.55
41.79 18.76 39.45 80.48 9.17 10.35 37.51 18.11 44.38 58.25 17.87 23.87
7.94 9.74 82.32 30.19 12.75 57.06 6.14 7.69 86.17 17.77 12.97 69.26
141
Appendix Table A5.3 Intertemporal Poverty Headcount Rate using the JR approach (%), by Region
Permanent US$1.25 US$2 half of median government/official
Location
Income (US$) Persistent Transient Total Persistent Transient Total Persistent Transient Total Persistent Transient Total
NCR 2,244.39 0.00 0.44 0.44 3.10 4.61 7.72 0.00 0.38 0.38 1.75 3.11 4.86
CAR 1,092.29 7.10 7.64 14.74 38.11 8.65 46.76 4.56 5.66 10.22 13.77 9.04 22.81
Region 1 1,222.25 4.53 4.81 9.33 25.99 7.74 33.73 2.72 4.40 7.12 15.57 7.57 23.13
Region 2 1,243.33 6.43 5.58 12.01 30.22 6.85 37.07 3.93 4.87 8.80 11.76 7.86 19.62
Region 3 1,384.29 0.92 2.28 3.20 14.07 7.01 21.08 0.70 1.85 2.55 8.06 6.35 14.41
Region 4-A 1,554.14 4.18 2.45 6.63 17.99 4.60 22.59 2.46 2.16 4.62 10.90 5.15 16.05
Region 4-B 810.75 25.27 8.14 33.41 59.25 6.49 65.75 21.23 8.21 29.44 34.95 9.04 44.00
Region 5 1,091.79 19.68 6.50 26.18 48.06 7.26 55.32 14.64 6.66 21.29 31.08 9.73 40.81
Region 6 1,075.84 12.19 7.45 19.64 42.59 6.06 48.66 7.19 8.92 16.10 19.86 7.51 27.37
Region 7 965.06 22.41 6.80 29.21 51.46 3.53 54.98 18.53 7.39 25.92 33.08 6.80 39.89
Region 8 1,056.29 18.02 7.85 25.87 47.49 6.38 53.88 13.83 7.29 21.11 24.59 11.51 36.10
Region 9 935.81 35.01 5.98 40.99 60.59 4.31 64.90 30.17 7.03 37.20 40.15 6.72 46.87
Region 10 1,077.15 21.76 5.80 27.56 48.02 4.04 52.06 16.63 6.95 23.58 34.95 5.67 40.63
Region 11 1,011.30 18.54 6.54 25.08 47.17 5.28 52.45 14.84 5.85 20.69 28.10 6.78 34.89
Region 12 885.82 12.38 9.72 22.10 46.43 7.60 54.04 8.04 9.47 17.51 28.89 9.84 38.74
ARMM 553.51 34.22 10.12 44.34 82.60 2.47 85.08 24.51 12.77 37.28 36.87 14.19 51.06
Caraga 755.41 24.73 7.78 32.51 60.39 7.50 67.89 18.47 9.35 27.82 42.77 8.81 51.57
Source: Author’s computations using household expenditure per capita data from the longitudinal subsample of FIES 2003, 2006 and 2009
142
Appendix Table A5.4 Standard Errors of Intertemporal Poverty Headcount Rate
using the Components Approach (%), by Household Characteristics
Rural 0.83 0.70 0.91 0.88 0.60 0.79 0.76 0.72 0.90 0.90 0.73 0.88
Urban 0.54 0.65 0.81 0.96 0.81 1.11 0.46 0.59 0.73 0.76 0.80 1.02
NCR 0.00 0.56 0.56 1.23 1.86 2.13 0.00 0.52 0.52 0.85 1.70 1.86
Luzon 0.66 0.69 0.90 1.02 0.81 1.11 0.57 0.64 0.82 0.88 0.81 1.06
Visayas 1.32 1.22 1.54 1.56 1.02 1.48 1.17 1.24 1.51 1.46 1.25 1.56
Mindanao 1.24 1.01 1.32 1.28 0.78 1.18 1.15 1.04 1.31 1.31 1.02 1.29
Male-headed hhld 0.58 0.53 0.72 0.77 0.55 0.78 0.52 0.52 0.68 0.69 0.59 0.77
Single 1.38 3.05 3.31 3.42 3.45 4.61 1.33 2.26 2.61 1.89 3.87 4.19
Married 0.59 0.52 0.72 0.77 0.54 0.78 0.52 0.52 0.68 0.69 0.59 0.78
Widowed/Separated/
0.98 1.34 1.59 1.77 1.39 1.99 0.76 1.28 1.44 1.44 1.44 1.86
Others
Primary school 0.95 0.82 1.05 1.03 0.74 0.93 0.87 0.80 1.03 1.04 0.85 1.02
College 0.24 0.52 0.57 0.65 1.23 1.37 0.00 0.35 0.35 0.20 0.87 0.89
Family size ≤ 3 0.63 0.97 1.12 1.25 1.23 1.54 0.54 0.90 1.02 0.87 1.25 1.42
3 < Family size ≤ 5 0.62 0.70 0.89 1.04 0.78 1.14 0.52 0.69 0.83 0.82 0.80 1.05
5 < Family size ≤ 7 1.04 0.89 1.26 1.36 0.95 1.39 0.93 0.88 1.19 1.21 1.02 1.37
7 < Family size ≤ 9 1.92 1.70 2.18 2.16 1.50 1.96 1.75 1.64 2.12 2.13 1.86 2.14
Family size > 9 3.21 2.54 3.45 3.40 2.29 3.12 2.93 2.73 3.42 3.44 2.48 3.34
Hhld head's age ≤ 35 1.27 1.01 1.45 1.51 0.96 1.48 1.17 1.04 1.41 1.42 1.11 1.51
35 < Hhld head's age ≤ 44 1.08 0.94 1.30 1.40 0.98 1.40 0.94 0.96 1.24 1.26 1.07 1.41
Hhld head's age > 44 0.58 0.67 0.83 0.92 0.73 1.01 0.50 0.63 0.76 0.75 0.76 0.96
0.78 0.66 0.93 0.97 0.66 0.99 0.71 0.67 0.89 0.89 0.73 0.98
2.35 1.94 2.29 1.81 1.26 1.42 2.23 2.09 2.34 2.34 1.87 1.95
0.62 0.75 0.92 1.08 0.86 1.18 0.51 0.68 0.82 0.86 0.89 1.11
143
Appendix Table A5.5 Standard Errors of Intertemporal Poverty Headcount Rate
using the Spells Approach (%), by Household Characteristics
Rural 0.84 0.67 0.91 0.87 0.57 0.79 0.80 0.66 0.90 0.91 0.67 0.88
Urban 0.56 0.63 0.81 0.99 0.77 1.11 0.50 0.56 0.73 0.82 0.74 1.02
NCR 0.24 0.50 0.56 1.48 1.69 2.13 0.24 0.46 0.52 1.04 1.61 1.86
Luzon 0.68 0.67 0.90 1.04 0.78 1.11 0.62 0.60 0.82 0.94 0.73 1.06
Visayas 1.35 1.18 1.54 1.55 0.97 1.48 1.27 1.14 1.51 1.49 1.18 1.56
Mindanao 1.26 0.96 1.32 1.27 0.73 1.18 1.20 0.96 1.31 1.31 0.95 1.29
Male-headed hhld 0.60 0.51 0.72 0.77 0.52 0.78 0.56 0.48 0.68 0.71 0.54 0.77
Single 1.57 2.96 3.31 3.55 3.32 4.61 1.57 2.10 2.61 2.03 3.82 4.19
Married 0.60 0.50 0.72 0.78 0.52 0.78 0.56 0.48 0.68 0.71 0.54 0.78
Widowed/Separated/
1.06 1.29 1.59 1.80 1.31 1.99 0.85 1.22 1.44 1.50 1.37 1.86
Others
Primary school 0.96 0.80 1.05 1.03 0.71 0.93 0.91 0.74 1.03 1.05 0.79 1.02
College 0.24 0.52 0.57 0.71 1.20 1.37 0.00 0.35 0.35 0.18 0.87 0.89
Family size ≤ 3 0.69 0.93 1.12 1.32 1.15 1.54 0.61 0.86 1.02 1.06 1.13 1.42
3 < Family size ≤ 5 0.66 0.67 0.89 1.05 0.74 1.14 0.59 0.63 0.83 0.87 0.73 1.05
5 < Family size ≤ 7 1.07 0.85 1.26 1.38 0.88 1.39 1.00 0.79 1.19 1.26 0.93 1.37
7 < Family size ≤ 9 1.93 1.69 2.18 2.16 1.50 1.96 1.85 1.51 2.12 2.16 1.75 2.14
Family size > 9 3.27 2.37 3.45 3.39 2.25 3.12 3.02 2.61 3.42 3.45 2.29 3.34
Hhld head's age ≤ 35 1.30 0.95 1.45 1.51 0.88 1.48 1.25 0.93 1.41 1.45 0.99 1.51
35 < Hhld head's age ≤ 44 1.10 0.91 1.30 1.41 0.93 1.40 1.01 0.87 1.24 1.30 1.00 1.41
Hhld head's age > 44 0.60 0.65 0.83 0.94 0.71 1.01 0.54 0.59 0.76 0.80 0.70 0.96
0.80 0.63 0.93 0.98 0.63 0.99 0.75 0.63 0.89 0.91 0.68 0.98
2.37 1.82 2.29 1.84 1.34 1.42 2.34 1.82 2.34 2.31 1.78 1.95
0.65 0.73 0.92 1.10 0.81 1.18 0.58 0.63 0.82 0.94 0.80 1.11
144
Appendix Table A5.6 Standard Errors of Intertemporal Poverty Headcount Rate using the JR approach (%), by Region
US$1.25 US$2 half of median government/official
Location
Persistent Transient Total Persistent Transient Total Persistent Transient Total Persistent Transient Total
NCR 0.00 0.56 0.56 1.23 1.86 2.13 0.00 0.52 0.52 0.85 1.70 1.86
CAR 2.10 2.89 3.26 3.50 2.76 3.32 1.74 2.62 2.97 2.79 2.83 3.43
Region 1 1.65 1.86 2.33 2.60 2.16 2.68 1.38 1.82 2.18 2.37 2.10 2.69
Region 2 1.69 2.24 2.60 2.86 2.05 2.91 1.44 2.02 2.36 2.18 2.36 2.83
Region 3 0.66 1.31 1.44 1.90 1.82 2.32 0.50 1.19 1.28 1.44 1.80 2.13
Region 4-A 1.16 1.21 1.61 1.97 1.59 2.24 0.96 1.07 1.40 1.72 1.43 2.06
Region 4-B 3.34 2.77 3.49 3.20 2.27 2.66 3.15 2.90 3.51 3.51 2.73 3.33
Region 5 2.63 2.15 2.85 2.82 2.12 2.51 2.44 2.08 2.79 2.79 2.39 2.76
Region 6 1.93 1.92 2.39 2.47 1.76 2.34 1.52 2.05 2.32 2.15 2.00 2.47
Region 7 2.34 2.04 2.62 2.61 1.42 2.50 2.19 2.00 2.58 2.59 1.92 2.62
Region 8 2.65 2.50 3.06 3.05 2.19 2.83 2.41 2.40 2.98 2.77 2.80 3.03
Region 9 3.22 2.28 3.23 3.07 1.84 2.79 3.14 2.36 3.26 3.24 2.39 3.17
Region 10 2.94 2.14 3.13 3.11 1.64 3.01 2.71 2.31 3.09 3.13 1.98 3.10
Region 11 2.62 2.15 2.97 3.04 1.87 2.92 2.50 2.08 2.90 2.89 2.19 3.05
Region 12 2.47 2.66 3.04 3.01 2.11 2.72 2.04 2.63 2.95 2.97 2.56 2.95
ARMM 3.41 2.94 3.09 2.34 1.37 1.99 3.28 3.12 3.23 3.39 3.21 2.81
Caraga 3.23 2.64 3.40 3.21 2.45 2.63 3.00 2.74 3.39 3.41 2.67 3.08
145
Chapter 6 Who are Income Mobile?
6.1 Introduction
Chapter 5 has discussed mobility among low income range people and found that poverty
in the Philippines is mostly persistent in nature. This chapter extends the discussion by
examining the mobility patterns of people from other income segments. Here, mobility is
defined in terms of how fast income is growing. The discussion focuses on the relationship
between income inequality and income mobility by identifying the characteristics of those who
benefitted from economic growth through high positive income mobility and of those who were
left out because they experienced negative income mobility. This is an important analytical
exercise because as explained in the previous chapters, the high levels of inequality in the
Philippines could imply that different population groups benefit from economic growth at
different rates. In particular, I address the following questions:
(i) What are the characteristics of the income mobile households in the Philippines?
(ii) How do the income mobility outcomes of initially advantaged and initially
disadvantaged households differ?
In answering the first question, I will identify who has benefited from economic
development and who has been left out. I will also examine whether the initially disadvantaged
groups caught-up with the initially advantaged ones through faster income growth rates. For
the second research question, I will examine whether the impact of economic development on
initially advantaged and initially disadvantaged households differ when the country’s economy
is expanding or contracting.
Investigating how the benefits of economic growth accrue to different groups is usually
addressed by comparing the income growth rates of different segments of the population.
Examining the growth incidence curves (GIC) provides a good starting point for this analytical
exercise. We have learned earlier that a growth process is considered relatively pro-poor if it
allows the poor to catch-up with the non-poor through faster income growth rates resulting in
a downward sloping GIC. While several studies briefly examined this issue in the Philippines
(e.g. Balisacan & Pernia, 2002; Pernia, 2003; Schelzig, 2005; Aldaba 2009), most used data
from repeated cross-sectional surveys. Chapter 1 identified that the problem with working with
cross-sectional surveys is that when individuals are ranked according to their income in each
time period, the composition of a particular income quantile in the initial time period will not
146
exactly match the composition of the same quantile in a subsequent time period because people
move from one quantile to another over time. This process makes it difficult to infer whether
the initially poor or initially non-poor experienced faster income growth rates (Grimm 2007).
Another potential disadvantage of the GIC is that it implicitly ignores what happened in
between the start and end of the observation period. The analyses in this chapter address the
limitations of previous studies in several ways. First, I depart from the conventional approach
of measuring income mobility using growth rates between initial and final-period incomes
(Tabunda & Albert 2002; Reyes et al. 2011; Takahashi 2013) by incorporating the available
information between the start and end of the observation periods. This allows me to distinguish
people who experienced consistently positive or negative mobility from people who
experienced unstable income flows, an issue that has not been explored in previous studies
(Reyes et al. 2011; Takahashi 2013). Second, I go beyond the conventional approach of
examining pro-poorness of growth by testing the convergence (vs. divergence) and symmetry
(vs. asymmetry) of household income mobility following the approach of Fields et al. (2007).
Similar to the concept of pro-poor growth, the concepts of convergence and symmetry refer to
the effect that income mobility patterns over time have on the differences in income between
the initially advantaged and initially disadvantaged people. From a policy perspective
understanding these patterns would help us gauge the extent to which the high income
inequality in the country is a reflection of inequitable distribution of socio-economic
opportunities.
6.2 Different Patterns of Income Mobility

Variations in the effect of economic growth on people’s living standards can be explained
by differences in their socio-demographic characteristics, resource endowment, skills, risk
aversion, effort and luck (Morrisson 2006; Marrero & Rodriguez 2013; Ros 2013). As pointed
out earlier, there is more concern among policymakers when the observed inequality portrays
inequality of opportunities rather than inequality of outcomes. Inequality of opportunities could
lead to long episodes of segmentation between the advantaged and the disadvantaged groups
and thus, can undermine the country’s full economic potential (Braham, Rattansi & Skellington
1992; Pasha & Palanivel 2003) whereas if socio-economic opportunities are distributed
equally, inequality of outcomes would only arise due to variation in efforts (Arrow, Bowles &
147
Durlauf 2004; Kenworthy 2004).59 Hence, despite diversity being woven in the fabric of the
socio-economic development process, there is much interest in understanding what causes
socio-economic inequalities, especially in a developing country like the Philippines where
rapid economic growth is accompanied by persistently high income inequality.
To determine the extent to which income inequality in the Philippines is characterized by
inequality of opportunities, it is important to examine how the incomes of different population
groups with varying levels of (initial) advantage change over time. In this context, the income
mobility process can be generally classified as (i) convergent or divergent; and (ii) symmetric
or asymmetric. Income mobility is said to be convergent when incomes of the initially
disadvantaged are growing at least as fast as their initially advantaged counterparts and it is
divergent when the initially disadvantaged group receives disproportionately less benefits from
the observed mobility process (Shorrocks & van der Hoeven 2004; Grimm et al. 2007).
There are several factors that can contribute to convergent income mobility. For instance,
if economic growth expands the access of initially disadvantaged to credit markets, then the
additional capital can unleash the growth potential of the poor leading to faster income growth
rates. Similarly, macro-level policies on government spending and progressive taxation may
also contribute to faster income growth rates among the poor (Pintus 2008). Analogously, a
divergent income mobility process could be attributed to capital market imperfection wherein
the initially disadvantaged systematically confronts borrowing constraints which in turn,
prevents them from reaping the benefits of economic growth (Galord 1996; Banerjee & Duflo
2003; Ravallion 2012). On the other hand, the movement of the additional capital created by
growth could also be perfectly fluid in which case, a person’s initial resources will not have a
significant effect on his/her subsequent income growth. In general, convergent income mobility
can be linked to the concept of pro-poor growth while divergent mobility can be associated
with poverty traps and cumulative advantage.
Solely relying on a converging income mobility process cannot guarantee that the poor
will have adequate resources such as financial capital, education and employment that would
assure that they will never experience poverty again. In particular, even if the income mobility
pattern is convergent (or divergent) for a specific time period, it is not always the case that the
same pattern will persist over time. For instance, a convergent income mobility spell may be
followed by a divergent income mobility spell, or vice-versa. Hence in addition to convergence,
59Kenworthy (2004) examined the hypothesis that economic growth is always accompanied by increasing inequality. He
concluded that there is no necessary trade-off between equality and economic growth as long as an optimal balance of policy
options are combined to create a fair economy.
148
it is also important to examine the symmetry of income mobility to be able to understand how
people’s income mobility prospects change over time. A mobility process is said to be
symmetric when the group that experienced better (inferior) income mobility outcomes during
a specific time period experiences inferior (better) mobility outcomes in the subsequent period.
A good example of a symmetric income mobility process is when the rich benefits
disproportionately more during episodes of economic growth but they also lose more during
episodes of economic turmoil (Fields et al. 2007). It can be observed during financial markets-
induced crises when the rich bear its negative impact more than the poor because of their higher
exposure to credit markets.60 Analogously, an income mobility process could also be
considered symmetric when the poor benefits more during episodes of economic growth but
they also lose more during periods of economic uncertainties probably because they have
limited access to social safety nets that can cushion them from large income losses.
Testing whether income mobility is converging or diverging and whether it is symmetric
or asymmetric will help us understand how the economic development processes in the
Philippines allow the initially disadvantaged to catch up with the rich, or whether these
processes systematically exclude them from reaping the benefits of economic growth.
6.3. Methods
6.3.1 Classifying Households According to Income Mobility Trajectories
In this chapter, income mobility is measured in terms of changes in log per capita
household consumption derived from the FIES. Convergence and symmetry of mobility are
gauged in terms of how fast people’s incomes are growing with respect to its initial levels.
Instead of simply looking at growth rates from 2003 to 2009, I estimate the growth rates from
2003 to 2006 and 2006 to 2009, separately to unmask interesting features about the household
income flows that may otherwise be hidden if I simply look at the income differences between
2003 and 2009. Compared to the approach used in Chapter 1, this analytical strategy further
capitalizes on the longitudinal feature of the data. For instance, it is possible that some
households that experienced high income growth rates between 2003 and 2009 also
experienced very volatile income flows. As noted in Chapter 1, this is not necessarily a
desirable outcome especially when households are averse to income fluctuations. Furthermore,
it is possible that a high (low) income growth observed in 2006-2009 might offset a low (high)
income growth observed in 2003-2006 which in turn may be mistakenly classified as
60Whether the rich or the poor suffer more during economic crises is a debatable issue. Some argue that the poor suffer more
because the rich are more likely to be compensated by government bail outs (Halac & Schmukler 2004).
149
immobility if one simply relies on the income growth rate in 2003-2009. Estimating the growth
rates for 2003-2006 and 2006-2009 separately also allows me to examine how income mobility
changes over time and thus, test whether it is symmetric or asymmetric.
Figure 6.1 shows the top view of the density plot of the annualized growth rates between
2003 and 2006 in the x-axis and the annualized growth rates between 2006 and 2009 in the y-
axis. The plot reveals a negative correlation between the two sets of growth rates, i.e., faster
income growth between 2003 and 2006 tends to be followed by slower income growth between
2006 and 2009, and vice-versa. It also shows that the density peaks near the origin which means
that a significant fraction of the households experienced consistently slow income growth from
2003 to 2009.
Figure 6.1 Joint Distribution of Income Trajectories, 2003-2006 and 2006-2009
45
Annualized income growth (%) 2006-2009
30
15
-15
-30
-45
-45 -30 -15 0 15 30 45
Annualized income growth (%), 2003-2006
I follow a heuristic approach in grouping households that have homogeneous income

mobility trajectories. In particular, households that experienced slow to moderate income
growth (at most +/- 5% per year) in both 2003-2006 and 2006-2009 periods are grouped in the
first cluster.61 Households that observed consistently positive or consistently negative growth
rates, wherein at least one growth rate exceeds 5%, are classified under the second or third
cluster, respectively. Lastly, households that experienced highly positive income growth (>5%)
in 2003-2006 yet highly negative income growth (< -5%) in 2006-2009 are classified in the
61
The median absolute income growth rate for 2003-2006 and 2006-2009 is about 9% per year.
150
fourth cluster while households that experienced highly negative growth (< -5%) in 2003-2006
followed by highly positive growth (> 5%) in 2006-2009 are classified in the fifth cluster. As
illustrated in Figure 6.2, the first cluster corresponds to households with very modest income
growth. The second and third clusters include households that experienced consistently upward
and downward mobility, respectively. The last two clusters correspond to households that
experienced high transitory income fluctuations. Section 6.3.4 provides the details about the
empirical strategy on how these clusters are used to test convergence and symmetry of income
mobility.
Figure 6.2 Different Types of Income Trajectories, 2003-2009
Cluster1 Cluster2 Cluster3
2003 2006 2009 2003 2006 2009 2003 2006 2009
Cluster4 Cluster5
2003 2006 2009 2003 2006 2009
6.3.2 Measures of Socio-economic Advantage

Since the objective of this study is to examine the extent to which a household’s initial
level of socio-economic advantage predicts its subsequent income growth trajectory, it is
essential to provide a measure of socio-economic advantage. To do this, I group the households
using two methods. First, I use the quintiles of the observed income in 2003. In general,
grouping households according to quantiles is a common approach in income distributional
analysis (RC 2003). However, although this approach is useful for capturing how income is
appropriated into different segments of the society, it is unable to capture polarization or the
implicit clustering of individuals into groups (Chakravarty & Ambrosio 2010). While both
income inequality and polarization are concerned of the variability of the income distribution,
high income inequality does not always imply a “divided” or “polarized” society (Gochoco-
151
Bautista, Bautista, Maligalig & Sotocinal 2013).62 Thus, in addition to examining inequality, it
is also important to study polarization because a segmented society is usually prone to conflict
due to skewed distribution of opportunities (Gasparini et al. 2008). To capture polarization, I
follow the approach proposed by Liao (2006) which entails fitting latent cluster models on
initial income in 2003.63 Model-based clustering is one of the sophisticated statistical tools that
has been increasingly used by researchers to stratify population units based on various
characteristics of interest. Unlike conventional clustering methods, model-based clustering
assumes that the underlying population is made up of different clusters, each following a
different probability distribution (Stahl & Sallis 2012). In other words, the data is assumed to
be a realization from a specific mixture probability density function and this reduces the
clustering task into estimating the parameters of the assumed mixture distribution. One of the
main advantages of using this approach in empirical application is that it allows researchers to
find optimal clusters even with limited prior information about how the units are clustered in
theory (Vermunt & Magidson 2002). Compared to conventional clustering methods, model-
based clustering uses a less arbitrary approach in minimizing within-cluster and maximizing
between cluster-variations (Vermunt & Magidson 2002; Liao 2006). Furthermore, unlike group
membership according to quintiles, the choice of the optimal number of clusters in model-
based clustering is less arbitrary because it is based on the values of the Bayesian Information
Criterion computed from different candidate models.
In empirical studies, it is common to find initial incomes to be negatively correlated with
subsequent income growth (Khor & Pencavel 2008). However, as pointed out in Chapter 1,
income data from household surveys is usually subject to measurement errors (Fields et al.
2003, Forbes 2000, Khor & Pencavel 2008) and if left unaddressed, may lead to spurious
correlation between income mobility and initial income. For instance, underestimated initial
incomes may lead to mean reversion and the process would erroneously portray a convergent
income mobility. To address this issue, I also consider the household’s permanent income as
an alternative monetary measure of advantage. For each household, I approximate permanent
income by taking its longitudinal average income from 2003, 2006 and 2009.
62 For example, for an n-individual society where one individual has Z units of income (Z > n-1) while each of the n-1
individuals has one unit of income only, the resulting inequality will be very high but polarization is low. Liao (2006) provided
a more detailed discussion on how the notion of polarization can produce different trends of income variability than Gini-
based measures of inequality.
63 I used the Mclust package available in R in estimating latent cluster models.
152
6.3.3 Other Correlates of Income Mobility
In examining the impact of initial socio-economic advantage on income mobility, I
control for the effect of gender, education, location as well as demographic and economic
events. A demographic event is defined as changes in household composition while economic
events refer to changes in income sources. Demographic events may affect income mobility
systematically in several ways. First, it is a common assumption that individuals within a
household pool their resources. Thus, income is expected to move in the same direction as the
change in the number of household members who are engaged in paid employment. However,
if income is fixed but the number of non-working members increases, then the need to allocate
the pooled resources among more people might eventually manifest as negative income
mobility. This is because these households are likely to have less savings. In turn, lower savings
implies higher vulnerability to unexpected income shocks arising from illness, unemployment,
among other factors that lead to reduced income flows. In general, the extent of negative effect
of an additional non-working member depends on his/her age. An additional dependent child
may limit the ability of women to engage in paid employment due to the amount of time needed
for child rearing, leading to a reduced income flow for a significant period of time. On the other
hand, the negative impact of an additional member could be less severe if the additional
member is of working age because he/she has the potential to contribute to generation of
additional income or provide unpaid work around the household should the need arise in the
future. Moreover, a significant portion of incomes of households from developing countries is
generated from earnings related to paid employment (Dicken 2011). However, income mobility
can’t be solely determined by the number of household members employed because each
working member may be employed in different economic sectors which offer varying levels of
income opportunities. In other words, the source of income is also an important factor that
could explain one’s income trajectories.
6.3.4 Statistical Models of Income Mobility

To examine the convergence and symmetry of the income mobility regime that transpired
in the Philippines over the past decade, I estimate a multinomial logistic model wherein the
dependent variable corresponds to the propensity to be classified in each of the five clusters
and the independent variables correspond to the different indicators of socio-economic
advantage, as shown in (6.1).
𝑐𝑙𝑢𝑠𝑡𝑒𝑟𝑗
𝑙𝑜𝑔 ( 𝑐𝑙𝑢𝑠𝑡𝑒𝑟1
) = 𝛽𝑗 𝑋𝑖𝑖𝑛𝑐 + 𝜃𝑗 𝑊𝑖𝑛 𝑛−𝑖𝑛𝑐
+ 𝑗 𝑍𝑖
𝑣 𝑛𝑡𝑠
(6.1)
153
where 𝑝𝑐𝑙𝑢𝑠𝑡 𝑟𝑗
denotes the probability of falling in cluster j = 2,.., 5, while 𝑋𝑖𝑖𝑛𝑐 denotes a
household’s initial monetary advantage, 𝑊𝑖𝑛 𝑛−𝑖𝑛𝑐
denotes a household’s initial non-
monetary advantage, 𝑍𝑖 𝑣 𝑛𝑡𝑠
corresponds to the various demographic and economic events. To
account for the potential varying impact when income is measured in terms of actual observed
income or permanent income, two variants of (6.1) are estimated:
𝑖𝑛𝑐 𝑣 𝑛𝑡𝑠
𝑙𝑜𝑔 ( 𝑐𝑙𝑢𝑠𝑡𝑒𝑟1
) = 𝛽𝑗 𝑋𝑖2003 + 𝜃𝑗 𝑊𝑖2003 + 𝑗 𝑍𝑖 (6.2)
𝑖𝑛𝑐 𝑣 𝑛𝑡𝑠
𝑙𝑜𝑔 ( 𝑐𝑙𝑢𝑠𝑡𝑒𝑟1 ) = 𝛽𝑗 𝑋𝑖𝑎𝑣 + 𝜃𝑗 𝑋𝑖2003 + 𝑗 𝑍𝑖 (6.3)
In the context of the hypotheses about income mobility described in the previous section,
the signs and the magnitude of the estimates for 𝛽𝑗 and 𝜃𝑗 after controlling for 𝑍𝑖 𝑣 𝑛𝑡𝑠
can be
used to determine whether income mobility is converging or diverging and whether it is
symmetric or asymmetric. Recall that the first cluster corresponds to nil income growth
throughout the observation period while the second and third cluster correspond to consistently
positive and consistently negative growth rates, respectively. The fourth and fifth clusters
correspond to a steep change in the income growth trajectories. Since convergence refers to the
initially disadvantaged group catching up with the initially advantaged group, then I can argue
that the income mobility in the Philippines is convergent throughout the past decade if the value
of either 𝛽2 or 𝜃2 is higher for the initially disadvantaged households than the initially
advantaged group. The mobility process can also be considered convergent if the value of either
𝛽3 or 𝜃3 is lower for the initially disadvantaged households than the initially advantaged group.
In other words, convergence occurs when the initially disadvantaged experience higher growth
than the initially advantaged, or the initially disadvantaged experience less decline in growth
than the initially advantaged. On the other hand, income mobility is said to be symmetric if the
values of either 𝛽4 , 𝜃4 , 𝛽5 or 𝜃5 are significantly different between the initially advantaged and
disadvantaged groups.

6.4.1 Trends in Income Inequality and Polarization
From the Lorenz curves derived from the distribution of income for each survey wave
and the distribution of the longitudinally-averaged income, we can see that over the past
154
decade, both cross-sectional and long-run inequality barely moved (Figure 6.3).64 More recent
studies also show that the country’s observed inequality has been accompanied by high levels
of polarization or stratification of individuals into different income segments (Gochoco-
Bautista et al. 2013). The results presented in Table 6.1 confirm this finding. The numbers
under the column labelled as “Total” correspond to the estimated value of the Gini coefficient
for each of the survey year while the numbers under columns labelled as “%within” and
“%between” correspond to the percentage share of the variability of incomes within and
between segments that were formed using latent cluster analysis to the total value of the Gini
coefficient, respectively. Here, I find that at least 70% of the observed cross-sectional
inequality and about 80% of long-run inequality can be attributed to polarization.
Which income source contributes significantly to the observed inequality? To answer
this question, I adopt the method proposed by Shorrocks (1982) which entails doing the
following steps. Suppose a household’s (total) income is denoted by Yi and Yik refers to the
income from the kth income source. Thus,
𝑌𝑖 = ∑𝑘 𝑌𝑖𝑘 (6.4)
Shorrocks denotes by sk the relative factor inequality weight or the proportion of income
inequality that can be attributed to the kth income source. Technically, Shorrocks showed that
sk is equal to the covariance between the total income and the income from kth source divided
by the variance of the total income, i.e.,
𝑣(𝑌,𝑌𝑘 )
𝑠𝑘 =
𝜎𝑌2
such that ∑𝑘 𝑠𝑘 = (6.5)
Table 6.2 presents the estimates of the factor inequality weight 𝑠𝑘 (multiplied by 100%)
for each income component. The results suggest that variations in employment income account
for approximately 85% of the total inequality. Interestingly, the contribution of variations in
income from self-employment or entrepreneurial income to total inequality seems to be
increasing over the years. This pattern is probably driven by the impact of the global financial
crisis (GFC) which started in 2008. As jobs were lost during the GFC, a significant fraction of
household earnings derived from wage employment shifted to entrepreneurial or self-
employment (Yap, Reyes and Cuenca, 2009). I turn to this issue in Chapter 8.
64
The value of the Gini coefficient is equal to the area below the line of perfect equality and above the Lorenz
curve wherein higher values suggest higher inequality.
155
Figure 6.3 Income Inequality in the Philippines, 2003-2009
1
.8
average
.6
2003
.4
2006
2009
.2
0
0 .2 .4 .6 .8 1
Percentiles (p)
6.4.2 Income Mobility and Inequality

In this section, I examine how initial advantage affects income mobility prospects. If
advantage is gauged in terms of income, the GIC is a good graphical tool for examining pro-
poorness of growth. As discussed in Chapter 1, there are two ways of deriving GICs. The first
approach entails comparing the income of a household with a certain income rank based on the
distribution of initial period incomes with another household with the same income rank based
on the distribution of final period incomes. The resulting curve is the conventional GIC. The
alternative approach is to derive the IGIC by computing the income growth rates of the same
respondents and plotting these growth rates with the quantiles of the initial income. The solid
lines in Figure 6.4 represent the IGICs while the broken lines represent the conventional GICs.
Since the slopes for the IGICs are more negative than the slope of the conventional GICs, it
implies that the development process has worked to the advantage of the initially poor more
than what we can perceive based on conventional GICs. This is consistent with the result of
the simple simulation experiment using pseudo-panel data presented in Chapter 2 which
suggests that IGIC is likely to portray a more (relative) pro-poor growth than the conventional
GIC.
156
Table 6.1 Decomposition of Inequality by Income Clusters
2003 2006 2009 Permanent Income
Location
Total %within %bet Total %within %bet Total %within %bet Total %within %bet
Philippines 42.84 30.00 70.00 44.28 15.68 84.32 42.15 28.96 71.04 41.14 15.96 84.04
Urban 39.26 38.62 61.38 42.18 43.89 56.11 40.38 27.89 72.11 38.43 28.73 71.27
Rural 39.28 28.47 71.53 38.8 27.09 72.91 37.43 27.36 72.64 36.28 27.68 72.32
NCR 37.12 38.95 61.05 42.78 39.81 60.19 38.65 46.6 52.4 36.93 33.37 66.63
Luzon 39.66 28.47 71.53 40.66 28.34 71.66 38.63 28.53 71.47 37.55 27.94 72.06
Visayas 41.73 37.14 62.86 42.98 38.68 61.32 42.40 26.04 73.96 40.36 26.01 73.99
Mindanao 41.96 27.25 72.75 42.02 26.01 73.79 41.70 27.45 72.55 39.89 27.02 72.98
Table 6.2 Decomposition of Inequality by Income Source

Permanent
Income Source 2003 2006 2009
Income
Wage income 48.45 40.04 29.28 41.88
Entrepreneurial income 38.56 43.39 62.97 45.63
Asset income 7.34 5.43 3.31 5.87
Income from transfers 0.44 1.88 0.92 1.07
Remittance income 4.39 8.15 2.93 4.86
Other income 0.82 1.10 0.59 0.68
Total 100.00 100.00 100.00 100.00
Source: Author’s computations using household income per capita data from the longitudinal
157
One of the main limitations of using IGIC is that it examines only two income vectors at
a time. As explained earlier, this can be problematic if one wants to differentiate households
that have experienced volatile income flows from households that have experienced more
stable income changes. Table 6.3 summarizes how different levels of income growth rates are
distributed in each time period. If short-distance move is defined as absolute income growth
rate of less than 5%, it would account for less than one-third of the total observed mobility in
2003-2006 as well as in 2006-2009. On the other hand, medium-distance moves or absolute
income growth rates between 5% to 20% account for more than half of the observed mobility
while long-distance moves or absolute income growth rates exceeding 20% contribute to about
13% of the total observed mobility in 2003-2006 and 2006-2009. Interestingly, the distribution
of growth rates in 2003-2009 is less varied wherein about half of the observed mobility is
characterized by short-distance moves, 48% are medium-distance moves and only 2% are long-
distance moves. A possible reason for this is that the 2003-2006 growth rates offset the 2006-
2009 growth rates. Table 6.4 provides evidence for this hypothesis by showing that there is a
non-negligible number of households that experienced consistently positive or consistently
negative growth rates. Overall, positive and negative changes in household income are both
common throughout the observation period suggesting that the development process has
created both “winners” and “losers”.
6.4.3 Testing Convergence, Divergence and Symmetry of Income Mobility

Using Income as a Measure of Advantage
Tables 6.5 and 6.6 show the distribution of income trajectories from 2003 to 2009, by
income quintile and income cluster. When initial income in 2003 is used, the latent cluster
analysis produced two clusters which I labelled as “Poor” and “Non-poor” in Table 6.5 and
when longitudinally-averaged income is used, the method produced three clusters which I
labelled as “Poor”, “Middle” and “Rich” in Table 6.6. If initial (monetary) advantage was
independent of income mobility, the expected value in each cell should be approximately the
same as the overall distribution of income trajectories depicted in Table 6.4. However, the
results are characterized by mixed patterns. For instance, if households are grouped according
to actual income in 2003, I find that the middle 60% households were more likely to be
classified under the first cluster than the poorest 20% and richest 20% households. In terms of
the groups formed by latent clustering method, I find that the poor are significantly more likely
to fall in the second cluster while the non-poor are significantly more likely to fall in the third
158
Table 6.3 Distribution of Income Growth Rates (%)
annualized growth (g) 2003-2006 2006-2009 2003-2009
-20% ≥ g 7.07 5.56 0.55
-20% < g ≤ -10% 16.21 11.39 7.36
-10% < g ≤ -5% 14.12 10.43 13.72
-5% < g ≤ 5% 29.7 31.88 50.52
5% < g ≤ 10% 12.42 14.86 17.09
10% < g ≤ 20% 13.94 17.99 9.97
20% < g 6.55 7.9 0.79
Total 100 100 100
Table 6.4 Distribution of Income Trajectories (%)

Type of income trajectory %Population
Cluster 1: slow to moderate growth 10.64
Cluster 2: generally positive income growth 30.84
Cluster 3: generally negative income growth 24.48
Cluster 4: high positive growth in 2003-2006, high
14.11
negative growth in 2006-2009
Cluster 5: high negative growth in 2003-2006, high
19.93
positive growth in 2006-2009
Total 100
Table 6.5 Distribution of Income Trajectories (%), by Segments of Initial Income

Cluster Cluster Cluster Cluster Cluster Total%
Group
1 2 3 4 5
(Initial) quintile 1 9.75 54.38 9.16 11.40 15.31 100
Poor cluster 11.34 39.02 18.23 12.99 18.41 100
Non-poor cluster 9.72 20.27 32.55 15.56 21.90 100
All 10.64 30.84 24.48 14.11 19.93 100
2003, 2006 and 2009.
159
Figure 6.4 Anonymous and Non-Anonymous Growth Incidence Curves, 2003-2009
20 2003-2006 20 2006-2009 20 2003-2009
15 15 15
10 10 10
5 5 5
0 0 0
-5 -5 -5
-10 -10 -10
-15 -15 -15
-20 -20 -20
0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100
percentile percentile percentile
160
group. On the other hand, when households are grouped according to longitudinally-averaged
income, it is the poorest 20% households who were most likely to be classified under the first
cluster.
Regardless whether households are grouped according to initial income in 2003,
longitudinally-averaged income or income quintile or clusters formed from latent cluster
analysis, the results suggest that the poorest group is more likely to be classified under the
second cluster than the rest of the population. In other words, the poorest (on various
definitions) experienced the best income mobility outcomes. On the other hand, I find mixed
patterns when looking at households that experienced consistently negative income growth
rates. In particular, when households are grouped according to initial income, the propensity to
be classified under the third cluster increases as one moves up the income ladder. However,
when households are grouped according to longitudinally-averaged income, middle income
households had the highest risk of experiencing consistently downward mobility. Lastly, I find
that the richest households based on initial income in 2003 were more likely to be classified
under the fourth and fifth clusters. However, when longitudinally-averaged income is used, the
rich households were more likely to be classified under the fourth cluster but the poor
households were more likely to be classified under the fifth cluster.
In terms of the income mobility patterns, the results provide empirical support for
(unconditional) convergence of mobility when households are grouped according to either
initial income in 2003 or longitudinally-averaged income, because poor households have the
highest probability to be in the generally positive income growth cluster while the non-poor
have the highest probability to be in the generally negative income growth cluster. In addition,
the results also provide evidence for (unconditional) symmetry of mobility when households
are grouped according to longitudinally-averaged income because this suggest that the rich
households were more likely to be in the high positive growth in 2003-2006 and high negative
growth in 2006-2009 cluster while the poor households were more likely to be in the high
negative growth in 2003-2006 and high positive growth in 2006-2009.
161
Table 6.6 Distribution of Income Trajectories (%), by Segments of Permanent Income
Cluster Cluster Cluster Cluster Cluster Total
Group
1 2 3 4 5 %
(Ave.) quintile 1 12.50 37.83 19.55 7.64 22.48 100
(Ave.) quintile 2 11.01 30.68 25.81 12.15 20.35 100
(Ave.) quintile 3 10.66 29.02 27.37 13.23 19.71 100
(Ave.) quintile 4 10.82 29.56 25.98 16.02 17.63 100
(Ave.) quintile 5 8.11 26.58 24.22 21.80 19.29 100
(Ave.) Poor cluster 11.58 34.81 22.52 9.58 21.50 100
(Ave.) Middle income 100
cluster 10.99 28.80 26.88 14.10 19.23
(Ave.) Rich cluster 9.12 27.80 24.65 19.82 18.62 100
All 10.64 30.84 24.48 14.11 19.93 100
2003, 2006 and 2009.
Table 6.7 Distribution of Income Mobility by Household Characteristics (%)

Group Cluster Cluster Cluster Cluster Cluster Total
(based on 2003 data) 1 2 3 4 5 %
Female-headed hhld 7.21 30.92 26.33 17.23 18.31 100
Male-headed hhld 11.15 30.83 24.20 13.65 20.17 100
Marital status of hhld head:
4.19 21.59 32.26 26.07 15.88
Single 100
Married 11.23 31.23 24.10 13.44 20.01 100
Widowed/Separated/Others 7.05 29.32 26.21 17.46 19.96 100
Educational attainment of
hhld head:
Primary education 10.88 32.01 23.11 12.68 21.33 100
Secondary education 10.59 29.62 26.22 14.75 18.81 100
Tertiary education 9.66 31.29 22.29 18.07 18.69 100
Family size: 1 to 3 7.77 18.71 38.79 15.44 19.30 100
4 to 5 12.29 25.96 25.79 15.22 20.75 100
6 to 7 10.47 35.47 20.73 13.67 19.66 100
8 to 9 10.56 40.00 17.21 12.59 19.63 100
10 or more 8.87 45.85 16.61 10.03 18.64 100
Rural 10.13 34.60 22.00 12.90 20.38 100
Urban 11.16 26.97 27.03 15.37 19.47 100
NCR 8.16 21.27 28.28 18.50 23.79 100
Luzon 11.08 29.61 25.94 14.65 18.72 100
Visayas 10.14 35.32 22.27 12.59 19.68 100
Mindanao 11.25 33.01 22.04 12.66 21.03 100
Non-Agriculture 10.48 28.89 26.28 15.19 19.16 100
Agriculture 11.07 36.27 19.48 11.11 22.07 100
All 10.64 30.84 24.48 14.11 19.93 100
2003, 2006 and 2009.
162
Socio-demographic variables
I use geographic location, household head’s sex, educational attainment, marital status,
household size and main source of income as control variables. As pointed in Chapter 1, these
variables are correlates of socio-economic well-being used in the existing literature. Table 6.7
shows the distribution of the types of income trajectories for each group (based on 2003 data).
Here, I find that single, female-headed households experienced more volatile income flows
than their married, male counterparts. The results also suggest that households with highly
educated heads were more likely to experience very high income growth rates in 2003-2006
but they also had the highest risk to incur high income losses in 2006-2009. In contrast,
households headed by primary educated individuals were more likely to experience high
income losses in 2003-2006 and high income gains in 2006-2009. Significant variations in
income mobility outcomes are also apparent when households are grouped according to
household size. In particular, households with more members were more likely to experience
consecutive episodes of upward mobility while smaller-sized households were more likely to
experience consecutive episodes of downward mobility. Furthermore, I also find that rural
agricultural households, especially those from Visayas and Mindanao had better income
mobility outcomes. Overall, the results suggest that initially disadvantaged households
especially with respect to family size, geographic location and employment sector experienced
faster income growth rates than advantaged households.
6.4.4 Estimated Statistical Models

In the previous section, I find evidence that income mobility outcomes differ in terms of
the marginal distribution of income status and other socio-demographic characteristics. This
section measures the statistical significance of each of these factors in explaining mobility in
the presence of other factors. Furthermore, it also examines the significance of demographic
and economic events in explaining the variations in income mobility.
Table 6.8 shows the coefficients of the multinomial logistic models based on (6.3) and
(6.4) for the monetary indicators. The full regression results are provided in Appendix Table
A6.1. The coefficients are interpreted as follows. For instance, based on the first entry in Table
6.8, the multinomial logit for households from the second quintile relative to those in the
poorest quintile is 1.1 unit lower for being in the cluster of households that experienced
consistently upward mobility (cluster 2) relative to the cluster that experienced nil income
mobility (cluster 1). The other numbers can be interpreted analogously. Since the coefficients
163
for cluster 2 decrease as the income quintiles increase and the coefficients for cluster 3 increase
as the income quintiles increase, we can conclude that the poorest 20% households had the
highest (logged) odds to experience consistently upward mobility (cluster 2), followed by the
middle income households and lastly by the richest 20% households. On the other hand, the
richest quintile had the highest odds of experiencing consecutive episodes of downward
mobility (cluster 3), followed by the middle-income households and lastly by the poorest
quintile. These results support the finding described in the previous sections that households
from the poorest quintile had experienced generally better income mobility outcomes than
households from the richest quintile. Notably, the differences in the odds to experience either
consistently upward or consistently downward mobility became less pronounced when
longitudinally-averaged income was used as the measure of advantage rather than initial
income. Furthermore, the data based on the longitudinally-averaged income also suggest that
the richest two quintiles had the highest odds of experiencing the most volatile income
movements (clusters 4 and 5).
Appendix Table A6.1 also shows the impact of the control variables. When both income
and socio-demographic variables are included in the model, the only significant patterns are
that higher educational attainment (of the household head) and lower dependency ratio were
positively correlated with the propensity to experience better income trajectories. Furthermore,
after adding the different indicators of demographic and economic events in the models, I find
that an additional non-working age family member is correlated with inferior income mobility
outcome while an increase in the number of employed members improves a household’s
income mobility prospects. Overall, the results of the estimated models suggest that while
initial advantage is a significant determinant of a household’s income trajectory, it only
explains a small fraction of the variations in the income mobility outcomes. Changes in
household composition and employment outcomes provide additional information in
predicting a household’s income trajectory.
In summary, the empirical investigation presented in this chapter leads to mixed findings.
First, if advantage is measured in terms of initial income (in 2003), I find that the households
from the richest quintile had the lowest propensity to experience slow to moderate income
changes and were most likely to experience consistently downward mobility throughout the
observation period. Furthermore, initially advantaged households had the highest propensity to
experience consistently upward mobility. Second, if advantage is measured in terms of
longitudinally-averaged income, I still find that the richest quintile tend to be the least immobile
and were most likely to experience the most erratic income fluctuations. In particular, the
164
richest quintile had the highest propensity to experience very high income growth rates in 2003-
2006, a period when average income was decreasing and very high income losses in 2006-
2009, a period when average income was increasing. In addition, the poorest quintile had the
lowest propensity to experience consistently downward mobility. Nevertheless, although the
results suggest that advantage is a significant determinant of income mobility, I also find that
demographic changes (e.g., changes in household composition) and economic events (e.g.,
employment transitions) are also important determinants of mobility.
6.5 Summary and Discussion

How does income segmentation affect income mobility? Does economic growth allow
initially disadvantaged people to catch up through faster income growth or are they left out
because of the cumulative effect of advantage over time? These are the questions that I tried to
address in this paper. The results provided in the last two sections show that income advantage
is an important correlate of subsequent income trajectories. In particular, initial income has a
negative correlation on income growth rates such that households starting with lower initial
income were more likely to experience higher income growth rates than those who had higher
initial income. However, this result needs to be interpreted with care because it is possible that
those who were either below or above their permanent income in 2003 only regressed towards
their longitudinally-averaged income in the subsequent years. In other words, the consistently
significant negative relationship between initial income and income growth rates may simply
be an artefact of the regression to the mean phenomenon as previous studies suggest that initial
income’s explanatory power can be a mix of genuine income dynamics and measurement errors
(Fields et al. 2003). To examine the robustness of the findings, I also considered using the
longitudinally-averaged income instead of initial income as a measure of advantage. After
doing this, I still find that the lower income households experienced (slightly) better income
mobility outcomes. However, their edge over higher income households was much smaller
when longitudinally-averaged income was used. Although these patterns portray convergence,
it can be argued that part of this convergence can be attributed to temporal income fluctuations.
Furthermore, the results also point to symmetry of mobility based on initial incomes and
longitudinally-averaged incomes. In particular, the data shows that based on initial income, the
richest households had the greatest propensity to experience the highest income losses during
economic contraction in 2003-2006 and highest income gains during economic expansion in
165
Table 6.8 Regression Coefficients of Multinomial Logistic Models
(Reference category: Cluster 1 (slow to moderate growth))
Initial income in 2003 Permanent income
Income Segment
cluster 2 cluster 3 cluster 4 cluster 5 cluster 2 cluster 3 cluster 4 cluster 5
Cluster
(base: 1st quintile)
2nd quintile -1.077*** .8009*** -.4979*** .3609** -0.0101 .4144*** .536*** 0.2296
3rd quintile -1.434*** 1.573*** -.6894*** .6826*** 0.1281 .4823*** .633*** 0.264
4th quintile -1.79*** 1.849*** -.8973*** .9757*** 0.2813 .4459** .9259*** .4173**
5th quintile -2.233*** 2.604*** -.7946*** 1.677*** .4677** .6626*** 1.536*** .899***
Cluster
(base: Poor cluster)
Middle income cluster 0.0901 0.1671 0.3166** 0.1171
-0.7121* 0.8753*** -0.259* 0.6719***
Rich cluster .3053* 0.1537 0.7754*** 0.4317**
166
2006-2009. In contrast, based on longitudinally-averaged income, the data suggest that the
richest quintile had the highest probability of observing very high income growth rates even if
the rest of the population experienced decreasing incomes in 2003-2006 and incurred income
losses when the rest of the population observed positive income growth in 2006-2009.
In terms of policy implications, the result that income trajectories of the poor, middle
income and rich households are statistically different, reiterates that the impact of economic
development is not uniform. Thus, policies should be tailor-fitted according to the diverse
circumstances confronting different population groups. Similar to the findings presented in the
previous chapters, I find evidence to suggest that low income households are more likely to
experience better income mobility outcomes than the rest of the population even after
controlling for temporary income fluctuations. Nevertheless, the finding that significant
fraction of low income households experienced income losses during the observation period
suggests that existing poverty reduction programs could be improved further. This re-echoes
what I have argued in Chapter 5 that to speed up the poverty reduction, it is important that
intervention programs should be responsive to the long-term human capital development needs
of the chronically poor and social protection needs of the transiently poor.
Like the poor, the significant gains experienced by some middle income households were
offset by the losses incurred by others. This contributed to the slow income growth of middle
income households. If such trend continues, this may push the country to a middle income trap
like many countries in Latin America (Jankowska, Nagengast & Perea 2012). If the middle
income households remain stagnant, it will be difficult for the Philippines to really take-off
because a strong middle class is usually the engine for growth. To minimize this danger, one
of the steps that the government can take is to create more jobs of better quality not only for
college graduates but also for non-college graduates which comprise a significant fraction of
the middle class. Although this is easier said than done as job creation often requires
strengthening the competitiveness of local firms in the global production chain which usually
comes at the expense of quality of employment, this feat is not impossible as seen in the
experiences of other Asian countries like Singapore and South Korea which have improved the
quality of jobs held by workers without college degrees (Li 2002). Economists reckon that the
first step is to build a strong manufacturing sector because without a strong manufacturing
sector, non-college graduates are often left to take low paying jobs in the services sector which
can hardly sustain upward mobility in the long-run (Usui 2011, 2012). However, this may also
require upgrading the skills of workers in the traditional manufacturing sector so that they will
167
remain competitive in the labour market as the country transitions from traditional to modern
manufacturing. For higher income households, this study finds that they experienced the most
erratic income fluctuations. Although higher income households may be better-off in handling
income fluctuations because they suffer less from liquidity constraints than lower income
households, it is important that policies should still aim to minimize these volatilities. One way
to address this issue is to facilitate a socio-economic climate conducive for business as majority
of high income households rely on income from entrepreneurial activities. A WB study shows
that the business regulations in the country are among the most complicated and costliest in
Southeast Asia (WB 2013). This should prompt policymakers to review how the existing rules
and regulations in doing business in the country can be simplified.
Interestingly, my findings that demographic and economic events are significant
correlates of income trajectories resonate some of the advances in the literature of socio-
economic stratification. Traditionally, sociologists and economists have been interested in
understanding the patterns of social segmentation due to income, social class, gender or
educational level and how these factors shape a person’s socio-economic prospects. Lately, the
research focus has shifted to the importance of life course events as predictors of income
trajectories (Vandecasteele 2010). This calls for the need to collect more relevant data on life
course events to be able to assess their structuring effect on a person’s socio-economic well-
being.
168
Appendix Table A6.1 Regression Coefficients of Multinomial Logistic Models for Income Growth Trajectories in the Philippines
Initial Income in 2003 Permanent Income
Variable
Cluster 2 Cluster 3 Cluster 4 Cluster 5 Cluster 2 Cluster 3 Cluster 4 Cluster 5
Male-headed hhld -0.2998 -0.2744 0.0235 -0.1082 -0.2951 -0.2558 0.0189 -0.0911
Hhld head's age 0.0247 0.0092 0.0190 -0.0032 0.0110 0.0205 0.0050 0.0010
Hhld head's age squared -0.0003 0.0000 -0.0001 0.0000 -0.0001 -0.0001 0.0000 0.0000
Marital status of hhld head
(base = Married)
Single -0.0722 -0.2082 -0.1912 -0.1505 -0.2944 -0.0685 -0.3789 -0.0867
Separated/widowed/others 0.1257 -0.1301 0.1157 0.2155 -0.0780 -0.0317 -0.0624 0.2591
Hhld head's educational attainment
Primary education .4796*** -.321** .3327** -.2492** 0.0618 -0.0326 -0.0229 -0.1224
Tertiary education .9064*** -.9734*** 0.3552 -.5875** 0.0462 -.4324* -0.3180 -0.3676
Region (base = NCR)
Luzon -0.0324 0.0056 -0.1405 -0.2897 0.1175 -0.0822 -0.0413 -0.3216
Visayas 0.0737 0.1643 -0.2562 -0.0160 .4972** -0.0741 0.0587 -0.1204
Mindanao -0.1536 0.0964 -0.3340 -0.1249 0.2978 -0.1606 0.0193 -0.2248
Urban -0.0728 -.3324*** -0.0768 -.4153*** -.326*** -0.1309 -.2855** -.3209***
Hhld type
(base = Single Family)
Extended family 0.1876 0.1757 0.0754 -0.0360 0.0606 .266* -0.0348 0.0052
Two or more non-related
13.0900 12.2300 12.6800 12.8000 13.8700 13.6700 13.5300 14.0900
individuals
-1.301*** .8153** -1.555*** 0.4440 -0.3063 0.1972 -.7377* 0.2106
are young
Family size -.05809* 0.0442 -.08264** 0.0524 0.0474 -0.0329 0.0098 0.0202
169
Appendix Table A6.1 Regression Coefficients of Multinomial Logistic Models for Income Growth Trajectories in the Philippines
Variable
Main source of income is
-.2153* 0.0700 -.3374** 0.1834 0.1230 -0.1632 -0.0355 0.0800
Agriculture
Hhld head is employed -0.0443 -0.0753 -0.5284 0.0957 -0.1522 -0.0175 -0.5917 0.1653
At least one hhld member is
.356*** -0.2200 0.1607 -0.0786 0.0283 -0.0046 -0.1129 0.0149
working abroad
Hhld income quintile
(base = 1st quintile)
2nd quintile -1.077*** .8009*** -.4979*** .3609** -0.0101 .4144*** .536*** 0.2296
3rd quintile -1.434*** 1.573*** -.6894*** .6826*** 0.1281 .4823*** .633*** 0.2640
4th quintile -1.79*** 1.849*** -.8973*** .9757*** 0.2813 .4459** .9259*** .4173**
5th quintile -2.233*** 2.604*** -.7946*** 1.677*** .4677** .6626*** 1.536*** .899***
0.1016 0.3492 .415* .4722** 0.1962 0.2934 .4838** .4457*
are employed
0.0151 -0.0474 0.0371 -0.0656 -0.0429 -0.0074 -0.0180 -0.0452
members with permanent job
0.0417 0.1503 .363** -0.0411 -0.1728 .2905* 0.1765 0.0157
members with formal job
-.3544** -0.1265 -0.2858 -0.1841 -.2788* -0.1751 -0.2310 -0.2169
members with multiple jobs
170
Appendix Table A6.1 (con’t) Regression Coefficients of Multinomial Logistic Models for Income Growth Trajectories in the Philippines
Variable
Hhld head's sex changed (2003-2006) 0.2852 0.0852 .6397** 0.3313 0.2402 0.0693 .588** 0.3163
Hhld head's sex changed (2006-2009) -.3587* -0.0628 -.5687** -.4354* -0.3414 -0.0884 -.5462** -.4567**
Hhld head's marital status changed (2003-
-0.1066 0.1113 -0.2411 -0.2356 -0.0630 0.1322 -0.2136 -0.2133
2006)
Hhld head's marital status changed (2006-
0.1633 -0.1201 0.2943 0.2741 0.1831 -0.1087 0.3007 0.2919
2009)
Hhld head's educational attainment improved
0.2655 -.4082** 0.1668 -0.0437 0.0396 -0.2846 -0.0335 -0.0076
(2003-2009)
Hhld head's educational attainment
-0.0638 0.3076 0.0452 0.2497 0.1301 0.1560 0.1987 0.1793
deteriorated (2003-2009)
Change in family size (2003-2006) -.2441*** .2704*** -.2545*** .1894*** -.2154*** .2552*** -.2144*** .1886***
Change in family size (2006-2009) -.2515*** .2358*** .2061*** -.2308*** -.2393*** .2334*** .2238*** -.2245***
Change in proportion of young members
-0.1521 0.1559 -1.073** 0.5103 0.0853 0.1086 -.8799** 0.4970
(2003-2006)
Change in proportion of young members
-1.004*** .8568** 0.2630 -0.0610 -.7299** .6831* 0.5123 -0.1159
(2006-2009)
Change in proportion of employed members
0.3623 -.7166** -0.2402 -0.1501 0.4545 -.7677** -0.1960 -0.1707
(2003-2006)
Change in proportion of employed members
0.1369 -.9959*** -.7275*** 0.2938 0.1748 -.9815*** -.7338*** 0.3062
(2006-2009)
Main source of income changed (2003-2006) 0.0245 0.2347 0.2020 0.1158 0.1130 0.1874 .2823* 0.1058
Main source of income changed (2006-2009) 0.1641 0.2315 0.0937 .4979*** 0.2184 0.1659 0.1622 .4776***
Constant 2.375*** -1.0700 1.2440 -0.0516 0.7808 0.0758 -0.1508 0.3540
Source: Author’s computations using data from the longitudinal subsample of FIES 2003, 2006 and 2009.
Note: The dependent variable is based on household expenditure per capita. The results based on the latent cluster analysis are qualitatively similar. To save space, I don’t present the results
here.
171
Chapter 7 What Drives Income Distribution Dynamics in the Philippines?
7.1 Introduction
The previous two chapters examined the dynamics of poverty and inequality by taking
into account the mobility of incomes. A good understanding of how much various factors affect
poverty and inequality is important for strategic planning and policy making as it allows socio-
economic planners devise policy interventions that could help economic growth achieve
maximum impact on reducing socio-economic deprivation. For instance, if one finds that
changes in employment income drive upward mobility, labour market policies that promote
growth in sectors where most of the poor are should become the focus. On the other hand, if
economic shocks drive downward mobility, policymakers should strengthen social safety nets.
In the Philippines, several studies have attempted to identify why poverty and inequality remain
high despite faster economic growth (e.g., Balisacan & Hill 2003; Schelzig 2005; ADB 2007;
Aldaba 2009) by identifying factors that correlate with these two phenomena. In the previous
chapters, I have also implicitly focused on correlations between mobility and various socio-
economic variables. However, solely relying on correlations make it hard to gauge the extent
to which perturbations in different factors would affect the distribution of household income.
For example, although many of the existing studies in the recent years suggest that sub-optimal
employment outcomes highly correlate with higher poverty (ILO 2009; ADB 2011b), they are
silent about how much of the observed changes in poverty levels can actually be attributed to
the changes in employment outcomes. The main objective of this chapter is to contribute to
the existing literature in identifying proximate determinants of poverty and inequality dynamics
in the Philippines. Using a general Shapley (1953)-based accounting method proposed by
Shorrocks (2013), the analysis presented in this chapter departs from the conventional
correlation-based approaches by carrying out a series of counterfactual simulations to
decompose the changes in poverty and inequality into the contribution of changes in various
income correlates. While there are also limitations in the decomposition approach proposed by
Shorrocks (2013), the result of such an accounting tool is easier to interpret and facilitates a
more straightforward ranking of the relative importance of each factor in driving poverty and
inequality because the estimated contributions sum up to the observed changes in poverty and
inequality compared to the conventional correlation-based approaches. This exercise may be
considered as a head start to better understand how to prioritize policy intervention programs
to induce better household income distribution outcomes in the country.
172
In identifying the factors that have contributed to the observed household income
distribution dynamics, I examine the extent to which changes in poverty and inequality depend
on the changes in people’s socio-economic capital or to changes in the economic returns to
these capital. Simply put, a socio-economic capital (SEC) can be viewed as an economic tool
that a person can use to extract the available wealth in the society to be able to improve his/her
well-being. The type of education, employment and assets held are examples of SECs.65 In
general, each SEC is valued differently. For example, having a college education does not
necessarily have the same impact on a person’s well-being as having a small parcel of land. I
refer to this value as socio-economic returns (SER). In addition to employment, many studies
have highlighted the importance of having higher skill set through better educational
qualification in promoting upward mobility (Greenstone et al. 2013, Morgan et al. 2006). Some
studies, particularly in the Philippines, have also stressed the limited access to basic social
services and productive assets as underlying cause of poverty and inequality (Balisacan 2007).
However, how changes in the returns to various forms of capital contribute to the evolution of
household income distribution in the Philippines remains an empirical issue. For instance, as
the supply of a specific form of socio-economic capital increases, it is tempting to expect for
its corresponding economic returns to decline assuming that the demand for such capital
remains fixed. This potential trade-off between capital and economic returns and the fact that
either demand or supply of socio-economic capital hardly remains constant relative to the other
make it less straightforward to infer how poverty and inequality would change over time. It
may lead to either poverty reduction if low income households are acquiring additional capital
faster than economic returns are dropping or increasing poverty if economic returns are
deteriorating faster than the rate at which low income households are acquiring additional
capital. Alternatively, poverty reduction will be much faster or lower than expected if both
socio-economic capital and returns to capital are simultaneously increasing or decreasing,
respectively. Counterfactual analysis allows me to investigate which of these scenarios hold in
the Philippines. In particular, I address the following questions:
(i) Are changes in households’ socio-economic capital and/or changes in returns to
capital important in explaining the evolution of poverty and inequality in the
Philippines?
65 In other sociological literature, education is considered as an endowment while employment is considered as a type of
functioning (i.e., capacity to translate an endowment to resources that can be used directly to improve one’s well-being). In
this study, I considered both education and employment as different types of socio-economic capital to account for the fact
that people have different capacities to make endowments function towards improving one’s living standards.
173
(ii) What are the socio-economic factors that have contributed significantly to changes
in poverty and inequality in the Philippines over the past decade?
Like in other chapters, I use the longitudinal subsample data from the FIES-LFS to
answer these questions. Throughout the study, estimates are presented at the national and
(broad) regional levels.
7.2 Concepts and Methods

7.2.1 Drivers of Income Distribution Dynamics
Following the convention used in the previous chapters, I use the (log) household
expenditure per capita as the main measure of well-being and I refer to this as income. To be
able to measure the contribution of changes in SECs and changes in SERs to the observed
trends in poverty and inequality, equation (7.1) decomposes (log) per capita income as a
stochastic function of several correlates of a household’s well-being that are typically used in
the existing literature (Canlas et al. 2009).
𝑐 ℎℎ𝑙𝑑𝑐 ℎℎ𝑙𝑑𝑐
𝑌𝑖𝑡 = 𝛽𝑡𝑙 𝑐𝑎𝑡𝑖 𝑛
𝑋𝑖𝑡𝑙 𝑐𝑎𝑡𝑖 𝑛 + 𝛽𝑡 𝑋𝑖𝑡 +
𝑙 𝑦 𝑙 𝑦
𝛽𝑡 𝑋𝑖𝑡 + 𝛽𝑡𝑠𝑣𝑐𝑠 𝑋𝑖𝑡𝑠𝑣𝑐𝑠 + 𝛽𝑡𝑎𝑠𝑠 𝑡𝑠 𝑋𝑖𝑡𝑎𝑠𝑠 𝑡𝑠
+ 𝜀𝑖𝑡 (7.1)
Similar to how I categorized the proximate determinants of poverty dynamics in Chapter

5, the SECs are broadly grouped into (i) (geographic) location, (ii) education, (iii) employment,
(iv) access to (basic) services and (v) physical assets.66 Although all of these SECs are
important, identifying which of them have the most significant impact on household income
distribution outcomes will enable policymakers prioritize intervention programs. In a
developing country like the Philippines, setting policy priorities and channelling the limited
resources available to areas where interventions could have optimal impact is critical.
How does the relationship between SEC and SER affect household income distribution
outcomes? It is worth pointing out that simply increasing households’ capital levels would not
necessarily guarantee better living standards (King, Montenegro, & Orazem 2012; Schultz
1975). For instance, if the labour force had higher stockpile of skills, it is not absolutely
consequential that this would result in upward economic mobility across the board unless the
demand for better-skilled workers also increases. A higher supply of skilled workers with a
fixed demand for such type of labour would likely result in lower SERs. The same can be said
about the other types of SECs. In this simple example, (absolute) poverty would increase if
66As mentioned earlier, there are other forms of socio-economic capital (e.g., health, social networks, etc.) that can influence
the household income distribution based on the existing literature but they are not included here due to data limitations.
174
SER falls faster than the rate at which SEC is increasing for low income households and it
would decrease if SEC increases faster than the rate at which SER is falling. On the other hand,
inequality would increase when SEC is increasing disproportionately faster in high income
households or SER is decreasing disproportionately faster in low income households. The
following section outlines the methodology for estimating the contribution of each of these
factors on income distribution dynamics, separately.
7.2.2 Estimating the Contribution of SECs and SERs to the Evolution of the Income
Distribution
Since the pioneering work of Oaxaca (1973) and Blinder (1973) who proposed methods
for decomposing group differences in income into various components, substantial progress
has been made in terms of understanding what contributes to income distributional variations
across space and over time. The main idea behind the Oaxaca-Blinder method is to decompose
income differentials (between groups) into factors that are attributable to differences in SECs
and variations in the SERs. To illustrate the approach, assume the income of individual i from
the gth group, denoted by 𝑌𝑖(𝑔) , is a function of his/her SEC 𝑋𝑖(𝑔) , SER 𝛽 (𝑔) , and an unobserved
error term 𝜀𝑖(𝑔) as shown in Equation 7.2.67 For simplicity, suppose we have two groups, g = 0,
1. The main objective of the Oaxaca-Blinder decomposition method is to explain the difference
in group averages denoted by 𝑌̅ ( )
𝑌̅ (0). This is done by constructing income for one group,
denoted by 𝑌̅ (𝑐) , by assuming that it has the same income structure (i.e., same SER) as the other
group as shown in Equation 7.3. Equation 7.4 shows that the difference 𝑌̅ ( )
𝑌̅ (0) can be
arithmetically expressed as a sum of two components where the first term corresponds to the
gap in the average SEC in each group while the second term corresponds to the variation in the
SER.
(𝑔) (𝑔) (𝑔) (0) ( )

𝑌𝑖 = 𝛽 (𝑔) 𝑋𝑖 + 𝜀𝑖 𝑌̅ (0) = 𝛽̂ (0) 𝑋𝑖 and 𝑌̅ ( )
= 𝛽̂ ( ) 𝑋𝑖 (7.2)
(0)
𝑌̅ (𝑐) = 𝛽̂ ( ) 𝑋𝑖 (7.3)
𝑌̅( ) 𝑌̅ (0) = (𝑌̅ ( ) 𝑌̅ (𝑐) ) + (𝑌̅ (𝑐) 𝑌̅ (0) )
( ) (0) (0) (0)
= (𝛽̂ ( ) 𝑋𝑖 𝛽̂ ( ) 𝑋𝑖 ) + (𝛽̂ ( ) 𝑋𝑖 𝛽̂ (0) 𝑋𝑖 )
= 𝛽̂ ( ) (𝑋̅ ( ) 𝑋̅ (0) ) + (𝛽̂ ( ) 𝛽̂ (0) )𝑋̅ (0) (7.4)
67 Here, income is expressed in the natural logarithmic form.
175
Since its inception, the Oaxaca-Blinder decomposition technique has been used extensively to
estimate the separate contributions of group differences in outcomes of interest with respect to
observable characteristics like sex, education, race, and location. Nevertheless, although the
method was originally proposed to explain income discrimination between two groups for a
fixed time period, the procedure can also be applied to explain temporal changes in average
income of the same group. In general, the Oaxaca-Blinder decomposition method is very
straightforward to apply as it only entails estimation of the coefficients of a linear regression
model and the sample means of the underlying independent variables. However, the approach
has two main shortcomings. First, it is limited to explaining differences in average income
while differences in other parts of the income distribution are left unexplained. Second, the
decomposition depends on the choice of a reference group. For example, when estimating
separate wage regressions for five geographic locations, the results where the first geographic
location is left-out would not necessarily be the same when the last geographic location were
left-out. This portrays an identification problem wherein the results depend on an arbitrarily
chosen reference group (Jones & Kelly 1984; Oaxaca & Ramson 1999). Over the years, several
alternative decomposition methodologies have been proposed to address these limitations. To
save space, I do not discuss them here but see Bourguignon, Ferreira & Lustig (2004) and
Bourguignon & Ferreira (2008). More recently, Shorrocks (2013) provide a unified framework
for different decomposition methods aiming to assess the contribution of a set of factors which
together account for the observed value of some aggregate statistic.
This study adopts the procedure proposed by Shorrocks (2013) also known as the
Shapley-Shorrocks (SS) approach using the Stata implementation developed by Azevedo,
Nguyen & Sanfelice (ANS) (2012).68 To illustrate the procedure, suppose we treat households
as the unit of analysis and assume that there are two time periods. For notation purposes, I
express (log) income Yit as a function of C components where each component is denoted by
𝐹𝑖𝑡𝑐 , c = 1, 2, …, C; t= 0, 1 (7.5) and the term 𝑀(𝑌𝑡 ) is used to denote a specific characteristic
feature of the household income distribution. The main interest is to decompose the change in
the characteristic feature of the income distribution between time 0 and time 1, 𝑀(𝑌 ) 𝑀(𝑌0 ),
into the contribution of changes 𝐹𝑖𝑐 - 𝐹𝑖0𝑐 . As argued by ANS (2012), this can be done by
simulating the income distribution by changing each 𝐹𝑖𝑡𝑐 one at a time. The step-by-step
procedure is outlined below.
68 The Stata routine is called ADECOMP.
176
𝑌𝑖𝑡 = 𝑓(𝐹𝑖𝑡 , 𝐹𝑖𝑡2 , … . , 𝐹𝑖𝑡 − , 𝐹𝑖𝑡 ) (7.5)
𝑀(𝑌𝑖𝑡 ) = ∅(𝑓(𝐹𝑖𝑡 , 𝐹𝑖𝑡2 , … . , 𝐹𝑖𝑡 − , 𝐹𝑖𝑡 )) (7.6)
Shapley-Shorrocks’ Algorithm for

Estimating the Contribution of 𝐹 𝑐 on 𝑀 (𝑌 ) 𝑀0 (𝑌0 )
Step #1: Using the formula provided below, compute the counterfactual income distributions
at the initial time period and the corresponding parameter of interest M(Y0)(c) for each factor
Fc.
2
𝑀(𝑌0 )(0) = ∅ (𝑓(𝐹𝑖0 , 𝐹𝑖0 , … . , 𝐹𝑖0− , 𝐹𝑖0 )) = 𝑀(𝑌0 )
𝑀(𝑌0 )( ) = ∅(𝑓(𝐹𝑖 , 𝐹𝑖0 2
, … , 𝐹𝑖0− , 𝐹𝑖0 ))
𝑀(𝑌0 )(2) = ∅(𝑓(𝐹𝑖 , 𝐹𝑖2 , … , 𝐹𝑖0− , 𝐹𝑖0 ))
:
𝑀(𝑌0 ) ( − )
= ∅(𝑓(𝐹𝑖 , 𝐹𝑖2 , … , 𝐹𝑖 − , 𝐹𝑖0 ))
𝑀(𝑌0 )( ) = ∅(𝑓(𝐹𝑖 , 𝐹𝑖2 , … , 𝐹𝑖 − , 𝐹𝑖 ) = 𝑀(𝑌 )
Step #2: Compute the contribution of Fc by subtracting M1(Y)(c-1) from M1(Y)(c).
𝑜 𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜 (𝐹𝑖𝑐 𝑐
𝐹𝑖0 ) = 𝑀(𝑌0 )(𝑐) 𝑀(𝑌0 )(𝑐− )
(7.7)
𝑀(𝑌0 )(𝑐)
− 𝑀(𝑌0 )(𝑐−1)
% 𝑜 𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜 (𝐹𝑖𝑐 𝑐
𝐹𝑖0 )= (7.8)
𝑀(𝑌1 )− 𝑀(𝑌0 )
Step #3: Repeat Steps #1 and #2 for all possible orderings of Fc’s and then take the average of
(7.7) and (7.8).
At this point, important remarks are in order. First, like the Oaxaca-Blinder
decomposition method, the procedure outlined in the first two steps is path-dependent. Suppose
the income measure 𝑌𝑖𝑡 is expressed as a function of 𝐹𝑖𝑡𝑐 ’s and the characteristic feature of the
income distribution is some function M() of Yit, the idea behind the SS algorithm is to construct
a counterfactual distribution of income by changing the values of the 𝐹𝑖𝑡𝑐 from the observed
value at the initial time period to the observed value at the succeeding time period, one at a
time. In the example above, I started chronologically from 𝐹𝑖𝑡 to 𝐹𝑖𝑡 . Thus, the values of (7.7)
and (7.8) depend on this specific ordering of the factors. However, had I started from 𝐹𝑖𝑡 to 𝐹𝑖𝑡
or followed any other ordering, the results would have been different. To address this issue, the
third step entails computing the contribution of each factor across all possible permutations or
“paths” and using the average to estimate the factor’s contribution on 𝑀 (𝑌 ) 𝑀0 (𝑌0 ).
Second, the approach entails estimating the contribution of one factor at a time by holding
the values of all other factors constant. Hence, the decomposition methodology does not reflect
economic equilibrium because it employs a simplistic assumption that each factor can be
177
changed one at a time while the rest can be held fixed (Azevedo et al. 2013). Nevertheless, the
potential interactions between factors are partially taken into account by estimating the
contribution of a specific factor as the difference between the cumulative counterfactuals.
Third, unlike the Oaxaca-Blinder method and other conventional decomposition tools
which are mostly based on the means, the SS algorithm flexibly accommodates quantiles,
variance and any other characteristic features of an income distribution. Although the
methodology can be used to explain the temporal differences in various forms of 𝑀(𝑌𝑡 ), this
study defines 𝑀𝑡 (𝑌𝑡 ) in terms of poverty and inequality only, in particular, I focus on US$2
poverty gap and Gini coefficient (i.e., ∅(𝑓) = ⋯ ).69
Fourth, to be able to construct counterfactual income distributions, the SS algorithm
requires panel data. If repeated cross-sectional data is available, the algorithm can be modified
by making additional assumptions as outlined in Azevedo et al. (2013).
To estimate the contribution of the changes in SEC and SER to poverty and inequality
dynamics using the SS algorithm, each of the 𝑋𝑖𝑡𝑐 (SEC) and the parameter 𝛽𝑡𝑐 (SER) as well as
the error term 𝜀𝑖𝑡 can be considered as one of the 𝐹𝑖𝑡𝑐 ’s. Note that each SEC could have multiple
indicators, for example, access to services can be measured in terms of access to either
electricity, clean water or sanitary toilet, estimation of (7.7) and (7.8) could be very
computationally-intensive due to the iterative nature of the SS algorithm if each indicator is
treated as a separate 𝐹𝑖𝑡𝑐 . To address this issue, I reduce the dimension of (7.1) by constructing
an index for each SEC by following the approach outlined in UN (2005) and estimating a
regression model (of income) and using the corresponding coefficients as weights for the index.
In particular, I regress (log) income on the various indicators of SECs. Since I am interested to
measure the impact of changes in SEC levels to poverty and inequality dynamics, I do not want
the changes in the SEC indices to be artificially contaminated by the changes in the weights of
the component indicators. Thus, I use the data from the initial survey year only to derive the
weights for each component indicator. These weights are then multiplied to the value of each
component indicator for the initial survey year and the succeeding time periods. The resulting
indices are then used as inputs for the SS algorithm. Although the indicators included in the
construction of the SEC indices in this study are similar to the ones commonly used in the
existing literature (Montgomery et al. 2000; Aldaba 2009), these were chosen on an ad-hoc
basis, subject to data availability and the results of descriptive analysis. In general,
Montgomery et al. (2000) argued that in the empirical literature, indicators are usually chosen
69 Results for other poverty and inequality indices are provided in Appendix Tables A7.2 and A7.3.
178
on an ad-hoc basis due to lack of “best practice” approach of selecting indicators that can proxy
living standards comprehensively.
Furthermore, I treat the model residuals as a separate component that gauges the level of
socio-economic shocks. In general, while variations in household incomes across space and
over time can be mostly explained by differences in stock of socio-economic capital and
economic returns, incomes could also fluctuate significantly due to unexpected shocks. As
pointed out in the previous chapters, a quick review of the Philippines’s economic history
would reveal that socio-economic shocks (e.g., environmental disasters, financial crisis, etc.)
have been prominent features of the country’s development landscape (Bayudan-Dacuycuy &
Lim 2013) but not much has been said about the magnitude of impact of these shocks on
poverty and inequality dynamics using a longitudinal perspective in the country. By treating
the model residuals as an approximate measure of shock, I can explicitly gauge how much of
the changes in poverty and inequality observed in the past decade are attributable to shocks in
household incomes, after accounting for the changes in SECs and SERs.70
7.2.3 Constructing Indices of Socio-Economic Capital

In constructing the SEC indices (i.e., Location, Education, Employment, Services and
Assets), I derive the weights by estimating several regression models with the (log) income as
the dependent variable and the various indicators of SEC that are available from the survey as
independent variables. On the basis of preliminary analyses, I drop indicators that are not
statistically significant and have counterintuitive signs of model coefficients to be able to come-
up with sound and parsimonious SEC indices. The final SEC index Location consists of four
dummy variables: (i) whether the household is living in urban area, (ii) whether the household
is living in the National Capital Region (NCR), (iii) whether the household is living in Luzon
and (iv) whether the household is living in Visayas. The index Education has three sub-
component indicators: (i) proportion of working-age household members with primary
education, (ii) proportion of working-age household members with secondary education, and
(iii) proportion of working-age household members with post-secondary education. Similarly,
the index Employment consists of three indicators: (i) proportion of working-age household
members who are employed, (ii) proportion of employed household members working in the
non-agriculture sector, (iii) proportion of employed household members with formal
70 Here, since the model residuals are used to approximate shocks, these also contain household-specific effects and other
factors that were not controlled in the model.
179
employment.71 The index Services has four dummy variables: (i) whether household has
electricity at home, (ii) whether household has water faucet at home, (iii) whether household
has a sealed-toilet facility and (iv) whether household has closed-pit toilet facility. Lastly, the
index Assets consists of four dummy variables: (i) whether household owns house/lot, (ii)
whether household owns a refrigerator, (iii) whether household owns a phone and (iv) whether
households owns a car.72 Overall, although the resulting indices are not comprehensive, they
provide a good starting point for a more nuanced understanding of how changes in SEC levels
interplay with the changes in SER in driving household income distribution dynamics.
7.3 Results
7.3.1 Drivers of Household Income Distribution Dynamics in the Philippines
The objective of this section is to examine whether the observed changes in poverty and
inequality can be attributed to changes in households’ SECs or changes in the SERs. As pointed
out earlier, the household income distribution dynamics is potentially shaped by how much the
pace at which SECs and SERs are changing differ from each other.
Figure 7.1 Temporal Changes in the Levels of Socio-Economic Capital

Location Education Employment
.071
6.48
.072
.07
6.47
.07
.069
6.46
.068
.068
6.45
.066
2003 2006 2009 2003 2006 2009 2003 2006 2009

year year year
Services Assets
.13
.16
.125
.14
.12
.12
.115
.1
.11
2003 2006 2009 2003 2006 2009

year year
FIES 2003, 2006 and 2009.
71 Here, formal employment refers to jobs held by government employees, professionals and wage workers employed in
private businesses. Further details are provided in Chapter 8.
72 As pointed out in Chapter 5, the variables that make up the Assets and Services indices could be considered as proxy
measures of material deprivation. Since they could be endogenous with income poverty, I avoid inferring causal relationships
in most of the succeeding discussions.
180
Figure 7.1 summarizes the evolution of the distribution of each SEC index over time.
The bars correspond to the mean levels of each SEC while the bands correspond to 95%
confidence intervals. In the case of Location, the distribution does not change because I am
using data from households that did not move residential location throughout the observation
period. On the other hand, I observe no changes in Education and Employment. This is
consistent with the findings from previous studies which have attributed the low growth
elasticity of poverty to its lack of enabling capacity to expand economic opportunities for the
poor (Aldaba 2009). In contrast, significant improvements can be observed in Services from
2003 to 2009 and in Assets across all survey years.
Figure 7.2 Temporal Changes in the Socio-Economic Returns to Capital

RLocation REducation REmployment
1
1.6
1.4
.995
.9
1.2
.99
2003 2006 2009 2003 2006 2009 2003 2006 2009

year year year
RServices RAssets
1
1
.98
.9
.96
.8
.94
.7
.92
.6
2003 2006 2009 2003 2006 2009

year year
FIES 2003, 2006 and 2009.
To estimate the SERs, I regress (log) per capita income on the SEC indices for each
survey year. The coefficients of the SECs are used as estimates of the SERs. Figure 7.2 shows
how these SERs have changed over the past decade. Except for REducation and REmployment,
the results provide empirical support to the hypothesis that improved SEC levels usually lead
to lower SERs. In the case of REducation, there is a slight downward trend but the changes are
not as remarkable as that of other SERs. Interestingly, I find that the REmployment have
uniformly increased over the past ten years.
181
The estimated contribution of the changes in SECs and SERs to poverty and inequality
dynamics are presented in Figure 7.3.73 The bars represent how much each factor has
contributed to the increase/decrease in poverty and inequality. Positive values indicate
inflationary impact while negative values indicate deflationary impact on our income
distributional measures. The number on top of each bar indicates the total change in poverty or
inequality observed during the period under consideration.
Between 2003 and 2006, the results of the counterfactual simulations based on the SS
algorithm suggest that the SEC levels in terms of Education, Employment and Services had
minimal inflationary effect on the overall poverty gap. In particular, the observed changes in
Education, Employment and Services would have increased the overall poverty gap by 0.1, 0.4
and 0.5 percentage points, respectively, if all other factors remained constant. On the other
hand, the observed changes in Assets had negative effect on poverty from 2003 to 2006. In
particular, the changes in Assets would have coincided with reduction in poverty gap by 2.4
percentage points if all other factors were held fixed. In terms of the changes in SERs, I find
that the changes in REducation and RServices between 2003 and 2006 had strong correlation
with increase in poverty. In particular, the observed changes in REducation and RServices
would have coincided with an increase in the overall poverty gap of 3.7 and 2.4 percentage
points, respectively, if the values of all other components were held constant during this period.
Similarly, the changes in RLocation and RAssets had increasing, albeit slightly weaker,
correlation on poverty gap. In contrast, the changes in REmployment had a strong correlation
with poverty reduction, contributing to a 3.6 percentage point reduction in poverty gap between
2003 and 2006, ceteris paribus.
On the other hand, the increase in the Gini coefficient from 42.8 in 2003 to 44.3 in 2006
can be mostly attributed to changes in SEmployment. Changes in Education and Employment
also contributed positively to higher inequality during this period. However, this was largely
offset by the inequality-reducing impact of changes in Services, Assets, RLocation,
REducation, RServices and RAssets.
From 2006 to 2009, the US$2 poverty gap dropped from 16.3 to 13.6. The poverty-
inflationary impact of the changes in RLocation, REmployment, RServices and RAssets have
been largely offset by the changes in SECs, particularly Assets and Employment which together
have contributed to a 3.0 percentage point reduction in US$2 poverty gap while the reduction
in inequality during this period could be mostly attributed to the changes in Assets.
73 The estimates are provided in Appendix Tables A7.2 and A7.3.
182
Figure 7.3 Estimates Contribution of Different Factors on Poverty and Inequality
-1.7
+0.88
5
+1.44 -0.7
-2.6
-2.1
0
-5
Education Employment Services Assets Shocks

-10
Returns Education Returns Employment Returns Services Returns Assets Returns Location
0 povgap
povgap 03-06
03-06 povgap
20
povgap 06-09
06-09 povgap
40 03-09
povgap 03-09 gini
gini03-06
60
03-06 gini
gini 06-09
80
06-09 gini 03-09 100
FIES 2003, 2006 and 2009.
In summary, the results suggest that the changes in poverty gap between 2003 and 2009
coincided mostly with changes in returns to education, returns to access to basic services,
returns to employment and levels of asset ownership. The last two factors are positively
correlated with poverty reduction while the first two factors are positively correlated with
increase in poverty. Interestingly, this has occurred at the backdrop of trivial changes in human
capital (i.e., education and employment). While the results of these study have not established
causal relationship with poverty and inequality, the findings seem to depart from the
conventional wisdom that the underlying cause of the country’s poverty and inequality in the
1980s and 1990s is the limited access to basic social services and productive assets (Balisacan
2007). Instead, the results may be indicative of the need to improve human capital outcomes.
For instance, given the way how Employment index has been constructed, the finding that it
did not contribute significantly to poverty reduction suggests that the poor did not experience
improvements in their chance to be employed in the non-agriculture and formal sectors. This
portrays a labour market segmentation wherein the poor workers continuously experience
difficulty in moving to formal, non-agriculture sectors. Since more productive sectors require
higher levels of skills, the stagnant education levels, which can be used to proxy skills, could
probably explain why a significant fraction of poor workers were unable to move away from
less productive sectors. Nevertheless, the finding that those who successfully transitioned to
formal and non-agriculture jobs have experienced improved living standards due to higher
183
economic returns of working in these sectors highlight the importance of improving
employment outcomes for tackling poverty in the Philippines.
The results also confirm that the contribution of changes in the SECs and SERs to poverty
and inequality generally offset one another. This usually happens when the demand for a
specific type of SEC is fixed. To explicitly show this, I summed up the contribution of SEC
and SER for the five correlates of well-being considered in this study and present the results in
Table 7.1. Here, I find that assets and employment outcomes have contributed to lower poverty
gap, leading to a 4.2 and 3.6 percentage point reduction, respectively. However, this gain has
been partially offset by education and services outcomes. In terms of inequality, both SECs
and SERs have generally contributed to a reduction in inequality. Assets and services outcomes
have the highest poverty-reducing impact while employment outcomes have contributed to
increasing inequality.
Table 7.1 Trade-off between Socio-Economic Capital

and Socio-Economic Returns, 2003-2009
Poverty Gap (%) Gini (%)
Factor Total Total
SEC SER SEC SER
Contribution Contribution
Location 0.00 0.74 0.74 0.00 -0.57 -0.57
Education 0.00 3.34 3.34 0.30 -0.07 0.23
Employment 0.00 -3.57 -3.58 0.22 1.92 2.14
Services -1.02 2.69 1.68 -0.59 -1.21 -1.80
Assets -4.65 0.50 -4.15 -1.48 -1.31 -2.80
Total
-5.67 3.70 -1.97 -1.55 -1.24 -2.80
Contribution
FIES 2003, 2006 and 2009.
In addition to SECs and SERs, I also disentangle the contribution of socio-economic

shocks on the observed changes in poverty and inequality. This computational exercise is
important because previous studies suggest that household income is subject to different forms
of socio-economic risks. For instance, Dercon & Krishnan (2002) noted that income from
employment may be heavily affected by ill-health or financial crisis-induced unemployment.
Income transfers may be reduced due to uncertain access to public goods. Income reduction
and value of assets, especially in the agriculture sector, may deteriorate due to war, theft,
uncertainty in land tenure or environmental shocks like earthquakes or typhoons. While the
impact of these shocks is usually transient, it can also have long-term effects on a household’s
future economic prospects (Albert, Elloso & Ramos 2009). Worryingly, socio-economic
184
shocks may push poor and economically vulnerable households to further risk-induced poverty
traps. In the Philippines, there are several sources of socio-economic shocks. Environmental
hazards are good example. On average, about 20 tropical cyclones hit the country every year
(PAG-ASA 2014) and these cost about 0.8% of GDP in damages (Oxford Economics 2013).
Other sources of shocks that are commonly experienced by Filipino households are brought by
illness, accident, unemployment, and economic crises (Albert et al. 2009).
7.3.2 Robustness Checks

Regional Estimates
In this section, I briefly examine the regional variations in terms of the contribution of
SECs, SERs and socio-economic shocks to poverty and inequality dynamics over the past
decade. As pointed out in the earlier chapters, the Philippines consists of three major island
groups, Luzon, Visayas and Mindanao (Figure 3.1 in page 84). Although NCR is within Luzon,
I separate the two in our analysis because NCR differs significantly from the rest of Luzon in
terms of average income levels.
The results of the counterfactual simulations by region are presented in Appendix A7.3.
Although the list of major contributing factors to the observed poverty and inequality dynamics
is similar across regions, there are some spatial differences that are worth pointing out. First,
the changes observed in SECs and SERs have significantly bigger impact on poverty in poorer
regions of Visayas and Mindanao while socio-economic shocks played a more pronounced role
in driving the changes in poverty and inequality in NCR and Luzon. Second, for low income
households in Visayas and Mindanao, the level of SECs improved much faster than the rate at
which its corresponding SERs declined, thus contributing to reduction in poverty gap. The
same can be said for Luzon although the offsetting effect between its SEC and SER was
stronger, leading to lower reduction in poverty gap. In contrast, poverty gap in NCR slightly
increased between 2003 and 2009 and this can be explained by SERs declining faster than the
rate at which SECs of low income households increased.
Other Measures of Poverty and Inequality

To examine the robustness of the results to the type of poverty and inequality indicators
used, I also estimate the contribution of the changes in SECs, SERs and economic shocks to
household income distribution dynamics using the proportion of population with income below
US$2 a day (headcount poverty rate) and the average squared income shortfall (poverty
severity) as alternative measures for poverty and the Theil coefficient as an alternative measure
185
for inequality. The estimates are presented in the appendix. The results based on poverty gap
and Gini coefficient are mostly similar with the results for poverty severity and Theil
coefficient, respectively. However, there are some remarkable differences when one looks at
US$2 headcount poverty rates. In particular, the impact of economic shocks are more
pronounced when US$2 headcount poverty rates are used instead of poverty gap. For example,
it has been mentioned earlier that the economic shocks had minimal deflationary effect on
poverty gap between 2003 and 2006. However, when poverty headcount is used, I find that
socio-economic shocks had a significant inflationary impact, contributing to a 2.2 percentage
point increase in poverty gap between 2003 and 2006. This is equivalent to a +73% contribution
to the observed increase in poverty headcount during this period, compared to its –3.9%
contribution to the observed increase in poverty gap. A possible reason for this is that many of
those who fell into poverty due to economic shocks between 2003 and 2006 were households
that had incomes that were just a little lower than the poverty line. In such case, headcount
poverty is more sensitive to capture these changes than poverty gap. On the other hand, the
impact of socio-economic shocks on poverty between 2006 and 2009 is consistent, regardless
of the poverty measure used. In terms of the qualitative results about income inequality, I did
not find significant differences when inequality is measured using Theil index instead of the
Gini coefficient.
7.3.3 Potential Limitations of the Accounting Exercise74

The decomposition approach adopted in this study is not a perfect tool for analysing
determinants of income mobility. For instance, it falls short in capturing general equilibrium
effects that can affect income distribution dynamics. A good example is a policy initiative that
raises average wages. If higher wages also increase the prices of basic commodities up to the
point that the purchasing power of people is where it was before the policy was implemented,
it will be hard for such a decomposition exercise to capture this process. Another potential
limitation of this study is that the measurement of socio-economic capital and returns to capital
falls short in capturing the exact economic meaning of these concepts. If the statistical models
suffer from severe omitted variable bias, then it will be difficult to assume that the model
coefficients capture the socio-economic returns to capital. Taking into account all these
limitations, adequate caution should be taken from inferring causal relationships from the
74
I thank the external referees who reviewed this thesis for pointing out the issues of the decomposition approach
adopted in this chapter.
186
results presented in this chapter. At best, it can be considered as a modest advance in probing
beyond correlations.
7.4 Summary and Discussion

To be able to devise policies and intervention programs that could address poverty and
inequality effectively, socio-economic planners need to understand the factors that shaped the
household income distribution in the country. In this chapter, I use counterfactual simulations
as an accounting tool to approximate the contribution of various factors to changes in poverty
and inequality over the past decade and in turn, direct us to priorities for policy. I classify the
hypothesized correlates into three broad factors: socio-economic capital, socio-economic
returns to capital and socio-economic shocks. Analysis of the survey data suggests that while
the correlates of income poverty and inequality are diverse, there is empirical evidence that the
higher levels of ownership of assets and higher economic returns to formal, non-agricultural
employment are correlated to lower income poverty while much work needs to be done so that
income poverty reduction would coincide with education, employment and access to basic
services. The results also re-echo the findings in the previous chapters about the existence of
strong offsetting forces that lead to small changes in poverty and inequality at the aggregate-
level over the past decade. In particular, I find that while the levels of socio-economic capital
increased in some cases, the corresponding economic returns also declined at approximately
the same pace. This departs from conventional wisdom that only portrays income poverty and
inequality as a simple lack of socio-economic capital of those who are at bottom of the social
pyramid because the changes in the various forms of capital held by Filipino households
interact with the changes in its corresponding economic returns in driving the household
The results of this chapter point to the need to ensure that the welfare-improving effect,
i.e., the changes in SERs, do not work to the disadvantage of the poor is probably as important
as providing access to SECs. There are several ways to do this. In terms of human capital, it is
important that socio-economic planners provide enabling opportunities for the poor to get
access to skills needed in higher-productivity sectors for the country’s poverty reduction to
speed up (ADB 2012b). At the same time that workers are stockpiling skills, it is also important
that economic growth would be used to create high quality jobs continuously so that the
economic returns to formal and non-agricultural employment will not deteriorate as the supply
of high skilled workers increases. This is discussed further in the next chapter. On the other
hand, the finding that the returns to basic services dropped faster than the rate at which access
187
to basic services increased, contributing to higher income poverty, could be indicative of higher
cost that low income households have to pay to access basic services due to the hike in
electricity tariffs and expanded value added tax in utilities which started in 2006. Thus, to
strengthen the correlation of increased access to basic services with poverty reduction, it is
important to minimize the cost needed to provide such services. This can be done by investing
more on infrastructure that can make the delivery of such services more efficient. However,
although there are signs of improvement, the availability of key infrastructure in the country
compares unfavourably with that in many of its Southeast Asian neighbours at present (ADB
2007, WB 2014).75 Nevertheless, given the high economic growth and higher liquidity in the
financial market nowadays, the government can respond to this problem by initiating more
infrastructure investment and providing a socio-economic environment that will attract non-
government players to play more actively in this role. In terms of access to assets, I find that
access to assets increased much faster than the rate at which the returns to asset ownership
dropped which in turn, contributed to lower poverty and inequality. For policy-makers, the
challenge is to provide an economic environment that will sustain this trend by ensuring that
access to productive assets is equitable and knowledge on how to use these assets for income-
generation is easily accessible to everyone.
By using the residuals from the estimated models as proxy to socio-economic shocks,
this chapter has also briefly examined the impact of shocks to poverty and inequality. At the
national-level, the results suggest that shocks have smaller impact on poverty gap relative to
the contribution of the changes in SECs and SERs between 2003 and 2006. In contrast, the
impact of shocks on the change in poverty gap between 2006 and 2009 is comparable with the
impact of changes in other factors, particularly the changes in returns to access to basic services
and returns to asset ownership. In addition, socio-economic shocks have also contributed to
increasing inequality. To some extent, this could mean that the shocks experienced by Filipino
households over the past decade had debilitating impact for the poor. To minimize the adverse
impact of economic shocks on poor and vulnerable households, social safety nets should be put
in place. Often, this is the responsibility of the government. However, some studies suggest
that the efforts of the government fall short in this respect. For instance, an ADB report
surmised that despite the country being used to environmental disasters, the relief provided
during such episodes remains inadequate (ADB 2007). Some studies also suggest that the weak
75 According to the 2013-2014 Global Competitiveness Index compiled by World Economic Forum, the Philippines is ranked
96th out of 148 countries based on the Infrastructure pillar. Its Southeast Asian neighbours rank higher: Malaysia (29th),
Thailand (47th) and Indonesia (61st) (WB 2014).
188
impact of the social protection programs in poverty reduction can be partially explained by the
low coverage and limitations in targeting appropriate recipients (ADB 2007, Bird and Hill
2009, Reyes et al. 2011). When formal social safety nets are not working effectively, low
income households would often turn to informal risk sharing networks where funds are raised
through gifts and loans among members (Fafchamps & Lund 2003). However, informal risk-
sharing is not always optimal for the poor (Fafchamps & Gubert 2007). In particular, although
some loans made through this channel are usually subjected to zero interest rates or do not have
to be repaid fully, others expect much higher payments leading the poor to further debts
(Platteau 1997). In addition, members of a risk-sharing network may have a hard time raising
funds if all of them are experiencing income shocks (Landmann, Vollan & Frolich 2012).
Furthermore, the funds raised through this channel may only cover a fraction of the income
shocks (Townsend 1994). Thus, it is important that policymakers examine the effectiveness of
both formal and informal social safety nets that exist today. Nevertheless, this topic warrants
further investigation using more sophisticated statistical tools to be able to better understand
the different short and long-run effects of shocks as it is highly probable that the different socio-
economic capital partially absorb the shocks with different timings.
There are also some limitations in this study. First, as the socio-economic correlates of
poverty and inequality that were used here could be considered as measures of material
deprivation. Although I used counterfactual simulations to measure their contribution to
changes in poverty and inequality, it is still difficult to conclude causal relationships. Second,
I might be more appropriate to examine the impact of changes in socio-economic returns to
capital using longer observation period.
Nevertheless, the findings of this chapter are sufficient to highlight that the problem on
poverty and inequality cannot be addressed by simply increasing the levels of socio-economic
capital of the people living at the bottom of the social hierarchy. Without any intervention, the
benefits of higher levels of socio-economic capital may just be washed out by lower economic
returns. Thus, socio-economic planners should devise policies that would ensure that economic
growth translates to improvement in socio-economic capital and creation of more opportunities
where this capital can be used more productively. Throughout this process, the importance of
providing access to social safety nets should not be taken for granted. In particular, although
the results suggest that economic shocks between 2003 and 2006 did not contribute
significantly to the observed changes in poverty gap, it drove US$2 headcount poverty rate to
increase. Between 2006 and 2009, shocks contributed to higher poverty, regardless of the
poverty index being used. In addition, economic shocks also contributed to higher income
189
inequality for all periods. Given that the Philippines has a wide range of social safety nets in
place (Ortiz 2001; Bird & Hill 2009; Reyes et al. 2011), the finding that income shocks have
pushed (headcount) poverty and inequality up, should prompt socio-economic planners to re-
evaluate the effectiveness of existing social protection programs. If left unaddressed, socio-
economic shocks may deter the country’s economic development.
190
Appendix Table A7.1 Descriptive Statistics of Variables Included in SEC Indices
2003 2006 2009
SEC Indicators Mean / Standard Mean / Standard Mean / Standard
Proportion Deviation Proportion Deviation Proportion Deviation
LOCATION Urban 0.49 0.5 0.49 0.5 0.49 0.5
NCR 0.09 0.29 0.09 0.29 0.09 0.29
Luzon 0.46 0.5 0.46 0.5 0.46 0.5
Visayas 0.22 0.41 0.22 0.41 0.22 0.41
(Mindanao) 0.23 0.42 0.23 0.42 0.23 0.42
(Proportion of working age hhld
members who have at most 0.36 0.37 0.35 0.36 0.32 0.35
EDUCATION
primary education)
Proportion of working age hhld
members who have at most 0.41 0.35 0.42 0.33 0.43 0.32
secondary education
Proportion of working age hhld
members who have postsecondary 0.23 0.32 0.24 0.32 0.25 0.32
education
EMPLOYMENT
members working in the non- 0.65 0.43 0.65 0.47 0.69 0.44
agriculture sector
members with formal employment 0.27 0.37 0.28 0.37 0.32 0.39
arrangement
0.69 0.4 0.73 0.39 0.74 0.37
members with permanent jobs
191
Appendix Table A7.1 (con’t) Descriptive Statistics of Variables Included in SEC Indices
2003 2006 2009
SEC Indicators Mean / Standard Mean / Standard Mean / Standard
Proportion Deviation Proportion Deviation Proportion Deviation
Has access to electricity at
0.78 0.41 0.83 0.38 0.87 0.34
BASIC SERVICES home
Has access to water faucet at
0.44 0.5 0.44 0.5 0.49 0.5
home
Has access to water-sealed
0.72 0.45 0.75 0.43 0.81 0.4
toilet facility at home
Has access to closed pit toilet
0.1 0.29 0.08 0.28 0.06 0.24
facility at home
Owns a house/lot 0.72 0.45 0.76 0.42 0.75 0.43
ASSETS
Owns a refrigerator 0.37 0.48 0.4 0.49 0.4 0.49

Owns a phone 0.31 0.46 0.54 0.5 0.71 0.45
Owns a car 0.12 0.33 0.18 0.39 0.25 0.43
192
Appendix Table A7.2 Estimated Contribution of Different Factors on Changes in Poverty and Inequality in the Philippines
2003-2006 2006-2009 2003-2009
Factor
FGT(0) FGT(1) FGT(2) Gini Theil FGT(0) FGT(1) FGT(2) Gini Theil FGT(0) FGT(1) FGT(2) Gini Theil
Location 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Economic returns to location 0.90 0.38 0.19 -0.29 -0.54 0.84 0.39 0.19 -0.29 -0.53 1.81 0.74 0.37 -0.57 -0.96
Education 0.04 0.10 0.09 0.27 0.29 -0.26 -0.12 -0.06 0.02 -0.14 -0.19 0.00 0.03 0.30 0.21
Economic returns to education 6.95 3.71 2.16 -0.08 -0.14 -0.49 -0.27 -0.15 0.01 0.01 6.54 3.34 1.88 -0.07 -0.12
Type of employment 0.23 0.40 0.30 0.40 0.39 -1.15 -0.50 -0.28 -0.25 -0.51 -0.52 0.00 0.06 0.22 -0.02
Economic returns to type of employment -7.68 -3.61 -1.91 2.02 3.89 0.13 0.07 0.03 -0.04 -0.07 -8.11 -3.57 -1.85 1.92 3.54
Access to services -0.69 -0.47 -0.28 -0.29 -0.71 -0.77 -0.45 -0.26 -0.25 -0.38 -1.64 -1.02 -0.58 -0.59 -1.10
Economic returns to access to services 5.82 2.44 1.24 -1.16 -1.94 0.52 0.25 0.13 -0.10 -0.16 6.60 2.69 1.33 -1.21 -1.95
Assets held -5.55 -2.31 -1.12 -0.23 -0.19 -4.45 -2.46 -1.34 -1.34 -2.66 -10.03 -4.65 -2.38 -1.48 -2.80
Economic returns to assets 0.80 0.27 0.12 -0.90 -1.53 0.56 0.22 0.10 -0.45 -0.74 1.47 0.50 0.22 -1.31 -2.13
Unobserved factors 2.21 -0.03 -0.42 1.71 2.18 0.71 0.29 0.04 0.56 1.96 2.75 0.26 -0.33 2.11 3.83
Total change 3.03 0.88 0.35 1.44 1.69 -4.36 -2.59 -1.60 -2.13 -3.20 -1.32 -1.71 -1.24 -0.68 -1.51
193
Appendix Table A7.3 Estimated Contribution of Different Factors on Changes in Poverty and Inequality, by Region
National Capital Region
2003-2006 2006-2009 2003-2009
Factor
Location 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Economic returns to location 1.61 0.32 0.10 0.00 0.00 1.37 0.38 0.12 0.00 0.00 2.44 0.59 0.19 0.00 0.00
Education 0.40 -0.06 -0.03 -0.22 -0.34 -0.15 0.02 0.00 0.08 -0.43 0.39 0.06 0.00 -0.01 -0.61
Type of employment 0.08 0.04 0.00 -0.17 -0.38 -1.05 -0.16 -0.03 -0.78 -1.57 -0.52 -0.13 -0.04 -0.39 -0.79
Access to services -0.08 -0.01 0.01 -0.02 -0.11 -0.51 -0.14 -0.06 -0.18 -0.26 -0.49 -0.15 -0.05 -0.19 -0.34
Assets held -1.35 -0.44 -0.15 -0.30 -1.14 -4.30 -0.67 -0.17 -1.40 -2.27 -4.21 -0.92 -0.29 -1.38 -2.80
Unobserved factors 0.84 0.05 -0.03 5.77 10.56 -1.00 -0.05 0.01 -1.48 -3.60 -0.29 -0.15 -0.06 3.34 5.19
Total change 5.64 0.66 0.15 5.66 9.36 -5.25 -0.50 -0.10 -4.13 -8.64 0.39 0.15 0.05 1.53 0.71
Luzon
2003-2006 2006-2009 2003-2009
Factor
Location 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Education 0.12 0.13 0.09 0.29 0.21 -0.30 -0.09 -0.04 0.07 0.00 -0.13 0.03 0.04 0.34 0.15
Type of employment 0.21 0.30 0.20 0.37 0.40 -1.36 -0.48 -0.22 -0.12 -0.33 -0.77 -0.06 0.02 0.17 -0.18
Assets held -6.24 -2.28 -1.02 -0.22 1.03 -4.48 -2.03 -0.99 -1.10 -1.94 -10.94 -4.24 -1.95 -1.41 -1.50
Unobserved factors 1.85 0.15 -0.09 0.99 -2.67 1.18 0.43 0.10 0.02 0.06 3.12 0.73 0.09 1.22 -1.68
Total change 2.92 1.09 0.56 1.00 -1.18 -3.88 -1.82 -1.03 -2.03 -3.58 -0.96 -0.73 -0.48 -1.03 -4.76
194
Appendix Table A7.3 (con’t) Estimated Contribution of Different Factors on Changes in Poverty and Inequality, by Region
Visayas
2003-2006 2006-2009 2003-2009
Factor
Location 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Education -0.12 0.15 0.15 0.60 0.78 -0.30 -0.20 -0.11 0.12 0.07 -0.36 -0.01 0.03 0.68 1.01
Type of employment 0.01 0.50 0.40 0.68 0.62 -0.86 -0.67 -0.39 -0.37 -0.68 -0.41 -0.06 0.04 0.35 -0.09
Assets held -5.45 -2.50 -1.32 0.64 0.07 -4.31 -3.11 -1.80 -1.36 -2.12 -10.23 -5.61 -3.03 -0.53 -1.58
Unobserved factors 2.12 0.01 -0.35 0.29 2.86 1.49 0.05 -0.29 2.09 5.56 3.06 0.10 -0.55 2.20 7.78
Total change 1.70 0.94 0.56 1.25 3.06 -3.55 -3.87 -2.61 -0.59 1.15 -1.85 -2.93 -2.05 0.66 4.21
Mindanao
2003-2006 2006-2009 2003-2009
Factor
Location 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Education -0.11 0.05 0.09 0.56 0.82 -0.17 -0.18 -0.10 -0.12 -0.21 -0.40 -0.09 0.01 0.45 0.62
Type of employment 0.53 0.64 0.51 0.59 0.61 -1.05 -0.50 -0.38 -0.07 0.69 -0.12 0.21 0.20 0.49 1.06
Assets held -5.98 -2.96 -1.55 0.20 -0.69 -4.59 -3.46 -2.11 -1.05 -2.51 -10.41 -6.11 -3.52 -0.92 -3.18
Unobserved factors 3.62 -0.49 -1.34 -0.46 0.75 -0.29 0.36 0.27 2.01 8.64 2.96 -0.39 -1.07 1.62 9.34
Total change 3.44 0.48 -0.16 0.06 0.19 -5.73 -3.78 -2.40 -0.31 4.98 -2.28 -3.30 -2.56 -0.25 5.16
195
Chapter 8 Multiple Jobholding and Socio-Economic Mobility in the Philippines
8.1 Introduction
The results presented in the previous chapters have highlighted the importance of employment
in inducing income mobility in the Philippines. Ideally, labour being one of the few assets available
to everyone, should be a vehicle for upward mobility, especially for the poor. However, when a
country’s labour market is highly segmented, the poorest of the poor may be trapped in long episodes
of low productivity and precarious employment. In other words, being employed is not a sure ticket
out of poverty. This is particularly true in developing countries whose labour markets largely operate
outside the periphery of government regulation. In addition to this informal economy, non-standard
employment arrangements are also increasing as globalization takes a stronghold on labour markets.
Worryingly, sparse data suggest that workers with non-standard jobs are also prone to sub-optimal
social protection coverage and work under precarious conditions (Addabbo & Solinas 2012; Ebisui
2012). Nevertheless, non-standard jobs can also have potential benefits. For instance, structured and
predictable flexibility associated with non-standard employment may enable workers to outline better
work patterns that are more compatible with their other personal responsibilities. This dualistic nature
and lack of a universally-accepted definition of non-standard employment makes it difficult to infer
whether its emergence helps in promoting upward economic mobility or contributes to increasing
labour market segmentation.76 In general, while policymakers need to better understand non-standard
employment arrangements to be able to expand social mobility prospects for workers relying on such
kinds of jobs, the literature is limited especially in developing countries (Ruyter et al. 2009).
The Philippines provides a relevant case study for examining the relationship between non-
standard employment and socio-economic mobility. We have seen from the previous chapters that
despite high economic growth, significant improvement in the overall income distribution remains
elusive as reflected in the slow pace of poverty reduction and persistently high income inequalities.
At the same time, labour market trends suggest that this occurs at the backdrop of stagnant job
creation. The country has one of the highest unemployment rates in South East Asia (about 7.0% in
2012).77 On the other hand, more than half of its employed population relies on jobs outside the formal
economy (ILO 2012). If many of these jobs have non-standard employment arrangements, then the
emergence of non-standard jobs provides opportunities to participate in economic activities for
76 “Standard and non-standard employment arrangement” terms can hardly be characterized with a precise legal meaning and have no
universally-accepted definition. However, some literature recognizes the following characteristics of standard employment: 1)
indefinite or permanent; 2) full-time; and to some extent, 3) done at the employer’s workplace. Given these, the literature identifies
three main sources of non-standard employment: casualization, informalization, and externalization.
77 The latest unemployment rates in other Southeast Asian countries are as follows: Indonesia (6.6%, 2011), Malaysia (3.0%, 2012),
Singapore (2.8%, 2012), Vietnam (1.8%, 2012), Lao PDR (1.4%, 2005), Thailand (0.7%, 2012) and Cambodia (0.2%, 2008) (WDI
2014).
196
workers who would have been unemployed otherwise. Nevertheless, it is still important to examine
the quality of employment of non-standard workers in the Philippines. For instance, if non-standard
jobs are systematically characterized by inferior working conditions, this may offset the job creation
benefits of non-standard employment in the long-run. For this study, I examine the case of multiple
job holding or pluriactivity as a form of non-standard employment. Owing to its conceptual
simplicity, the incidence of multiple job holding is a simple but valid indicator of the prevalence of
non-standard employment (Riddell & St-Hilaire 2002).78 Moreover, I distinguish constrained from
non-constrained pluriactivity to be consistent with the perceived dualistic nature of non-standard
employment.
Although labour force data suggest that a significant fraction of the Philippines’s employed
population are relying on multiple jobs, about 14.3% in 2009, the characteristics and working
conditions of multiple job holders have not been examined extensively in the existing literature.
Using the merged FIES-LFS, this chapter seeks to answer the following questions:
(i) What are the characteristics of multiple job holders? How do they differ from single job
holders?
(ii) Does multiple job holding improve a person’s socio-economic mobility prospects?
8.2. Theoretical Model for Multiple Jobholding and Income Mobility

8.2.1 Determinants of Multiple Jobholding
In general, evidence is mixed about whether multiple jobholding constitutes a temporary
phenomenon or a more permanent feature of the labour market, particularly in industrialized countries
(Wu, Baimbridge & Zhu 2009; Panos, Pouliakas & Zangelidis 2011; Casacuberta & Gandelman
2012). Traditionally, multiple job holding is seen as a temporary strategy to address sub-optimal
levels of utility derived from one’s primary job (Perlman 1966; Shisko & Rostker 1976; Krishnan
1990) or as a hedge against the risk of unemployment (Bell, Hart & Wright 1997). In other words,
workers engage in multiple jobs to avoid experiencing downward economic mobility. However,
recent evidence from industrialized countries suggests that multiple job holding can also be used to
develop further expertise and acquire new skills, which in turn, may lead to better occupational
outcomes (Panos et al. 2011). This type of labour supply behaviour can be part of a worker’s portfolio
of long-term strategies for career growth. Whether this also applies in developing countries is unclear
as this multiple job holding has not been studied extensively outside industrialized countries.79
78Technically, multiple job holding can be a combination of standard and non-standard employment (i.e., office employee with a full-
time day job and another part-time night job). Aside from multiple job holding, other indicators of non-standard employment include
part-time, self-, and short-tenure employment (Riddell & St-Hilaire 2002; De Bruin & Dupuis 2004).
79 Theisen (2009) argues that studies on developing countries’ labour markets usually start under the presumption that multiple job
holding is not a norm. This probably contributes to the dearth of studies examining this type of labour supply behaviour in developing
countries.
197
Multiple job holding potentially has both negative and positive aspects for workers. It may
provide additional income particularly useful for emergency purposes (Danzer 2011) and give
additional satisfaction especially when the second job is related to one’s personal interests (Renna &
Oaxaca 2006). It may also increase one’s productivity as it provides opportunities to acquire new
skills and develop expertise (Panos et al. 2011). Hence, pecuniary and non-pecuniary factors may
drive people to engage in multiple jobs. However, multiple job-holding has also some potential
disadvantages for workers. A second job may lessen one’s productivity by diverting a worker’s focus
to a multitude of tasks. Having multiple jobs may also mean less time for finding more productive
employment prospects. Moreover, this type of labour supply behaviour may have adverse
consequences for one’s health and family relationships if having multiple jobs mean working longer
hours (Alam, Biswas and Hassan 2009). Thus, even though multiple job holding has a potential to
provide more economic opportunities and to strengthen labour force, it may also increase workers’
vulnerability to socio-economic uncertainties.
The perceived positive and negative impacts of multiple job holding gave rise to a number of
theoretical models about the determinants of multiple job holding. Such theoretical models include:
hours constraint model, target income model, main job insecurity model, and heterogeneous job
portfolio model. The hours constraint model by Shisko & Rostker (1976), Bell et al. (1997), Conway
& Kimmel (1998) and Wu et al. (2009) provides a springboard to understand the other approaches.
According to the hours constraint model, workers usually aim to maximize their “utility” or the
level of satisfaction from consuming goods, services, or leisure. Consider an average worker with a
well-behaved utility function denoted by:
Utility = f(C, L) (8.1)
where C is a composite consumption good and L is (time spent for) leisure. The value of consumption
is usually subject to a budget constraint equivalent to an individual’s wage and non-wage income.
This can be represented as:
C = W + NW (8.2)
where W corresponds to wage income and NW to non-wage income. The income from work is subject
to hours constraint, i.e., the number of hours available to any worker is finite,
W = (T – L)*w
= h*w (8.3)
where T is the worker’s hours constraint, h is the number of hours spent for work and w is the hourly
income rate. Graphically, this can be represented by indifference curves and budget constraints. An
indifference curve is the combination of income and leisure which an individual would accept to
maintain a given level of utility while a budget constraint is the combination of goods and services
198
that the worker can avail given his/her income budget. Note that the slope of the budget constraint is
equal to the income rate.
We can distinguish between two types of pluriactivity: constrained and non-constrained
(Averett 2010). In constrained pluriactivity, the quality of the second job is usually inferior to the
quality of the main job because under this scenario, a worker is willing to take any additional job to
make ends meet. In contrast, in non-constrained pluriactivity, the quality of the second job can be at
par or better relative to the primary job.80 Figure 8.1 illustrates these concepts under the hours-
constraint model. In particular, the figure illustrates three levels of utility that can be attained by a
worker, depending on employment circumstances (Averett 2010). In this example, I3 denotes the
highest level of utility that can be attained by a worker with an (hourly) income rate w1 who spends
h1 + h2 hours for employment. In the left side of Figure 8.1, the curve I1 denotes the lowest level of
utility that the same worker can attain if he/she only spends h1 hours for employment. This happens
when the main job prevents him/her from working for H = h1 + h2 hours. However, the left side of
the figure also suggests that this worker can still attain a higher level of utility denoted by I2 if he/she
is willing to take on a second job even if it offers a lower (hourly) income rate. This represents
constrained pluriactivity in the sense that this pluriactive worker will earn less compared to a single
job holder with basically the same qualification (and thus the same wage rate w1) who also works for
H1+H2 hours. Under the hours constraint model, an inferior second job is one with a lower hourly
wage rate than the primary job. By contrast, the right side of Figure 8.1 represents non-constrained
pluriactivity wherein the wage rate offered in the second job is higher. In other words, a multiple job
holder will earn more relative to his counterpart with only one job. This may happen when the job
quality (in this case, income-related dimensions) of the secondary employment exceeds the quality
of the primary job. The case of a university professor or researchers engaged in part-time outside
consultancy projects or workers accepting a part-time employment as a second job while waiting for
a full-time employment opportunity to open up are some instances when this could happen.
In addition to the hours-constraint model, there are several other models identifying the factors
that drive workers into pluriactivity. For example, the target income model suggests that workers will
allocate work on different jobs to meet a specific income goal assuming that jobs offer different
pecuniary and non-pecuniary benefits (Lundborg 1995; Wu et al. 2009). This perspective is supported
by the findings of Krishnan (1990) who concluded that the propensity to take on multiple jobs
80 In rural areas, a good example of constrained multiple job holding is the combination of agricultural production with small-scale
non-farm entrepreneurial activities. In urban areas, workers may avail part time employment in different elementary occupations (e.g.,
working as an office cleaner in the morning and as a construction labourer in the evening). On the other hand, a university professor
doing part time consultancy jobs in the industry is an example of non-constrained pluriactivity.
199
Figure 8.1 Constrained and Non-constrained Pluriactivity
Source: Adopted from Averett (2010)
200
declines as the level of income received from primary job increases. On the other hand, according to
the main job insecurity model, workers whose main jobs are vulnerable or exposed to high risk of
termination may actively participate in dual job holding to cushion the effects of possible
unemployment (Bell et al. 1997). Boheim & Taylor (2004), for instance, found that the presence of a
permanent work contract in the primary job, as a proxy indicator of job security, reduces the
propensity of looking for a secondary job. Danzer (2011) also provided empirical support for the
main job insecurity model by concluding that having a secondary economic activity can be used as a
coping strategy to smooth income and ensure uninterrupted employment during wage shocks.
Alternatively, according to the heterogeneous job portfolio model, some workers may find incentives
to take more than one job because different jobs are not perfect substitutes. This implies that the wage
paid and utility lost from foregone leisure may not adequately reflect the benefits and costs of working
(Conway & Kimmel 1998; Wu et al. 2009). For example, Renna & Oaxaca (2006) found that some
workers have personal preferences for job differentiation, wherein they derive varying levels of
satisfaction from different occupations.
The determinants of multiple job holding described above can be summarized by estimating a
logistic labour-supply model,
𝑚𝑢𝑙𝑡
𝑙 ( 𝑖𝑡
0 ) = 𝜃𝑋𝑖𝑡
𝑖𝑡 (8.4)
where pit0 denotes the probability of having a single job while pitmult denotes the probability of taking
multiple jobs, and Xit is a vector of factors affecting the ith worker’s labour supply behaviour. We can
further generalize (8.4) to distinguish constrained from non-constrained pluriactivity by estimating a
multinomial logistic model denoted by:
𝑙
𝑙 ( 𝑖𝑡0 ) = 𝜃𝑋𝑖𝑡
𝑖𝑡 (8.5)
where l = 1, 2, pit1 denotes the probability of being engaged in constrained pluriactivity, while pit2
denotes the probability of being engaged in non-constrained pluriactivity.
8.2.2 Socio-Economic Mobility and Multiple Job Holding
In the previous chapters, each household member is assigned with the same level of welfare as
measured by the household expenditure per capita. To be able to examine how individual-level
occupation affects the socio-economic mobility prospects of a person, the analysis presented in this
chapter departs from that convention. In particular, this chapter uses two measures of socio-economic
mobility: mobility of employment income and occupational mobility. By using the person’s earnings
from employment rather than household expenditure per capita, I can focus on individual-level
mobility rather than joint socio-economic mobility of household members. On the other hand,
201
including occupational mobility in this context is also important for several reasons. First, even if it
were true that multiple job holding provides opportunities to acquire new skills as hypothesized, it is
possible that being equipped with new skills could take time before it translates to higher employment
income. In some cases, the new skills acquired from pluriactivity first open up employment
opportunities before leading to income mobility. Second, the fact that occupational mobility
indicators are often less prone to measurement errors than income mobility indicators make the latter
an attractive alternative indicator of socio-economic mobility.
Income Mobility
Standard human capital models express income as a function of skills or the capacity to
contribute to production for a given rental rate of each unit of skill (Bowles, Gintis & Osborne 2001).
Skill is traditionally measured in terms of years of education and labour market experience. Mincer
(1974) first formalized this mathematical relationship by expressing log income as a sum of a linear
function of years of education and a (quadratic) function of labour market experience such that:
𝑙 (𝑌𝑖𝑡 ) = 𝛼 𝐸𝑖𝑡 + 𝛼2 𝐿𝑖𝑡 + 𝛼3 𝐿2𝑖𝑡 + 𝜀𝑖𝑡
(8.6)
where Yit is the income of the ith worker at time t, α1 measures the rate of return to education for each
year in school Eit, α2 and α3 measure the return rate for each year of labour market experience Lit and
εit is the stochastic disturbance term. Assuming that the return rates to education and labour market
experience are fixed, this model suggests that income growth is a function of change in human capital
stock Eit and Lit,
 ln Yit  1Eit   2 Lit   3 L2it   it
(8.7)
However, previous studies show that these conventional indicators of human capital stock only
explain a small fraction of the total variation in individual income (Atkinson, Bourguignon &
Morrisson 1988). In other words, workers with the same amount of human capital can have
substantially different incomes. Similarly, changes in human capital stock over time would also
explain little variation in income growth. Over the years, researchers have identified that factors like
intergenerational reproduction of advantage (e.g., proxied by parental education), unobserved
individual heterogeneity (e.g., unobserved variations in effort) and spatial externalities could have
stronger effect on income and income growth (Engel, Rigobon, & Ferreira 2007) than changes in
schooling and labour market experience. In this study, I extend (8.7) to account for the quality of
employment within the context of multiple job holding. As mentioned earlier, I hypothesize that the
income mobility-effect of multiple job holding is asymmetric. For instance, for non-constrained
multiple job holders, I suspect that the arguments of Paxson & Sicherman (1996) and Panos et al.
(2011) may apply. In particular, secondary employment may accelerate the accumulation of skills
202
through additions to the job portfolio. This is because non-constrained multiple job holders have high
quality jobs that could induce positive human capital spill-over effects between primary and
secondary employment leading to higher productivity. Standard microeconomic theory suggests that
increasing productivity will also be compensated with faster income growth (Solow 1956). On the
other hand, for constrained pluriactive workers, I suspect that this type of labour supply behaviour is
not significantly correlated with increased income mobility. By definition, constrained pluriactive
workers do not maximize their income potential since the second job is usually paid at a lower wage
rate. In more equal societies where economic growth is uniformly distributed, having a lower initial
income may be correlated with faster income growth. However, for societies with high levels of
inequality, the inferior quality of the second job is less likely to induce income mobility for
constrained multiple job holders. To formally test these hypotheses, this chapter estimates a standard
income mobility model denoted by:
𝑌𝑖𝑡
𝑙 ( )= 𝑌𝑖𝑡− + 𝛼 ∆𝐸𝑖𝑡 + 𝛼2 ∆𝐿𝑖𝑡 + 𝛼3 ∆𝐿2𝑖𝑡 + 𝛽 𝑙 𝐿𝑆𝑖𝑡𝑙 + 𝜉𝑖𝑡
𝑌𝑖𝑡−1 (8.8)
where LSlit is the ith worker’s labour-supply behaviour l where l = 1 represents constrained
pluriactivity and l = 2 represents non-constrained pluriactivity. The parameter βl measures the impact
of specific type of labour supply behaviour on income mobility after controlling for initial income
and changes in human capital stock.
Occupational Mobility
To be able to examine occupational mobility, it is important to provide a yardstick of job

quality.81 In general, identifying the features of quality employment is not straightforward as the
concept may have different meanings for varying levels of development (ADB 2011b). For Filipinos,
findings from the World Values Survey show that income and job security are among the most
important factors that individuals identify when asked about the qualities they look for in a job. Except
for few factors, a stylized pattern also emerges where those in higher income brackets demand more
job benefits. However, other than people’s subjective beliefs about job attributes that are associated
with high quality jobs, there are limited objective data that can capture all of the multidimensional
features of job quality. For instance, the LFS only collects basic information about occupation type,
wages, and income. A way around this problem is to link the concept of employment quality with the
concepts of formal and informal jobs, that is jobs covered by the formal labour market regulations,
and those operating outside of such regulations. In this context, one can associate high quality
81Under the hours-constraint model of multiple job holding, job quality is gauged with respect to income levels. In other words, I can
distinguish constrained from non-constrained pluriactivity by comparing the hourly wage rate of one’s primary and secondary job.
However, I decided to use the concept of formal and informal jobs to provide a more multi-dimensional concept of job quality.
203
employment with having a formal job and low quality employment with having an informal job.
Certainly, this normative assumption is not without limitations. In some cases, skilled workers
voluntarily enter the informal economy for prospects of higher economic returns. In other words,
participation in the informal economy could also be an optimal choice for some workers who are
capable of getting jobs in the formal economy. This represents voluntary informal employment. On
the other hand, workers who have no choice but to take on low quality jobs in the informal economy
due to the lack of skills and structural barriers on entry to the formal sector represent involuntary
informal employment82. Nevertheless, empirical evidence from the Philippines as well as other
developing countries suggest that a significant number of informal workers are trapped in jobs with
inferior working conditions (WB 2010; ADB 2011b). With significantly lower income, informal
workers in the country are more likely to fall into poverty. In addition, a lack of social protection
coverage exposes them to greater socio-economic risks that may eventually lead to chronic poverty.
This provides a good motivation to use formal and informal employment as a rough measure of
quality of employment
To examine the relationship between occupational mobility and labour supply behaviour,
occupational mobility can be defined as a multinomial outcome which assumes a value of 0 if a
worker keeps the same type of job for two consecutive survey waves, 1 if a worker moves from an
informal main job to a formal main job and 2 if a worker moves from a formal main job to an informal
main job.
𝐽
𝑙 ( 𝑖𝑡0 ) = 𝑗 𝑌𝑖𝑡− + 𝛼 𝑗 ∆𝐸𝑖𝑡 + 𝛼2𝑗 ∆𝐿𝑖𝑡 + 𝛼3𝑗 ∆𝐿2𝑖𝑡 + 𝛽𝑗𝑙 𝐿𝑆𝑖𝑡𝑙 + 𝜉𝑖𝑡 (8.9)
𝑖𝑡
where pit0 denotes the probability of staying in the same type of employment arrangement between
time t and t+1, pit1 denotes the probability of moving from an informal to a formal main job while
pit2 denotes the probability of moving from a formal to an informal main job.
8.3. Data and Implementation of Concepts

8.3.1 Merged FIES-LFS
The analyses are based on the data from the FIES-LFS conducted by the NSO. As pointed out
in Chapter 3, the LFS is a quarterly survey that collects information on household members’
employment. Unlike in other chapters where households are used the main units of analysis and they
are weighted proportionally to the household size, this chapter uses individual-level data. In
particular, the data of all working-age members who are employed for at least two consecutive waves
from the 6,519 households that appear in all three waves of FIES-LFS (2003, 2006 and 2009) is
82Kucera & Xenogiani (2009) provide a good discussion of voluntary and involuntary informal employment by comparing the quality
of formal and informal jobs.
204
used.83 Although this data comprises a balanced sample of households, it does not have complete
longitudinal information for every member since individuals moving out of a sample household are
not tracked over time. As explained in Chapter 3, survey weight adjustments are used to account for
the potential bias that may be induced by attrition in all computations.
8.3.2 Measuring Socio-Economic Mobility
While the labour force survey collects various indicators of labour market participation of all
sampled household members, the survey collects earnings data from workers in wage or salaried
employment only. Employers, self-employed and unpaid family workers do not report any income in
LFS. Thus, income mobility can only be estimated for wage workers. In this context, income mobility
is defined as the annualized growth in wage workers’ earnings.84 On the other hand, occupational
mobility is gauged in terms of formal-informal job transitions.
According to the 17th International Conference of Labour Statisticians, jobs are considered
informal if the corresponding employment relationship is, “in law or in practice, not subject to labour
legislation, income taxation, social protection or entitlement to certain employment benefits (advance
notice of dismissal, severances of pay, paid annual or sick leave, etc)” (ILO 2004). In other words,
informal work refers to jobs which are typically outside formal labour regulation. However,
implementing this definition is not straightforward given data constraints. 85 To operationalize this
definition in LFS, this study adopts a classification system that is similar to the one used in Heriawan
(2004) and Martinez et al. (2014). In particular, the criteria used in distinguishing formal and informal
jobs are based on cross tabulating employment status and type of occupation (Table 8.1). Formal
workers correspond to all employers with permanent workers. This includes self-employed workers,
who are assisted by family members in the non-agriculture sector. Formal workers also include all
government employees and those who are employed as professionals. The rest are classified as
informal workers.
83
BLES defines the working age population as the household population 15 years and over (BLES 2011).
84 Although it is possible to use the household expenditure per capita as income measure for non-wage workers, it would be hard to
directly link the impact of multiple job holding on mobility if such income measure is used.
85 Depending on the available data, there are various ways of implementing the official definition and each approach may produce
different estimates. For example, by treating casual workers and unpaid family workers as informally employed, Cuevas et al. (2009)
estimated that at least 29% of all the employed in Indonesia are working in the informal economy based on 2007 Sakernas. However,
most sources put the estimates at a much higher level. For example, using the presence or absence of work contracts and information
about bookkeeping records reported in the 2009 Informal Sector Survey, ADB and BPS (2011) estimated that 89% of the employed
population in the predominantly agricultural province of Yogyakarta is informal, while 76% of total employment in the predominantly
urbanized province of Banten has informal employment arrangements. Furthermore, OECD (2010) estimates the informal employment
in Indonesia to be at least 70%. On the other hand, according to BPS Indonesia, about 68% of Indonesians were employed in the
informal economy in 2009.
205
Table 8.1 Definition of Formal and Informal Employment
Agriculture, Production
Administrative animal and related
Professional, Clerical and
and Services husbandry, workers,
technical and related Sales workers Others
managerial worker forest, Transport
related worker workers
worker fishermen, operators and
hunters labourers
Own account worker F F F I I I I I
Self-employed
assisted by family F F F F F I F I
worker
Employer F F F F F F F F
Government worker F F F F F F F F
Private worker
Casual worker in
agriculture F F F I I I I I
Casual worker in
non-agriculture
Unpaid family
I I I I I I I I
worker
Source: Martinez et al. (2014)
8.3.3 Distinguishing between Constrained and Non-Constrained Pluriactivity

As mentioned earlier, constrained pluriactivity refers to instances when a worker is willing to
take a second job with inferior quality relative to the characteristics of his/her first job. Non-
constrained pluriactivity refers to the opposite case. However, being able to implement this definition
depends on data availability. For instance, Martinez et al. (2014) defined that a multiple job holder in
Indonesia is engaged in constrained pluriactivity (relative to a single job holder with the same type
of primary job) if he/she is either (i) holding a formal main job and an informal secondary job, or (ii)
holding two informal jobs. On the other hand, a multiple job holder is engaged in non-constrained
pluriactivity if he is either (iii) holding two formal jobs or (iv) holding an informal main job and a
formal secondary job. However, following this definition in the Philippine context is problematic for
two reasons. First, the LFS data provide limited information about the second job which prevents me
from classifying whether the second job has a formal or an informal arrangement. Second, defining
constrained and non-constrained pluriactivity in terms of formal and informal job could create
circularity problems since our measure of occupational mobility also depends on the
formality/informality of a worker’s job. In this study, I follow an indirect approach. A worker is
considered to be in constrained pluriactivity if he/she has multiple jobs and comes from a household
who are consuming more than what they are earning (i.e., household expenditure exceeds household
income). This definition is premised on the argument that liquidity constraints affect occupational
choices (Giannetti 2011). In particular, those who are exposed to higher risks of liquidity constraint
may not have the leisure to choose better quality jobs.
Following the definitions outlined above, survey estimates show that about 7 in 10 workers
were informally employed in their main jobs. Among the employed in 2009, approximately 11% had
206
multiple jobs. Of these multiple job holders, about 43% are in constrained pluriactivity while the 57%
are in non-constrained pluriactivity.

8.4.1 Background on the Philippines’s Labour Market Over the Past Decade
One possible reason why the high incidence of poverty and pervasive income inequalities have
remained prominent features of the Philippines’s development process despite strong economic
growth rates is that the quality of jobs held by workers at the bottom of the income pyramid has not
improved significantly. Previous studies show that to be able to move forward into a higher and
sustained level of development, it is important to expand good quality employment opportunities to
the poor (ADB 2011b, WB 2013 and 2014). This section examines how the quality of employment
in the country has changed over the past decade.
Table 8.2 provides a summary of the employment trends based on key labour market indicators
since 2003. On the positive side, one can find a noticeable drop in the proportion of the labour force
without jobs during this period. Despite this progress, underemployment rates or the proportion of
employed persons who are either looking for a second job, a new job with longer work hours or wants
additional work hours in their current jobs, increased. In a developing country like the Philippines,
the underemployment rate is probably a more telling indicator than unemployment rate because the
poor which comprises a significant fraction of the population cannot afford to remain unemployed
for extended period of time. The results also portray a declining trend in labour participation rates for
both men and women. This is in sharp contrast to the trends observed in previous years when labour
participation rates, especially among women, were increasing (KILM 2014). Survey estimates
suggest that labour participation rate among men dropped from 82% in 2003 to 79% in 2012 while
the proportion of working age women entering the labour market declined from 51% to 50% during
the same period. Taken in a comparative context, although the Philippines’ labour market can be
characterized with higher participation rate, higher incidence of unemployment and
underemployment are more prominent features of its labour market structure compared to other Asian
countries (Montalvo 2006).
Tables 8.3 to 8.5 describe the distribution of the proportion of workers employed by production
sector, occupation group and type of employment, respectively. Interestingly, while agriculture
remains to be the sector with the highest contribution to total employment, its share has dropped from
35% in 2003 to 30% in 2012. The declining role of agriculture sector has translated to an expanding
employment in service-oriented sectors whose share to total employment increased by 5.3 percentage
points over the past decade. On the other hand, the contribution of the industry sector has become
207
Table 8.2 Trends in Key Labour Market Indicators, 2003-2012
Employment Indicator 2003 2006 2009 2012
Labour Force (in million) 34570.8 35464.1 37894 40432
Labour participation rate, total
66.7 64.2 64 64.2
(% of total population ages 15+)
Labour participation rate among men 82.2 79.3 78.7 78.5
Labour participation rate among women 51.4 49.3 49.4 50
Unemployment rate
11.4 8 7.5 7
(% of the labour force)
Underemployment rate
17.5 21.5 19.7 20.9
(% of the employed)
Source: Author’s computations using data from the longitudinal subsample of FIES-LFS 2003, 2006, 2009, BLES
data and KILM (2014)
stagnant as its share to total employment decreased by roughly 0.5 percentage points. In terms of
occupations, the past decade has seen a moderate increase in the number of workers holding
managerial positions. This is also accompanied by a consistent increase in the share of clerical and
sales. On the other hand, the proportion of employed who are production workers (i.e., trades and
related workers, plant and machine operators and assemblers) has declined significantly while the
proportion of labourers and unskilled workers has observed a small increase. In terms of type of
employment, the previous decade has witnessed a significant shift from self-employment to wage and
salaried employment. In particular, self-employment dropped by 7 percentage points from 2003 to
2012 while the proportion of employed in wage and salaried jobs increased by the same amount.
However, the country continues to operate with a significant share of unpaid family work. From 2003
to 2012, the proportion of employed people in unpaid work barely changed from 10% to 9%. Overall,
while non-agricultural employment is expanding, the pace of reduction of employment in agriculture
sector has been much slower compared to the marked shift from agricultural to non-agricultural
employment that transpired before the Asian financial crisis (WDI 2014). In addition, the increasing
role of the services sector to total employment can be mostly attributed to the higher proportion of
persons employed in low-paying service-oriented jobs.
Tables 8.6 and 8.7 provide further insights on how the quality of employment in the country
has evolved over the years. For instance, the estimates suggest that real incomes of workers with
wage and salary jobs increased by approximately 1.7% per year from 2003 to 2012. Paid workers
from family-operated activities noted the fastest annual income growth (3.3%) while those working
for private households experienced the lowest rate of increase in income (0.9%). Furthermore, there
has also been a gradual increase in the proportion of the labour force who have formal employment
arrangements. Interestingly, the proportion of those who take multiple jobs, an approximate indicator
208
of the prevalence of non-standard employment arrangements, comprise a non-negligible portion of
the labour force and more importantly, have shown signs of increasing trend.
In summary, a quick examination of key labour force indicators reveals that unlike its macro-
economic growth, the country’s performance in the employment front portrays a mixed picture. On
the positive side, the statistics show increasing non-agricultural and formal employment. However,
the improvement in the quality of jobs held by those who are at the bottom of the occupational ladder
has been less remarkable with the unemployment and informal employment rates remaining high. In
other words, the issue is less about a significant fraction of the country’s population not having jobs
but more on the observed pattern that many of those who are employed remain in low quality
employment. Worryingly, a quick examination of the labour force survey also reveals that moving
into better jobs is not an easy task. For instance, only about three in five of the initially non-employed
(i.e., unemployed and not in the labour force) reported having a job in the succeeding wave.
In addition, not everyone who gets a job always remain employed, wherein approximately 10%
of those who initially had a job were found to be either unemployed or not in the labour force in the
following survey period. Furthermore, I also find that only about one in five who were initially
Table 8.3 Distribution of Workers (%), by Production Sector of Main Job

Production Sector 2003 2006 2009 2012
Agriculture 35.4 34.7 32.8 30.4
Mining 0.4 0.4 0.5 0.7
Manufacturing 9.9 9.3 8.4 8.3
Electricity, Gas and Water 0.3 0.4 0.4 0.4
Construction 5.4 5.2 5.4 6
Wholesale and Retail Trade 18.5 18.5 19.6 18.8
Hotels and restaurants 2.6 2.7 3.1 4.1
Transport, storage and
7.7 7.9 7.6 8.1
communication
Financial intermediation 1 1 1.1 1.1
Real estate, renting, and business
2.2 2.3 3.1 3.3
activities
Public administration and defence,
4.7 4.4 5.1 5.2
compulsory social security
Education 3 3.1 3.2 3.4
Health and social work 1.2 1.1 1.2 1.2
Other community, social and
2.8 2.4 2.6 2.6
personal service activities
Private households with employed
4.9 4.9 5.9 6.1
persons
Source: Author’s computations using data from the longitudinal subsample of FIES-LFS 2003,
2006, 2009 and BLES data.
209
Table 8.4 Distribution of Workers (%), by Main Occupation
Occupation 2003 2006 2009 2012
Officials of government and special
interest organizations, corporate
12.3 12.1 14.5 16.1
executives, managers, managing
proprietors, and supervisors
Professionals 4.3 4.3 4.4 4.9
Technicians and associate
2.8 2.7 2.7 2.8
professionals
Clerks 4.3 4.9 5.3 5.7
Service workers and shop and
9.3 9.8 10.7 12.6
market sales workers
Farmers, forestry workers and
18.6 17.6 15.4 12.7
fishermen
Trades and related workers 9.2 8.1 7.7 6.8
Plant and machine operators and
7.6 7.7 6.1 5.3
assemblers
Labourers and unskilled workers 31.3 32.3 32.7 32.9
Special occupations 0.4 0.4 0.4 0.2
Source: Author’s computations using data from the longitudinal subsample of FIES-LFS 2003, 2006,
2009 and BLES data.
Table 8.5 Distribution of Workers (%), by Status of Main Employment

Type of worker 2003 2006 2009 2012
Wage and salary workers 53 53.4 55.8 60.2
Private household 5.7 5.9 5.6
Private establishment 39.4 41.3 46.1
Government 7.8 8.2 8
Family owned business 0.5 0.3 0.4
Self-employed 37.1 35.1 33.6 30.4
Own-account worker 30.4 29.4 26.9
Employer 4.7 4.2 3.5
Unpaid family worker 10 11.5 10.6 9.4
Source: Author’s computations using data from the longitudinal subsample of FIES-LFS
2003, 2006, 2009 and BLES data.
210
Table 8.6 Average Daily Basic Pay of Wage and Salary Workers
(in 2005 PPP US$)
Type of worker 2003 2006 2009 2012
All Wage and Salary
10.97 11.7 12.78 12.76
Workers
Private household 5.15 5.22 5.76 5.59
Private establishment 10.3 11.32 12.41 12.02
Government /
Government 18.78 19.48 21.04 22.42
Corporation
Family-operated
7.32 7.31 8.17 9.86
business
Source: Author’s computations using data from the longitudinal subsample of FIES-LFS
2003, 2006, 2009 and BLES data.
Table 8.7 Distribution of Employment Status (%), 2003-2009

Type of worker 2003 2006 2009
Single job holder with
22.88 23.09 25.66
formal job
Multiple job holder with
2.59 2.53 2.87
formal main job
Single job holder with
56.44 55.86 53.28
informal job
Multiple job holder with
6.78 7.27 7.12
informal main job
Unemployed 11.31 11.23 11.08
Source: Author’s computations using data from the longitudinal subsample
of FIES-LFS 2003, 2006, 2009
employed in the informal sector finds a job in the formal sector in the following survey wave. These
results set the tone for the need to investigate the mechanisms through which social mobility can be
facilitated. In this study, I examine the case of non-standard arrangements, particularly, multiple job
holding.
8.4.2 Discussion of Empirical Results
Descriptive Trends
Survey estimates suggest that multiple job holding is a significant part of total production in the
Philippines, increasing from 10.4% in 2003 to 11.1 in 2009. In 2009, empirical data suggests that
about 62% of pluriactive workers in the Philippines were in paid employment in their main jobs while
the remaining 38% were in self-employment (including unpaid family work). On the other hand,
about 57% of multiple job holders held their main jobs in the agricultural sector, 10% in industry and
33% in service-oriented sectors. Furthermore, 65% of multiple job holders were engaged in
elementary occupations for their main jobs (including agricultural work), while 15% were in sales
211
and other service oriented positions. About 20% of multiple job holders in the sample were engaged
in professional, administrative and managerial jobs.
Interestingly, 63% of pluriactive workers reported a different secondary occupation, while 37%
had the same line of work for their main and secondary jobs. Table 8.8 summarizes what kind of work
multiple job holders in the Philippines take as their second jobs. In particular, agricultural work as a
secondary activity is quite common among workers especially those holding elementary occupations
in their primary jobs. Conversely, combining agricultural work is least common among professionals,
technical workers and those holding administrative and managerial positions. This is not surprising
considering that professionals and technical workers are more likely to be in urban areas, where
agricultural employment is not common. Furthermore, agricultural workers and labourers are more
likely to be engaged in the same occupation for their second jobs.86
Determinants of Multiple Job Holding87

To identify the determinants of pluriactivity, equations (8.4) and (8.5) are estimated using
logistic regression with robust standard errors to adjust for the correlation among repeated
observations for the same individual. The results show that Filipino men are more likely to have a
second job than their female counterparts (Table 8.9). This is different from the findings in other
countries which typically report that women are more likely to get a second job than men.
Nevertheless, the gender difference in multiple job holding rates has slightly decreased over the years
with the proportion of pluriactive women increasing from 8.1% in 2003 to 9.7% in 2009, while that
of men increased only from 16.3% to 16.8%. Household composition also seems to matter. As the
family size increases, the propensity to take multiple jobs tends decrease but the average age of other
household members is negatively correlated with the propensity to be pluriactive. Moreover, the
burden of getting a second job is usually left to the head of the household.
Less educated workers are more likely to get a second job in the Philippines. For instance,
those who only had primary education were approximately 1.5 times more likely to get a second job
than those who had secondary or college education. On the other hand, there is a declining propensity
to get a second job as an individual moves up the income ladder – a pattern consistent with the target
income model of pluriactivity. In particular, workers from the poorest 20% households are
approximately three times more likely to get a second job than workers from the richest 20%
households. Nevertheless, the fact that as many as 8% from richest quintile are also engaged in
86
An interesting avenue for future research is to focus on multiple job holding in the agriculture sector using detailed
income data from agricultural sources. In particular, future research may examine the interaction between “push” and
“pull” factors and how this affects an individual’s income mobility prospects through pluriactivity.
87
Because of the limited number of survey waves, the statistical models do not control for individual-specific effects.
212
Table 8.8 Distribution of Multiple Job Holders (%), by
Type of Occupation in Main and Secondary Jobs
Officials of Plant and
Government Technicians Trades and Machine Laborers and
Special Service Agricultural
Main Job \ Secondary Job and Special Professionals and Associate Clerks Related Operators Unskilled # obs
Occupations Workers Workers
Interest Professional Workers and Workers
Organizations Assemblers
Special Occupations 0 38.46 0 0 0 0 41.03 0 0 18.46 5

Officials of Government and
Special Interest 0 26.96 0.17 1.48 3.14 8.81 30.98 6.72 5.58 16.14 151
Organizations
Professionals 0 32 6 25.6 0 11.2 22.4 0 2.48 0.54 22
Technicians and Associate
0 2.11 0 9.31 0 19.61 13.73 17.65 7.84 30.39 28
Professional
Clerks 0 20.9 0 15.67 5.22 4.85 40.3 0 2.95 9.7 29
Service Workers 0 20.18 0 2.39 0 10.7 21.1 12.23 5.81 27.83 42
Agricultural Workers 0.17 9.85 0 0.83 1.33 3.87 45.56 6.31 3.24 28.83 652
Trades and Related
0 5.73 0 1.86 0 7.28 40.56 10.84 7.74 26.01 83
Workers
Plant and Machine
1.73 15.45 0 1.91 0 2.24 39.43 9.96 12.4 16.87 53
Operators and Assemblers
Laborers and Unskilled
0.47 4.47 0.43 2.27 0.31 5.62 29.09 6.31 2.52 48.56 420
Workers
Source: Author’s computations using 2003 data from the longitudinal subsample of FIES-LFS 2003, 2006, 2009
Note: Detailed information about the secondary job is not available from 2006 round onwards.
213
multiple job holding suggests that dual job holding is not always a matter of constrained pluriactivity
as previously inferred. Multiple job holding rates also differ across geographic locations. Workers
from urban areas are less likely to take multiple jobs compared to their rural counterparts. In
particular, self-employed agricultural workers are more likely to be pluriactive than paid workers in
the non-agriculture sectors.
The results also confirm the hours-constraint hypothesis. In general, multiple job holders are
more likely to work for less than 35 hours in their main jobs compared to single job holders.
Interestingly, while the prevalence of multiple job holding is generally higher among those engaged
in fewer hours of work in their primary job, multiple job holding is still quite high for those who are
working for at least 35 hours per week in their primary jobs.88 In particular, about 9.5% of those who
are working for at least 35 hours per week in their main jobs are engaged in multiple economic
activities.
Consistent with the findings from existing literature, the estimated models show that the
motivation to have multiple jobs in the Philippines is generally associated with the presence of
constraints in one’s primary job. Both income and non-income factors make up such constraints. For
instance, both wage employment and the number of hours worked in a person’s main job decrease
the propensity to be pluriactive while living in rural agricultural areas and being engaged in the
agriculture sector tend to increase the propensity of an individual to take multiple jobs. Nevertheless,
as pointed out earlier, multiple job holding is not always a case of constrained pluriactivity which
could be indicative that the determinants of constrained and non-constrained pluriactivity are
different. Estimation of (8.5) allows me to examine this hypothesis. The results suggest that higher
educational attainment increases the propensity to be engaged in non-constrained pluriactivity but
decreases the odds of becoming a constrained multiple job holder. Wage employment decreases the
odds of falling into constrained pluriactivity while self-employment increases it.
Relationship between Economic Mobility and Multiple Job Holding

I argued earlier that being employed is not a sure ticket out of poverty and in most cases, the
quality of jobs held is important. However, landing a job with satisfactory quality that is enough to
lift poor workers out of poverty is often a function of origins, skills, effort and luck (Piketty 1995;
Kochar 1999). To some extent, a worker’s decision to be pluriactive could be a sign of effort that is
motivated by the desire to improve one’s living standards. Moreover, recent evidence from
88Compared to industrialized countries, I consider 15% as a high proportion of the population with multiple jobs. In industrialized
countries, the incidence of multiple job holding is about 5% to 10% (Australian Bureau of Statistics (ABS) 2009; Wu et al. 2009;
Campbell 2011;).
214
Table 8.9 Regression Coefficients of Logistic and Multinomial Logistic Models on the
Propensity to Take Multiple Jobs
Logistic Model Multinomial Logistic Model
Variable Non-constrained Constrained
Pluriactive
pluriactivity pluriactivity
Urban -.54*** -.43*** -.74***
Hhld head .58*** .59*** .53***
Male .26*** .28*** .28***
Age .083*** .081*** .087***
Age squared -.00089*** -.00085*** -.00095***
Educational attainment
(base = primary education)
secondary education -.026*** -.049*** 0.0093
tertiary education -.072*** 0.0045 -.25***
Main job is formal .53*** .5*** .55***
Employer in main job
Wage/salaried job -.041*** -0.0012 -.13***
Self-employed .19*** .13*** .25***
Unpaid family work .22*** .056*** .4***
Main job is in agriculture sector
Manufacturing -.52*** -.44*** -.64***
Services sectors -.42*** -.29*** -.6***
Number of hours in main job -.023*** -.022*** -.026***
Wants to work more hours 1.2*** 1.2*** 1.2***
Family size -.014*** -.011*** -.021***
Has spouse .2*** .092*** .4***
Average age of other household
-.0099*** -.0039*** -.021***
members
Income quintile
2nd quintile -.14*** -.17*** -.091***
3rd quintile -.093*** -.037*** -.13***
4th quintile -.1*** .032*** -.3***
5th quintile -.2*** -.096*** -.39***
Intercept -3.2*** -4*** -3.8***
Source: Author’s computations using data from the longitudinal subsample of FIES-LFS 2003, 2006, 2009
Notes: *** - p< 0.01, ** p <0.05, * - p < 0.1
industrialized countries suggest that pluriactivity provides a good venue to acquire new skills or
improve existing ones which could eventually open up an avenue of better economic opportunities.
Nevertheless, this type of labour supply behaviour may not always result in a worker’s improved
living standards through acquisition of new skills. For one, high inequalities lead to labour market
segmentation wherein access to high quality jobs is limited to a privileged few. This makes the
relationship between socio-economic mobility and multiple job holding an empirical issue.
215
After holding other factors such as changes in sectoral transitions fixed, the results suggest that
among workers in wage or salaried jobs, non-constrained multiple job holders experienced faster
income growth than either constrained multiple job holders or single job holders. Given that the
statistical model from which this conclusion has been drawn is based on data from workers who
remain in wage or salaried jobs in two consecutive waves only, one should be cautious in concluding
that multiple job holding leads to economic mobility that would allow the poor workers to catch-up
with the rest. In the Philippines, more than half of the poorest 40% are workers who are either self-
employed or engaged in unpaid family work. To include them in the analyses, I also estimated the
occupational mobility model described in Section 8.2. The results suggest that after holding other
factors fixed, having multiple jobs is weakly correlated with higher income growth but strongly
correlated with formal to informal or informal to formal job transitions. Interestingly, compared to
single job holders, unconstrained pluriactivity decreases the odds of moving from informal to formal
jobs and increases the odds of moving from formal to informal jobs. On the other hand, constrained
pluriactivity increases the odds of both informal to formal and formal to informal job transitions. In
other words, the results are indicative that the impact of multiple job holding on Filipino workers’
prospects of economic mobility is mixed. For some, multiple job holding leads to faster income
growth while for others, this type of labour supply behaviour increases occupational mobility but the
accompanying mobility is not necessarily an upward mobility. There are several possible
explanations for this. The most intuitive explanation is that having multiple jobs serves as a coping
response and tool to avoid experiencing more severe forms of poverty during times of economic
uncertainties. It could also be the case that some multiple job holders are more interested in the non-
pecuniary benefits of having multiple jobs that is not adequately captured in the model of occupational
mobility. For example, a worker may take a second job that is related to his/her personal hobbies. In
some cases, having multiple jobs may also lead to more flexible schedule that would allow one to
balance work and other personal responsibilities. However, it is hard to test this hypothesis due to
data limitations. Another possible reason why pluriactivity is giving mixed signals in terms of its
relationship with socio-economic mobility is that our data only allows us to estimate mobility between
two time periods that are three years apart. It is possible that the effect of multiple job holding
gradually tapers off over time. If this is true, then we may need to rely on longitudinal data which are
collected more frequently to be able to draw more conclusive inferences.
8.5 Conclusion and Policy Implications

Amidst rapid economic development, the Philippines confronts the challenges of jobless
growth, slow reduction of poverty and income inequalities. These trends could have adverse
216
Table 8.10 Regression Coefficients of Economic Mobility Models
Income Mobility
Multinomial Logistic Model
Model
Variable
Informal to Formal Formal to Informal
Wage growth
Main Job Transition Main Job Transition
Urban 0.014 -.11*** 0.0061
Hhld head -.029* -.066*** .088***
Male .046*** -0.0055 -.032*
Age 0.0002 .041*** .073***
Age squared -0.0000094 -.00036*** -.0006***
Educational attainment
(base = primary education)
secondary education -0.018 .2*** .17***
tertiary education -0.029** -0.038*** .15***
Sector of employment remained
(main job)
Moved from agriculture to non-
.055*** 1.7*** -.46***
agriculture
Moved from non-agriculture to
-.12*** -.14*** 1.5***
agriculture
Change in family size 0.00049 .01*** .0044*
Change in the proportion of hhld
-0.034 .13*** -.36***
members who are employed
Income quintile
2nd quintile .041** .36*** .22***
3rd quintile .065*** .52*** .24***
4th quintile .088*** .88*** .34***
5th quintile .15*** .73*** .057***
Single job holder
Non-constrained multiple job
.032* -.065*** .26***
holder
Constrained multiple job holder 0.0034 .023** .36***
Intercept -0.0025 -3.6*** -4.6***
Source: Author’s computations using data from the longitudinal subsample of FIES-LFS 2003, 2006, 2009
Notes: The dependent variable for the wage growth model is employment earnings.
*** - p< 0.01, ** p <0.05, * - p < 0.1
consequences for the long-term growth prospects of a country, as well as for a broad range of human
welfare outcomes (Wilkinson and Pickett 2009).
Increasing inequalities and labour market segmentation are mutually reinforcing. Thus, labour
market policies are important tools for addressing the adverse consequences of increasing
inequalities. For this, policymakers need sufficient data to identify the vulnerable workers. Previous
studies show that a bulk of these vulnerable workers have non-standard employment arrangements.
An important form of non-standard employment practices that has been identified in the literature is
217
multiple job holding. Previous studies suggest that this labour supply behaviour is typically used as a
coping mechanism against risk of unemployment or income shortfall. However, recent evidence from
industrialized countries suggests that it can also be used as a means to move into better occupations.
The present paper adds to the existing literature by offering a detailed examination of multiple job
holders in the Philippines – a country for which the empirical evidence of the patterns of pluriactivity
is scarce.
The analysis of a nationwide panel data from the FIES-LFS, reveals that multiple job holding
is quite prevalent in the Philippines, accounting for more than 10% of the employed between 2003
and 2009. In addition, the results suggest that men, especially those who are head of households,
those less educated, rural agricultural workers, underemployed and workers from the lower income
quintile are more likely to get a second job. This confirms that the propensity to be pluriactive is
primarily driven by socio-economic constraints. In particular, almost half of multiple job holders are
engaged in constrained pluriactivity, that is having multiple jobs to be able to make ends meet. Despite
the fact that constrained pluriactivity can be a potent tool for avoiding socio-economic vulnerabilities
in the short-run, the estimates provided in this chapter show that it is not strongly correlated with
better mobility outcomes. Moreover, there is some evidence to suggest that highly skilled workers
who are already on top of the socio-economic hierarchy observe immediate increase in income from
multiple job holding. If this pattern holds for other forms of non-standard employment, then it can be
argued that non-standard employment contributes to further segmentation of the labour market
wherein the benefits are being channelled disproportionately to those who are already on top of the
socio-economic hierarchy. In this context, the challenge confronting policymakers is to ameliorate
the expanding labour market segmentation in the country. This entails improving the working
conditions of individuals with non-standard jobs, particularly those who combine multiple precarious
employment. For instance, a system could be considered to compensate for the risks associated with
non-standard jobs using higher incomes or provision of other non-pecuniary benefits, across all
segments of the labour market. Finally, to be able to better respond to the needs of researchers and
policy-makers, there is a need to improve existing data collection systems on labour market
indicators. In particular, there is a need to collect more reliable indicators of working conditions
within the context of continuously changing labour markets.
218
Chapter 9 Evaluating the Feasibility of Using Pseudo-Panel Data to Measure Income Mobility
9.1 Introduction
Using the Philippines as case study, the previous chapters underscored the usefulness of income
mobility analysis for understanding how income distribution evolves over time. Despite the
advantages of examining income mobility, it was almost an uncharted field in developing countries
until recently and most of the existing studies (including this thesis) cover short periods of time only
(Fields 2011). One of the main factors that contribute to the dearth of income mobility studies in
developing countries is the high cost of collecting panel data from where information on the income
of each person or household, and changes in income over time, can be derived (Deaton 1997;
Bourguignon, Goh & Kim 2004; Antman & McKenzie 2007; Fields et al. 2007; Cuesta, Nopo &
Pizzolitto 2011). In this chapter, I evaluate the performance of different pseudo-panel estimation
techniques that have been proposed in the literature in measuring different income mobility indicators
when panel data is not available. Moreover, I further develop these methods to estimate a wider range
of mobility indices in addition to the ones that these methods were originally designed to measure.
Over the years, a number of researchers have attempted to reconcile the need for examining
income mobility with the lack of panel data by exploiting the information provided by other data
sources. In particular, methodological research has focused on exploring the usefulness of repeated
cross-sectional surveys because many developing countries regularly collect income (and non-
income) indicators of development at the micro-level through cross-sectional surveys. The main issue
in working with cross-sectional surveys in the context of income mobility analysis is that it is not
straightforward to measure the amount of temporal dynamics in each person’s or household’s income
given that cross-sectional surveys do not follow the same set of respondents over time.
To address this issue, researchers proposed a methodology called pseudo-panel estimation
which entails constructing synthetic panels from repeated cross-sectional survey data. There are
several variants of pseudo-panel estimation. Four of the most recent developments in the literature
are the methods proposed by Bourguignon, Goh & Kim [BGK] (2004), Antman & McKenzie [AM]
(2007), Dang, Lanjouw, Luoto & McKenzie [DLLM 1] (2011) and Dang, Lanjouw, Luoto &
McKenzie [DLLM 2] (2014). These methods employ different procedures for creating synthetic
panels and are originally designed to measure different aspects of income mobility. For instance, the
AM approach is designed to examine the origin-independence perspective by extending the procedure
initially proposed by Deaton (1985) which entails grouping all observations into different mutually
exclusive and exhaustive cohort groups. The characteristics of interest are then averaged for each
cohort group, and in turn, the cohort averages serve as the pseudo-panels. On the other hand, the
BGK, DLLM 1and DLLM 2 approaches are designed to measure poverty dynamics or the amount of
219
income movements at the low-income range. All three approaches involve estimating structural
models to impute the unobserved incomes of respondents from a specific cross-sectional survey wave
while maintaining the original respondents as the units of analysis. The main objective of this study
is to demonstrate how these four methods can be extended to be able to estimate a wider array of
income mobility measures.89 In particular, the chapter addresses the following questions:
(i) Are the proposed pseudo-panel techniques useful for measuring income mobility? In general,
which of the techniques are most desirable?
(ii) Does the performance of the pseudo panel techniques’ depend on the type of income mobility
measure being estimated?
(iii)How can the existing pseudo-panel techniques be improved?
9.2 Developments in Income Mobility Estimation using Repeated Cross-Sectional Data

Several researchers have proposed a variety of statistical techniques which use information
from repeated cross-sectional surveys to create synthetic panel data. This section reviews the
methodological developments in measuring income mobility using pseudo-panel methods. The
discussions begin with a review of the classical pseudo-panel estimation methodology pioneered by
Deaton (1985) followed by the presentation of four contemporary pseudo-panel approaches that are
designed to answer specific income mobility-related questions.90
9.2.1 What is Pseudo-Panel Estimation?
In general, the pseudo-panel approach refers to a class of statistical and econometric methods
that use repeated cross-sectional surveys to estimate indices or models that are typically suitable for
longitudinal studies. For exposition, consider the following time-indexed static model
Yit  X it  f i   it
i = 1, 2,…,N, T= 1, 2, …, T (9.1)
where Yit is the response outcome for unit i at time t, Xit is a vector of explanatory variables, fi is an
unobserved unit specific effect and εit is a random disturbance term. Conventionally, the parameters
of this model can be estimated using fixed-effects (FE) or random effects (RE) methods when
longitudinal data is available.91 However, estimation complexities arise when only repeated cross-
89
A similar study has been done by Cruces, Fields & Violaz (2013) which examined the usefulness of AM, BGK and DLLM 1 methods
in estimating various indices of mobility using Chilean panel data. This chapter extends Cruces et al. (2013) in several ways. First, in
addition to the three methods that they examined, I also included DLLM 2 in the analysis. Second, I used a wider set of income mobility
indices. Third, I explained in greater detail how these pseudo-panel estimation approaches can be extended to construct synthetic micro-
based income data.
90
Another strand of literature in pseudo-panel estimation is based on Age-Period-Cohort (APC) models. The APC models
are generally used for assessing the impact of age, period and cohort for demographic and epidemiological variables such
as disease incidence or mortality rates. Since the APC model is not used for poverty estimation, I do not discuss it here.
Interested readers may refer to Yang, Fu & Land (2004) and Mason and Wolfinger (2002).
91 Even in the presence of genuine panel data, ordinary least squares (OLS) estimation of (9.1) will produce inconsistent estimates for
β if fi is correlated with Xit . On the other hand, even if fi can be reasonably assumed to be orthogonal with Xit, OLS estimators will
still be non-optimal due to the serial correlation induced by the term fi. In lieu of this, a fixed effects (FE) (or a random effects in the
220
sectional survey (RCS) data is available because in such case, Yit and Xit are not fully observed. As
an alternative to (9.1), one can consider the following model,
Yi (t )t  X i (t )t  f i (t )   i (t )t
i(t) = 1, 2,…,Nt, T= 1, 2, …, T (9.2)
Notice the change in subscripts used to denote the sampled units when comparing (9.1) and
(9.2). The subscript i(t) corresponds to the ith respondent at the tth cross-sectional survey wave while
the subscript t correspond to the time period t. Conventional RCS designs imply that i(τ)t ≠ i(Ψ)t for
every pair (i(τ)t, i(Ψ)t) in {1,2,…,Nt}, t = {τ, Ψ}. In other words, the ith sampled unit at time period τ
is not necessarily the same with the ith unit at time period Ψ. If we simply use the pooled RCS data
and proceed to ordinary least squares (OLS) estimation of the model, the parameter estimates will be
inconsistent if fi(t) is correlated with Xi(t)t. One way of addressing this issue is to find suitable
instruments for Xi(t)t, i.e., variables that are correlated with Xi(t)t but are asymptotically uncorrelated
with the unobserved terms of (1). In a seminal work, Deaton (1985) proposed an approach which uses
cohort-averaging as an indirect form of instrumentation. In particular, Deaton (1985) proposed the
following model,
Y ct   X ct  f ct   ct c = 1, 2,…, C , T = 1, 2,…, T (9.3)

where
nct nct nct nct
1 1 1 1
Y ct 
nct
Yit , X ct 
i 1 nct
 X it , f ct 
i 1 nct

i 1
f i ,  ct 
nct

i 1
it
This approach groups the sampled units into C mutually exclusive classes such that each class
is always represented from every cross-sectional survey round and that class membership is fixed
over time. Since we do not observe the same set of units in RCS, the term fct is not fixed over time.92
Thus, we cannot readily rely on conventional panel data estimation techniques such as the FE
estimator to difference out this term. However, when nc is sufficiently large (in proportion to Nc, the
number of individuals in the population who are in the cth class), Deaton (1985) argued that we can
conveniently assume that the term fct will be constant. In such case, it will be straightforward to
remove this term using data transformation. Moreover, Deaton (1985) introduced further adjustments
to the conventional FE estimator to take into account that Yct and Xct are error-ridden estimators of
their population counterparts. Furthermore, Verbeek & Nijman (1993) proposed a general class of
estimators that can be considered when estimating (2). This class of estimators employs a “within
transformation” on the pseudo-panel and adjusts the moment matrices in the least squares to account
for measurement error due to data aggregation. Verbeek & Nijman (1993) also improved Deaton’s
case that fi is uncorrelated with Xit) estimator can be considered. The FE estimator implements a data transformation that removes the
correlation between the explanatory variables X it with the unobserved terms of (9.1).
92 Note that at the population-level, this term is fixed under the assumption that the population is closed.
221
estimator after showing that the latter performed poorly in terms of the mean squared error when the
cohort sample size is small.
The static model described in (9.1) can be extended to include a lagged term of the dependent
variable (9.4). There are two estimation issues for (4). First, the term Yi(t)t-1 is unobserved when using
RCS dara. Second, fi(t) is likely to be correlated with both Yi(t)t-1 and Xi(t)t prompting the need to find
suitable instruments.
Yi (t )t  Yi (t )t 1  xi (t )t  f i (t )   i (t )t
(9.4)
To be able to estimate (4), Moffitt (1993) ignored the term fi(t) and proposed using an instrument
for Yi(t)t-1 in the form of (9.5) where Wi(t)t-1 is vector of exogenous variables whose historical time-
series are provided in the data and Zi(t-1) is a vector of time-invariant exogenous variables. In other
words, Moffitt’s (1993) idea is to first estimate a static model and use the predicted values as
instrument for Yi(t)t-1.
  
yi (t )t 1  1Wi (t )t 1   2 Z i (t 1)
(9.5)
Moffitt’s approach is anchored on a strong data requirement that the historical time-series of
Xi(t)t is observed. This is hardly satisfied in most of the existing RCS designs. In turn, Collado (1997)
improved Moffitt’s (1993) approach by using less stringent data requirements and going back to the
conventional cohort-based approach. In particular, Collado (1997) proposed a Generalized Method
of Moments (GMM) estimator corrected for measurement error for the following model,
yc(t )t  yc(t 1)t 1  xc(t )t  f c(t )   c(t )t

(9.6)
Unlike Moffitt (1993), Collado (1997) did not assume that the unit-specific effects cancel out
at the cohort-level. The author also argued that in the cohort-based approach, there is a trade-off
between the number of cohort groups and the sample size per group. In particular, when the number
of sampled units per cohort becomes large, the issue on measurement error becomes less relevant.
However, this may have potential costs on efficiency since in finite sample sizes, increasing the
number of units per cohort calls for fewer cohort groups to be formed.
Like Moffitt (1993), Girma (2000) departed from the conventional cohort-based approach and
argued that a unit-based estimation method (i.e., maintaining the original observations as the units of
analysis) will better optimize the use of available information from repeated cross-sectional data.
Although he still grouped the units into cohorts, he did not involve transforming the data to cohort
averages. Instead, he argued that different units within the same cohort (even across different time
periods) exhibit non-zero correlations. In turn, such information can be used to find a suitable
instrument when estimating (9.4). In particular, he proposed a pairwise quasi-differencing approach
for the following model,
222
yi (t )t  y j (t 1)t 1  xi (t )t  f i (t )  i (t )t
(9.7)
where i and j are units from the same cohort group. Implicitly, (9.7) suggests that any past and present
value of y and x can be used as instruments. Without imposing other conditions, this would create an
infinite number of candidate instruments. However, subsequent studies argued that relying on
arbitrarily chosen units from the same cohort as instruments could be a noisy approximation of the
unobserved value of yi(t)t-1 which might lead to inaccurate estimation of (9.4) (Verbeek & Vella 2005).
McKenzie (2004) extended (9.4) to allow for different covariate effects across cohorts. This
heterogeneous dynamic pseudo-panel model can be denoted by (9.8) and its corresponding cohort-
level model is denoted by (9.10). The author also argued that a GMM estimator similar to the one
adopted by Collado (1997) which is consistent as the number of cohort groups increases, may not
work since the number of parameters to be estimated also increases with the former. Instead
McKenzie (2004) used an approach analogous to the Arellano-Bond estimator typically used in
genuine panel models wherein 𝑦̅𝑐(𝑡−2)𝑡−2 is used as an instrument for 𝑦̅𝑐(𝑡− )𝑡−2 which in turn as
unbiased estimator of 𝑦̅𝑐(𝑡)𝑡−2 . Although this instrumentation approach addresses the bias arising from
the measurement error induced by not observing the same individuals for each time period, the author
pointed out that this estimator may be less efficient relative to the OLS estimator. In other words, an
OLS estimator may still be superior (with lower variability) unless the number of time periods and
the cohort sample sizes are both large.
yi (t )t   c yi (t )t 1   c xi (t )t  f i (t )   i (t )t
(9.8)
yc(t )t   c yc(t )t 1   c xc(t )t  f c(t )   c (t )t

(9.9)
Inoue (2008) further extended the discussion of pseudo-panel estimation of dynamic models by
considering a model that contains time-invariant unit-specific and (cohort) group-specific fixed
effects denoted by (10) where Zc(t) are cohort-level explanatory variables and δc is the time-invariant
group specific effect. Inoue (2008) proposed a GMM-based estimator for (9.10) which is consistent
under some stringent orthogonality and rank conditions.
yc(t )t   c yc(t )t 1  xc(t )t  Z c(t )  f c (t )   c   c(t )t

(9.10)
In summary, there are two broad types of pseudo-techniques that have been proposed in the
literature. The first type or what I refer to as Type I method in the succeeding discussions, uses cohort-
averages as a form of instrumentation. In particular, all sampled units are grouped into mutually
exclusive and exhaustive cohort classes. The cohort averages of the characteristics of interest are then
used as the analytical units. In this context, the cohort averages act as the pseudo-panels. The
approaches proposed by Deaton (1985), Verbeek & Nijman (1993), Collado (1997), McKenzie
(2004) and Inoue (2008) can be considered as Type I methods. On the other hand, what I refer to as
223
Type II methods maintain the original sampling units as the analytical units. Following this definition,
the approaches developed by Moffitt (1993) and Girma (2000) can be considered as Type II methods.
Overall, each type has its own advantages and limitations. For instance, the main advantage of Type
I method is that its underlying statistical theory, particularly how the model parameters can be
estimated consistently, has been discussed extensively in the literature (Verbeek 2008). However,
aggregating the units into cohorts can lead to significant loss of information. In particular, it becomes
less straightforward to examine variations of the characteristics of interest within cohorts. On the
other hand, Type II method addresses this limitation as it maintains the original observations as the
units of analysis. However, unlike the Type I methods, the underlying statistical theory of Type II
methods has not been extensively discussed in the literature. Over the years, both methods have been
used in different empirical applications. In the next section, I discuss four recently proposed pseudo-
panel techniques falling under either Type I or Type II method, that are specifically designed to
measure income mobility.
9.2.2 Estimation of Income Mobility using Pseudo-Panel Data

Antman & McKenzie’s (AM) Approach
As pointed out in Chapter 1, one of the ways of viewing mobility is to conceptualize it as the
temporal dependence between previous and current income. There are two ways of measuring
temporal dependence. First, we can estimate the correlation between previous and current income.
Subtracting this correlation from unity yields the Hart’s index described in Chapter 1. Alternatively,
we can use income elasticity which entails expressing each unit’s current income Yi(t) as a function of
its lagged income Yi(t)t-1, a vector of socio-demographic characteristics Xi(t)t and a unit-specific effect
fi(t). This is the approach followed by Antman & McKenzie (2005 and 2007) in measuring income
mobility in Mexico. Equation (9.11) shows the underlying statistical model. Here, α is the mobility
parameter of interest. In general, a large absolute value for α portrays strong temporal dependence,
i.e., low mobility, while small values mirror weak relationship between previous and current incomes,
i.e., high mobility.
yc(t )t  yc(t 1)t 1  xc (t )t  f c(t )t  c(t )t

(9.11)
where
λ𝑐(𝑡)𝑡 = 𝛼[𝑌̅𝑐(𝑡)𝑡− 𝑌̅𝑐(𝑡− )𝑡− ]
McKenzie (2004) argues that the term λc(t)t can be ignored when the number of sampled units for
every cohort is sufficiently large. Noticeably, (9.11) is an extension of (9.6). Thus, as can be inferred
from the discussions in the previous section, consistent estimation of the parameters of (9.11) depends
224
on the assumptions about the unobserved unit-specific effect as well as the sample size for each
cohort.93 Following Antman & McKenzie (2005 and 2007), a number of studies have applied this
approach in estimating the temporal dependence-based concept of income mobility (Calonico 2006;
Navarro 2006; Cuesta et al. 2011).
Bourguignon, Goh & Kim’s (BGK) Approach

Bourguignon, Goh & Kim (2004) proposed a method of estimating the probability of falling
into poverty using the following model,
Yi (ct )t   tc X ic(t )t   ic(t )t
 ic(t )t   c ic(t )t 1  eic(t )t (9.12)
𝑐 𝑐
where 𝑌𝑖𝑡(𝑡) is the income of unit i from (cohort) group c at time t, 𝑋𝑖𝑡(𝑡) is a vector of explanatory
𝑐
variables, 𝛽𝑡𝑐 is the corresponding vector of covariate effects and 𝜀𝑖(𝑡)𝑡 is and AR(1) error term such
𝑐 𝑐 𝑐
that 𝑉(𝑒𝑖(𝑡)𝑡 ) = 𝜎 2𝑐𝑡 . With RCS data, 𝜀𝑖(𝑡)𝑡 and 𝜀𝑖(𝑡)𝑡− are not observed simultaneously. Nevertheless,
the authors argued that the parameters of (9.12) can be estimated using RCS data using the variance
of the residuals as shown in (9.13).
 2ct  (  c ) 2V ( ic(t )t 1 )   ect
2
(9.13)
In particular, for each group c and time t, (9.12) can be estimated using OLS. The variance of the
resulting residuals from (9.12) can then be used to estimate (9.13) while the residuals of (9.13) can
as estimates of 𝜎 2𝑐𝑡 . Given these parameter estimates and under the assumption that
be used
𝑐
𝑒𝑖(𝑡)𝑡 ~ (0, 𝜎 2𝑐𝑡 ), then the probability of falling into poverty at time t+1 is given by
c c
z   2 ˆ c ˆc
t 1 xi ( t ) t 1    i ( t ) t
P(Y  z | x , xˆ
c
i ( t ) t 1 , ˆ ,  ect 1 )  (
c c
i ( t ) t 1
c
t 1 )
i (t )t
ˆ ect 2
1
(9.14)
While the estimation methodology is quite straightforward to implement, there are several issues with
this approach. First, estimating heterogeneous models with varying parameters across cohort groups
reduces the effective sample size. If some cohort groups comprise only few observations, then the
corresponding parameter estimates might not be reliable. Second, to be able to estimate the
probability of falling into poverty in the future, the formula calls for the availability of estimates for
𝑐 𝑐
𝛽𝑡+ , 𝑋𝑖𝑡+ and 𝜎 2𝑐𝑡+ . In the absence of this information, a simple approach is to assume that these
parameters and variables are time-invariant throughout the observation period.
93
In the next section, I adopt the simplifying assumption that conditional on previous income, there is no persistent unit-specific effect.
Hence, the model parameters can be estimated using OLS.
225
Dang, Lanjouw, Luoto & McKenzie’s (DLLM 1) Approach
Like Bourguignon, Goh & Kim, Dang, Lanjouw, Luoto & McKenzie (2011) focused on
measuring dynamics in the low income range. In particular, consider the following models,
Yi (1)1  1 X i (1)1   i (1)1 (9.15)
Yi ( 2) 2   2 X i ( 2) 2   i ( 2) 2 (9.16)
where Y is individual (or household) income and X is a vector of individual (or household)
characteristics whose values are fixed over time. These models can be estimated using two waves of
RCS. However, to be able to estimate indicators of poverty dynamics, we need either Yi(1)2 or Yi(2)1,
both of which are unobserved in RCS data. Thus, the main idea behind the DLLM approach is to
impute the values of Yi(2)1 or Yi(1)2 using the information provided in (15) and (16). Without loss of
generality, I will focus on the imputation of Yi(2)1.Following the approach initially developed by
Elbers, Lanjouw and Lanjouw (2003) for small area estimation of poverty, DLLM (2011) proposed
the “out-of-sample” imputation formula depicted in (9.17) which assumes that the explanatory
variables are constant over time (i.e., Xi(t)t = Xi(t)t+1).

Yi ( 2)1  ˆ1 X i ( 2)1  v~i ( 2)1
 ˆ1 X i ( 2) 2  v~i ( 2)1 (9.17)
In addition to β1 and Xi(2)2, (17) calls for an estimate of the error term vi(2)1. To do this, DLLM (2011)
first assumed that (vi(2)1 and vi(2)2) ~ BVN(0, ∑ϑ) such that94
𝜎𝜗2 𝜌𝜎𝜗 𝜎𝜗2
∑ϑ = [ 2 ] (9.18)
𝜌𝜎𝜗 𝜎𝜗2 𝜎𝜗2
The parameters 𝜎̂𝜗2 and 𝜎̂𝜗2

2
can be estimated from (9.15) and (9.16). On the other hand, the authors
adopted a naïve approximation for ρ by assuming that it is either equal to zero or one. This produces
lower and upper bounds for different indicators of poverty dynamics (DLLM 2011) as shown below.
𝑧−𝛽1 𝑋𝑖(2)2 𝑧−𝛽2 𝑋𝑖(2)2 𝑧−𝛽1 𝑋𝑖(2)2 𝑧−𝛽2 𝑋𝑖(2)2

∅(
𝜎𝜗1
,
𝜎𝜗2
| 𝜌 = 0) ≤ P(𝑌̂i(2)1< z, Yi(2)2< z) ≤ ∅( 𝜎𝜗1
,
𝜎𝜗2
|𝜌= ) (9.19)

∅(
𝜎𝜗1
,
𝜎𝜗2
|𝜌= ) ≤ P(𝑌̂i(2)1< z, Yi(2)2> z) ≤ ∅( 𝜎𝜗1
,
𝜎𝜗2
| 𝜌 = 0) (9.20)
∅(
𝜎𝜗1
,
𝜎𝜗2
|𝜌 = ) ≤ P(𝑌̂i(2)1> z, Yi(2)2< z) ≤ ∅( 𝜎𝜗1
,
𝜎𝜗2
| 𝜌 = 0) (9.21)
∅(
𝜎𝜗1
,
𝜎𝜗2
| 𝜌 = 0) ≤ P(𝑌̂i(2)1> z, Yi(2)2> z) ≤ ∅( 𝜎𝜗1
,
𝜎𝜗2
|𝜌 = ) (9.22)
94
DLLM (2011) also proposed an analogous non-parametric approach.
226
Dang, Lanjouw, Luoto & McKenzie’s (DLLM 2) Approach
The DLLM 1 approach which entails constructing lower and upper bounds using ρ = 0 and ρ =
1 for poverty dynamics is intuitive. A value of zero for ρ implies that after accounting for correlates
of income, the temporal fluctuations in each unit’s income are independent. This (temporal)
independence is expected to induce more mobility, i.e., more movements into or out of poverty. On
the other hand, if ρ is equal to one, then there is perfect inertia in the temporal fluctuations in one’s
income which is expected to minimize mobility, i.e., less movements into or out of poverty.
The good thing about DLLM 1 approach is that it maintains the original observation as the unit
of analysis, making it straightforward to estimate the bounds depicted in (9.19) to (9.22) for different
sub-population groups which in turn, enriches the analysis.95 In addition, since this approach allows
for heterogeneous covariate effects wherein the parameters β1 and β2 are estimated separately, it may
be able to capture structural changes in the income distribution96. However, one of the obvious
limitations of DLLM 1 method is that it does not provide point estimates for the different poverty
indicators. In addition, the width of the bounds depend on how much income variation can be
attributed to differences in time-invariant individual or household characteristics. As the model fit
improves (higher R2), the bounds become narrower (DLLM 2011). However, R2 is high when the
underlying income distribution regime is rigid wherein differences in time-invariant characteristics
are the primary determinants of income variation. In other words, when much of the income variations
arise from factors other than these fixed characteristics, the DLLM 1 approach may not provide
optimal estimates of poverty dynamics. To address this issue, Dang et al. (2014) proposed a point
estimator for ρ which they derived as follows
𝑜𝑣(𝑌𝑖(2) , 𝑌𝑖(2)2 )
𝜌(𝑌𝑖(2) , 𝑌𝑖(2)2 ) =
√𝑉(𝑌𝑖(2) )𝑉((𝑌𝑖(2)2 )
𝑜𝑣(𝛽 ′ 𝑋𝑖(2) + 𝜗𝑖(2) , 𝛽2′ 𝑋𝑖(2)2 + 𝜗𝑖(2)2 )

𝜌(𝑌𝑖(2) , 𝑌𝑖(2)2 ) =
√𝑉(𝑌𝑖(2) )𝑉((𝑌𝑖(2)2 )
𝛽 ′ 𝑉(𝑋𝑖(2) )𝛽2 + 𝜌√𝜎𝜗21 𝜎𝜗22

=
√𝑉(𝑌𝑖(2) )𝑉((𝑌𝑖(2)2 )
ρ(𝑌𝑖(2) , 𝑌𝑖(2)2 )√V(Yi )V(Yi2 ) β′ V(X i )β2

𝜌=
σ𝜗1 𝜎𝜗2
95
Intuitively, this is subject to the sample size available for the sub-population group under consideration.
96
For instance, if the level of importance that a (social) opportunity structure attributes to fixed individual characteristics
like race, ethnicity, religion and parental education changes significantly over time, then such changes are implicitly
incorporated in the estimation of the model parameters.
227
̂ ′1 V(Xi )β
̂y̅c1 y̅c2 √V(Yi1 )V(Yi2 )−β
ρ ̂2
ρ̂ = ̂ ϑ2
̂ ϑ1 σ
(9.23)
σ
As shown in (9.23), the point estimator is a function of the parameter estimates of (9.15) and (9.16),
variance of the observed incomes as well as the variance of the cohort means of the observed incomes.
To arrive at this formula, the authors used the correlation between the two sets of cohort averages as
a rough approximation of 𝜌(𝑌𝑖(2) , 𝑌𝑖(2)2 ). Furthermore, it is straightforward to show that when β̂′ = β̂2 ,
ρ̂ can be re-expressed as a function of the correlation between the cohort means and the coefficient of
determination of each cross-sectional model. In addition to (9.24), Dang et al. (2014) also provided
more informative lower and upper bounds for 𝜌 as shown in (9.25) and (9.26). In turn, formula
analogous to (9.19) to (9.22) can be derived using the estimated values of 𝜌.
2 2
̂y
ρ ̅ c2 − √𝑅1 𝑅2
̅ c1 y
𝜌̂ = (9.24)
√( −𝑅12 )( −𝑅22 )
𝜌̂𝐿𝐵 = ρ
̂̅ ̅
yc yc2 (9.25)
′
̂ V(X )β
̂
β i 2
𝜌̂𝑈𝐵 = (9.26)
√V(Yi )V(Yi2 )
9.2.3 Extending Pseudo-Panel Methods to Measure Broad Class of Income Mobility

Measures
Can the pseudo-panel estimation methods discussed in the previous section be used to measure
a broader class of mobility indicators other than those for which these methods are originally designed
to measure? A previous study by Cruces et al. (2013) asked the same question and found that the AM,
BGK and DLLM 1 did not perform well in capturing the income mobility patterns in the Chilean
panel data. Whether this provides conclusive proof undermining the usefulness of pseudo-panel
methods in estimating income mobility merits further investigation. For one, the Chilean panel data
is not representative of the population of data sets to which the pseudo-panel techniques can be
applied. In other words, this panel data could have specific characteristics that make pseudo-panel
techniques less attractive. To further investigate this issue, I completed a similar exercise using the
panel data from the Philippines.
As explained earlier, the AM, BGK, DLLM 1 and DLLM 2 methods are designed to answer
different income mobility-related questions. In particular, the AM approach answers the question,
“up to what extent can previous income predict current income?” On the other hand, the BGK
approach is designed to answer the question, “what is the risk of falling into poverty in the future,”
while both DLLM 1 and DLMM 2 approaches answer “what is the probability of staying, moving
into or moving out of poverty?” (Cruces et al. 2013). Similar to Cruces et al. (2013), the objective of
this study is to examine the feasibility of using these approaches in estimating a wider array of income
228
mobility measures. To do this, it is essential to first construct a micro-based pseudo-panel data of
incomes. For simplicity, suppose we have two cross-sections denoted by {Yi(1)1} and {Yi(2)2}. The
main task is to provide imputed values for either {Yi(1)2} or {Yi(2)1} that can be used to estimate any
mobility measure denoted by M(Yi(1)1, Yi(1)2) (or M(Yi(2)1, Yi(2)2)). This section provides the step-by-
step procedures of extending the algorithms of AM, BGK, DLLM 1 and DLLM 2 to be able to
estimate other mobility indices which these techniques were not originally designed for.
AM Approach
Step 1: For each time period t = 1, 2, group all sampled units into different cohort groups.97
Step 2: Compute the average income of each cohort. Do the same for other characteristics of interest.
Step 3: Estimate the model yc (t )t  yc (t 1)t 1   c (t )t using OLS (Note: This model can be expanded to
include other control factors).
Step 4: Compute the variance of the residuals V(𝜀̂𝑐(𝑡)𝑡 ).
Step 5: Estimate 𝑌̂𝑖( )2 = 𝛼̂𝑌𝑖( ) + 𝜀̃𝑖( )2 where 𝜀̃𝑖( )2 is a randomly drawn data point from N(0,
Var(𝜀̂𝑐(𝑡)𝑡 )).
Step 6: Estimate the mobility measure M(𝑌𝑖( ) , 𝑌̂𝑖( )2 ).
Step 7: Repeats Steps 5 and 6 for R times.
Step 8: Take the average and standard deviation of M(𝑌𝑖( ) , 𝑌̂𝑖( )2 ) across all iterations.
BGK Approach
Step 1: For each time period t = 1, 2, 3, group all sampled units into different cohort groups.
Step 2: For each cohort group c, estimate 𝑌𝑖(𝑐 ) = 𝛽𝑡𝑐 𝑋𝑖(𝑐 ) + 𝜀𝑖(𝑐 ) , 𝑌𝑖(2)2 𝑐 𝑐
= 𝛽𝑡𝑐 𝑋𝑖(2)2 𝑐
+ 𝜀𝑖(2)2 and
𝑐 𝑐 𝑐 𝑐
𝑌𝑖(3)3 = 𝛽𝑡 𝑋𝑖(3)3 + 𝜀𝑖(3)3.
Step 3: Retrieve the residuals 𝜀̂𝑖(𝑐 ) , 𝜀̂𝑖(2)2
𝑐 𝑐
, 𝜀̂𝑖(3)3 from the models estimated in Step 2. Compute their
2 2 2
respective variances 𝜎̂𝜀𝑐1 , 𝜎̂𝜀𝑐2 , 𝜎̂𝜀𝑐3 .
Step 4: For each cohort c, estimate  2ct  (  c ) 2 V ( ic(t )t 1 )   ect
2
.
Step 5: From the model in Step 4, retrieve the residuals 𝜎̂ 2𝑐𝑡 .
Step 6: Estimate 𝑌𝑖(𝑐 )2 = 𝛽̂2𝑐 𝑋𝑖(𝑐 ) + 𝜌̂𝑐 𝜀̂𝑖(𝑐 ) + 𝑒̃𝑖(𝑡)𝑡
𝑐 𝑐
where 𝑒̃𝑖(𝑡)𝑡 is a randomly drawn data point from
N(0, 𝜎̂ 2𝑐𝑡 )98.
Step 7: Estimate the mobility measure M(𝑌𝑖( ) , 𝑌̂𝑖( )2 ).
Step 8: Repeats Steps 6 and 7 for R times.
Step 9: Take the average and standard deviation of M(𝑌𝑖( ) , 𝑌̂𝑖( )2 ) across all iterations.
DLLM 1 Approach
Step 1: For each time period t, estimate 𝑌𝑖(𝑡)𝑡 = 𝛽𝑡 𝑋𝑖(𝑡)𝑡 + 𝜗𝑖(𝑡)𝑡 . Retrieve the parameter estimates 𝛽̂𝑡 ,
and the residuals 𝜀̂𝑖(𝑡)𝑡 .
Step 2: Compute the mean and the variance of the residuals, 𝜇̂ 𝜗𝑡 and 𝜎̂𝜗2𝑡 .
Step 3: Set the residual correlation 𝜌̂𝑗 , j {LB, UB}, such that 𝜌̂𝐿𝐵 = 0 and 𝜌̂𝑈𝐵 = .
Step 4: Sort the residuals 𝜀̂𝑖(2)2 from lowest to highest.
97 As discussed earlier, cohort grouping should be mutually exclusive and exhaustive. In addition, cohort membership should be fixed
over time. Furthermore, it is ideal to strike a balance between the number of cohorts and the sample size per cohort. A common
approach used in the literature is to form cohorts based on gender and year of birth.
98 This assumes that 𝑋 𝑐 𝑐
𝑖( ) = 𝑋𝑖( )2
229
̂ 𝜗 ) where
Step 5: Draw n2 pairs of residuals (𝜀̃𝑖(2) , 𝜀̃𝑖(2)2 ) from BVN(0, ∑
𝜎̂𝜗2 𝜌̂𝑗 𝜎̂𝜗 𝜎̂𝜗2
̂ 𝜗= [
∑ ]
2
𝜌̂𝑗 𝜎̂𝜗 𝜎̂𝜗2 𝜎̂𝜗2
Rank the residual pairs (𝜀̃𝑖(2) , 𝜀̃𝑖(2)2 ) in ascending order according to the values of 𝜀̃𝑖(2)2 .
𝑗
Step 6: Pair the first element 𝜀̃𝑖(2) of each sorted residual pair (𝜀̃𝑖(2) , 𝜀̃𝑖(2)2 ) with the sorted 𝜀̂𝑖(2) .
𝑗 𝑗
Step 7: For each j {Est, LB, UB}, estimate 𝑌̂𝑖(2) = 𝛽̂ 𝑋𝑖(2)2 + 𝜀̃𝑖(2) .
𝑗
Step 8: Estimate the mobility measure Mj(𝑌̂𝑖(2) , 𝑌𝑖(2)2 ).
Step 9: Repeats Steps 5 to 8 for R times.
𝑗
Step 10: For each j {LB, UB}, take the average and standard deviation of Mj(𝑌̂𝑖(2) , 𝑌𝑖(2)2 ) across all
iterations.
DLLM 2 Approach
Step 1: For each time period t = 1, 2, group all sampled units into different cohort groups. Compute
the correlation of 𝑌̅𝑐( ) and 𝑌̅𝑐(2)2 and denote it by ρ̂y̅c1y̅c2 .
Step 2: Compute the variances V(Yi(1)1) and V(Yi(2)2).
Step 3: For each time period t, estimate 𝑌𝑖(𝑡)𝑡 = 𝛽𝑡 𝑋𝑖(𝑡)𝑡 + 𝜗𝑖(𝑡)𝑡 . Retrieve the parameter estimates 𝛽̂𝑡 ,
residuals 𝜀̂𝑖(𝑡)𝑡 , and the coefficients of determination 𝑅𝑡2 .
Step 4: Compute the mean and the variance of the residuals, 𝜇̂ 𝜗𝑡 and 𝜎̂𝜗2𝑡 .
Step 5: Compute the residual correlation 𝜌̂ 𝑠𝑡 , 𝜌̂𝐿𝐵 , and 𝜌̂𝑈𝐵 .
′
ρ
̂̅ (Yi )V(Yi2 )
√V ̂ V(X )β
β ̂
y i 2
c yc2
̅
𝜌̂ =
𝑠𝑡
σ
̂ϑ σ
̂ ϑ2
𝜌̂𝐿𝐵 = ρ
̂̅ ̅
yc yc2
′
̂ V(X )β
β ̂
i 2
𝜌̂𝑈𝐵 =
√V(Yi )V(Yi2 )
Step 6: Rank the residuals 𝜀̂𝑖(2)2 from lowest to highest.

̂ 𝜗 ) where
Step 7: Draw n2 pairs of residuals (𝜀̃𝑖(2) , 𝜀̃𝑖(2)2 ) from BVN(0, ∑
𝜎̂𝜗2 𝜌̂𝑗 𝜎̂𝜗 𝜎̂𝜗2
̂ 𝜗= [
∑ ]
2
𝜌̂𝑗 𝜎̂𝜗 𝜎̂𝜗2 𝜎̂𝜗2
Rank the residual pairs (𝜀̃𝑖(2) , 𝜀̃𝑖(2)2 ) in ascending order according to the values of 𝜀̃𝑖(2)2 .
𝑗
Step 8: Pair the first element 𝜀̃𝑖(2) of each sorted residual pair (𝜀̃𝑖(2) , 𝜀̃𝑖(2)2 ) with the sorted 𝜀̂𝑖(2) .
𝑗 𝑗
Step 9: For each j {Est, LB, UB}, estimate 𝑌̂𝑖(2) = 𝛽̂ 𝑋𝑖(2)2 + 𝜀̃𝑖(2) .
𝑗
Step 10: Estimate the mobility measure Mj(𝑌̂𝑖(2) , 𝑌𝑖(2)2 ).
Step 11: Repeats Steps 7 to 10 for R times.
𝑗
Step 12: For each j {Est, LB, UB}, take the average and standard deviation of Mj(𝑌̂𝑖(2) , 𝑌𝑖(2)2 ) across
all iterations.
9.3 Data
Consistent with the existing literature on pseudo-panel data, the final sample is restricted to
households whose heads were born between 1933 and 1978, or equivalently, those who were aged 25
to 70 in 2003. The reason for doing this is to be able to have a relatively in-scope population. If I
230
included younger or older households, I would be dealing with households that can change
dramatically in the succeeding time periods. It would then be difficult to identify which population is
being represented by the results. From the panel sample consisting of 6519 households, I drew
independently a smaller (50%) random subsample for each wave, to create cross-sectional data to be
used in pseudo-panel estimation. I decided not to use the full cross-sectional data of FIES so that I
can treat the estimates based from the panel subsample as the actual values of the target “population”
parameters.
Following the convention in the previous chapters, the household expenditure per capita is used
as the income measure. For the AM approach, the cohorts are constructed by grouping households
according to the heads’ year of birth and gender. In particular, I followed the approach of Cruces,
Fields & Viollaz (2013) which uses two-year span to be able to strike a balance between the number
of cohorts and sample size per cohort. In estimating the conditional models, I included the cohort’s
average family size, average proportion of household members less than 15 years old, average
proportion of household members who are employed and proportion of households relying on
agricultural income. For the BGK, DLLM 1 and DLLM 2 approaches, the income correlates included
in the model are provincial dummies, household head’s age and its square, gender of the household
head and educational attainment of the household head. Furthermore, estimates of poverty dynamics
are based on US$2/day poverty line.
9.4 Discussion of Empirical Results

Following the procedures outlined in Section 9.2.3, I estimated four measures of poverty
dynamics which include the proportion of population moving into poverty, moving out of poverty,
staying in poverty and staying non-poor. In addition, I also estimated seven of the most commonly
used income mobility indices as described in Chapter 1. They cover three different mobility concepts
– movement (average rank jump, Fields-Ok’s, King’s indices), temporal dependence (Hart’s index)
and equalizer of income (CDW, Fields’ and Shorrocks’ indices). Before proceeding to the discussion
of the results, some remarks are in order. First, when computing income mobility from time t to t+1,
I always chose a reference period. For the chosen reference period, the actual income data from FIES
is used. On the other hand, the income values for the other time period are imputed following the
approach outlined in Section 9.2.3. In general, the choice of reference period is different for each
method. For the AM and BGK approaches, the initial time period is always chosen as the reference
period while the income values for the final time period are imputed. In contrast, the DLLM 1 and
DLLM 2 approaches use the final time period as the reference while the income values for the initial
time period are imputed. Given that the final sample for each cross-sectional wave is representative
of the same population, I suspect that the differences in the estimates will not depend on the choice
231
of reference period. Second, unlike the original AM, BGK, DLLM 1 and DLLM 2 methods which
strictly use either parametric or non-parametric procedures, I adopt a semi-parametric approach. As
outlined in Section 9.2.3, the semi-parametric algorithm for estimating income mobility entails
iteratively drawing random disturbance terms from the Gaussian distribution with pre-specified
parameters. For each mobility index, every iteration corresponds to an estimated value. The results
provided in the succeeding discussions are based on 100 replicates. Third, I derived a point estimate
for the DLLM 1 approach by taking the midpoint of the lower and upper bounds. Fourth, to be able
to fine-tune the estimates for the DLLM 1 and DLLM 2 approaches, I introduce a structure preserving
technique that takes into account the rank of the residuals from the reference period. Lastly, there are
three components that contribute to the estimated standard error of the mobility estimates - sampling
error, model error and the iterative sampling procedure for the stochastic disturbance term.
Dynamics in the Low Income Range

As noted in Chapter 5, the gross outflow from US$2/day poverty from 2003 to 2009 is
approximately 10% of the population while the gross inflow accounts for 9% (Table 9.4). The
proportions of the population who remained poor and non-poor during these two periods are 34% and
47%, respectively. Tables 9.1 to 9.3 compare the estimated proportion of each category of poverty
status across the different pseudo-panel methods. The performance of the AM approach is not
consistent. For some years, the pseudo-panel estimates are reasonably close to the estimates based on
actual panel data but there are cases when both the unconditional and conditional pseudo-panel
estimates are quite different from the proportions estimated from the actual panel data. Interestingly,
the unconditional estimates for the poverty outflow and inflow are generally lower than the actual
panel data-based estimates while the proportion of persistent poverty and non-poverty are consistently
higher. A slightly different pattern emerges when we look at the conditional estimates of the AM
approach. In particular, pseudo-panel estimates of poverty outflow and persistence of non-poverty
are consistently higher while poverty inflow and poverty persistence are consistently lower than the
panel estimates. On the other hand, the performance of the BGK approach yields mixed results. For
instance, while the pseudo-panel estimates for poverty inflow and persistence of non-poverty are quite
close to the estimates derived from actual panel data, the pseudo-panel estimates for poverty outflow
and persistence of poverty are quite disparate. Compared to the AM and BGK approaches, the DLLM
1 and DLLM 2 methods performed better. For instance, the estimates using the actual panel data fall
in between the estimated lower and upper bounds produced by DLLM 1. While these bounds could
be restrictively wide as pointed out by Cruces et al. (2013), taking its midpoint yields estimates that
are quite close to the actual panel-based estimates. Furthermore, the bounds estimated using the
DLLM 2 method are much narrower compared to that of the DLLM 1. This is expected given that
232
the DLLM 2 approach uses more informative estimates of the residual correlation. The point estimates
using the DLLM 2 approach are also reasonably close to the actual panel-based figures. Nevertheless,
it is worth noting that the point estimates of the DLLM 2 approach are always lower for movements
into and out of poverty and always higher for poverty immobility.
Other Dimensions of Income Mobility

As pointed out in the earlier chapters, from 2003 to 2009, average per capita consumption barely
moved, changing by approximately 2%. This is also accompanied by a small reduction in income
inequality where the Gini coefficient changed from 0.44 in 2003 to 0.43 in 2009. However, turning
to a broader set of income mobility measures, I find a much more dynamic income distribution over
the six-year period, especially when viewed in terms of income movements. In particular, the mean
absolute percentage change in per capita consumption is about 36%. This is equivalent to a 14-step
change in income ranks, on the average. On the other hand, mobility is less pronounced when viewed
in terms of temporal dependence and equalizer of income. For instance, the correlation of the
logarithm of incomes in 2003 and 2009 is about 0.8. Furthermore, the observed income mobility
reduces long-run inequality by only 6%.
Tables 9.4 to 9.6 compare the estimated values of different mobility indices using the actual
panel and pseudo-panel data. For the AM approach, I find that the estimates derived from the
conditional models are closer to the actual panel-based figures for indices designed to gauge
movement and temporal dependence of incomes. In contrast, the unconditional models performed
better than the unconditional models in estimating income mobility indices under the equalizer of
income perspective. On the other hand, the pseudo-panel estimates computed using the BGK
approach are at least twice as high than the values estimated using actual panel data. Turning to the
DLLM 1 and DLLM 2 approaches, I find that the values of all mobility indices using actual panel
data fall in between the lower and upper bounds estimated using the pseudo-panel approach. Unlike
the indices of poverty dynamics for which DLLM (2011) provided a theoretical proof that the
approach produces valid lower and upper bounds, we have not done so for the mobility indices
considered in this section. Thus, this result is encouraging. In other words, it provides us reason to
believe that the DLLM 1 and DLLM 2 approaches can also be useful for estimating mobility indices
other than what they were originally designed to measure. Furthermore, comparing the midpoint of
the bounds derived using DLLM 1 with the DLLM 2’s point estimator, I find that for 2003-2006 and
2006-2009 periods, the estimated values from DLLM 1 are quite a bit closer to the movement-based
indices computed using the actual panel data. For the rest, the DLLM 2 estimates performed better.
In addition, I find that the midpoints from the DLLM 1 approach consistently overestimate mobility
233
while the DLLM 2’s point estimates are usually lower than the values of indices computed from
actual panel data.
In summary, the results of these analyses allow me to address the three research questions:
1. Are the proposed pseudo-panel techniques useful for measuring income mobility? In general,
which of the pseudo-panel techniques are most desirable?
Overall, the results using Philippine data suggest that the DLLM 1 and DLLM 2 approaches
performed reasonably well in estimating poverty dynamics and other measures of mobility. On the
other hand, the AM and BGK approaches provided satisfactory results for selected indicators only;
the proportion of population moving into poverty and proportion of population remaining non-poor
for BGK and indices under the equalizer of income perspective for AM. There are several possible
explanations for this. For instance, the pseudo-panel approaches considered in this study are not
originally designed for estimating varied income mobility measures. Rather, these methods are
proposed for specific mobility indicators only. If the structural parameters of the underlying models
are not flexible enough, it would not be surprising to note that the pseudo-panel estimators will not
always perform well for all types of income mobility measures. Of the four pseudo-panel methods
considered here, I argue that the DLLM 1 and DLLM 2 approaches use more flexible model
specifications than the AM and BGK methods.
2. Do the pseudo panel techniques’ performance depend on the type of income mobility measure
being estimated?
My empirical findings suggest that in most cases, the pseudo-panel methods performed better
when estimating indices under the mobility as income movement perspective. On the other hand,
indices measuring temporal dependence of income and its inequality-reducing effect are harder to
impute using the pseudo-panel methods considered here. A possible reason why this is the case is that
unlike the temporal dependence and equalizer of income-based measures, most of the movement-
based indices are less sensitive to the detailed features of the joint temporal distribution of incomes.
3. How can the existing pseudo-panel techniques be improved?

Compared to the findings by Cruces et al. (2013) using the Chilean panel data, the use of
pseudo-panel methods especially the DLLM 1 and DLLM 2 approaches in estimating income
mobility in the Philippines produced more encouraging results. I suspect that the differences in the
characteristic features of the income distributions of the two countries contribute to the diverging
findings about the usefulness of the pseudo-panel methods. For instance, the Chilean panel data spans
234
a ten-year period while the Philippine data covers a shorter six-year period. Given this, it is reasonable
to expect that there is more mobility in the Chilean panel data than in the Philippine data. However,
as I have initially pointed out, the use of time-invariant variables in DLLM 1 and DLLM 2 are
probably more suitable when the true income mobility regime is low because it imputes the value of
permanent income. This could potentially justify why I have noted more satisfactory results than
Cruces et al. (2013) found. Nevertheless, it is difficult to provide a conclusive explanation without
doing further studies. To be able to move forward, I recommend doing simulation studies that will
outline a more objective characterization of the performance of each pseudo-panel method considered
in this paper.
There are several areas for improvement that could be explored further. First, it would be
worthwhile to extend the BGK, DLLM 1 and DLLM 2 algorithms to allow the use of time-varying
correlates of income in the model specification. Adding time-varying correlates may significantly
improve the prediction power of the models and in turn, improve the pseudo-panel estimates of
income mobility. Second, incorporating structural preserving techniques in the existing pseudo-panel
algorithms may prove useful in estimating a wider array of income mobility indices. At present, it
appears that income mobility measures which are sensitive to the overall structure of the income
distribution are harder to impute than indices which focus on capturing movements of individual
incomes. Third, statistical inference will enrich the income mobility analysis because it will allow
comparison of income mobility across space and over time. However, much of the existing
discussions are centred on estimation of mobility indices. Thus, further research is needed to be able
to provide a theoretical framework that will serve as a practical guide for conducting statistical
inference in the context of income mobility estimation using pseudo-panel data.
9.5 Summary and Future Directions

Measures of income mobility are commonly used in socio-economics literature as analytical
tools for examining the evolution of the income distribution. In general, proponents of these measures
believe that incorporating a longitudinal perspective enriches the analysis of income distribution. In
particular, they argue that income mobility measures provide a more dynamic perspective of the
evolution of a country’s living standards than simply examining changes in cross-sectional indices of
poverty and inequality over time. Panel data that tracks the incomes of the same set of households or
individuals is the appropriate data source for measuring income mobility. However, factors like cost
and risks of attrition often lead to the use of cross-sectional data.
Unlike panel surveys, cross-sectional surveys do not necessarily follow the same set of
households or individuals. Recently, several pseudo-panel estimation methods have been proposed in
235
estimating different concepts of mobility. The development of pseudo-panel methods reconciles the
need for incorporating a longitudinal perspective when examining income distribution with the
absence or lack of panel data. In particular, it offers researchers with the opportunity to depart from
conventional methods of examining static indicators of well-being and delve deeper into the multi-
dimensional issue of equality of socio-economic opportunities using cross-sectional survey data.
There are several ways of creating synthetic or pseudo-panels out of repeated cross-sectional
data. This chapter reviewed four recent developments in the pseudo-panel estimation of income
mobility literature. The first method proposed by AM (2007) entails transforming the unit-level data
into cohort averages. These cohort averages serve as the synthetic panels. On the other hand, the
approaches proposed by BGK (2004), DLLM (2011) and DLLM (2014) estimates income models
while maintaining individuals or households as the units of analysis. Originally, these methods are
designed to answer varying concepts of mobility. In this study, I outlined algorithms which extend
these approaches to be able to measure a wider array of income mobility indices. The results suggest
that the proposed methods by DLLM (2011) and DLLM (2014) which employ the weakest
assumption about the structural parameters and functional form of the income models performed
satisfactorily in terms of estimating different mobility concepts. Nevertheless, several areas for
improvement remain. These include exploring techniques that would accommodate time-varying
correlates for the BGK, DLLM 1 and DLLM 2 income models, employing structural preserving
strategies to provide better estimates of mobility indices which are sensitive to the overall structure
of the income distribution and outlining the framework for carrying out statistical inferences.
236
Table 9.1 Estimates of Poverty Dynamics in the Philippines, 2003-2006
PSEUDO-PANEL
Indicator PANEL
AM (A) AM (B) BGK DLLM 1 (A) midpt DLLM 1 (B) DLLM 2 (A) DLLM2 (B) DLLM2 (C)
Poverty outflow 6.19 2.65 5.52 19.21 2.81 8.41 14 6.25 7.79 10.61
Poverty inflow 9.11 2.18 7.76 11.58 1.98 10.18 18.38 5.98 8.39 12.68
Stay in Poverty 35.08 39.16 33.57 22.12 25.1 33.3 41.5 30.8 35.08 37.49
Stay Non-Poor 49.62 56.02 53.15 47.09 42.52 48.12 53.72 45.92 48.73 50.27
Source: Author’s computations using household expenditure per capita data from the longitudinal subsample of FIES 2003, 2006, 2009

PSEUDO-PANEL
Indicator PANEL
AM (A) AM (B) BGK DLLM 1 (A) midpt DLLM 1 (B) DLLM 2 (A) DLLM2 (B) DLLM2 (C)
Poverty outflow 10.49 0.85 6.92 22.84 8.9 13.36 17.82 11.43 13.03 14.61
Poverty inflow 5.93 8.39 10.95 11.31 0.38 8.42 16.46 4.33 7.32 10.39
Stay in Poverty 33.7 34.82 32.26 20.36 23.69 31.73 39.77 29.76 32.83 35.82
Stay Non-Poor 49.88 55.94 49.88 45.48 42.03 46.49 50.95 45.24 46.82 48.42

PSEUDO-PANEL
Indicator PANEL
AM (A) AM (B) BGK DLLM 1 (A) midpt DLLM 1 (B) DLLM2 (A) DLLM2 (B) DLLM2 (C)
Poverty outflow 10.22 1.15 1.19 20.63 7.07 11.79 16.52 10.59 12.3 13.43
Poverty inflow 8.58 7.48 7.41 11.54 0.62 8.81 16.99 5.91 9.25 11.17
Stay in Poverty 31.05 33.85 33.92 20.7 23.16 31.34 39.53 28.97 30.89 34.24
Stay Non-Poor 50.15 57.52 57.48 47.13 43.34 48.06 52.78 46.42 47.55 49.26
237
Table 9.4 Estimates of Income Mobility in the Philippines, 2003-2006
PSEUDO-PANEL
Indicator PANEL
AM (A) AM (B) BGK DLLM1 (A) midpt DLLM1 (B) DLLM2 (A) DLLM2 (B) DLLM2 (C)
Movement
Ave. Rank Jump 11.78 3.51 10.17 23.13 4.17 13.63 23.1 8.87 11.71 16.7
Fields-Ok 31.77 9.33 28.18 126.55 11.87 36.4 60.92 24.07 31.44 44.52
King 29.87 8.68 24.85 61.91 13.75 32.12 50.49 24.28 30.29 40.01
Temporal dependence
Hart 15.08 1.2 12.9 72.8 2.39 27.01 51.63 8.32 14.05 27.82
Equalizer of income
CDW 1.61 -0.88 7.83 -32.82 -2.62 2.22 7.06 -1.44 -0.33 2.45
Fields 1.31 -0.91 8.63 -27.91 -3.46 1.91 7.27 -2.31 -1.22 1.78
Shorrocks 6.58 0.56 6.37 41.64 1.17 11.71 22.25 3.82 6.36 12.35

PSEUDO-PANEL
Indicator PANEL
Movement
Ave. Rank Jump 11.8 4.91 13.34 24.35 4.33 14.23 24.13 10.9 14.38 17.83
Fields-Ok 31.59 17.63 35.26 142.41 19.81 40.6 61.38 31.14 38.71 46.72
King 29.63 9.88 30.1 62.05 13.39 31.52 49.65 27.45 33.89 39.84
Temporal dependence
Hart 15.27 2.27 19.66 77.81 2.4 28.88 55.36 12.35 20.81 31.41
Equalizer of income
CDW 5.43 9.57 11.5 -22.86 -0.43 4.75 9.93 1.74 3.41 5.46
Fields 5.37 9.66 11.51 -16.91 -1.08 4.93 10.93 1.22 3.1 5.45
Shorrocks 6.66 2.56 9.51 47.12 1.09 12.46 23.83 5.54 9.24 13.85
238
PSEUDO-PANEL
Indicator PANEL
Movement
Ave. Rank Jump 13.9 5.1 5.1 23.61 5.15 14.5 23.86 11.9 15.57 17.63
Fields-Ok 36.13 16.19 16.19 129.34 17.95 39.13 60.3 32.16 40.63 45.54
King 32.76 10.56 10.6 60.28 16 32.84 49.69 29.55 36.26 39.73
Temporal dependence
Hart 19.86 2.52 2.53 74.8 3.41 28.81 54.21 14.66 24.34 30.77
Equalizer of income
CDW 4.72 6.88 6.86 -20.91 -0.76 4.13 9.01 1.53 3.41 4.64
Fields 4.7 7.68 7.65 -18.68 -1.58 4.32 10.21 0.97 3.14 4.63
Shorrocks 8.69 2 2 44.76 1.55 12.48 23.42 6.57 10.83 13.62
239
Appendix Table A9.1 Standard Error of Estimates of Poverty Dynamics, 2003-2006
(standard deviation across iterations)
PSEUDO-PANEL
Indicator
AM (A) AM (B) BGK DLLM 1 (A) DLLM 1 (B) DLLM 2 (A) DLLM2 (B) DLLM2 (C)
Poverty outflow 0.2534 0.2286 0.2333 0.1546 0.5406 0.4751 0.5733 0.5244
Poverty inflow 0.2652 0.2669 0.2697 0.2865 0.5965 0.4719 0.5303 0.596
Stay in Poverty 0.2652 0.2669 0.2333 0.5965 0.2865 0.596 0.5303 0.4719
Stay Non-Poor 0.2534 0.2286 0.2697 0.5406 0.1546 0.5244 0.5733 0.4751

PSEUDO-PANEL
Indicator
Poverty outflow 0.1604 0.3033 0.2629 0.5229 0.5994 0.6331 0.608 0.6562
Poverty inflow 0.4075 0.2902 0.2152 0.0856 0.6226 0.4414 0.5588 0.5147
Stay in Poverty 0.4075 0.2902 0.2629 0.6226 0.0856 0.5147 0.5588 0.4414
Stay Non-Poor 0.1604 0.3033 0.2152 0.5994 0.5229 0.6562 0.608 0.6331

PSEUDO-PANEL
Indicator
Poverty outflow 0.2404 0.2152 0.2227 0.3586 0.6503 0.5557 0.5706 0.6197
Poverty inflow 0.3927 0.39 0.2814 0.1017 0.5658 0.4808 0.5234 0.6384
Stay in Poverty 0.3927 0.39 0.2227 0.5658 0.1017 0.6384 0.5234 0.4808
Stay Non-Poor 0.2404 0.2152 0.2814 0.6503 0.3586 0.6197 0.5706 0.5557
240
Appendix Table A9.4 Standard Error of Estimates of Income Mobility, 2003-2006
PSEUDO-PANEL
Indicator
AM (A) AM (B) BGK DLLM1 (A) DLLM1 (B) DLLM2 (A) DLLM2 (B) DLLM2 (C)
Ave. Rank Jump 0.0644 0.0724 0.0799 0.0581 0.3343 0.1589 0.2152 0.286
Fields-Ok 0.1619 0.1336 0.3415 0.2303 0.8638 0.3897 0.4724 0.7788
King 0.201 0.1306 0.1835 0.3335 0.7313 0.6343 0.7348 0.6688
Hart 0.0371 0.1039 0.1763 0.0663 1.283 0.2561 0.4382 0.8969
CDW 0.2281 0.0832 0.553 0.2475 0.808 0.3507 0.4218 0.631
Fields 0.244 0.1072 0.4779 0.3303 0.7042 0.4582 0.5394 0.6992
Shorrocks 0.0179 0.0495 0.1085 0.0471 0.548 0.1261 0.2008 0.4104

PSEUDO-PANEL
Indicator
Ave. Rank Jump 0.0777 0.1026 0.0798 0.0368 0.3356 0.1863 0.2448 0.291
Fields-Ok 0.1963 0.2119 0.2984 0.8715 0.8957 0.5402 0.5716 0.745
King 0.212 0.213 0.2366 0.289 0.7626 0.559 0.6368 0.7498
Hart 0.0694 0.2192 0.1765 0.0749 1.3768 0.3781 0.627 0.9206
CDW 0.1672 0.1226 0.6153 0.2981 1.0647 0.4849 0.6193 0.7303
Fields 0.1871 0.1401 0.497 0.3486 0.8058 0.5625 0.6505 0.6995
Shorrocks 0.0683 0.0982 0.1368 0.0344 0.5921 0.17 0.2715 0.4013
241
PSEUDO-PANEL
Indicator
Ave. Rank Jump 0.086 0.085 0.0786 0.0505 0.3344 0.2157 0.2695 0.3188
Fields-Ok 0.232 0.229 0.3121 0.5556 0.8119 0.4715 0.6141 0.7583
King 0.223 0.2143 0.2305 0.2588 0.6528 0.5806 0.6293 0.6621
Hart 0.0784 0.078 0.1695 0.0813 1.2445 0.4731 0.7185 0.9356
CDW 0.2004 0.201 0.5555 0.2296 0.7547 0.389 0.5784 0.5644
Fields 0.2442 0.243 0.5096 0.2926 0.631 0.4636 0.6451 0.6106
Shorrocks 0.0679 0.0723 0.1424 0.039 0.5633 0.2169 0.3089 0.4213
242
Chapter 10 Summary and Conclusion
10.1 Introduction
The primary objective of this study was to measure and examine income mobility in the
Philippines using longitudinal data from the redesigned Family Income and Expenditure Survey and
Labour Force Survey. The study highlights the importance of taking a longitudinal perspective when
examining a country’s income distribution and going beyond cross-sectional indicators of mean
income, poverty and inequality. One of the main advantages of examining income mobility using
longitudinal data over the conventional cross-sectional perspective that is commonly adopted in many
income distribution studies is that the former is able to differentiate between persistent and transient
poor, and between those who experience stable income stream and those who have fluctuating income
flows. Empirical evidence of these patterns provides more accurate insights for policy planning.
10.2 Motivation and Research Goals

Over the recent years, the Philippines has shown strong economic growth which even exceed
economists’ initial growth forecasts. From 2009 to 2012, for instance, its GDP per capita grew at
annual rate of 4.1% (WDI 2014). Due to this apparent rosy economic performance, several major
global international credit rating agencies awarded the country an investment grade. As improved
credit ratings usually translate to lower debt interest payments, experts forecast that the Philippines
will attract more foreign investment and encourage stronger domestic consumption (ADB 2013c).
These factors can potentially propel the country into a virtuous economic growth regime in the
coming years, a welcome outcome for a country that has long been regarded as the “Sick Man of
Asia” due to slow economic growth since the 1980s. However, such an outcome is not pre-ordained
considering that average income, poverty and inequality are not improving. This could be indicative
that the benefits of growth bypass those who are most disadvantaged. Using the longitudinal data that
has recently become available through the country’s improved household survey system, this study
describes income mobility patterns in the country and identifies the offsetting forces that contribute
to trivial changes in the Philippines’s conventional income indicators despite its transition into faster
economic growth regime over the past decade.
The first three chapters of this study set the tone by reviewing the important analytical tools
typically used for examining income mobility of Filipino households. The following six chapters are
motivated by specific research questions, which have attracted a lot of interest among policymakers
and yet have been given little attention in empirical research due to limited panel data needed to carry
out such type of research.
243
10.3 Main Findings
There is significant income mobility in the Philippines despite stagnant income poverty rates and
inequality levels.
One of the key findings of this study is that the distribution of household income in the
Philippines is much more dynamic than what is commonly presumed due to slow poverty reduction
and stubbornly high levels of income inequality. For instance, as demonstrated in Chapter 4, more
than half of the household population moved into a different income quintile from 2003 to 2009. This
implies that a simple examination of temporal trends in cross-sectional indicators of poverty and
inequality in the Philippines will fall short in providing a comprehensive picture of how the country’s
income distribution has evolved throughout the years.
There are various offsetting forces that contribute to minimal changes in cross-sectional income
poverty and inequality.
The results from Chapter 4 provide evidence that for every household that experienced upward
(absolute) income mobility of a certain magnitude, there is approximately one household that
experienced downward income mobility of the same magnitude. Furthermore, the analysis in Chapter
7 reveals that improvement in socio-economic capital levels seem to have been washed out by the
deterioration of economic returns to those capitals.
Filipino households have experienced widely varied income mobility trajectories, with the poor
households experiencing slightly faster income growth rates than the rest.
Results from Chapters 4 and 6 indicate that lower income households were more likely to
experience faster income growth rates than the rest of the population. In particular, if Filipinos were
classified according to their initial household incomes, Chapter 6 demonstrates that lower income
households were more likely to experience consecutive episodes of upward income mobility while
middle income households were more likely to register slower income movements. Higher income
households experienced the most erratic income mobility trajectories. Although these mobility
patterns portray convergence of income and relatively pro-poor growth, temporal income fluctuations
partially drive this result. In particular, when households are classified according to longitudinally-
averaged income instead of their initial incomes, evidence of income convergence becomes weak.
Majority of poor Filipinos are persistently disadvantaged but economic vulnerability also
exacerbates the country’s problem on poverty.
Although there is evidence that lower income households observed faster income growth,
Chapter 5 shows that poverty in the Philippines is still mostly characterized by long episodes of
244
income shortfall below the poverty line. Furthermore, a significant fraction of the Filipinos who are
not persistently poor are economically vulnerable and are at risk of falling into poverty from time to
time. For instance, about 10% to 15% of the non-poor households (based on the US$2/day poverty
line) in a specific year transitioned into poverty in the succeeding survey period. Furthermore, the
proportion of the poor who are in transient poverty increases as the poverty line decreases or as the
poverty measure becomes more sensitive to the illfare of the poorest of the poor.
The quality of employment plays key role in facilitating a positive income mobility regime.
The statistical models used in this study identify employment to be one of the most important
correlates of income mobility. Results from Chapters 5 to 8 highlight that holding good jobs
minimizes the risk of falling into long episodes of poverty while transition into better jobs increases
the odds of upward income mobility.
Pseudo-panel estimation provides an alternative tool for examining welfare dynamics in the
absence of panel data.
Examining welfare dynamics in developing countries is often constrained by the lack of suitable
panel data that track the living standards of people over time. The pseudo-panel estimation approach
discussed in Chapter 9 which use repeated cross-sectional data to impute income trajectories can be
considered as a welcome addition to the modern methods of mobility analysis.
10.4 Broad Policy Implications

The findings underscore the need for more effective policies that will facilitate more sustainable
gains in poverty reduction and equitable distribution of socio-economic opportunities created by
economic growth. There is no one-size-fits-all policy that can be implemented to meet this objective.
In a country like the Philippines where the economic growth is not distributionally-neutral, the focus
should be on finding ways to make growth more responsive to poverty reduction and equally
distributed socio-economic opportunities.
There are several channels through which growth can be more inclusive for the persistently
poor and more sustainable for those who periodically move into and out of economic hardships. To
achieve a pro-poor growth, changes in the socio-demographic structure of the Philippines need to be
examined to ensure that economic returns to higher skills will not deteriorate. In other words, as the
country invests in socio-economic capital development, such an initiative has to be buttressed by an
effective management of the economic returns to these forms of capital. This study also finds that
employment plays a key role in driving household income distribution outcomes. Worryingly,
unemployment rate in the Philippines remains high at 7 percent. In addition to the need to generate
245
more jobs, the country also confronts the challenge of ensuring that these will be productive and
good-quality jobs. In addition, since poverty is still very much an agricultural phenomenon, creating
more jobs in the non-agriculture sector where labour productivity is much higher could have the
enabling capacity of reducing poverty rates. However, this is easier said than done. I have mentioned
in Chapter 8 that if the current labour market trends continue, economists from ILO forecast that
employment-to-population ratio in the Philippines will decrease by 0.1 percentage point between
2010 and 2015 while labour productivity is expected to drop by 1.1 percentage point over the same
period. This is likely to lead to minimal changes in poverty rates in the coming years unless the
bottlenecks toward the creation of more vibrant employment opportunities can be addressed so that
poor people are able to use labour as a vehicle out of economic dearth.
Since living conditions in developing countries are usually plagued by socio-economic risk, the
role of income shocks in the evolution of the household income distribution should also be
underscored. My findings that income shocks in the country have a poverty-increasing effect from
2003 to 2009 may be a cause of concern. Given that access to adequate insurance and social protection
coverage facilitates effective management of risks and their negative consequences for income
distribution outcomes, it is important to ensure that social protection systems are working. However,
studies suggest that while the Philippine government has a wide range of programs offering social
safety nets especially to the poor and the vulnerable segments of the population, most of these
programs are fragmented and thus, do not provide sustainable protection from socio-economic risks.
If left unaddressed, income shocks may continue to have debilitating effects on the poor. This prompts
the need to evaluate the effectiveness of existing social protection infrastructure in the country.
10.5 Limitations and Future Directions

This study has advanced the existing socio-economic development literature in the Philippines
by providing a benchmark for examining income mobility. However, it has a number of limitations
that are also worth pointing out. First, income mobility is not a perfect measure of equality of socio-
economic opportunities. For instance, this study showed that mobility can be inflated by transitory
income fluctuations and thus, it will not be safe to assume that higher levels of mobility are always
desirable. The existing literature has proposed a number of alternative measures of equality of
opportunities, some of which are based on calculating the income growth rates with a declining
weight on growth amongst the rich (Palmisano & Peragine 2014). Future research can explore how
these alternative measures fare relative to the mobility measures discussed in this study. Second, well-
being is measured mainly in terms of household expenditure per capita. Since the 1970s, there has
been a lot of contention on how well-being should be measured. One of the commonly used
alternatives to expenditure is income (or earnings). Chapters 1 and 3 have discussed the advantages
246
and disadvantages of using one over the other and identified this study’s motivation for using
consumption expenditure. Nevertheless, I have carried out robustness analysis using income rather
than expenditure data. What I find from these robustness checks is that there are slightly higher levels
of mobility when income is used instead of expenditure. This can be attributed to the fact that the
temporal distribution of income is less stable than expenditure. All other results are qualitatively
similar. What I have not done in this thesis, however, is to probe beyond the income dimension and
look at non-monetary measures of well-being. Over the recent years, there has been increasing
recognition that understanding living standards and well-being requires shifting the focus of inquiry
from one-dimensional income-based poverty measures to multidimensional poverty measures that
tap other important life domains. This approach emerged from the paradigm on social exclusion and
deprivation proposed by Townsend (1979) and Sen’s notion of functioning and capabilities (Sen
1985). However, much of the empirical application of this framework has focused on data from
developed countries. Some of the results emerging from these studies suggest that there is a relatively
low degree of overlap between income poverty and multidimensional poverty. If income poverty is
not necessarily the same as multidimensional poverty, this difference should affect how poverty-
reduction programs are designed. Whether the same pattern holds true in the context of developing
countries like the Philippines is a promising avenue for future studies. This can be done if we improve
the existing data collection systems in developing countries to incorporate not only a longitudinal
perspective but also shift from the conventional income-based measures to more holistic and more
direct measures of living standards. In the case of the Philippines, combining the data from the Annual
Poverty Indicators Survey which collects information about non-monetary measures of well-being
with the FIES-LFS data could be explored, although much care should be taken to ensure the
comparability of data between these surveys.
Furthermore, to be able to advance research about poverty and disadvantage, it is important to
collect more contextually relevant indicators. For instance, there are some studies from developed
countries following a life-course perspective which find that the relationship between disadvantage
and social status is becoming weaker while life course events are becoming more important
determinants of socio-economic pathways. To be able to determine how life course events shape the
poverty risks of people in the context of developing countries, we need to start collecting such data.
The third potential limitation of this study is the limited number of time points used in the
analyses. Nationwide panel data is scarcely available in the Philippines. Some longitudinal studies
which have significantly longer observation periods cover only limited areas. Since I worked with
limited time points, there has been a preference to use relatively simple approaches, particularly when
estimating the mobility of permanent and transitory income. Nevertheless, in most of my analyses, I
247
examined the robustness of the results to chosen methodologies and measurement parameters up to
the extent that could be afforded using the available data.
The fourth potential limitation relates to how I addressed the issue of non-coverage bias when
working with the panel subsample of FIES-LFS. As explained in Chapter 3, I reweighted the panel
subsample to make it comparable with the cross-sectional sample. While the resulting average
income, poverty and inequality rates are comparable to that of the full cross-sectional sample, the
standard errors are also slightly inflated not only because of the smaller sample sizes but also due to
the use of additional survey weight adjustments. Thus, when comparing groups with respect to some
key characteristics of interest, larger differences are needed to be able to detect significant findings.
The fifth limitation is that most of the mobility estimates are presented at the national and broad
regional levels only. This approach is mainly dictated by the sample size limitations. There is a need
for future studies to provide a more disaggregated set of mobility estimates especially on poverty
dynamics as a number of previous studies have highlighted the significant spatial variations in the
Philippines in terms of socio-economic development. Perhaps, a better approach is to present mobility
estimates at the administrative level (e.g., village, municipal or provincial levels), which local
government units could use as inputs for policy planning. However, this would require the use of
other computationally-intensive statistical techniques such as small area estimation which is beyond
the scope of this study.
Sixth, there is only limited discussion in the thesis on how the mobility estimates for the
Philippines can be viewed in an international comparative context. If the country is to be the next
Asian Tiger, it is important to gauge where the Philippines stands in terms of the socio-economic
mobility relative to other strong candidates within Developing Asia. This analysis is reserved for
future research.
Finally, this study is neither a full account of the economic history of the Philippines over the
past decade nor a comprehensive diagnostic or prescriptive examination of what went wrong and
what has to be done for rapid economic growth to translate into significant poverty reduction and
equitable distribution of opportunities. It is an initial study to investigate the usefulness of utilising
longitudinal data and provide a benchmark for future. In addition, the attempt to showcase the
usefulness of taking a longitudinal perspective when examining a broad range of topics about the
income distribution may have led to a less detailed discussion of the policy implications of the results.
In summary, the Philippines has made a substantial progress in accelerating economic growth
over the past decade. However, it seems to be failing short in achieving some of the goals set forth in
the MDGs, particularly in reducing poverty. For instance, in 1991, about 30% of the country’s
population lived with less than $1.25 a day and the proportion dropped to 18% in 2009 which is still
3 percentage points less than the 2015 MDG target of 15%. Although it is possible that the Philippines
248
will hit the target just in time based on the trends observed over the past ten years, many of its
neighbouring countries with similar or even slower pace of economic growth have attained their
respective poverty targets much earlier (UNDP 2013). This is indicative that there is ample room for
improvement for the country’s poverty reduction efforts. As we start tackling how to address this
“unfinished agenda” and reflect on how we should move forward after 2015, it is hoped that this study
will bring to the attention of researchers, policymakers and other key stakeholders the need to invest
in the collection and statistical analysis of appropriate indicators that would allow a more dynamic
examination of people’s well-being over time. This is the first step in identifying intervention policies
that could improve the living standards of those who remain extremely poor, minimize the
vulnerability of the transiently poor and ensure that the benefits of economic growth are accessible
for everyone.
249
References
Acock, A., & Hurlbert, J. (2011). Social networks, marital status and well-being. Social
Networks, 15(3), 309-334.
Adair, L., Guilkey, D., Bisgrove, E., & Guiltiano, S. (2002). Effect of childbearing on Filipino
women's work hours and earnings. Journal of Population Economics, 15(4), 625-645.
Adam, R., & Jane, J. (1995). Sources of income inequality and poverty in rural Pakistan. Washington,
DC: International Food Policy Research Institute Research Report 102.
Addawe, J., Martinez, A., & Perez, R. (2007). Responding to the Needs for Decentralized Planning:
Small Area Poverty Estimates. Conference Proceedings of the 10th National Convention of
Statistics, Manila, Philippines.
Addabbo, T. & Solinas, G. (2012). Non-Standard Employment and Quality of Work: Towards New
Forms of Measurement: The Case of Italy. Heidelberg: Physica-Verlag.
Alam, M., Biswas, K. and Hassan, K. (2009). A Test of Association between Working Hour and
Work Family Conflict: A Glimpse on Dhaka’s Female White Collar Professionals. International
Journal of Business and Management, 4(5), 27-35.
Albacea, E., & Gironella, A. (2003). Building Panel Data for Monitoring Poverty in the
Philippines. Manila: National Statistics Office.
Albert, Elloso, L. and Ramos, A. (2009). Toward Measuring Household Vulnerability to Income
Poverty in the Philippines. Philippine Journal of Development, 35(1), 23-53.
Aldaba, F. (2009). Poverty in the Philippines: Causes, Constraints and Opportunities. Manila: Asian
Development Bank.
Alkire, S. & Santos, M. (2013). A Multidimensional Approach: Poverty Measurement & Beyond.
Social Indicators Research, 112(2), 239-257.
Allanson, P. (2008). On the Characterisation and Measurement of the Welfare Effects of Income
Mobility from an Ex-ante Perspective. Discussion Paper in Economics No. 48. Scotland:
University of Dundee.
Ang, A., Sugiyarto, G., & Jha, S. (2009). Remittances and Household Behaviour in the
Philippines. Economics Working Paper, No. 188. Manila: Asian Development Bank.
Antman, F., & McKenzie, D. (2007). Earnings Mobility and Measurement Error: A Pseudo-panel
Approach. Economic Development and Cultural Change, 56(1), 125-161.
Araar, A., & Duclos, J. (2007). DASP: Stata package for doing distributive analysis. Available online
at http://dasp.ecn.ulaval.ca/.
Araar, A., Duclos, J., Audet, Makdissi, M., & P. (2009). Testing for Pro-poorness of Growth, with an
Application to Mexico. Review of Income and Wealth, 55(4), 853-881.
250
Arellano, M., & Bond, S. (1991). Some Tests of Specification for Panel Data: Monte Carlo Evidence
and an Application to Employment Equations. The Review of Economic Studies, 58(2), 277-297.
Arrow, Bowles and Durlauf (2004). Meritocracy and Economic Inequality. New Jersey: Princeton
University Press.
Asian Development Bank. (2007). Philippines: Critical development constraints. Country
Diagnostic Studies. Manila: Asian Development Bank.
Asian Development Bank. (2010a). The Rise of the Middle Class. Key indicators 2010 Special
Chapter. Manila: Asian Development Bank.
Asian Development Bank & Badan Pusat Statistik. (2011). The Informal Sector and Informal
Employment in Indonesia. Manila: Asian Development Bank.
Asian Development Bank. (2011a). Asia 2050: Realizing the Asian Century. Manila: Asian
Development Bank.
Asian Development Bank. (2011b). Towards Higher Quality Employment. Key Indicators 2011
Special Chapter. Manila: Asian Development Bank.
Asian Development Bank. (2012a). Global Crisis, Remittances and Poverty in Asia. Manila:
Philippines.
Asian Development Bank. (2012b). Asian Development Outlook 2012: Confronting Rising
Inequality. Manila: Asian Development Bank.
Asian Development Bank. (2013a). Key Indicators for Asia and the Pacific 2013. Manila: Asian
Development Bank.
Asian Development Bank. (2013b). Asian Development Outlook 2013 Update. Manila: Asian
Development Bank.
Asian Development Bank. (2013c). Asian Development Outlook 2013: Asia’s Energy Challenge.
Manila: Asian Development Bank.
Asian Development Bank. (2014). Poverty in Asia: A Deeper Look. Key Indicators 2014 Special
Chapter. Manila: Philippines.
Aslam, A., & Corrado, L. (2012). The Geography of Well-being. Journal of Economic
Geography, 12(3), 627-649.
Atkinson, A., Bourguignon, F. & Morrisson, C. (1998). Income Distribution and Wealth Inequalities:
Income Mobility. European Economic Review, 32, 619-632.
Atkinson, A., Bourguignon, F., & Morrisson, C. (1992). Empirical Studies of Earnings Mobility. In
J. Lesourne, & H. Sonnenschien (Eds.), Fundamentals of Pure and Applied Economics Volume
52. Philadelphia: Harwood Academic Publishers.
Atkinson, A. (1987). On the Measurement of Poverty. Econometrica, 55(4), 749-764.
251
Australian Bureau of Statistics. (2009). People With More Than One Job. Available online:
http://www.ausstats.abs.gov.au/ausstats/subscriber.nsf/LookupAttach/4102.0Publication24.09.
095/$File/41020_Multiplejobs.pdf
Averett, S. (2010). Moonlighting: Multiple Motives and Gender Differences. Applied Economics,
33(11), 1391-1410.
Azam, S., & Imai, K. (2009). Vulnerability and Poverty in Bangladesh. Australia-South Asia
Research Centre Working Paper No. 28. Canberra: Australian National University.
Azevedo, J., Nguyen, M. & Sanfelice, V. (2012). ADECOMP: Stata Module to Estimate Shapley
Decomposition by Components of a Welfare Measure. Software Components. Massachusetts:
Boston College Department of Economics.
Azevedo, J., Inchauste, G., Olivieri, S., Saavedra, J. & Winkler, H. (2013). Is Labour Income
Responsible for Poverty Reduction? A Decomposition Approach. World Bank Policy Research
Working Paper No. 6414.
Baker, P. (2004). Men’s Health and Economic Well-being. Local Economy, 19(3), 278-281.
Balisacan, A. (1994). Poverty, Urbanization, and Development Policy: A Philippine Perspective.
Manila: University of the Philippines Press.
Balisacan, A. (1997). In Search of Proxy Indicators for Poverty Targeting: Toward a Framework for
a Poverty Indicator and Monitoring System. Paper prepared for the National Statistics Office’s
Component in the UNDP-assisted Project: Strengthening Institutional Mechanisms for the
Convergence of Poverty Alleviation Efforts. Quezon City: University of the Philippines.
Balisacan, A. (2001). Poverty in the Philippines: An Update and Re-examination. Philippine Review
of Economics, 38(1), 15-52.
Balisacan, A. (2007). Why Does Poverty Persist in the Philippines? Facts, Fancies and Policies. In
Severino and Salazar (Eds.) Whither the Philippines in the 21st century? Singapore: Institute of
Southeast Asian Studies.
Balisacan, A. & Fujisaki, S. (1998). Growth, Poverty, and Income Inequality in the Philippines.
Tokyo: Institute of Developing Economies.
Balisacan, A., & Hill, H. (2003). The Philippine Economy: Development, Policies, and Challenges.
New York: Oxford University Press.
Balisacan, A., & Pernia, E. (2002). Probing Beneath Cross-National Averages: Poverty, Inequality,
and Growth in the Philippines. Economics Research Working Paper No. 7. Manila: Asian
Development Bank.
Balisacan, A., Piza, S., Mapa, D., Santos, C., & Ondra, D. (2010). The Philippine Economy and
Poverty during the Global Economic Crisis. Philippine Review of Economics, 41(1), 1-37.
252
Bane, M., & Ellwood, D. (1986). Slipping Into and Out of Poverty: The Dynamics of Spells. Journal
of Human Resources, 21, 1-23.
Banerjee, A., & Duflo, E. (2003). Inequality and Growth: What Can the Data Say? Journal of
Economic Growth, 8(3), 267-299.
Banerjee, A., & Duflo, E. (2007). The Economic Lives of the Poor. Journal of Economic
Perspectives, 21(1), 141-167.
Bardhan, P., Bowles, S., & Gintis, H. (2000). Wealth Inequality, Wealth Constraints and Economic
Performance. In Atkinson and Bourguignon Eds. Handbook on Income Distribution. North
Holland: Elsevier Press.
Barrios, E. (2007). Growth convergence and spending efficiency among Filipino households. School
of Statistics Working Paper No. 1. Manila: University of the Philippines Diliman.
Barrios, E., & Landagan, O. (2004). Geographic Distribution of the Poor: Is Poverty
Contaminating? Proceedings of the 9th National Convention of Statistics, Manila, Philippines.
Baulch, B. (2011). Household Panel Data Sets in Developing and Transition Countries. Chronic
Poverty Research Centre. Available online at
http://www.chronicpoverty.org/uploads/publication_files/Annotated_Listing_of_Panel_Dataset
s_in_Developing_and_Transitional_Countries.pdf
Baulch, B., & Hoddinott, J. (2000). Economic Mobility and Poverty Dynamics in Developing
Countries. Journal of Development Studies, 36(6), 1-24.
Bayudan-Dacuycuy, C., & Lim, J. (2013). Chronic and Transient Poverty and Vulnerability to
Poverty in the Philippines: Evidence Using a Simple Spells Approach. Social Indicators
Research, 118(1), 389-413.doi:10.1007/s11205-013-0409-5
Bell, D., Hart, R. & Wright, R. (1997). Multiple Job Holding as a “Hedge” Against Employment.
Discussion Paper No. 1626. London: Centre for Economic Policy Research
Benabou, R., & Ok, E. (2001). Social mobility and the demand for redistribution: POUM
hypothesis. Quarterly Journal of Economics, 117, 871-91.
Bersales, L. (2009). Issues on the Official Poverty Estimation Methodology in the Philippines:
Comparability of Estimates Across Space and Over Time. Discussion Paper No. 17. Manila:
Philippine Institute for Development Studies.
Bibi, S., Duclos, J., & Verdier-Chouchane, A. (2010). Assessing Absolute and Relative Pro-poor
Growth: An Application to the MENA Region. Working Paper No. 111. Tunis-Belvedere:
African Development Bank.
Bigman, D., & Fofack, H. (2000). Geographical Targeting for Poverty Alleviation: An Introduction
to the Special Issue. The World Bank Economic Review, 14(1), 129-145.
253
Bird, K., & Hill, H. (2009). Philippine Economic development: A Turning Point? Southeast Asian
Affairs, 1, 267-285.
Blinder, A. (1973). Wage Discrimination: Reduced Form and Structural Variables. Journal of Human
Resources, 8, 436-455.
Boheim, R. & Taylor, M. (2004). Actual and Preferred Working Hours. British Journal of Industrial
Relations, 42(1), 149-166.
Böheim, R., & Jenkins, S. (2006). A Comparison of Current and Annual Measures of Income in the
British Household Panel Survey. Journal of Official Statistics, 22(4), 733-758.
Bound, J., & Krueger, A. (1991). The Extent of Measurement Error in Longitudinal Earnings Data:
Do Two Wrongs Make a Right? Journal of Labour Economics, 9(1), 1-24.
Bourguignon, F. & Ferreira, F. (2008). Beyond Oaxaca-Blinder: Accounting for Differences in
Household Income Distributions. Journal of Economic Inequality, 6(2), 117-148.
Bourguignon, F., Ferreira, F., & Lustig, N. (2004). In Bourguignon F., Ferreira F. (Eds.), The
Microeconomics of Income Distribution Dynamics in East Asia and Latin America. Washington,
DC: The World Bank.
Bourguignon, F. Goh, C. & Kim, D. (2004). Estimating Individual Vulnerability to Poverty with
Pseudo-panel Data. Policy Research Working Paper 3375. Washington, DC: The World Bank.
Bourguignon, F. (2011). Non-anonymous Growth Incidence Curves, Income Mobility and Social
Welfare Dominance. Journal of Economic Inequality, 9(4), 605-627.
Bowles, S., Gintis, H. & Osborne, M. (2001). The determinants of income: A behavioural approach.
Journal of Economic Literature, 39(4), 1137-1176.
Braham, P., Rattansi, A., & Skellington, R. (1992). Racism and Anti-racism: Inequalities,
Opportunities and Policies. London: SAGE Publications Ltd and Open University Press.
Breen, R., & Moisio, P. (2004). Poverty Dynamics Corrected for Measurement Error. Journal of
Economic Inequality, 2, 171-191.
Brown, R., & Leeves, G. (2011). Comparative Effects of Migrants’ Remittances on Composition of
Recipient Household Income in Two Small, Island Economies. Applied Economics, 43(7), 3965-
3976.
(Philippine) Bureau of Labour and Employment Statistics. (2011). Primer on Labour Force Survey.
Available online at http://www.bles.dole.gov.ph/PUBLICATIONS/primers/LFS_April2011.pdf
Callander, E., Schofield, D., & Shrestha, R. (2011). Multidimensional Poverty in Australia and the
Barriers Ill Health Imposes on the Employment of the Disadvantaged. Journal of Socio-
Economics, 40(6), 736-742.
Campbell, J. (2011). Multiple Jobholding in States. US Monthly Labour Review.
254
Calónico, S. (2006). Pseudo-panel Analysis of Earnings Dynamics and Mobility in Latin
America. Working Paper No. 636. Washington, DC: Inter-American Development Bank.
Canlas, D., Khan, M. & Zhuang, J. (2009). Diagnosing the Philippine Economy: Toward Inclusive
Growth. Manila: Asian Development Bank.
Carter, M. (2000). Land Ownership Inequality and the Income Distribution Consequences of
Economic Growth. World Institute for Development Economics Research Working Paper No.
201. Helsinki: The United Nations University.
Casacuberta, C. & Gandelman, N. (2012). Multiple Jobholding: The Artist’s Labour Supply
Approach. Applied Economics, 44(3), 323-337.
CEIC. (2014). CEIC database. Available online at https://www.ceicdata.com/
Chakravarty, S. & Ambrosio, C. (2010). Polarization Orderings of Income Distributions. Review of
Income and Wealth, 56(1), 47-64.
Chakravarty, S., Dutta, B., & Weymark, J. (1985). Ethical indices of Income Mobility. Social Choice
and Welfare, 2(1), 1-21.
Chant, S. (1997). Women-headed Households: Poorest of the Poor?: Perspectives from Mexico, Costa
Rica and the Philippines. IDS Bulletin, 28, 26-48.
Chesher, A., & Schluter, C. (2002). Welfare Measurement and Measurement Error. Review of
Economic Studies, 69(2), 357-378.
Ching, F. (1993). Eye on Asia: 'The Sick Man of Asia'. Far Eastern Economic Review, 156(44), 51.
Christiaensen, L., & Shorrocks, A. 2. (2012). Measuring Poverty Over Time. Journal of Economic
Inequality, 10(2), 137-143.
Coclanis, P. (2013). Asia’s next tigers? Burma, the Philippines, and Sri Lanka. World Affairs, 175(6),
69.
Collado, M. (1997). Estimating Dynamic Models from Time Series of Independent Cross-Sections.
Journal of Econometrics, 82, 37-62.
Committee on the Elimination of Discrimination Against Women (2009). CEDAW in Action in
Southeast Asia: Philippines. Available online at http://cedaw-seasia.org/philippines.html
Conley, D. (2000). Sibling Sex Composition: Effects on Educational Attainment. Social Science
Research, 29(441), 457.
Conway, K. & Kimmel, J. (1998). Male Labour Supply Estimates and the Decision to Moonlight.
Labour Economics, 5, 135-166.
Creedy, J., & Wilhelm, M. (2002). Income Mobility, Inequality and Social Welfare. Australian
Economic Papers, 41(2), 140-150.
255
Cruces, G., Fields, G., & Viollaz, M. (2013). Can the Limitations of Panel Datasets be Overcome by
Using Pseudo-panels to Estimate Income Mobility? Working Paper. Bonn: Institute for the Study
of Labour.
Cuesta, J., Nopo, H., & Pozzolitto, G. (2011). Using Pseudo-panels to Measure Income Mobility in
Latin America. Review of Income and Wealth, 57(2), 224-246.
Cuevas, S., Mina, C., Barcenas, M. & del Rosario, A. (2009). Informal Employment in Indonesia.
Economics Working Paper No. 156. Manila: Asian Development Bank.
Dang, H., Lanjouw, P., Luoto, J., & McKenzie, D. (2011). Using Repeated Cross-Sections to Explore
Movements Into and Out of Poverty. World Bank Policy Research Paper No.5550. Washington,
DC: The World Bank.
Dang, H., & Lanjouw, P. (2014). Welfare Dynamics Measurement: Two Definitions of a
Vulnerability Line and Their Empirical Application. World Bank Policy Research Paper No.
6994. Washington, DC: The World Bank.
Dang, H., Lanjouw, P., Luoto, J., & McKenzie, D. (2014). Using Repeated Cross-Sections to Explore
Movements Into and Out of Poverty. Journal of Development Economics, 107, 112-128.
Danzer, A. (2011). Labour Supply and Consumption Smoothing When Income Shocks are Non-
insurable. Discussion Paper No. 5499. Bonn: Institute for the Study of Labour.
Dartanato, T., & Nurkholis. (2013). The Determinants of Poverty Dynamics in Indonesia: Evidence
from Panel Data. Bulletin of Indonesian Economic Studies, 49(1), 61-84
doi:10.1080/00074918.2013.772939
Datt, G. (1998). Computational Tools for Poverty Measurement and Analysis. Working Paper No.
50. Washington, DC: International Food and Policy Research Institute.
David, I., & Maligalig, D. (2001). Issues and Recommendations for Improving Poverty
Statistics. Bangkok: Paper Presented at the UNESCAP Working Group of Statistical Experts
Meeting, 2001.
De Bruin, A. & Dupuis, A. (2004). Work-life Balance? Insights from Non-standard Work. New
Zealand Journal of Employment Relations, 29(1), 21-37.
Deaton, A. (1985). Panel Data from Time Series Cross-Sections. Journal of Econometrics, 30, 109-
126.
Deaton, A. (1991). Saving and Liquidity Constraints. Econometrica, 59(5), 1221-1248.
Deaton, A. & Zaidi, S. (2002). A Guide to Aggregating Consumption Expenditures. Living Standards
Measurement Study Working Paper No. 135. Washington, DC: The World Bank.
Deaton, A. (1997). The Analysis of Household Surveys: A Microeconometric Approach to
Development Policy. Baltimore: Johns Hopkins University Press
256
Deaton, A. (2004). Measuring Poverty. Woodrow Wilson School of Public and International Affairs
Working Paper No. 170. New Jersey: Princeton University.
Deaton, A. (2005). Measuring Poverty in a Growing World. The Review of Economics and
Statistics, 87(1), 1-19.
Deaton, A., & Dupriez, O. (2011). Purchasing Power Parity Exchanges Rates for the Global
Poor. American Economic Journal: Applied Economics, 3, 137-166.
Dercon, S., & Krishnan, P. (2000). Vulnerability, Seasonality and Poverty in Ethiopia. Journal of
Development Studies, 36(6), 25-53.
Dicken, P. (2011). Global Shift: Mapping the Changing the Contours of the World Economy (Sixth
Edition). New York: The Guilford Press.
Dickens, R. (2000). The Evolution of Individual Male Earnings in Great Britain: 1975-95. Economic
Journal, 460, 27-49.
DiPrete, T., & Eirich, G. (2006). Cumulative Advantage as a Mechanism for Inequality: A Review
of Theoretical and Empirical Developments. Annual Review of Sociology, 32, 271-297.
Duclos, J. (2009). What is “Pro-Poor”? Social Choice and Welfare, 32(1), 37-58.
Duclos, J., Araar, A., & Giles, J. (2010). Chronic and Transient Poverty: Measurement and
Estimation, with Evidence from China. Journal of Development Economics, 91(2), 266-277.
Duncan, G., Gustaffson, B., Hauser, R., Schmaus, G., Messinger, R., Muffels, H., & Ray, J. (1993).
Poverty Dynamics in Eight Countries. Journal of Population Economics, 6, 215-234.
Duncan, G., & Hill, D. (1985). An Investigation of the Extent and Consequences of Measurement
Error in Labour Economic Survey Data. Journal of Labour Economics, 3(4), 508-522.
Dutrey, A. (2007). Successful Targeting? Reporting Efficiency and Costs in Targeted Poverty
Alleviation Programmes. Social Policy and Development Programme Paper No. 35. Geneva:
United Nations Research Institute for Social Development.
Ebisui, M. (2012). Non-standard Workers: Good Practices of Social Dialogue and Collective
Bargaining. Industrial and Employment Relations Department Working Paper No. 36. Geneva:
International Labour Organization.
Echavez, C., Montillo-Burton, E., McNivem, S., and Quisumbing, A. (2006). Many Paths to the Same
Moon? Moving out of Poverty in Bukidnon, Philippines. Report produced for Broadening Access
and Strengthening Input Market Systems (BASIS) Collaborative Research Support Program
(CRSP) Management Entity, University of Wisconsin-Madison. Available online at
http://crsps.net/wp-content/downloads/BASIS/Inventoried%202.24/13-2006-7-88.pdf
Eichengreen, B., Gupta, P., & Kumar, R. (2010). Emerging Giants: China and India in the World
Economy. New York: Oxford University Press.
257
Elbers, C., Lanjouw, J., & Lanjouw, P. (2003). Micro-level Estimation of Poverty and
Inequality. Econometrica, 71(1), 355-364.
Engel, E., Rigobon, R. and Ferreira, F. (2007). Intragenerational Income Mobility in Latin America.
Economia Spring 2007. Washington, DC: Brookings Institution Press.
Ericta, C., & Fabian E. (2009). A Documentation of the Philippines’ Family Income and Expenditure
Survey. Discussion Paper No. 18. Manila: Philippine Institute for Development Studies.
Essama-Nssah, B., & Lambert, P. J. (2009). Measuring Pro-poorness: A Unifying Approach with
New Results. Review of Income and Wealth, 55(3), 752-778.
Estudillo, J., Sawada, Y., and Otsuka, K. (2008). Poverty and Income Dynamics in Philippine
villages, 1985-2004. Review of Development Economics, 12(4), 877-890.
Fachinger, U., & Himmelreicher, R. (2012). Income Mobility – Curse or Blessing? Mobility in Social
Security Earnings: Data on West-German Men since 1950. Schmollers Jahrbuck, 132(2), 175-
203.
Fafchamps, M. & Lund, S. (2003). Risk Sharing Networks in Rural Philippines. Journal of
Development Economics, 71, 261-287.
Fafchamps, M. and Gubert, F. (2007). The Formation of Risk Sharing Networks. Journal of
Development Economics, 83(2), 326-350.
Ferreira, F., Messina, J., Rigolini, J., Lopez-Calva, L., Lugo, M., & Vakis, R. (2013). Economic
Mobility and the Rise of the Latin American Middle Class. The World Bank Latin American and
Carribean Studies. Washington, DC: The World Bank.
Fields, G. (2008). Income mobility. In L. Blume, & S. Durlauf (Eds.), The New Palgrave Dictionary
of Economics. New York: Palgrave Macmillan.
Fields, G. (2009). Income Mobility within a Generation: An Introduction to the State of the Art in
Latin America. Research for Public Policy Human Development HD-03. New York: United
Nations Development Programme.
Fields, G. (2010). Does Income Mobility Equalize Longer-term incomes? New Measures of an Old
Concept. Journal of Economic Inequality, 8, 409-427.
Fields, G. (2011). What We Know (and Want to Know) about Earnings Mobility in Developing
Countries. Available at http://digitalcommons.ilr.cornell.edu/workingpapers/154
Fields, G., Cichello, P., Freije, P., Menendez, S., &Newhouse, D. (2003). Household Income
Dynamics: A Four-Country Story. Journal of Development Studies, 40(2), 30-54.
Fields, G., Hernandez, R., Freije, S., Puerta, M., Arias, O., & Assuncao, J. (2007). Intragenerational
Income Mobility in Latin America. Economia, 7(2), 101-143.
258
Fields, G., & Ok, E. (1999a). The Measurement of Income Mobility: An Introduction to the
Literature. In J. Silber (Ed.), Handbook of income inequality measurement (pp. 557-596)
Norwell: Kluwer Academic Publishers.
Fields, G., & Ok, E. (1999b). Measuring Movement of Incomes. Economica, 66(264), 455-471.
Fields, G., & Puerta, G. (2010). Earnings Mobility in Times of Growth and Decline: Argentina from
1996 to 2003. World Development, 38(6), 870-880.
Fitzgerald, J., Gottschalk, P., & Moffitt, R. (1998). The Impact of Attrition in the Panel Study of
Income Dynamics on Intergenerational Analysis. Journal of Human Resources, 33(2), 300-344.
Forbes, K. J. (2000). A Reassessment of the Relationship between Inequality and Growth. The
American Economic Review, 90(4), 869-887.
Foster, J. (2009). A Class of Chronic Poverty Measures. In T. Addison, D. Hulme & R. Kanbur
(Eds.), Poverty Dynamics: Interdisciplinary Perspectives (pp. 59-76). Oxford: Oxford
University Press.
Foster, J., & Rothbaum, J. (2012). Mobility curves: Using Cut-offs to Measure Absolute Mobility.
George Washington University, Mimeo.
Foster, J., Seth, S., Lokshin, M., & Sajaia, Z. (Eds.). (2013). A Unified Approach to Measuring
Poverty and Inequality: Theory and Practice. Washington, DC: The World Bank.
doi:doi/book/10.1596/978-0-8213-8461-9
Foster, J., & Shorrocks, A. (1988). Poverty orderings. Econometrica, 56(1), 173-177.
Foster, J., Greer, J., & Thorbecke, E. (1984). A class of decomposable poverty
measures. Econometrica, 52, 761-766.
Fraley, C., Raftery, A. & Scrucca, L. (2014). MCLUST: R Package for normal mixture modelling for
model-based clustering, classification and density estimation including Bayesian regularization.
Available online at http://www.stat.washington.edu/mclust/.
Friedman, M. (1957). The permanent income hypothesis. In M. Friedman (Ed.), A Theory of the
Consumption Function (pp. 20-37) New Jersey: Princeton University Press.
Friedman, M. (1962). Capitalism and freedom. United States of America: Chicago: University of
Chicago Press.
Friedman, M., & Kuznets, S. (1954). Income from Independent Professional Practice. Cambridge:
National Bureau of Economic Research.
Fuwa, N. (2011). Should We Track Migrant Households When Collecting Panel Data?: Household
Relocation, Economic Mobility and Attrition Biases in the Rural Philippines. American Journal
of Agricultural Economics, 93(1), pp. 56-82.
Gaiha, R., & Deolalikar, A. (1993). Persistent, Expected and Innate Poverty: Estimates for Semi-arid
Rural South India, 1975-1984. Cambridge Journal of Economics, 17(4), 409-421.
259
Galord, E. (1996). Convergence?: Inferences from Theoretical Models. The Economic
Journal, 106(437), 1056-1069.
Gasparini, L., Horenstein, M., Molina, E., & Olivieri, S. (2008). Income Polarization in Latin
America: Patterns and Links with Institutions and Conflict. Oxford Development Studies, 36(4),
461-484.
Giannetti, M. (2011). Liquidity Constraints and Occupational Choice. Finance Research Letters,
8(1), 37-44.
Girma. S. (2000). A Quasi-Differencing Approach to Dynamic Modeling from a Time Series of
Independent Cross-Sections. Journal of Econometrics, 98, 365-383.
Geweke, J., & Keane, M. (2000). An Empirical Analysis of Earnings Dynamics among Men in the
PSID: 1968-1989. Journal of Econometrics, 96, 293-356.
Gini, C. (1912). Variabilità e mutabilità. In E. Pizetti & E. Salvemini (Eds.), Memorie Di
Metodologica Statistica. Rome: Libreria Eredi Virgilio Veschi.
Glewwe, P. (2012). How Much of Observed Economic Mobility is Measurement Error? IV Methods
to Reduce Measurement Error Bias, with an Application to Vietnam. World Bank Economic
Review, 26(2), 236-264.
Gochoco-Bautista, M., Bautista, C., Maligalig, D., & Sotocinal, N. (2013). Income Polarization in
Asia. Asian Economic Papers, 12(2), 101-136.
Goldthorpe, J. (2000). On sociology. Oxford: Oxford University Press.
Gottschalk, P., & Huynh, M. (2010). Are Earnings Inequality and Mobility Overstated? The Impact
of Non-classical Measurement Error. The Review of Economics and Statistics, 92(2), 302-315.
Gottschalk, P. (1997). Inequality, Income Growth, and Mobility: The Basic Facts. The Journal of
Economic Perspectives, 11(2), 21-40.
Gradin, C., Canto, O. & del Rio, C. (2008). Inequality, Poverty and Mobility: Choosing Income or
Consumption as Welfare Indicators. Investigaciones Economics, 32(2), 169-200.
Gradin, C., del Rio, C. & Canto, O. (2012). Measuring Poverty Accounting for Time. Review of
Income and Wealth, 58(2), 330-354.
Greenstone, M., Looney, A., Patashnik, J., & Yu, M. (2013). Thirteen Economic Facts about Social
Mobility and the Role of Education. Washington, DC: Brookings: The Hamilton Project.
Grimm, M., Klasen, S., & McKay, A. (2007). Determinants of Pro-poor Growth: Analytical Issues
and Findings from Country Cases. New York: Palgrave Macmillan.
Grimm, M. (2007). Removing the Anonymity Axiom in Assessing Pro-poor Growth. The Journal of
Economic Inequality, 5(2), 179-197.
Grootert, C. & Kanbur, R. (1995). The Dynamics of Welfare Gains and Losses: An African Case
Study, Journal of Development Studies, 31(4), 635-657.
260
Gustavsson, M. (2007). The 1990s Rise in Swedish Earnings Inequality – Persistent or
Transitory. Applied Economics, 39, 25-30.
Halac, M., & Schmukler, S. (2004). Distributional Effects of Crises: The Financial
Channel. Economia, 5(1), 1-67.
Harding, A., & Szukalska, A. (2000). Financial Disadvantage in Australia – 1999. The Unlucky
Australians? Sydney: The Smith Family and National Centre for Social and Economic
Modelling.
Hart, P. (1976). The Comparative Statics and Dynamics of Income Distributions. Journal of the Royal
Statistical Society (Series A), 139(1), 108-125.
Hasan, R., & Jandoc, K. (2010). Workers’ Earnings in the Philippines: Comparing Self-Employment
with Wage Employment. Asian Development Review, 27(1), 43-79.
Haughton, J., & Khandker, S. (2009). Handbook on Poverty and Inequality. Washington, DC: The
World Bank.
Hennigner, N., & Snel, M. (2002). Where are the Poor? Experiences with the Development and Use
of Poverty Maps. Arendal: World Resources Institute and United Nations Environment
Programme.
Heriawan, R. (2004). Improving the Quality of Informal Sector Statistics: Country Experience. The
7th Meeting of the Expert Group on Informal Sector Statistics, Report of the Delhi Group. New
Delhi: Budan Pusat Statistik.
Hilal, S., Sparreboom, T. & Meade, D. (2013). The Philippines Employment Projections Model:
Employment targeting and scenarios. Employment Working Paper No. 140. Geneva:
Hirschman, A., & Rothschild, M. (1973). The Changing Tolerance for Income Inequality in the
Course of Development. The Quarterly Journal of Economics, 87(4), 544-566.
Hoddinott, J. (2006). Shocks and their consequences across and within households in rural
Zimbabwe. Journal of Development Studies, 42(2), 301-321.
Holtz-Eakin, D., Newey, W., & Rosen, H. (1988). Estimating Vector Autoregressions with Panel
Data. Econometrica, 56(6), 1371-1395.
Housing and Urban Development Coordinating Council. (2008). Metro Manila Road Map for Urban
Renewal and Basic Services for the Poor. Manila: Housing and Urban Development
Coordinating Council and Asian Development Bank.
Hout, M., & DiPrete, T. (2006). What We Have Learned: RC 28's Contribution to Knowledge about
Social Stratification. Research in Social Stratification and Mobility, 24(1), 1-20.
Inoue, A. (2008). Efficient Estimation and Inference for Linear Pseudo-panel Data Models. Journal
of Econometrics, 142(1), 449-466.
261
International Labour Organization. (2012). Global Wage Report 2012/13 - Wages and Equitable
Growth. Geneva: International Labour Organization.
International Labour Organization. (2004). Multiple Jobholding. Conditions of Work and
Employment Programme Information Sheet No. WT-19. Geneva: International Labour
Organization.
International Labour Organization. (2009). Globalization and Informal Jobs in Developing Countries.
Geneva: International Labour Organization.
International Labour Organization. (2012). Statistical Update on Employment in the Informal
Economy. Geneva: International labour Organization.
Jain, R., & Sonnen, D. (2011). Social life networks. IT Professional, 13(5), 8-11.
Jalan, J., & Ravallion, M. (1998). Transient Poverty in Post-Reform Rural China. Journal of
Comparative Economics, 26, 338-357.
Jalan, J., & Ravallion, M. (2000). Is Transient Poverty Different? Evidence for Rural China. The
Journal of Development Studies, 36(6), 82-99.Jankowska, A., Nagengast, A., & Perea, J. (2012). The
Middle-Income Trap: Comparing Asian and Latin American Experiences. Development Centre
Policy Insights, 5(96), 1-2.
Jappelli, T. (1990). Who is Credit Constrained in the US Economy. Quarterly Journal of
Economics, 105, 219-234.
Jarvis, S., & Jenkins, S. (1998). How Much Income Mobility is there in Britain? Economic Journal,
Royal Economic Society, 108(447), 428-443.
Jefferson, P. (2012). The Oxford Handbook of the Economics of Poverty. New York: Oxford
University Press.
Jenkins, S. (2011). Changing Fortunes: Income Mobility and Poverty Dynamics in Britain. Oxford:
Oxford University Press.
Jenkins, S., & Van Kerm, P. (2009). Decomposition of Inequality Change into Pro-poor Growth and
Mobility. Paper presented at the 2009 UK Stata Users Group Meeting, London.
Jenkins, S., & Van Kerm, P. (2006). Trends in Income Inequality, Pro-Poor Income Growth and
Income Mobility. Oxford Economic Papers, 58(3), 531-548.
Jones, F.&Kelly, J. (1984). Decomposing Differences between Groups: A Cautionary Note on
Measuring Discrimination. Sociological Methods and Research, 12, 323-343.
Justino, P.,& Verwimp, P. (2013). Poverty Dynamics, Violent Conflict, and Convergence in
Rwanda. Review of Income and Wealth, 59(1), 66-90.
Kakwani, N., & Pernia, E. (2000). What is pro-poor growth? Asian Development Review, 16(1), 1-
22.
262
Kakwani, N., & Son, H. (2004). Pro-poor Growth: Concept, Measurement and Application.
International Poverty Centre Working Paper No. 1. Andar: United Nations Development
Programme.
Kanbur, R. (1987). Measurement and Alleviation of Poverty: With an Application to the Impact of
Macroeconomic Adjustment. International Monetary Fund Staff Papers, 34(1), 60-85.
Kaufmann, F. (1970). Sicherheit als soziologisches und sozialpolitisches Problem. Untersuchungen
zu einer Wertidee hochdifferenzierter Gesellschaften. Stuttgart: Enke.
Kearl, J. & Pope, C. (1983). The Life Cycle in Economic History. The Journal of Economic History.
43(1), 149-158.
Kenworthy, L. (2004). Egalitarian Capitalism: Jobs, Incomes, and Growth in Affluent Countries.
New York: Russell Sage Foundation.
Key Indicators of the Labour Market. (2014). Key Indicators of the Labour Market Database. Geneva:
Khandker, S., & Haughton, J. (2009). Handbook on Poverty and Inequality. Washington, DC: The
World Bank. doi:10.1596/978-0-8213-7613-3
Khor, N., & Pencavel, J. (2008). Income Mobility of Individuals in China and the United States. The
Economics of Transition, 14(3), 417-458.
Kind, H. (2000). The Philippines – The Sick Man of Asia? Economic Development in the Philippines
After 1946. Working Paper No. 24. Bergen: Foundation for Research in Economics and Business
Administration.
King, M. (1983). An Index of Inequality with Applications to Horizontal Equity and Social
Mobility. Econometrica, 51(1), 99-115.
Kochar, A. (1999). Smoothing Consumption by Smoothing Income: Hours-of-Work Responses to
Idiosyncratic Agricultural Shocks in Rural India. Review of Economics and Statistics, 811, 50-
61.
Krebs, T., Krishna, P., & Maloney, W. (2012). Income risk, income mobility and welfare. Policy
Research Working Paper Series 6254. Washington, DC: The World Bank.
Krishnan, P. (1990). The Economics of Moonlighting: A Double Self-Selection Model. Review of
Economics and Statistics, 72(2), 361-367.
Krugman, P. (1992). The Rich, the Right and the Facts. The American Prospect, 11, 19-31.
Kucera, D. & Xenogiani, T. (2009). Persisting Informal Employment: What Explains It? J. Jutting, J.
Laiglesia (Eds.), Is Informal Normal? Towards More and Better Jobs in Developing Countries.
Paris: Organisation for Economic Co-operation and Development.
Kurosaki, T. (2006). The Measurement of Transient Poverty: Theory and Application to Pakistan.
Journal of Economic Inequality, 4(3), 325-345.
263
Kurita, K., & Kurosaki, T. (2011). Dynamics of Growth, Poverty, and Inequality: A Panel Analysis
of Regional Data from Thailand and the Philippines. Asian Economic Journal, 25(1), 3-33.
Kuznets, S. (1955). Economic Growth and Income Inequality. American Economic Review, 65, 1-28.
Lanjouw, P. & Ravallion, M. (1995). Poverty and household size. Economic Journal, 105, 1415-
1434.
Lemmi, A. & Betti, G. (2006). Fuzzy Set Approach to Multidimensional Poverty Measurement. New
York: Springer.
Li, W. (2002). Capitalist Development and Economism in East Asia: The Rise of Hong Kong,
Singapore, Taiwan, and South Korea. Routledge Studies in the Growth Economies of Asia Ed.).
New York: Routledge Publishing.
Liao, T. (2006). Measuring and Analyzing Class Inequality with the Gini Index Informed by Model-
Based Clustering. Sociological Methodology, 36, 201-224.
Lillard, L., & Willis, R. (1978). Dynamic aspects of earning mobility. Econometrica, 46(5), 985-
1012.
Lobao, L., Hooks, G., & Tickamyer, A. (2007). The Sociology of Spatial Inequality. New York:
SUNY Press.
Lorenz, M. (1905). Methods of measuring the concentration of wealth. Publications of the American
Statistical Association, 9(70), 209-219.
Lundborg, P. (1995). Job Amenity and the Incidence of Double Work. Journal of Economic
Behaviour and Organization, 26, 273-287.
Maasoumi, E. (1998). On mobility. In D. Giles and A. Ullah (Ed.), Handbook of Applied Economic
Statistics (pp. 119-176). New York: M. Dekker Inc.
MacLachlan, R., Gilfillan, G., Gordon, J. (2013). Deep and Persistent Disadvantage in Australia. Staff
Working Paper. Canberra: Productivity Commission.
Magno, C. (2011). Analysis of the Basic Education of the Philippines – Implications for the K to 12
Program. Report prepared for SEAMEO INNOTECH and Australian Aid (AusAid). Available
online at
http://www.academia.edu/3814475/Analysis_of_the_Basic_Education_of_the_Philippines_
Maligalig, D., Caoli-Rodriguez, R., Martinez, A., & Cuevas, S. (2014). Education Outcomes in the
Philippines. In M. Concepcion (Ed.), Millennium Development Goals and Beyond: Are We
Making Progress? Volume 2. Manila: National Academy of Science and Technology.
Manasan, R. (2009). Reforming Social Protection Policy: Responding to the Global Financial Crisis
and Beyond. Discussion Paper No. 22. Manila: Philippine Institute for Development Studies.
Marrero, G., & Rodriguez, J. (2013). Inequality of Opportunity and Growth. Journal of Development
Economics, 104, 107-122.
264
Martinez, A. (2013). Small Area Estimation with a Multivariate Spatial Temporal Model. The
Philippine Statistician, 61(2), 1-17.
Martinez, A., Lucio, E. & Villaruel, M. (2014). Examining a Weight Reallocation Method for Small
Area Estimation of Poverty. Accepted for publication in Electronic Journal of Applied Statistical
Analysis.
Martinez, A., Western, M., Haynes, M., Tomaszewski, W., & Macarayan, E. (2014). Multiple
Jobholding and Income Mobility in Indonesia. Research for Social Stratification and Mobility,
37, 91-104.
Mason, W. & Wolfinger, N. (2002). Cohort Analysis. In International Encyclopedia of the Social &
Behavioral Sciences, Smelser, N. & Baltes, P. (Eds.) New York: Elsevier.
McCulloch, N., & Baulch, B. (1999). Distinguishing the Chronically from the Transitorily Poor:
Evidence from Rural Pakistan. Institute of Development Studies Working Paper No. 97. Sussex:
University of Sussex.
McGillivray, M. (2006). Measuring Non-Economic Well-Being Achievement. In M. McGillivray
(Ed.), Inequality, Poverty and Well-Being. (pp. 208-240). London: Palgrave Macmillan.
McKenzie, D. (2004). Asymptotic Theory for Heterogeneous Dynamic Pseudo-Panels. Journal of
Econometrics, 120, 235-262.
Medalla, F. & Jandoc, K. (2009). Philippine GDP Growth After the Asian Financial Crisis: Resilient
Economy or Weak Statistical System? The Philippine Review of Economics, 46(1), 1-34.
Meyer, B., & Sullivan, J. (2003). Measuring the Well-being of the Poor Using Income and
Consumption. Journal of Human Resources, 38(Special Issue), 1180-1220.
Mincer, J. (1974). Schooling, experience and income. New York: Columbia University Press.
Mitra, A. (1992). Urban Poverty: A Rural Spill-Over? Indian Economic Review, 27, 403-419.
Moffitt, R. (1993). Identification and Estimation of Dynamic Models with a Time Series of Repeated
Cross-Sections. Journal of Econometrics, 59(1), 99-124.
Moffitt, R., & Gottschalk, P. (2002). Trends in the Transitory Variance of Earnings in the United
States. Economic Journal, 112, 68-73.
Montalvo, J. (2006). Regional Evolutions in Labour Markets in the Philippines. A Dynamic
Approach. Journal of Asian Economics, 17, 448-477.
Montgomery, M., Gradnolati, M., Burke, K. & Paredes, E. (2000). Measuring Living Standards with
Proxy Variables. Demography, 37(2), 155-174.
Morgan, S., Grusky, D., & Fields, G. (2006). Mobility and Inequality: Frontiers of Research in
Sociology and Economics. California: Stanford University Press.
Morrisson, C. (2006). Institutions, Factor Endowment and Inequality in Ghana, Kenya and
Senegal. Economic Studies in Inequality: Social Exclusion and Well-being, 1, 309-329.
265
Moser, C. (2006). Asset-based Approaches to Poverty Reduction in a Globalized Context: An
Introduction to Asset Accumulation Policy and Summary of Workshop Findings. Global
Economy and Development Working Paper. Washington, DC: Brookings Global Economy and
Development.
National Commission on the Role of Filipino Women. (2004). Report on the State of Filipino Women,
2001-2003. Manila: National Commission on the Role of Filipino Women.
National Economic and Development Authority. (2011). Philippine Development Plan, 2011-2016.
Available online at http://www.neda.gov.ph/?p=1128.
National Economic and Development Authority. (2013). Statement of Secretary Balisacan at the
Quarter 1, 2013 Performance of the Philippine Economy, Press Conference.
National Statistical Coordination Board. (2003). Technical Notes on the Official Poverty Statistics in
the Philippines.
National Statistical Coordination Board. (2012). Fishermen still the Poorest Sector in 2009. Press
Release on Official Poverty Statistics for the Basic Sectors. Available online at
http://www.nscb.gov.ph/pressreleases/2012/PR-201206-SS2-01_pov2009.asp
National Statistical Coordination Board. (2013a). Philippine Statistical Yearbook 2013. Manila:
National Statistical Coordination Board.
National Statistical Coordination Board. (2013b). Poverty incidence unchanged, as of first semester
2012, Press Release. Available online at http://www.nscb.gov.ph/poverty/defaultnews.asp
National Statistics Office. (2003). 2003 Master Sample Brochure. Manila: National Statistics Office.
National Statistics Office. (2015). Technical Notes on the Labour Force Survey. Available online at
http://census.gov.ph/content/technical-notes-labour-force-survey-lfs.
Navarro, A. (2006). Estimating Long Term Earnings Mobility in Argentina with Pseudo-panel
Data. Revista De Analisis Economico, 35, 65-90.
Nussbaum, M., & Sen, A. (1993). The Quality of Life. Oxford: Clarendon Press.
Oaxaca, R. (1973). Male-Female Wage Differentials in Urban Labour Markets. International
Economic Review, 14, pp. 693-709.
Oaxaca, R. & Ransom, M. (1999). Identification in Detailed Wage Decompositions. The Review of
Economics and Statistics, 81(1), pp. 154-157.
Ofstedal, M., Reidy, B., & Knodel, J. (2004). Gender Differences in Economic Support and Well-
Being of Older Asians. Journal of Cross-Cultural Gerontology, 19, 165-201.
Ongsotto, R. & Ongsotto, R. (2002). Philippine History Module-based Learning. Manila: REX
Bookstore.
Organisation for Economic Co-operation and Development. (2010). OECD Economic Survey of
Indonesia. Paris: Organisation for Economic Co-operation and Development.
266
Organisation for Economic Co-operation and Development. (2013). How's Life? Executive Summary
Report of the Better Life Index. Paris: Organisation for Economic Co-Operation and
Development.
Ortiz, I. (2001). Social Protection in Asia and the Pacific. Manila: Asian Development Bank.
Palmisano, F., & Peragine, V. (2014). The Distributional Incidence of Growth: A Social Welfare
Approach. Review of Income and Wealth. doi:10.1111/roiw.12109
Palmisano, F., & Van de Gaer, D. (2013). History Dependent Growth Incidence: A Characterization
and an Application to the Economic Crisis in Italy. SERIES Working Paper No. 45. Bari:
Università degli Studi di Bari “Aldo Moro”.
Panos, G., Pouliakas, K. & Zangelidis, A. (2011). Multiple Jobholding as a Strategy for Skill
Diversification and Labour Market Mobility. Essex Business School Working Paper. Essex:
University of Essex.
Park, C. (2007). A Dictionary of Environment and Conservation. Oxford: Oxford University Press.
Pasha, H., & Palanivel, T. (2003). Pro-Poor Growth and Policies: The Asian Experience. The Pakistan
Development Review, 42(4), 313-348.
Paxson, C.& Sicherman, N. (1996). The Dynamics of Dual Jobholding and Job Mobility. Journal of
Labour Economics, 14(3), 357-393.
Perlman, R. (1966). Observations on Overtime and Moonlighting. Southern Economic Journal, 33(2),
237-244.
Pernia, E. (2003). Pro-Poor Growth - What is it and How is it Important? Economics Research
Development Policy Brief No. 17. Manila: Asian Development Bank.
Pernia, E. (2008). Migration, Remittances, Poverty and Inequality in the Philippines. School of
Economics Discussion Papers No. 1. Manila: University of the Philippines Diliman.
Philippine Atmospheric Geophysical and Astronomical Administration (PAG-ASA). (2014).
Available online at http://kidlat.pagasa.dost.gov.ph/
Piketty, T. (1995). Social Mobility and Redistributive Politics. The Quarterly Journal of Economics,
110(3), 551-584.
Platteau, J. (1997). Mutual Insurance as an Elusive Concept in Traditional Rural Communities.
Journal of Development Studies, 33(5), 714-717
Ramos, X. (2003). The Covariance Structure of Earnings in Great Britain, 1991-1999. Economica,
70(278), 353-374.
Ravallion, M. (1994). Poverty Comparisons. Chur: Harwood Academic Publishers.
Ravallion, M. (2009). How Relevant is Targeting to the Success of an Anti-poverty Program? The
World Bank Research Observer, 24(2), 205-231.
267
Ravallion, M. (2011). On multidimensional indices of poverty. Journal of Economic Inequality, 9(2),
235-248.
Ravallion, M. (2012). Why Don’t We See Poverty Convergence? The American Economic
Review, 102(1), 504-523.
Ravallion, M. (2014). The ADB Says Poverty Is Rising in Asia: I Have My Doubts. Available online
at http://www.cgdev.org/blog/adb-says-poverty-rising-asia-i-have-my-doubts
Ravallion, M., & Chen, S. (2003). Measuring Pro-poor Growth. Economics Letters, 78(1), 93-99.
Ravallion, M., & Jalan, J. (1996). Growth Divergence Due to Spatial Externalities. Economic
Letters, 53, 227-232.
Ravallion, M., & Wodon, Q. (1997). Poor Areas or Only Poor People? Policy Research Working
Paper No. 1798. Washington, DC: The World Bank.
Renna, F. & Oaxaca, R. (2006). The Economics of Dual Jobholding: A Job Portfolio Model of Labour
Supply. Discussion Paper No. 1915. Bonn: Institute for the Study of Labour.
Reyes, C. (2002a). The Poverty Fight: Have We Made an Impact? Discussion Paper Series No. 20.
Manila: Philippine Institute for Development Studies.
Reyes, C. (2002b). Impact of Agrarian Reform on Poverty. Philippine Journal of Development, 29(2),
63-131.
Reyes, C., & Tabuga, A. (2012). Conditional Cash Program in the Philippines: Is it Reaching the
Extremely Poor. Discussion Paper Series No. 42. Manila: Philippine Institute for Development
Studies.
Reyes, C., Tabuga, A., Mina, C., Asis, R., & Datu, M. (2011). Dynamics of Poverty in the Philippines:
Distinguishing the Chronic from the Transient Poor. Discussion Paper Series No. 31. Manila:
Philippine Institute for Development Studies.
Riddell, W. & St-Hilaire, F. (2002). Adapting public policy to a labour market in transition. Ottawa:
Carleton University Press.
Rodriguez, E., & Tiongson, E. (2001). Temporary Migration Overseas and Household Labour
Supply: Evidence from Urban Philippines. International Migration Review, 35(3), 709-725.
Roemer, J. (1998). Equality of Opportunity. Massachusetts: Harvard University Press.
Rohdes, N., Tang, K., & Rao, P. (2013). Distributional Characteristics of Income Insecurity in the
U.S., Germany, and Britain. Review of Income and Wealth, 60, 159-176.
Ros, J. (2013). Rethinking Economic Development, Growth, and Institutions. New York: Oxford
University Press.
Rubin, D. (1987). Multiple Imputation for Nonresponse in Surveys. New York: Wiley.
268
Ruyter, A., Singh, A. Warnecke, T. & Zammit, A. (2009). Core vs. Non-core Standards, Gender and
Developing Countries: A Review with Recommendations for Policy and Practice. Paper
presented at the ILO conference on Decent Work. Geneva, Switzerland.
Santos-Paulino, A. & Wan, G. (2010). The Global Impact of the Southern Engines of Growth: China,
India, Brazil and South Africa. Policy Brief No. 6. Tokyo: United Nations University.
Santos, A. (2008). Press Statement on the 2006 Official Poverty statistics. Office of the Secretary,
Philippine National Economic Development Authority. Available online at
http://www.nscb.gov.ph/pressreleases/2008/NEDA%20statement.pdf
Schelzig, K. (2005). Poverty in the Philippines: Income, Assets and Access. Manila, Philippines:
Asian Development Bank.
Schluter, C., & Trede, M. (2003). Local Versus Global Assessment of Mobility. International
Economic Review, 44(4), 1313-1335.
Schultz, T. (1975). The Value of the Ability to Deal with Disequilibria. Journal of Economic
Literature, 13(3), 827-846.
Sen, A. (1976). Poverty: An Ordinal Approach to Measurement. Econometrica, 44, 219-231.
Sen, A. (1981). Poverty and Famines: An Essay on Entitlements and Deprivation. Oxford: Clarendon
Press.
Sen, A. (1999). Development as freedom. Oxford: Oxford University Press.
Shapley, L. (1953). A Value for N-person Games. In Contributions to the Theory of Games Volume
2, Kuhn, H. and Tuckers, A. (Eds.) Annals of Mathematical Studies, 28, 307-317.
Shin, D., & Solon, G. (2009). Trends in Men's Earnings Volatility: What does the Panel Study of
Income Dynamics show? Journal of Public Economics, 95, 973-982.
Shisko, R. and Rostker, B. (1976). The Economics of Multiple Jobholding. The American Economic
Review, 66(3), 298-308.
Shorrocks, A. (1982). Inequality Decomposition by Factor Components. Econometrica, 50(1), 193-
212.
Shorrocks, A. (2013a). Ranking income distributions. Economica, 50, 3-17.
Shorrocks, A. (2013b). Decomposition Procedures for Distributional Analysis: A Unified Framework
Based on the Shapley Value. The Journal of Economic Inequality, 11(1), 99-126.
Shorrocks, A., & van der Hoeven, R. (2004). Growth, inequality, and poverty: Prospects for pro-
poor economic development. Oxford: Oxford University Press.
Shorrocks, A., & Wan, G. (2008). Ungrouping income distributions: Synthesising samples for
inequality and poverty analysis. Working Paper No. 16. Tokyo: United Nations University and
World Institute for Development Economics Research.
269
Shorrocks, A. (1978). Income Inequality and Income Mobility. Journal of Economic Theory, 19(2),
376-393. doi:10.1016/0022-0531(78)90101-1
Sinn, H. (1981). Capital Income Taxation, Depreciation Allowances and Economic Growth: A
Perfect-Foresight General Equilibrium Model. Journal of Political Economy, 41(3), 295-305.
Solon, G. (2001). Model of Intergenerational Mobility Variation Over Time and Place. In Corak, M.
(Eds.) Generational Income Mobility in North America and Europe. (pp. 38-47) Cambridge:
Cambridge University Press.
Solow, R. (1956). A Contribution to the Theory of Economic Growth. Quarterly Journal of
Economics, 70(1), 65-94.
Sorokin, P. (1927). Social Mobility. Harper’s Social Science Series. Minneapolis: University of
Minnesotta.
Sorokin, P. (1959). Social and Cultural Mobility. New York: Free Press.
Sta. Ana, S.& Varona, E. (2012). Computation of Spatial Index for the Poor. Manila: Southeast Asian
Regional Centre for Graduate Study and Research in Agriculture and National Statistics Office.
Stahl, D., & Sallis, H. (2012). Model-based Cluster Analysis in Computational Statistics.
Computational Statistics, 4(4), 341-358.
Tabunda, A., & Albert, J. (2002). Philippine Poverty in the Wake of the Asian Financial Crisis and
El Nino. In Khandker, S. (Ed.), Impact of the East Asian Financial Crisis: Revisited. Manila:
World Bank Institute and Philippine Institute for Development Studies.
Takahashi, K. (2013). Pro-poor Growth or Poverty Trap: Estimating Intergenerational Income
Mobility in Rural Philippines. Discussion Paper No. 380. Chiba: Japan External Trade
Organization Institute of Developing Economies.
Teguh, D., & Nurkholis, N. (2013). The Determinants of Poverty Dynamics in Indonesia: Evidence
from Panel Data. Bulletin of Indonesian Economic Studies, 49(1), 61-84.
Theil, H. (1967). Economics and information theory. Chicago: Rand McNally and Company.
Theisen, T. (2009). Multiple job holding in Tanzania. In Kanbur, R. and Svejnar, J. (Eds.), Labour
Markers and Economic Development. New York: Routledge Publishing.
Tilak, J. (2002). Higher Education Under Structural Adjustment. In Banerjee, & Richter
(Eds.), Economic institutions in India (pp. 289-341) New York: Palgrave Macmillan.
Tocqueville, A. ((1856) 1986). L’Ancien régime et la revolution (The Old Regime and the
Revolution). Translated by Bonner, J. New York: Harper and Brothers Publishers
Townsend, P. (1979). Poverty in the United Kingdom: A Survey of Household Resources and
Standards of Living. Harmondsworth: Penguin Publishing.
Townsend, R. (1994). Risk and Insurance in Village India. Econometrica, 62(3), 539-591.
270
Trade Union Congress of the Philippines. (2009). President Arroyo approves Salary Standardization
Law. News article available at http://www.tucp.org.ph/news/index.php/2009/06/president-
arroyo-approves-salary-standardization-law/.
United Nations. (2014). The Millennium Development Goals Report 2014. New York: The United
Nations.
United Nations Development Programme. (1990). United Nations Human Development Report 1990.
New York: United Nations Development Programme.
United Nations Development Programme. (2008). Empowered and Equal: Gender Equality Strategy
2008-2011. New York: United Nations Development Programme.
United Nations Development Programme. (2013). United Nations Human Development Report 2013.
New York: United Nations Development Programme.
United Nations Development Programme. (2013). Asia-Pacific Aspirations: Perspectives for a Post-
2015 Development Agenda. New York: United Nations Development Programme.
United Nations Economic and Social Commission for Asia and the Pacific. (2009). System of
National Accounts 2008. Bangkok: The United Nations.
United Nations Economic and Social Commission for Asia and the Pacific. (2013). Statistical
Yearbook for Asia and the Pacific 2013. Bangkok: The United Nations.
Usui, N. (2011). Transforming the Philippine Economy: Walking on Two Legs. Economics Working
Paper No. 252. Manila: Asian Development Bank.
Usui, N. (2012). Taking the Right Road to Inclusive Growth Industrial Upgrading and Diversiﬁcation
in the Philippines. Manila: Asian Development Bank.
Vandecasteele, L. (2010). Life Course Risks or Cumulative Disadvantage? The Structuring Effect of
Social Stratification Determinants and Life Course Events on Poverty Transitions in Europe.
European Sociological Review, 27(2), pp. 246-263.
Van Kerm, P. (2002). Tools for the analysis of income mobility in Stata. Available online at
http://econpapers.repec.org/paper/bocdsug02/9.htm.
Van Kerm, P. (2006). Comparisons of Income Mobility Profiles. Institute for Social Economic
Research Working Paper Series No. 36. Sussex: University of Essex.
Van Kerm, P. (2009). Income Mobility Profiles. Economics Letters, 102(2), 93-95.
doi:10.1016/j.econlet.2008.11.022
van Praag, B., Hagenaars, A., & van Eck, W. (1983). The influence of classification and observation
errors on the measurement of income inequality. Econometrica, 51(4), 1093-1108.
Verbeek, M.& Nijman, T. (1993). Minimum MSE estimation of a regression model with fixed effects
from a series of cross-sections. Journal of Econometrics, 59, 125-136.
271
Verbeek, M. (2008). Pseudo-panels and repeated cross-sections. The econometrics of panel
data, advanced studies in theoretical and applied econometrics, 46, 369-383.
Vermunt, J., & Magidson, J. (2002). Latent class cluster analysis. In J. Hagenaars, & A. and
McCutchen (Eds.), Applied latent class analysis (pp. 89-106). Cambridge: Cambridge University
Press.
Virola, R. (2008). Presentation on the 2006 official poverty statistics in the Philippines. Available
online at
www.nscb.gov.ph/pressreleases/2008/FINAL%20-%20rav%20presentation,%205mar08.pdf.
Virola, R. (2010). On questions about Philippine GDP estimates. The Philippine Review of
Economics, 47(2), 119-144.
Wade, R. (2001). Global inequality: Winners and losers. The Economist, 359(8219), 72-74.
World Bank. (2001). Philippines Poverty Assessment Volume I: Main Report. Washington: The
World Bank.
World Bank. (2004). World Development Report 2004: Making Services Work for Poor People.
Washington, DC: The World Bank.
World Bank. (2010). Fostering More Inclusive Growth Main Report. Manila: The World Bank
Philippine Country Office.
World Bank. (2012). PovcalNet: An Online Poverty Analysis Tool. Available online at
http://iresearch.worldbank.org/PovcalNet/index.htm.
World Bank. (2013). Philippine Economic Update - Accelerating Reforms to Meet the Jobs
Challenge. Manila: The World Bank Philippine Country Office.
World Bank. (2014). World Development Report 2014: Risk and Opportunity - Managing Risk for
Development. Washington, DC: The World Bank.
World Bank. (2014). Philippine Economic Update – Pursuing Inclusive Growth through Sustainable
Reconstruction and Job Creation. Manila: The World Bank Philippine Country Office.
World Development Indicators. (2014). World Development Indicators Database. Available online
at http://data.worldbank.org.ezproxy.library.uq.edu.au/data-catalog/world-development-
indicators
World Economic Forum. (2006). The Global Gender Gap Report. Geneva: World Economic Forum.
World Health Organization & United Nations International Children’s Emergency Fund. (2008).
Progress on Drinking Water and Sanitation. Special Focus on Sanitation. Geneva: World Health
Organization and United Nations Children’s Fund Joint Monitoring Programme for Water
Supply and Sanitation
272
Worts, D., McDonough, P., & Sacker, A. (2010). Re-assessing Poverty Dynamics and State
Protections in Britain and the US: The Role of Measurement Error. Social Indicators
Research, 97(3), 419-438.
Wu, Z., Zhu, Y., Baimbridge, M. (2009). Multiple Jobholding in the United Kingdom: Evidence from
the British Household Panel Survey. Applied Economics, 41(21), 2751-2766.
Yang, Y., Fu, W., & Land, K. (2004). A Methodological Comparison of Age-Period-Cohort Models:
The Intrinsic Estimator and Conventional Generalized Linear Models. In Sociological
Methodology Volume 34, Stolzenberg (Ed.) Boston: Blackwell Publishing.
Yap, J., Reyes, C., & Cuenca, J. (2009). Impact of the Global Financial and Economic Crisis in the
Philippines. Discussion Paper No. 30. Manila: Philippine Institute for Development Studies.
Yaqub, S. (2000). Poverty Dynamics in Developing Countries. Development Bibliography No. 16.
Brighton: Institute of Development Studies.
Zandvakili, S. (2002). Trends in Earnings Inequality among Young Adults. Review of Social
Economy, 60(1), 93-107.
Zhuang, J., Kanbur, R., & Rhee, C. (2014). Rising Inequality in Asia and Policy Implications.
Working Paper No. 463. Tokyo: Asian Development Bank Institute.
273

A Longitudinal Approach To Measuring Income Mobility Among 4shnv3v8xp

Uploaded by

Copyright:

Available Formats

A Longitudinal Approach To Measuring Income Mobility Among 4shnv3v8xp

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Longitudinal Approach To Measuring Income Mobility Among 4shnv3v8xp

Uploaded by

Copyright:

Available Formats

Open access Dissertation DOI:10.14264/UQL.2015.

A longitudinal approach to measuring income mobility among Filipino households

Published on: 28 May 2015

Rising Income Inequality in China : A Race to the Top

Income Mobility and Income Inequality in Rural China

Income Inequality, Mobilityy and Economic Insecurity in Australia

A New Data Set Measuring Income Inequality

Share this paper:

A thesis submitted for the degree of Doctor of Philosophy at

Martinez, A. & Perales, F. (2014). The Dynamics of Multidimensional Poverty in Australia.

Publications included in this thesis

Australian and New Zealand Standard Research Classifications (ANZSRC)

Fields of Research (FoR) Classification

Chapter 1 Analytical Tools for the Analysis of Income Mobility ...................................... 25

Table 1.1 Differences Between Income Mobility Perspectives ............................................... 34

Figure 1.1 Relationship Between Different Income Mobility Indices…...............................39

A2.1 Constructing Synthetic Pseudo-Panel Data of Income ...................................... 71

ADB – Asian Development Bank

1.2 What is Income Mobility and Why it is Important?

1.3 How Do We Measure Income Mobility?

1.3.1 What Characterizes Mobility of Incomes?

Table 1.1 Differences Between Income Mobility Perspectives

1.3.2 Analytical Tools for Measuring Income Mobility

Table 1.2 Sample Income Transition Matrix

2nd Quintile 0.1 0.4 0.35 0.1 0.05

3rd Quintile 0.05 0.2 0.45 0.2 0.1

4th Quintile 0 0.05 0.35 0.45 0.15

Richest Quintile 0 0.05 0.15 0.25 0.55

In addition to using a transition matrix, there are a number of alternative ways of

1.4.1 Measuring Intertemporal Poverty

Income Movement Average Rank Jump | 𝑟𝑎 𝑜𝑚𝑒𝑖𝑡 𝑟𝑎 ( 𝑜𝑚𝑒𝑖𝑡− )|

Poverty Persistence { 𝑜𝑚𝑒𝑖𝑡 ,} 𝑒𝑟𝑒 {𝑖 𝑜𝑚𝑒𝑖𝑡− }

Poverty Inflow { 𝑜𝑚𝑒𝑖𝑡 ,} 𝑒𝑟𝑒 {𝑖 𝑜𝑚𝑒𝑖𝑡− }

Time dependence Hart 𝑜𝑟𝑟𝑒𝑙( 𝑜𝑚𝑒𝑖𝑡 , 𝑜𝑚𝑒𝑖𝑡− )

Mobility as Equalizer of (∑ 𝑜𝑚𝑒𝑖𝑡)

Chakravarty, Dutta and 𝐼(𝐼𝑛𝑐 )

Duclos, Araar & Giles (2010) Approach

𝛼 (𝑔𝑖 ) = 𝑖𝛼 (𝑔𝑖 ) 𝑖 (𝑔𝑖 ) (1.7)

15 When a household is risk neutral (α = 1), transient poverty will be equal to 0.

Measuring Intertemporal Poverty using the Spells Approach

Conventional Spells Approach

Gradin, del Rio and Canto (GRC) (2012) Approach

1.4.2 Income Mobility-Adjusted Pro-Poor Growth Assessment

Figure 1.2 Sample Growth Incidence Curve

Source: Author’s computations using simulated data.

1.4.3 Measuring Dynamics of Income Inequality

1.4.4 Identifying Correlates of Income Mobility

1.5 Analytical Considerations When Examining Income Mobility

1.5.1 Correcting for Measurement Error and Data Contamination

measurement errors on cross-sectional estimates of poverty and inequality.

1.5.2 Decomposing Income into Permanent and Transitory Components

Table 2.1 Socio-Economic Indicators for Selected Southeast Asian Countries

The Republic of the Philippines is an archipelago consisting of 7,107 islands located in

2.3 Poverty, Inequality and Pro-poor Growth Patterns in the Philippines

Figure 2.1 Comparison of Estimates of Growth from

Source: WDI (2014) and Povcalnet (2014)

2.3.2 Longitudinal Perspective

Source: Author’s computations using simulated data from Povcalnet.

Figure 2.5 Individual Income Growth in the Philippines, 1985-2009

Correlation of income ranks 0.77* 0.8*