Joe Dec2008 Consumptionlaborsupply
Joe Dec2008 Consumptionlaborsupply
Joe Dec2008 Consumptionlaborsupply
Journal of Econometrics
journal homepage: www.elsevier.com/locate/jeconom
article info a b s t r a c t
Article history: We present a new econometric model of aggregate demand and labor supply for the United States.
Available online 16 September 2008 We also analyze the allocation full wealth among time periods for households distinguished by a
variety of demographic characteristics. The model is estimated using micro-level data from the Consumer
JEL classification:
Expenditure Surveys supplemented with price information obtained from the Consumer Price Index. An
C81
D12
important feature of our approach is that aggregate demands and labor supply can be represented in
D91 closed form while accounting for the substantial heterogeneity in behavior that is found in household-
level data. As a result, we are able to explain the patterns of aggregate demand and labor supply in the
Keywords: data despite using a parametrically parsimonious specification.
Consumption © 2008 Elsevier B.V. All rights reserved.
Leisure
Labor
Demand
Supply
Wages
1. Introduction employed, we impute the opportunity wages they face using the
wages earned by employees.
The objective of this paper is to present a new econometric Cross-sectional variation of prices and wages is considerable
model of aggregate consumer behavior for the United States. The and provides an important source of information about patterns
model allocates full wealth among time periods for households of consumption and labor supply. The demographic characteristics
distinguished by demographic characteristics and determines the of households are also significant determinants of consumer
within-period demands for leisure, consumer goods, and services. expenditures and the demand for leisure. The final determinant
An important feature of our approach is the development of a of consumer behavior is the value of the time endowment for
closed form representation of aggregate demand and labor supply households. Part of this endowment is allocated to labor market
that accounts for the heterogeneity in household behavior that activities and reduces the amount available for consumption in the
is observed in micro-level data. Aggregate demand functions are form of leisure.
important components of general equilibrium models that are We employ a generalization of the translog indirect utility
used to analyze the macroeconomic consequences of a broad function introduced by Jorgenson et al. (1997) in modeling
spectrum of public policies. household demands for goods and leisure. This indirect utility
We combine expenditure data for over 150,000 households function generates demand functions with rank two in the sense
from the Consumer Expenditure Surveys (CEX) with price infor- of Gorman (1981). The rank-extended translog indirect utility
mation from the Consumer Price Index (CPI) between 1980 and function proposed by Lewbel (2001) has Gorman rank three. We
2006. Following Slesnick (2002) and Kokoski et al. (1994), we ex- present empirical results for the original translog demand system
ploit the fact that the prices faced by households vary across re- as well as the rank-extended translog system and conclude that the
gions of the United States as well as across time periods. We use rank three system more adequately represents consumer behavior
the CEX to construct quality-adjusted wages for individuals with although the differences are not large.
different characteristics that also vary across regions and over time. Our model of consumption and labor supply is based on two-
In order to measure the value of leisure for individuals who are not stage budgeting and is most similar to the framework described
and implemented by Blundell et al. (1994) for consumption goods
alone. The first stage allocates full wealth, including assets and
∗
Corresponding author. the value of the time endowment, among time periods using the
E-mail addresses: [email protected] (D.W. Jorgenson), standard Euler equation approach introduced by Hall (1978). Since
[email protected] (D.T. Slesnick). the CEX does not provide annual panel data at the household level,
0304-4076/$ – see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.jeconom.2008.09.011
D.W. Jorgenson, D.T. Slesnick / Journal of Econometrics 147 (2008) 326–335 327
we employ synthetic cohorts, introduced by Browning et al. (1985) Gorman showed that if demands are consistent with consumer
and utilized, for example, by Attanasio et al. (1999), Blundell et al. rationality, the matrix {bij (ρ)} has rank that is no larger than three.1
(1994) and many others. We assume that household preferences can be represented by a
We introduce our model of consumer behavior in Section 2. We translog indirect utility function that generates demand functions
first consider the second stage of the model, which allocates full of rank three. Lewbel (2001) has characterized such a utility
consumption among leisure, goods, and services. We subsequently function to be of the form:
ρk ρk ρk
present the first stage of the consumer model that describes the
0 0
1
allocation of full wealth across time periods. In Section 3 we discuss (ln Vk )−1
= α0 + ln αp + ln Bpp ln
Fk 2 Fk Fk
data issues including the measurement of price and wage levels
−1
ρk ρk
that show substantial variation across regions and over time. In
0 0
Section 4 we present the estimation results for the rank-two and + ln BpA Ak − ln γp (1)
Fk Fk
rank-three specifications of our second-stage model. We present
estimates of price and income elasticities for goods and services, where we assume Bpp = B0pp , i0 BpA = 0, i0 Bpp i = 0, i0 αp = −1 and
as well as leisure. We find that the wage elasticity of household i0 γp = 0.
labor supply is essentially zero, but the compensated elasticity To simplify notation, define ln Gk as:
is large and positive. Leisure and consumer services are income
ρk ρk ρk
0 0
1
elastic, while capital services and nondurable goods are income ln Gk = α0 + ln αp + ln Bpp ln
inelastic. Perhaps most important, we find that the aggregate Fk 2 Fk Fk
demands and labor supplies predicted by our model accurately
ρk
0
replicate the patterns in the data despite the (comparatively) + ln BpA Ak . (2)
simple representation of household labor supply. Fk
Finally, we estimate a model of the inter-temporal allocation Application of Roy’s Identity to Eq. (1) yields budget shares of the
of full consumption. We partition the sample of households into form:
17 cohorts based on the birth year of the head of the household.
ρk
1
There are 27 time series observations from 1980 through 2006 for wk = αp + Bpp ln + BpA Ak + γp [ln Gk ]2 (3)
all but the oldest and youngest cohorts and we use these data to D(ρk ) Fk
estimate the remaining unknown parameters of the Euler equation where D(ρk ) = −1 + i0 Bpp ln ρk .
using methods that exploit the longitudinal features of the data. With demand functions of this form, aggregate budget shares,
denoted by the vector w, can be represented explicitly as functions
2. Modeling consumption behavior of prices and summary statistics of the joint distribution of full
expenditure and household attributes:
P
We assume that household consumption and labor supply are Fk wk P
allocated in accord with two stage budgeting. In the first stage, k 1 Fk ln Fk
w = = αp + Bpp ln ρ − i Bpp P
0
D(ρ)
P
full expenditure is allocated over time so as to maximize a lifetime Fk Fk
utility function subject to a full wealth constraint. Conditional on k
Fk (ln Gk ) 2
P P
the chosen level of full expenditure in each period, households Fk Ak
allocate expenditures across consumption goods and leisure so as + BpA + γp P .
Fk Fk
to maximize a within-period utility function.
To describe the second stage model in more detail, assume that
households consume n consumption goods in addition to leisure. 2.1. The inter-temporal allocation of consumption
The within-period demand model for household k can be described
using the following notation: In the first stage of the household model, full expenditure Fkt is
xk = (x1k , x2k , . . . , xnk , Rk ) are the quantities of goods and allocated across time periods so as to maximize lifetime utility Uk
leisure. for household k:
ρk = (pk , pLk ) are prices and wages faced by household k. These (
T
"
(1−σ )
#)
X
−(t −1) Vkt
prices vary across geographic regions and over time. max Uk = Et (1 + δ) (4)
wik = pik xik /Fk is the expenditure share of good i for household Fkt
t =1
(1 − σ )
k.
subject to:
wk = (w1k , w2k , . . . , wnk , wRk ) is the vector of expenditure
T
shares for household k.
(1 + rt )−(t −1) Fkt ≤ Wk
X
Ak is a vector
P of demographic characteristics of household k.
t =1
Fk = pik xik + pLk Rk is the full expenditure of household
k where pLk is the wage rate and Rk is the quantity of leisure where rt is the nominal interest rate, σ is an inter-temporal
consumed. curvature parameter, and δ is the subjective rate of time
In order to obtain a closed-form representation of aggregate preference. We expect δ to be between zero and one and the
demand and labor supply, we use a model of demand that within-period utility function is logarithmic if σ is equal to one.
is consistent with exact aggregation as originally defined by The first order conditions for this optimization yield Euler
Gorman (1981). Specifically, we focus on models for which the equations of the form:
aggregate demands are the sums of micro-level demand functions ∂ Vkt ∂ Vk,t +1 (1 + rt +1 )
rather than the typical assumption that they are generated by (Vkt )−σ = Et (Vk,t +1 )−σ .
∂ Fkt ∂ Fk,t +1 (1 + δ)
a representative consumer. Exact aggregation is possible if the
demand function for good i by household k is of the form: (5)
J
X
xik = bij (ρ)ψj (Fk ).
1 See Blundell and Stoker (2005) for further discussion.
j =1
328 D.W. Jorgenson, D.T. Slesnick / Journal of Econometrics 147 (2008) 326–335
If the random variable ηkt embodies expectational errors for sources. While the BLS provides time series of price indexes for
household k at time t, Eq. (5) becomes: different cities and regions, they do not publish information on
price levels. Kokoski et al. (1994) (KCM) use the 1988 and 1989
∂ Vkt
(Vkt )−σ CPI database to estimate the prices of goods and services in 44
∂ Fkt urban areas. We use their estimates of prices for rental housing,
∂ Vk,t +1 (1 + rt +1 ) owner occupied housing, food at home, food away from home,
= (Vk,t +1 )−σ ηk,t +1 . (6) alcohol and tobacco, household fuels (electricity and piped natural
∂ Fk,t +1 (1 + δ)
gas), gasoline and motor oil, household furnishings, apparel, new
We can simplify this equation by noting that, for the rank three vehicles, professional medical services, and entertainment.4 Given
specification of the indirect utility function given in Eq. (1), we price levels for 1988 and 1989, prices both before and after this
obtain: period are extrapolated using price indexes published by the BLS.
∂ Vkt Vkt Most of these indexes cover the period from December 1977 to the
= (−D(ρkt ))[1 − (γp0 ln ρkt )∗ Gkt ]−2 . present at either monthly or bimonthly frequencies depending on
∂ Fkt Fkt
the year and the commodity group.5
The last term in the square bracket is approximately equal to one These prices are linked to the expenditure data in the CEX.
in the data, so that taking logs of both sides of Eq. (6) yields: Although KCM provide estimates of prices for 44 urban areas across
∆ ln Fk,t +1 = (1 − σ )∆ ln Vk,t +1 the US, the publicly available CEX data do not report households’
cities of residence in an effort to preserve the confidentiality of
+ ∆ ln(−D(ρk,t +1 )) + ln(1 + rt +1 ) − ln(1 + δ) + ln ηkt . (7)
survey participants. This necessitates aggregation across urban
Eq. (7) serves as the estimating equation for σ and the subjective areas to obtain prices for the four major Census regions: the
rate of time preference δ . Northeast, Midwest, South and West. Because the BLS does not
collect nonurban price information, rural households are assumed
3. Data issues to face the prices of Class D-sized urban areas.6
In the United States, the only comprehensive sources of The primitive observational unit in the CEX is a ‘‘consumer
information on expenditure and labor supply are the CEX unit’’, and expenditures are aggregated over all members. We
published by the Bureau of Labor Statistics. These surveys choose to model labor supply at the same level of aggregation
are representative national samples that are conducted for the by assuming that male and female leisure are perfect substitutes
purpose of computing the weights in the CPI. The surveys were when measured in quality-adjusted units. The price of leisure (per
administered approximately every ten years until 1980 when they efficiency unit) is estimated using a wage equation defined over
were given every year. Detailed information on labor supply is ‘‘full time’’ workers, i.e. those who work more than forty weeks
provided only after 1980 and, as a result, we use the sample per year and at least thirty hours per week. The wage equation for
that covers the period from 1980 through 2006. Expenditures are worker i is given by:
recorded on a quarterly basis and our sample sizes range from X X
between 4000 and 8000 households per quarter. To avoid issues ln PLi = βjz zji + βjs (Si∗ zji )
related to the seasonality of expenditures, we use only the set of j j
m
P
3.4. Measuring quality-adjusted household leisure where Rkt = m Rkt is total household leisure computed as the
sum over all adult members.
For workers, estimates of the quantity of leisure consumed are In Fig. 2a we present tabulations of per capita full consumption
easily obtained. The earnings of individual m in household k at time (goods and household leisure) as well as per capita consumption
t are: (goods only). For both series, expenditures are deflated by price
and wage indexes that vary over time and across regions. Over
kt Hkt ,
m
Ekt = pLt qm m
the period from 1980 through 2006, per capita consumption grew
where pLt is the wage at time t per efficiency unit, qm
kt is the quality
at an average annual rate of 1.1% per year compared to 1.0% per
m year for per capita full consumption. Fig. 2b shows the average
index of the worker, and Hkt is the observed hours of work. With
observations on wages and the hours worked, the quality index for level of quality-adjusted leisure consumed per adult. The average
worker m is: annual hours increased by approximately 18% over the 26 years
m
Ekt from 2656 in 1980 to 3177 in 2006. Fig. 2c shows that the inclusion
qm
kt = m
. of household leisure has the effect of lowering the dispersion
pLt Hkt
in consumption in each year. The variance of log per capita full
If the daily time endowment is 14 h, the household’s time consumption is approximately 25% lower than the variance of log
kt ∗ (14) and
endowment measured in efficiency units is Tktm = qm per capita consumption. The trends of the two series, however, are
leisure consumption is Rm
kt = qm
kt ( 14 − H m
kt ). similar.
330 D.W. Jorgenson, D.T. Slesnick / Journal of Econometrics 147 (2008) 326–335
Table 1
Sample summary statistics (Sample size: 154,180).
Variable Mean Standard error Minimum Maximum
Table 2
Price and income elasticities (Reference household: Two adults, Two children, NE Urban, Male, White, Full expenditure = 100 K).
Good Uncompensated price elasticity Compensated price elasticity Full expenditure elasticity
Rank 2 Rank 3 Rank 2 Rank 3 Rank 2 Rank 3
7 In the calculations of the wage elasticities, unearned income is assumed to be Both the rank two and rank three demand systems are
zero the value of the time endowment is equal to full expenditure. consistent with exact aggregation and provide closed form
Table 4
Aggregate budget shares.
Year Sample shares Fitted shares R-squared Aggregation factors
Price Expenditure Demographics
Nondurables
1980–1981 0.1145 0.1074 0.1273 0.3985 −0.3009 0.0098
1985–1986 0.0993 0.1003 0.1609 0.4009 −0.3090 0.0084
1990–1991 0.0967 0.0990 0.1793 0.4051 −0.3141 0.0080
1995–1996 0.0898 0.0892 0.2198 0.3996 −0.3181 0.0077
2000–2001 0.0846 0.0852 0.1910 0.4011 −0.3235 0.0076
2005–2006 0.0864 0.0845 0.1806 0.4055 −0.3279 0.0068
Capital services
1980–1981 0.0956 0.1162 0.0296 0.2100 −0.0141 −0.0797
1985–1986 0.1134 0.1178 0.1003 0.2103 −0.0143 −0.0782
1990–1991 0.1186 0.1213 0.1292 0.2132 −0.0145 −0.0774
1995–1996 0.1222 0.1240 0.1161 0.2161 −0.0150 −0.0771
2000–2001 0.1306 0.1272 0.1226 0.2193 −0.0154 −0.0766
2005–2006 0.1403 0.1344 0.1134 0.2255 −0.0155 −0.0756
Consumer services
1980–1981 0.0566 0.0561 0.0018 −0.0439 0.1202 −0.0202
1985–1986 0.0626 0.0668 0.0111 −0.0370 0.1236 −0.0199
1990–1991 0.0706 0.0678 0.0317 −0.0379 0.1258 −0.0201
1995–1996 0.0734 0.0750 0.0318 −0.0326 0.1270 −0.0193
2000–2001 0.0724 0.0747 0.0420 −0.0350 0.1289 −0.0192
2005–2006 0.0748 0.0678 0.0245 −0.0434 0.1308 −0.0195
Leisure
1980–1981 0.7333 0.7203 0.1506 0.4354 0.1948 0.0902
1985–1986 0.7247 0.7151 0.1532 0.4257 0.1997 0.0897
1990–1991 0.7141 0.7119 0.1804 0.4197 0.2028 0.0895
1995–1996 0.7146 0.7117 0.1791 0.4170 0.2060 0.0887
2000–2001 0.7124 0.7129 0.1758 0.4147 0.2100 0.0882
2005–2006 0.6985 0.7133 0.1458 0.4124 0.2126 0.0883
332 D.W. Jorgenson, D.T. Slesnick / Journal of Econometrics 147 (2008) 326–335
Table 5
Group budget shares of leisure (Male household head with 2 or more adults).
Year At least 1 child No children
Sample share Fitted share R-Squared Sample share Fitted share R-Squared
Table 6
Synthetic cohorts.
Cohort Cohort birth year Average no. observ. Range of no. observ. Years covered
and Yt and Dt are defined similarly for the full expenditure and
demographic components of the aggregate demand system:
Fig. 3b. Age profile of per capita consumption. Fig. 4. Age profile of ln Vk .
where
X X
∆ ln Fc ,t +1 = ln Fk,t +1 − ln Fk,t
kε c kε c
X X
∆ ln Vc ,t +1 = ln Vk,t +1 − ln Vk,t
kε c kε c
X X
∆ ln(−D(ρc ,t +1 )) = ln(−D(ρk,t +1 )) − ln(−D(ρk,t ))
kε c kε c
Table 7
Parameter estimates — intertemporal model.
Least squares estimates
OLS Weighted OLS Random effects
Variable Estimate SE Estimate SE Estimate SE
We find that the cross-sectional and inter-temporal variation and incomes. While demographic characteristics have significant
of prices and wages is substantial which allows us to estimate impacts cross-sectionally, they show little movement over time
the price and wage elasticities very precisely. We find that wage and have little impact on temporal variations of aggregate
elasticities of labor supply are negative but close to zero while demand.
the price elasticities of demand for nondurables and consumer
services are price inelastic. As important, household heterogeneity Appendix
is important in explaining consumption patterns and these effects
would be missed if micro-level data are not used to estimate See Tables A.1–A.3.
the joint determination of labor supply and goods demand. For
example, we find that the numbers of adults and children in each References
household have an important impact on the allocation of full
consumption between leisure and goods. Attanasio, O., Banks, J., Meghir, C., Weber, G., 1999. Humps and bumps in lifetime
consumption. Journal of Business and Economic Statistics 17 (1), 22–35.
The within-period model of consumer behavior has been Attanasio, O., Weber, G., 1995. Is consumption growth consistent with intertempo-
extended by utilizing a less restrictive approach for representing ral optimization? Evidence from the consumer expenditure survey. Journal of
income effects. We estimate a translog demand system of Gorman Political Economy 103 (6), 1121–1157.
Blundell, R., Browning, M., Meghir, C., 1994. Consumer demand and the life-cycle
rank three and compare it with demand functions that are of rank allocation of household expenditures. Review of Economic Studies 61, 57–80.
two. We find that the average income and price elasticities of goods Blundell, R., Pashardes, P., Weber, G., 1993. What do we learn about consumer
and services, as well as leisure, are very similar. However, over demand patterns from micro data?. American Economic Review 83 (3),
570–597.
the entire range of full consumption, the new rank three translog Blundell, R., Stoker, T., 2005. Heterogeneity and aggregation. Journal of Economic
demand system better describes the income effects than the earlier Literature 43 (2), 347–391.
rank two system implemented by, for example, (Jorgenson and Browning, M., Deaton, A., Irish, M., 1985. A profitable approach to labor supply and
consumer demands over the life cycle. Econometrica 53 (3), 503–543.
Slesnick, 1997).
Browning, M., Hansen, L., Heckman, J., 1999. Micro data and general equilibrium
The most important and novel feature of our model of the models. In: Taylor, J., Woodford, M. (Eds.), Handbook of Macroeconomics, vol.
joint determination of leisure and goods demand is that it is 1A. North-Holland, Amsterdam.
consistent with exact aggregation. As illustrated by Jorgenson Gorman, W., 1981. Some Engel curves. In: Deaton, A. (Ed.), Essays in the Theory and
Measurement of Consumer Behavior. Cambridge University Press, Cambridge.
and Wilcoxen (1998), aggregate commodity demands and labor Hall, R., 1978. Stochastic implications of the life cycle-permanent income
supply play crucial roles in general equilibrium models that are hypothesis: Theory and evidence. Journal of Political Economy 86 (6), 971–988.
used to evaluate the macroeconomic consequences of energy and Jorgenson, D., Lau, L., Stoker, T., 1997. Transcendental logarithmic model of
aggregate consumer behavior. In: Jorgenson, D. (Ed.), Aggregate Consumer
environmental policies. As noted by Browning et al. (1999), the Behavior. The MIT Press, Cambridge, pp. 203–356.
challenge is to capture the heterogeneity of household behavior Jorgenson, D., Slesnick, D., 1997. Aggregate consumer behavior and household
in a tractable way in developing aggregates that can be used for equivalence scales. In: Jorgenson, D. (Ed.), Measuring Social Welfare. The MIT
Press, Cambridge, pp. 219–252.
macroeconomic policy evaluation. Jorgenson, D., Wilcoxen, P., 1998. Energy, the environment, and economic growth.
The exact aggregation framework presented in Section 4 In: Jorgenson, D. (Ed.), Energy, the Environment, and Economic Growth. The MIT
incorporates this heterogeneity while also encompassing the price Press, Cambridge, pp. 1–88.
Kokoski, M., Cardiff, P., Moulton, B., 1994. Interarea Price Indices for Consumer
and income variation that is included in traditional models based Goods and Services: An Hedonic Approach. BLS Working Paper 256.
on the highly over-simplified theory of a representative consumer. Lewbel, A., 2001. The Rank Extended Translog, Boston, Department of Economics,
The aggregate demand functions can be represented in closed form Boston College.
Slesnick, D., 2001. Consumption and Social Welfare. Cambridge University Press,
and are obtained by simply summing over all of the households in
Cambridge.
the sample. We find that the aggregate expenditure patterns and Slesnick, D., 2002. Prices and regional variation in welfare. Journal of Urban
leisure demands are largely determined by movements in prices Economics 51 (3), 446–468.