MANAGERIAL AND DECISION ECONOMICS, VOL.

18, 2731 (1997)

Two-stage Budgeting as an Economic

Decision-making Process for
Spanish Consumers
Jose Alberto Molina
Department of Economic Analysis, University of Zaragoza, Spain
Two-stage budgeting as an economic decision-making process for consumers is
illustrated by its application to new data. After showing that the Spanish observed
behaviour is consistent with utility maximization, we estimate, for each stage, the
elasticities of a dynamic AIDS model, which allow us to explain the economic decisions of
consumers. The main findings show that all goods are normal and display decreasing
demands, as theory predicts. In particular, transport, in the first stage, and the purchase of
personal vehicles, in the second, show the highest expenditure effects. Moreover, transport
1997 by
and public transport display the highest Marshallian own-price elasticities.
John Wiley & Sons, Ltd.

Managerial and Decision Economics, 18, 2731, 1997

No. of Figures: 0. No. of Tables: 3 No. of Refs: 15

INTRODUCTION
It is well known that the most important economic
decisions taken by consumers correspond to the
allocation of expenditure to specific consumption
goods. In order to give a suitable response,
consumers employ the two-stage budgeting process
(Strotz, 1957; Gorman, 1959). This method postulates that agents allocate total expenditure first to
broad groups of goods, based on a price index for
each group, and then further allocate expenditure
within each of these groups, based on group
individual prices and group expenditures.
The objective of this paper is to apply the twostage budgeting method as an economic decisionmaking process to a new data set. Specifically, we use
Spanish annual time-series of expenditures and prices
for the period 1964 to 1992. First, however, we carry
out a nonparametric analysis, based on the revealed
preference theory developed by Varian (1982, 1983),
in order to show that the observed behaviour of
Spanish consumers is consistent with the utility
maximization hypothesis. We then choose, as the

CCC 0143-6570/97/010027-05 $17.50

1997 by John Wiley & Sons, Ltd.

demand model for the parametric analysis, a dynamic

version of the linear Almost Ideal Demand System,
proposed by Deaton and Muellbauer (1980a) and
later used, for example, by Blanciforti and Green
(1983), Fujii, Khaled and Mak (1985) and Mergos
and Donatos (1989). This model is first estimated
using total expenditure disaggregated into five main
groups of goods.We then estimate the same demand
system, but now employing only data for one of these
groups, dividing the expenditure in this group into
three specific goods.

THE TWO-STAGE BUDGETING

PROCESS: THE AIDS MODEL
Let us now assume a rational consumer who first
decides with respect to the purchase of five aggregate
consumption goods, namely, food, beverages and
tobacco (F), clothing and footwear (C), housing (H),
transport (T) and other goods and services (O); and,
second, with respect to the three specific goods that
are included in the transport group, that is, the
purchase of personal vehicles (T1), the maintenance

28

J. A. MOLINA

of personal vehicles (T2) and public transport (T3). In

what follows qi denotes the quantity demanded of ith
good, pi represents the corresponding nominal price
and y denotes total expenditure (or income) on the
five aggregated goods. Thus, the utility function
corresponding to the two-stage budgeting process of
the consumer can be written as u F[uF(qF), uC(qC),
uH(qH), uT(qT1, qT2, qT3), uO(qO)], where ug( 1 ) are
sub-utility functions that depend on a subset qg of one
or more goods. F[ 1 ] and the sub-utility functions
ug( 1 ) satisfy the classical conditions of monotonicity
and quasi-concavity.
The consumer first maximizes the utility function
u F[uF(qF), uC(qC), uH(qH), uT(qT), uO(qO)], subject
to the budget constraint and, second, once he or she
has determined the quantity (qT) and expenditure (yT)
destined to a particular aggregate good, in this case,
transport, he should then allocate this expenditure
among the specific goods which are included in the
aggregate, thus maximizing u uT(qT1, qT2, qT3).
To determine the functional form of the demand
equations resulting from the constrained maximization of the utility functions for both stages, we use a
dynamic version of the Almost Ideal Demand System
(AIDS), following Deaton and Muellbauer (1980b),
which specifies the initial constant parameter to be a
linear function of the lagged endogenous variable and
a time trend. Therefore, this dynamic specification
generates the following demand equations:
wit

ai

awit

ai t
33

n
P

gij log pjt


bi log yt a0

n
P

k
1
2

ak

n P
n
P
k

awkt

ak t log pkt
33

gkj log pkt log pjt

i 1; . . . ; n

where wi is the budget share of the ith good and ai , a,

ai , gkj and bi are parameters. The basic demand
restrictions are expressed in terms of the coefficients:
3

33

(1)

ai

1 a;

n
P
i

ai

33

n
P
i

wit

ai

bi log yt

awit

ai t
33

n
P

n
P
j

gij

n
P
i

bi

gij log pjt

wjt log pjt

uit i 1; . . . ; n

Due to the adding-up restriction, the covariance

matrix is singular and the likelihood function
undefined. The usual procedure followed in this
study has been to omit one of the equations at each
stage, to estimate the remaining system and to
calculate the parameters in the omitted equation via
the adding-up condition. In our case, the omitted
equations are other goods and services in the first
stage and public transport in the second.
Model (2) is technically a simultaneous equation
system and, therefore, has been estimated by using
the SURE method of Zellner (1962) employing the
TSP. Because of the nature of our data (time series)
we have tested the individual first-order autocorrelation (AR=MA1) by means of the Godfrey (1978)
Lagrange multiplier test and the joint autocorrelation
using the Harvey (1982) test. We have then tested for
dynamic heteroscedasticity by means of the Engle
(1982) test. Furthermore, the theoretical restrictions
of the system have been tested using the Wald test.
As the estimated results from the model are
presented in the form of elasticities, the expressions
corresponding to the linear AIDS model (2) are: (1)
expenditure:
ei

0;

bi
;
wi

and (2) uncompensated price:

homogeneity:
n
P
j

(3)

The initial specification of model (1) generates

equations that are non-linear in their parameters. To
avoid the non-linear estimation, our paper follows
Deaton and Muellbauer (1980a) and uses the Stone
(1954) index. With this transformation, and adding an
error term, uit, that captures taste shifts, measurement
errors in the dependent variable and the effects of
omitted variables, the stochastic version of the
dynamic linear AIDS is:

adding-up:
n
P

(2)

ESTIMATION PROCEDURE

gij

and symmetry: gij gji.

eij

dij

gij bi wj

wi
wi

where dij is the Kronecker delta.

Managerial and Decision Economics, 18: 2731 (1997)

# 1997 by John Wiley & Sons, Ltd.

29

TWO-STAGE BUDGETING AS AN ECONOMIC DECISION-MAKING PROCESS

EMPIRICAL RESULTS
Table 2.
Before describing the parametric empirical results,
our first purpose is to show that the Spanish observed
behaviour is consistent with the utility maximization
hypothesis. In order to achieve this, we use the
nonparametric approach derived from the revealed
preference theory. This procedure has the advantage
over the parametric methods of not requiring ad hoc
functional specifications for demand equations. On
the basis of observed and measurable magnitudes, the
nonparametric approach allows us to check whether a
given data set is consistent with the neoclassical
model of consumer behaviour. Nonparametric methods have been developed to test data for consistency
with utility maximization by means of several
axioms, with the most general of these being the
so-called Generalized Axiom of Revealed Preference
(GARP). By using the software routine described by
Varian (1985), which is especially designed to
directly test the GARP, we have shown that both
data sets have been generated by a maximizing
behaviour consumer.
The results of the estimation are reported in Tables
1 to 3. In Table 1 we show the Godfrey test, Harvey
test and Engle test values. As can be seen, none of the
equations, except clothing and footwear, exhibit firstorder autocorrelation problems. In addition to this
individual test, we have tested the joint autocorrelation by using the Harvey test, observing that the
values of this test are clearly lower than the critical
values. Therefore, we can reject the hypothesis of
autocorrelation in both demand models at the
conventional 5% level of significance. Moreover,
we have also tested and rejected the dynamic
heteroscedasticity problems. Hence, both specifica-

Table 1.

Theoretical Hypotheses Tests

Wald

w20 05

First stage
Homogeneity (4 d.f.)
Homogeneity and symmetry (14 d.f.)

48.84
77.23

9.49
23.68

Second stage
Homogeneity (2 d.f.)
Homogeneity and symmetry (5 d.f.)

9.39
12.94

5.99
11.07

tions are clearly acceptable from an econometric

point of view.
The test values of the theoretical hypotheses for
both stages are reported in Table 2. The values for
homogeneity and joint homogeneity and symmetry
are greater than their critical values at the 5% level of
significance. Both hypotheses are, therefore, rejected.
These results are in accordance with those reported in
other papers which have estimated the AIDS model
(Deaton and Muellbauer, 1980a; Blanciforti and
Green, 1983; Mergos and Donatos, 1989).
In Tables 3(a) and 3(b) we show the elasticities
evaluated at the mean point of the explanatory
variables. These values are reasonable in signs and
magnitude and, for the most part, are individually
significant. Table 3(a) reflects the expenditure and
price effects for the first stage. With respect to the
expenditure effects, we can note that all values are
positive, that is, all commodities are normal goods,
and statistically significant at the 5% level. Clothing
and footwear and transport are luxuries; by contrast,
food, beverages and tobacco, housing, and other
goods and services (which include medical expenses
and education, among others) are necessities. The
expenditure effect on transport, 1.69, is the highest

Specification Tests
Autocorrelation

Godfrey

w20 05

First stage
Food, beverages and tobacco
Clothing and footwear
Housing
Transport
Other goods and services

0.27
5.08
3.45
2.37
1.47

3.84
3.84
3.84
3.84
3.84

4.24

Second stage
Purchase of personal vehicles
Maintenance of personal vehicles
Public transport

2.00
2.67
1.82

3.84
3.84
3.84

1.37

# 1997 by John Wiley & Sons, Ltd.

Harvey

w20 05
:

Dynamic
heteroscedasticity
Engle
w20 05
:

9.49

0.02
0.42
3.15
0.08
0.09

3.84
3.84
3.84
3.84
3.84

5.99

0.51
0.17
0.28

3.84
3.84
3.84

Managerial and Decision Economics, 18: 2731 (1997)

30
Table 3(a).

J. A. MOLINA

Elasticities: First Stage

Expenditure

Food, beverages and tobacco

Clothing and footwear
Housing
Transport
Other goods and services

0.772a
(10)
1.214a
(10)
0.953a
(12)
1.697a
(9)
0.919a
(13)

Price
Food

Clothing

0.264a
(2.9)
a
0.333
(2.1)
a
0.244
(2.7)
a
0.465
(2.1)
a
0.462
(7)

0.096
(1.7)
a
0.546
(5.7)
a
0.152
(2.5)
0.161
(1.1)
0.045
(0.8)

Housing

Transport

Other

0.110a
(2.3)
a
0.495
(6.3)
a
0.194
(3.8)
0.046
(0.4)
a
0.112
(2.9)

0.061
(1)
0.092
(0.9)
a
0.142
(2.3)
a
0.824
(5.8)
0.032
(0.5)

0.169
(2.4)
0.111
(0.8)
0.116
(1.2)
a
0.490
(2.9)
a
0.500
(7.4)

The figures in parentheses are t-values.

Significant at the 5% level.

Table 3(b).

Elasticities: Second Stage

Expenditure

Purchase of personal vehicles

Maintenance of personal vehicles
Public transport

Price

Total

Specific

Purchase

1.951a
(16)
1.732a
(28)
1.130a
(12)

1.150a
(16)
1.021a
(28)
0.666a
(12)

0.082
(0.8)
a
0.460
(8.6)
a
0.266
(3.6)

Maintenance

Public

0.849a
(5.8)
a
0.512
(6.7)
0.064
(0.6)

0.086
(0.5)
a
0.208
(2.7)
a
0.530
(5.7)

The figures in parentheses are t-values.

Significant at the 5% level.

among all values and more than twice as high as that

for food, beverages and tobacco, 0.77.
Own-price elasticities are all negative, as the
theory predicts, and significant at the 5% level. The
values below one, in absolute terms, of all uncompensated own-price elasticities, indicate that the five
demands are price inelastic, with food, beverages and
tobacco appearing as the category which is most
insensitive to its own price, 0.26; by contrast,
transport is the good which is the most sensitive to its
own price, 0.82. The final results of interest concern
the cross-price values. We can observe that 10 crossprice elasticities are significant and, second, that all
effects are weak, which is suggested by the fact that
all values, in absolute terms, are smaller that one.
Seventeen out of 20 values are negative, indicating
complementary goods.
As regards the second stage, Table 3(b) shows the
expenditure and price effects. First, we have obtained
the specific expenditure elasticities, calculated with
respect to total transport expenditure, rather than to
total expenditure. These values indicate that the
purchase and the maintenance of personal vehicles

are luxuries, with public transport being a necessity

good. Moreover, we can calculate the effects with
respect to total expenditure by using eiY eiYT eTY ,
where eiY denotes total expenditure elasticity for the
ith transport good, eiYT represents the elasticity of the
ith transport item with respect to total transport
expenditure and eTY denotes the expenditure elasticity of transport with respect to total expenditure
or income. As eTY 1.69, total effects for the
purchase of personal vehicles, the maintenance of
personal vehicles and public transport are 1.95, 1.73
and 1.13, respectively. Therefore, we can conclude
that all three transport groups clearly behave as
luxuries.
All uncompensated own-price elasticities are
negative and two of them (the maintenance of
personal vehicles and public transport) are significant
at the conventional 5% level. Every transport demand
is inelastic, with the purchase of personal vehicles
being the most insensitive to its own-price, 0.08. As
regards cross-price effects, both personal transport
goods show complementary relations, significant at
the 5% level. Further, we find asymmetric substitut-

Managerial and Decision Economics, 18: 2731 (1997)

# 1997 by John Wiley & Sons, Ltd.

TWO-STAGE BUDGETING AS AN ECONOMIC DECISION-MAKING PROCESS

ability between private and public

we observe that both relations,
personal vehiclespublic transport
portthe maintenance of personal
substitutability.

transport, that is,

the purchase of
and public transvehicles, exhibit

SUMMARY AND CONCLUSIONS

This paper describes the two-stage budgeting process
as an economic decision-making process for consumers, applying it to the Spainsh economy for the
period 1964 to 1992. After showing, by means of the
revealed preference theory, that the observed behaviour of Spanish consumers is consistent with the
utility maximization hypothesis, we use a dynamic
version of the linear Almost Ideal Demand System in
order to estimate both stages of the two-stage
budgeting process. We first divide total expenditure
into five main groups of goods, and then disaggregate
a particular expenditure into various specific goods.
As regards the first stage, we observe that all goods
are normal and, in particular, the expenditure
elasticity of transport has the highest value, being
more than twice as high as that for food, beverages
and tobacco. Every own-price effect is negative, as
theory predicts, with transport appearing as the
category which is most sensitive to its own-price.
The results of the second stage show that all
specific transport goods are luxuries with respect to
total expenditure, with the value corresponding to the
purchase of personal vehicles being the most
sensitive to income, whereas the most inelastic is
public transport. All own-price elasticities are
negative, with the purchase of personal vehicles
being the most insensitive to its own-price. As
regards cross-price effects, both personal transport
goods show complementary relations and we detect
asymmetric substitutability between private and
public transport goods.

# 1997 by John Wiley & Sons, Ltd.

31

REFERENCES
L. Blanciforti and R. Green (1983). An almost ideal
demand system incorporating habits: an analysis of
expenditures on food and aggregate commodity groups.
The Review of Economics and Statistics, 65, 51115.
A. Deaton and J. Muellbauer (1980a). An almost ideal
demand system. The American Economic Review, 70,
31226.
A. Deaton and J. Muellbauer (1980b). Economics and
Consumer Behaviour, Cambridge: Cambridge University
Press.
R. F. Engle (1982). Autoregressive conditional heteroscedasticity with estimates of the variance of United
Kingdom inflation. Econometrica, 50, 9871007.
E. T. Fujii, M. Khaled and J. Mak (1985). An almost ideal
demand system for visitor expenditures. Journal of
Transport Economics and Policy, 19, 16171.
L. G. Godfrey (1978). Testing against general autoregressive and moving average error models when the
regressors include lagged dependent variables.
Econometrica, 46, 12931301.
W. M. Gorman (1959). Separable utility and aggregation.
Econometrica, 27, 469-81.
A. Harvey (1982). A test of misspecification for systems of
equations. Discussion Paper No. A31, London School of
Economics Econometrics Programme.
G. J. Mergos and G. S. Donatos (1989). Consumer
behaviour in Greece: an application of the almost ideal
demand system. Applied Economics, 21, 98393.
R. Stone (1954). Linear expenditure systems and demand
analysis: an application to the pattern of British demand.
The Economic Journal, 64, 51127.
R. H. Strotz (1957). The empirical implications of a utility
tree. Econometrica, 25, 26980.
H. Varian (1982). The nonparametric approach to demand
analysis. Econometrica, 50, 94573.
H. Varian (1983). Nonparametric tests of consumer
behaviour. Econometrica, 52, 57997.
H. Varian (1985). Nonparameric Demand Analysis (NONPAR).
A. Zellner (1962). An efficient method of estimating
seemingly unrelated regressions and tests for aggregation
bias. Journal of the American Statistical Association, 57,
34868.

Managerial and Decision Economics, 18: 2731 (1997)