DE ECONOMIST 136, NR.

3, 1988

A M O D E L OF REAL AND F I N A N C I A L H O U S E H O L D B E H A V I O U R

BY

ELMER STERKEN*

1 INTRODUCTION

Most macroeconometric models contain separate equations for consumption

and portfolio behaviour. Once a consumption function has been specified, sav-
ing is treated as a residual. Saving is the difference between current income and
current consumption. Portfolio theory on the other hand takes wealth, and
hence saving, as given. The interrelationship between consumption and port-
folio theory has been ignored. Theoretical insights however stress the impor-
tance of integrated modelling of consumption and portfolio decision making.
An example is the monetarist view of integrating the consumption of durables
into the total wealth position, and by that, into the portfolio model of
households. In this paper we will stress the necessity of integrated modelling of
real and financial household behaviour, both in a theoretical and an empirical
context. Consumption and portfolio decisions are taken simultaneously. Our
empirical work consists of an integrated model of Dutch household behaviour.
Other Dutch macroeconometric models, like FREIA-KOMPAS (Van den Berg
et al. (1987)), M O R K M O N (De Nederlandsche Bank (1984)) and CCSO (see
Kuipers, Kuper and Sterken (1987)), specify consumption functions, which in-
clude a wealth variable, without explaining the composition of household
wealth. In our analysis we stress the influence of the composition of financial
wealth on consumption planning (see Purvis (1978) and Pissarides (1978)).
Integrated modelling of expenditure and portfolio behaviour is necessary to
explain wealth and liquidity effects on consumption behaviour and expenditure
effects on financial behaviour. In the standard Brainard-Tobin (1968) Pitfalls
approach to describe the financial sphere of the economic system, no explicit
attention has been paid to the quantity of transmission to and from the real sec-
tor. The accounting framework of real and financial transactions of different
economic sectors (firms, households, government, financial institutions, etc.)
takes account of the transmission of income shortages from the real sector to

* University of Groningen. The author would like to thank Professor Th. van de Klundert, Pro-
fessor S.K. Kuipers, Mr. G.H. Kuper and Dr. N.S. Kroonenberg for their comments on an earlier
version of this paper.
318 E. STERKEN

the monetary sector. This form of balance sheet transmission has been used by
all large scale macroeconometric models. However, no behavioural transmis-
sion has been built into the models, because no explicit attention has been given
to the underlying consumer optimization process.
In the literature there exists a large number of studies on consumption
behaviour. The same holds for studies on portfolio behaviour. In section 2 a
brief survey of the literature on these topics will be given. Section 2 also con-
tains a survey of integrated modelling of consumption and portfolio
behaviour. A detailed survey of integrated modelling of expenditure and finan-
cial decisions can be found in Owen (1986). Our basic model is the model pro-
posed by Purvis (1978). This model can be derived f r o m an explicit optimizing
framework, and has been shown by Parkin, Cooper, Henderson and Danes
(1975).
One of the problems of estimating a Purvis-type model for The Netherlands
is the absence of appropriate statistical information on the wealth position of
households. Van Loo (1983) makes some assumptions on financial asset
holdings of households. For our study it was, however, necessary to disag-
gregate the financial stock holdings of households a bit further. In Sterken
(1987a) the results of a detailed description of the Dutch capital market for the
period 1957-1983 are shown. This information has been supplemented with
estimates of the liquidity holdings of households and the demand for short-
term bank credit by households. The data can be found in the appendix. Sec-
tion 3 contains the estimations for a system of equations on consumption and
portfolio behaviour of Dutch households in the period 1958-1983. Two model
versions, based on different income concepts, are used. First, we include actual
wage and non-wage income as independent variables. They represent the cur-
rent income restriction on consumption and saving decisions. In this approach
it has been assumed that short-term income considerations dominate long-term
permanent income ideas. Consumers are believed to have a short-term
planning horizon (one period in this case), so intertemporal substitution is not
believed to influence current consumption behaviour. Secondly, we include
lagged consumption as a proxy for permanent income and delete current in-
come variables. In this approach, long-term income considerations dominate
short-term income restrictions, and intertemporal substitution plays a role in
real household behaviour.

2 INTEGRATED MODELLING OF CONSUMPTION AND PORTFOLIO BEHAVIOUR

In this section we will discuss the literature on modelling the consumption func-
tion and portfolio behaviour. Both models that separate and integrate con-
sumption and portfolio decisions are reviewed. After that we will discuss in
detail the model, which will be used as a f r a m e w o r k for describing Dutch
household behaviour.
The consumption function is the object of a large number of studies in the
 REAL AND FINANCIAL HOUSEHOLD BEHAVIOUR 319

literature. In theory and in econometric practice different specifications have

been developed and estimated. Okker and Den Haan (1987) present a recent
survey of estimation results for The Netherlands. Blinder and Deaton (1985)
show evidence for the United States. Davidson, Hendry, Srba and Yeo (1978)
and Molana (1987) list results for the British economy.
Until recently modelling the consumption function in theory involved choos-
ing between two ways of putting Fisher's theory of intertemporal optimization
into operation: Milton Friedman's permanent income hypothesis (PIH) or
Modigliani-Brumberg's life cycle hypothesis (LCH). The P I H and L C H lead to
similar econometric specifications, which makes testing both theories against
each other impossible. Testing the P I H or L C H against the standard Keynesian
consumption function is impossible too, because relevant data on permanent
and transitory income components are lacking. Friedman (1956) argues that
permanent consumption depends on permanent income in a stable way. Devia-
tions from permanent levels can only be temporary and are uncorrelated over
time. Modigliani and Brumberg (1954) arrive at a similar result. Consumption
is planned over the consumer's life time and depends on expected discounted
future earnings. Modigliani and Brumberg distinguish the observable com-
ponents of the income variable: labour income and value of assets, while Fried-
man relies on less observable aspects of income.
Blinder and Deaton (1985) list several criticisms of the traditional P I H and
L C H models. First, they mention the Lucas critique. Lucas (1976) points out
that consumption and income should be estimated jointly, because, under the
assumption of rationally formed expectations, the P I H does not lead to a struc-
tural relationship between consumption and income. Besides the traditional
dichotomy between permanent and transitory income, a distinction between
anticipated and unanticipated income should be made. Hall (1978) shows that,
under rational expectations, only surprises in permanent income will affect
current income. Barro (1974) argues that, if people can freely transfer income
across generations, household's net worth (including government debt) does
not produce shifts in the consumption pattern. The difficulty is that consumers
know that sooner or later government debt has to be paid back in the form of
taxes and will take account of their future obligations by lowering current con-
sumption. Another critique, mentioned by Blinder and Deaton, is the omission
of intertemporal substitution from descriptions of consumer behaviour. In-
tertemporal substitution is often represented by the inclusion of a (real) interest
rate in the consumption function. The real interest rate is however often ex-
cluded from the consumption function on empirical grounds. Blinder and
Deaton test the three forms of criticism mentioned above. They conclude that
the Lucas critique is relevant for consumption behaviour in the U.S. Testing
the Barro hypothesis on Ricardian equivalence leads to less unambiguous
results. Intertemporal substitution is not supported by empirical tests, on data
of the U.S. economy. Dutch evidence on the Lucas critique has not yet been
reported. De Haan and Zelhorst (1987) test the Barro hypothesis and review
320 E. STERKEN

empirical work on this subject. They conclude that the results of testing the
Barro hypothesis depend heavily on test statistics and testing procedures. In-
tertemporal substitution could be of some relevance in the Dutch case, because
some major macroeconometric models contain consumption functions, which
include the real interest rate (see FREIA-KOMPAS, MORKMON and CCSO).
The integration of financial variables in the consumption function has been
a popular research topic too. Pigou (1943) discusses the influence of a decreas-
ing price level, which would induce a rise in the real value of assets holdings, on
consumption (which is best known as the real balance effect (Patinkin (1965)).
Metzler (1951) argues that changes in interest rates cause a revaluation of flows
of services over the lifetime of assets. Mishan (1958) argues the existence of two
types of wealth effects on expenditure. The first argument is a scaling argu-
ment. A large amount of wealth will induce less need for saving and increase
consumption. The second type of wealth effect on expenditure is an effect in
which unexpected or exogenous changes in money balances, which lead to ex-
cess money holdings, spill over into expenditure and lead to acquisition of
other assets. This is a disequilibrium effect resulting from differences between
desired and actual asset holdings. Laidler (1984) uses this concept in explaining
the influence of the monetary buffer stock on private consumption. Pissarides
(1978) proves that, if transaction costs are included in the economic model,
both liquidity and wealth effects influence consumption. Empirical studies on
the integration of wealth effects into the consumption function have been made
by Patinkin (1965), Ando and Modigliani (1963), Deaton (1972) and Davidson
et al. (1978) amongst others. Empirical evidence for the Dutch economy has
been provided by Knoester (1980) for Laidler's buffer stock approach and by
Okker and Den Haan (1987).
The modelling of portfolio behaviour is based on the famous Pitfalls paper
by Brainard and Tobin (1968). Brainard and Tobin emphasize the need for in-
tegrated modelling of financial transactions with explicit recognition of
balance sheet restrictions and market equilibrium conditions. Brainard and
Tobin show that balance sheet restrictions require a careful treatment of the
estimation of asset demand equations. If n asset demand equations are to be
estimated and the n assets form the known wealth of a portfolio holder, only
n-1 asset demand equations can be analyzed independently, while the n-th de-
mand equation is the mirror image of all other demand equations. Brainard
and Tobin emphasize the need for dynamic portfolio models, as proposed by
Markowitz (1959), in order to take account of dynamic effects. Parkin (1970)
shows that, if the Brainard-Tobin model is founded on a micro optimization
model in which long-run static behaviour is derived from utility maximization
and adjustment behaviour has been derived from cost minimization, then the
portfolio model parameters are restricted by symmetry, homogeneity and con-
cavity conditions. Although the introduction of parameter restrictions can be
attractive, because they reduce the number of parameters to be estimated, some
objections can be raised against this kind of restriction (see Sterken (1987b)).
 REAL AND F I N A N C I A L H O U S E H O L D BEHAVIOUR 321

Despite these doubts, the Parkin representation of the Brainard-Tobin model

has been popular for econometric purposes. For The Netherlands we mention
M O R K M O N , F R E I A - K O M P A S and Van Loo (1983) (a detailed survey of
foreign macroeconometric models of the financial system would be beyond the
scope of this paper but can be found in Sterken (1987c)). Integration of expen-
diture components in portfolio models has not been studied extensively. The
Keynesian transactions motive can be translated into the inclusion of each ex-
penditure category in the financial demand equations. The monetarist view
that total wealth consists of financial assets, durable consumption goods and
h u m a n capital, and that all relative prices affect portfolio choices, implies the
introduction of prices in financial models.
Most macroeconometric models assume that wealth of households is ex-
ogenous in the portfolio model of consumers. An exception is the model of the
United States by Fair (1984), in which consumption and savings plans are ex-
plicitly derived f r o m an optimization framework. Dutch macroeconometric
models, like F R E I A - K O M P A S and M O R K M O N , do not disaggregate the
private sector into households and firms. M O R K M O N includes wealth of the
total private sector in the consumption function, whereas F R E I A - K O M P A S
computes total private sector wealth and disaggregates into households, firms
and institutional investors on an a d h o c basis. The formation of wealth by
households can not be separated, however, f r o m the consumption decision.
Consumption implies the level of savings and therefore the mutation in total
wealth. This explains the need for integrated expenditure-portfolio models.
Integrated models of portfolio and consumption behaviour are all based on
the initial Brainard-Tobin framework. The Brainard-Tobin P i t f a l l s model con-
tains the essential elements of the wealth allocation version of the so-called
Y A L E approach and has been the starting point for the development of in-
tegrated models of expenditure and portfolio behaviour. For this reason we
will shortly discuss the Brainard-Tobin model and use it later on as a yardstick
when we develop our integrated consumption-portfolio model.
Brainard and Tobin assume that target asset holdings are determined by:

A;= ~ firiX/ r = 1..... n (2.1)

i=0

where A r = target asset holding or the r-th asset;

X 0 - income;
X,,, - total wealth at the end of the period;
X i = other explanatory variable (e.g. interest rates); i~O, rn.

Brainard and Tobin postulate homogeneity of the first degree in wealth:

A;= flriYi X m r = 1. . . . . n (2.2 t

\i=0
322 E. S T E R K E N

Dynamic behaviour has been introduced by assuming a multivariate stock ad-

justment model:

~Ar= ~ drs[A*~-As(-1)l+fr~Xm r,s=l ..... n (2.3)

s=l

End-of-period wealth X m equals ~ sn=1As. The additional A X m term is includ-

ed to separate the allocation of new flows (current savings) from the realloca-
tion of existing asset holdings. So the inclusion of the A X m term allows for
explicit flow effects on portfolio adjustments. Both e x a n t e and e x p o s t balance
sheet restrictions hold, but have a different impact, r=
n 1Ar*_
- Xm, the e x a n t e
restriction, is, according to Purvis (1978), not a budget constraint, but a ra-
t i o n a l desires hypothesis. E x p o s t , the actual end-of-period stocks of assets
must be constrained by total wealth: ~ ~= 1Ar = X m . Because ~ X m = ~ rn=1Z~4r
n
and &A r = ~ s= 1Yrs[As* - A s ( - 1)], with Yrs= d~s +fr, the following expression
holds:

~Sm = ~ ~ °triSi - ~ ~ )Prs'Zs(- 1 ) + ~ ~r (2.4)

r-li 0 r=ls=l r=l

where ~ri = ~ •rsflsi; i = 0.... , m

s--1
•r = disturbance term.

Equation (2.4) holds if and only if:

~l)Iri=O i = 0 ..... m-- 1

r=l
n

E ~rm = 1
r=l
n

7rs=l S = I ..... n
r-1
n

Z er=0
r-1

These constraints guarantee the aggregation restriction on the standard P i t f a l l s

model.
The Brainard-Tobin P i t f a l l s model assumes that households make their
consumption-savings decision and then decide on their desired allocation of the
predetermined wealth amongst n assets in the portfolio. Purvis (1978) proposes
an integrated model of savings and portfolio decisions, which interrelates con-
sumption and portfolio decisions. Purvis assumes a discrete approximation to
a theoretical intertemporal continuous optimizing model constructed by
Parkin, Cooper, Henderson and Danes (1975). A discrete approximation is
necessary, because it is Purvis' and our purpose to estimate the model with
standard estimation techniques. Parkin et al. assume that consumers maximize
 REAL AND FINANCIAL HOUSEHOLD BEHAVIOUR 323

the following utility index:

I = i e-atU(C" W . A , A ) d t (2.5)

where I = utility index;

a = rate of time discount;
C = consumption;
A = vector of asset holdings;
W = matrix of financial asset prices;
U(.) = instantaneous utility function.

Changing asset levels f r o m the o p t i m u m leads to a lower utility index. The utili-
ty function for individual households includes cost of adjustment, which is
represented by the inclusion of the term W . A . Mukherjee and Zabel (1974)
show that consumption and portfolio decisions are interrelated if transaction
costs are included in the utility function. Parkin et al. assume static price expec-
tations, static exogenous asset level expectations and a steady growth of income
in their model. Linearity of the demand equations, as has been assumed by
Purvis, can be achieved if the utility function is of the Stone-Geary type - in
general any quasi-homothetic utility function will result in linear demand rela-
tions [see T h o m a s (1981)]. This leads to the following specification:
m - 1

A; = 2 flriXi r= l,...,n (2.6)

i=0

Purvis assumes that wealth X m does not influence the long run asset demand
functions. He argues that the composition of wealth, not the wealth level, is im-
portant in determining the optimal decision paths. Substituting (2.5) in (2.3)
and adding a disturbance term er yields the following model:
m ] n
~4r= ~ ~iXi - ~ 7~sAs(-1)+er r = l ..... n (2.7)
i=0 s=l

where c~ri= ~ 7,sflsi i = 0 .... , rn - 1

s=l

The consumption function has the same functional form:

/~7 1 n
C = ~ biXi+ ~ e ~ A s ( - l ) + e 0 (2.8)
i-0 s=l

Purvis' model (2.7)-(2.8) can be seen as the solution of the optimization

problem (2.5) under the continuous time restriction that consumption and sav-
ings add up to income: X 0 = C + ~ ~= 1AA,.. Substitution of (2.7) and (2.8) into
this restriction gives:
m ] n m - 1

i=0 s 1 r 1 i 0 r Is 1 r=0
324 E. STERKEN

This condition holds if and onlY if:

n
b0 + ~ a~0 = 1
r-1
n
b i + ~ c¢,.i = 0 i = 1..... m - 1 (2.9)
r=l
n
es - ~ y,.s=O s=l,...,n
r=l

~ ~;r=0
r-0

Income is divided amongst consumption and asset accumulation. Changes in

interest rates or lagged asset holdings lead to reallocation of income within the
total of asset accumulation and consumption. Although Smith (1978) and
Owen (1986) make some critical remarks concerning the lack of hierarchy in
decision making and the exclusion of a wealth constraint, we will use model
(2.7),(2.8) as our reference model. The reasons for choosing the Purvis model
are its proper theoretical foundation and its relatively simple empirical
specification. Hierarchy of decision making is a problem which is, at a macro
level, difficult to describe and test empirically. The inclusion of a wealth con-
straint, as proposed by Owen, leads to serious identification problems (see
Owen (1986)). Empirical models of integrated consumption and portfolio deci-
sions are presented by Purvis (1975), Clements (1976), Backus and Purvis
(1980), Kohli and McKibbin (1982) and Owen (1986). The results of Kohli and
McKibbin are incorporated in a large-scale macroeconometric model of the
Australian economy RBII. We will not discuss these empirical studies in detail,
but compare our results with the results of other studies in section 3. A review
of empirical studies is given by Owen (1986).

3 EMPIRICAL RESULTS

In this section we will show estimation results for the Purvis model applied to
the Dutch economy over the period 1958-1983. Before listing our results we
will discuss some specifications of the Dutch consumption function and the
model of portfolio behaviour of households, which has been developed by Van
Loo (1983). After presenting our own results, we will compare the results,
where possible, with the results f r o m other investigations.
The Central Planning Bureau (see Okker and Den H a a n (1987)) relates con-
sumption to disposable wage income, disposable transfer income, disposable
non-wage income, real interest rate, real wealth of households, revaluation of
wealth and to a u t o n o m o u s consumption. In this specification account has been
taken of the fact that some components of income are fully consumed by
definition (inputed rent on owner-occupied houses, benefits related to medical
services and administration costs of institutional investors). The real interest
rate has been measured as an average of short-term and long-term interest rate.
 REAL AND FINANCIAL HOUSEHOLD BEHAVIOUR 325

Wealth of households is part of the wealth of the private sector and is not
described by a behavioural equation in the model. MORKMON, the macro-
econometric model of De Nederlandsche Bank (1984), includes in the con-
sumption function disposable wage and transfer income, disposable non-wage
income, the nominal capital market interest rate, real wealth of the private sec-
tor and the price index of private consumption. A characteristic of this
specification is the inclusion of the nominal capital market interest rate instead
of the real interest rate. The wealth variable contains the wealth of the private
sector as a whole. Wealth of households is not described by a behavioural equa-
tion in the model. Consumption and savings decisions are made separately in
MORKMON. The Center for Cyclical and Structural Research (CCSO) uses a
similar consumption function (see Kuipers, Kuper and Sterken (1987)).
Knoester (1980) follows the buffer stock approach, suggested by Laidler
(1984). Knoester uses a standard consumption function, with two income
categories (wage and transfer income and other disposable income), the real in-
terest rate and inflation as explanatory variables. Besides these explanatory
variables, Knoester includes a monetary buffer [?r- 1;1,where B,. represents the
amount of base money and V the output of enterprises. The monetary buffer
can be interpreted as a kind of liquidity variable. Other models of the Dutch
economy do not differ fundamentally from the specifications mentioned
above. G R E C O N (see Voorhoeve (1986)) includes a consumer credit variable
in the consumption function, which can be considered as a first step in incor-
porating a wealth component.
Van Loo (1983) describes the financial behaviour of Dutch households. He
assumes that households decide on savings and investments prior to, and in-
dependent of, the distribution of savings over the portfolio. Implicit weak
separability and hierarchy of decision making in real and financial decisions
have been assumed. Three endogenous assets are analyzed: liquid assets,
domestic bonds and mortgage loans. Van Loo's results suggest that the gap be-
tween the bond yield and the yield on mortgage loans affects portfolio deci-
sions. Purchase of houses influence the demand for liquid assets and mortgage
loans, while change in private consumption has an impact on demand for liquid
assets and domestic bonds. Private consumption has not been explained by the
model, however, because the model only describes the financial sector.
Empirical results for the Dutch economy on integrated modelling of real and
financial decisions of households have not yet been given. Most models treat
the consumption decision separately from the portfolio decision. Another
disadvantage of standard modelling is that specifying the consumption func-
tion with aggregate wealth as an explanatory variable assumes that transaction
costs are equal for all assets which are part of total wealth. It is also strange to
assume that, for instance, wealth and liquidity of households are strongly cor-
related over time, so that only one of them can be included in the consumption
function, as Okker and Den Haan (1987) argue.
In matrix notation the Purvis (1978) model, which will be used in this sec-
326 E. STERKEN

tion, becomes:

bo...bm_l I CI,," en Al!-l) CO

17
0~10-" "~lm-1

OJnO.,.OJnm_l
÷
-Yll'"-Yln
: :

-Ynl'"-Ynn
I
LAd-l)
_}_ ~1

(3.1)

The restrictions (2.9) complete the model. In our study we estimate two ver-
sions of the Purvis model, based on two different theoretical income concepts.
First, we use current income (in two categories, to be specified below), which
represents the current restriction on consumption and financial transactions.
Secondly, we interpret lagged consumption as permanent income. Lagged con-
sumption is assumed to represent permanent income, because consumption in
the previous period has been based on the permanent income hypothesis and so
represents all relevant information on previous permanent income. Actual per-
manent income is assumed to differ from previous permanent income by a ran-
d o m variable only (see Hall (1978)). In the second type of model intertemporal
transactions are important, and the current income restriction is believed to be
less relevant. Consumers are able to optimize current and future consumption
and are not restricted by current income. The second type of model can be seen
as a long-term model specification, whereas the first model describes short-
term behaviour. In line with this argument we expect that liquidity considera-
tions have more impact on private consumption in the current income model
than in the permanent income model. In the permanent income model, in-
tertemporal substitution, represented by the influence of interest rates, will
have more impact on consumption than in the current income model.
In the empirical analysis, the model (3.1) and (2.9) has been implemented as
follows (all variables in real terms):

C = private consumption of households;

A1 .... ,A17 = short-term bank loans to households (BCR), liquidity de-
m a n d by households (LIQ), demand for shares by
households (SH), demand for bonds by households
(BON), demand for long-term time and savings deposits
by households (LTD), demand for mortgage loans by
households (ML) and demand for other net assets by
households (ONA).
X 0 = two income variables:
1. actual income consisting of two categories: wages, salaries and
social benefits fie) and other disposable income (n/8);
2. permanent income, represented by C_ 1;
 REAL AND FINANCIAL HOUSEHOLD BEHAVIOUR 327

Xi = yields on various assets and liabilities (rBCR, rLro, rsH, rBON,rLTD)

and inflation Pc.

Liabilities (short-term bank credit (BCR) and mortgage loans (ML)) are
multiplied with - 1 and can thus be treated as assets. We assume that other net
assets do not have an interest rate. Because the interest rate on mortgage loans
is strongly correlated with the interest rate on bonds, we do not include the
mortgage loan interest rate. Other net assets include non-financial wealth, like
real estate and accumulated contractual savings. The change in the sum of all
assets and liabilities equals disposable income minus private consumption.
We have estimated model (3.1) with OLS and Zellner's SUR (Zellner (1962)).
OLS has been applied to estimate the model without restrictions on the para-
meters, except for additive restrictions. Where restrictions on the parameters
have been imposed, the model has been estimated with Zellner's SUR. SUR
takes account on the interdependency between the model equations by assum-
ing that the residuals of the model equations are correlated. SUR is equivalent
to GLS in this type of model. Where current endogenous variables are included
as explanatory variables, 2SLS and 3SLS are more appropriate. In our model,
only lagged endogenous variables are included, so SUR is equivalent to 3SLS.
Denton (1978) shows that OLS yields unbiased results in case all explanatory
variables have been included in each equation of the model. The equation for
O N A has not been estimated, because its parameter estimates follow from the
parameter estimates of all other equations and the additive constraints. These
results are therefore not reported in the tables. Tables 1 and 2 present an over-
view of the OLS parameter estimates f o r the current income and permanent in-
come model, respectively.
From Tables 1 and 2 it can be seen that the current income model outper-
forms the permanent income model, except for the demand for shares. In the
consumption equation, the impact of liquidity holdings in the current income
model is larger than in the permanent income model, while the real domestic
yield on equity has more influence on consumption in the permanent income
model. These results confirm our theoretical assumptions. Both permanent in-
come and current income are important explanatory variables in the demand
for short-term bank credit and the demand for liquidity. The demand for long-
term financial assets is not influenced by income considerations.
F r o m the estimates of the current income model it can be seen that private
consumption is determined by short-term financial stocks, current income and
the real interest rate on short-term bank credit. The demand for short-term
bank credit does not depend on interest rate developments, but on portfolio ad-
justment considerations. The demand for liquidity depends on current income,
inflation and interest rates. The demand for shares depends on the stock of li-
quidity, the stock of private real estate and contractual savings and the interest
rate on short-term bank credit. The interest rate on short-term bank credit can
be interpreted as an alternative yield and as a cost factor. The estimation results
328 E. S T E R K E N

I I I I I I I I ~ w ~
i~1 ~
I

I I

©
<1

O I I

II I

I I

0
I

.a
< 8
I I II l i t I ~ w ~
I I I
t~

8
- T - _ T

I I
 REAL AND FINANCIAL HOUSEHOLD BEHAVIOUR 329

~ 0 ~ ~ ~ ~ ~ 0

I
~1 ~ ~1

©
Z

Z
Z

[.....

II I1[
I I ~ I
St'3
r,

©
1

.< I It I ft

I II
7'-7'7

77~77-7
~2
1 I
330 E. STERKEN

for the demand for bonds are somewhat similar to the results obtained for the
demand for shares. It should be noted, however, that some differences exist,
for instance in the magnitude and significance of the elasticity of the real yield
on domestic shares. Neither long,term time and savings deposits and mortgage
loans have a unique set of significant explanatory variables. The rate of the
speed of adjustment is high for the demand for shares and rather low for all
other assets and liabilities, including the demand for liquidity. However,
because other empirical studies find a higher rate of speed of adjustment for li-
quidity, it could be concluded that parameter estimates might be biased by the
inclusion of so many explanatory variables in each equation. The permanent
income model shows results similar to those of the current income model with
respect to the financial asset demand equations. The main differences can be
seen in the consumption function.
It should be noted that only 26 (25) of the 105 (98) parameter estimates of the
current income model (permanent income model) are significant at a 95 percent
confidence level. If Tobin's gross s u b s t i t u t i o n proposition is adopted, implying
o w n interest elasticities to be positive and cross interest elasticities to be
negative, 16 (15) of the 30 parameters appear to have a theoretically wrong sign
(the consumption equation has not been included in this comparison). Other
restrictions, like symmetry and homogeneity in interest rate elasticities, and the
[ - 1,0]-interval restriction on the adjustment parameters, which are frequently
made in other empirical investigations, are not supported by our estimation
results. Parameter restrictions like these are dependent on the underlying
statistical structure of expected interest rates and the form of the utility func-
tion (see Sterken (1987b)). Strong assumptions on statistical properties of the
expectations and the form of the utility function have to be made in order to in-
sure the homogeneity or symmetry restrictions. We do not impose these
assumptions, but estimate the model and test the significance of the parameters
instead.
The results of the OLS estimates are not completely satisfactory. First, in-
terest rate effects are in conflict with the s u b s t i t u t i o n theorem. Secondly, many
parameters are not significant at a 95 percent confidence level. Third, some
rates of speed of adjustment seem to be low (for instance, liquidity). For this
reason, we restrict our model by a p r i o r i setting parameters to zero. The param-
eters to be restricted are interest elasticities with a theoretically wrong sign
and highly insignificant parameters (as obtained in the OLS estimation). The
results obtained with this model are presented in Tables 3 and 4 for the current
and permanent income model, respectively. The parameters have been
estimated with SUR. The results of the SUR estimation are more satisfactory
than the results of the unrestricted OLS model. The current income model
yields better results than the permanent income model for the equations for the
demand for short-term bank credit, liquidity and bonds. The demand for
shares can be explained in a more satisfactory way by the permanent income
model. The demand for private consumption depends on current income,
 REAL AND F I N A N C I A L H O U S E H O L D BEHAVIOUR 331

M c~o o d c~M
I

x,O
I I oo
I

I l l I o

© I

Z
©

II

©
c.t?

,..a
=
0

II I

I
e¢3

<
I I II II

77_TT-7
"~ eq
I I
332 E. S T E R K E N

&
i~ oo ~-~ ¢',,i ~1-
¢--,1 '.~ "~,- ,.....~ o •,
<1 o ~ o o
oo

o ~eq

©
¢'-,i i-,-

cr~
<1
©
L)
z

Z
Z t¢~ O
~1" o o
< O
O

~ o~
I .o

~ t",,1 O ~D
<1 O
<

¢-q

"7 7 -7~"7
 REAL AND FINANCIAL HOUSEHOLD BEHAVIOUR 333

short-term bank credit, liquidity, shares, bonds and inflation in the current in-
come model. In the permanent income model, private consumption depends on
permanent income, shares, long-term time and savings deposits, inflation and
the real domestic yield on shares. The demand for short-term bank credit
depends on income and portfolio adjustments only. The demand for liquidity
is determined by income (permanent or current income), inflation, the dif-
ference between short-term and long-term interest rates and on the lagged stock
of bonds. The demand for shares depends on income, inflation and portfolio
adjustments. The interest rate on short-term bank credit has a negative impact
on the demand for shares. The demand for bonds depends on the difference
between the capital market interest rate and the interest rate on short-term
b a n k credit. The long-term interest rate influences the demand for long-term
deposits and mortgage loans. It is remarkable that the lagged stock of other net
assets (real estate) determines the demand for mortgage loans and shares in
both models. The speed of adjustment for the demand for shares is high in both
models. C o m p a r e d with the OLS results the speeds of adjustment for other
assets and liabilities are somewhat higher in the SUR results.
To compare our results with other empirical studies on integrated modelling
of consumption and household behaviour is a difficult task. A comparison of
the value of parameter elasticities in foreign studies is hardly possible. The only
point of comparison is the statistical fit of the model. If we compare our results
with, for instance, those obtained by Owen (1986), we observe some
similarities. The unrestricted model of Owen has a similar a m o u n t of insignifi-
cant parameters as our model. Owen has obtained a somewhat better fit, due
to the inclusion of variables for capital income gains and statistically exact real
estate variables.
Dutch empirical work on the consumption function arrives at similar values
for parameters as our estimates. The wealth effects are different in each model,
depending on the definition of the wealth variables. We find a significant im-
pact of short-term wealth components and securities in the current income
model and stocks of shares and long-term time and savings deposits in the per-
manent income model. The results obtained by Van Loo for the financial
behaviour of households indicate a high speed of adjustment of the demand for
liquidity (0.83), which is much higher than in our model (0.33). The speed of
adjustment of the demand for bonds (0.17 according to Van Loo) is similar to
our findings for the permanent income model. In the current income model a
higher speed of adjustment for the demand for bonds has been found (0.54).
Van Loo finds no lagged adjustment for the demand for mortgage loans, while
we find a coefficient of 0.50. An important difference with Van Loo is our in-
clusion of five different interest rates, while Van Loo only includes the dif-
ference between the interest rate on bonds and on mortgage loans.
334 E. STERKEN

4 CONCLUSION

A simple model of household behaviour has been presented. Households are

assumed to maximize expected utility derived from real consumption and real
asset holdings. It has been assumed that different transaction costs are con-
nected with different assets. As Purvis (1978) and Pissarides (1978) show, both
consumption and saving decisions depend on both real and financial variables.
The model proposed by Purvis is attractive, both from a theoretical and an em-
pirical point of view. First, real and financial household decisions are derived
from an integrated optimization model. Secondly, a few simplifying assump-
tions (static expectations and partial adjustment) yield a relatively simple em-
pirical model. Thirdly, the Purvis model can be estimated by means of standard
techniques. We estimated the Purvis model with two income concepts. First,
the current income approach has been applied. Secondly, the permanent in-
come concept has been used, where lagged private consumption is assumed to
reflect information on permanent income.
One of the problems with implementing the Purvis model is lack of data. In
the appendix we describe the construction of our data. Appropriate data on
capital gains and real estate are still lacking. Despite these omissions in the
data, our empirical results indicate that the Purvis model can be used to
describe Dutch household behaviour. The current income hypothesis yields
somewhat better results than the permenent income model. This could be inter-
preted as domination of short-term income and liquidity considerations over
long-term intertemporal optimization arguments in the Dutch case. Private
consumption depends on lagged wealth effects, both in the current income and
in the permanent income model. In the current income model short-term finan-
cial assets are important, while in the permanent income model only long-term
assets are significant explanatory factors. It is remarkable that the demand for
shares can be explained better by means of the permanent income model,
whereas the demand for short-term bank credit and liquidity can be explained
better by means of the current income model.
The integrated modelling approach is successful, because it explains both
consumption and savings and allows for a more detailed description of finan-
cial transmission. Our paper shows that a detailed description of the financial
transmission can explain consumption. Both current and permanent income
variables are important in describing financial household behaviour. Therefore
it seems promising to incorporate our approach into the construction of large
scale macroeconometric models.
 REAL AND F I N A N C I A L H O U S E H O L D BEHAVIOUR 335

APPENDIX
A D E S C R I P T I O N OF THE DATA

In this appendix we describe the sources and construction of the data. After
this description a full listing of data for the r a w variables is listed for the
variables that do not have a standard reference.
The most severe data problems consist in the construction of the wealth com-
ponents of households. In Sterken (1987a) a full description of the capital
market assets of households is presented. The demand for bonds by households
is the residual item on the market for bonds, where, besides the financial in-
stitutions and the foreign sector, the government is an important sector. The
demand for long-term time and savings deposits by households has been set to
90 percent of the total demand for long-term deposits by the private sector (the
figure of 90 percent has been taken over from Van Loo (1983)). The demand
for mortgage loans by the private sector has been multiplied by the percentage
of newly registered mortgage loans on houses with respect to the total of newly
registered mortgage loans. The demand for bank credit by households equals
the demand for consumer credit from private banks, which can be found in the
statistical appendices of the publications of De Nederlandsche Bank. The de-
mand for liquidity has been computed as the sum of the demand for currency,
demand deposits and short-term time and savings deposits. These components
are based on the statistical assumptions made by Van Loo (1983). Special effort
has been spent on the calculation of the asset holdings of households. The
assumption is that households are the residual demanding sector. The demand
by other financial institutions (private banks, institutional investors and invest-
ment trusts, mainly) is described in Sterken (1987a). Net demand by the foreign
sector has been computed by totaling net asset streams from Table 7 from the
statistical appendices of the Annual Reports of De Nederlandsche Bank. The
stock of assets held by the foreign sector is adjusted for the stock exchange in-
dex published by the Central Bureau of Statistics. The starting point is the
stock of 66.9 billion of guilders at the end of 1985 (see Van Nieuwkerk (1987)
and Van Erp, Hasselman and Nibbelink (1987)). The supply of shares is given
by the market value of shares, published by the Central Bureau of Statistics.
The financial assets of households are listed below.
Private consumption has been taken from the National Accounts. The same
is true for the wage and non-wage income categories and the price index of
private consumption. The interest rates on short-term bank credit, liquidity,
shares, bonds and long-term deposits are taken from Table 9 of the Annual
Reports of De Nederlandsche Bank and are described in Sterken (1986).
336 E. STERKEN

SH BON ML L TD BCR LIQ

1957 5070 9207 3171 6189 0 4443.1

1958 9040 9193 2814 7093 7 4892.9
1959 11859 8944 3084 8203 14 5173.3
1960 12988 7860 3534 9539 23 5522.8
1961 14783 7459 4215 10848 34 5874.8
1962 14425 7682 5139 12264 46 6365.7
1963 17035 7701 5885 13963 59 6973.8
1964 18168 8449 7079 15556 64 7641.3
1965 18089 9360 7624 17825 75 8300.4
1966 14246 10001 8111 19763 97 8787.2
1967 18074 9936 8356 22532 161 9444.6
1968 23986 10295 11039 25118 250 10471.1
1969 23647 10558 11294 28437 429 11303.3
1970 20840 10632 12841 32314 536 12442.9
1971 16659 11315 15291 37678 628 13867.6
1972 20925 12475 19807 42755 837 15874.8
1973 15609 13963 22881 46306 1071 17090.2
1974 9616 14475 25243 49497 1380 20203.0
1975 16934 17673 37174 60423 1714 22299.7
1976 17517 19136 45004 66452 2170 25409.3
1977 19261 21462 61653 79483 2748 27449.8
1978 19588 24091 78034 97422 3386 29174.5
1979 21521 14932 84406 108810 3736 31053.3
1980 24001 19183 86539 118594 3696 33092.3
1981 19026 23661 92704 130225 3636 33583.9
1982 21221 38876 97034 136038 3869 36367.2
1983 34061 47248 109215 137286 3980 40057.0

A l l figures are in m i l l i o n s o f guilders.

REFERENCES

Ando, A. and F. Modigliani (1963), 'The Life Cycle Hypothesis of Saving: Aggregate Implications
and Tests,' American Economic Review, 53, pp. 55-84.
Backus, D. and D.D. Purvis (1980), 'An Integrated Model of Household Flow-of-funds
Allocations,' Journal of Money, Credit and Banking, 12, pp. 400-421.
Barro, R.J. (1974), 'Are Government Bonds Net Wealth?,' Journal of Political Economy, 82, pp.
1095-1117.
Berg, P.J.C.M. van den, G.M.M. Gelauff and V.R. Okker (1987), 'The FREIA-KOMPAS Model
for the Netherlands: A Quarterly Macroeconomic Model for the Short and Medium Term,'
Occasional Paper, 39, The Hague.
Blinder, A. arid A. Deaton (1985), 'The Time Series Consumption Function Revisited,' Brookings
Papers on Economic Activity, 2:1985, pp. 465-521.
 REAL AND FINANCIAL HOUSEHOLD BEHAVIOUR 337

Brainard, W.C. and J. Tobin (1968), 'Pitfalls in Financial Model Building,' American Economic
Review, 58, pp. 99-122.
Central Planning Bureau (1985), 'FREIA-KOMPAS '85, Een kwartaalmodel voor Nederland voor
de korte en middellange termijn,' Monografie, 28, The Hague.
Clements, K.W. (1976), 'A Linear Allocation of Spending-power System: A Consumer Demand
and Portfolio Model,' Economic Record, 52, pp. 182-198.
Davidson, J., D.F. Hendry, F. Srba and S. Yeo (1978), 'Econometric Modelling of the Aggregate
Time-series Relationship between Consumers' Expenditure and Income in the United
Kingdom,' Economic Journal, 88, pp. 661-692.
Deaton, A. (1972), 'Wealth Effects on Consumption in a Modified Life-cycle Model,' Review of
Economic Studies, 39, pp. 443-453.
De Nederlandsche Bank (1984), 'MORKMON, een kwartaalmodeI voor macroeconomische
beleidsanalyse,' Monetaire Monografie6n, 2, Deventer.
Denton, F.T. (1978), 'Single Equation Estimators and Aggregation Restrictions when Equations
have the Same Set of Regressors,' Journal of Econometrics, 8, pp. 173-179.
Erp, F.A.M. van, B.H. Hasselman and A.G.H. Nibbelink (1987), 'De waarde van bet
aandelenbezit; Enkele ramingen,' Werkdocument, 21, The Hague.
Fair, R.C. (1984), Specification, Estimation, and Analysis of Macroeconometric Models,
Harvard.
Friedman, M. (1956), 'The Quantity Theory of Money: A Restatement,' in: M. Friedman (ed.),
Studies in the Quantity Theory of Money, University of Chicago Press.
Haan, J. de and D. Zelhorst (1987), 'The Empirical Evidence on the Ricardian Equivalence
Hypothesis,' Kredit und Kapital, forthcoming.
Hall, R.E. (1978), 'Stochastic Implications of the Life Cycle-Permanent Income Hypothesis:
Theory and Evidence,' Journal of Political Economy, 86, pp. 971-987.
Knoester, A. (1980), Over geld en economische politiek, Leiden.
Kohli, U.R. and W.J. McKibbin (1982), 'Are Government Deficits the Prime Cause of Inflation?,'
Journal of Policy Modelling, 4, pp. 279-309.
Kuipers, S.K., G.H. Kuper and E. Sterken (1987), 'The CCSO Annual Model of the Dutch
Economy,' Paper presented at The Second Annual Congress of the European Economic
Association and The 1987 European Meeting of the Econometric Society, Copenhagen, 1987.
Laidler, D.E.W. (1984), 'The Buffer Stock Notion in Monetary Economics,' Economic Journal,
94 (Supplement), pp. 1-28.
Loo, P.D. van (1983), A Sectoral Analysis of the Dutch Financial System, Leiden.
Lucas, R.E. (1976), 'Econometric Policy Evaluation: A Critique,' in: K. Brunner and A.H.
Meltzer (eds.), The Phillips Curve and Labor Markets, Carnegie-Rochester Conference Series
on Public Policy, 1, pp. 19-46.
Markowitz, H.M. (1959), Portfolio Selection: Efficient Diversification of Investments, New York.
Metzler, L.A. (1951), 'Wealth, Saving and the Rate of Interest,' Journal of PoliticalEconomy, 59,
pp. 96-116.
Mishan, E.J. (1958), 'A Fallacy in the Interpretation of the Cash Balance Effect,' Economica, 25,
pp. 106-118.
Modigliani, F. and R.E. Brumberg (1954), 'Utility Analysis and the Consumption Function: An
Interpretation of Cross-section Data,' in: K.K. Kurihara (ed.), Post-Keynesian Economics, New
Brunswick.
Molana, H. (1987), 'Role of Personal Wealth, Relative Price and Interest Rate in the Aggregate
Consumption Function: British Evidence,' Discussion paper 8704, Glasgow.
Mnkherjee, R. and E. Zabel (1974), 'Consumption and Portfolio Choices with Transaction Costs,'
338 E. STERKEN

in: M. Balch, D. McFadden and S. Wu (eds.), Essays on Economic Behaviour under

Uncertainty, Amsterdam.
Nieuwkerk, M. van (1987), 'Kapitaalexport,' Economisch-Statistische Berichten, 72, pp. 352-358.
Okker, V.R. and R.J.A. den Haan (1987), 'De consumptiefunctie in FK'85,'
Onderzoeksmemorandum, 29, The Hague.
Owen, D. (1986), Money, Wealth and Expenditure, Cambridge.
Parkin, M. (1970), 'Discount House Portfolio and Debt Selection,' Review o f Economic Studies,
27, pp. 469-497.
Parkin, M., R.J. Cooper, J.F. Henderson and M.K. Danes (1975), 'An Integrated Model of
Consumption, Investment and Portfolio Decisions,' Reserve Bank o f Australia, Papers in
Monetary Economics, Volume II.
Patinkin, D. (1965), Money, Interest, and Prices: An Integration o f Monetary and Value Theory,
New York.
Pigou, A.C. (1943), 'The Classical Stationary State,' Economic Journal, 53, pp. 343-351.
Pissarides, C.A. (1978), 'Liquidity Considerations in the Theory of Consumption,' Quarterly
Journal o f Economics, 92, pp. 279-296.
Purvis, D.D. (1975), 'Portfolio and Consumption Decisions: Towards A Model of the
Transmission Process,' Reserve Bank o f A ustralia, Papers in Monetary Economics, Volume II.
Purvis, D.D. (1978), 'Dynamic Models of Portfolio Behaviour: More on Pitfalls in Financial
Model Building,' American Economic Review, 68, pp. 403-409.
Smith, G. (1978), 'Dynamic Models of Portfolio Behavior: Comment on Purvis,' American
Economic Review, 68, pp. 410-416.
Sterken, E. (1986), 'Een monetair model voor NederIand,' CCSO-reeks, 1, Groningen.
Sterken, E. (1987a), 'Een statistiek van de Nederlandse kapitaalmarkt,' CCSO-reeks, 4,
Groningen.
Sterken, E. (1987b), 'Should Portfolio Model Parameters be Restricted or Not?,' Mimeo,
Groningen.
Sterken, E. (1987c), 'Macroeconometric Models of the Financial System,' Mimeo, Groningen.
Thomas, R.L. (1981), 'Wealth and Aggregate Consumption,' The Manchester School, 49, pp.
129-152.
Voorhoeve, W. (1986), Short-term Economic Forecasting in Practice: The GRECON Case,
Groningen.
Zellner, A. (1962), 'An Efficient Method of Estimating Seemingly Unrelated Regression and Tests
for Aggregation Bias,' Journal o f the American Statistical Association, 57, pp. 348-368.

Summary

A MODEL OF REAL AND FINANCIAL HOUSEHOLD BEHAVIOUR

This paper shows that the traditional specification of the consumption function in Dutch macro-
econometric models is unsatisfactory. In the traditional approach the fact that the consumption
decision is taken simultaneously with financial decisions has been ignored. If both the consump-
tion function and the asset demand equations are modelled simultaneously, then in the Dutch case
financial stocks have influence on private consumption. Also it has been shown that the income
concept is important in describing household behaviour. In a current income model short-term
financial considerations are important in the consumption decision. In a permanent income model
long-term financial considerations influence private consumption.