7

CHAPTER 2

CONSUMPTION THEORY

Much of the most insightful empirical work in macroeconomics

over the past twenty years has been concerned with consumption ...
(Romer 1996:309)

2.1 INTRODUCTION

This chapter commences with an account of the relevant economic theory of consumption
expenditure, to support the theoretical derivation for a model of private consumption.
Reference to a number of studies on private consumption expenditure is made, paying
particular attention to the effect of aspects like wealth, prices, liquidity constraints and
expectations on consumption. The specification of consumption functions in some well­
known international macro-models are compared to conclude the chapter. In Chapter 5, the
South African situation is evaluated against the backdrop of the above analysis when an
empirical estimation of private consumption expenditure functions is presented.

2.2 THE KEYNESIAN VS THE NEW CLASSICAL APPROACH

A Keynesian economist thinks about consumption theory in terms of private domestic

behavioural relations underlying the IS schedule. The effects of income and interest rates on
consumption would be stressed and adding the LM schedule would complete the model. A
new classical economist on the other hand, would specify a production function and then
would allow prices and interest rates to adjust to clear all markets (Abel 1990:726).

A number of different theories of consumption have been developed in response to the

deficiencies in the simple Keynesian consumption function. Since the 1950s, economic
models of consumption behaviour have explicitly recognised that in making consumption
decisions, consumers consider their lifetime resources rather than simply their current
income. Both the life-cycle model of Modigliani and Brumberg (1954) and Ando and
 8

Modigliani (1963) and the permanent-income model of Friedman (1957) are based on the
notion that consumers prefer smooth streams of consumption over time. Hall and Taylor
(1993:278) refer to these theories jointly as the forward-looking theory of consumption.

2.3 THE UNDERLYING CHOICE-THEORETIC FRAMEWORK

The life-cycle and permanent-income hypotheses, which are the major theories of
consumption behaviour, both relate consumption to lifetime income.

The underlying choice-theoretic framework emphasises that a consumer has an

intertemporal utility function that depends on consumption in every period of life. A first
principle of microeconomics is that consumers structure their consumption plans to

maximise their satisfaction or utility. According to Abel (1990:729), the consumer

maximises utility subject to a single lifetime budget constraint. There is no static or period­
by-period, budget constraint that requires consumption in a period to equal the income in
that period. It is access to capital markets that allows consumers to choose a sequence of
consumption expenditures over time, which will be smoother than the sequence of income.
Access to perfect capital markets would allow consumers to borrow or lend as much as they
would like, given an exogenous interest rate.

Varian (1993: 179-92) examines consumer behaviour by considering the choices involved in
saving and consumption over time - the consumer's intertemporal choices. The shape of
the consumer's indifference curves would indicate his tastes for consumption at different
times. Well-behaved preferences, where the consumer is willing to substitute some present
consumption for future consumption, would be the most reasonable. How much he is
willing to substitute, depends on his subjective pattern of consumption. Convexity of
preferences is very natural in this context, since it means that the consumer would rather
have an 'average' amount of consumption each period than a lot today and nothing
tomorrow or vice versa. The consumer's optimal combination of consumption in any two
periods, say, is where the budget line is tangent to an indifference curve.
 9

Thomas (1993:253) makes it clear that "there is no way III which we can derive the
absolute income hypothesis (the Keynesian consumption function) from a traditional
microeconomic analysis". The feasible approach towards an analysis of consumer
behaviour would therefore have to be conducted within a new classical framework.

2.4 THE BASIC FORWARD~LOOKING THEORY OF CONSUMPTION

The basic notion that individual consumers are forward-looking decision-makers is

embodied in the life-cycle and permanent income theories. The life-cycle theory derives
its name from its emphasis on a family looking ahead over its entire lifetime. The
permanent income theory is named for its distinction between permanent income, which a
household expects to be long-lasting, and transitory income, which is expected to quickly
disappear. In practice the theories differ primarily in the types of equations used to express
the basic idea of forward-looking consumers and how they are implemented empirically.

2.4.1 The lif~cycle hypothesis

The analysis of section 2.3 corresponds with the work of Fisher (1907). It was later
adopted and generalised by Modigliani and Brumberg (1954) in their life-cycle hypothesis.
They assumed that a household plans its lifetime consumption pattern so as to maximise the
total utility it obtains from consumption during its lifetime.

Assuming that households do not plan to leave assets to their heirs, algebraically a
household of age T maximises a utility function of the form

(2.1)

where C i (i=T,T+ 1,T+2,...L) is planned consumption at age i and L is the household's

expected age at 'death'. Since the household plans to exactly exhaust its resources during
its lifetime, (2.1) is maximised subject to the lifetime budget constraint
 10

N ye L C
A T - 1 + YT + I
i=T+l (1 + r)
I i-T =I
i=T (1 + r)
I i-T • (2.2)

where

AT_I = non-human wealth (i.e. physical and financial assets) carried over
from the household's (T-1)th year
YT = household's earned or non-property income at age T
ye household's expected non-property income at age i
I

r = the interest rate and

N household's age at retirement.

Modigliani and Brumberg adopt the simplifying assumption that the utility function (2_1) is
homothetic l . This implies that the planned current consumption is given by

(2.3)

where W T is the household's total expected lifetime resources at age T. That is, it is the
sum of all the terms on the left-hand side of the budget constraint (2.2)
N ye
WT = A T - 1 + YT + I
i=T+l (1 + r)
I i-T • (2.4)

Similarly, planned consumption in future years is given by

i = T + 1, T + 2,00', L . (2.5)

The YiS in equations (2.3) and (2.5) will depend on the rate of interest and the household's
tastes and preferences. However, they will also depend on the age of the household.
Because resources are to be exhausted during its lifetime, the nearer the household is to

I Graphically, this would mean that the slopes of the indifference curves are the same along any straight line
drawn through the origin. Thus for a given rate of interest, as Wr increases and the budget line shifted
outward parallel to itself, the optimal ratio Cr/C r + 1 remains unchanged regardless of the magnitude of Wr .
The ratio Cr/C r + 1 will however depend on the tastes of the consumer, as represented by the precise form of
his indifference map, and on the rate of interest.
 11

'death', the larger the proportion of its resources it plans to expend during any given year.
,
The important aspect of (2.3) and (2.5) is that the YiS are independent of the magnitude of
WT' Thus the household keeps the ratios of its planned consumption expenditures on any
two future years unchanged no matter what the size of its lifetime resources.

Ando and Modigliani (1963) adopted equation (2.3) for empirical estimation from
aggregate time series data. Problems in estimating the expected non-property income for
consumers meant that their final equation simply involved regressing aggregate
consumption C t , on aggregate current non-property income YI' and the aggregate wealth of
consumers At • l . Their most important finding was that, for annual US data for 1929-59,
At- l was a significant determinant of Ct. The marginal propensity to consume (MPC) out of
net worth was estimated to be in the region 0.07 to 0.10. Their aggregate consumption
function was therefore of the form C t nAt-! + pYt •

2.4.2 The permanent-income hypothesis

Although the permanent-income hypothesis shares many similarities with the life-cycle
hypothesis, the former was developed independently and found its first definite form in the
work of Friedman.

Friedman generalizes the two period case to an 'indefinitely long horizon' rather than to a
remaining life-span. Friedman also introduces the concepts of present period planned of
permanent consumption, CP, and permanent income, Yp.

According to Friedman, permanent consumption would be a function of present period total

wealth W, and the rate of interest:

CP = q(W,r). (2.6)

Total wealth in the Friedman formulation is the discounted sum of all future receipts,
including income from non-human assets. Wealth in period t would be
 12

YI I Yt 2 Y+
Y +-+-+ + + I 3 3 + ... (2.7)
I 1 + r (1 + r) 2 (1 + r)

where YI is total expected receipts in period t.

Friedman also makes use of the simplifying assumption that the consumer's utility function
is homothetic, and equation (2.6) then becomes

(2.8)

where the factor of proportionality q, is dependent on the consumers' tastes and on the rate
of interest.

Theoretically, permanent income is defined as the maximum amount a consumer could

consume while maintaining his wealth intact. It is in fact the return on wealth, i.e. yP
= rW. Equation (2.8) can thus be rewritten as

(2.9)

where q = rk.

The quantity k in equation (2.9) depends on the tastes of the household and on the rate of
interest. However, under conditions of uncertainty, Friedman also introduces an additional
motive for saving - the need to accumulate a reserve of wealth for contingencies. Since
human wealth constitutes a less satisfactory reserve than non-human wealth, the proportion
of permanent income consumed, k, is made to depend, also, on the proportion of total
wealth which is held as non-human wealth. For a given rate of interest, this ratio is
directly proportional to the ratio of non-human wealth to permanent income for which
Friedman uses the symbol w. Thus we have

(2.10)

where u is a portmanteau variable reflecting consumers' tastes.

 13

When attempts are made to relate the permanent income hypothesis to actual data, obvious
problems are faced. Current or 'measured' income is clearly different from the theoretical
concept of permanent income and even if adequate 'flow of services' data on current or
'measured' consumption were available, this would still differ from planned or permanent
consumption. According to Friedman, measured income, Y, consists of two components ­
a permanent component YP, and a transitory component yt. Measured consumption C, is
similarly divided into permanent consumption, C P, and transitory consumption C. Thus we
have

Y =y P + y t
and (2.11)

The empirical definition of yP is that it is the normal or expected unfortuitous income of the
consumer. This roughly corresponds to the theoretical definition but is purposely left
vague by Friedman since "the precise line to be drawn between permanent and transitory
components is best left to the data themselves, to be whatever seems to correspond to
consumer behaviour" (Friedman 1957:23). In practice, permanent income would be
whatever quantity the consumer regarded as determining his planned consumption. The
transitory component of income is to be regarded that which arises from accidental or
chance occurrences, while permanent and transitory consumption may be interpreted as
planned and 'unplanned' consumption respectively. Based on Friedman's assumption that
yt is uncorrelated with C, any unforeseen increment in income does not result in unplanned
consumption. This is obviously open to debate. Friedman however justifies this premise
by pointing out that even if income is other than expected, the consumer would tend to stick
to his consumption plan, but adjust his asset holdings.

2.4.3 Comparison of the life-cycle and permanent-income hypotheses

From the above analysis, it is clear that there are basic similarities between the life-cycle
and permanent-income hypotheses. According to the life-cycle hypothesis, a change in
current income Y T. will influence current consumption CT, only to the extent that it changes
WT' the household's expected lifetime resources. Normally, changes in YT' unless they

lead to revisions in expectations concerning future income, i.e. to changes in the yjes , can

be expected to have little influence on current consumption unless the household is near
'death'. Similarly, in Friedman's model an increase in current income influences current
 14

consumption only to the extent that it changes Wand, hence, permanent income.
Furthermore, in both cases, the 'proportionality postulate' is not vital. In the case of the
life-cycle hypothesis, current consumption would remain a function of total lifetime
resources, although the relationship would no longer be one of strict proportionality. In the
permanent income hypothesis, cP remains a function of Wand hence, of permanent income
rather than current income.

There are also relatively minor, but clear differences between the models. The
annuitisation of total wealth means that the stock of non-human assets does not appear
explicitly in Friedman's consumption function. However, Friedman does distinguish
between the influences of human and non-human wealth on consumption. The factor of
proportionality, k, i.e. average propensity to consume (APC) out of permanent income, is
dependent on the ratio of the two. Non-human wealth - physical and financial assets ­
appears explicitly in the formulation of wealth in the life-cycle model, equation (2.4).
Finally, in the life-cycle model, the household merely looks ahead towards the end of its
life. Friedman's annuitisation of total wealth suggests that his household has an infinite life
or at any rate attaches as much importance to the consumption of its heirs as its own. One
of the major implications of the life-cycle hypothesis is that saving is done by consumers
when they are working to provide for consumption when they are retired. The implication
will not be captured in a model in which the consumer lives and earns income forever.

2.5 A REVIEW OF RESEARCH ON AGGREGATE CONSUMPTION

EXPENDITURE

This section highlights the development in the empirical application of the basic forward­
looking theories.

2.5.1 Early time series estimations under the permanent-income hypothesis

Friedman (1957) adopted a distributed lag formulation with geometrically declining

weights, when suggesting that, for aggregate time series data, permanent income in period t
may be estimated by
 15

O<A<l. (2.12)

The argument is that consumers will assign the largest weight to their current income in
assessing their permanent income and successively declining weights to past income.
According to Thomas (1993:263) this formulation stresses the 'expected' nature of
permanent income. Equation (2.12) implies that permanent income is determined by an
adaptive expectations hypothesis. In this case, it is

(2.13)

Thus differences between permanent and measured income lead to an adjustment in the
perceived level of permanent income. The extent of the adjustment depends on the size of

A.

Friedman's time series work was, however, based on versions of equation (2.12). He
computed various time series for yPt using a different value for A in each case, truncating
after 16 terms. Using annual real per capita US data for 1905-51, he ran regressions of the

form C = &. + Pyt

t
for each yPt series and chose the value of A which provided the closest

fit. The highest R2 was obtained for A=0.33. For this equation, the intercept term was

p
insignificant with a very low t-ratio and the estimate of k was =0.88. This supported the
hypothesis that the relationship between c P and yP was one of proportionality. The value
A

obtained for p was also close to the observed APC for the period.

It is commonly known that Friedman's estimating procedure can be simplified by

application of the Koyck transformation. Evans (1969) estimated this version, using US
annual data for 1929-62. He obtained the following result:

C t = 0.280Yt + 0.676C H
(2.14)
(0.041) (0.052)
 16

Interpreting (2.14) in terms of Friedman's hypothesis yields A= 1-0.676 0.324 and a value
for k=0.2801A=0.86. Both these estimates were very close to those obtained by
Friedman.

Since the data series are non-stationary, the disturbance term in (2.14) is almost certain to
be autocorre1ated, the combination of autocorrelated disturbance terms and a lagged
dependent variable means the OLS estimators will be biased and inconsistent. This
problem is addressed in section 4.2.

2.5.2 The effect of prices and inflation on consumption - the influential study by
Davidson, Hendry, Srba and Yeo (1978): an error correction approach

The mid-1970s saw a breakdown of the form of consumption function discussed in the
previous section. These functions implied a constant savings ratio and could not explain its
rise during the 1970s. Davidson, Hendry, Sarb and Yeo (1978) found econometric support
for the presence of an inflation term in the equation and also concluded that the rise in
inflation in the 1970s was the factor behind the rising savings ratio. Their study was also
highly influential in introducing the error correction approach to econometrics - the first
study that attempted to deal with non-stationarity in the data and the spurious correlation
problem. The study of Davidson et al. also represented the first explicit use in this area of
the Hendry type general to specific methodology.

Davidson et al. concentrated on three previous studies of non-durable consumption and

personal disposable income for UK data - that of Hendry (1974), Ball et al. (1975) and
Wall et al. (1975) - and tried to explain why these previous investigators came to such
widely different conclusions.

Their analysis proceeded by noting seven potential explanations for the main differences
between the three studies, namely the choice of (i) data series, (ii) methods of seasonal
adjustment, (iii) other data transformations, (iv) functional forms, (v) lag structures, (vi)
diagnostic statistics and (vii) estimation methods. Even after 'standardising' the models on
a common basis for (i)-(iv), the models still seemed to lead to different conclusions.
Remaining possible reasons could only be (v)-(vii).
 17

The above standardisation by sample period, etc., enabled Davidson et al. to 'nest' the
three competing hypotheses as special cases of a general hypothesis or estimation equation.
This enabled them to test, on purely statistical grounds, which provided the best description
of the relationship between income and consumption. The best model appeared to be that
of Wall et al., which was of the form

(2.15)

with LlC[ and LlYt the quarterly changes in consumption and income.

Davidson et al. pointed out that this statistically preferred equation (2.15) had some
unacceptable economic properties: first, the equation had no static equilibrium solution.
Also, the equation implied that the adjustment of consumption to any change in income is
complete after just two quarters, and moreover, was apparently independent of any
disequilibrium in the previous levels of the variables C[ and Yt. The model thus accounted
only for short-run behaviour. Davidson et al. (op. cit. :686) resolved the last of the above
problems by adopting an error correction approach and presented the following error
correction model in log-linear form (Ll 4 denotes the four period or annual difference for
quarterly data):

Ll 4C[ = +0.47 1 Ll4 Yt - 0.21LlILl4 Y[ - 0.1O(C t _ 4

(0.04) (0.05) (0.02)
(2.16)
+ 0.0ILl4D~ 0.13Ll 4 Pt - 0.28Ll 1Ll 4 Pt
(0.003) (0.07) (0.15)

Davidson et ai. invoked the Deaton hypothesis 2 by adding price variables to the equation in
order to rectify the consistent overprediction of consumption during the period 1971-75,
when there was a steady increase in the UK propensity to save.

2 Deaton (1978) argued that it is accelerating inflation that reduces consumer expenditure. Davidsonet al.
followed him by including first and second differenced forms of the price variable in their specifications of
UK consumption.
 18

Hendry and Ungern-Sternberg (1980), indicated that the importance of the variables ~4Pt

and (C t_r Yt-4) can be explained in terms of a wealth effect. They began by noting that
equations such as (2.16) have a major flaw as a complete account of the dynamic behaviour
of flow variables. Since Ct and Yt are rarely equal, this means that some latent asset stock
must be changing and changes in this stock may itself affect the change in Ct. This, of
course, is merely another way of saying that wealth effects may influence consumption.

Wealth, or 'cumulative saving' effects, were introduced into their model by assuming that
consumers seek to maintain constant ratios not only between consumption and income, but
also between the latent asset stock and income. They proposed an 'equilibrium
relationship', At BYt , where At is the latent asset stock or wealth variable.

The disequilibrium 'costs' or 'losses' are incurred if Ct or At differs from their equilibrium
values. The consumer is assumed to minimise a quadratic function of these losses subject
to a budget constraint. This eventually leads to an equation that is a generalisation of
(2.16), since it will contain two equilibrium errors (C t_l - Yt- J) and (A t _J- Yt- J). The latter
reflects the extent of the previous period disequilibrium between asset stock and income.

Drobny and Hall (1989) established that the conventional variables used in the Davidson et
al. specification do in fact not constitute a cointegrating vector. It also failed to perform
well over the first five years of the 1980s. They contributed the failure partly to relative
income effects within the overall distribution of incomes. If, as empirical results suggest,
the propensity to consume of higher-rate income tax payers is different from that of
standard-rate taxpayers, large changes in tax differentials (such as those of 1979 in the
United Kingdom) may have large effects on aggregate expenditure on non-durable goods.
They therefore introduced a tax rate differential variable and, together with disposable
income and the wealth-to-income ratio, they found a cointegrating vector to exist. They
further added a dummy variable for announced VAT changes to the long-run equilibrium
equation.

Their preferred specification for the error correction model, based on the two-step Engle­
Granger estimation procedure was (op. cit. :459):
 19

I1C t = 0.008 + 0.3311Y t

0.121111Y t +0.007D t -0.20111\ -0.14I1C t _ 1 O.l 9Z t-I
(0.002) (0.07) (0.05) (0.002) (0.05) (0.092) (0.073)
(2.17)

with Zt-l the lagged residual from the co integrating regression.

Prior to the study by Davidson et al., Branson and Klevorick (1969) added a price variable
to the simple life-cycle hypothesis to test for the effect of 'money illusion' in US
consumption. Economic theory suggests that the price coefficient in the consumption
function should be equal to zero. A rise in the price level, with real income and real wealth
remaining constant, must imply an equiproportionate rise in money income and money
wealth and hence should not change consumption expenditure. If the price coefficient is
positive, then this would imply that consumers are exhibiting the phenomenon commonly
known as 'money illusion'. A positive coefficient means that a rise in P l , with Yl and W l
constant, results in a rise in consumption. Consumers must therefore be treating the
equiproportionate rise in money income and money wealth as if it were a rise in real
income and real wealth and 'not noticing' the rise in prices. A negative coefficient,
however, implies some sort of reverse illusion. In the face of equiproportionate changes in
the price level, money income and money wealth, consumers reduce consumption. This
suggests that they believe that their real income and wealth have fallen when in fact they
have not. In some way they are 'noticing' the rise in prices, but not the equiproportionate
rises in money income and money wealth. Branson and Kleverick found their equivalent of
the price coefficient to be significantly greater than zero and concluded that a significant
degree of money illusion existed in the US consumption function.

The rapid inflation first experienced by many Western economies during the 1970s led to a
number of attempts to establish links between the inflation rate, rather than the price level,
and consumption. The economic rationale for the inclusion of inflation was as a proxy for
the inflation loss on liquid assets, but Deaton (1987) suggested an alternative explanation in
that variable inflation created uncertainty, and hence a decision to postpone consumption.
Greater uncertainty regarding future real income during times of high inflation will lead to
greater precautionary savings. Subsequent work by Hendry and Ungern-Sternberg (1981)
 20

specifically included a liquid assets term and argued that the inflation loss on this variable
should be deducted from the income variable.

2.5.3 Wealth effects

The second main challenge to the consumption function came in the mid-1980s, when this
time models of the type discussed in section 2.5.2 failed to explain the sharp fall in the
savings-ratio in the United Kingdom. According to Bai and Whitley (1997:70), modellers
reacted to the impact of forecast failures in the late 1980s by assuming that financial
deregulation had increased liquidity of physical assets held by the personal sector.
Deregulation resulted in the personal sector increasing its debt-to-income ratio. At the
same time this boost to demand stimulated a rise in asset prices (especially house prices)
but the ratio of debt to assets rose in spite of asset appreciation. Physical wealth began to
appear in consumption equations. The equations then began to appear more like the life­
cycle models of consumption rather than the original Keynesian form.

Although researchers agreed all along that wealth is an important determinant of

consumption, the reason why earlier studies did not include explicit wealth variables was
basically a lack of data on total non-human wealth. In instances where researchers
attempted to include some measure of non-human wealth, they had to rely either on the
liquid asset component of such wealth, or construct their own series from past data on
saving. Examples include Zellner et al. (1965), and Townsend (1976) who found liquid
asset variables to be important determinants of consumption, and Stone (1973) who

constructed wealth data for the United Kingdom using the relationship W[ = Wo + I:=t Sj ,

where Wo refers to wealth in some 'bench-mark' year for which data is available and Sj to
saving in year i.

The study of Ball and Drake (1964) was the first study to explicitly pay attention to the
precise type of consumer behaviour that may cause wealth variables to be important. In the
Ball-Drake model, individuals are assumed to be short-sighted in the face of uncertainty,
and their basic motive for saving is a broad precautionary one. The arguments in the
consumer's utility function are current real consumption and real non-human wealth. That
 21

is U t U(C I' WI) , where Wt is wealth at the end of the (short planning) period over which
utility is maximised and Ct is consumption during that period. That is, the consumer does
not ignore the future but safeguards against its uncertainties by accumulating wealth. The
more wealth he accumulates, the more secure he feels and, given his rate of consumption,
the more utility he derives. The future is therefore allowed for without making the possibly
unrealistic assumption of intertemporal utility maximisation required by the life-cycle and
the permanent-income hypotheses.

However, these models of consumption behaviour in turn failed to explain the rise in the
savings-ratio that occurred in the United Kingdom in 1990-91 and research began to focus
on the forward-looking behaviour of consumers, going back to the original life-cycle model
adjusted for forward-looking behaviour by Hall (1978) and subsequently adjusted by
Hayashi (1982) to deal with liquidity constraints.

2.5.4 Liquidity constraints and credit constraints

The assumption in the basic forward-looking theory is that consumers have access to
perfect capital markets and can borrow or lend at an exogenous rate of interest, while in
reality, a substantial fraction of consumers is unable to consume as much as predicted by
the forward-looking theory because they are unable to finance their desired level of
consumption. These consumers are said to be liquidity constrained if they are unable to
maintain expenditure by liquidating financial assets, or credit constrained if they are unable
to borrow as much as they would like to at the prevailing interest rate.

The importance of liquidity constraints from the viewpoint of macroeconomics is that the
relation between consumption and contemporaneous income is generally different for
liquidity constrained consumers than it is for consumers who do not face binding liquidity
constraints. The literature on liquidity constraints on consumption suggests that aggregate
consumption responds to changes in both permanent and current income. This is equivalent
to distinguishing between forward-looking consumers (the wealth constrained), who smooth
their consumption according to the life-cycle hypothesis and backward-looking or credit­
constrained consumers (liquidity constrained), whose consumption is restricted by their
current incomes (Bai and Whitley 1997: 73).
 22

Liquidity constraints were first introduced by Hall and Mishkin (1982) and tested by
Hayashi (1982) and Campbell and Mankiw (1991). The basic notion is that many
households have small initial values of assets and are unwilling or unable to borrow.
Hence their current consumption is constrained by current income. Aggregate consumption
is then given by the behaviour of both unconstrained and constrained households:

(2.18)

The importance of liquidity constraints m consumption behaviour has been widely

recognised; important empirical implementations include Campbell and Mankiw (1989),
Abel (1990), Darby and Ireland (1993) and Campbell and Deaton (1989). They assume that
some consumers can borrow or lend at an exogenous rate of interest while others, who
would like to increase their current borrowing in order to increase current consumption, are
unable to do so because of liquidity constraints.

Bai and Whitley (1997) utilises this insight as foundation for a model that behaves quite
differently in the two groups of consumers.

The consumption of forward-looking consumers in their model is given by

(2.19)

where CIt is the consumption of forward-looking consumers, Aft is net financial and
physical wealth (real non-human wealth) at time t, and Hft is human wealth at time t for
forward-looking consumers.

The credit-constrained consumers' consumption is equal to their non-property income:

(2.20)

so that aggregate consumption is a linear function of total wealth of the two types of
consumers:

(2.21)
 23

Flavin (1985) and Muelbauer and Bover (1986), also concluded that the addition of a
liquidity constraint does improve the explanation of the data. The liquidity constraint
contains a shadow price which operates as an interest rate, so that a consumer faced with a
liquidity constraint will behave as if faced with a higher interest rate. Current consumption
therefore becomes more expensive and consumers substitute future consumption.

2.5.5 The role of expectations in consumption

During the twenty years after the Second World War, the adaptive expectations hypothesis
enjoyed considerable popularity as a simple and apparently sensible model of how
economic agents form expectations. Under the permanent income hypothesis, permanent
income can be regarded as determined by an adaptive expectations process, i.e. through
estimation of equations such as (2.13). The deficiencies of the adaptive expectations
hypothesis, however, gradually became more apparent. It was pointed out that to model
expectations of a variable adaptively, implies irrational behaviour on the part of economic
agents if that variable grows at a constant rate over time. The agents' forecast error will
consistently be positive. If the agent continues forming expectations adaptively under these
circumstances, he will soon realise he is consistently underpredicting the variable under
consideratioin and a rational individual will start taking this into account when he makes his
predictions.

The 1980s saw the introduction of rational expectations (also termed model-consistent
expectations) into a number of forecasting models. In simple terms, rational expectations
implies that agents have access to all relevant information and make the best possible use of
it when forming expectations regarding any variable. Relevant information, of course,
includes knowledge of government policy aims. Hall and Garratt (1992b) point out that the
economic evidence for the importance of expectations is almost uniformly based on the
weak form of rational expectations (i.e. that agents do not make systematic mistakes),
rather than the strong form (that they use a particular model to form their expectations). It
is clearly a significant step to go from the statement that agents are 'on average' correct in
their expectations to the much stronger one that they use a particular model which they
believe completely. Hall and Garratt learnt from practical implementation of rational
expectations within a forecasting context, that the presence of rational expectations in
 24

models tends to cause jumps in the initial period value of the variable. This occurs because
agents anticipate the future and therefore make all the required adjustments in the current
period.

Learning is a natural assumption, which overcomes the complications of the rational

expectations hypothesis, and also avoids the need for the unrealistic information assumption
of the strong rational expectations hypothesis. It is based on the notion that expectations
are formed by intelligent agents who are not fully informed but are able to learn from their
environment as time progresses. A specific expectations rule is therefore set up and the
parameters of this rule are allowed to change in a way which represents the process of the
agents learning about the economy.

With regard to modelling consumption expenditure, price expectations may hypothetically

influence current consumption. Agents have a rule for forming expectations about the next
period's consumer prices, which might involve such factors as past values of, say, interest
rates, the exchange rate as well as lagged values of the consumer price index. They use
this rule with their best estimate of the parameters attached to the rule to form an
expectation of consumer prices in the next period and this affects consumption in the
current period. During the next period, agents are able to observe the actual outcome for
consumer prices and to compare it with their original expectation, which then gives them a
direct measure of their expectation error. Given this error, the parameters of the
expectations rule may then be revised to produce a better forecast for the following period.
By repeating this procedure over a number of periods, the agents come to learn a set of
parameters which will give them the best forecast of consumer prices; thus they learn about
the structure of the model.

In practice, a set of 'rolling' regressions may be performed, using OLS, each period adding
the latest expectation error to the data set, or alternatively, a more sophisticated mechanism
based on the Kalman filter may be used for the variable parameter estimation of the
expectations rule. Price expectations formation and the implementation of the price
expectations variable in the consumption function are addressed in Chapter 3 and Chapter 5
respectively.
 25

2.5.6 Comparing specifications in different macro-models

A brief overview of the specification of consumption functions for some of the main UK
macroeconometric models (Whitley 1994:34) can be found in Table 2.1. In addition, the
approach towards addressing the notion of expectations is also reported. Inevitably,
focussing on models at one point in time runs the risk that the description might be outdated
at present. However, it may be believed that while details may change, the underlying
features remain very much the same.

According to Whitley, many accounts of consumption equations written by the modellers

themselves describe the consumption equation as based on the life-cycle hypothesis. This is
a rather vague description in practice, since the empirical models include explana£ions of
consumption by income alone, a combination of real incomes and wealth, and wealth alone.
From Table 2.1 it is evident that there seems to be consensus on the inclusion of a
combination of wealth and income, with wealth usually defined as the sum of financial and
physical assets, in the determination of consumption spending. In addition, either real or
nominal interest rates are included. This suggests a life-cycle approach modified by
liquidity constraints.

2.6 CONCLUSION

This chapter reviewed the literature on consumption theory, with specific reference to the
development of the forward-looking theory of consumption.

The most common approach in the 1970s was to treat consumers as constrained in their
purchase decisions by current income. The mid-1970s saw the breakdown of this form of
equation in the face of rising inflation world-wide and a rise in the savings ratio,
specifically in the United Kingdom. This led to the introduction of price or inflation
variables to consumption functions. Researchers also became aware of the problem of non­
 26

Table 2.1 A comparison of specifications of consumption functions

Model I Consumption I Expectations

London Business School Real income , housing and i Exchange rate either
(LBS) financial wealth, nominal forward looking or
interest rates, learning approach
demographic factors
National Institute for Non-credit-financed Rational in exchange rate,
Economic and Social spending a function of prices, wages, stock
Research (NIESR) financial wealth, real building
! income and real interest
rates
Liverpool University Private sector demand Rational
(LPL) determined by wealth and
long-term interest rates
I HM Treasury (HMT) Total wealth, inflation- Adaptive or implicit
adjusted income, real
interest rates
Bank of England (BE) Financial wealth, housing Adaptive or implicit
wealth, real income,
mortgage equity
withdrawal
Oxford Economic Income, total wealth, Adaptive or implicit
Forecast (OEF) nominal interest rates,
relative price
durables/non-durables

Source: Whitley (1994:235).

 27

stationary data and spurious regressions and the error correction approach was adopted in
the late 1970s. The second main challenge to consumption specifications came in the mid­
1980s when models failed to explain the sharp fall in the savings ratio in the United
Kingdom. This gave rise to the explicit introduction of wealth variables in consumption
equations. When models of consumption behaviour again failed to explain the rise in the
savings ratio that os:;curred in 1990-91 in the United Kingdom, the forward-looking theory
of consumption was adapted to deal with liquidity constraints. The most recent
development in consumption specification has been to introduce a price expectations
variable into the behavioural equation. This has been done in this study and is reported in
Chapter 5.