Alternate Models of Consumption Behaviour. A Cross-National Analysis : The U.S., U.K. and Israël ; n°1 ; vol.26, pg 123-134

Warren L.Young

Description

Revue économique - Année 1975 - Volume 26 - Numéro 1 - Pages 123-134
12 pages

Sujets

• Sociologie, société et politique

Informations

• Publié par: REVUE_ECONOMIQUE
• Publié le: 01 janvier 1975
• Nombre de lectures: 43
• Langue: English
• Poids de l'ouvrage: 1 Mo

Extrait

Monsieur Warren L. Young
Alternate Models of Consumption Behaviour. A Cross-National
Analysis : The U.S., U.K. and Israël
In: Revue économique. Volume 26, n°1, 1975. pp. 123-134.
Citer ce document / Cite this document :
Young Warren L. Alternate Models of Consumption Behaviour. A Cross-National Analysis : The U.S., U.K. and Israël. In: Revue
économique. Volume 26, n°1, 1975. pp. 123-134.
http://www.persee.fr/web/revues/home/prescript/article/reco_0035-2764_1975_num_26_1_408196MODELS OF CONSUMPTION BEHAVIOUR ALTERNATE
A Cross -National Analysis :
the US, UK and Israel, 1953-1970
In this paper, we consider three major hypotheses of aggregate non
durable consumption behaviour in cross-national perspective and compare
their explanatory ability and forecasting performance over the period 1953-
1970. It is recognized, at the outset, that much work on the « consumption
function » has been done ; perhaps the field has even been overstudied.
We feel, however, that the cross-national framework within which we intend
to operate and the application of alternate behavioural models in this
regard does present a different, if not somewhat new approach to the pro
blem. Now, a great deal of effort has been expended in both theoretical and
empirical studies revolving around the « consumption function ». Some of
these studies, however, have been « fishing expeditions » for a « best fit »
rather than for the most reasonable and valid explanation of consumption
behaviour as such. In contrast to this, we have chosen to utilize a priori
specifications reflecting three major theoretical propositions regarding con
sumption behaviour : (a) absolute income, (b) permanent income, and (c)
relative income hypotheses, thereby enabling us to take into account both
general and case specific factors that would result in one hypothesis being
more acceptable than another in each of the cases dealt with or over all of
the cases involved. Moreover, as a result of this approach, we can compare
their explanatory ability and forecasting performance by and across cases in
order to gauge the relative efficacy of each hypothesis in both case-specific
and general contexts. Finally, while not a strict « test » of the hypotheses
themselves, our results would give some indication as to the « power » of each
hypothesis, such that a composite scale to rank hypotheses in order of
validity can be constructed. REVUE ECONOMIQUE 124
I
The three specifications we use were derived by Evans and applied, origi
nally, in the US case over the period 1929-41/1946-64. We have, however, l
imited consideration to the post-Korean War period, i.e. 1953/4-1969/70 over
which — with certain exceptions — more or less « steady » economic growth
occured in all three cases, although at different rates. Evans original specifi
cations are of the following form : [1]
(a) Absolute Income C = a + bY + d C_x
C /AY^ -/ C
(a) Permanent Income
Y \ Y / V Y
C / Y°\ / C
(c) Relative Income = a + b I j + d I
y \ y / \ x
where C is consumption of food, non-durables, and services, Y is personal
disposable income, and Y° is peak previous income, with AY = Y — Y_.l5
all at constant prices, base 1964 in the Israeli case, 1963 in the UK, and 1958
in the US case. In addition, there is a one year lag on the (C/Y) ratios and C
values as specified. [2]
It should be noted, however, that there is a fundamental anomaly inherent
in Evans permanent and relative income specification set as given above.
The problem, simply put, is that when disposable income exhibits a constant
secular upward trend, then Y° = Y_1? such that Y°/Y = Y^/Y. But A Y/Y
= (Y — Y_l)/Y =1 — (Y_x/Y). The regression coefficients for the two spe
cifications, therefore, would be of equal magnitude, varying only in the cons
tant and the sign of the co-efficient of the income term itself, as the sign of
the lagged C/Y ratios would be the same, resulting in identical mpc's being
generated both in the short run and long term. We have, therefore, inverted
the relative income term in our case, with a negative co-efficient value for
the revised expected as a result, thus obtaining the revised
C / Y \ / C
specification : = a — b I ) + d \
Y° / \ Y Y V
The regression results for the period 1953-1969 are given in table 1 below,
along with standard errors of estimate in parentheses below the regression
co-efficients, co-efficients of multiple determination adjusted for degrees of
freedom (Ra), and durbin-watson co-efficients in the respective case (dw).
The table also gives short run and long term mpc values generated by the
regressions themselves. A cursory glance at the table reveals, in the first
instance, that the permanent and relative income specifications have quite
similar explanatory ability but the absolute specification has a higher
R2 in each case. In the US case, however, the dw value hints at first order ALTERNATE MODELS OF CONSUMPTION BEHAVIOUR 125
autocorrelation of residuals in the absolute income specification, while in the
UK vase, the value is considerably lower than that obtained in both perma
nent and relative income specifications. As for the mpc values generated, they
all seem reasonably acceptable, even though both the short run and long term
values for the permanent income specification are higher than those generated
by the relative and absolute income models — although these may be biased
downwards, as Evans has noted elsewhere. With regard to forecasting per
formance, however, somewhat different results are obtained, as will be seen
below, where true ex-post and ex-ante forecasts are presented for the period
1966-1970.
TABLE 1
MPC
■^ ■ —
_ SHORT TERM =„ , Country and specification no
UK
ABS : C = 1162 .927 + .302 Y + .599 C^ ... ,302 .724 .99 1.64 (1)
(.061) (.081) C
PIH = .023 — .504 (A Y/Y) + .982 (C/Y)^ .371 .778 .93 1.84 (2)
(.093) (.068) C
DEF = .501 — .473 (Y/Y°) + .980 (C/Y)_x .349 .714 .93 1.85 (3)
(.087) (.068)
US
ABS : C = 125 .684 + .418 Y + .440 C_, 418 .724 .99 1.05 (1)
(.084) (.121) C
PIH = .069 — .345 (A Y/Y) + .927 (C/Y)^ .454 .753 .90 2.49 (2)
(.063) (.081) C
DEF = .387 — .319 (Y/Y°) + .928 (C/Y_3 .344 .733 .90 2.47 (3)
(.059) (.081)
Israel
ABS : C = 232 .689 + .349 Y + .590 C_x 349 .762 .99 1.55 (1)
(.077) (.099) C
PIH = .024 — .457 (A Y/Y) + 1.010 (C/Y)_T .456 .858 .60 (2) 1.48
(.144) (.210) C
DEF = .415 — .386 (Y/Y°) + 1.001 (C/Y)_j ,364 .775 .58 1.57 (3)
(.126) (.212)
where ABS = the absolute income specification (a) given above,
PIH = the permanent income (b),
= the relative income (c), this form being a revised version DEF
of an equation originally proposed by Duesenberry-Eckstein-Fromm as
indicated by Evans. REVUE ECONOMIQUE 126
II
It is common practice in reporting ex-post forecasting results to cite
the values predicted by the regression as the forecasted values of the depen
dent variable. While there is some justification in using results obtained in this
manner, the predicted value generated by the regression equation is not
actually a true ex-post forecast, for the « future » values of the indepen
dent variable substituted into the regression equation are not forecasts, but
are the actual future values themselves. In order to compare ex-post regres
sion forecasts with true ex-ante forecasts, then, one must substitute an
independently obtained forecasted value for the exogenous variable into the
regression equation and compare the result obtained with the actual value of
the dependent variable for the period to which the forecast is made. Theref
ore, in order to obtain consistent ex-post, and ex-ante forecasts, the regression
equation must be re-estimated to each successive base forecasting period and
new forecasts for both independent and dependent variables continuously
made. As such, autoregressive income equations of the form Yt = a + b Yt ..-,
have been estimated to each base period and income projections made accor
dingly. The projected values are then substituted into the regression equations
that have been re-estimated at each base period (in our case 1965-1969),
and consumption forecasts for one year ahead obtained accordingly. Thus,
both the ex-post (1966-69) and ex-ante (1970) forecasts are consistently based,
and as such can be validly averaged and compared by and over the cases in
volved. The results of the income projection technique are presented in table
2 below, which compares the projected and actual values over the period
1966-1970 and gives average absolute projection errors by country :
TABLE 2
Income Projections, 1966-1970 :
percentage error and average absolute error by country
Projection for
Country
1966 1967 1968 1969 1970 Average
US . . 0.03 2.8 0.6 2.5 0.8 1.3
— 1.7 UK 1.1 1.1 1.7 0.4 1.2
— 0.6 — 7.7 — 1.0 — 0.6 Israel . . 9.6 3.9
Now, the error in the dependent variable forecast should be somewhat
greater, on average, when using our method than when actual future values
for the independent variables are substituted in the regression equat