AN ALTERNATIVE
While I am in general agreement with Haidacher's
recommendations, I do not think he goes far enough.
Use of time trends and varying paramenter methods
merely treat the symptoms, not the cause. The same
could be said about including lagged variables in the
model if these variables are included in an ad hoc fash-
ion.
In contrast to Haidacher, I would recommend mod-
ifying the existing theory or model to account for the
hypothesized structural shift or misspecification. In this
way, the MH includes the source of the hypothesized
misspecification, and when nested within the AH, it
becomes an hypothesis which can in principle be re-
futed by the data. This suggestion is not new, but can
be traced back at least two decades to Frederick Waugh
when he wrote,
It is high time we develop new theories and concepts of
value that are testable by statistical analyses. If statistical
findings fail to confirm the theories inherited from our
predecessors, should we struggle to invent elaborate
methods to reconcile the facts with the theory? Rather like
the physical scientists, we should modify theoretical con-
cepts to make them fit the observed facts in the actual
marketplace, (p. 7)
This, of course, does not mean we should necessarily
throw out the neoclassical paradigm. What it means to
me is that we should not stop once we have obtained a
set of demand parameter estimates. We should follow
the diagnostic procedures outlined by Haidacher and
then, if necessary, respecify and reestimate the model.
A case where this procedure has proven useful is in
testing the general restrictions of consumer behavior.
These tests have consistently led to rejection of the
homogeneity restriction and, in some cases, the sym-
metry restriction (Deaton and Muellbauer, chapter 3).
What are we to conclude from these tests? That utility
maximization is incompatible with consumer behav-
ior? That taste changes make the restrictions incom-
patible with the data? Not necessarily. Most analysts
have focused on possible causes of the misspecifica-
tion including functional form misspecification (Gal-
lant), dynamic misspecification (Deaton and
Muellbauer), joint allocation of labor supply and goods
expenditure (Barnett), and aggregation over con-
sumers (Sexauer).
With respect to functional form misspecification,
Gallant estimates an essentially unbiased functional
form based on a multivariate fourier series expansion.
He then uses this to test for functional form misspeci-
fication in the translog specification. He rejects the
translog relative to the fourier form and concludes that
tests based on the translog bias the results toward re-
jection of the general restrictions. Deaton and Muell-
bauer, in testing the general restrictions with the Almost
Ideal Demand System, found that when they first-dif-
ferenced their equations and included intercepts in the
equations, the incidence of serial correlation and re-
jection of homogeneity went down. Since the inter-
cepts were significant in most instances (suggesting
trend effects), they concluded that this was strong evi-
dence of dynamic misspecification. Further work by
Blanciforti and Green tends to confirm this finding.
In contrast to these studies, Bamett argues that lei-
sure is the shift variable causing the apparent taste
change found in aggregate-goods demand estimates.
When goods expenditure and labor supply are esti-
mated jointly, he finds the price of leisure is a signif-
icant variable in these equations, and the shift variables
become insignificant.
Sexauer, in exploring the effects of demographic
shifts and income distributional changes on food-away-
from-home expenditures, concludes that
. . . some of the behavior which economists normally at-
tribute to taste changes [on food-away-from-home ex-
penditures] can actually be quantified as being a result of
compositional shifts, (p. 1055)
While all these studies differ somewhat in method-
ology and data sources, they have one thing in com-
mon: they attempt to isolate the source of the
misspecification and then modify the existing theory
to account for this misspecification.
By way of summary, I would like to return to the
fundamental question of whether we should entertain
the possibility that taste changes are a cause of struc-
tural shifts in demand equations. In this context, a well-
known paper by Stigler and Becker makes some rele-
vant points. Their maintained hypothesis (p. 76) is
“that tastes neither change capriciously nor differ im-
portantly between people.” This interpretation is im-
portant, they argue, because
an explanation of economic phenomena that reaches a dif-
ference in tastes between people or times is the terminus
of the argument: the problem is abandoned at this point to
whoever studies and explains tastes (psychologists? an-
thropologists? phrenologists? sociologists?). On our pre-
ferred interpretation, one never reaches this impasse: the
economist continues to search for differences in prices or
incomes to explain any differences or changes in behav-
ior. (p. 76)
Stigler and Becker then go on to argue that such phe-
nomena as addiction, custom and tradition, advertis-
ing, and fashions and fads can be explained by relative
prices and income with stable tastes. What is impor-
tant here, they argue, is definition of the direct objects
of choice to the consumer (i.e., what he is deriving sat-
isfaction from) and the form of the household produc-
tion functions relating the market goods to these
commodities. While some of the examples they pre-
sent may seem trivial or appear unimportant to agri-
cultural economists, the possibilities for household
production theory in explaining demand behavior for
food and agricultural commodities seem endless. More
importantly, their basic message has important impli-
cations for how we assess structural change in demand
for food. That is, we should not abandon our search for
economic explanations in favor of interpretations of
taste changes until we are satisfied that we have ex-
plored the numerous subtle forms that prices and in-
come can take in explaining demand behavior.
40