meat demand at quarterly frequencies. Although this is consistent with the approach followed by
most studies on habit formation under rational expectations, it is not innocuous. It is plausible
that consumption at nearby dates is substitutable (durable) while habits develop over a longer
time span. In Becker and Murphy, it was shown that the degree of habit persistence is positively
related to the rate at which the habit stock depreciates. Becker (1996) defined traditions as mild
habits whose habit stocks depreciate more slowly and are likely to be related to behaviors in the
more distant past. In the case of food, the rate of depreciation of its habit stock may be quite low
compared to substances that are clearly addictive to many people, such as cigarettes. For instance,
once the household learns a new recipe, its knowledge capital may not quickly dissipate. It follows
that it is desirable to account for consumption experiences in the more distant past than just one
quarter ago.
Therefore, a more fruitful formulation of the problem may be to explicitly model short-run
durabilities and long-run habit formation (see, for example, Heaton, 1995). This requires additional
lags of consumption to enter the utility function. As Heaton pointed out, these additional terms
imply a larger MA structure in the error term of (9). It makes the estimation of the asymptotic
covariance matrix of the GMM estimator more difficult. A practical solution to this problem is to
follow Heaton (1995)’s Simulated Methods of Moments approach. Exploration of this possibility is
outside of the scope of this paper but is on our research agenda.
Finally, it is important to realize that our results are conditional on the chosen functional form.
Future research should consider other specifications of the utility function and the way that food
safety variables enter the household preferences.
16