of ARCH effects. Under the null the test statistic ARC H (κ) is χ2(κ) distributed where κ is the lag
order of the ARCH effect. These test results confirm the presence of conditional heteroscedasticity
in all residual series. Also reported in table (2) are Ljung-Box test results for autocorrelation in the
error terms. The test statistic LB(κ) is χ2 (κ) distributed under the null of no serial correlation,
where κ is the lag order of the serial correlation. The Ljung-Box test may show spurious evidence
of serial correlation when the series is conditionally heteroscedastic. Diebold’s (1987) correction for
ARCH effects was applied to the calculation of these Ljung-Box statistics. The corrected Ljung-Box
statistic with eight lags cannot reject the null of no serial correlation for the ∆4eit's at the 10%
level.
5 Concluding Remarks
The objective of this paper has been to test for rational habit formation and the effects of food safety
information on U.S. meat consumption. We consider a multiple-good version of the household pref-
erences allowing for intertemporal nonseparabilities. Under rational expectations the representative
household maximizes the life-time utility taking into account the effects of its current consumption
decisions upon future utilities. Habits provide such a mechanism through which current levels
of consumption could affect future utilities. To investigate the effects of food safety information
on meat demand, food safety news articles from the popular press were compiled into information
indices. These indices are then used to approximate the “true” consumer perceptions of food safety.
U.S. quarterly data on meat consumption are used to estimate the model. The degree of
habit persistence for beef increased around 1998, while there were no discernible shifts in the
degree of habits for pork and poultry. During the post-1998 period habit persistence dominates
inventory adjustment in beef consumption. It implies that the long-run price and quality elasticities
of beef are larger than their short-run counterparts. However, the intertemporal optimization
nature of the consumer’s problem means that these elasticities cannot be easily calculated. In fact,
the consumer’s problem is highly non-linear and has to be solved numerically. Once this step is
complete, consumption responses to price and food safety shocks can be simulated under various
expectations schemes.
In the empirical analysis, we have specified that only one-period lagged levels of consumption
enter the current utility function and hoped this would be enough to capture habit formation in
15