preferences follow. Habits may arise because it is costly for the household to adjust consumption in
response to changes in the economic environment. These costs may involve the cost of learning and
perfecting new recipes and any psychological disutilities from switching to new cuisine. Conversely,
inventory adjustment may be a result of hoarding on the part of the household to take advantage of
supermarket specials or an appetite for food diversities. Finally, additive separability for any two
goods (i, j) is implied if bij = 0. It is important to test habit formation in meat consumption in a
demand system framework as all empirical studies of meat demand decisively reject separabilities
across meat types.
To incorporate the effects of food safety information on the household preferences, the parameter
ai is further specified to be a linear function of the household’s perception of meat safety. That is,
ai = ai0 + ai1 bst + ai2pst + ai3cst
(7)
where bst , pst and cst are, respectively, consumer perceptions of the safety of beef, pork and
poultry. Because there is no way to determine a priori the most appropriate way of representing
these perception variables with our news indices. We examine the performance of a few alternative
representations in more detail in the model results section.
4.2 Estimation strategy
When estimating a dynamic rational expectations model, the Generalized Method of Moments
(GMM) of Hansen (1982) is the natural choice. Use the direct utility function (6) to parameterize
the marginal utility of consumption (3)
ait lnxjt lnxit-1 lnxit+1
(8)
MUit =--+ > bij—+ + bi--+ βEt bi---—
mit mit mit mit
j=1
for i = 1, 2 and 3. Next, use (8) and (5) to derive an estimable form of the FOC
eit =
a1t lnxjt lnx1t-1 lnx1t+1
---+ > , bij—+ + bi------+ bιβ------
m1t m1t m1t m1t
j=i
ait
mit
3
+ bij
j=i
lnx jt
mit
lnxit-i
+ bi-------
mit
+ biβ
lnxit+1
mit
(9)
for i equal to 2 (pork) and 3 (poultry), where mi is the expenditure on the ith meat. The expecta-
tions in (5) are replaced by realizations less innovations. These innovations are expectation errors
11