region regarding serial correlation (especially in the AIDS model). Overall, the
models appear to do a better job of explaining milk, soft-drink and coffee and tea
demand than juice demand. All own-price parameters (except the one for coffee and
tea in model J) in models I and J are statistically significant (at the 5% level unless
noted otherwise) with correct signs, while only few of them are statistically
significant in models F - H, indicating the advantage of using demand system over
single equation. The own-advertising parameter is statistically significant only for
soft drinks in models F, G, and J. The price-advertising interaction term is found
weakly significant (at the 10% level) for soft drinks and coffee and tea in model F,
significant for soft drinks in model G, and significant for milk, soft drinks, and
coffee and tea model J. Furthermore, most of the models show higher proportion of
population under age five leads to higher demand for milk, and more dining out
(higher Fafh) leads to higher demand for soft drinks and lower demand for milk, and
coffee and tea, which are all consistent with expectation.
Base on the estimates in table 3, we calculate the own price and advertising
elasticities, compute interaction-related rotation effects according to the formulae in
table 1, and report them in table 4. Wald statistics for the null hypothesis that the
estimated interaction effects are jointly zero are also reported. 9 We report own
advertising elasticities when γi or dii is found significant and report interaction-
related rotation effects when γi is found significant. The linear, semi-log, and AIDS
models reject the null hypothesis that the estimated interaction effects are jointly
zero at the 5% level based on the Wald statistics. To put the results of rotation effect
into perspective, we focus on interpreting the results of the AIDS model given the
14