Empirical models of meat demand: How do they fit out of sample?

Out of sample tests were performed on two meat demand models: the linear, relative-price real-income specification and the almost ideal demand system. Variants of each model included seasonal effects and habit formation. Estimation procedures included single equation OLS, seemingly unrelated regression, and autocorrelation corrections. The models were tested over five different time periods. The study found that no particular model consistently outperformed the others. This was true over different time periods, across commodities, and across specifications and estimation techniques. A X2 goodness of fit test could not reject the hypothesis that outcomes were uniformly distributed by classes. There was little statistical difference between the top five model types. At the 5% level, the best predicting model was not significantly different from the top four contenders in a majority of the 15 test cases.