This analysis gives a number of applications of Deaton and Muellbauer (1980a)’s almost ideal demand (AID) system to annual and seasonally unadjusted quarterly household consumption data obtained from the Norwegian national accounts. Parameter estimates from the expenditure systems are utilized to generate Marshallian price and income elasticities for 10 non-durable consumption categories. The analysis compares the explanatory power of the alternative expenditure system specifications by means of the likelihood dominance criterion for model selection proposed by Pollak and Wales (1991), which shows that the dynamic linear approximate AID system incorporating habits is preferred to the other specifications. However, the evidence indicates that the homogeneous static linear approximate AID system is the only specification that (i) does not suffer from lack of precisely estimated parameters and (ii) yields results that are interpretable and empirically plausible.
The analysis then evaluates how the linear approximate AID system incorporating habits performs out-of-sample. We also examine whether simpler expenditure system specifications are more suitable for forecasting. Based on the Diebold-Mariano test proposed by Diebold and Mariano (1995) we conclude that our preferred dynamic specification does not yield more accurate predictions than the dynamic linear approximate AID system incorporating habits without cross-price effects or the random walk model. The evidence also suggests that the focus of dynamic forecast analyses of non-durable consumption categories should be on obtaining accurate price predictions, as prices account for most of the variation in the commodities’ expenditure shares.