Comparing the Performance of Linear and Non-Linear Models to Explain Almost Ideal Demand System

In most of empirical studies based on almost ideal demand system (Aids), the elasticity of the price and income estimated by these equations resulted to some sensitive policy making recommendations in microeconomics and macroeconomics. It is in such a case that there is some doubt about reliability of linear estimation of such models. In this study, the performance of linear and non-linear almost ideal demand system is under the investigation. For this purpose, seemingly unrelated regression (SURE) method will be applied to estimate linear model and multilayered feed forward neural network (MFNN) is used to estimate a non-linear one. The results indicate that multilayered feed forward neural network is associated with less error than the linear model, and consequently, leads to a better estimation of almost ideal demand system. This result creates some hesitate on application of Stone price index for linear zing estimation of almost ideal demand system. Therefore, it is suggested that feed forward neural network will be applied to estimate almost ideal demand systems.