Modeling household behavior in a CGE model: linear expenditure system or indirect addilog?

We try to argue that in a computable general equilibrium model, household preferences should be modeled by the indirect addilog system (IAS) rather than by the frequently used linear expenditure system (LES). Both systems have the same data requirement and are as easy to implement, but IAS provides for a richer description of preferences. Contrarily to LES, its Engel curves are non-linear and it allows for inferior commodities, elastic demand and gross substitution. LES assigns zero utility to households with expenditure below a positive minimum value, whereas IAS assigns a positive utility, provided zero expenditure is replaced by a small positive number. In micro simulation models where the results of a macro CGE model (with one representative household) are used at micro level, this constitutes a clear advantage of IAS. In the framework of an expenditure survey, we find overwhelming statistical evidence that the IAS indirect utility function is likely to be (much) closer to the true indirect utility function than LES. Consequently, expenditure elasticities and welfare changes are likely to be (much) better estimated by IAS. Simulations with a CGE model for Palestine show that price responses and equivalent variation are considerably higher for IAS than for LES.