Estimating censored and non homothetic demand systems: the generalized maximum entropy approach

The econometric estimation of zero censored demand system faces major difficulties. The virtual price approach pioneered by Lee and Pitt (1986) in an econometric framework is theoretically consistent but empirically feasible only for homothetic demand system. It may even fail to converge depending on initial conditions. In this paper we propose to expand on this approach by relying on the generalized maximum entropy concept instead of the Maximum Likelihood paradigm. The former is robust to the error distribution while the latter must stick with a normality assumption. Accordingly the econometric specification of censored demand systems with virtual prices is made easier even with non homothetic preferences defined over several goods. Illustrative Monte Carlo sampling results show its relative performance.