California Agricultural Experiment Station Giannini Foundation for Agricultural Economics

The generalized maximum entropy (GME) estimator was introduced by Golan et al. as a way to overcome two empirical problems that hamper traditional econometrics: multi-collinearity and ill-posed models. Despite its recent origin, several papers based on the GME approach have appeared already in this journal (Paris and Howitt; Miller and Plantinga; Lence and Miller; Zhang and Fan). The distinguishing feature of the GME es-timator consists in its requirement of a specific assumption and non-sample information about parameters and error terms. In particular, its implementation relies on subjective in-formation about the range of variation of parameters and error terms that must be pro-vided by the researcher. For reference ease, we state the GME estimator of a classical linear statistical model following the notation of Golan et al. (chapter 6): The linear model to estimate is given by (1) y X u, u ~I ID(0, 2I), where the dimensions of the various components are y ~ (T 1),u ~ (T 1), ~ (K 1) and X ~ (T K). The vector y and the matrix X constitute sample information while the vector represents parameters to estimate and the vector u contains random distur-bances. The principal assumption of the GME formalism is that a parameter k is re-garded as the mathematical expectation of some discrete support values Zkm, m 1,...,M, such that (2) k Zkmpkmm 1 M where pkm 0, k 1,...,K, m 1,...,M, are probabilities and pkm 1m