MINIMUM SAMPLE SIZE REQUIREMENTS FOR MAXIMUM LIKELIHOOD ESTIMATION OF SOME DEMAND MODELS

Abstract

The method of maximum likelihood is rather demanding, in the sense that a large number of observations is required in order to prevent the likelihood function from becoming unbounded. In this paper precise conditions are formulated such that the likelihood function associated with the Rotterdam model will be bounded under (a) conditions of homogeneity and symmetry and (b) the negativity condition. In addition it is shown that, surprisingly enough, the number of observations that is required to estimate the linear expenditure system is not smaller than the number that is required to estimate the Rotterdam model.