No label defined (Q4512130)

Posterior analysis of econometric models invariably requires complete specification of the distribution of dependent variables given covariates and parameters. In the vast majority of situations, however, specification of distributions is largely a problem of dealing with nuisance parameters. Although, in general, the distribution is not of interest itself and it is not possible to derive useful conclusions from the data unless one specifies precisely the distribution. More often than not choosing a specification is motivated by convenience rather than economic theory itself or other considerations. From the classical theory viewpoint some nonparametric alternatives have been proposed but attention has been restricted to expansions or other functional approximations to the distribution of error terms. Assuming a parametric form for the regression function nonparametric or semi-parametric modeling of the density or likelihood has received relatively little attention in the classical literature and no attention at all in the Bayesian literature. In the Bayesian field very little is known about posteriors derived from nonparametric or semi-parametric likelihood functions. Whenever it is reasonable to assume that disturbances of statistical models are independent a simple approach is available and the purpose of this paper is to explore this nonparametric approach. Specifically, the paper proposes to approximate the unknown distribution by its nonparametric counterpart -- a step function -- and treats the points of the support and the corresponding density values as parameters whose values should be determined based on the available data. The present paper also shows that posterior distributions of parameters of interest, posterior odds and predictive densities, can be computed easily by using this nonparametric approach to the problem of formulating the likelihood function. The paper proposes Markov chain Monte Carlo methods to perform posterior analysis, and applies the new method to an analysis of stock returns.