Minimum distance estimation in linear models with long-range dependent errors

Abstract

This paper discusses the asymptotic representations of a class of L 2-distance estimators based on weighted empirical processes in a multiple linear regression model when the errors are a function of stationary Gaussian random variables that are long-range dependent. Unlike the independent errors case, the limiting distributions of the suitably normalized estimators are not always normal. The limiting distributions depend heavily on the Hermite rank of a certain class of random variables. Some ‘goodness of fit’ tests for specified error distribution are also considered.