Testing for Heterogeneous Parameters in Least-Squares Approximations

Abstract

This paper suggests tests for the heterogeneity of parameters in linear least-squares estimation. The tests are based on the properties of resampled estimates, and test the hypotheses that the parameters have a common mean, or that they are independently and identically distributed. The tests can be viewed as the analogue of those based on recursive residuals, in cross-sectional models. We analyse the properties of tests based on jack-knifed estimates in the linear regression model, and compare their performance in a small empirical application.