FINITE-SAMPLE BEHAVIOR OF TESTS FOR GROUPED HETEROSKEDASTICITY

When the regression error variance is constant within subsets of data but differs across the subsets, grouped heteroskedasticity exists. Most tests for grouped heteroskedasticity are based on an asymptotic chi2 variable, and how well they perform in small samples is unknown. In this study the performance of several of these, along with Breusch-Pagan tests, is examined using Monte Carlo methods. Generally, it is found that the tests which tend to be the easiest to apply and use the fewest assumptions are at least as good as somewhat more elaborate procedures.