Rejection rates of bootstrapped and exact heteroskedasticity tests in response to skedastic function and normal or skewed disturbances

Abstract

This study focuses on examining the power of bootstrapped and exact tests to detect variance heterogeneity in linear models via Monte Carlo. Three independent MC schemes were set up corresponding to distinct forms of heteroskedasticity, degree of heteroskedasticity and with or without normality. The tests produced unequal power in the simulated combinations; some performed poorly, when exposed to the unusual type of heteroskedasticity from the point of the test. The Bamset LR test had adequate power for most skedastic functions and should be extensively used. Park’s test had inadequate power in most scenarios. The power of White’s test varied strongly depending on the skedastic function. The bootstrapped tests had generally lower power and were often oversized, although the only resampled Harrison and McCabe test had outstanding power in several MC schemes. Reduced power and increased Type I Error Rates were noticed in most exact or resampled tests with skewed disturbances.

Keywords:

• Bootstrap
• Heteroskedasticity test
• Monte Carlo simulation
• Normality
• Power
• Test size

Funding

The authors have declared that no direct funding was associated with the research reported in this article.