Working Paper: NBER ID: w17698
Authors: Jerry A. Hausman; Christopher J. Palmer
Abstract: Since the advent of heteroskedasticity-robust standard errors, several papers have proposed adjustments to the original White formulation. We replicate earlier findings that each of these adjusted estimators performs quite poorly in finite samples. We propose a class of alternative heteroskedasticity-robust tests of linear hypotheses based on an Edgeworth expansions of the test statistic distribution. Our preferred test outperforms existing methods in both size and power for low, moderate, and severe levels of heteroskedasticity.
Keywords: heteroskedasticity; finite samples; edgeworth expansion; bootstrap
JEL Codes: C01; C12
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| Methods used for estimating standard errors (C51) | Accuracy of hypothesis testing (C12) |
| Existing adjusted estimators (C51) | Size distortions in hypothesis testing (C12) |
| Second-order bootstrap method (C69) | Rejection rates of null hypotheses (C12) |
| Traditional White standard errors (C21) | Overrejection under null hypothesis (C12) |
| Second-order bootstrap method (C69) | Size properties and power (C29) |
| Second-order bootstrap method (C69) | Rejection frequencies with high heteroskedasticity (C21) |