Using heteroscedasticity-non-consistent or heteroscedasticity-consistent variances in linear regression

C. Y.(Chor yiu) Sin, Cheng Few Lee

Research output: Contribution to journalArticlepeer-review

3 Scopus citations

Abstract

The properties of the heteroscedasticity non-consistent variances and heteroscedasticity consistent variances are reviewed. Unlike the related existing results, the following cases are discussed separately: (i) the cases where the explanatory variables are strictly exogenous; and (ii) the cases where the explanatory variables may or may not be strictly exogenous. The latter cases allow weakly dependent explanatory variables such as those generating from an autoregressive process. New results on the original robust variance (denoted by HC0) and its variants (denoted by HC1, HC2, HC3, HC4 and HCj) are derived. In particular, the followings are shown: (i) the ordering of the original robust variance and its variants; (ii) the asymptotic equivalence among different variants of robust variance; and (iii) under quadratic form of heteroscedasticity (with mesokurtic/leptokurtic normalized error) or GARCH(1,1)-error, non-robust variance rejects more often than robust variance. Simulation studies suggest HC4 by and large does not over-rejects or mildly under-rejects.

Original languageEnglish (US)
Pages (from-to)117-142
Number of pages26
JournalEconometrics and Statistics
Volume18
DOIs
StatePublished - Apr 2021

All Science Journal Classification (ASJC) codes

  • Statistics and Probability
  • Economics and Econometrics
  • Statistics, Probability and Uncertainty

Keywords

  • Asymptotic properties
  • Finite-sample properties
  • non-robust variance
  • Robust variance
  • Strictly exogenous
  • Weakly dependent

Fingerprint

Dive into the research topics of 'Using heteroscedasticity-non-consistent or heteroscedasticity-consistent variances in linear regression'. Together they form a unique fingerprint.

Cite this