Parameter-invariant unbiased estimation of individual variances and their pairwise products

Citation:

Anatolyev, Stanislav and Korobka, Yaroslav (2026) "Parameter-invariant unbiased estimation of individual variances and their pairwise products", Economics Letters, vol. 268, pp. 113145

Abstract:

Recent literature proposed variants of estimators of conditional error variances that are robust to the presence of many regressors. One of the proposals produces unbiased estimates, which, however, are sensitive to regression parameter values. We propose its modification based on cross-fitting and averaging over random sample splits, which preserves the unbiasedness but eliminates that sensitivity. Then, we extend the cross-fitting machinery to unbiased estimation of pairwise variance products and to testing for many restrictions. We verify properties of associated estimators and tests in a simulation setup with many covariates.

Paper in accepted form:

PIUE.pdf

Supplimentary appendix:

PIUE-SA.pdf