Size Distortions in Robust Estimators: Implications for Asset Pricing

Abstract

We evaluate the reliability of HAC estimators in typical asset pricing applications. Through simulations, we show that these estimators often produce inflated t-statistics and invalid inference when applied to return series with autocorrelation and heteroskedasticity—common features in anomaly returns. To address this, we introduce SHARFS, a simulation-based inference procedure that estimates p-values using empirically calibrated null data-generating processes. Unlike traditional methods, SHARFS provides valid finite-sample inference even under complex return dynamics. Applying our method to 212 documented anomalies, we find that standard estimators substantially overstate significance: many strategies deemed significant by HAC methods fail to pass our more robust test. Our results challenge the credibility of conventional inference in empirical finance and call for a shift toward simulation-based methods.

Keywords: anomalies, asset pricing, autocorrelation, heteroscedasticity, robust estimation

JEL Classification: C12, C14, C21, C58, G12