Title: HAC standard errors for robust estimators
Authors: Christophe Croux - Edhec Business School (France) [presenting]
Abstract: Robust regression methods give reliable estimates in presence of outliers. Most of the research in robust regression focuses on point estimation, although the statistical inference part has been developed as well and is available in statistical software. We discuss the problem of obtaining valid standard errors for robust regression estimators when the error terms are heteroscedastic or serially correlated. Such standard errors are called HAC standard errors. We collect existing formulas and discuss up to what extent the HAC standard errors are robust with respect to outliers, including leverage points. We also obtain new results, as we could quantify the loss in efficiency when using a HAC standard error when in fact there is no need for it.