Model selection for regression with heteroskedastic and autocorrelated errors

This paper develops new model selection criteria for regression with heteroskedastic and autocorrelated errors. We prove the selection consistency of the introduced criteria and evaluate their performance by simulation. The results suggest that the new criteria may bring significant improvement relative to the traditional criteria. Besides, we discuss how the idea behind the new criteria can apply to model selection for a general class of M-estimation models.

► We develop new model selection criteria for regression with heteroskedastic and autocorrelated errors. ► We prove the selection consistency of the new criteria and evaluate their performance by simulation. ► The simulation results suggest that the new criteria may bring significant improvement relative to the traditional criteria. ► We discuss how to extend the idea behind the new criteria to model selection for a general class of M-estimation models.