HAC變異數矩陣的新估計方法

Abstract

We propose using recursive residuals to compute HAC estimators. Simulations show that the recursive residuals-based HAC estimator has smaller bias than the conventional HAC estimator, and the tests based on this estimator performs very well in terms of size and power. In contrast to the KVB robust test, it pays a slight price of power loss but delivers accurate finite sample size. In some cases, the tests based on the proposed HAC estimator even has more accurate finite sample size than that of KVB. In contrast to the prewhitened HAC estimator, the tests based on the proposed HAC estimator still works well when nonlinear dependency or heteroskedasticity is present.