A1798
Title: Optimal tuning parameter selection in heteroscedasticity and autocorrelation robust inference
Authors: Zhou Zhou - Zhejiang University (China) [presenting]
Abstract: The selection of a block size, window size, or bandwidth parameter is universal and important in various Heteroscedasticity and Autocorrelation Robust (HAR) inference problems. Most current proposals for choosing this parameter are ad hoc without theoretical justification of their performance or optimality. An investigation into the asymptotic bias and variance of a fundamental HAR procedure for a general class of piecewise locally stationary time series is presented. As a result, an easy to implement and asymptotically optimal tuning parameter selection algorithm is proposed. The investigation also reveals that analogous tuning parameter selection algorithms for stationary time series are biased when applied to nonstationary time series.
