Abstract

We develop bootstrap procedures for the realized Laplace transform of volatility and associated statistics. First, we demonstrate that a naive wild bootstrap fails for the realized Laplace transform of volatility. Next, we propose a modified wild- as well as a local Gaussian bootstrap and established their first-order asymptotic validity. Motivated by the superior performance of the local Gaussian bootstrap procedure in finite samples, we use Edgeworth expansions to compare its theoretical accuracy with existing first-order feasible asymptotic theory. Our cumulants expansions demonstrate that the local Gaussian bootstrap is able to mimic the higher-order bias of the studentized test statistic, for which second-order asymptotic refinements are obtained. Not surprisingly, our Monte Carlo simulation study shows that the local Gaussian bootstrap outperforms the modified wild bootstrap and existing (first-order) feasible inference theory in finite sample. This is joint work with Ulrich Hounyoy and Rasmus T. Varneskov.

About the Speaker

Dr. Zhi LIU is an associate professor in University of Macau. He obtained his PhD degree in Statistics from the Hong Kong University of Science and Technology in 2011. Before joining the Department of Mathematic of University of Macau in August 2012, he served as assistant professor in Economics at the Xiamen University, China, from September 2011 to July 2012. His current research interests include Empiricial Likelihood, Financial Econometrics, Bioinformatics, etc.

Poster