A modified bootstrap for kernel-based specification test with heavy-tailed data

This paper provides a new resampling strategy to improve the finite sample performance of a nonparametric kernel-based specification test in the presence of heavy-tailed error terms. Based on the test statistic of Li and Wang (1998), we propose to generate the bootstrapped samples using a modified wild bootstrap. This new method matches all moments of the error terms if the error has a symmetric distribution and matches the first and all even moments when error distribution is asymmetric around zero. This new resampling method has better finite sample performance than the traditional one when the distribution of the error terms is symmetric and heavy-tailed. (c) 2020 Elsevier B.V. All rights reserved.