Block Length Choice for the Bootstrap of Dependent Panel Data -- a Comment on Choi and Shin (2020)

TL;DR

This paper discusses how an adaptive block length choice improves the performance of a bootstrap-based change-point test in dependent panel data, challenging previous claims about size distortions.

Contribution

It shows that with proper data-adaptive block length selection, existing tests can better handle mild temporal dependence, reducing size distortions.

Findings

Adaptive block length improves test accuracy

Sharipov et al.'s test handles mild dependence better

Reduces size distortion compared to previous claims

Abstract

Choi and Shin (2020) have constructed a bootstrap-based test for change-points in panels with temporal and and/or cross-sectional dependence. They have compared their test to several other proposed tests. We demonstrate that by an appropriate, data-adaptive choice of the block length, the change-point test by Sharipov, Tewes, Wendler (2016) can at least cope with mild temporal dependence, the size distortion of this test is not as severe as claimed by Choi and Shin (2020).