Bootstrapping Hill estimator and tail array sums for regularly varying time series (Q2040068)

The paper provides a multiplier block bootstrap procedure for tail array sums of heavy-tailed time series. Tail array sums contain, among others, the Hill estimator, the expected shortfall, the extremogram and the tail spectral process. The available data is assumed to be obtained from regularly varying functions of Markov chains, with the exact setup presented in Assumption 1 on pages 1414--1415. Section 1.3 explains that the naive multiplier bootstrap will be inconsistent, and what kind of adjustments need to be made. Section~2 collects a number of results from earlier articles, written in part by the second author [Stochastic Processes Appl. 129, No. 11, 4209--4238 (2019; Zbl 1448.60114); Ann. Stat. 38, No. 4, 2145--2186 (2010; Zbl 1210.62051); Ann. Appl. Probab. 8, No. 3, 868--885 (1998; Zbl 0939.60007)], that are useful in the considered theoretical framework. Section~3 contains the main contribution of the paper, which is the announced multiplier block bootstrap procedure and its consistency. Section~3.1 is an application to two particular examples: the Hill estimator and the extremogram. The proof of the main result is provided in Section~3.2. The finite-sample performance of the proposed procedure is examined in Section~4 on both of these particular examples of tail array sums.