Resampling methods for dependent data (Q1407179)

At the beginning, a brief description of the ``bootstrap principle'' is given and and the use of resampling methods is advocated for estimating what are called ``level-2`` (and ``higher-level'') parameters in this book. Chapter 2 sketches the historical development of bootstrap methods and describes various types of bootstrap methods that have been proposed in the context of dependent (temporal) data. The next two chapters establish consistency of various block bootstrap methods for estimating the variance and the distribution function of the sample mean and extend these results to general classes of statistics, including \(M\)-estimators and differentiable statistical functionals. A number of numerical examples is given. Chapter 5 deals with the comparison of different block bootstrap methods. It starts with a numerical comparison; then some theoretical results are presented. Chapter 6 deals with Edgeworth expansions and second-order properties of block bootstrap methods for normalized and studentized statistics under dependence. The important problem of selecting the optimal block size empirically is addressed next. The following two chapters treat bootstrap based on independent and identically distributed innovations in popular time series models, such as the autoregressive processes and deal with the frequency domain bootstrap. Chapter 10 describes properties of block bootstrap and subsampling methods for a class of long-range dependent processes. Chapter 11 treats two special topics -- viz., extremums of dependent random variables and sums of heavy-tailed dependent random variables. As in the independent case, here the block bootstrap fails if the resample size equals the sample size. A description of the random limit is given in these problems, but the proofs are omitted. Chapter 12 considers resampling methods for spatial data under different spatial sampling designs. It also treats the problem of spatial prediction using resampling methods. There are mainly two target audiences for the book. The first five chapters of the book give full details of the proofs of the theoretical results and step-by-step instructions for implementation of the methodology. This part of the book can be used as a text for a graduate level course. For the first part, familiarity with only basic concepts of theoretical statistics is assumed. The second part of the book (Chapters 6-12) is written in the form of a research monograph, with frequent reference to the literature for the proofs and for further ramification of the topics covered. This part is primarily intended for researchers in statistics and econometrics.