Moving block and tapered block bootstrap for functional time series with an application to the \(K\)-sample mean problem (Q2325383)

The authors consider Hilbert space-valued stochastic processes \(\mathbb{X} =(X_t, t \in \mathbb{Z})\), which satisfy ``certain stationarity and weak dependence properties'', and investigate bootstrap methods to approximate the distributions of associated statistics. The two central bootstrap methods are the moving block bootstrap (MBB) and the tapered block bootstrap (TBB), with the latter being an improved version of the first down-weighting ``observations at the beginning and at the end of each resampled block'' thus improving ``the bias properties of the MBB''. In particular, the authors ``prove consistency of the MBB and of the TBB for the sample mean function''. Moreover, they show that ``these bootstrap methods provide consistent estimators of the covariance operator of the sample mean function estimator''. The authors also apply the bootstrap methods to an ``equality of the mean functions'' testing problem and provide ``numerical simulations and a real-life data example''. The paper will be interesting for researchers working in the area of mathematical statistics, in particular on theory for Hilbert-space valued stochastic processes. Its style is rigorous mathematical statistics. Some required assumptions on the processes seem strong, but the authors mention that for instance assumption 1 is ``satisfied by many linear and non-linear functional time series models considered in the literature''.