B0155
Title: Resampling techniques for dependent functional data
Authors: Han Lin Shang - Australian National University (Australia) [presenting]
Abstract: The bootstrap is a useful method in functional data analysis for estimating the distribution of an estimator or test statistics by resampling data or a model estimated from data. However, the bootstrap validity heavily depends on whether the functional data are generated from an independent random variable or a time series. The work is concerned with the application of the bootstrap to functional time series. We extend some commonly used methods that have been proposed in the univariate time series literature to functional time series context, including Markovian bootstrap, nonparametric residual bootstrap, and the residual bootstrap for parametric models. Illustrated by a series of simulation studies and real-world applications, we examine the estimation accuracy of an estimator obtained from different resampling techniques using the notion of interval score. We argue that methods for implementing the bootstrap with functional time series are not as well understood as methods for independent functional random variables.
