Online Program

Bootstrapping Time Series Data (303052)

Keywords: bootstrap, time series, R

The bootstrap is a powerful technique for estimating statistical parameters from sampled data. Applying the bootstrap to time series data presents unique problems, however, and requires specialized techniques. The central problem is that most time series data contain some correlation structure—weak or strong, known or unknown. The bootstrapping algorithm must honor that structure. A naive bootstrap does not, but, fortunately, there are several alternative algorithms that can preserve and explore the structure. This presentation will explain and demonstrate bootstrapping algorithms for time series data. Topics will include review of bootstrapping concepts and basic algorithms, special issues in bootstrapping time series data, the naive bootstrap algorithm, the block resampling algorithm, bootstrapping parametric models and hierarchical models of time series data, the maximum entropy bootstrap, and bootstrapping for seasonal data. All concepts will be illustrated with R code and graphical plots, using real-life data where possible.