Time-series Bootstraps

Bootstraps for time-series data come in a variety of forms. The three contained in this package are the stationary bootstrap (StationaryBootstrap), which uses blocks with an exponentially distributed lengths, the circular block bootstrap (CircularBlockBootstrap), which uses fixed length blocks, and the moving block bootstrap which also uses fixed length blocks (MovingBlockBootstrap). The moving block bootstrap does not wrap around and so observations near the start or end of the series will be systematically under-sampled. It is not recommended for this reason.

StationaryBootstrap(block_size, *args, **kwargs)

Politis and Romano (1994) bootstrap with expon distributed block sizes

CircularBlockBootstrap(block_size, *args, ...)

Bootstrap using blocks of the same length with end-to-start wrap around

MovingBlockBootstrap(block_size, *args, **kwargs)

Bootstrap using blocks of the same length without wrap around


Estimate optimal window length for time-series bootstraps