SETAR model selection -- a bootstrap approach (Q2488425)

The authors propose a technique of order selection in the self-exciting threshold autoregressive (SETAR) time series models which is based on the bootstrap estimate of the true prediction power of the model. This approach is compared with some versions of Akaike's information criteria (AIC) via simulations. Application to the classical Canadian lynx data is also considered. The authors conclusion is that an unbiased bootstrap procedure outperforms AIC.