Performance of Alternative Forecasting Methods for Setar Models

Preprint
Publisher Website
Google Scholar
    • preprint
      • Published in RePEc
Abstract
Five alternative forecasting methods used for SETAR modeling are compared with each other, and relative to mis-specified linear AR models, using Monte Carlo simulation. The results show that for forecasting beyond 1-step ahead, the method that uses Monte Carlo to generate forecasts out-perform the other five methods, when the SETAR model is assumed known. However, with parameter uncertainty the bootstrap method sometimes dominates the Monte Carlo method. The alternative forecasting methods are then used to generate multi-period forecasts of US GNP from the SETAR models, and these forecasts are compared to those from linear models. our results highlight the need for the forecast period to contain 'nonlinear features' if the nonlinear model is to out-perform the simpler linear model

This publication has 0 references indexed in Scilit: