Title: Bootstrap prediction intervals for weighted SETAR forecasts
Authors: Francesco Giordano - University of Salerno (Italy)
Marcella Niglio - University of Salerno (Italy) [presenting]
Abstract: The generation of prediction intervals is not always an easy task, mainly when the interest is focused on nonlinear time series models whose predictor distribution is not standard. After presenting a new predictor for the Self Exciting Threshold AutoRegressive (SETAR) model, we focus the attention on the generation of the prediction interval. In more detail, the new predictor, that we call weighted SETAR predictor, is obtained as a weighted mean of the past observations. The weights are obtained from the minimization of the Mean Square Forecast Errors (MSFE). We propose a residual bootstrap procedure to build prediction intervals that are then compared to other approaches largely used in the literature though a Monte Carlo study. In particular, the coverage of the different prediction intervals is examined, considering SETAR models with increasing degree of complexity and nonlinearity.