Sieve Bootstrap Prediction Intervals for Contamined BIP-ARMA Processes

In this paper we present the construction of prediction intervals for time series based on the sieve bootstrap technique, which does not require the distributional assumption of normality that most parametric techniques impose. The construction of prediction intervals in the presence of innovation outliers does not have distributional robustness, leading to undesirable increase in the length of the prediction interval. In the analysis of financial time series it is common to have irregular observations that have different types of isolated and group outliers. For this reason we propose the construction of prediction intervals based in the winzorised residuals and bootstrap techniques for time series. The algorithm used for the construction of prediction interval is based in the AR-sieve bootstrap technique for non-parametric linear models. This method is compared using a Monte Carlo study with other proposal recently published in the literature obtaining favorable results in terms of a metric based in the interval length and coverage.