Simultaneous calibrated prediction intervals for time series

This paper deals with simultaneous prediction for time series models. In particular, it presents a simple procedure which gives well-calibrated simultaneous predictive intervals with coverage probability equal or close to the target nominal value. Although the exact computation of the proposed intervals is usually not feasible, an approximation can be easily obtained by means of a suitable bootstrap simulation procedure. This new predictive solution is much simpler to compute than those ones already proposed in the literature based on asymptotic calculations. An application of the bootstrap calibrated procedure to first order autoregressive models is presented.