Improved frequentist prediction intervals for ARMA models by simulation

Abstract

[Introduction] In a traditional approach to time series forecasting, prediction intervals are usually computed as if the chosen model were correct and the parameters of the model completely known, with no reference to the uncertainty regarding the model selection and parameter estimation. The parameter uncertainty may not be a major source of prediction errors in practical applications, but its effects can be substantial if the series is not too long. The problems of interval prediction are discussed in depth in Chatfield (1993, 1996) and Clements & Hendry (1999). [Continues; please see the article]