Semiparametric Block Bootstrap Prediction Intervals for Parsimonious Autoregression

This paper investigates the research question of whether principle parsimony carries over into interval forecasting, and proposes new semiparametric prediction intervals that apply block bootstrap to first-order autoregression. The AR(1) model is parsimonious in which error term may be serially correlated. Then, utilized resample blocks consecutive observations account for serial correlation. Monte Carlo simulations illustrate that, general, proposed outperform traditional based on nonparsimonious models.