Inflation forecasting with rolling windows: An appraisal

We examine the performance of rolling windows procedures in forecasting inflation. We implement rolling windows augmented Dickey–Fuller (ADF) tests and then conduct a set of Monte Carlo experiments under stylized forms of structural breaks. We find that as long as the nature of inflation is either stationary or non-stationary, popular varying-length window techniques provide little advantage in forecasting over a conventional fixed-length window approach. However, we also find that varying-length window techniques tend to outperform the fixed-length window method under conditions involving a change in the inflation process from stationary to non-stationary, and vice versa. Finally, we investigate methods that can provide early warnings of structural breaks, a situation for which the available rolling windows procedures are not well suited.