Forecasting Italian Inflation with Large Datasets and Many Models

The aim of this paper is to propose a new method for forecasting Italian inflation. We expand on a standard factor model framework (see Stock and Watson (1998)) along several dimensions. To start with we pay special attention to the modeling of the autoregressive component of the inflation. Second, we apply forecast combination (Granger (2000) and Pesaran and Timmermann (2001)) and generate our forecast by averaging the predictions of a large number of models. Third, we allow for time variation in parameters by applying rolling regression techniques, with a window of three-years of monthly data. Backtesting shows that our strategy outperforms both the benchmark model (i.e. a factor model with does not allow for model uncertainty) and additional univariate (ARMA) and multivariate (VAR) models. Our strategy proves to improve on alternative models also when applied to turning point prediction