Forecasting Regional GDP in Italy

This paper explores the usefulness of factor and bootstrap aggregation forecasting in predicting regional GDP in Italy. We use methods designed to target the set of potential predictors. We compute the mean square forecasting error (MSE) by using direct multi-step forecasts for the period 2004-2005. Our findings can be summarized as follows. First, factor and bagging forecasts generally show lower mean square forecasting error than the mean square error of the autoregressive AR(3) model used as a benchmark. Secondly, bagging methods seem to produce similar MSE as factor augmented models, especially for predicting aggregate GDPs. In synthesis, our analysis shows that using factors and bagging methods reduces the prediction mean squared error relative to standard forecasting methods