Modeling the distribution of key economic indicators in a data-rich environment: new empirical evidence

Abstract

This study explores the ability of a large number of macroeconomic variables to forecast the mean, quantiles and density of key economic indicators. In the baseline case, we construct the forecasts using an autoregressive model. We then consider several general specifications that augment the time series model with macroeconomic information, either directly using the full set of predictors, through targeted-factors, targeted-predictors or forecast combinations. Our findings suggest that aggregating information across quantiles leads to improved estimates of the conditional mean. Overall, augmenting the autoregressive model with macroeconomic variables through methods that perform variable selection or account for non-linearities improves predictive performance. This increase in out-of-sample performance arises from the improved estimation of the lower and middle part of the distribution.