Forecasting in data-rich environments

Abstract

Stock and Watson (1998 and 1999) developed a factor-model approach which allows for big data sets to be systematically reduced to a few explanatory factors. In this paper two other methods are proposed. The first one, Partial Least Squares is imported from the Chemometrics literature. The second one, which is based on the Combination of Forecasts literature is a modification of Stock and Watson’s method. We will call this method Principal Components Combination. These methods are compared in an empirical application to inflation. We conclude that the method with the best overall performance is the Principal Components Combination.