Macroeconomic forecasting with matched principal components

Abstract

This article proposes an improved method for the construction of principal components in macroeconomic forecasting. The underlying idea is to maximize the amount of variance of the original predictor variables that is retained by the components in order to reduce the variance involved in estimating the forecast model. This is achieved by matching the data window used for constructing the components with the estimation window. Extensive Monte Carlo simulations, using dynamic factor models, clarify the relationship between the achieved reduction in forecast variance and various design parameters, such as the observation length, the number of predictors, and the length of the forecast horizon. The method is also used in an empirical application to forecast eight key US macroeconomic time series over various horizons, where the components are constructed from a large set of predictors. The results show that the proposed modification leads, on average, to more accurate forecasts than previously used principal component regression methods.