An Empirical Comparison of Methods for Forecasting Using Many Predictors

PDF531 KB

Abstract:

This paper provides a simple shrinkage representation that describes the
operational characteristics of various forecasting methods that are applicable when there
are a large number of orthogonal predictors (such as principal components). These
methods include pretest methods, Bayesian model averaging, empirical Bayes, and
bagging. We then compare these and other many-predictor forecasting methods in the
context of macroeconomic forecasting (real activity and inflation) using 131 monthly
predictors with monthly U.S. economic time series data, 1959:1 - 2003:12. The
theoretical shrinkage representations serve to inform our empirical comparison of these
forecasting methods.

Website

Last updated on 07/24/2012