Revisiting Useful Approaches to Data-Rich Macroeconomic Forecasting
In this paper it is shown theoretically and in simulations that under a variety of different unobserved factor structures, Partial Least Squares (PLS) and Bayesian regressions provide a better fit for a target variable relative to Principal Components (PC) regrssion. Empirically, PLS and Bayesian regressions usually have better out-of-sample performances than PC regression. Published in Computational Statistics & Data Analysis, 2016.