Mean-Plus-Noise Factor Models: An Empirical Exploration

In this paper I propose an alternative factor model whose common factor follows a mean-plus-noise process. While the existing methods for factor models should be valid for this model, their performance may be misleading in small samples. Because the likelihood function of this model is not tractable, I estimate this model by Bayesian methods. When applied to U.S. macroeconomic time series, we find that the estimated mean-plus-noise factor is useful for predicting real personal income, industrial production and unemployment at long prediction horizons. ∗Address correspondence to: Department of Agricultural and Resource Economics, North Carolina State University, Campus Box 8109, Raleigh, NC 27695-8109. Email: [email protected]. Phone: 919-515-5969.