Estimating cross-section common stochastic trends in nonstationary panel data

Abstract

This paper studies large-dimension factor models with nonstationary dynamic factors, also referred to as cross-section common stochastic trends. We consider the problem of estimating the dimension of the common stochastic trends and the stochastic trends themselves. We derive the rates of convergence and the limiting distributions for the estimated common trends and for the estimated loading coefficients. Generalized dynamic factor models with nonstationary factors are also considered. Cointegration among the factors is permitted. The method is applied to the study of employment fluctuations across 60 industries for the U.S. We examine the hypothesis that these fluctuations can be explained by a small number of aggregate factors. We also test whether some observable macroeconomic variables are the underlying factors.