A Monte Carlo comparison of estimating the number of dynamic factors

Based on a Monte Carlo simulation, this study compares the finite sample performance of five of the most widely used methods for estimating the number of dynamic factors. The simulation results show that although the performance is affected by the data generating process, the methods proposed by Hallin and Lika (J Am Stat Assoc 102(478):603-617, 2007) and Bai and Ng (Bus Econ Stat 25(1):52-60, 2007) generally outperform the others. Specifically, Amengual and Watson's (J Bus Econ Stat 25(1):91-96, 2007) method is sensitive to cross-sectional correlation, and Breitung and Pigorsch's (Oxf Bull Econ Stat 75(1):23-36, 2013) estimator is sensitive to the overestimation of the number of static factors. The results of this study are further supported by an empirical application to a Chinese macroeconomic dataset.