Selection Of The Latent Factor Number Of High Dimension Panel Data Model And Its Robustness Analysis

High Dimension latent factor models gains its popularity in economic,financial analysis, ecology statistics,medical statistics due to its advantage of simplifying the analysis of large scale data, while to correctly specifying the number of factors (r) is still a fundamental issue for the application of factor models. In recent years,there have been a rapidly growing literature about the determination of the number factors published journals of statistics and econometrics.In this paper, a comparative issue about finite sample properties of three common static latent factor estimation method of panel data is studied by Monte Carlo numerical simulations. By the way,we propose two methods which are based on the ratios of two adjacent estimated eigenvalues.Monte Carlo simulation illustrates that these two estimators are more robustness than others in weak factors existing case. In consideration of the above criteria assumes that the number of factors does not change as the sample size changes,so we have developed guidelines for the above information in order to adapt the data under the case of variable factors. Finally, take the CSI300Index constituent stocks which have226return series of288days for example to construct the balance panel data set, and then investigate the factor numbers of CSI300stocks yield with above methods.We conclude that two factors which are highly relevant with CSI Index300explain the most proportion of series variation. Further research with the criterion we have developed in this paper imply that no economic structure changes exists in our sample data during the study period.