Comparison Of The Selection Methods Of Approximate Factor Model Factors And Their Simulation

In the context of the big data era,the data dimension has grown geometrically.The factor model is a good choice for dealing with high-dimensional data.The selection of the number of factors is always a hot issue.In recent years,a large number of factor selection methods have been frequently proposed by statisticians.In this paper,Monte Carlo simulation test is used to compare the selection methods of the four static approximate factor model factors.Based on the two adjacent eigenvalue ratios,three new methods for selecting the number of factors are proposed,named WR,IER1 and IER2,respectively.The WR combines the advantages of the information criteria ER and GR and weights them to average.The last two methods take into account that there may be dominant factors in the common factors.The comparison of neighboring eigenvalues may lead to underestimation of the number of factors.Therefore,a function is used to compress the eigenvalues towards 0 to reduce the different eigenvalues.The huge gap between them to avoid underestimating the situation.Monte Carlo simulations show that the new criteria,especially the latter two when there are dominant factors,have better robustness and adaptability than other criteria.Lastly,the 367 companies listed on the A-shares were listed on the market and 328 consecutive trading days were used to build a balanced panel data set.By comparing the number of different factor selection methods,we concluded that there are two common factors.