다요인모형의 주식수익률 예측 성과 비교

This study compared the portfolio predictive power of three models: the traditional Fama-French 3-factor and 5-factor models, and the dynamic factor model estimated by Kalman filter. In this study, six combined portfolios were constructed by selecting two variables among the four factor variables representing size, value, profitability, and investment, and then compared the prediction performance for the rate of return of the six combined portfolios using these three models. Each combined portfolio has 64 sub-portfolios in which each variable is divided into 8 categories according to the size of the factor variables. Each factor variable and portfolio return were obtained using stock returns considering both the monthly capital gains and expected dividend yields of KOSPI and KODAQ companies from July 1998 to March 2022. The predictive power of each model was compared through the Diebold-Mariano statistic. The results of the empirical analysis show that the dynamic factor model has excellent predictive power compared to the traditional Fama-French 3-factor and 5-factor models in all six combined portfolios. It was also found that Fama-French 3-factor model had better predictive power than the 5-factor model. The results of this study show that approaches that increase the number of factors do not always have good predictive power, and that dynamic factor models that consider the state variables have better predictive performance than traditional Fama-French models that rely on past return data.