Heterogeneity Research And Application Of Fama-French Three Factor Model Based On Machine Learning Method

China's securities market has made great progress in the great historical process of reform and opening up.From the official launch of gem in 2009 to the launch of scientific innovation edition and pilot registration system in 2019,China has completed the establishment of multi-level capital market,and the domestic capital market has entered a new cycle of accelerating gear shift.Based on the pivotal role of the securities market,asset pricing has long been a hot issue in the investment analysis of the securities market.An effective asset pricing model is of great significance for asset allocation,risk management and post investment evaluation.Among them,the influencing factors of stock return have attracted scholars' attention.The CAPM model proposed by Sharp reflects the pricing relationship between the return of a single security and the return of a market portfolio under the condition of market equilibrium.It is the pillar of modern financial market price theory and has strong theoretical significance,but it is not enough to explain the empirical results.In order to make up for this defect,Fama&French come up with the famous three factor theory,and in this theory,they pointed out that the market risk factor,market value factor(SMB)and book to market ratio factor(HML)have more significant explanatory power on excess return than CAPM.After that,some scholars constantly optimize and improve the model to make it glow with new vitality.This article changes the traditional research thinking of supplementing the factors in the three-factor model,and proposes to improve and optimize the factor coefficients of the three-factor model.First,the factor coefficients of the three-factor model are optimized from constants to time-varying functions,and theoretical derivation is carried out.Secondly,based on the real monthly data of China's capital market from 2000 to 2019,the coefficient function in the factor model is estimated and characterized,which confirms the rationality and effectiveness of the coefficient function of the factor,which is conducive to improving the factor model's impact on the securities market.The explanatory power of the excess return rate is to construct a three-factor model that is more suitable for the Chinese market.Finally,based on the three types of coefficient functions in the obtained three-factor model,the paired distances between stocks are measured,and the group recognition clustering algorithm in machine learning is used to classify securities,reducing the selection when constructing investment portfolios.The workload of stocks provides theoretical basis and data support for investors' investment decisions.The main research conclusions of this paper are as follows:(1)The factor coefficients in the three-factor model change with time.Optimizing the coefficients in the three-factor model into a time-varying function is more in line with the actual situation of the Chinese stock market;(2)The coefficient function of each stock has different forms,including constant type,diagonal type,curve type,and Oscillating type,so the change of the factor coefficient function in the time series varies from stock to stock.(3)The market risk factors,market value factors,and book-to-market value ratio factors in the three-factor model have different interpretations of the excess returns of different stocks,at the same time,these three factors also explain the excess returns of a given stock.It changes with time,so in the domestic research field of factor model applicability,there will be differences in conclusions due to different sample stock selections and different research periods.(4)The optimized three-factor model can be widely used in stock classification,and can simplify the process of investment portfolio construction through group recognition clustering algorithm.