Research On Asset Pricing Based On Cluster Grouping

With the development of China’s economy,the scale of China’s capital market is becoming larger and larger,and it is necessary to continue to have a profound understanding of the financial market.Since the capital asset pricing model proposed by Sharpe(1964),many factors have been discovered to explain the return of assets,but they still can not explain the return of assets well.With the deepening of research,more and more factors are included in the scope of pricing research to explain various cross-section anomalies.The study of these factors is difficult to accurately explain the return of assets,that is to say,to some extent,there is a certain deviation in the pricing of stock returns using the observed data.This is because the use of observable data indicators to construct factors will be disturbed by data noise;And the significance of a single factor is also different in different historical periods.The residual item after fitting each factor combination using the traditional pricing model contains information that cannot be explained by the original pricing model.Therefore,this paper uses its time series correlation as a measure of the similarity of each factor to mine the hidden information between different factors that has not been explained by the original pricing model,and uses this information to cluster the data to obtain unobservable factors.Therefore,in order to study the impact of those unobservable factors on the return on assets,this paper uses a method of clustering a large number of financial time series based on high-dimensional panel data with grouping factor structure,and captures the similarity level of each time series by establishing an interactive fixed effect factor model,the sensitivity of observable factors and the sensitivity of unobservable factor structure.This method allows the correlation between observable factors and unobservable factors,as well as the cross section and sequence dependence and heteroscedasticity in the error structure.This paper first introduces this method,and then uses this method to carry out the corresponding empirical test on the abnormal factors of China’s stock market.Under the existing factor data,the MacroPCA estimation method is used to cluster and group a large number of factor data in the market,select the best estimation result through the corresponding model discrimination method,and then carry out relevant verification.The empirical result test found that,without grouping the data in advance,there are indeed unobservable group-specific factors that group the data,and these unobservable group-specific factors affect the return of assets,which is reflected in the different coefficients of the constant items before and after grouping,and the factor data containing potential factors after grouping is better than the factor data before grouping.Then we use cross section regression to test these potential factors,and find that most of the potential factors found by our method are significant.Secondly,we use the factor data after grouping to compare with the original attribution of the factor data before grouping,and find that the existence of these potential factors blurs the boundary between the known factors,affects the pricing of assets,and makes the known factors not pricing the return of stocks well.This also shows to some extent that the traditional asset pricing model ignores the impact of potential factors,and there are certain deficiencies in the accuracy of asset pricingThe idea of the model construction in this paper may be of great significance for the correct pricing of the return on assets.This paper uses the method of robust estimation to find out the unobservable potential factors on the one hand,and prove that the residual item contains factors that were not considered in the previous model.On the other hand,the method of clustering and grouping a large number of factors in the market and then asset pricing is a powerful supplement to the existing asset pricing model.