| Preference measurement plays an important role in asset price pricing,portfolio construction and risk management.In fact,investors’ trading behavior is a reflection of their own risk preference.We hope to better study the basic rules of the capital market and explain market anomalies by studying investors’ risk preference or trading behavior.Existing in previous studies of various measurement methods of risk preference,most scholars for the measure of portfolio risk is based on macovei,mean-variance model and improved model,but based on high dimensional asset price to measure risk appetite is another kind of mainstream thinking,both at home and abroad a large number of studies have shown that asset prices contains the information of the appetite for risk.In this paper,the dependence of 28 industry blocks is studied based on the first level industry classification of Shenyin and Wanguo.The study is the dynamic interdependence between two industries,compared with simple correlation,dynamic interdependence is a big step forward.Therefore,DCC-GARCH model was used in this paper to analyze the pair-dependence between industries.Before that,the three-factor model was first used to model the data,and the residual was used to calculate the dependence between the two industries,and the conditional dynamic dependence between 378 industries was obtained.Two feature extraction methods,principal component analysis(PCA)and autoencoder,are compared,and the information content of the extracted index is measured.The empirical results show that the conditional dynamic dependence performance of 378 industries in this paper accurately reflects the interdependence between industries,and the effect of autoencoder is slightly better than that of principal component analysis in feature extraction of data,which solves the problem that it is difficult to extract effective information from high-dimensional data.Article proves that these indicators contains the information of the appetite for risk,and further proves that the measure index has rich market information and asset condition of extracting dynamic dependency factor compared to the popular macroeconomic forecasting factor,and fluctuation of market price has stronger ability to predict and explain ability,and of the dependence of the extraction to join a yuan linear model,It can significantly improve the prediction accuracy of univariate models with popular macroeconomic predictors. |
