Research On The Impact Of Dependency Structure On Stock Earnings Forecasts

Traditional financial investors are perfectly reasonable and assume that investors are risk averse.Behavioral finance theory holds that investors’ trading behavior is mainly affected by risk preference.In the capital market,the degree of investors’ risk preference is affected by many factors.Under the influence of macro-economy,market environment,corporate characteristics and investor sentiment,it produces individual differences and heterogeneity,and then affects investment returns and individual stock prices.The change of investors’ risk preference is used to explain various uncertain phenomena in the asset market,and has a significant impact on stock returns and asset prices.Therefore,measuring risk appetite is of great significance to asset pricing.Recently,some studies have gradually shown that there is an obvious correlation between cross market and cross variety dependence and investors’ risk preference,which provides a certain basis for us to measure risk preference from the perspective of asset dynamic dependence,but there are also obvious limitations.The main point is that these studies do not take into account high-dimensional dependencies,only two or a few assets in the measurement of risk appetite.These years,with the development of financial market and the popularity of procedural trading,the asset allocation of portfolio is becoming more and more complex,which makes the scope of asset dependence more extensive.So,that’s why we want to research asset dependence from a high-dimensional perspective.In view of this,this paper will further discuss the correlation between dependence and risk preference on the basis of measuring the dynamic dependence of high-dimensional assets.This paper predicts stock returns——based a graph autoencoder approach,using the stock price and characteristic information.Using the graph data structure,the dependence between stocks and the dependence between features are retained.Among them,the dependence between stocks adopts the excellent representatives of measuring high-dimensional dynamic dependence: dcc-garch model and copula model.The models and experiments in this paper give us results like this: the AUC of the graph self encoder(vgae)model proposed is 0.9322,which is better than other machine learning methods.It is confirmed that the model with dependency can better predict the return of stocks.At the same time,it is concluded that using graph data structure can better extract data features and better predict the return of stocks.