Forecasting The Direction Of Stock Return Based On Financial News

The research on the direction of stock returns is of great significance for stock investors,especially for retail investors in China.Direction prediction is actually a binary problem and Logistic model is usually used to predict the direction.However,the traditional parameter models are hard to avoid model misspecification.Non-parameter models can solve this problem well,and can better capture the features of the financial market,which is a complex system.Therefore,this paper studies the direction prediction of stock returns based on non-parametric method.With the development of economy,the Internet technology is also developed rapidly.The mining technology is no longer limited to the structured data.Heterogeneous data analysis technology is widely applied in various fields.In recent years,words such as "Financial big data" and"Internet finance" have emerged,and the internet platform has become a place for information exchange and expression of emotion.However,most of the information on the internet is in text form.In the financial field,financial news covers a lot of important information of the stock market,thus it is of great significance to study the direction of stock returns from the perspective of financial news.Harvey and Oryshchenko(2012)proposed a nonparametric kernel density theory,which was applied to the estimation of NASDAQ exponential density.The results show that the method can fully describe the characteristics of financial assets.Gu et al.(2018)proposed a time-varying factor weighted nonparametric density function model(F-TVF)based on the theory.In this study,a dictionary of financial news is constructed,based on which an emotional index is constructed,and the emotional index is applied to the F-TVF model.Due to the uncertainty of accuracy,this paper introduces the scoring rules as also a evaluation method.In order to reach the double optimal in the accuracy and the scoring rules,this paper constructs a new prediction model based on the scoring rules--the score weighted model.In the empirical study,the daily price data of the Shanghai Composite Index is applied.Logistic model,time-varying non-parametric density function model(TVF)based on time-varying density function estimation theory,time-varying factor weighted non-parametric density function model(F-TVF)with influence factor and scores weighted model are used to predict.The direction of stock returns is forecasted and the trading strategy is simulated.The results show that the financial news sentiment index is correlated with the stock returns,and the F-TVF model based on the sentiment index has a significant improvement compared with the traditional Logistic model and TVF model,while the scores weighted model has the best prediction effect.