Blending News Text And Economic Policy Uncertainty To Forecast The Company's Unexpected Earnings

In recent years,researchers have explored the existence of Pose Earnings Announcement Drift(PEAD)and its causes,indicating that investors need to pay attention not only to the information disclosed in the company's announcement and grasp the value of the information conveyed in the company's announcement,but also to pay attention to more information,broaden their information perspective,and seize the opportunity to invest.Based on the above background,this paper takes listed companies in China's A-share market as a research sample,obtains relevant data from the first quarter of 2008 to the fourth quarter of 2020,and uses LSTM and transformer models on the basis of fundamental company data,respectively,and integrates the news text features quantified by XL-transformer model and economic policy uncertainty index to predict company The experiments show that adding news text features or economic policy uncertainty index can improve the model prediction accuracy,among which adding economic policy uncertainty index improves the model prediction effect slightly;news headlines have better prediction effect compared with news content,and news content can further improve the model prediction accuracy as a supplement to news headlines;adding news text features and then adding The addition of economic policy uncertainty index after adding news text features does not improve the model prediction accuracy,probably because the construction of economic policy uncertainty index also relies on text data,and there is information overlap between the two;meanwhile,the transformer model has better prediction effect compared with the LSTM model,but the improvement of the effect is limited;in short,both news text and economic policy uncertainty contain value that is not conveyed in the company fundamental data In short,both news text and economic policy uncertainty contain value information that is not conveyed in company fundamental data.This paper further uses the predicted values of the best model with the cumulative abnormal returns constructed by CAR method to prove the existence of PEAD anomalies in China's A-share market through both descriptive and regression analyses;and further analyzes PEAD anomalies in different quarters and different industries by considering the differences in PEAD anomalies in different quarters and different industries to provide support for investors to construct investment strategies at a more granular level.The results indicate that PEAD anomalies exist in China's A-shares as a whole,in different quarters and in different sectors,with similar movement trends in different quarters and in different sectors as the whole,most of which are under-reactive to information in the first 30 days,and investors react more strongly to bad news;the peak of the positive group in the first quarter and the electrical industry is higher than the overall one,which can reach 5% and 4% respectively;the second quarter and the electrical industry The bearish group peaks at around 30 days across the 0 axis,between0.5% and 1%.