Blending News Text and Economic Policy Uncertainty to Forecast the Company’s Unexpected Earnings

Abstract

Employing Chinese A-share market data, this study explores how news text and economic policy uncertainty (EPU) can be combined to predict a company’s unanticipated earnings using the XL (extra long) Transformer and long short term memory (LSTM) models. The results show that adding news text features or the EPU index can improve the model’s predictive performance. However, adding the EPU index improves the model prediction performance by a tiny amount. Next, news headlines have better predictive performance relative to news content. Meanwhile, as a supplement to news headlines, news content can further improve predictive performance. Finally, the XL-Transformer model has better predictive performance than the LSTM model, but the improvement in the effect is limited.