Multi-Source Sentiment Integration with Machine Learning for A-Share Earnings Forecast Windows: A Proposed Implementation Framework

Abstract

In China A-share market, to make Alpha generation, we should not only rely on traditional factor models, but also learn to use those wrong pricing driven by emotions during earnings forecast windows. At this time, there are many retail investors in the market and the information is asymmetric. We reviewed 42 intellectual studies and put forward a unified method framework. With the knowledge of behavioral finance, market microstructure and machine learning, we have built a four-layer implementation pipeline, which integrates three different emotional sources: textual disclosures, order flow and social media, and uses advanced ML architectures. These studies found that hybrid architectures performed better than the single method in the accuracy and risk index of direction prediction. These benefits are mainly concentrated in the high-concern forecast window with the largest emotional fluctuation. However, there are two restrictions in actual operation: the handling fee of 0.15-0.25% per transaction will reduce the total Sharpe ratio by 35-50%, so the minimum threshold must reach 1.5; Moreover, the performance of the strategy depends on the market situation-78% can make money in a bull market and only 45% in a bear market. This study provides practical suggestions for quantitative traders, including how to capture emotional signals more carefully, choose the appropriate machine learning algorithm according to different time, and how to reduce transaction costs and adapt to market changes.