An Interpretable Framework for Stock Trend Forecasting

Abstract

Stock trend forecasting plays a critical role when investing in the stock market. Comparing with traditional technical analysis and fundamental analysis, deep learning models own better forecasting performance. However, the poor interpretability of deep learning models brings lots of limitations to its practical application since the lack of interpretability increases the investment risk. In this paper, we propose a graph-based framework, which owns good interpretability while maintaining forecasting performance. Specifically, the framework explains the stock returns by dividing the returns into multiple parts: 1) the part related to individual stock; 2) the part related to company business; 3) the part related to the corresponding industry; 4) the part associated with the whole market. Extensive experiments on real-world Chinese stock market data have demonstrated the effectiveness of our proposed framework for stock trend forecasting. Afterward, we try to illustrate the interpretability of our framework.