Deep Learning for Multi-factor Models in Regional and Global Stock Markets with Interpretability

Abstract

Many studies have been undertaken with machine learning techniques to predict stock returns in terms of time-series prediction. However, from the viewpoint of the cross-sectional prediction, there are few examples that verify its profitability in regional and global stock markets. Recently, a method known as deep learning, which achieves high performance mainly in image recognition and speech recognition, has attracted attention in the machine learning field. In this paper, we examine the effectiveness of stock return prediction in the cross-section based on a multi-factor model using deep learning in regional and global stock markets. The result shows that deep learning models outperformed representative machine learning models in terms of risk-adjusted return in both regional and global stock markets. In addition, we present the application of layer-wise relevance propagation (LRP) for deep learning models to decompose attributes of the predicted return and determine which factor contributes to prediction.