Deep Learning from Stock Index Fundamentals

This paper employs a deep learning approach for linking stock market fundamentals to trading signals via neural networks. With an average accuracy of ~54% the model predicts whether markets go up or down on the subsequent trading day. Coupling the prediction to a binary long/cash strategy yields superior return characteristics and significant outperformance against various stock market indices. Predictability is detected for 28 country indices by including local and global fundamental data. I further find stronger predictive power of global fundamentals as compared to local ones