Deep Learning for Algorithmic Trading: Enhancing MACD Strategy

Transaction automation has always received widespread attention in the field of financial research. As one of the most popular technical indicators of traders, the Moving Average Convergence Divergence(MACD) indicator sometimes performs worse than expected in unstable financial markets. In this paper, we use Residual Networks to improve the effectiveness of traditional trading MACD algorithm in technical analysis. The rationale behind our research is that deep learning networks can learn market behavior and be able to estimate whether a given trading point is more likely to succeed. We verify our strategy (MACD-KURT) which is based on the combination of Residual Networks prediction and technical analysis on CSI300 index constituent stocks in the Chinese market, and the results show that the strategy based on the combination of Residual Networks prediction and technical analysis is better than the one based on technical analysis alone, ether in strategy's return or risk control.