An application of deep learning for trade signal prediction in financial markets

We know algorithms for predicting price movement direction and time are in practical use, despite being disputed at a theoretical level. In this study, we analyze the benefits of various machine learning algorithms to the price movement direction prediction problem, on selected stocks from the U.S. stock markets. To this end, we generate an array of features known to be beneficial in technical analysis of securities, and show the efficacy of several supervised learning methods. Lastly, we demonstrate that Stacked Denoising Auto-Encoders, an example of “deep learning” that has grown popular in recent years, yields significant prediction power.