Machine Learning for Predicting Price Movements of Stocks

Abstract

Currently, about 80% of the stock market trades are from automated systems. With the rise of systematic trading and individual investing, there is significant opportunity to apply novel machine learning techniques to this space. Since stocks are incredibly volatile, even the slightest advantage in predicting their movement is helpful for financial analysts and real-world traders. The purpose of this project is to design and test various machine learning algorithms in predicting price movement of an asset. To simplify the problem for feasibility purposes, my goal was to accurately predict the percent change in price in the next timestep given a specific lookback period. Since stock prices can be defined as a time-series function, I mainly focused on using Recurrent Neural Networks (RNNs) given their proficiency in time-based predictions. I combined this network with technical indicators which are currently used by traders in the real world to refine this prediction.