Invested! Stock market prediction with Neural Networks

In this post, we attempt to find a way to systematically beat the market by applying a Bayesian Neural Network, an artificial neural network using Bayesian inference, and a Long Short Term Memory network (LSTM), a version of an artificial recurrent neural network. We apply these networks to predict the next-day prices or returns of the SPY ETF. These predictions determine if we should hold our shares or sell them. Through this process, we were able to construct an active strategy with a significantly higher Sharpe ratio than the Sharpe ratio of SPY over the same period. While we set out to achieve higher returns and less volatility than the market, we achieved similar returns but with far less volatility. The implications and process of outperforming the S&P 500 on a risk-adjusted basis will be discussed within the following analysis. Modern Portfolio Theory states that the best investment you can make is one into index funds unless you hold a consistent comparative advantage. These funds are essentially a proxy holding a portfolio of the whole market.