Using Neural Network Models to Enhance a Novel Trend-Following Strategy for Market Index Investing

Abstract

Trend-following strategies that use purely technical indicators have been shown to be profitable in the long term. Choosing an optimal trend indicator and being able to detect and even forecast trends should further enhance these strategies. With these objectives in mind, we were able to achieve two major results in this study: (1) we developed a novel technical indicator – the 200-day simple moving median (SMM) – as well as a long-only, trend-following trading strategy that significantly and consistently outperformed the S&P 500 index in terms of average annualized return, Sharpe ratio, volatility, and maximum drawdown; and (2) we designed a neural network model, which was found to further enhance our strategy’s performance. Aside from these results, we also discovered that: (1) contrary to popular belief, a strategy that used the popular 200-day simple moving average indicator was not necessarily able to outperform the S&P 500 index; and (2) the neural network model we designed achieved higher prediction accuracy when it was used to predict long-term trends (i.e. the 200-day simple moving median) instead of next-day S&P 500 closing values.