Artificial Intelligence for Long-Term Investing

Abstract

This paper presents the results of using a novel Artificial Intelligence (AI) model for long-term investing. The AI model takes various financial data as input signals and tries to determine an optimal portfolio allocation. In these experiments, the AI model considers the stocks of 40 US companies, as well as the S&P 500 index and US government bonds with one-year maturity. The portfolio is rebalanced annually. Between 1995 and 2015, the equal-weighted rebalancing of these 42 assets outperformed the S&P 500 by 5-6% (percentage points) per year on average. The AI model outperformed the equal-weighted rebalancing by 12-13% (percentage points) per year on average, and the AI model outperformed the S&P 500 by about 18% (percentage points) per year on average. It is uncertain and probably unrealistic that this performance advantage of the AI model will continue in the future, but it seems feasible that some combination of AI models could work reasonably well for long-term investing (aka. low-frequency trading).

Keywords: Portfolio Optimization, Artificial Intelligence, Machine Learning