Robo-advisors: Machine learning in trend-following ETF investments

Abstract

We examine an application of machine learning to exchange traded fund investments in the U.S. market. To find how the changes in exchange traded fund prices are associated with expected market fundamentals, we propose three parsimonious risk factors extracted from various U.S. economic and market indicators. Based on the information set including these three factors, we build a predictive support vector machine model that can detect long or short investment signals. We find that the high probability of an upward momentum from our forecasting model suggests a long exchange traded fund signal, whereas the low probability of a downward momentum indicates a short exchange traded fund signal. We further design an algorithmic trading system with the support vector machine factor model. We find that the trading system shows practically desirable and robust performances over in-sample and out-of-sample trading periods.