Can a trading strategy based on predictions from a nonlinear support vector machine outperform a passive investor holding the S&P500 index?

Abstract

In this empirical research, we compare the forecasting performance of a supervised Support Vector Machine technique to a passive buy-and-hold strategy on the S&P500 index. By introducing two investment strategies, we find evidence that the application of a nonlinear Support Vector Machine can be superior to linear regression models, as well as to a passive buy-and-hold strategy. The Support Vector Machine model generates both excess returns and reduced volatility for the period between 2013 to 2019. However, when comparing the prediction results of a Support Vector Machine model to that of a linear regression model during the Great Recession, the results are ambiguous, although both models have proven to explicitly outperform the passive buy-and-hold approach.