Can tree-structured classifiers add value to the investor?

We analyse the investor welfare gain of including tree-structured classifiers’ predictions about the relative performance of stock vs. cash. The CART, bagging, and random forest methods select the VIX level and momentum, the earning bond yield level and momentum, and the detrended risk-free rate as the most important state variables to predict the outperformance of the S&P 500 vs. cash out-of-sample. These tree-structured classifiers’ predictions are used as a binary state variable to estimate optimal investor portfolios that also deliver out-of-sample higher Sharpe ratios and certainty equivalent return gains than competing portfolio strategies that exclude them.