Classification outperforms regression throughout matched machine studying fashions in portfolio building. A stacking ensemble of gradient boosted timber, random forest, and neural community yields a value-weighted annualized Sharpe ratio of two.08 for classification and 1.39 for regression. This outperformance strengthens with class granularity and persists throughout subsamples and after transaction prices. Spanning assessments present that classification retains economically giant alphas after we management for regression, whereas regression alphas shrink considerably as soon as we management for classification. These outcomes point out that classification extracts extra return info than matched regression. Our diagnostics hint classification’s benefit to extra exact separation of return deciles.


