The Virtue of Complexity in Return Prediction

The paper theoretically proves that simple models with few parameters understate return predictability compared to complex models where parameters exceed observations.

It empirically documents this 'virtue of complexity' in U.S.

equity market return prediction across various asset classes, advocating for machine learning in expected return modeling.

The findings challenge parsimony principles and demonstrate improved out-of-sample performance and Sharpe ratios with higher model complexity.