Working Paper: NBER ID: w30217
Authors: Bryan T. Kelly; Semyon Malamud; Kangying Zhou
Abstract: Much of the extant literature predicts market returns with “simple” models that use only a few parameters. Contrary to conventional wisdom, we theoretically prove that simple models severely understate return predictability compared to “complex” models in which the number of parameters exceeds the number of observations. We empirically document the virtue of complexity in US equity market return prediction. Our findings establish the rationale for modeling expected returns through machine learning.
Keywords: Return Prediction; Machine Learning; Portfolio Construction
JEL Codes: C1; C45; G1
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| model complexity (C52) | expected out-of-sample forecast accuracy (C53) |
| model complexity (C52) | portfolio performance (G11) |
| increased model complexity (C52) | improved performance (D29) |
| high-complexity regime (p > t) (P10) | expected out-of-sample forecast accuracy (C53) |
| high-complexity regime (p > t) (P10) | portfolio performance (G11) |
| model complexity (C52) | predictive outcomes (C52) |
| high-dimensional models (C52) | outperform simpler models (C52) |
| nontrivial shrinkage (C24) | enhance Sharpe ratio (G11) |