by Shihao Gu, Dacheng Xiu
Compares machine-learning methods for estimating asset risk premia and finds that trees and neural networks capture nonlinear interactions among signals including momentum, liquidity, and volatility.
Summary by SelfMadeQuant from the paper's listing. Check the original for exact claims.
Continue on SelfMadeQuant
Start with the foundations, then use the reading focus to connect the paper to further study.
Preserve time order and avoid future information in training.
Interpret exposures, residuals and uncertainty in fitted coefficients.
Work with returns and recognise their empirical limitations.
Account for dependence and overlapping observations.
Examine which information contributes to predictions.
Connect factor definitions to portfolio construction.
Learn statistical learning, establish simple baselines, and evaluate financial predictions without leaking the future.
Use an auditable backtest before evaluating a learned signal.
Membership required