Micro Alphas
提出微阿尔法概念，即单个不显著的弱信号组合后可预测股权风险溢价，并基于弹性网络模型构建策略，在公共基金中实现超额收益。
The authors introduce the concept of micro alphas: weak signals that may not reach statistical significance individually but can generate predictability for the equity risk premium when combined. Unlike traditional predictability studies that require predictors to demonstrate persistent statistical significance over the full sample period, the authors argue that such a requirement is unnecessarily restrictive. More recent work has adopted a more nuanced approach, but candidate variables are still typically evaluated against criteria that only partially reflect their potential contribution in combination with other signals. Borrowing from the machine learning literature, the authors apply feature transformations and selection within a cross-validated walk-forward framework to capture potentially nonlinear and time-dependent effects. This leads to a strategy based on an elastic net model that aggregates these signals. The strategy has been implemented in a public fund and has delivered consistent excess returns relative to the S&P 500. Potential extensions and improvements to the model are discussed.