Correcting Long-Side Failure in High-Frequency Equity Factors
Summary
The document discusses why high-frequency equity factors may appear strong in isolation but fail to improve stock-selection portfolios. Even after orthogonalization against common equity factors, a signal can have a high information coefficient and strong long-short returns while the relationship between factor values and future returns is weaker or different among long-side stocks.
To address this, it recommends reweighting groups when computing IC so that the long side receives more emphasis. It also describes modeling nonlinear factor-return relationships with higher-order polynomial terms or radial basis function expansions followed by a linear model, which can behave like piecewise regression. The summary reports that polynomial expansions improved optimized portfolio performance, but supplies no detailed empirical setup. It warns that added dimensions can reduce estimate stability, cause rank deficiencies that interfere with tracking-error constraints, and raise overfitting risk when monthly history is limited.
Key ideas
- Aggregate factor metrics can mask poor performance among stocks selected for long positions.
- Long-side-weighted IC can help screen factors for their intended portfolio use.
- Polynomial features can capture nonlinear links between factor exposures and future returns.
- Radial basis expansion with a linear model offers a data-driven piecewise approximation.
- Higher-dimensional models can become unstable, rank-deficient, or overfit limited data.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.