Orthogonal Alpha Factors and Risk-Aware Factor Weighting
Summary
This research note discusses how to turn equity alpha signals into portfolio forecasts and combine multiple factors. It suggests converting factor z-scores into expected returns using average results from repeated cross-sectional regressions, and favors Spearman rank correlation for information-coefficient testing because its significance assessment does not require normally distributed variables. Whether to neutralize factor exposures should depend on the portfolio’s actual risk controls. The note reports that, in its A-share tests, purified alpha information performed better than risk-adjusted IC, partly because the latter favored larger stocks.
To reduce redundant signals and collinearity, the proposed incremental orthogonalization process screens factors and uses the resulting orthogonal residual factors for IC-to-IR weighting. An alternative is to weight the screened factors’ original z-scores directly, though that approach is more sensitive to factor count and correlation and should be tested for each factor library. The note argues that some robust optimization methods may sacrifice too much alpha and fit poorly with markets where shorting and leverage are difficult. It provides conclusions rather than detailed empirical tables, and flags model failure and extreme-market shocks as risks.
Key ideas
- Alpha forecasts depend on which risk exposures the portfolio controls.
- Spearman rank correlation is presented as a more robust choice than Pearson correlation for factor IC significance testing.
- Risk neutralization should match the portfolio’s exposure controls rather than be applied indiscriminately.
- Screening factors and weighting orthogonal residuals by IC-to-IR can reduce redundancy and collinearity.
- Weighting original factor scores may work, but its sensitivity to factor count and correlation calls for separate testing.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.