Orthogonalizing Equity Factors in Multi-Factor Stock Selection
Summary
This research summary explains why correlated stock-selection factors can make portfolio exposures difficult to control. Factor correlations can vary over time, so even an equally weighted combination may carry an unintended concentration in a factor such as size. The report proposes removing linear overlap by regressing each factor on factors already included and using the residuals as orthogonalized factor values. Adding factors sequentially can produce a mutually orthogonal set.
The summary reports that orthogonalized portfolios outperformed their original counterparts on composite information coefficient and its information ratio, long-only excess return, portfolio information ratio, and relative monthly win rate across the factor sets discussed. Improvements from equal-weight models weakened as more factors were added, while a return-maximizing model appeared to lessen that effect. It also considers choosing the orthogonalization order dynamically using predictive-model R-squared; this approach helped, but less than a fixed order, and gains weakened when less effective factors were included. The evidence is reported qualitatively here, without the underlying tables or full methodology. Systematic market, liquidity, and policy risks remain relevant.
Key ideas
- Changing cross-sectional correlations can create unintended factor exposures in multi-factor portfolios.
- Regression residuals can remove a factor’s linear relationship to previously selected factors.
- Sequential orthogonalization was reported to improve several portfolio and factor metrics in the tested settings.
- A dynamic ordering based on predictive-model R-squared showed weaker gains than a fixed ordering.
- The summary omits detailed supporting data and notes market, liquidity, and policy risks.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.