Skip to content
All library documents

Empirical Comparison of Six Methods for Combining Equity Factors

Article BigQuant
Read the original

Summary

This report compares six ways to combine equity factors: equal weighting, historical factor returns, historical Rank IC, maximum IC, maximum IC_IR, and principal component analysis. It applies the methods within six factor groups in a Chinese A-share framework, evaluating them with cross-sectional regressions, Rank IC and IC_IR, and layered portfolio tests. Factor exposures are processed with outlier handling and industry and market-cap neutralization; portfolio tests use monthly rebalancing. The report finds that maximum IC_IR and maximum IC often improve on equal weighting, while equal weighting and the first principal component tend to produce more stable composites. Historical return weighting is effective in some groups.

The evidence is a historical study covering April 2007 through November 2018, so the rankings may change in other market conditions. The report highlights estimation risks: correlated factors can make covariance estimates unreliable, and short lookback windows provide weak inputs. It recommends choosing a method according to factor relationships and notes that factor combination is optional. A twelve-month window performs well in many tested combinations, but other windows sometimes fare better.

Key ideas

  • The report evaluates equal weighting, historical return and IC weighting, maximum IC and IC_IR optimization, and PCA.
  • Maximum IC and maximum IC_IR generally improve selected performance measures relative to equal-weight composites in the tested factor groups.
  • Equal weighting is inherently stable, while the first principal component is often the most stable non-equal-weight composite.
  • Maximum IC_IR optimization depends on estimates of future IC and the IC covariance matrix, which can be unreliable with short histories or highly correlated factors.
  • A twelve-month lookback works well in many cases, but the best window and combination method vary by factor group.
  • The findings summarize historical A-share tests and do not establish future performance.

Tags

Factor investingPortfolio constructionStatisticsBacktestingChina markets

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.