Factor attribution disaggregates a portfolio's total return into the portion explained by exposures to systematic risk factors (market, size, value, momentum, quality, etc.) and the portion that is idiosyncratic — unexplained by those factors and therefore genuine alpha.
Without factor attribution, a strategy that simply tilts toward well-known premia (e.g., value stocks) will show strong backtested returns that are compensation for systematic risk exposure, not skill.
Attribution frameworks
- Brinson-Hood-Beebower — asset-class level allocation vs selection decomposition
- Fama-French factor model — market, SMB, HML as explanatory factors; the time-series residual is alpha (Jensen's alpha)
- Barra model — commercial, finer-grained decomposition using style, industry, and country factors across many periods
- Signal-level attribution — tracing active returns to each individual signal in a composite forecast, using IC contributions or Shapley values to identify which inputs drove performance