How do you decompose alpha to distinguish genuine manager skill from luck and factor exposure?

Alpha decomposition is the process of separating a manager's excess return into its component sources to determine how much (if any) represents genuine investment skill. The key insight is that not all alpha is created equal.

Sources of Apparent Alpha:

1. Factor Exposure (Systematic Alpha) — Returns from tilting toward factors like value, momentum, size, or quality. These can be replicated cheaply with passive factor funds, so they don't represent unique skill.

1. Genuine Security Selection (Idiosyncratic Alpha) — Returns from picking individual securities that outperform their factor-expected returns. This is the 'real' skill component.

1. Timing Alpha — Returns from correctly varying factor exposures over time (increasing value tilt before value rallies)

1. Luck — Random variation that appears as alpha over short periods but doesn't persist

The Decomposition Framework:

Total Active Return = Factor Alpha + Selection Alpha + Timing Alpha + Residual

Using a multi-factor regression:

R_p - R_f = α + β₁(Market) + β₂(Size) + β₃(Value) + β₄(Momentum) + ε

The regression intercept (α) after controlling for factor exposures represents the manager's true security selection skill. If the pre-regression alpha was 3.0% but post-regression alpha is only 0.5%, most of the apparent outperformance came from factor tilts.

Practical Example:

Manager at Whitfield Capital reports 2.5% annual alpha over the S&P 500 over five years.

Factor analysis reveals:

• Value tilt (overweight cheap stocks): +1.2% contribution
• Small-cap tilt: +0.6% contribution
• Momentum exposure: +0.3% contribution
• True selection alpha (regression intercept): +0.4%

So 84% of the apparent alpha came from factor exposures that could be replicated with a $30/year factor ETF, and only 0.4% represents genuine skill.

Testing for Luck vs Skill:

• Statistical significance — Is the alpha t-statistic above 2.0? With five years of monthly data (60 observations), a 0.4% annual alpha may not be statistically different from zero.
• Persistence — Does the alpha persist out-of-sample? Split the track record into halves and test whether first-half alpha predicts second-half alpha.
• Breadth — A manager with consistent small alphas across many positions (high breadth) is more likely skilled than one with a few lucky concentrated bets.

Exam Application: Expect vignettes showing a manager's returns alongside factor returns, and questions asking whether the alpha is genuine or factor-driven. Also expect questions on whether a short track record of outperformance is statistically meaningful.

Deepen your performance evaluation expertise in our CFA Level III community.