In the Capital Asset Pricing Model, alpha is the intercept of the return regression: Alpha = Portfolio Return - (Risk-Free Rate + β × Market Excess Return). If a portfolio returns 15% in a year when the market returns 12% and the risk-free rate is 4%, with a beta of 1.0, the alpha is 15% - (4% + 1.0 × 8%) = 3%. This 3% represents return from factors other than market exposure — potentially skill, research edge, factor tilts not captured by beta, or luck.

Multi-factor alpha is more demanding and more accurate. Measuring alpha relative to Fama-French five-factor model exposure (market, size, value, profitability, investment) or including momentum as a sixth factor eliminates the inflation of apparent alpha from known factor exposures. A manager generating strong returns primarily from loading on value and momentum factors is not generating true alpha — they are generating factor returns that any rules-based strategy could capture at lower cost. True alpha persists after controlling for all known systematic risk factors.

Genuine alpha in equity markets typically comes from one of four sources: informational edge (access to, or superior interpretation of, legally available information), analytical edge (superior financial modeling or industry expertise), behavioral edge (exploiting systematic mistakes of other investors), or structural edge (transacting where large institutional players cannot — micro-cap stocks too small for institutions, special situations too complex for generalist funds).

The informational source has been substantially compressed by Reg FD (which eliminated selective disclosure by public companies to favored analysts), algorithmic trading that instantly prices public information, and satellite/alternative data that has been commoditized. Analytical edge — building genuinely superior financial models — remains available but requires deep domain expertise. Behavioral edge is the most durable source: systematic investor errors (overreaction, anchoring, recency bias) persist because they are rooted in human psychology, not information processing.

Distinguishing genuine skill from luck requires long observation periods. With annual performance data, 5 years of positive alpha provides weak statistical evidence of skill — the confidence interval is wide. Academic consensus suggests 15-20 years of excess returns are needed for strong statistical confidence. Practically, the approach is to supplement return track records with process evaluation: is the manager's strategy clearly defined and consistently applied? Are their performance patterns consistent with the stated strategy's characteristics? Does alpha correlate with periods when their edge should be most effective?

The Sharpe Ratio and Information Ratio are practical alpha-quality metrics. The Information Ratio (IR) = Alpha / Tracking Error (the standard deviation of excess returns relative to benchmark). A consistent IR above 0.5 is considered good; above 1.0 is excellent. The persistence of the Information Ratio across different market regimes — not just bull markets — is the clearest evidence of durable alpha generation versus strategy-specific luck. In practice, fewer than 20-25% of actively managed mutual funds deliver positive after-fee alpha over 10-year periods.