Jensen's alpha explained

What Jensen's alpha measures

Raw outperformance — fund return minus benchmark return — confounds skill with risk taking. A fund with beta 1.4 should be expected to beat a beta-1.0 index in rising markets and lag in crashes. Jensen's alpha isolates the unexpected portion of return: what remains after subtracting the return CAPM says you deserved for bearing systematic (market) risk.

Positive alpha means the manager delivered more than fair compensation for beta. Negative alpha means they underperformed relative to the risk they took. Zero alpha is consistent with CAPM efficiency: you got exactly what the market priced for your exposure. Alpha is expressed in return units (annualized percentage points), not as a ratio — unlike Sharpe or Treynor, which divide excess return by a risk measure.

The formula

For portfolio p over a measurement window:

αp = Rp − [Rf + βp × (Rm − Rf)]

Where R_p is the portfolio return, R_f is the risk-free rate, β_p is portfolio beta vs the market benchmark R_m, and the bracketed term is CAPM expected return. The difference is Jensen's alpha.

Rearranged, alpha equals average excess return minus beta times the market risk premium:

αp = (Rp − Rf) − βp × (Rm − Rf)

Example: a fund returns 12%, T-bills yield 4%, the S&P 500 returns 10%, and estimated beta is 1.1. CAPM expected return = 4% + 1.1 × (10% − 4%) = 10.6%. Jensen's alpha = 12% − 10.6% = +1.4% per year. The manager beat fair compensation by 1.4 percentage points.

Statistical significance

Jensen's original papers tested whether alpha was statistically different from zero using regression. Run a time-series regression of portfolio excess returns on market excess returns:

Rp,t − Rf,t = α + β × (Rm,t − Rf,t) + εt

The intercept is Jensen's alpha; its t-statistic tells you whether alpha is distinguishable from luck. Institutional reviewers often require |t| > 2 (roughly 95% confidence) and 36+ months of data before crediting skill. Short windows produce noisy alphas that flip sign with one bad quarter.

Estimation choices that move alpha

Two analysts can compute different alphas for the same fund because inputs are not standardized. Document your assumptions before comparing managers.

Beta estimation

Beta is typically estimated via OLS regression of fund excess returns on benchmark excess returns over 36–60 months. Choices matter:

• Benchmark: S&P 500 for U.S. large-cap; MSCI ACWI for global mandates. Wrong benchmark inflates or deflates beta.
• Frequency: Monthly returns are standard; weekly adds observations but increases microstructure noise.
• Window: Rolling 36-month beta reacts to style drift; 60-month smooths cycles but lags recent risk changes.
• Leverage adjustment: levered funds need de-levered beta or alpha is distorted.

Risk-free rate

Use the same horizon as your return measurement. Annual alpha audits typically use 3-month T-bill yields averaged over the period, or a constant maturity series. Mismatching (e.g., 10-year Treasury for a 1-year alpha) shifts results by tens of basis points.

Total return vs price return

Include dividends and distributions in R_p and R_m. Price-only series understate equity returns and bias alpha downward for dividend-heavy funds.

Jensen's alpha vs other risk-adjusted metrics

Jensen's alpha and Treynor ratio are closely linked: for a given beta, higher alpha implies higher Treynor. Alpha is additive across time (roughly); Treynor is a ratio better suited for ranking. Use Sharpe when beta understates true risk — alternatives, private credit, or concentrated stock pickers often have low equity beta but high total volatility. Use information ratio when the manager's contract references a specific benchmark (e.g., Russell 1000 Growth) rather than the broad market.

Beyond single-factor CAPM

CAPM alpha assumes all non-market risk diversifies away. Empirically, factor exposures (size, value, momentum, quality) explain part of what single-factor alpha labels as skill. Multi-factor models extend the regression:

Rp,t − Rf,t = α + βmkt × MKTt + βsmb × SMBt + βhml × HMLt + εt

The intercept is multi-factor alpha — return unexplained by market, size, and value factors. A manager with positive CAPM alpha but negative Fama-French alpha may simply be a disguised small-cap value tilt, not a stock picker. Institutional due diligence increasingly reports both.

For hedge funds and alternatives, factor models may include momentum (UMD), bond market (TERM), or credit (DEF) factors. The principle is the same: alpha is only as credible as the risk model behind it.

Harbor Capital active equity sleeve worked example

Harbor Capital audits three large-cap managers annually. Manager B's five-year record (annualized, total return):

• Manager B return (R_p): 13.8%
• S&P 500 return (R_m): 11.7%
• 3-month T-bill average (R_f): 4.2%
• Estimated beta vs S&P 500: 1.35 (60-month monthly regression)

Step 1 — CAPM expected return:
E[R_p] = 4.2% + 1.35 × (11.7% − 4.2%) = 4.2% + 1.35 × 7.5% = 14.3%

Step 2 — Jensen's alpha:
α = 13.8% − 14.3% = −0.5% per year

Step 3 — Raw active return for context:
R_p − R_m = 13.8% − 11.7% = +2.1% (looks like skill)

The gap between +2.1% raw and −0.5% Jensen alpha is the beta story: Manager B ran 35% more market exposure than the index. In a period when equities returned 11.7%, that leverage mechanically contributed ~2.6% of excess return (0.35 × 7.5% market premium). The stock-selection component was negative.

Harbor's regression t-stat on alpha was −0.8 — not statistically significant, but directionally concerning. Trustees placed Manager B on a one-year watch: maintain allocation but require quarterly alpha reports and cap beta at 1.20. Manager A, with beta 0.95 and return 12.4%, posted Jensen alpha of +0.9% (t = 2.1) and received the incremental allocation.

When to use Jensen's alpha

Common pitfalls

• Confusing raw outperformance with alpha. Beating the S&P 500 is not alpha if beta was above 1.0 in a rising market.
• Ignoring statistical significance. +1.5% alpha over 18 months with t = 0.9 is noise, not skill.
• Wrong benchmark beta. A global fund regressed on the S&P 500 alone produces meaningless alpha.
• Survivorship bias in fund databases. Reported alphas among live funds overstate average skill; dead funds had negative alpha.
• Single-factor alpha on factor-tilted managers. A value manager's CAPM alpha partly reflects the value premium, not selection.
• Fee-blind alpha. Gross alpha before fees flatters managers; net alpha is what LPs experience.
• Non-normal return tails. CAPM regression assumes well-behaved residuals; crisis periods violate this — alpha estimates become unstable.
• Style drift without rolling beta. A fund that shifted from 0.8 to 1.3 beta mid-period needs window-matched estimation.

Allocator checklist

• Compute Jensen's alpha over at least 36 months (60 preferred) of monthly total returns.
• Match risk-free rate horizon to the return measurement period (typically 3-month T-bills).
• Estimate beta vs the fund's stated equity benchmark, not a convenience index.
• Report alpha t-statistic alongside point estimate; require |t| > 2 before crediting skill.
• Compare gross and net alpha after management and performance fees.
• Run multi-factor regression (Fama-French + momentum) to strip style tilts from alpha.
• Pair alpha with Treynor and Sharpe for a complete risk-adjusted picture.
• For active mandates, also compute information ratio vs the contractual benchmark.
• Document beta window and regression method in investment policy statements.
• Re-audit alpha after manager style drift, benchmark changes, or team turnover.

Key takeaways

• Jensen's alpha is return above CAPM fair compensation — the intercept from regressing excess returns on market excess returns.
• Positive alpha suggests skill; negative alpha suggests poor risk-adjusted results even when raw benchmark outperformance looks positive.
• Alpha depends on beta, risk-free rate, and benchmark choices — document assumptions before comparing managers.
• Pair with t-statistics and multi-factor models to separate luck, factor tilts, and genuine selection ability.
• Use information ratio for mandate-relative skill and Sharpe when total volatility matters more than beta.

Related reading

• Capital Asset Pricing Model (CAPM) explained — beta, expected return, and the security market line that defines fair compensation
• Treynor ratio explained — excess return per unit of beta, the ratio counterpart to Jensen's alpha
• Information ratio explained — active return per tracking error for mandate-relative manager evaluation
• Sharpe ratio explained — excess return per total volatility when idiosyncratic risk matters