Jensen's Alpha

The formula in one line

α  =  R_p  −  [ R_f  +  β_p (R_m − R_f) ]

Where R_p is the portfolio's realized return, R_f the risk-free rate, R_m the market return, and β_p the portfolio's beta. The bracket is the CAPM expected return — what the portfolio "should" have earned given its systematic risk. Alpha is the gap between realized and predicted. Positive α means the portfolio outperformed its risk-adjusted expectation; negative α means it underperformed.

Unlike Sharpe and Treynor (which are ratios), Jensen's alpha is an absolute number — measured in percentage points per year. A 2% alpha means 200 basis points of risk-adjusted outperformance. A −1% alpha means the manager destroyed 100 basis points of value relative to a passive same-β benchmark.

Worked example: A growth fund's alpha

A growth fund earned 15% last year. The market returned 12%, the risk-free rate was 3%, and the fund's beta (from a 3-year monthly regression) is 1.2.

CAPM expected return = R_f + β(R_m − R_f)
                     = 3% + 1.2 × (12% − 3%)
                     = 3% + 1.2 × 9%
                     = 3% + 10.8%
                     = 13.8%

Jensen's α = R_p − CAPM expected
           = 15% − 13.8%
           = +1.2%

The fund delivered 1.2 percentage points of risk-adjusted excess return — modest but positive. To know if it's skill or luck, compute the standard error: with 36 months of data, the typical SE on a fund's α is about 1.5%, so a 1.2% α has t-stat ~ 0.8 — not statistically distinguishable from zero. Three years isn't enough to call it skill. With 10 years of data the SE drops to about 0.8%, and 1.2% α would have t-stat ~ 1.5 — still not significant at 95% confidence. This statistical noise is why Jensen's 1968 paper concluded that most "alpha" reported by mutual funds was sampling variation, not skill.

The geometric picture — above the Security Market Line

Plot expected return against beta. CAPM equilibrium produces a straight line — the Security Market Line — passing through (0, R_f) with slope equal to the equity risk premium. Every fairly priced asset lies on the SML; mispriced assets sit above or below.

Position on SML diagram	Interpretation	Trade implication
Above the line	Realized return > CAPM expectation; α > 0	Undervalued — buy
On the line	Realized return = CAPM expectation; α = 0	Fairly priced
Below the line	Realized return < CAPM expectation; α < 0	Overvalued — sell or avoid

Alpha is the vertical distance from the SML to the fund's actual coordinate in beta-return space. A portfolio with β = 1.0 and R = 15% in a market where R_m = 10% and R_f = 3% sits 5 percentage points above the SML — α = 5%. The visual makes it obvious why CAPM defines mispricing as "off the line."

Jensen's α vs other performance metrics

Metric	Form	Risk control	What it captures
Jensen's α	Absolute (% per year)	β (systematic)	Excess return above CAPM expectation
Sharpe ratio	Ratio	σ (total)	Reward per unit of total vol
Treynor ratio	Ratio	β (systematic)	Reward per unit of systematic risk
Information ratio	α / σ(ε)	Residual (idiosyncratic)	α earned per unit of tracking error
Appraisal ratio	α / σ(ε)	Residual	Stock-picking skill, scaled by precision
Fama-French 3-factor α	Absolute	β_mkt, β_SMB, β_HML	Excess above 3-factor expectation
Carhart 4-factor α	Absolute	+ β_MOM	Excess above 4-factor expectation (incl. momentum)

Jensen's α is the workhorse for absolute manager evaluation: how many basis points of skill, in plain percent. Sharpe and Treynor rank funds by reward-to-risk; α tells you the dollar amount of risk-adjusted outperformance for a given allocation. Multi-factor extensions strip away factor-tilt "alpha" — a value tilt that earned a 3% premium has zero Fama-French α once you control for HML.

Michael Jensen, 1968, and the indexing revolution

Michael C. Jensen completed his University of Chicago PhD in 1968 with a dissertation titled The Performance of Mutual Funds in the Period 1945-1964. He computed CAPM-residual alpha for 115 actively managed U.S. mutual funds over a 20-year window. The headline finding: the average fund's alpha was approximately −1% per year before fees, and roughly −1.5% net of fees. After accounting for survivorship bias, the average actively managed dollar underperformed a passive index strategy.

The result was scandalous in 1968 — and is still controversial in active-management trade publications today. It motivated Jack Bogle to launch the Vanguard 500 Index Fund in 1976, the first retail index fund. Fast-forward to 2024: passive index funds now hold more U.S. equity AUM than active funds. Jensen's dissertation, which most contemporary critics tried to bury, won.

Jensen later joined Harvard Business School, did foundational work on agency theory (Jensen-Meckling 1976), and remained one of the central figures in modern empirical finance.

Implementation in 15 lines of Python

import numpy as np
import statsmodels.api as sm

def jensen_alpha(portfolio_returns, market_returns, rf_per_period, periods_per_year=12):
    """portfolio_returns, market_returns: arrays at same frequency
       rf_per_period: scalar or array, same frequency
       Returns: (alpha_annualized, beta, t_stat_alpha)"""
    p_ex = np.asarray(portfolio_returns) - rf_per_period
    m_ex = np.asarray(market_returns) - rf_per_period
    X = sm.add_constant(m_ex)
    model = sm.OLS(p_ex, X).fit()
    alpha_period = model.params[0]
    beta = model.params[1]
    t_alpha = model.tvalues[0]
    return alpha_period * periods_per_year, beta, t_alpha

# Example: 5-year monthly fund returns
alpha_yr, beta, t = jensen_alpha(fund_rets, mkt_rets, rf=0.0025)
print(f"α = {alpha_yr:.2%}/yr · β = {beta:.2f} · t-stat = {t:.2f}")

The t-statistic is crucial. With 60 monthly observations, t > 2 is roughly the 95% confidence threshold. Industry shorthand: "skill-significant" α requires a t-stat of 2+ on at least 5 years of monthly data. Anything below that is statistically indistinguishable from zero.

Famous alphas across investing history

Manager / fund	Period	Annualized α	Notes
Berkshire Hathaway (Warren Buffett)	1965–2024	≈ 8% / yr	59-year compound; the canonical persistence case
Renaissance Medallion	1988–2024	≈ 30%+ gross	Closed to outside money; secretive
Magellan Fund (Peter Lynch)	1977–1990	≈ 13% / yr	Retired voluntarily at his peak
Tiger Cubs (Robertson, Coleman, Mandel, etc.)	1990s–2010s	3-7% / yr	Long-short equity; mostly faded post-2010
Average active US equity mutual fund	1990–2024	≈ −1% / yr (after fees)	Jensen's 1968 finding holds in modern data
Top-decile active mutual fund	1990–2024	≈ 2-3% / yr	Selected ex-post; ex-ante persistence is weak
Index fund (Vanguard 500)	1976–2024	≈ 0% (by construction)	α = 0 ± fees; the cheap passive benchmark

The 8% Berkshire alpha needs context. Frazzini, Kabiller, and Pedersen (2018) decomposed Berkshire's α into two factor exposures — quality (high-return-on-equity, low-leverage) plus 1.6x effective leverage from float — leaving residual "pure skill" alpha of about 3%, still extraordinary but smaller than the headline 8%. The lesson: factor-model decomposition often shrinks reported CAPM alpha.

Real-world applications

• Manager selection at pension funds. CalPERS, Yale, Norway sovereign wealth all evaluate candidate managers on multi-factor α (typically 5-7 factor) over 5-10 year horizons. CAPM α is the baseline screen; factor α determines hire/fire.
• Mutual fund disclosure. Morningstar publishes CAPM α on its fund tearsheets alongside Sharpe and Sortino. Required disclosure varies by jurisdiction.
• Hedge-fund seeding. First-loss-capital allocators (Allianz, Tishman Speyer, Investcorp) screen new managers on prior 3-year α with t-stat ≥ 1.5 as a baseline qualification.
• Performance fees. Many institutional mandates pay performance fees on positive α (over a CAPM-adjusted benchmark) rather than raw returns. This protects clients from paying premium fees for what is really beta exposure.
• Academic research. Cross-sectional α distributions are the test bed for theories of market efficiency. Fama-French (2010) found that most U.S. mutual fund α distributions are consistent with zero skill plus sampling noise.
• Smart-beta and factor ETF assessment. The marketing claim "factor X delivered α" is usually wrong — factor exposure isn't α. Proper evaluation uses Fama-French 5-factor or AQR factor models to back out true residual skill.

Common pitfalls

• Single-factor α is upper-bound. Including size, value, momentum, profitability, and investment factors typically shrinks CAPM α by 30-70%. A "5% α" on Fama-French 5-factor is far more impressive than a 5% CAPM α.
• Short windows give noisy α. 12-month α is noise. Need 5+ years monthly minimum, ideally 10+, with explicit t-stat reporting.
• Survivorship bias inflates reported α. Failed funds drop out of databases; cross-sectional α distributions are systematically upward-biased by 1-2% in academic studies.
• Risk-free rate mismatch. Use the same-currency, same-frequency risk-free rate. A common bug: subtracting annual R_f from monthly returns inflates monthly α by a factor of 12.
• Non-linear strategies. CAPM α assumes constant β. Strategies with option-like payoffs (merger arb, vol trading) have time-varying β, and a single-β regression mis-attributes return to α. Use Henriksson-Merton or Treynor-Mazuy specifications.
• Confusing α with outperformance. Beating the S&P by 5% with a 1.5β portfolio is mostly β, not α. The CAPM-adjusted excess is much smaller. Always decompose before declaring skill.