Carhart four-factor model explained

Why momentum belongs in factor attribution

The Fama-French three-factor model explains returns with market beta, size (SMB), and value (HML). It handles many value and small-cap tilts well, but leaves a systematic hole: funds that buy stocks with strong recent relative performance often show positive unexplained alpha under FF3 even when they are simply harvesting the momentum premium.

Momentum is economically distinct from growth investing. Growth tilts use fundamentals (earnings growth, price-to-earnings); momentum uses past price returns, typically skipping the most recent month to reduce short-term reversal noise. A fund can be growth-heavy with negative momentum (falling leaders) or momentum-positive without classic growth characteristics. Ignoring UMD misattributes trend exposure as manager skill.

Carhart's contribution was empirical: adding UMD to FF3 explained persistence in mutual fund performance rankings that CAPM and FF3 alone could not. For modern allocators evaluating quant funds, ETF rotation strategies, or growth managers who implicitly ride winners, the four-factor model is often the minimum honest attribution specification.

The four factors: MKT, SMB, HML, and UMD

The time-series regression for portfolio or fund excess returns is:

Ri − Rf = αi + βi(Rm − Rf) + si SMB + hi HML + mi UMD + εi

• Market (MKT-RF): Excess return of the market portfolio over the risk-free rate.
• SMB (Small Minus Big): Return spread between small-cap and large-cap stock portfolios.
• HML (High Minus Low): Return spread between high and low book-to-market (value minus growth) portfolios.
• UMD (Up Minus Down): Return spread between prior winners and prior losers — the momentum factor. Also labeled WML (winner minus loser) in some data libraries.

Factor returns are long-short portfolio spreads. A fund with m = 0.6 behaves as if it holds a 60% notional exposure to the momentum long-short factor on top of its other tilts. Loadings are estimated jointly; adding UMD often reduces apparent HML and SMB loadings because momentum correlates with size and style rotations across market cycles.

Reading momentum loadings

• m > 0.5 — strong momentum tilt (quant trend, growth leaders, 12-1 relative strength screens).
• m ≈ 0 — style-neutral on momentum; FF3 may suffice for broad passive funds.
• m < 0 — contrarian or mean-reversion exposure (rare in long-only equity; more common in certain hedge strategies).
• m > 0, h < 0 — growth-momentum profile common in tech-heavy rallies.
• m > 0, h > 0 — value-momentum combo (sometimes called “proactive value”); can be volatile in style reversals.

How the UMD factor is constructed

Kenneth French's momentum factor series (often used in Carhart regressions) follows a standard methodology:

1. Universe: U.S. common stocks on NYSE, AMEX, and NASDAQ with sufficient price history.
2. Formation period: Rank stocks by cumulative return from month t−12 to t−2 (eleven months, skipping the most recent month to avoid microstructure reversal).
3. Breakpoints: 30th and 70th percentiles of the momentum signal split losers, middle, and winners.
4. Size control: Independent size sort (median NYSE market cap) creates six portfolios; UMD combines small and large winner legs minus loser legs.
5. Value-weighting: Stocks within portfolios are value-weighted; the factor is rebalanced monthly.

Practitioner implementations vary: some use 6-month lookbacks, risk-adjusted momentum (return divided by volatility), or industry-neutral ranks. Regression results are only comparable when fund returns and factor series share the same calendar, currency, and momentum definition. Mixing a French UMD series with a proprietary 6-month signal breaks interpretability.

Momentum crashes and non-normal tails

UMD earns positive average premiums over decades but suffers sharp drawdowns when crowded momentum trades unwind — famously in 2009 and during violent style rotations. A fund with high m loading inherits this tail risk even if monthly alpha looks stable in calm regimes. Stress-test attribution with sub-periods that include 2009, March 2020, and 2022 growth-to-value rotation windows before sizing momentum-heavy allocations.

Interpreting alpha after controlling for momentum

The intercept α is Jensen's alpha after four factors. Many funds that beat benchmarks on raw returns show negligible Carhart alpha once UMD is included.

• FF3 vs Carhart: If α drops sharply when UMD enters, the fund's edge was likely momentum exposure, not security selection.
• Statistical significance: Report t-statistics with Newey-West standard errors; momentum factor returns are autocorrelated and skewed.
• R²: Adding UMD often raises explained variance for growth and quant funds by 5–15 percentage points versus FF3 alone.
• Rolling windows: Momentum loadings spike in trending markets and compress in choppy ranges; 36-month rolling regressions reveal regime dependence.

Pair regression with holdings analysis: high m should correlate with overweight positions in prior 12-month outperformers. Mismatch between regression and holdings suggests timing skill, leverage, or options overlays not captured by static factor models.

Harbor Capital Carhart attribution refactor

Harbor's tactical equity sleeve rebalanced quarterly into the top two quintiles of 12-1 month relative strength within the Russell 1000, with sector caps and a 5% turnover buffer. Marketing emphasized “adaptive stock selection.” LP due diligence requested factor attribution.

1. FF3 baseline — Positive α of 95 bps (t = 2.3) over 72 months; HML loading −0.22 (growth tilt), SMB 0.08. Appeared skilled.
2. Carhart extension — UMD loading 0.72; α fell to 18 bps (t = 0.5). HML moved to −0.31 as momentum explained overlap with growth leadership.
3. Holdings verification — Average holding had been in top momentum decile at purchase; 78% of excess return vs Russell traced to UMD factor replication.
4. Fee benchmarking — Compared to momentum ETF (MTUM-class) net of fees; active fee premium hard to justify at 18 bps alpha.
5. Disclosure update — Prospectus reframed strategy as “systematic momentum exposure with risk overlays” rather than “proprietary tactical selection.”

Outcome: Sleeve retained investors who wanted momentum beta; fee negotiation tightened. Lesson: Carhart attribution prevents paying active fees for passive factor exposure.

Carhart vs FF5 and six-factor models

Fama and French's 2015 five-factor model adds profitability (RMW) and investment (CMA) but still omits momentum. Practitioners often run:

• Carhart (4-factor): FF3 + UMD — best for momentum-heavy, growth-trend, and quant funds.
• FF5: MKT, SMB, HML, RMW, CMA — best for quality and profitability tilts without explicit trend rules.
• Six-factor: FF5 + UMD — broadest equity attribution when funds combine quality screens with momentum ranking.

Running FF5 without UMD on a momentum fund can leave fake alpha; running Carhart without RMW/CMA can mislabel quality funds as skilled stock pickers. Match the model to the mandate, not the narrative.

Model decision table

Common pitfalls

• Paying active fees for momentum beta — Run Carhart before fee negotiations on tactical or growth funds.
• Using FF3 on quant funds — UMD loadings above 0.4 are common; unexplained alpha is often an artifact.
• Confusing momentum with growth — High P/E and high past returns are correlated but not identical; regression separates them.
• Ignoring momentum crash risk — Positive historical alpha can vanish in one reversal quarter; report worst-drawdown sub-periods.
• Short regression windows — Fewer than 48 monthly observations produce unstable UMD loadings; prefer 60+ months.
• Survivorship-biased fund samples — Dead momentum funds disappear from databases, overstating average alpha (see survivorship bias guide).
• Mixing momentum definitions — French UMD vs 6-month rank vs risk-adjusted momentum are not interchangeable.
• International funds with U.S. UMD — Use regional momentum factors when available; U.S. UMD misattributes global trend funds.

Production checklist

• Download consistent monthly factor returns (French library: FF3 + Momentum).
• Align fund returns and factors on same calendar (month-end, USD, net of fees).
• Run FF3 and Carhart side by side; document how α and loadings change.
• Use Newey-West standard errors on time-series regressions.
• Report R², adjusted R², and observation count with every table.
• Plot rolling 36-month UMD loadings for tactical mandates.
• Stress-test sub-periods including known momentum crash windows.
• Bridge regression to holdings: average 12-1 momentum rank at purchase.
• Add RMW and CMA (six-factor) when quality screens overlap with trend rules.
• Compare net alpha to momentum ETF net of fees before approving active premium.

Key takeaways

• The Carhart four-factor model adds momentum (UMD) to Fama-French three-factor attribution, exposing trend exposure that FF3 mislabels as alpha.
• UMD is built from winner-minus-loser long-short portfolios using 12-2 month cumulative returns, value-weighted and rebalanced monthly.
• Momentum loadings above 0.5 describe quant trend, growth-leadership, and tactical rotation funds — not necessarily security-selection skill.
• Harbor Capital's refactor showed Carhart attribution aligns fee discussions with measurable momentum beta rather than marketing narratives.
• Use six-factor (FF5 + UMD) when funds blend quality screens with momentum ranking; match the model to the mandate.

Related reading

• Fama-French three-factor model explained — SMB, HML, and the base Carhart extends
• Momentum investing explained — practitioner trend rules, crashes, and ETF implementation
• Fama-French five-factor model explained — RMW, CMA, and six-factor extensions
• Jensen's alpha explained — interpreting the regression intercept