In brief

The Fama–French three-factor model expands the Capital Asset Pricing Model by adding two stock-return factors to the market factor:

  • Market (Rm − Rf): the return of the broad stock market above the risk-free rate;
  • SMB, or Small Minus Big: the return difference between diversified portfolios of smaller and larger companies; and
  • HML, or High Minus Low: the return difference between diversified portfolios of high and low book-to-market stocks.

The model asks whether a portfolio’s return can be explained by exposure to the broad market, smaller companies, and value stocks. It is widely used to evaluate portfolios and investment managers, but it does not promise that small or value stocks will outperform. Factor premiums can be weak or negative for long periods, and the answer depends on the market, sample, construction method, and period being studied.

I think of the model as a better diagnostic tool than a prediction machine. It helps answer, “What kinds of risk or return patterns did this portfolio load on?” It cannot tell us with certainty what those factors will earn next.

Why researchers moved beyond CAPM

The Capital Asset Pricing Model, or CAPM, describes expected return using one source of systematic risk: exposure to the market. Its equation uses a single market beta.

That simplicity made CAPM influential, but the historical evidence did not fit its central prediction especially well. In their 1992 paper, “The Cross-Section of Expected Stock Returns”, Eugene Fama and Kenneth French studied U.S. stock returns and reported that company size and book-to-market equity helped explain differences in average returns, while the relation between market beta and average return was weak in their tests.

Their 1993 paper, “Common Risk Factors in the Returns on Stocks and Bonds”, turned that evidence into a practical stock-return model with three factors: market, size, and book-to-market equity.

The important shift was not “CAPM is useless.” It was that one market beta did not appear to capture every recurring pattern in the data. The three-factor model gave researchers a richer benchmark.

The equation

For a portfolio or stock i, the time-series form is commonly written as:

Rᵢ − Rf = αᵢ + βMKT(Rm − Rf) + βSMB(SMB) + βHML(HML) + εᵢ

Here is what each term means:

  • Rᵢ − Rf: the investment’s return above the risk-free rate;
  • αᵢ: the portion of average excess return not explained by the three factors in the regression;
  • βMKT: sensitivity to the market excess return;
  • βSMB: sensitivity to the size factor;
  • βHML: sensitivity to the value factor;
  • εᵢ: the remaining period-specific return that the model does not explain.

Unlike the familiar CAPM expected-return equation, this version is often used as a regression. Historical returns are compared with historical factor returns to estimate the portfolio’s factor loadings and alpha.

That distinction matters. A regression can describe what a portfolio behaved like during a sample. It does not establish that the same loadings, factor premiums, or alpha will persist.

The three factors in plain English

1. Market excess return

The market factor is the broad stock market’s return minus the risk-free rate. It carries over the central idea from CAPM: investors holding risky stocks are exposed to market-wide movements that diversification within the stock market cannot eliminate.

A market loading near 1 means the portfolio historically moved about one-for-one with the chosen market factor in the regression. A loading above or below 1 indicates greater or smaller estimated sensitivity.

This is still an estimate. It depends on the market proxy, data frequency, sample period, and portfolio behavior during that period.

2. SMB: Small Minus Big

SMB is a long-short factor return. It represents the return on diversified small-stock portfolios minus the return on diversified large-stock portfolios.

  • A positive SMB loading means the portfolio behaved more like smaller companies.
  • A negative SMB loading means it behaved more like larger companies.
  • A loading near zero means the regression found little size tilt.

SMB is not the return of one small-cap index, and it is not the same as owning a small-cap ETF. It is a constructed difference between groups of stocks.

3. HML: High Minus Low

HML represents high book-to-market stocks minus low book-to-market stocks. A high book-to-market ratio means accounting book equity is large relative to the company’s market value; these stocks are commonly called value stocks. Low book-to-market stocks are commonly called growth stocks.

  • A positive HML loading indicates value-like behavior.
  • A negative HML loading indicates growth-like behavior.
  • A loading near zero indicates little measured value or growth tilt.

HML is also a constructed long-short factor. It should not be read as the return of every stock labeled “value.”

How SMB and HML are constructed

The exact construction is more careful than simply splitting a stock list in half. Kenneth French’s official U.S. factor methodology describes six value-weighted portfolios formed from company size and book-to-market equity:

Low book-to-market (growth) Neutral High book-to-market (value)
Small Small Growth Small Neutral Small Value
Big Big Growth Big Neutral Big Value

The published factors are then built from those portfolio returns:

SMB = average of the three small portfolios − average of the three big portfolios

HML = average of Small Value and Big Value − average of Small Growth and Big Growth

This construction helps separate size from value. Small companies appear in the growth, neutral, and value groups, while both small and big companies contribute to HML.

The details are not identical across every region. The official data library documents the eligible securities, breakpoints, currency treatment, risk-free series, and portfolio-rebalancing rules for each dataset. Those choices can affect the resulting factor returns.

Reading factor loadings

Factor loadings are slopes estimated from historical data. They describe how strongly the investment’s excess returns moved with each factor during the chosen sample.

Estimated loading A reasonable interpretation What it does not prove
Market beta 1.05 Slightly more market-sensitive than a loading of 1 in this sample That the portfolio will gain 1.05% whenever the market gains 1%
SMB beta 0.40 A measurable small-company tilt That every holding is a small company
HML beta 0.70 A meaningful value tilt That value will outperform next year
Alpha near zero The three factors explained the average return reasonably closely in the fitted sample That the strategy has no value or will never outperform

Loadings can change. A fund manager can alter holdings, an index can rebalance, companies can grow, valuation ratios can move, and the regression itself can change when the sample window changes.

A worked three-factor example

Consider a hypothetical monthly example. These figures are teaching inputs, not current factor estimates or forecasts.

Assume:

  • risk-free return: 0.25%;
  • market excess return: 0.80%;
  • SMB return: −0.20%;
  • HML return: 0.40%;
  • market loading: 1.05;
  • SMB loading: 0.40;
  • HML loading: 0.70; and
  • assumed alpha: 0.

The model-implied excess return is:

1.05 × 0.80% + 0.40 × (−0.20%) + 0.70 × 0.40% = 1.04%

Adding the hypothetical 0.25% risk-free return gives:

Model-implied total return = 1.29% for the month

The contributions make the result easier to understand:

Component Calculation Contribution
Market 1.05 × 0.80% 0.84%
Size 0.40 × −0.20% −0.08%
Value 0.70 × 0.40% 0.28%
Total excess return 1.04%
Risk-free return 0.25%
Model-implied total return 1.29%

The negative size contribution does not mean the portfolio lacked a small-company tilt. The loading was positive, but the hypothetical SMB factor return was negative that month. Exposure and payoff are different concepts.

Alpha: what remains after the factors

Alpha is the regression intercept. In a historical performance analysis, it is the average return left over after accounting for the model’s factor exposures.

Suppose a portfolio had positive CAPM alpha but little or no alpha in the three-factor model. One possible interpretation is that its apparent CAPM outperformance came from persistent small-company or value exposure rather than security-selection skill.

That is useful, but alpha is not a verdict. Its interpretation depends on:

  • whether the benchmark model is appropriate;
  • the length and quality of the return history;
  • statistical uncertainty;
  • fees, taxes, and trading costs;
  • stale or changing factor exposures; and
  • whether the analysis was designed before or after seeing the results.

A positive estimated alpha can be noise. A zero alpha can still accompany a useful low-cost portfolio that delivered the factor exposure its investor wanted.

What the historical evidence does—and does not—say

The 1992 Fama–French study examined U.S. data and found that size and book-to-market equity captured substantial cross-sectional variation in average stock returns in its samples. The 1993 paper found that market, size, and book-to-market factors captured shared variation in stock returns and provided a stronger empirical description than a market-only model.

Those findings are historical, not guarantees. The model does not say:

  • SMB and HML must be positive every year;
  • every small or value stock has a higher expected return;
  • the premiums will be identical across countries;
  • the original sample estimates are forecasts; or
  • factor exposure eliminates the possibility of long underperformance.

Kenneth French’s Data Library publishes current research datasets, regional factors, construction details, and historical archives. The existence of revised archives is a useful reminder that research databases and construction processes can change. A backtest should identify the exact dataset and version it uses.

Risk explanation or behavioral pattern?

There are two broad ways to interpret size and value premiums.

One interpretation is risk-based: small and high book-to-market companies may be exposed to economic distress or other risks that investors require compensation to bear.

Another is behavioral or market-based: investors may systematically overreact, underreact, extrapolate growth too far, or face institutional constraints that allow pricing patterns to persist.

The three-factor regression can measure exposure without settling that debate. It describes return covariation; it does not, by itself, prove why a factor earned a premium.

Three-factor model compared with other models

Model Factors Main use Important omission
CAPM Market Simple expected-return baseline and market-risk estimate Size, value, momentum, profitability, and investment patterns
Fama–French three-factor Market, size, value Explaining diversified equity returns and style exposure Momentum, profitability, and investment factors
Carhart four-factor Market, size, value, momentum Performance attribution when momentum exposure matters Profitability and investment factors
Fama–French five-factor Market, size, value, profitability, investment Broader equity-return attribution Momentum; the model’s authors also document remaining weaknesses

More factors do not guarantee a better answer. Adding factors can improve fit while increasing complexity, estimation error, overlap, and the temptation to explain past noise. The right benchmark depends on the portfolio and the question being asked.

Five common mistakes

1. Calling SMB and HML investable indexes

They are research factors built from long-short portfolio returns. A real ETF has fees, implementation rules, trading costs, tax consequences, and exposures that may not match the academic factor.

2. Treating a positive loading as a promised premium

A portfolio can have a positive HML loading while HML produces a negative return. Exposure describes sensitivity; it does not guarantee compensation on a schedule.

3. Comparing regressions that use different data

Daily versus monthly returns, different risk-free rates, different market regions, and different sample windows can produce different loadings and alphas.

4. Ignoring statistical uncertainty

A coefficient is an estimate. Its standard error and the stability of the result matter. A small positive alpha or loading may not be meaningfully different from zero.

5. Using the model as a complete portfolio plan

The model does not choose an asset allocation, emergency reserve, tax strategy, time horizon, or spending plan. It describes equity-return patterns; it does not replace financial planning.

How I would use the model

I would use the three-factor model for attribution, comparison, and questions.

For attribution, I would ask how much of a fund’s historical behavior came from the market, size, and value factors rather than treating all excess return as manager skill.

For comparison, I would put similar funds on the same date range, return frequency, factor dataset, and risk-free series. Otherwise, differences in methodology can look like differences in investment quality.

For questions, I would look for unstable loadings, unexplained alpha, concentration, fees, turnover, and whether the academic factor has a realistic investable counterpart. I would also compare the three-factor result with CAPM and a broader factor model to see whether the conclusion depends heavily on one specification.

I would not select a portfolio simply because a historical regression shows attractive factor exposure. The investor still has to endure tracking error, stretches of underperformance, costs, taxes, and the possibility that the historical premium does not repeat.

Bottom line

The Fama–French three-factor model extends CAPM with two important historical stock-return patterns: size and value. Its market, SMB, and HML loadings provide a clearer vocabulary for discussing why diversified portfolios behaved differently.

Its greatest strength is diagnosis. It can show that apparent outperformance was associated with systematic factor exposure rather than a mysterious source of skill.

Its greatest limitation is the same one shared by every historical model: the future does not have to reproduce the sample. Factor loadings are estimates, factor premiums vary, and a good explanation of past returns is not a guaranteed forecast.

References