Momentum & factors

The Fama-French five-factor asset pricing model

Momentum & factors · Published 2026-10-25

Asset pricing models try to explain why some stocks earn higher average returns than others. The Fama-French models are the most widely used empirical answer. Their five-factor version, published in 2015, is now a standard tool for testing whether a strategy's returns are new or simply a repackaging of known factors. This article explains the model, how its factors are built, and how researchers use it.

From CAPM to three factors

The Capital Asset Pricing Model, developed in the 1960s by William Sharpe, John Lintner and others, says a stock's expected excess return depends only on its beta with the market. Empirical tests in the following decades found patterns that beta could not explain. Small companies earned more than their betas implied, and so did "value" stocks with high book-to-market ratios.

Eugene Fama and Kenneth French (1993) responded with a three-factor model: the market excess return, a size factor (SMB, small minus big) and a value factor (HML, high minus low book-to-market). Each factor is the return of a long-short portfolio. A stock's or portfolio's exposure to each factor is estimated by regression, and the model says those exposures, not beta alone, explain average returns.

Adding profitability and investment

Research in the 2000s identified two further patterns. Firms with high operating profitability earned higher average returns than firms with low profitability (Novy-Marx, 2013, among others), and firms that invested aggressively, growing their total assets quickly, earned lower returns than conservative investors. Fama and French (2015) added two factors to capture them:

The theoretical motivation comes from the dividend discount model: holding other things equal, a firm's expected return is higher when its book-to-market ratio is higher, when its expected profitability is higher, and when its expected investment is lower.

Ri − Rf = ai + bi(RM − Rf) + siSMB + hiHML + riRMW + ciCMA + ei

If the model fully explains a portfolio's returns, the intercept ai, often called alpha, should be indistinguishable from zero.

How the factors are built

Stocks are sorted at the end of each June into two size groups using the median NYSE market capitalisation, and independently into three groups on each characteristic (book-to-market, operating profitability or investment) using NYSE 30th and 70th percentiles. The intersections form value-weighted portfolios, and each factor is an average of long-short differences across them. The details, including the exact definitions of operating profitability and investment, are documented on the Kenneth French Data Library, which publishes the factor returns for the U.S. and several international regions.

A notable finding: HML becomes redundant

In the 2015 paper, the value factor HML was largely explained by the other four factors in U.S. data from 1963: value stocks tend to be profitable, conservatively investing firms, so once RMW and CMA are included, HML adds little. Fama and French noted that HML could be dropped without loss for describing average returns over that sample, while keeping it can still be useful for describing portfolio tilts.

Where is momentum?

The five-factor model does not include momentum. Fama and French acknowledged that it does not explain momentum returns, and many researchers simply add the momentum factor (UMD) as a sixth factor. Fama and French (2018) later examined models including momentum, noting their own reservations about adding factors without a theoretical basis. In practice, a six-factor regression is common when testing equity strategies.

Using the model to test a strategy

Factor regressions are one of the most practical tools for a researcher. The procedure:

  1. Compute the monthly excess returns of the strategy (returns minus the risk-free rate, which the Data Library also provides).
  2. Download the factor returns for the same months and region.
  3. Regress the strategy's excess returns on the factors with ordinary least squares, using standard errors robust to heteroskedasticity and autocorrelation.
  4. Read the loadings to understand what the strategy is exposed to, and the intercept to see what is left unexplained.

A strategy that appears to beat the market may turn out to be a small-cap value tilt with a positive alpha that vanishes once SMB and HML are included. Conversely, a significant intercept after controlling for all factors is evidence, not proof, that the strategy captures something different. The statistical bar for such claims is high: Harvey, Liu and Zhu (2016) argue for t-statistics above 3 for new factors.

Criticisms and limits

Takeaway

The five-factor model is best seen as a measuring instrument. Before believing that a backtest has found an edge, run the regression and check whether the returns are explained by known exposures. Our strategy creation guide includes this step in its checklist.

Sources