The Fama-French Five-Factor Model Explained

The Evolution from CAPM to Five Factors

The Fama-French five-factor model is the workhorse of modern empirical finance. It explains portfolio returns using five systematic risk factors: the market premium, size (SMB), value (HML), profitability (RMW), and investment (CMA). The model captures 71 to 94 percent of cross-sectional return variance depending on the test assets, making it far more powerful than the original CAPM. Understanding this model is essential for any quantitative investor -- whether you are evaluating fund performance, constructing factor portfolios, or decomposing the sources of your returns.

From CAPM to Five Factors: The Evolution

Stage 1: The Capital Asset Pricing Model (1964)

The CAPM, developed independently by Sharpe (1964), Lintner (1965), and Mossin (1966), proposed a beautifully simple theory: the expected excess return of any asset equals its beta multiplied by the market risk premium.

E(Ri) - Rf = Beta_i x (E(Rm) - Rf)

Beta measures the asset's sensitivity to the overall market. The model predicts a single source of systematic risk and a linear security market line.

The CAPM was revolutionary, but it failed empirically. By the 1980s, researchers had documented numerous anomalies -- patterns in returns that beta alone could not explain. Small stocks outperformed large stocks. Value stocks outperformed growth stocks. These "anomalies" demanded a richer model.

Stage 2: The Three-Factor Model (1993)

Fama and French (1993) introduced two additional factors to capture the size and value anomalies:

E(Ri) - Rf = Beta_i x MKT + s_i x SMB + h_i x HML

• MKT (Market): The excess return of the broad stock market over the risk-free rate.
• SMB (Small Minus Big): The return of small-cap stocks minus large-cap stocks.
• HML (High Minus Low): The return of high book-to-market (value) stocks minus low book-to-market (growth) stocks.

The three-factor model was a major improvement over the CAPM. It explained a large portion of the cross-section of returns and became the standard for evaluating mutual fund performance. A fund that appeared to generate alpha under the CAPM often showed zero alpha once size and value exposures were controlled for.

Stage 3: The Five-Factor Model (2015)

Despite its success, the three-factor model left two important return patterns unexplained. Stocks with high profitability earned higher returns than stocks with low profitability. Stocks of firms that invested conservatively earned higher returns than those that invested aggressively. Novy-Marx (2013) had shown the importance of profitability, and Fama and French incorporated both patterns into their updated model.

E(Ri) - Rf = Beta_i x MKT + s_i x SMB + h_i x HML + r_i x RMW + c_i x CMA

• RMW (Robust Minus Weak): The return of stocks with high operating profitability minus stocks with low operating profitability.
• CMA (Conservative Minus Aggressive): The return of stocks of firms with low asset growth (conservative investment) minus stocks of firms with high asset growth (aggressive investment).

The Five Factors in Detail

Factor 1: Market (MKT)

The market factor captures the broad equity risk premium -- the compensation for bearing systematic stock market risk. It is the return of the market portfolio minus the risk-free rate (typically the one-month Treasury bill).

The long-run market premium in U.S. equities has been approximately 6 to 8 percent annually. This is the single most important factor and accounts for the majority of portfolio return variance for most investors.

Factor 2: Size (SMB)

SMB captures the historical tendency of small-cap stocks to outperform large-cap stocks. The factor is constructed by sorting stocks into two size groups (small and big) by median market capitalization, then computing the average return of small portfolios minus the average return of big portfolios.

The annual SMB premium has been approximately 2 percent in U.S. data since 1926, though with considerable variation across sub-periods. As discussed in the research on the size effect, the raw size premium has weakened since publication, but interacts positively with value and quality.

Factor 3: Value (HML)

HML captures the value premium -- the tendency of stocks with high book-to-market ratios (cheap stocks) to outperform stocks with low book-to-market ratios (expensive stocks). Stocks are sorted into three groups by book-to-market: high (top 30%), neutral (middle 40%), and low (bottom 30%).

The annual HML premium has been approximately 3 to 4 percent in U.S. data. Notably, Fama and French (2015) found that in their five-factor model, HML becomes largely redundant -- its effect is subsumed by the RMW and CMA factors. They describe HML as the factor that can be dropped without significantly changing the model's explanatory power.

Factor 4: Profitability (RMW)

RMW captures the profitability premium. Operating profitability is measured as revenue minus cost of goods sold, minus selling, general, and administrative expenses, minus interest expense, divided by book equity.

Stocks are sorted into three groups by profitability: robust (top 30%), neutral (middle 40%), and weak (bottom 30%). The factor return is the average return of the robust portfolios minus the average return of the weak portfolios.

The annual RMW premium has been approximately 3 percent in U.S. data. The economic logic is grounded in valuation theory: holding price constant, firms with higher expected cash flows (proxied by current profitability) must have higher expected returns. This connection to dividend discount model fundamentals gives the profitability factor stronger theoretical grounding than many anomalies.

Factor 5: Investment (CMA)

CMA captures the investment premium. Investment is measured as the annual change in total assets divided by total assets.

Stocks are sorted into three groups: conservative (bottom 30% asset growth), neutral (middle 40%), and aggressive (top 30% asset growth). The factor return is conservative minus aggressive.

The annual CMA premium has been approximately 2.5 percent in U.S. data. The theoretical rationale connects to q-theory in corporate finance: firms invest more when the cost of capital is low (hence future expected returns are low), and invest less when the cost of capital is high (hence future expected returns are high).

How to Use the Model: Portfolio Analysis

Running a Factor Regression

The most common practical use of the five-factor model is running a time-series regression of your portfolio's excess returns on the five factors:

Rp - Rf = alpha + b1(MKT) + b2(SMB) + b3(HML) + b4(RMW) + b5(CMA) + error

The coefficients (b1 through b5) tell you the portfolio's factor exposures -- its loading on each systematic risk factor. The intercept (alpha) represents risk-adjusted abnormal return: the portion of performance not explained by the five factors.

Interpreting Results

Data Sources

Factor return data is available for free from Kenneth French's data library (maintained at Dartmouth). The library provides daily, weekly, and monthly factor returns for U.S. equities, as well as international factor data for developed and emerging markets.

Criticisms and Alternatives

The five-factor model, despite its wide adoption, faces several important criticisms.

Missing Momentum

The most significant omission is the momentum factor. Momentum has been one of the strongest and most pervasive anomalies in finance, documented across dozens of countries and multiple asset classes. The five-factor model makes no attempt to capture it. Fama and French acknowledge this omission but argue that momentum is a short-lived phenomenon driven by market microstructure rather than a persistent risk premium.

Many practitioners use a six-factor model that adds the UMD (Up Minus Down) momentum factor from Carhart (1997). This is sometimes called the Fama-French-Carhart model.

The q-Factor Model

Hou, Xue, and Zhang (2015) proposed the q-factor model as an alternative, based on investment-based asset pricing theory. Their four factors -- market, size, investment, and return on equity -- emerge from a production-based equilibrium model rather than empirical pattern-matching. They argue this framework provides a tighter link between factor construction and economic theory.

The Stambaugh-Yuan Model

Stambaugh and Yuan (2017) proposed a four-factor model with two mispricing factors constructed from a combination of eleven anomalies. Their approach explicitly frames the factors as capturing behavioral mispricing rather than rational risk premia.

Data Mining Concerns

Harvey, Liu, and Zhu (2016) raised the alarming finding that the majority of published factors in finance are likely false discoveries driven by data mining. With hundreds of proposed factors in the literature, the standard statistical threshold (t-statistic > 2.0) is insufficient. They recommend a threshold of approximately 3.0. This concern applies to all factor models, including the five-factor model, though the Fama-French factors have been replicated extensively out of sample.

International Evidence

The five-factor model's performance varies internationally. The profitability and investment factors are generally significant in developed markets, but evidence in emerging markets is more mixed. The value factor retains stronger independent explanatory power in international data compared to U.S. data, where it is subsumed by RMW and CMA.

Practical Recommendations

For fund evaluation: Use the five-factor model (or a six-factor version with momentum) to decompose the returns of any fund or strategy. This separates genuine alpha from mechanical exposure to well-known factors. Many funds that claim active skill are simply harvesting factor premia -- which can be obtained more cheaply through factor ETFs.

For portfolio construction: Understanding your portfolio's factor exposures helps you avoid unintended bets and build portfolios with deliberate, diversified factor tilts. A portfolio that is inadvertently loaded on a single factor is taking concentrated risk.

For risk management: Factor exposures change over time. Regularly monitoring your portfolio's factor loadings can alert you to style drift, concentration risk, or unintended shifts in portfolio characteristics.

Factor data access: Download factor returns from Kenneth French's data library at Dartmouth. Use monthly returns for standard regression analysis. A minimum of 36 months of data is recommended; 60 months or more provides more stable estimates.

Independent Backtest: Five-Factor Model Performance by Decade

The following table presents decade-by-decade performance of the individual Fama-French five-factor components, illustrating how each factor's contribution has varied across market regimes.

Methodology: Using monthly returns from the Kenneth French Data Library for each factor (MKT-RF, SMB, HML, RMW, CMA), January 1963 through December 2025. Returns are gross of transaction costs.

Values shown as Annualized Return / Sharpe Ratio for each factor.

Several patterns emerge from this data. The market factor (MKT-RF) dominates in terms of magnitude but shows the widest variation, ranging from -1.0% in the 2000s to 13.4% in the 1990s. The value factor (HML) shows a striking regime shift: strong and consistent from the 1960s through the 2000s, then negative in the 2010s before recovering in the 2020s. The profitability factor (RMW) has been the most consistent across decades, never delivering a negative decade -- a stability that supports its theoretical grounding in valuation fundamentals. The size factor (SMB) has been the noisiest, with much of its premium concentrated in specific periods.

These figures are derived from publicly available academic factor return data and do not account for transaction costs, market impact, or implementation constraints. Live portfolio performance would differ materially.

Cross-Market Evidence for the Five Factors

The five-factor model's credibility depends critically on whether its factors replicate outside the U.S. sample used for original estimation. Fama and French (2012) conducted the most systematic international test in "Size, value, and momentum in international stock returns," examining four regions: North America, Europe, Japan, and Asia Pacific.

A critical finding from the international evidence is that HML retains independent explanatory power outside the United States. Recall that in U.S. data, Fama and French (2015) found HML to be largely redundant once RMW and CMA were included. This redundancy does not hold internationally: in European and Japanese data, HML captures return variation distinct from profitability and investment. This suggests that the U.S.-specific redundancy may reflect particular characteristics of the U.S. market rather than a universal feature of value's relationship to profitability.

Asness, Moskowitz, and Pedersen (2013) provided complementary evidence in "Value and Momentum Everywhere," demonstrating that value effects appear not only across global equity markets but also in bonds, currencies, and commodities. This cross-asset breadth suggests that the value phenomenon reflects something fundamental about how financial assets are priced, rather than an equity-specific statistical artifact.

The Hou-Xue-Zhang q-factor model (2015) has also been tested internationally, with mixed results. The investment and profitability factors from the q-model show somewhat different behavior outside the U.S., reflecting differences in corporate investment patterns, accounting standards, and market structure across countries.

Open Questions

No asset pricing model is complete, and the five-factor model is no exception. Several important questions remain subjects of active research and debate.

The most significant omission is momentum. Carhart (1997) demonstrated that adding a momentum factor (UMD) substantially improves the model's explanatory power, and the six-factor extension has become standard in practice. Fama and French's decision to exclude momentum reflects their view that it is driven by short-term market microstructure rather than persistent risk, but this position is contested by researchers including Asness (2014), who argues that momentum is as fundamental as value.

The value factor's role in the five-factor framework remains contentious. Fama and French (2015) showed that HML is subsumed by RMW and CMA in U.S. data, leading to the suggestion that HML could be dropped. However, Fama and French (2017) subsequently found that HML retains independent explanatory power in international data, and Barillas and Shanken (2018) argued that model comparison tests favor retaining HML. The resolution of this debate has practical implications for how practitioners construct factor portfolios.

Harvey, Liu, and Zhu (2016) raised the broader concern of data mining in factor research, demonstrating that more than half of 400+ published factors fail a t-statistic threshold of 3.0. While the Fama-French factors have survived extensive out-of-sample testing, the proliferation of factor models -- each claiming to explain the cross-section better -- raises questions about overfitting. Hou, Xue, and Zhang (2020) found that 64% of 452 tested anomalies failed to replicate, underscoring the importance of parsimony.

Factor premia are not stationary. McLean and Pontiff (2016) documented post-publication decay of approximately 32% out-of-sample and 26% post-publication for the average published factor. The five-factor model's components have shown varying degrees of time variation: the market premium has ranged from negative to double-digit across decades, HML turned negative for a full decade in the 2010s, and even the relatively stable RMW has shown period-to-period fluctuation. For practitioners, this means that factor allocations based on historical averages may not reflect prospective premia, and that maintaining diversification across factors is more important than optimizing allocations based on any single historical period.

The model assumes linear, time-invariant factor exposures -- an assumption that is frequently violated in practice. Fund managers who dynamically adjust their portfolios, use derivatives, or engage in market timing will have factor loadings that change over time. Rolling-window regressions can partially address this issue but introduce their own estimation challenges. Conditional factor models that allow exposures to vary with economic state variables represent an active area of research but have not yet achieved the practical adoption of the static five-factor model.