How Fama-French Factor Models Explain What CAPM Misses
A single beta cannot explain why cheap, small, and highly profitable companies have historically outrun expensive, large, unprofitable ones by more than market risk alone predicts. Multifactor models were built specifically to close that gap, and understanding how they are constructed changes how you should read any fund's stated risk exposure.
Beyond a single factor
The Capital Asset Pricing Model reduces every source of priced risk to one number: sensitivity to the overall market. Researchers testing that model against decades of stock data kept finding the same leftover pattern: after accounting for market beta, small companies still earned more on average than large ones, and statistically cheap companies (measured by a high book value relative to market price) still earned more than statistically expensive ones. Since these differences persisted after controlling for beta, they represented either compensation for risks the single-factor model was missing, or a market inefficiency, and either way, they were too large and too consistent to dismiss as noise.
The response was to build additional factors directly out of these patterns and treat each one as its own risk exposure. A size factor, commonly called SMB for "small minus big," is constructed as the return of a portfolio long small-company stocks and short large-company stocks. A value factor, HML for "high minus low" book-to-market, is long statistically cheap stocks and short statistically expensive ones. Later extensions added a profitability factor (long highly profitable firms, short weakly profitable ones) and an investment factor (long conservatively investing firms, short aggressively expanding ones), since these characteristics also kept showing up as unexplained by the original three factors.
It is worth being precise about what each factor is actually measuring, since the labels are easy to misread. The size factor is not simply "small companies do better," it is the return difference between a diversified basket of small companies and a diversified basket of large companies, constructed to be roughly market-neutral in its overall exposure. The value factor similarly is not "cheap stocks are good," it is the spread between statistically cheap and statistically expensive companies, sorted by a valuation ratio, held long and short simultaneously. This construction matters because it isolates the characteristic being studied from the overall direction of the market: a positive historical average for HML tells you cheap stocks beat expensive stocks on average, independent of whether the market as a whole went up or down over the same period.
Building and reading the model
The general regression extends the single-factor model by adding one term per additional factor: excess return = alpha + (beta_market × market premium) + (beta_size × SMB) + (beta_value × HML) + residual, with further terms added for however many factors the model includes. Each beta in this equation is estimated the same way as a plain CAPM beta, by regressing the fund or stock's historical returns against the corresponding factor's historical returns.
Consider Fund A, whose regression against a three-factor model produces a market beta of 1.0, an SMB loading of 0.3 (a mild tilt toward smaller companies), and an HML loading of 0.4 (a moderate value tilt). If the annualized factor premiums over the relevant period were 6% for the market, 2% for SMB, and 3% for HML, the model's expected excess return is (1.0 × 6%) + (0.3 × 2%) + (0.4 × 3%) = 6% + 0.6% + 1.2% = 7.8%. If Fund A's actual realized excess return over the same period was 9.0%, its multifactor alpha is 9.0% minus 7.8%, or 1.2%, a smaller and more defensible skill claim than the fund would show if judged against the market factor alone.
The second example illustrates the opposite outcome. Fund B, a large-cap value fund, has a market beta of 0.95, a slightly negative SMB loading of −0.10 (reflecting its large-cap tilt), and a strong HML loading of 0.55. Using the same factor premiums of 5.5% (market), 1.5% (SMB), and 4.0% (HML) for this period: expected excess return = (0.95 × 5.5%) + (−0.10 × 1.5%) + (0.55 × 4.0%) = 5.225% − 0.15% + 2.2% = 7.275%. If Fund B's actual excess return was 6.8%, its alpha is 6.8% minus 7.275%, or −0.475%: a fund that looked like it was earning its keep through a value tilt was actually underperforming its own factor-adjusted benchmark once that tilt is properly priced in.
It is useful to notice what these two examples have in common and where they diverge. Both funds carry meaningful, deliberate style tilts away from the broad market, and in both cases the multifactor model absorbs a large share of the fund's raw return through those tilts rather than crediting it to manager skill. The difference lies entirely in the sign and size of what is left over after that absorption: Fund A's process added value beyond its factor exposures, while Fund B's did not. A raw performance chart showing both funds' total returns, without this decomposition, would tell an investor almost nothing about which manager was actually adding value through security selection versus simply riding a factor tailwind or headwind.
What the evidence shows
Historical data spanning most of the twentieth and early twenty-first centuries show a value premium (HML) averaging somewhere in the range of 3% to 5% annually over long samples, though with enormous variability, including decade-long stretches, most visibly through the late 1990s and again in the years surrounding 2020, when growth stocks meaningfully outperformed value stocks and the premium ran negative. The size premium (SMB) shows a similar pattern of a positive long-run average with periods, particularly since the late 1980s, where it has been weak, inconsistent, or negative, prompting real debate over whether the historical size premium was partly a product of data and measurement issues in earlier decades rather than a durable, structural phenomenon. Profitability and investment factors, introduced later specifically to absorb return patterns the original three factors left unexplained, have shown more consistent statistical significance in the samples where they were tested, though they still carry the same out-of-sample uncertainty that affects every factor derived from historical data.
Momentum, a separate and widely used factor built from a stock's own recent trailing return rather than a valuation or size characteristic, deserves particular mention because of how it behaves relative to the others. Momentum strategies, long recent strong performers and short recent weak performers, have shown some of the most statistically robust historical premia across markets and time periods studied, yet also some of the sharpest, fastest drawdowns of any major factor, including a well-documented sudden reversal around 2009 when previously beaten-down stocks rallied hard and momentum portfolios suffered outsized losses in a matter of weeks. This combination of strong average performance and occasional violent reversal makes momentum a useful illustration of a broader point: a factor's long-run average premium and its risk profile are two separate pieces of information, and a strategy can be well documented empirically while still being genuinely uncomfortable, even dangerous when leveraged, to hold through its worst stretches.
A separate and important empirical finding concerns factor crowding. Once academic and practitioner attention identifies a factor and asset managers build products around it, some of the historical premium tends to shrink, plausibly because more capital chasing the same characteristic compresses the very mispricing or risk compensation that produced the premium in the first place. This does not mean factors stop working, but it is one reason realized factor premiums since a strategy becomes widely known have, in several documented cases, run below their pre-discovery historical averages.
Using factor tilts in a real portfolio
For a working investor, or a high-earning professional evaluating a fund lineup, multifactor models offer a more honest lens than a single CAPM alpha for judging whether a manager's outperformance reflects skill or simply a persistent style tilt available cheaply elsewhere. A fund charging an active-management fee for what is mostly a value and small-size tilt is charging for something that low-cost, rules-based factor funds now replicate at a fraction of the cost. Before paying an active fee, it is worth asking what factor exposures actually drive a fund's historical returns, and whether the same tilts are available through a lower-cost index or factor-based product.
Factor tilts also carry real portfolio-construction consequences: value and small-size tilts have historically increased volatility and lengthened the periods of relative underperformance an investor must tolerate, which matters enormously for anyone with a shorter time horizon or lower tolerance for tracking a benchmark loosely. A tilt that looks attractive on a 30-year historical average can still produce a genuinely uncomfortable five- or ten-year stretch of relative underperformance in the middle of a career or a retirement.
There is also a tax dimension worth flagging for taxable accounts specifically. Factor strategies, particularly value and momentum, typically require more frequent rebalancing than a plain market-weight index fund, as stocks migrate in and out of the cheap or high-momentum baskets that define the factor. That higher turnover generates more realized capital gains along the way, which matters for after-tax return in a standard brokerage account even when the pre-tax factor premium looks attractive on paper. Holding factor tilts inside tax-advantaged retirement accounts, where turnover does not trigger a current tax bill, sidesteps this problem entirely and is generally the more efficient placement when the choice is available.
Actionable breakdown
- Reading a fund's factor exposure
- Ask for a regression against market, size, and value.
- Check profitability and investment loadings if disclosed.
- Compare stated alpha against a multifactor, not single-factor, model.
- Deciding whether to tilt
- Confirm you can hold through multi-year underperformance.
- Compare active fees against low-cost factor fund alternatives.
- Avoid stacking correlated tilts that add cost without diversification.
- Setting expectations
- Treat historical premiums as ranges, not guarantees.
- Expect crowding to compress premiums once factors are popular.
- Rebalance on a schedule, not in reaction to recent factor returns.
Common pitfalls
Confusing a factor tilt with manager skill: as the Fund A and Fund B examples show, a positive raw return can hide negative multifactor alpha, and a fund's outperformance may simply be its factor exposure showing up during a favorable stretch for that factor.
Abandoning a tilt after a bad multi-year stretch: factor premiums have historically shown long periods of underperformance even when the full-sample average remained positive; switching out after underperformance risks locking in losses right before mean reversion.
Overpaying for a fund that just replicates a known factor: once a factor is well documented, low-cost, rules-based products typically capture most of its historical premium without an active management fee.
Placing a high-turnover tilt in a taxable account by default: factor strategies with meaningful rebalancing turnover generate more realized gains than a plain index fund; where possible, house them in tax-advantaged accounts instead.
The bottom line
Multifactor models explain a materially larger share of the differences in average returns across stocks than a single market beta ever could, but every added factor still carries the same out-of-sample uncertainty and requires the same patience to hold through its inevitable bad stretches, so size any tilt to what you can genuinely tolerate over a full market cycle.
The Fama-French three-factor model · Tests of the multifactor CAPM and APT · A multifactor APT · Value investing basics · Factor investing guide