What Multifactor Models Actually Explain About Returns
A single measure of market sensitivity, beta, often fails to explain why two stocks with similar market risk post very different returns over time. Multifactor models solve this by breaking a stock's return into several separate, independently priced risk exposures instead of forcing everything through one number.
Why one factor was never going to be enough
A single-factor model, CAPM chief among them, assumes market risk is the only systematic force affecting every security's return, and that everything left over is pure noise, unique to that one company, that diversification erases. In practice, groups of stocks move together for reasons that have nothing to do with the broad market's direction on a given day. Energy companies rise and fall together with oil prices regardless of what the S&P 500 is doing. Banks move together with interest rate expectations. Exporters move together with currency swings. These shared, sector- or theme-level movements are systematic in the sense that diversifying across many stocks within the same exposure does not remove them, yet a single market beta captures none of this structure explicitly.
A multifactor model responds by allowing more than one systematic source of risk to enter the pricing equation, each with its own sensitivity, called a factor beta or factor loading, and its own separately estimated risk premium, the extra expected return investors demand for bearing exposure to that specific factor.
The problem is not that CAPM's insight was wrong; it is that CAPM's insight was incomplete. Diversifiable, firm-specific risk really is unrewarded, and market risk really is compensated. What CAPM's single factor cannot express is that there can be more than one kind of undiversifiable, compensated risk operating at once, and that two stocks with identical market betas can carry very different amounts of these other, equally systematic exposures. A regional bank and a semiconductor exporter might each have a market beta near 1.0, yet one is dominated by domestic interest rate risk and the other by global currency and cyclical demand risk, two entirely different stories a single number cannot tell apart.
The general multifactor formula
The general form extends the single-factor security market line directly: expected return = risk-free rate + (beta1 × factor1 premium) + (beta2 × factor2 premium) + (beta3 × factor3 premium) + .... Each factor beta is estimated the same way a market beta is, typically through a regression of the security's historical returns against a constructed series representing that factor, such as the return difference between small and large companies, or between high and low valuation companies. Each factor premium is estimated separately, usually from the long-run historical average return earned by a portfolio designed to have exposure to that one factor and nothing else.
It is worth being explicit about what this regression actually estimates and what it does not. A factor beta tells you how sensitive a security's historical returns have been to a given factor's movements; it does not, by itself, tell you why that sensitivity exists, and it does not guarantee the same sensitivity will hold going forward if the underlying business changes. A regional bank's interest-rate sensitivity, for instance, depends on the specific structure of its loan book, fixed versus floating rate, short versus long duration, which can shift meaningfully after a merger, a change in lending strategy, or a shift in the broader competitive environment for deposits, all of which can move the true factor exposure well before enough new return data accumulates to reveal the shift in an updated regression.
The number of factors used varies by model and by purpose: some practitioner models use three or four, others used in institutional risk management can include a dozen or more, spanning valuation, quality, momentum, volatility, profitability, and various macroeconomic and sector groupings, chosen to match the specific risks the institution actually needs to monitor and report on.
Macroeconomic factors versus fundamental factors
Multifactor models generally fall into two families. Macroeconomic factor models use observable economy-wide variables directly, unexpected changes in inflation, industrial production growth, credit spreads, or interest rates, as the factors themselves, on the theory that these are the actual economic forces investors are compensated for bearing. Fundamental factor models, more common in practice, instead use characteristics of the securities themselves, market capitalization, book-to-market ratio, recent price momentum, profitability, as proxies for exposure to underlying risks that may be harder to observe directly but show up reliably in return patterns. The Fama-French family of models and its many extensions are the best known examples of the fundamental factor approach, and they dominate both academic research and much of the factor-fund industry built around them.
A third, less common family, statistical factor models, extracts factors directly from the return data itself using a technique called principal components analysis, without specifying in advance what any given factor represents economically. This approach can capture return patterns a researcher might not have thought to include as a named characteristic, and it tends to fit historical data very well by construction, but the resulting factors are often difficult to interpret or explain in plain economic terms, which makes them harder to trust going forward and harder to communicate to a client or investment committee than a factor with an obvious label like size or value.
The math, worked through twice
Suppose the risk-free rate is 4%, and a technology growth stock has a market beta of 1.1 against a market premium of 5.5%, a size factor exposure of -0.3 (it behaves like a large company, so this works against it) against a size premium of 2%, and a value factor exposure of -0.5 (it is a growth stock, trading well above book value) against a value premium of 3%. Expected return = 4% + (1.1 × 5.5%) + (-0.3 × 2%) + (-0.5 × 3%) = 4% + 6.05% - 0.6% - 1.5% = 7.95%. A single-factor CAPM estimate using only beta would have produced 4% + 6.05% = 10.05%, more than two full points higher, because it ignores the stock's negative exposure to both the size and value premiums.
Now consider a small industrial company with a market beta of 0.9, a size exposure of 0.7 (it is genuinely small), and a value exposure of 0.6 (it trades cheaply relative to book value). Expected return = 4% + (0.9 × 5.5%) + (0.7 × 2%) + (0.6 × 3%) = 4% + 4.95% + 1.4% + 1.8% = 12.15%. Here the single-factor CAPM estimate, 4% + 4.95% = 8.95%, understates the multifactor figure by more than 3 points, because it misses the additional compensation the model attributes to the company's small size and cheap valuation.
What the evidence shows
The empirical case for multifactor models rests on their consistently higher explanatory power for the cross-section of average stock returns compared with CAPM alone, replicated across many countries and multi-decade sample periods since the original Fama-French research. This does not mean every added factor is permanent or reliably tradable; several factors that looked robust in early samples have weakened, some to the point of near disappearance, once they became widely known and widely traded, a pattern consistent with the idea that at least part of a factor premium can reflect a temporary reward for holding an under-followed exposure rather than a permanent structural risk premium. Momentum in particular has shown notably higher volatility in its premium, including sharp, sudden reversals during market turning points, than the steadier value and size premiums.
An additional, often overlooked strand of evidence involves how factor premiums have historically behaved after their academic discovery was widely publicized. Some studies comparing pre-publication and post-publication average returns for a given factor find a meaningful decline in the premium after the underlying paper became widely known and widely traded on, consistent with the idea that publicizing a factor invites enough capital chasing it to partially compress the very premium being described. This does not mean published factors stop working entirely, most still show a positive, if smaller, premium in more recent data, but it is a caution against assuming any documented factor premium will persist at its original historical magnitude indefinitely into the future.
Reading a fund's factor exposures
For a retail investor, the most direct use of multifactor thinking is diagnostic: when a fund reports beating a broad benchmark, a factor decomposition can reveal whether the outperformance came from genuine security selection or simply from a tilt toward small, cheap, or high-momentum stocks, an exposure that a low-cost, rules-based factor fund can now replicate directly without paying active management fees for it. This matters enormously for cost: paying a percentage point or more in annual fees for what turns out to be a simple, replicable value tilt is paying active-management prices for something increasingly available at index-fund prices. Many fund providers now publish a regression-based factor decomposition alongside standard performance reporting, breaking a fund's historical return into a market component, a size component, a value component, and a residual, unexplained component; that residual, not the fund's headline total return, is the number that most honestly represents what an investor is paying active fees to obtain.
Actionable breakdown
- Ask what factors a fund actually holds, not just its label.
- Check published factor loadings if the fund provider discloses them.
- Compare returns against factor-adjusted benchmarks, not the S&P 500 alone.
- A small-cap value fund should be judged against a small-cap value index.
- Diversify across factors, not just across individual stocks.
- No single factor outperforms in every market environment; they tend to move in different cycles.
- Weigh cost against what a factor tilt is actually buying you.
- Low-cost factor index funds now replicate most documented tilts.
Common pitfalls
A frequent error is confusing a larger number of factors with a better model; adding factors that were selected because they happened to fit past data well, rather than because they represent a coherent economic story, tends to overfit history and predicts poorly going forward. A second pitfall is paying active-management fees for a factor exposure a cheap, rules-based fund already delivers. A third is assuming factor premiums are stable and always positive; several, momentum especially, have gone through multi-year stretches of negative or flat returns even while remaining positive over the full multi-decade sample. A fourth pitfall is ignoring correlation between factors included in the same model; size and value have shown meaningful correlation with each other in some markets and periods, which can inflate a model's apparent explanatory power without adding much genuinely new information beyond what either factor would provide on its own.
The bottom line
Multifactor models replace a single blurry risk number with several sharper, economically grounded ones, letting investors see what is genuinely driving a portfolio's returns and judge, with real evidence rather than a marketing claim, whether they are paying fairly for it.
Related reading: factor investing, understanding investment risk, arbitrage pricing theory, the Fama-French three-factor model, building a multifactor APT model.