Portfolio and risk · Portfolio theory
Factor models: explaining a portfolio’s returns with several betas
How multi-factor models extend CAPM with size, value and other exposures, shown by regressing six funds on three factor proxies built from ETFs, 2005 to 2026.
On this page
A factor model explains an investment's returns as the sum of its exposures to a small number of common influences, called factors, plus a remainder. The Capital asset pricing model (CAPM): what return a beta should earn is the simplest case: one factor, the market, and one exposure, beta. Multi-factor models add more, such as whether the holdings are small or large companies and whether they are cheap or expensive relative to their fundamentals.
The point is diagnostic. Two funds can have the same market beta and behave quite differently because one leans towards small value companies and the other towards large growth companies. A factor model measures those leanings from returns alone, and it revises the answer to the question every investor in an active fund should ask: after accounting for the kinds of risk this fund took, was anything left over?
The idea
Think of a fund's monthly returns as a recipe. The market supplies most of the flavour: when stocks in general rise 3%, most stock funds rise by something near 3%. But the recipe has other ingredients. A fund full of small companies gets an extra push in months when small companies beat large ones, and a drag when they lag. A factor model works out the quantity of each ingredient from the taste.
The figure shows three exposures for six funds. All six have a market exposure not far from 1: they are stock funds. They differ sharply on the other two. QQQ and ARKK have large negative value exposures, meaning they behave like the opposite of a value fund. Consumer staples (XLP), real estate (VNQ) and the low-volatility fund (USMV) lean the other way.
How it works
Each factor other than the market is a difference between two portfolios: a long position in one group and a short position in another. Fama and French's size factor (written SMB, small minus big) is the return of small companies minus that of large ones. Their value factor (HML, high minus low) is the return of stocks with a high ratio of book value to price minus that of stocks with a low one. Mark Carhart added momentum (recent winners minus recent losers) in 1997, and Fama and French added profitability and investment in 2015. Kenneth French publishes all of these series, monthly, back to 1926 for the original three.
The proxies used in this lesson
The published factors are built from thousands of individual stocks. This lesson uses a rougher version made from three ETFs, so that every number can be traced to Finabase's own data:
- Market: SPY's return minus the Treasury bill rate.
- Size: IWM (small companies) minus SPY.
- Value: VTV (large value) minus SPY.
These are approximations. The value proxy in particular compares value with the whole market, not with growth stocks, so its exposures are larger in magnitude than published HML loadings would be. The mechanics and the conclusions are the same.
A worked example on six funds
The data: monthly total returns from January 2005 (or the fund's first full month) to October 2026.
| Fund | One factor: beta | Alpha a year | Explained | Three factors: market | Size | Value | Alpha a year | Explained |
|---|---|---|---|---|---|---|---|---|
| RSP (S&P 500, equal weight) | 1.08 | −0.8% | 90% | 1.07 | 0.25 | 0.56 | +0.2% | 96% |
| QQQ (Nasdaq-100) | 1.13 | +3.0% | 82% | 1.03 | 0.12 | −1.15 | +2.9% | 92% |
| XLP (consumer staples) | 0.56 | +2.4% | 47% | 0.69 | −0.33 | 0.90 | +1.8% | 62% |
| VNQ (real estate) | 1.07 | −3.1% | 54% | 1.05 | 0.34 | 0.64 | −1.8% | 59% |
| USMV (low volatility) | 0.69 | +1.1% | 72% | 0.80 | −0.20 | 0.61 | −0.1% | 81% |
| ARKK (active growth) | 1.76 | −3.8% | 52% | 1.22 | 1.34 | −2.43 | +2.2% | 71% |
"Explained" is the regression's R², the share of the variation in the fund's monthly returns accounted for by the factors. USMV starts in December 2011 and ARKK in December 2014.
Take the first row. RSP holds the S&P 500's companies in equal amounts, which means far more in the smaller and cheaper members than the index has. Under CAPM it shows a beta of 1.08 and an alpha of −0.8% a year, as if it had been slightly mismanaged. The three-factor regression tells a different story: a market exposure of 1.07, plus a quarter-unit of size and half a unit of value. Both of those factors lost money over the period. Once they are accounted for, the alpha is +0.2%, which is zero for practical purposes. The fund did exactly what its construction implies.
USMV shows the reverse case. Its CAPM alpha of +1.1% looks like a reward for choosing calm stocks. With the value exposure included, it disappears.
ARKK is the most striking. Its single-factor beta of 1.76 suggested a fund nearly twice as risky as the market. Three factors show a market exposure of 1.22 with enormous size and anti-value exposures: it behaved like a leveraged bet on small, expensive companies. Its alpha flips from −3.8% to +2.2%, and neither figure is reliable, since even three factors explain only 71% of its returns.
Factors do not always pay
Fama and French found in the early 1990s that, in US data from 1963, small companies and value companies had earned more than CAPM predicted, and proposed that the extra return was a reward for extra risk. The period since has been unkind to both.
| Proxy | Average a year, 2005–2026 | 2005–2015 | 2016–2026 | Volatility |
|---|---|---|---|---|
| Size (IWM − SPY) | −1.3% | +0.4% | −3.1% | 9.6% |
| Value (VTV − SPY) | −1.1% | −0.4% | −1.8% | 5.4% |
A dollar in the small-company fund ended the period worth 62% of a dollar in SPY. A factor premium is an average over many decades with long stretches on the wrong side, and there is live disagreement about whether the historical premiums were compensation for risk, the result of investor mistakes that have since been corrected, or partly an accident of the data. Factor investing: five factor funds against the S&P 500, tested covers the investing side of that argument.
Using it
- Read a fund's exposures before its alpha. A fund that reports beating its index may simply hold smaller or cheaper stocks than the index. The exposures say what was bought; the alpha says whether anything else was added.
- Check for hidden overlap. Two funds with different names and similar factor exposures are the same bet. A portfolio of XLP, USMV and a dividend fund is three versions of one position.
- Compare an active fund with a cheap factor fund. If an active manager's returns are fully explained by market, size and value exposures, those exposures can be bought through index funds at a fraction of the fee.
- Use exposures to anticipate behaviour. A portfolio with a large negative value exposure will have a bad time in a period like 2022, when expensive growth stocks fell hardest, whatever its individual holdings are.
Where it fails or misleads
- Alpha is model-dependent. The table shows alphas changing sign when factors are added. There is no single correct alpha, only alpha relative to a stated model.
- Factors are chosen after looking at the data. Hundreds of factors have been proposed in academic journals. Many were found by searching, and a large share have not held up in later data or in other countries.
- Exposures drift. A regression over twenty years gives an average. A fund that changed its approach, or stocks that grew from small to large, will have exposures today that differ from the estimate.
- Published factors cannot be bought. They are long-short portfolios with no trading costs. Real funds are long-only and pay to trade, so they capture part of a factor at best.
- Unexplained is not the same as skill. A low R² means the model fits the fund badly, as with real estate here (59%). The leftover return then reflects missing factors as much as ability.
- The proxies in this lesson are crude. ETF differences are noisier and narrower than the published series. Use French's data library for serious work.
Compared with its neighbours
| Model | Factors | Typical use |
|---|---|---|
| CAPM | Market | Cost of equity; a first measure of alpha |
| Fama–French three-factor | Market, size, value | Evaluating stock funds |
| Carhart four-factor | Those three plus momentum | Mutual-fund performance studies |
| Fama–French five-factor | Market, size, value, profitability, investment | Academic asset pricing |
| Returns-based style analysis | A set of index funds (large growth, small value, bonds…) | Working out what mix of indexes a fund resembles |
| Risk models (commercial) | Dozens: industries, countries, styles | Institutional risk management |
How Finabase does it
Finabase does not run factor regressions and does not report factor exposures, R² or multi-factor alpha for any fund or portfolio.
It offers a holdings-based view of the same question. An ETF's Holdings tab lists its largest positions with their weights, the share of the fund in its top ten, and its breakdown by sector and country. A fund with a heavy weight in technology and a high top-ten concentration will show a negative value exposure in a regression; one spread across utilities, staples and health care will show the opposite. The screener can filter stocks by market capitalization and by P/E ratio, which are close to the characteristics the size and value factors are built from (the published value factor uses book value to price).
The comparison tab gives the returns-based view informally. Overlaying IWM, VTV and SPY, as in the second figure, shows which style led over the chosen range, and adding a fund to the same chart shows which of them it moved with.
Frequently asked questions
What is a factor, exactly?
A source of return that many securities share. In practice it is defined as the return of one group of stocks minus another: small minus large, cheap minus expensive, recent winners minus recent losers.
How is a factor model different from CAPM?
CAPM has one factor. A multi-factor model adds others, so it can tell apart funds that have the same market beta and different styles. CAPM is the special case in which the extra exposures are ignored.
Does a positive factor exposure mean higher returns?
Only if the factor itself earns a positive return in the period. Exposure to value was a drag from 2005 to 2026. Exposure describes behaviour; it promises nothing.
What does a negative exposure mean?
That the fund behaves like the opposite side of the factor. A value exposure of −1.15 for QQQ means it tends to rise when growth stocks beat value stocks, and fall behind when value leads.
Where can I get real factor data?
Kenneth French's data library publishes monthly and daily returns of the standard factors for the US and other regions, free of charge. A spreadsheet's regression tool is enough to estimate a fund's exposures from them.
Summary
A factor model writes a fund's return as alpha plus its exposures to a few shared sources of return, estimated by regression. With proxies for size and value added to the market, the share of an equal-weight S&P 500 fund's returns explained rose from 90% to 96% and its apparent underperformance vanished; a low-volatility fund's apparent outperformance vanished too. Exposures describe what a fund is. They do not predict that it will do well: the size and value proxies both lost ground to the market over 2005 to 2026. And because adding factors changes alpha, any claim of alpha should come with the model it was measured against.
Check your understanding
4 questions on this lesson. Each answer comes with the reason, and nothing is scored.
A fund has a CAPM alpha of +1.1% a year. After adding size and value factors, its alpha is −0.1%. What is the most reasonable reading?
- The manager’s skill disappeared
- The apparent outperformance came from exposure to those factors, not from selection beyond them
- The three-factor model is wrong
Show the answer
B. The apparent outperformance came from exposure to those factors, not from selection beyond them Alpha is whatever the model leaves unexplained. When extra factors account for it, the earlier alpha was a style exposure that the single-factor model could not see.
What is the size factor (SMB) in the Fama–French model?
- The market value of the fund
- The return of small companies minus the return of large companies
- The number of stocks a fund holds
Show the answer
B. The return of small companies minus the return of large companies Each non-market factor is a difference between two portfolios. SMB is small minus big; HML is high book-to-price minus low.
From 2005 to 2026 a small-company fund trailed SPY by 1.3% a year. What does that say about factor premiums?
- The size factor never existed
- A factor’s long-run average can include decades in which it is negative
- Small companies always underperform
Show the answer
B. A factor’s long-run average can include decades in which it is negative Historical premiums are averages over very long periods. An exposure is not a promise, and whether the premiums will persist is debated.
QQQ has a value exposure of about −1.15 in this lesson’s regression. How will it tend to behave in a year when value stocks beat growth stocks by a wide margin?
- It will tend to lag the market
- It will tend to lead the market
- The exposure has no effect
Show the answer
A. It will tend to lag the market A negative value exposure means the fund behaves like the growth side. When value leads, that exposure subtracts from its return relative to the market.
Sources and further reading
- Eugene F. Fama and Kenneth R. French, “Common Risk Factors in the Returns on Stocks and Bonds,” Journal of Financial Economics, vol. 33, no. 1 (1993)
- Mark M. Carhart, “On Persistence in Mutual Fund Performance,” Journal of Finance, vol. 52, no. 1 (1997)
- Stephen A. Ross, “The Arbitrage Theory of Capital Asset Pricing,” Journal of Economic Theory, vol. 13, no. 3 (1976)
- William F. Sharpe, “Asset Allocation: Management Style and Performance Measurement,” Journal of Portfolio Management, vol. 18, no. 2 (1992)
- Kenneth R. French, Data Library — the published factor return series
- Finabase chart data — monthly bars and dividends of SPY, IWM, VTV, RSP, QQQ, XLP, VNQ, USMV and ARKK through October 2026; regressions computed by Finabase
