CAIA® factor models: what happens to apparent fund alpha?

On this page

A Fama–French model asks how much of a portfolio’s excess return moves with broad, observable equity return patterns. The three-factor version allows for the market, size and value; the five-factor version also allows for profitability and investment. A larger set of factors can change the return left unexplained, but that remainder is evidence to investigate, not automatic proof of manager skill.

What is being explained?

Start with the fund return after subtracting the return on a risk-free reference over the same period. That excess return is the quantity on the left side of the model. The factors on the right are returns on particular portfolios or return spreads, rather than labels attached directly to a fund. A sensitivity, often called a loading or beta, tells you how strongly the fund’s returns have moved with each factor across observations. Multiplying a loading by its matching factor return gives a contribution for an illustrative period. The intercept estimated from a regression across many periods is the model’s alpha.

This distinction matters for a CAIA® candidate evaluating an alternative manager. A fund can have a positive reported return yet have no positive return after its market and style exposures are considered. It can also have a negative total return during a weak market while outperforming what its estimated exposures would suggest. Neither observation is enough to declare skill. The comparison needs a consistent time window, appropriate factor series, sensible estimation and an assessment of uncertainty around the intercept.

Read the three-factor model before adding more

The market factor is broad equity market return above the risk-free reference. Small minus big, abbreviated SMB, compares returns on smaller and larger companies. High minus low, or HML, compares high book-to-market companies with low book-to-market companies. A positive SMB loading means the fund tends to behave more like the small-company side of that spread; a negative loading points the other way. A positive HML loading indicates a value orientation in this model. These signs describe co-movement with constructed return series, not a direct count of small or value holdings.

The CAIA Association’s curriculum page makes clear that candidates should anchor their preparation in the current official curriculum. CFA Institute’s research digest describes the market, SMB and HML as the original model’s three factors and explains that the five-factor version adds profitability and investment. Use these source descriptions to understand the framework, while following your own examination year’s assigned curriculum for examinable detail.

In notation, a three-factor regression puts the fund’s excess return on the left and the market, SMB and HML returns on the right, each with a coefficient. The intercept is estimated jointly with those coefficients. An error term remains for each observation because an empirical model does not fit every month exactly. If you are shown a table of exposures and factor returns, check first whether the task asks for one period’s fitted return, one period’s unexplained return, or an estimated multi-period alpha. Those are related calculations with different interpretations.

What the extra factors add

The five-factor version adds robust minus weak, RMW, a profitability return spread, and conservative minus aggressive, CMA, an investment return spread. Positive RMW sensitivity means the fund tends to move with the profitable-company side of that factor; positive CMA sensitivity means it tends to move with the conservative-investment side. Profitability here is based on an accounting measure used in building the factor portfolios. Investment refers to growth in corporate assets in that construction. Neither factor is a promise that those shares will outperform in the next period.

Adding factors changes the comparison being made. What appeared unexplained under the three-factor model may be associated with profitability or investment exposure. Yet the loadings and intercept should normally be re-estimated when the regression specification changes. The value factor can overlap with the added factors, so a coefficient’s size or statistical strength may move. CFA Institute’s digest of the original five-factor research specifically discusses that overlap; its later international-test digest also cautions that different models can lead to different performance inferences. Thus an apparent fall in alpha is a hypothesis about omitted style exposure, not a mechanical law that holds for every fund and dataset.

Work an illustrative attribution

Consider an invented one-period example, with every percentage and loading supplied only to show arithmetic. A fund’s excess return is 10%. Suppose its market loading is 1 and the market excess return is 5%. Its SMB loading is 0.5 with an SMB return of 2%, and its HML loading is 1 with an HML return of 1%. The three contributions are 5, 1 and 1 percentage points, so the fitted three-factor excess return is 7%. The one-period unexplained return is therefore 3 percentage points.

Now suppose, strictly for illustration, that the fund also has an RMW loading of 0.5 and an RMW return of 2%, and a CMA loading of 1 with a CMA return of 1%. Those add one percentage point each. Keeping the original three loadings fixed in this classroom arithmetic, the five-factor fitted return is 9%, leaving one percentage point unexplained. The extra factors account for two points that the shorter specification left in its residual. This is an invented decomposition, not observed fund performance or an estimated five-factor regression.

The qualification is essential. In real work, you would re-estimate all loadings on a common sample, use matching factor definitions and ensure the risk-free and fund returns refer to the same dates and frequency. The difference between the two estimated intercepts need not equal the simple two-point addition above. A single residual is also not alpha: alpha is the intercept of the regression, accompanied by estimation uncertainty. If a question supplies estimated coefficients and asks for the predicted return for one month, multiply and add; if it asks whether a manager generated persistent abnormal performance, the single month cannot answer it.

Why an allocator should care

Factor exposure is useful because it makes a manager comparison more demanding. Two funds with the same return may have taken different market and style risks. If one fund’s return mostly follows familiar factors, an allocator can ask what distinctive process, if any, remains after allowing for those exposures. Conversely, a low loading on these equity factors does not mean a strategy is risk-free: leverage, illiquidity, derivatives or a missing factor may matter. A model measures only the risks it contains and only as well as the data permit.

Check whether the factor definitions fit the portfolio. A global equity fund should not casually be judged using a factor set constructed for a different market if the difference materially affects interpretation. Monthly returns can conceal exposures that change within the month. A short history makes coefficient estimates noisy; a strategy that changes style can make a full-sample average misleading. Fees and trading costs also matter when comparing an investor’s realised return with a model fitted to gross or net observations. State which return you are explaining before attributing the residual.

The factors are empirical descriptions of return patterns, not a causal diagnosis of why a particular manager earned money. A positive RMW loading might come from deliberate profitable-company selection, a sector tilt or correlation with another omitted characteristic. A nonzero estimated intercept might reflect genuine process value, a model gap, measurement error or chance. Examine stability across reasonable specifications and periods before drawing an investment conclusion. For exam purposes, that discipline is the point of applying both models rather than memorising five acronyms.

A practical order for an exam calculation

First identify the dependent return and the risk-free adjustment. Second match each factor return to its own loading, preserving any negative sign. Third add the factor contributions before comparing the sum with the observed excess return. Fourth name the result precisely: a fitted return, a period residual or a regression intercept. Finally, interpret the change between model specifications as conditional on the factors and sample, not as a verdict on a person. The Omni methodology page is useful for structuring practice around explanation and checking; the official CAIA curriculum remains the authority for your examination year.

The most common error is to call all unexplained performance skill. Another is to treat SMB, HML, RMW or CMA as percentages of holdings instead of returns to factor portfolios. A third is to assume five-factor alpha must always be smaller. The model can expose an omitted systematic tilt, but its estimate depends on sample, factor correlations and the way a fund actually behaves. Use the calculation to sharpen a question about performance, then test the question with evidence.