In the single-factor era, returns are pinned to one characteristic. In practice, stock moves are jointly driven by valuation, size, momentum, and more—so one factor rarely explains enough.
That is why we need multi-factor models. This chapter covers three core frameworks—APT, the Barra model, and Fama–French—the main pillars of factor investing.
APT (Arbitrage Pricing Theory) was introduced by Stephen Ross in 1976.
APT’s main contribution is a linear multi-factor framework. Its core claim is:
In math:
where:
Note: APT does not prescribe which factors to use—it only supplies a framework in which returns can be decomposed this way. Factor choice still needs economic logic.
In practice, treating APT as a catch-all formula and stuffing in many ad hoc factors often leads to severe overfitting. APT is a scaffold, not a recipe; factor selection needs economic grounding.
When building APT-style arbitrage strategies, highly correlated factors can make $\beta$ estimates unstable. Orthogonalize factors first, then regress—results usually improve markedly.
If APT is the theory, Barra is built for implementation. Developed by MSCI Barra, it is the multi-factor risk model most widely used by institutions. Factors fall into three groups:
Barra’s core equation is:
where:
The main distinction between Barra and APT:
APT is a pricing model—it explains expected returns. Barra is a risk model—it explains return variation. Barra asks why the book fell 3% today, not why long-run annualized return is 12%.
In Barra risk attribution, industry factors often account for a large share of volatility. A common practice follows: check industry tilts first, then style exposures.
Barra assumes relatively stable factor returns. A-share style regimes flip often—large-caps surged in 2017; small-caps led in 2021. Estimating the factor covariance matrix from the past three years can then be a poor match for today’s market structure.
Fama and French’s three-factor model (1993) is a landmark in academic asset pricing:
The three factors:
They later added two factors for the five-factor model (2015):
The five-factor model is widely used for performance attribution. Example: a fund returns 20% in a year—market 10%, SMB 5%, HML 3%, and the residual 2% as stock-selection skill.
Fama–French decomposes returns cleanly. If a manager claims stock-picking skill but most of the return comes from small-cap exposure, the alpha is less impressive than it looks.
In hedge-fund due diligence, the Fama–French five-factor model is a standard check on historical track records. Significant positive alpha suggests real skill; insignificant alpha often means the manager simply rode a style.
A compact comparison:
| Dimension | APT | Barra | Fama–French |
|---|---|---|---|
| Purpose | Pricing (expected return) | Risk attribution | Performance attribution |
| Factor source | Unspecified | Style + industry + country | Market + size + value + profitability + investment |
| Typical use | Theoretical derivation | Portfolio risk management | Fund performance analysis |
| Complexity | Low | High | Medium |
| Data needs | Low | High (industry classification) | Medium (fundamentals) |
Practical guidance:
A concrete selection logic for live work:
Take a quant long–short equity book of about RMB 1bn. Start from three questions:
Using Fama–French alone for risk control is common but weak: five factors miss industry risk. Switching to Barra usually makes risk attribution far more accurate.
Bottom line: the right tool multiplies effort; the wrong one wastes it.
APT is the theoretical scaffold—returns as a linear combination of factors. Barra is the risk model—managing style and industry exposures. Fama–French is the pricing / attribution workhorse—for performance and alpha analysis. They complement each other; mature quant teams typically use all three.