Factor models can be classified by whether they infer exposures from an investment’s return stream or compute them from the underlying holdings’ characteristics.
Return-based models
Infer factor exposures by regressing a portfolio’s return time series against factor return series. Only the portfolio’s return stream is needed — no knowledge of what it holds.
Sharpe’s (1992) returns-based style analysis is the classic example: regress a fund’s returns on a set of asset class or style indices to infer its style exposures. The fama-french-three-factor, carhart-four-factor, and fama-french-five-factor models work this way when applied to evaluate a fund or portfolio.
Holdings-based models
Compute factor exposures directly from the characteristics of the underlying securities (M ratio, market cap, beta, industry membership, etc.). Full position-level holdings are required. MAC3 and other commercial risk models (Barra, Axioma) work this way: each stock gets an exposure vector from its fundamentals, and the portfolio’s exposure is the holdings-weighted sum.
Trade-offs
| Return-based | Holdings-based | |
|---|---|---|
| Input required | Return series only | Full holdings |
| Transparency | Opaque (inferred) | Exact decomposition |
| Frequency | Limited by return history | Point-in-time |
| Use case | Evaluating external managers, funds with undisclosed holdings | Internal risk management, portfolio construction, optimization |
| Weakness | Exposures estimated with noise and lag; can’t detect recent style drift | Requires position-level data; sensitive to factor definition choices |
Different questions answered
Return-based analysis asks “what factors does this portfolio behave like?” Holdings-based analysis asks “what factors does this portfolio actually hold?”
A fund may appear to have value exposure in a return-based analysis simply because its returns happen to correlate with HML, even if none of its holdings have high book-to-market ratios. Conversely, a holdings-based model will show the exact value tilt from each position but cannot be applied to a fund whose holdings are not disclosed.
Relationship to factor model construction
This distinction overlaps with, but is not identical to, the time-series vs. cross-sectional construction distinction (see overview):
- Time-series models (traded factor portfolios, cross-sectional regression of returns) are the standard academic approach and naturally lend themselves to return-based analysis
- Cross-sectional / fundamental models (characteristic-based exposures, daily regression for factor returns) are the commercial risk model approach and are inherently holdings-based
Statistical factor models (PCA-derived) are a third category that operates purely on the return covariance matrix without economic interpretation.
Commercial landscape
Return-based platforms
- Factor Lens: 18-factor multi-asset model using hierarchical residualization. Cloud platform for manager evaluation, portfolio analytics, and risk decomposition. The dominant return-based platform for institutional investors.
- Liquid Factor Models (Rosenthal 2024): 10-factor model built from futures, bonds, and ETFs. Designed for hedging overlays. Academic proposal with practical focus.
- MPI Stylus Pro: Dynamic Style Analysis (DSA), an enhancement of Sharpe’s RBSA with rolling windows and dynamic factor selection. Since 1992.
- Zephyr StyleADVISOR: The first commercial RBSA software (1992). Performance analysis, peer group analysis, style attribution.
- Morningstar: Offers both return-based and holdings-based style analysis for mutual fund evaluation.
Holdings-based platforms
- Bloomberg MAC3: 14 style factors, equity-focused, daily updates, term structure of risk.
- MSCI Barra (GEM, USE): The original commercial fundamental factor model (Rosenberg 1974). Multiple equity models by region.
- Axioma (Qontigo): Offers both fundamental and statistical risk models with robust estimation.
- Northfield: Multi-asset risk models.
- R-Squared / FactSet: Risk models integrated into portfolio management platforms.
Return-based models dominate manager evaluation and alternatives analysis. Holdings-based models dominate internal risk management and portfolio optimization.