Definition
A factor regression relates an asset or portfolio excess return to one or more factor returns and estimates an intercept, loadings and residual. The loadings summarize conditional co-movement in the fitted sample. Inference requires assumptions about the error process, including possible heteroskedasticity or autocorrelation.
How it is calculated
Asset i excess return at t = alpha_i + beta_i transposed times the factor-return vector F_t + residual epsilon_i,t.
| Symbol | Meaning | Unit |
|---|---|---|
| R | asset return | return per period |
| R_f | risk-free return | return per period |
| alpha | regression intercept | return per period |
| beta | factor loading vector | dimensionless |
| F | factor return vector | return per period |
| epsilon | residual notation | return per period |
| y | response | return units |
| X | design matrix with a dimensionless intercept column and return-valued factor columns | mixed regression units |
| b | coefficient vector with a return-valued intercept and dimensionless loadings | mixed regression units |
| e | residual | return units |
-
Arrange the response, factor observations and coefficients in regression form.
-
Fit the coefficients by minimizing squared residuals.
-
Subtract the fitted factor contribution from observed excess return to obtain the residual.
-
Choose standard errors that match the residual-dependence assumptions.
Interpretation
The regression asks how much of a portfolio's movement lines up with named return streams. What remains is the fitted residual, not automatically skill or unexplained economics. A coefficient is conditional on the factors, sample and specification used to estimate it.
Assumptions and conventions
-
The factor definitions and observation timing are fixed.
-
The regression design is appropriate for the estimand and inference method.
-
Residual diagnostics and dependence-robust errors are considered where needed.
Returns, alpha and residual use return per period; beta is dimensionless when factor and asset returns share units.
Limitations and common misreadings
-
An intercept is not automatically causal alpha.
-
A small residual does not prove the factor model is complete.
References
- Eugene F. Fama, Kenneth R. French. (1993). Common risk factors in the returns on stocks and bonds. Journal of Financial Economics. Section 3, time-series regressions. doi:10.1016/0304-405X(93)90023-5. Source

