Definition
Factor identification is the set of assumptions and variation that make a factor exposure or factor effect distinguishable from competing explanations. In a regression, fit alone can estimate coefficients without establishing that the factor has a unique structural interpretation. Identification depends on timing, exclusion or pricing assumptions, rank, measurement and the comparison design.
How it is calculated
A simple rank condition for K factor regressors requires the expected regressor cross-product matrix to have rank K.
| Symbol | Meaning | Unit |
|---|---|---|
| rank | matrix rank | count |
| E | expectation | matrix units |
| X | regressor vector | factor units |
| K | number of regressors | count |
Interpretation
If two factor columns always move together, the data cannot tell their separate contributions apart. Identification is the reason a coefficient has a distinct meaning, not merely the reason an optimizer returns a number. A tidy fit can coexist with weak identification.
Assumptions and conventions
-
The regressors contain enough independent variation for the target coefficients.
-
The proposed interpretation includes assumptions beyond statistical fit where causal or structural claims are made.
Rank and K are counts; the matrix has squared regressor units.
Limitations and common misreadings
-
Low regression residuals do not establish unique factor interpretation.
-
A nonzero loading is not by itself causal evidence.
References
- Andrew Ang. (2014). Asset Management: A Systematic Approach to Factor Investing. Oxford University Press. Chapter 1, What Is a Factor? doi:10.1093/acprof:oso/9780199959327.001.0001. Source

