Definition
Heteroskedasticity means the conditional or unconditional variance of an error or response is not constant across observations or states. In finance, volatility can vary over time even when the mean equation is simple. Heteroskedasticity can leave coefficient estimates usable under conditions while making naive standard errors wrong.
How it is calculated
Conditional variance v(X_t,t) of error epsilon_t given X_t is nonconstant across conditioning states or observations.
| Symbol | Meaning | Unit |
|---|---|---|
| Var | conditional variance | squared error units |
| epsilon | error notation | response units |
| X | conditioning variables | predictor units |
| v | conditional variance function | squared error units |
Interpretation
Some observations arrive with a wide cloud of possible errors and others with a narrow cloud. Treating both clouds as equally noisy can distort uncertainty calculations. A volatility model or heteroskedasticity-robust inference addresses different parts of that problem.
Assumptions and conventions
-
The conditioning variables and error definition are specified.
-
Inference uses a variance estimator suited to the observed heteroskedasticity.
Conditional variance v is in squared error units.
Limitations and common misreadings
-
Heteroskedasticity is about changing variance, not necessarily changing mean.
-
Robust standard errors do not automatically fix a misspecified mean model.
References
- Halbert White. (1980). A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity. Econometrica. Sections 1–2. doi:10.2307/1912934. Source

