Sign up
ContentsHideShow

Quant glossary · Time series

GARCH

A GARCH model describes conditional variance as a function of a constant, recent squared innovations and its own recent variance.

Definition

A GARCH model describes conditional variance as a function of a constant, recent squared innovations and its own recent variance. GARCH(1,1) is the standard one-lag form. It models time-varying volatility, not the direction of returns, with omega > 0 and alpha, beta >= 0; usually alpha plus beta is below one for finite unconditional variance.

How it is calculated

σt2=ω+αϵt−12+βσt−12\sigma_t^2=\omega+\alpha\epsilon_{t-1}^2+\beta\sigma_{t-1}^2

Conditional variance at t = omega + alpha times the previous squared innovation + beta times the previous conditional variance.

Symbol Meaning Unit
sigma conditional standard deviation return per period
omega variance intercept return squared per period squared
alpha shock-response coefficient dimensionless
epsilon return innovation return per period
beta variance-persistence coefficient dimensionless
  1. Start with the conditional-variance recursion.

  2. Take unconditional expectations and equate innovation variance with conditional variance.

  3. Under the stated persistence condition, solve for the stationary variance.

Interpretation

A large recent shock raises the next variance estimate, and a high beta lets that elevation fade slowly. This is the model's memory of volatility clustering. The equation says nothing by itself about whether the next return is positive or negative.

Assumptions and conventions

  • Typically omega is positive and alpha and beta are nonnegative.

  • For the usual finite unconditional variance condition, alpha plus beta is less than one.

  • The innovation distribution and mean equation are specified separately.

sigma squared and omega are squared return units; alpha and beta are dimensionless.

Limitations and common misreadings

  • High volatility persistence is not the same as high expected return.

  • alpha plus beta below one is a finite-variance condition, not a fit certificate.

References

  1. Tim Bollerslev. (1986). Generalized autoregressive conditional heteroskedasticity. Journal of Econometrics. Section 2, equations defining the GARCH process. doi:10.1016/0304-4076(86)90063-1. Source

GARCH FAQ

What is GARCH?

A GARCH model describes conditional variance as a function of a constant, recent squared innovations and its own recent variance.

How is GARCH calculated?

Conditional variance at t = omega + alpha times the previous squared innovation + beta times the previous conditional variance.

What does GARCH not tell you or assume?

High volatility persistence is not the same as high expected return.

Browse all glossary termsTime series termsQuant research library

Definitions are educational and descriptive, not investment advice. Worked examples use labelled sample data.