Bayesian EGARCH Model
Bayesian Exponential Generalized Autoregressive Conditional Heteroscedasticity Model · Also known as: Bayesian EGARCH model, Bayesian Exponential GARCH, EGARCH with Bayesian estimation, B-EGARCH
The Bayesian EGARCH model combines Nelson's (1991) Exponential GARCH specification — which models the log of conditional variance and captures the leverage effect — with Bayesian posterior inference via Markov Chain Monte Carlo (MCMC). This allows full uncertainty quantification of all volatility parameters, including the asymmetry coefficient, without requiring large-sample normality of the estimates.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use Bayesian EGARCH when you need to model asymmetric volatility (leverage effect) in financial return series and want full posterior uncertainty over the asymmetry and persistence parameters — especially with short samples (fewer than 500 observations) where maximum likelihood estimates can be unreliable. It is also appropriate when incorporating prior knowledge about parameter ranges, when comparing competing EGARCH specifications via formal Bayesian model selection, or when propagating volatility uncertainty into downstream risk measures such as Value-at-Risk. Do not use it when a standard (frequentist) EGARCH is adequate and computation time is a binding constraint, or when the data clearly show no asymmetric volatility pattern — in that case a symmetric GARCH may suffice.
Strengths & limitations
- Captures the leverage effect: negative shocks raise volatility more than positive shocks of equal size.
- Log-variance formulation guarantees positive conditional variance without imposing inequality constraints on parameters.
- Full posterior inference provides credible intervals for all parameters, including the asymmetry coefficient, without relying on asymptotic approximations.
- Handles short samples better than maximum likelihood because priors regularise the estimation surface.
- Enables principled Bayesian model comparison across EGARCH, GARCH, and TGARCH specifications.
- Posterior predictive distributions give honest probabilistic volatility forecasts.
- MCMC sampling is computationally expensive relative to maximum likelihood EGARCH, particularly for high-frequency data or large models.
- Results depend on the choice of priors; poorly specified priors can distort inference, especially in small samples.
- Convergence of the Markov chain must be verified carefully; non-convergence produces misleading posteriors.
- The model still requires a correctly specified error distribution (e.g., Student-t); misspecification biases the volatility estimates.
- Interpretation requires familiarity with Bayesian concepts such as credible intervals and posterior predictive checks.
Frequently asked
What is the leverage effect and why does EGARCH capture it better than standard GARCH?
The leverage effect is the empirical tendency for negative return shocks to increase volatility more than positive shocks of the same magnitude. Standard GARCH uses squared residuals, so positive and negative shocks of equal size affect variance identically. EGARCH includes the signed standardised residual through the gamma coefficient, allowing asymmetric responses directly.
Why use Bayesian estimation instead of maximum likelihood for EGARCH?
Maximum likelihood provides point estimates with asymptotic standard errors that can be unreliable in short samples or when the likelihood surface is flat or multi-modal near the leverage parameter. Bayesian estimation yields a full posterior distribution over all parameters, gives honest uncertainty quantification, and can incorporate prior information — all advantages when samples are small or model uncertainty is high.
How many MCMC draws are needed for reliable inference?
This depends on the mixing rate of the sampler, but a common starting point is a burn-in of 5,000 draws followed by 20,000–50,000 retained draws with convergence checked via the Gelman-Rubin statistic (values below 1.1 indicate adequate convergence). Thin the chain if autocorrelation among draws is high.
What error distribution should I assume for the innovations?
Financial returns typically have heavier tails than a normal distribution, so a Student-t distribution for the standardised innovations is generally preferred. The degrees-of-freedom parameter can itself be assigned a prior and estimated from the data, providing a flexible tail-thickness estimate alongside the volatility dynamics.
How do I compare a Bayesian EGARCH to a Bayesian GARCH or TGARCH?
Use the Deviance Information Criterion (DIC) or Bayesian leave-one-out cross-validation (LOO-CV). Lower DIC or higher LOO expected log predictive density favours the better-fitting model. If model uncertainty is substantial, Bayesian model averaging over the competing specifications is also feasible.
Sources
- Nelson, D. B. (1991). Conditional heteroskedasticity in asset returns: A new approach. Econometrica, 59(2), 347–370. DOI: 10.2307/2938260 ↗
- Nakatsuma, T. (2000). Bayesian analysis of ARMA-GARCH models: A Markov chain sampling approach. Journal of Econometrics, 95(1), 57–69. DOI: 10.1016/S0304-4076(99)00029-9 ↗
How to cite this page
ScholarGate. (2026, June 3). Bayesian Exponential Generalized Autoregressive Conditional Heteroscedasticity Model. ScholarGate. https://scholargate.app/en/econometrics/bayesian-egarch
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- ARCH modelEconometrics↔ compare
- Bayesian DCC-GARCHEconometrics↔ compare
- Bayesian GARCH modelEconometrics↔ compare
- Bayesian TGARCHEconometrics↔ compare
- Bayesian VAR modelEconometrics↔ compare
- EGARCH modelEconometrics↔ compare