Bayesian ARCH Model
Bayesian Autoregressive Conditional Heteroskedasticity Model · Also known as: Bayesian ARCH, ARCH with Bayesian estimation, Bayesian conditional heteroskedasticity model, B-ARCH
The Bayesian ARCH model estimates Engle's Autoregressive Conditional Heteroskedasticity specification within a Bayesian framework. Instead of maximising a likelihood, it combines a prior distribution over the volatility parameters with the data likelihood to obtain a full posterior distribution, providing richer uncertainty quantification than classical maximum-likelihood ARCH.
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 the Bayesian ARCH model when you have a financial or macroeconomic time series with evidence of time-varying volatility and you want full posterior uncertainty over the volatility parameters — particularly in small or moderate samples where maximum-likelihood asymptotics are unreliable. It is well-suited when you have prior information about plausible parameter ranges, or when you need probabilistic forecasts of volatility with calibrated credible intervals. Avoid it when the sample is very large and computational speed is critical (classical MLE ARCH is faster), when there is no evidence of volatility clustering (a constant-variance model suffices), or when the volatility dynamics are better captured by a GARCH extension — the ARCH(q) order needed to adequately model persistence can become very large.
Strengths & limitations
- Provides full posterior distributions over ARCH parameters, enabling calibrated credible intervals without relying on asymptotic normality.
- Naturally incorporates prior information about volatility dynamics, which is valuable when the sample is short.
- Propagates parameter uncertainty through to volatility forecasts, yielding probabilistic forecast distributions rather than point estimates.
- Valid inference in small samples where maximum-likelihood standard errors and Wald tests can be unreliable.
- Facilitates Bayesian model comparison via marginal likelihoods or Bayes factors across competing ARCH orders.
- Computationally intensive: MCMC sampling is orders of magnitude slower than ML optimisation for large datasets.
- Posterior inference is sensitive to prior specification, especially when data are limited; poorly chosen priors can dominate the posterior.
- Pure ARCH models require high orders q to capture persistent volatility clustering; GARCH(1,1) is almost always more parsimonious.
- MCMC convergence must be diagnosed carefully; failing chains produce incorrect posteriors that can look superficially plausible.
- Software implementations are less standardised than classical ARCH packages, requiring greater econometric expertise to set up correctly.
Frequently asked
How is Bayesian ARCH different from classical ARCH?
Classical ARCH uses maximum likelihood to obtain point estimates and constructs confidence intervals via asymptotic theory. Bayesian ARCH treats parameters as random, combines a prior with the likelihood, and produces a full posterior distribution. The result is calibrated credible intervals valid in small samples and a natural mechanism for incorporating prior knowledge.
Should I use ARCH or GARCH in a Bayesian framework?
GARCH(1,1) is almost always preferred because it captures volatility persistence with far fewer parameters than a high-order ARCH. However, for analytical or pedagogical purposes — or when structural constraints make a pure ARCH model more interpretable — Bayesian ARCH remains a valid choice. Run both and compare via Bayes factors.
How do I choose the ARCH order q in a Bayesian setting?
Compute the partial autocorrelation function of squared residuals from a preliminary mean equation to identify candidate orders, then use Bayesian information criteria or Bayes factors computed from MCMC output to select among orders formally.
What priors are standard for Bayesian ARCH parameters?
A common choice is a truncated normal or gamma prior for the intercept alpha_0 (ensuring positivity) and truncated normal priors for the ARCH coefficients alpha_i (ensuring non-negativity and that their sum is below one to guarantee covariance stationarity). Hyper-parameters can be set to reflect expected volatility levels in the series.
What software can I use to estimate Bayesian ARCH models?
Stan (via RStan or PyStan) and JAGS/BUGS support custom ARCH specifications through probabilistic programming. The R package 'bayesforecast' and some functions in 'bayesm' also cover ARCH-family models with MCMC backends.
Sources
- Engle, R. F. (1982). Autoregressive conditional heteroscedasticity with estimates of the variance of United Kingdom inflation. Econometrica, 50(4), 987–1007. DOI: 10.2307/1912773 ↗
- Geweke, J. (1989). Exact predictive densities for linear models with ARCH disturbances. Journal of Econometrics, 40(1), 63–86. DOI: 10.1016/0304-4076(89)90030-4 ↗
How to cite this page
ScholarGate. (2026, June 3). Bayesian Autoregressive Conditional Heteroskedasticity Model. ScholarGate. https://scholargate.app/en/econometrics/bayesian-arch-model
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 EGARCHEconometrics↔ compare
- Bayesian GARCH modelEconometrics↔ compare
- Bayesian TGARCHEconometrics↔ compare
- DCC-GARCH modelEconometrics↔ compare
- GARCH ModelEconometrics↔ compare