DCC-GARCH Model (Dynamic Conditional Correlation)
Dynamic Conditional Correlation Generalized Autoregressive Conditional Heteroscedasticity Model · Also known as: DCC-GARCH, Dynamic Conditional Correlation GARCH, Engle DCC model, multivariate DCC
The DCC-GARCH model, introduced by Engle (2002), extends univariate GARCH to capture time-varying correlations between multiple financial time series. It decomposes the multivariate conditional covariance matrix into individual volatility processes and a dynamic correlation matrix, allowing correlations to fluctuate over time while remaining computationally tractable even with many series.
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.
+12 more
When to use it
Use DCC-GARCH when you need to model time-varying correlations among two or more financial or economic time series — for example, in portfolio risk management, contagion analysis, or volatility spillover studies. It is appropriate when individual series exhibit GARCH-type volatility clustering, and when correlations are suspected to shift over time (e.g., around crises). Do not use it when correlations are stable (a constant-correlation CCC-GARCH suffices), when data are non-financial and lack volatility clustering, when the sample is short (fewer than ~200 observations per series), or when the number of series is so large that even the two-step estimation becomes unstable.
Strengths & limitations
- Captures time-varying correlations without requiring a full unrestricted multivariate GARCH, which becomes intractable for large systems.
- Two-step estimation is computationally efficient and scales reasonably well as the number of series grows.
- Produces positive-definite covariance matrices by construction at every observation.
- Well-suited for modelling financial contagion, portfolio diversification benefits, and volatility spillovers.
- The DCC parameters a and b have clear economic interpretations: reaction speed and persistence of correlations.
- Assumes a single pair of scalar DCC parameters (a, b) governing all pairwise correlations, which may be too restrictive when different asset pairs have very different dynamics.
- The two-step estimation introduces efficiency loss relative to joint maximum likelihood estimation.
- Inference about DCC parameters relies on asymptotic results that may not hold well in small samples.
- Requires the number of observations to substantially exceed the number of series; large portfolios (many series) can destabilise the Q_t recursion.
Frequently asked
What is the difference between DCC-GARCH and CCC-GARCH?
In the Constant Conditional Correlation (CCC) model of Bollerslev (1990), the correlation matrix R is fixed over time; only individual variances evolve. DCC-GARCH relaxes this by letting R_t change each period according to the Q_t recursion. DCC is preferred when correlations visibly shift — for example, rising during crises — whereas CCC is simpler and sufficient when correlations are stable.
How do I choose between DCC-GARCH and a fully unrestricted multivariate GARCH (BEKK)?
BEKK-GARCH models every element of the covariance matrix separately, which quickly becomes overparameterised for more than three or four series. DCC-GARCH uses a parsimonious two-scalar (a, b) representation that scales to larger systems. Prefer DCC when you have five or more series or limited data; BEKK may be preferable for small systems where richer dynamics are estimable.
What distribution assumption should I use in the first-step GARCH?
The standard assumption is a normal (Gaussian) distribution, but financial returns often have fat tails. Using a Student-t or skewed-t distribution in the first-step GARCH generally yields better-specified standardized residuals and more reliable DCC estimates. Most econometric packages (Stata, R's rmgarch, Python's arch library) support these alternatives.
How many observations do I need for DCC-GARCH?
As a practical rule, at least 200 observations per series are recommended to obtain stable first-step GARCH estimates. For systems with many series, more data are needed because the Q-bar matrix (unconditional covariance of standardized residuals) must be estimated reliably.
Can DCC-GARCH handle asymmetric correlation dynamics?
The standard DCC model does not distinguish between positive and negative joint shocks. The asymmetric DCC (ADCC) extension by Cappiello, Engle, and Sheppard (2006) adds terms for negative co-shocks, capturing the stylised fact that correlations tend to rise more sharply in downturns than in upturns.
Sources
- Engle, R. F. (2002). Dynamic conditional correlation: A simple class of multivariate generalized autoregressive conditional heteroskedasticity models. Journal of Business and Economic Statistics, 20(3), 339-350. DOI: 10.1198/073500102288618487 ↗
- 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 ↗
How to cite this page
ScholarGate. (2026, June 3). Dynamic Conditional Correlation Generalized Autoregressive Conditional Heteroscedasticity Model. ScholarGate. https://scholargate.app/en/econometrics/dcc-garch-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
- EGARCH modelEconometrics↔ compare
- Granger Causality TestEconometrics↔ compare
- TGARCH modelEconometrics↔ compare
- Vector AutoregressionEconometrics↔ compare