Nonlinear DCC-GARCH Model (Asymmetric Dynamic Conditional Correlation)
Nonlinear Dynamic Conditional Correlation GARCH Model · Also known as: ADCC-GARCH, Asymmetric DCC-GARCH, NL-DCC-GARCH, Nonlinear Asymmetric DCC
The Nonlinear DCC-GARCH model extends Engle's (2002) Dynamic Conditional Correlation framework by allowing correlations to respond asymmetrically to negative versus positive return shocks. Proposed by Cappiello, Engle, and Sheppard (2006), it is the standard tool for measuring time-varying co-movement and contagion effects in multivariate financial time series when bad news is expected to increase correlations more than good news.
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 Nonlinear DCC-GARCH when you have two or more financial return series and believe that co-movement responds asymmetrically to bad news — common in equity–equity or equity–bond pairs during stress periods. It is appropriate for portfolio risk measurement, Value-at-Risk estimation, hedge ratio computation, and contagion analysis. Do not use it when series are not financial returns or lack GARCH-type volatility clustering; for a single series, a univariate asymmetric GARCH (GJR or EGARCH) is simpler and sufficient. If cross-series volatility spillovers (not just correlations) are the primary interest, BEKK-GARCH is more suitable.
Strengths & limitations
- Explicitly captures asymmetric correlation responses: negative joint shocks raise correlations more than equivalent positive shocks, matching empirical stylized facts.
- Preserves the computational tractability of the two-step DCC estimator, making it feasible for moderate dimensions (typically up to 10–30 series).
- Delivers a valid positive-definite covariance matrix at every time step, which is essential for portfolio optimization.
- Provides time-varying correlation estimates useful for dynamic hedging, asset allocation, and systemic risk monitoring.
- Well-grounded in the financial econometrics literature with a widely cited seminal reference (Cappiello et al., 2006).
- The two-step quasi-maximum-likelihood estimator is consistent but not fully efficient; standard errors of second-step parameters ignore first-step uncertainty unless corrected.
- The model does not capture volatility spillovers across series — only correlation dynamics; for spillover analysis, BEKK-GARCH or variance decomposition approaches are needed.
- Curse of dimensionality: estimation becomes numerically fragile as the number of series grows beyond roughly 20–30, because the parameter space expands and the likelihood surface flattens.
- Assumes that asymmetric effects enter only through joint negative innovations, which may not capture all forms of nonlinearity in correlations.
Frequently asked
What is the difference between DCC-GARCH and Nonlinear DCC-GARCH?
Standard DCC (Engle, 2002) uses only two scalar parameters (a and b) to drive correlation dynamics symmetrically. Nonlinear DCC adds a third matrix parameter G that activates only when both series have negative innovations, allowing correlations to react more strongly to joint bad news than to joint good news.
How do I test whether the asymmetric extension is needed?
Estimate both the standard DCC and the Nonlinear DCC on your data and compare them with a likelihood ratio test, or test the joint significance of the G matrix parameters using Wald tests. If G is not statistically significant, the added complexity of the nonlinear model is not warranted.
Can I apply this model to more than two series?
Yes. The model is inherently multivariate. However, numerical estimation becomes increasingly difficult beyond roughly 20–30 series because the parameter matrices grow and the likelihood surface can become flat. For very high-dimensional applications, dimensionality-reduction techniques or factor-based DCC variants are more practical.
What software can estimate Nonlinear DCC-GARCH?
The rmgarch package in R (by Ghalanos) provides a full implementation of the ADCC model. Stata's mgarch suite and MATLAB's Econometrics Toolbox also support DCC variants. Python users can implement it via the arch library combined with custom second-stage optimization.
Does the model guarantee a positive-definite covariance matrix?
Yes, provided the correlation-targeting constraints are satisfied and the scalar or matrix parameters obey the required regularity conditions (essentially that A, B, G are chosen so that Q_t stays positive definite). Practitioners should verify this numerically at each time step during estimation.
Sources
- Cappiello, L., Engle, R. F., & Sheppard, K. (2006). Asymmetric dynamics in the correlations of global equity and bond returns. Journal of Financial Econometrics, 4(4), 537–572. DOI: 10.1093/jjfinec/nbl005 ↗
- Engle, R. F. (2002). Dynamic conditional correlation: A simple class of multivariate generalized autoregressive conditional heteroskedasticity models. Journal of Business & Economic Statistics, 20(3), 339–350. DOI: 10.1198/073500102288618487 ↗
How to cite this page
ScholarGate. (2026, June 3). Nonlinear Dynamic Conditional Correlation GARCH Model. ScholarGate. https://scholargate.app/en/econometrics/nonlinear-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.
- DCC-GARCH modelEconometrics↔ compare
- EGARCH modelEconometrics↔ compare