Why DCC Models Do Not Capture Cross-Asset Volatility Spillovers

Summary

The document addresses whether an asymmetric dynamic conditional correlation model estimated with the R package rmgarch can provide full ARCH and GARCH parameter matrices that allow volatility spillovers across assets. Its answer distinguishes the DCC model’s correlation dynamics from models that directly model cross-asset effects in conditional variances.

In the explanation, each asset’s conditional variance in a standard DCC setup depends on its own past variance and shocks, while the model captures changing dependence through a separate correlation process. This structure does not represent spillovers in which one asset’s past volatility directly affects another’s. The answer points to BEKK as a contrasting multivariate GARCH approach capable of modeling such interactions. The document gives no estimation example, package instructions, or empirical comparison, and its brief answer does not explore model variants or implementation details; it is a conceptual distinction rather than a guide to extracting estimates.

Key ideas

• DCC models the time-varying correlations among assets separately from their conditional variances.
• In the described DCC structure, each asset’s conditional variance depends on its own history.
• Cross-asset volatility spillovers require a model that allows conditional variances to interact.
• BEKK is identified as a model that can represent volatility spillovers.

Tags

# VAR-aDCC full ARCH and GARCH parameter matrices in R


# VAR-aDCC full ARCH and GARCH parameter matrices in R












I am working with the `rmgarch` package in R and I estimated a VAR-aDCC model. Is there any way to extract the extended version of estimates (allowing for volatility spillovers)? More specific, I want the full ARCH and GARCH parameter matrices from the `dccfit` function of the `garch` package. I saw that in the `ccgarch` package there is this option, but the `dcc` package does not support the aDCC model.

Is this option available in the `rmgarch` package?

## Answer by Richard Hardy (score 0)

https://quant.stackexchange.com/a/33958

I do not think you can model spillovers with the DCC model. Unlike BEKK, in DCC the conditional variance of each asset depends only on its own past, not on the past of other assets' conditional variances. Thus no spillovers.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.