Rolling Correlations and Dynamic Copulas for Changing Co-Movement

Summary

The document asks how to measure co-movement when the relevant time frame is not fixed, mentioning concordance as a possible approach. The response offers two alternatives. A rolling-window correlation recalculates correlation across successive windows, allowing the analyst to observe how the relationship changes over time. The choice of window length affects what changes are visible, though the text gives no guidance on selecting it.

For a more model-based approach, the response suggests a dynamic conditional copula model. It describes this as structurally similar to GARCH models, with time-varying copula parameters, and says it can provide contemporaneous estimates of conditional correlations. The recommendation is brief and includes no worked example, comparison, or empirical evidence. It also notes that an available user-created implementation may need adaptation, so applying the method requires attention to the data, model specification, and implementation details.

Key ideas

• Rolling-window correlations can track how measured co-movement changes over time.
• The selected window length shapes the time variation that a rolling estimate reveals.
• Dynamic conditional copula models can estimate contemporaneous conditional dependence.
• The proposed copula approach is described as structurally related to GARCH modeling.
• The source gives no empirical comparison and cautions that implementation may require adaptation.

Tags

# Measuring co-movement at non-constant intervals


# Measuring co-movement at non-constant intervals












Correlation measures how much two series move together over a fixed interval. Are there any techniques that measure co-movement over a variable time frame? One technique I am aware of is concordance.

## Answer by John (score 3)

https://quant.stackexchange.com/a/3511

- One approach is to simply calculate the correlations over a rolling window. Changing the window can provide an indication as to how the correlations are changing over time.

- A more sophisticated approach is to estimate a DCC-Copula model. This type of model beras some similarity with Garch in its structure, except that it is for the paremeters of a copula model (I believe they were an adjustment to initial work done on multivariate DCC Garch models). This type of model could also provide you with contemporaneous estimates of conditional correlations. There is a user-created Matlab toolbox that can estimate these models (site: http://www.mathworks.com/matlabcentral/fileexchange/29303-dynamic-copula-toolbox-3-0). While it is user-friendly, you may need to adapt it to suit your needs...

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.