Evidence for Clustering in Asset Volatility and Correlations
Summary
This discussion asks whether correlations change more slowly than volatilities and whether clustering affects the two quantities differently. It cites the use of longer rolling windows for ex-ante correlation estimates than for volatility estimates in Betting Against Beta, then asks why correlation dynamics receive less attention despite the established study of volatility clustering.
The accepted answer offers a cautious perspective from portfolio management literature: forecasts of near-term returns may have little bearing on volatility and correlation estimates, and covariance structure may be more stable than investors assume. It acknowledges some evidence of regime changes in short-run currency volatility and correlations, while concluding that correlations do show some clustering. The answer does not provide a direct empirical test of the comparison, explain the cited estimation-window choice in detail, or establish that clustering affects correlations more than volatilities. Its contribution is therefore a qualified observation about possible stability and regime behavior rather than a quantitative finding or settled explanation.
Key ideas
- The question compares the persistence of correlation changes with volatility clustering.
- The cited strategy uses a longer estimation window for correlations than for volatilities.
- The answer argues that asset volatility and correlation structures may be more stable than investors expect.
- It recognizes some evidence of regime changes in short-run currency volatility and correlations.
- The response does not quantify whether clustering affects correlations more strongly than volatilities.
Tags
Full text
# How does clustering affect volatility and correlations differently? # How does clustering affect volatility and correlations differently? In Betting Against Beta - Frazzini & Pedersen (2014, JFE), the authors state that correlations appear to move more slowly than volatilities, which implies that the clustering phenomenon affects correlations much more than volatilities. It is also for this reason that the authors use a longer rolling window to estimate ex-ante correlations as compared to volatilities: Main Question: How true is this? Why has there not been extensive studies made on this? Since it is well-known that volatilities have clustering time series dynamics, why are correlations not as well studied. ## Answer by KaiSqDist (score 1, accepted) https://quant.stackexchange.com/a/82319 I would like to add a paragraph that sheds some light on this query (albeit slightly unrelated to mentioned Betting Against Beta paper). In Active Portfolio Management by Grinold & Kahn (2000), the authors specify in Chapter 10 (Forecasting Basics): > What can we conclude? The researcher who tries to forecast returns over the near horizon should ignore the slight impact of those forecasts on the volatility and correlation estimates for the assets. Asset allocators in particular should take note of this. Many asset allocators are seduced by the possibility of forecasting volatility and correlation along with returns. They believe that the market has changed and is obeying a new reality. The same force responsible for the exceptional returns is also changing the covariance structure. This is easier to imagine than to establish. There is some evidence of "regime changes" in short-run currency volatilities and correlations, however, in general there is more stability than instability in asset volatilities and correlations. Therefore to summarize, there is some evidence of correlations exhibiting "clustering" properties.
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.