Using Option-Implied Correlations to Check Correlation Stability

Summary

The document questions whether slowly changing estimated correlation matrices reflect genuinely stable markets or the smoothing effect of using a long rolling window. It highlights a potential implication for statistical arbitrage: strategies that rely on correlations or principal components may be using estimates that appear persistent partly because each new observation has little influence on the calculation.

The reply suggests comparing historical estimates with option-implied correlations, which it describes as depending on current-day option prices rather than a long return history. It asserts that these measures are stable, offering a possible cross-check on whether correlation stability is only an estimation artifact. No data, calculation details, or definition of the implied-correlation measure are included, so the claim is suggestive rather than evidence that realized correlations or trading relationships remain stable. The document does not explain how to turn the comparison into a profitable strategy.

Key ideas

• Long rolling windows can make estimated correlation matrices change slowly even if underlying relationships vary more quickly.
• Correlation estimates and principal components may be used to inform statistical-arbitrage strategies.
• Option-implied correlations are proposed as a cross-check based on current option prices.
• The reply claims implied correlations are stable but gives no supporting data or procedure.

Tags

# market change, correlation and estimation bias


# market change, correlation and estimation bias












I hear many quants sating that markets change very slowly. This "fact" is even presented as a justification of statistical arbitrage, for example, by affirming that correlations remain roughly the same for long periods, and then insight given by these correlations or by a PCA applied on the correlation matrix is valid through time.

My question is : indeed, correlation matrix does not change on a daily basis, but isn't that due to an estimation bias? When the estimation is based on last 250 days for example, any new day contribution is very small and does not dramatically change the estimator, and real correlation may be much more stochastic than the stable matrix estimated.

If this is the case, how come this artificially stable correlation matrix can give profitable trading strategies relying on it?

## Answer by phdstudent (score 2)

https://quant.stackexchange.com/a/19025

One easy way to cross-check that is to compute option implied correlations. Those correlations are model free and only depend on the current day option prices and they are indeed stable.

For a nice article on computing option implied correlations check Vilkov's website he has several articles discussing option implied correlations. http://www.vilkov.net/www/content/research

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.