Skip to content
All library documents

Why Correlation Alone Cannot Rebuild a Basket Implied Volatility Smile

Article Quant Q&A · Author: Animesh Saxena
Read the original

Summary

The document asks whether component implied volatility surfaces and historical correlations can be combined with the standard two-asset variance relationship to construct a basket’s strike-dependent implied volatility. The answer explains that the relationship describes aggregate variance under restrictive assumptions, but a single correlation estimate does not specify the full joint distribution of component returns needed to price basket options across strikes.

A Gaussian copula is suggested as an approximation for combining the component distributions, with the caveat that the result depends on the dependence model and input assumptions. The method is not an exact reconstruction: historical correlation may differ from market-implied dependence, and non-Gaussian return behavior and skew matter. The document gives a conceptual limitation and a modeling direction, but no numerical example, calibration procedure, or evidence comparing approximation accuracy.

Key ideas

  • A correlation-based variance formula does not by itself determine a basket’s strike-specific implied volatility surface.
  • Component volatility surfaces and a correlation matrix do not uniquely define the joint return distribution.
  • A Gaussian copula can be used as an approximate way to combine component distributions.
  • The approximation is sensitive to the dependence assumptions and may miss non-Gaussian behavior.

Tags

OptionsMulti-assetVolatilityDerivatives pricing
Full text
# How to calculate implied volatility smile of basket using correlations?


# How to calculate implied volatility smile of basket using correlations?












For a basket, the realized volatility can be calculated using:

$$\sqrt{\sigma_1^2 + \sigma_2^2 + 2 \sigma_1 \sigma_2 \rho}$$

If I have the volatility surface of two underlyings S1,S2 (strike space).

And for each point I calculate the vols using above formula, how accurate is the approximation? I can extend this to multiple assets using simple cholesky transformation.

Correlation used is historical correlation, and not implied correlation.

## Answer by onlyvix.blogspot.com (score 3)

https://quant.stackexchange.com/a/16845

The formula works for total variance, not "strike specific" variance that you need to construct basket vol surface from components, because single historical correlation (or correlation matrix) just does not provide enough information to uniquely reconstruct expected distribution of basket returns (unless for a trivial case where all components are gaussian, which is not what you're asking). The best you can do is to approximate.

Given the information you provided ( components surface + historical correlation/correlation matrix ) I would suggest using gaussian copula. Copula is a way to "combine" distributions of components into distribution of basket. Check out "Pricing Basket Options With Skew" http://wilmott.com/pdfs/100826qu.pdf Section 4.1 for a detailed description of such algorithm.

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.