I came across this presentation from volar.io. The authors show fitting examples for a flexible volatility smile parametrization in 5 to 8 parameters which is also able to fit the locally concave market implied volatility smiles around special events.

Does anybody know the details of their parametrization and can you provide a reference? In particular, is it a simple extension of their C3 parametrization where the Cn curve is given by

\begin{equation} \sigma^2(z) = \sigma_0^2 \left( 1 + \sum_{i = 1}^{n - 1} \frac{1}{n!} \xi_i z^n \right) \end{equation}


\begin{equation} z = \frac{\ln(K / F)}{\sigma_0 \sqrt{T}}. \end{equation}

I suppose this is not the case and there is more to it. Some reasons:

  • Their examples look very stable on the wings which I would not expect from higher order polynomials. While they do not show too much extrapolation, their C5 and C6 curves on slides 31 and 32 look fairly well behaved in the wings (where they loose some quality of fit though).

  • It might be possible that they define a lower and upper cutoff beyond which they use a different tail function (e.g. linear in variance) and impose smoothness in these points. However on slide 10, they explicitly write that they don't like "hacks" in the wings.

  • Another "goal" states on slide 10 is for no-arbitrage constraints to be easy to incorporate. In the above setup, absence of butterfly arbitrage at all strikes used for the calibration creates non-linear constraints for an otherwise nice linear problem.

  • 1
    $\begingroup$ Interesting question, I was not aware of such developments. I wonder if no (static) arbitrage conditions can easily be accounted for under such a parametrisation. $\endgroup$ – Quantuple Sep 28 '16 at 13:44
  • $\begingroup$ @Quantuple: That will exactly be my follow up question once we know how the parametrization looks like. It is usually not difficult to compute the density by just differentiating - either by hand (Mathematica) or using AD. The question is if we obtain sth. tractable that can be easily incorporated as a constraint in the fitting. Also - can it be ensured for all strikes or just a selected set? $\endgroup$ – LocalVolatility Sep 28 '16 at 13:53
  • $\begingroup$ Completely agree with you. $\endgroup$ – Quantuple Sep 28 '16 at 13:57
  • $\begingroup$ @LocalVolatility There seems more to it, they claim this can be done in real-time so Mathematica or AD is not a likely solution. $\endgroup$ – pyCthon Sep 30 '16 at 21:19
  • $\begingroup$ It's just a Taylor expansion, given its fitted using data in the wings, it's still going to be stable there. It's only extrapolation where it starts to become bad. $\endgroup$ – will Oct 29 '16 at 18:12

I am not a customer, and not familiar with the details of their parametrization, but on page 25 they imply that next 2 parameters (S5) are independent wing parameters.

Later on page 25 and 26 they imply that other parameters S6,S7,S8 are to introduce W-shaped wiggle for names like SPY, AAPL and GOOG.

  • 1
    $\begingroup$ I don't think that's what they're saying, more that the svi jump wings model has the two parameters to deal with the call and put wings, and that when they move up to 5 parameters they can do the same, but since the nature of the expansion is not the same, their parameters are not the independent. $\endgroup$ – will Oct 29 '16 at 18:18
  • $\begingroup$ I understand the nature of their S3 and the SVI5 parametrizations. What I am interested in is if the higher order parametrizations are extensions of those of if they just go back to polynomials with some linear wings attached. $\endgroup$ – LocalVolatility Oct 31 '16 at 0:18

The formula for S3 volatility curve is explicitly given in one of the presentations on Vola Dynamics website. In fact, it is functionally equivalent to SSVI curve. BTW, the company name and the website address have changed. See voladynamics.com


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.