In this paper Gatheral presents the following parametrization of the implied total variance $w(k,T) = \sigma_{BS}(k,T)^2T$ for each slice $k \mapsto w(k,T)$:

$$ w(k) = a + b\{\rho (k-m) + \sqrt{(k-m)^2 + \sigma^2} \}.$$

As far as I understand it, for each expiry $T$ one will have to calibrate a set of five parameters $\{a,b,m,\rho,\sigma\}$.

On the other hand I found the following article where in appendix A, a calibrated volatility surface is presented. But in their example there is an explicit dependence on $T$, so then they will have a "simple" specific expression for the whole volatility surface.

Another thing I've noticed when reading articles about various parametrizations that there seems to be some inconsistencies regarding implied total variance. Gatheral defines it as $\sigma_{BS}(k,T)^2T$ but I've seen in other articles people parametrizing on $\sigma_{BS}(k,T)^2$ or $\sigma_{BS}(k,T)$ instead.

Summarizing: My question is primarily if one has to calibrate SVI for each expiry slice or if it is possible to parametrizing the whole surface in a way such that the total number of parameter does not increase if more expiries are added.

  • $\begingroup$ The link to you "following article" is broke. It is better to write out explicitly the title and authors of any cited paper so that it is independent of any link. $\endgroup$
    – Hans
    Feb 20, 2020 at 2:08

2 Answers 2


Gatheral and Jacquier discuss this issue in section 4 of the paper. Instead of using the raw parameterization of the SVI, they use the natural parameterization of the total implied variance: $$ w(k) = \Delta + \frac{\omega}{2} \left\{ 1 + \zeta \rho (k - \mu) + \sqrt{(\zeta (k-\mu) + \rho)^2 + (1-\rho^2)} \right\} (\text{p. 61 of the published paper}) $$

In order to fit the entire surface of the total implied variance, they propose the following generalization. To ensure that the fit is free of arbitrage, they define the surface in terms of the log-moneyness and the at-the-money implied total variance $\theta_t := \sigma_{BS}^2(0,t)t$. The Surface SVI then has the form: $$ w(k,\theta_t) = \frac{\theta_t}{2} \left\{ 1 + \rho \phi(\theta_t) k + \sqrt{(\phi(\theta_t) k + \rho)^2 + (1-\rho^2)} \right\} (\text{p. 63 of the published paper}) $$ Where $\phi$ is a smooth function from $\mathbb{R}_{+}$ to $\mathbb{R}_{+}$ such that the limit $\lim_{t\rightarrow 0} \theta_t \phi(\theta_t)$ exists in $\mathbb{R}$. The parameters that you need to fit for the entire surface are therefore $\rho$ and whatever is needed to fit $\phi$. In practice, you need some interpolation to get $\theta_t$ because you almost never observe a log moneyness of exactly 0.

The function $\phi$ and the parameters have to satisfy certain restrictions for the parameterization to be free of arbitrage. The paper discusses these at length.

Heston and Jacquier propose two possible $\phi$ functions: $$\phi(\theta) = \frac{1}{\lambda \theta} \left( 1 - \frac{1-e^{-\lambda\theta}}{\lambda \theta} \right)$$ Which they call a Heston-like parameterization and the power law $$ \phi(\theta) = \eta \theta^{-\gamma}$$ A while back, I implemented the paper in MATLAB. In the end, I didn’t use the codes so they are not extensively tested. I have uploaded them to the file exchange. Maybe they are helpful to you: http://ch.mathworks.com/matlabcentral/fileexchange/49962-gatherals-and-jacquier-s-arbitrage-free-svi-volatility-surfaces

  • $\begingroup$ May I ask you a small question about your code? When you fit the whole surface (SSVI), why are you recalibrate the slice afterwards? according to ( arxiv.org/pdf/1204.0646.pdf ) you chould choose $\phi (\theta) = \frac{\eta}{\theta^\gamma(1+\theta)^{1-\gamma}}$, eq 4.5 on page 17. Given the constraint should result in a complete free of static arbitrage surface, or am I missing someting? $\endgroup$
    – math
    Aug 15, 2015 at 11:44
  • $\begingroup$ Thank you for the excellent answer! $\endgroup$
    – KaiSqDist
    Mar 12 at 16:57
  • $\begingroup$ Although this may be 10 years late, I thought I'd want to answer @math 's question, the slices are supposed to be recalibrated. The authors state it in the subsection "An example SVI calibration recipe", which explain that the point of the slice recalibration is not only to better fit the option prices but also to apply the penalties for interslice crossing (preventing CS arbitrage, take note that bf arb is already taken care of in the c_prime and v^tilda_prime). $\endgroup$
    – KaiSqDist
    Mar 14 at 22:28

For short maturity SPX option chain, the analytic form of the V-shape volatility smile has been fully worked out in my latest paper on SSRN. You can take a look.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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