As far as I can see on this website the stochastic volatilty models seem to be preferred to local volatility models, mainly due to the fact that stochastic volatility is 2D diffusive process whilst local volatility models are a 1D diffusive process.

Why do you see stochastic local volatility mixture models come up in academic papers? What are their advantages and disadvantages versus either a stochastic volatility model or a local volatility model.

Emphasis on exotic options modelling is also helpful.

  • 2
    $\begingroup$ stochastic local volatility mixture = realistic stochastic dynamics of the volatility (stoch vol component) + perfect fit to the current implied vol surface (local vol component) $\endgroup$ Mar 25, 2018 at 11:31
  • $\begingroup$ i wouldn't say that people on this site prefer stochastic volatility to local vol - perhaps a stochastic local vol model like this though. $\endgroup$
    – will
    Mar 25, 2018 at 13:19
  • $\begingroup$ when would you prefer local vol over stochastic vol then? $\endgroup$
    – Trajan
    Mar 25, 2018 at 15:32
  • $\begingroup$ Can anyone show how @AntoineConze's comment is true? I cant see this works mathematically $\endgroup$
    – Trajan
    Mar 25, 2018 at 15:51
  • 1
    $\begingroup$ Stochastic local volatility model means $dS_t/S_t = ... dt + \sigma_t L(S_t,t) dW_t$ with $\sigma_t$ your favorite diffusive process (Heston, ...) for the stochastic part of the volatility. Then the local part $L(S,t)$ is computed from "Dupire’s unified theory of volatility" $\sigma_{\text{local}}(S,t)^2=E[ (\sigma_t L(S_t,t))^2 | S_t=S]=E[ \sigma_t^2 | S_t=S] L(S_t,t)^2$. $E[ \sigma_t^2 | S_t=S]$ can be efficiently computed using for instance a 2D finite difference scheme for the Fokker Planck equation. $\endgroup$ Mar 25, 2018 at 17:42

1 Answer 1


Stochastic local volatility model means $dS_t/S_t=...dt+\sigma_t L(S_t,t)dW_t$ with $\sigma_t$ the stochastic part (modeled for instance as in the Heston model, or any other dynamics deemed appropriate) and $L(S_t,t)$ the local part.

The local part $L(S_t,t)$ is computed from "Dupire’s unified theory of volatility" which states that $$ σ_{\text{local}}(S,t)^2 =E[(σ_tL(S_t,t))^2|S_t=S] = E[σ_t^2|S_t=S]L(S,t)^2 $$ so that $$ \boxed{L(S,t)^2 = \frac{σ_{\text{local}}(S,t)^2}{E[σ_t^2|S_t=S]}} $$ $σ_{\text{local}}(S,t)$ is the local volatility computed from the implied volatility surface using the Dupire formula, and $E[σ_t^2|S_t=S]$ can be efficiently computed using for instance a 2D ADI finite difference scheme for the Fokker Planck equation (a.k.a. forward Kolmogorov equation) associated with the model.

With a a stochastic local volatility model you can therefore have a realistic stochastic dynamics for the instantaneous volatility while at the same time have a perfect fit of the current implied volatility surface, which means the model consistently prices vanilla options and, unlike pure local volatility models, does a decent job at pricing exotics that depend on the dynamics of future volatility.


Your Answer

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

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