Under $\mathcal{P}$, we have the Heston Model given by:

$$ d S_{t}=\mu S_{t} d t+\sqrt{\nu_{t}} S_{t} d W_{t}^{S},\\ d \nu_{t}=\kappa\left(\theta-\nu_{t}\right) d t+\xi \sqrt{\nu_{t}} d W_{t}^{\nu}. $$

Assume I estimated these parameters appering in the Heston model $\underline{\text{using only stock returns}}$ with MCMC.

My question is; how can I price vanilla options under a risk neutral measure $\mathcal{Q}$ now? Since these parameters are estimated under the stocks $\mathcal{P}$-dynamics, are these parameters useless for option pricing?

I know there exists fourier-transform pricing methods. But can I use fourier-transform pricing methods with these estimated parameters?


As you said, you estimated the $\mathcal{P}$ parameters but for option pricing, one needs the $\mathcal{Q}$ parameters. But there exists a transformation.

Under $\mathcal{P}$, Heston (1993) assumes \begin{align*} \mathrm{d}S_t &= \mu S_t\mathrm{d}t + \sqrt{v_t}S_t\mathrm{d}W_{1,t}^\mathcal{P}, \\ \mathrm{d}v_t &= \kappa(\theta-v_t)\mathrm{d}t + \xi\sqrt{v_t}\mathrm{d}W_{2,t}^\mathcal{P}, \end{align*} where $\mathbb{E}^\mathcal{P}[\mathrm{d}W_{1,t}^\mathcal{P}\mathrm{d}W_{2,t}^\mathcal{P}]=\rho\mathrm{d}t$.

These parameters include the market price of risk which equals zero in the risk-neutral world. Assume $\lambda(S_t,v_t,t)=\frac{\lambda}{\xi}\sqrt{v_t}$. Applying the two-dimensional Girsanov theorem, \begin{align*} \frac{\mathrm{d}\mathcal{Q}}{\mathrm{d}\mathcal{P}}\bigg|_{\mathcal{F}_t} =\exp\left(-\int_0^t\frac{\mu-r}{\sqrt{v_s}}\mathrm{d}W_{1,s}^\mathcal{P}-\int_0^t\frac{\lambda}{\xi}\sqrt{v_s}\mathrm{d}W_{2,s}^\mathcal{P}-\frac{1}{2}\int_0^t\frac{(\mu-r)^2}{v_s}+\frac{\lambda^2}{\xi^2}v_s\mathrm{d}s\right). \end{align*} This corresponds to \begin{align*} \mathrm{d}W_{1,t}^\mathcal{Q} &= \mathrm{d}W_{1,t}^\mathcal{P}+\frac{\mu-r}{\sqrt{v_t}}\mathrm{d}t \\ \mathrm{d}W_{2,t}^\mathcal{Q} &= \mathrm{d}W_{2,t}^\mathcal{P}+\frac{\lambda}{\xi}\sqrt{v_t}\mathrm{d}t \\ \end{align*} Now, similar to Black-Scholes, applying Ito's Lemma to $f(x)=\ln(x)$, we obtain under $\mathcal{Q}$ \begin{align*} \mathrm{d}\ln(S_t) &= \left(r-\frac{1}{2}v_t\right)\mathrm{d}t+\sqrt{v_t}\mathrm{d}W_{1,t}^\mathcal{Q}, \\ \mathrm{d}v_t &= \kappa^\mathcal{Q}(\theta^\mathcal{Q}-v_t)\mathrm{d}t + \xi\sqrt{v_t}\mathrm{d}W_{2,t}^\mathcal{Q}, \end{align*} where $\mathbb{E}^\mathcal{Q}[\mathrm{d}W_{1,t}^\mathcal{Q}\mathrm{d}W_{2,t}^\mathcal{Q}]=\rho\mathrm{d}t$ and $\kappa^\mathcal{Q}=\kappa+\lambda$ and $\theta^\mathcal{Q}=\frac{\kappa\theta}{\kappa+\lambda}$.

So, the good news is that you can convert the process (and its parameters) from the real-world into the risk-neutral world. It's also good that the vol-of-vol and correlation coefficient do not alter at all. The bad news is that the speed of mean-reversion and the long-term mean depend on the market price of volatility risk which requires further estimation.

| improve this answer | |
  • $\begingroup$ Further estimation how? Can you provide me some articles I can take a look at? $\endgroup$ – LocalMartingale Feb 23 at 19:21
  • $\begingroup$ Include the market price of risk in the process under $\mathcal{P}$ and use, for example, maximum likelihood estimation to find the different parameters. I ought to highlight that this however is non-standard. For good reasons, one almost always uses traded, liquid options to find parameters under $\mathcal{Q}$. $\endgroup$ – Kevin Feb 23 at 19:28
  • $\begingroup$ Yes I know. It is non-standard what I do. But my goal is to find the distribution of the parameters appearing in stochastic volatility models. Therefore I need to use sampling methods like MCMC. Point estimation doesn't give me any distribution. If you have any idea about how to handle this problem, I will appreciate it. $\endgroup$ – LocalMartingale Feb 23 at 19:33

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.