I am currently working on a paper for quoting option as a market maker from Bastien Baldacci , Philippe Bergault & Olivier Guéant

Without dwelling on details on how to obtain the HJB equation for this problem, I would like to know if the scheme I wrote for solving it numerically is viable or did I miss something.

I need to solve the following equation :

$$\begin{aligned} 0 = & \; \partial_{t}v(t,v,\mathcal{V}) + a_{\mathbb{P}}(t,v)\partial_{v}v(t,v,\mathcal{V}) + \frac{1}{2}v\xi^{2}\partial^{2}_{vv}v(t,v,\mathcal{V}) \\ & + \mathcal{V}\frac{a_{\mathbb{P}}(t,v) - a_{\mathbb{Q}}(t,v)}{2\sqrt{v}} - \frac{\gamma \xi^{2}}{8} \mathcal{V}^{2} \\ & + \sum_{Nb = 1}^{Nb = N}\sum_{l = a,l=b} \int_{R^{+}} z \, \mathbb{1}_{| \mathcal{V} - \phi(l)z\mathcal{V}^{I}| \leq \tilde{V}} H^{Nb,l}(\frac{v(t,v,\mathcal{V}) - v(t,v,\mathcal{V} - \phi(j)z\mathcal{V}^{i})}{z}) \mu(z)^{Nb,l} \end{aligned}$$

where H corresponds to a hamiltonian, $H(p) = \sup_{\delta > \delta_{\infty}} \Lambda(\delta)(\delta - p)$, $\delta_{\infty}$ is a constant. and terminal condition $v(T,v,\mathcal{V}) = 0$ ($\xi$, $\gamma$ are also some constant parameters).

I want to calculate this value function over a grid $[0,T] \times [ 0.1,0.2] \times [ -\tilde{V},+\tilde{V}]$. I denoted the time step $dt$, variance step $dv$, Vega $dV$. I denote $v^{k}_{i,j} = v(t_{k},v_{i},V_{j})$ I approximate the partial derviatives as follow :

$\partial_{t}v(t_{k},v_{i},\mathcal{V}_{j}) = \frac{v^{k}_{i,j} -v^{k-1}_{i,j}}{dt}$

$\partial_{v}v(t_{k},v_{i},\mathcal{V}_{j}) = \frac{v^{k}_{i+1,j} -v^{k}_{i-1,j}}{2dv}$

$\partial^{2}_{v}v(t_{k},v_{i},\mathcal{V}_{j}) = \frac{v^{k}_{i-1,j} -2 v^{k}_{i,j} + v^{k}_{i+1,j} }{dv^{2}}$

Then the scheme calculates the previous time step in function of the current time step.

Namely $$\begin{aligned} v^{k-1}_{i,j} = & \, v^{k}_{i,j} + dt \cdot \Big( a_{\mathbb{P}}(t_{k},v_{i}) \frac{v^{k}_{i,j} -v^{k}_{i-1,j}}{dv} \\ & + \frac{1}{2}v_{i}\xi^{2}\frac{v^{k}_{i-1,j} -2 v^{k}_{i,j} + v^{k}_{i+1,j} }{dv^{2}} + \mathcal{V}_{j} \frac{a_{\mathbb{P}}(t_{k},v_{i})- a_{\mathbb{Q}}(t_{k},v_{i})}{2 \sqrt{v_{j}}} \\ & - \frac{\gamma \xi^{2}}{8} \mathcal{V}_{j}^{2} + \sum_{Nb = 1}^{Nb = N}\sum_{l = a,l=b} \int_{R} z \mathbb{1}_{} H^{Nb,l}(\frac{v^{k}_{i,j} - v^{k}_{i,j-\phi(l)} }{z}) \mu(z)^{Nb,l} \Big) \end{aligned}$$

I don't know much about numerically methods for HJB equations so I just derived some explicit Euler scheme, is this correct or there are conditions for convergence and all ?

  • 1
    $\begingroup$ There will exist grid constants $dt/{dx^2}$ where this scheme is unstable. For example, if you make the appropriate terms (like $\xi$ ) constant or zero, then this reduces to the diffusion PDE, for which Euler is unstable when the grid constant is larger than 1/2. $\endgroup$
    – Brian B
    Nov 17, 2021 at 17:56
  • $\begingroup$ Yes I also think there must be some CFL condition for this type of equation but I do not know more about it, guess I'll try to apply something similar. The implicit scheme does not appeal to me at all anyway ;) Thanks man ! $\endgroup$
    – Kupoc
    Nov 18, 2021 at 8:11


Your Answer

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