0
$\begingroup$

I need to write down a code to simulate an SDE of the following type: $$ dX_t = - ( \eta_t X_t + \chi_t ) dt + \sigma dW_t, \ X_0=v_0 \quad t \in [0,T] $$ where $\eta_t$ is a deterministic function, let us say $\eta_t = t, t \in [0,T]$,$X_t \in \mathbb{R}^d$, $\sigma, v_0$ are two scalars, and $\chi_t$ is of the form $$ \chi_t^i = \int_t^T e^{\int_s^t \eta_u du}ds, \quad t \in [0,T], i =1 , \dots , d. $$ I also know that is possible to give a closed formula for the explicit solution which is $$ X_t = e^{- \int_0^t \eta_u du} v_0 - \int_0^t e^{-\int_s^t \eta_u du}\chi_s ds + \sigma \int_0^t e^{- \int_s^t\eta_ud_u} dW_s. $$ Just to be more specific with the physical meaning of $X_t$ it represents a velocity, hence I am also interested in numerically simulate $x_0 +\int_0^t X_s ds$ which is the position at time $t$.

Given all these not so clear premises, I don't know anything about simulating SDE and I was wondering if someone can help me in finding some reference to simulate such a problem with matlab.

$\endgroup$
1
$\begingroup$

To deep dive into numerical methods for SDEs, a good start will be the Euler-Mayurama method to simulate the solution of the SDE. The headlines of this method are the following: Denoting $X_t = \lbrace{X_t^j\rbrace}_{j\in\lbrace{1,...,d}\rbrace}$, we have \begin{align} X^j_{t_{i+1}} = X^j_{t_{i}} - (\eta_{t_{i}}X^j_{t_{i}}+\chi_{t_{i}})\Delta_{t_i} + \sigma(W^j_{t_{i+1}}-W^j_{t_{i}})\\ = X^j_{t_{i}} - (\eta_{t_{i}}X^j_{t_{i}}+\chi_{t_{i}})\Delta_{t_i} + \sigma\sqrt{t_{i+1}-t_i}\epsilon^j \end{align} Where $X_{t_0} = X_0$, $\Delta_{t_i} = t_{i+1}-t_{i}$ for $i = 0, ..., N$ and $\epsilon$ is a Gaussian vector with mean $0$ and co-variance $I_d$. It is an iterative method as the solution of the SDE is updated at each time step.

Another way will be to use the explicit solution as given in your post. The integral $\int_0^t\exp(-\int_s^t\eta_udu)\chi_sds$ can be computed using well-known numerical methods and the stochastic integral, i.e.$\left(\sigma\int_0^t\exp(-\int_s^t\eta_udu)dW_s\right)^j$, can be computed using the definition of the stochastic integral.

Once you have simulated $X_t$ using the methods described above, computing $x_0 +\int_0^t X_s ds$ could be done as follows: \begin{equation} x_0 +\int_0^t X_s ds = x_0 + \sum_{i=0}^N X_{t_i}\Delta_{t_i} \end{equation}

Note that the Euler-Mayurama method is well described in this document from Giles Pages (Chapter 7). You can also find many numerical methods applied in the probability field in the same doc.

$\endgroup$

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.