1
$\begingroup$

I need to simulate a process of the form

$$X_t=\int_0^t f(s,t)\mathop{dW_s}$$

where $f$ is deterministic and the integral is an Itô integral. I know I can simply take finite Itô sums of discrete increments of the Brownian motion driver, but I am wondering if there are more sophisticated approaches. Common methods such as the Euler-Maruyama method do not appear to be applicable because the integrand depends on the upper terminal $t$ and so $X_t$ is not an Itô process.

Are there known approaches for simulating this kind of process?

(Answers to related questions that would help to find relevant literature would also be useful: eg. Does this kind of process have a name? Is there a way of writing it as a SDE? If so, does that class of SDEs have a name?)


EDIT

The particular integral I'm interested in is the Molchan-Golosov representation of fractional Brownian motion. Up to a multiplicative constant, this is $$\int_0^t (t-s)^{H-1/2}F(1/2-H,H-1/2,H+1/2,\frac{s-t}{s})\mathop{dW_s}$$ where $H\in(0,1)$ and $F$ is the Gauss hypergeometric function.

In further reading I have found that processes of this form are known as Volterra processes, but I haven't found any discussion of simulation algorithms.

$\endgroup$
4
  • $\begingroup$ Can you provide some examples of the integrand $f(s,t)$? What sort of regularity does $f$ have? The choice of integration method seems to depend on this. $\endgroup$ – Nawaf Bou-Rabee Apr 28 '19 at 17:39
  • $\begingroup$ @NawafBou-Rabee Done. $\endgroup$ – Luke Thorburn Apr 28 '19 at 23:02
  • $\begingroup$ The first hit when entering "simulation fractional brownian motion" into Google is a whole thesis devoted to the problem. Have you read this? What do you want to achieve which the methods presented there don't? $\endgroup$ – Martin Hairer Apr 29 '19 at 0:04
  • $\begingroup$ @MartinHairer Yes, I have read that thesis. I'm not interested in simulating fBm in general, but in approximating well this particular representation of it by a (possibly complicated/algorithmic) function of its standard Bm driver. $\endgroup$ – Luke Thorburn Apr 29 '19 at 0:14
1
$\begingroup$

Here is an approximation scheme that uses a chain of independent Brownian bridges. For $t>0$ fixed, consider the following partition of the time interval $[0,t]$ $$ t_0 = 0 < t_1 < t_2 < \dots < t_{n} = t \;. $$ At these discrete values, compute a discretized Brownian motion $W_i = W(t_i)$ in the standard way $$ W_{i} = W_{ i-1} + \sqrt{t_{i+1}-t_i} \xi_i \;, i=1, \dots, n \;, $$ where the $\xi_i$'s are independent standard normal random variables. Let $\mathcal{G}_n$ denote the $\sigma$-field generated by this discretized Brownian motion. Then a pathwise accurate approximation to $X_t$ that converges in the $L^2$ sense is given by $$ \tilde X_t = \mathbb{E} \left( \int_0^t f(s,t) dW_s \mid \mathcal{G}_n \right) $$ and since a Brownian motion pinned at the $t_i$'s are independent Brownian bridges with mean $$ W_i + \frac{s-t_i}{t_{i+1}-t_i} (W_{i+1} - W_i) $$ we obtain $$ \tilde X_t = \sum_{i=0}^{n-1} \frac{W_{i+1}-W_i}{t_{i+1}-t_i} \int_{t_i}^{t_{i+1}} f(s,t) ds \;. $$ This approximation is based on Proposition 3.1 of the following paper.

Decreusefond, Laurent; Üstünel, Ali Süleyman, Stochastic analysis of the fractional Brownian motion, Potential Anal. 10, No. 2, 177-214 (1999). ZBL0924.60034.

$\endgroup$

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.