Unbiased simulation of stochastic differential equations (Q1704137)

The authors consider the process \(X_t\) defined as a unique strong solution of the multidimensional SDE \[ dX_t=\mu(t,X_t)dt+\sigma(t,X_t)dW_t, \] with \(X_0=x_0\), \(W_t\) being a \(d\)-dimensional Brownian motion, \(\mu:[0,T]\times \mathbb{R}^d\to \mathbb{R}^d\) and \(\sigma:[0,T]\times \mathbb{R}^d\to M^d\) mean and diffusion coefficients. The main focus is on the Monte Carlo approximation of the expectation \(V_0=E[g(X_{t_1},\dots,X_{t_n})]\) for some function \(G:\mathbb{R}^{d\times n}\to \mathbb{R}\) and a discrete grid \(0<t_1<\dots<t_n=T\). The suggested estimator is unbiased. The main idea is to use the Euler scheme solution as a solution to a regime switching SDE with some well-chosen coefficients. In order to compensate for the change of the coefficients of the SDE, a weight function obtained by the automatic differentiation technique induced by the Bismut-Elworthy-Li formula from Malliavin calculus is introduced. Some numerical examples are reported.