Multilevel Monte Carlo for Smoothing via Transport Methods

In this article we consider recursive approximations of the smoothing distribution associated to partially observed stochastic differential equations (SDEs), which are observed discretely in time. Such models appear in a wide variety of applications including econometrics, finance and engineering. This problem is notoriously challenging, as the smoother is not available analytically and hence require numerical approximation. This usually consists by applying a time-discretization to the SDE, for instance the Euler method, and then applying a numerical (e.g. Monte Carlo) method to approximate the smoother. This has lead to a vast literature on methodology for solving such problems, perhaps the most popular of which is based upon the particle filter (PF) e.g. [9]. It is well-known that in the context of this problem, that when filtering, the cost to achieve a given mean squared error (MSE) for estimates, the particle filter can be improved upon. This in the sense that the computational effort can be reduced to achieve this target MSE, by using multilevel (ML) methods [12, 13, 18], via the multilevel particle filter (MLPF) [16, 20, 21]. For instance, under assumptions, for the filter, some diffusions and the specific scenario of Euler discretizations with non-constant diffusion coefficients, to obtain a MSE of $\mathcal{O}(\epsilon^2)$ for some $\epsilon>0$ the cost of the PF is $\mathcal{O}(\epsilon^{-3})$ and the MLPF is $\mathcal{O}(\epsilon^{-2}\log(\epsilon)^2)$. In this article we consider a new approach to replace the particle filter, using transport methods in [27]. The proposed method improves upon the MLPF in that one expects that under assumptions and for the filter in the same case mentioned above, to obtain a MSE of $\mathcal{O}(\epsilon^2)$ the cost is $\mathcal{O}(\epsilon^{-2})$. This is established theoretically in an "ideal" example and numerically in numerous examples.