A numerical Fourier cosine expansion method with higher order Taylor schemes for fully coupled FBSDEs

Balint Negyesi, Cornelis W. Oosterlee · 2025

Read A numerical Fourier cosine expansion method with higher order Taylor schemes for fully coupled FBSDEs by Balint Negyesi, Cornelis W. Oosterlee (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-10988.txt.

Introduction

Balint Negyesi and Cornelis Oosterlee present a faster, sharper way to solve a class of paired equations that show up in control and finance. Their 2025 paper treats forward-backward stochastic differential equations, where one part runs with time and the other runs back from a known end state. Most solvers take crude steps for the forward part, which caps their accuracy.

The pair use finer Taylor-based steps for the forward part, including a scheme that lifts the error rate to first order for the whole system. A cosine series method fills in the averages the backward part needs at each step. Tests on equations from simple to fully linked cases confirm the higher rates. The authors say this is the first scheme to reach that mark for the full paired system.

Catalog record

CREATORBalint Negyesi, Cornelis W. Oosterlee
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE86 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-10988

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A numerical Fourier cosine expansion method with higher order Taylor schemes for fully coupled FBSDEs Balint Negyesi, Cornelis W. Oosterlee Abstract A higher-order numerical method is presented for scalar valued, coupled forward-backward stochastic differential equations. Unlike most classical references, the forward component is not only discretized by an Euler-Maruyama approximation but also by higher-order Taylor schemes. This includes the famous Milstein scheme, providing an improved strong convergence rate of order 1; and the simplified order 2.0 weak Taylor scheme exhibiting weak convergence rate of order 2. In order to have a fully-implementable scheme in case of these higher-order Taylor approximations, which involve the derivatives of the decoupling fields, we use the COS method built on Fourier cosine expansions to approximate the conditional expectations arising from the numerical approximation of the backward component. Even though higher-order numerical approximations for the backward equation are deeply studied in the literature, to the best of our understanding, the present numerical scheme is the first which achieves strong convergence of order 1 for the whole coupled system, including the forward equation, which is often the main interest in applications such as stochastic control. Numerical experiments demonstrate the proclaimed higher-order convergence, both in case of strong and weak convergence rates, for various equations ranging from decoupled to the fully-coupled settings. 1 Introduction In this paper we are concerned with solving the following fully-coupled system of forward-backward stochastic differential equations (FBSDE) \displaystyle X_{t} \displaystyle=x_{0}+\int_{0}^{t}\mu(s,X_{s},Y_{s},Z_{s})\mathrm{d}s+\int_{0}^{t}\sigma(s,X_{s},Y_{s},Z_{s})\mathrm{d}W_{s}, (1a) \displaystyle Y_{t} \displaystyle=g(X_{T})+\int_{t}^{T}f(s,X_{s},Y_{s},Z_{s})\mathrm{d}s-\int_{t}^{T}Z_{s}\mathrm{d}W_{s}, (1b) where all coefficients are deterministic, scalar valued functions, T>0 is a finite time horizon, and \{W_{t}\}_{t\in[0,T]} is a standard Brownian motion over an appropriate probability space. Such equations have an innate connection with second-order, parabolic, quasi-linear PDEs with terminal boundaries of the following form \displaystyle\begin{split}\partial_{t}u(t,x)+\frac{1}{2}\widetilde{\sigma}^{2}(t,x,u,\partial_{x}u)\partial_{xx}^{2}u(t,x)+\mu(t,x,u,\partial_{x}u\sigma)\partial_{x}u(t,x)+f(t,x,u,\partial_{x}u\sigma)&=0,\\ u(T,x)&=g(x),\end{split} (2) where \widetilde{\sigma} is connected to \sigma and the coefficients \mu,f,g coincide with those in ( 1 ) – see e.g. [ 26 , 9 ] . In particular, due to non-linear generalizations of the Feynman-Kac formula, the solution of ( 2 ) is related to the solution triple of ( 1 ) by the following relations \displaystyle Y_{t}=u(t,X_{t}),\quad Z_{t}=\partial_{x}u(t,X_{t})\sigma(t,X_{t},u(t,X_{t}))\eqqcolon v(s,X_{s}), (3) at least in the case when the diffusion coefficient does not depend on Z [ 18 , 26 ] . Fully-coupled FBSDEs naturally arise in applications of stochastic control, [ 25 , 19 ] . Solving such equations analytically is seldom possible in an analytical fashion and one most often has to resort to numerical approximations instead. The literature in the so called decoupled framework, i.e. when the coefficients \mu,\sigma do not depend on Y and Z are thoroughly studied, see e.g. [ 2 ] . However, things get more complicated in the coupled setting, i.e. when the coefficients \mu,\sigma in ( 1a ) take Y,Z as arguments. Due to the nature of the coupling, the forward simulation of ( 1a ) is not straightforward, as one needs to have approximated the backward equation’s solution pair ( 1b ) in order to do so beforehand. Several approaches have been proposed in the past decades, starting with the famous four step scheme [ 14 ] , built on the connection with the associated quasi-linear PDEs ( 3 ). Without the sake of completeness, we mention [ 1 , 16 , 6 , 15 , 26 ] and the references therein. Classical techniques include Monte Carlo techniques, branching diffusions, quantization algorithms [ 7 ] , Fourier cosine expansions. In particular, starting with the decoupled setting Ruijter and Oosterlee [ 20 ] proposed an algorithm for the numerical resolution of decoupled FBSDEs, built on the COS method [ 8 ] , where the corresponding conditional expectations are approximated by means of Fourier cosine expansions enabled by the fact that, at least in the Markovian framework, the characteristic function of the transition between two time steps is known in closed form, given a suitable discretization of ( 1a ). In their original work, the forward diffusion was assumed to be a standard Arithmetic Brownian motion, giving pathwise analytical solutions for ( 1a ), eliminating the need for discrete time approximations. Later on, they extended this framework in [ 21 ] to more general, state and time dependent drift and diffusion coefficients, up to discretizations by second-order Taylor schemes, including the Euler-Maruyama and Milstein schemes. In case of the former, their algorithm exhibited a strong convergence rate of order 1/2 , whereas in case of the latter, this was improved to order 1 . Nevertheless, these extensions were given in the decoupled setting, i.e. Y,Z did not enter the drift and diffusion coefficients in ( 1a ). In order to remedy this, following their ideas, Huijskens et al. in [ 11 ] extended their algorithm to the coupled FBSDE framework of ( 1 ). Therein, the forward and backward equations are first discretized by order 1/2 Euler schemes, and are subsequently decoupled by suitable decoupling fields which approximate the deterministic mappings in ( 3 ). Thereafter, they proposed three different strategies for finding the optimal decoupling fields: the so called explicit method , in which throughout the backward recursion, the decoupling relations at time t_{n} are replaced by the already computed approximations at time step t_{n+1} ; a local method where the decoupling relations are obtained via Picard iterations at each time step; and a global method where Picard iterations take place simultaneously, at all points in time similar to [ 1 ] . They found that the explicit method, inspired by [ 7 ] , is the most robust, giving strong convergence rate of order 1/2 , even in the case of fully-coupled systems, where the diffusion coefficient \sigma depends also on Z , as in ( 1a ). The purpose of the present paper is to combine [ 21 ] with [ 11 ] for the coupled system ( 1 ). Doing so, we extend the COS approximations of [ 11 ] to higher-order discretizations for fully-coupled FBSDE systems, such that strong convergence rates of order 1 and weak convergence rates of order 2 can be achieved. In particular, we generalize the discretization of the forward diffusion in ( 1a ) to second-order Taylor schemes , see e.g. [ 13 ] , including the Milstein - and simplified order 2.0 weak Taylor schemes. Subsequently, in order to preserve the higher-order convergence rates in the forward diffusion, we solve the corresponding backward recursions in the approximations of the backward equation by appropriate second-order schemes [ 29 , 5 ] , in order to obtain a suitable pair of decoupling fields ( 3 ). In this sense, our work is a generalization of [ 21 ] , extending the approaches proposed therein from the decoupled to the fully-coupled setting. The main challenge in doing so, is that higher-order corrections terms involve derivatives of the decoupling fields in ( 3 ). In order to remedy this, we capitalize on the fact that COS approximations are infinitely differentiable, and given an explicit scheme, these derivatives can be computed directly at each step in the backward recursion. This enables higher-order discretization schemes, up to strong convergence rates of order 1 with a Milstein -, and weak convergence rates of order 2 with a simplified order 2.0 weak Taylor

Read the complete text