Second order discretization of backward SDEs and simulation with the cubature method (Q2448692)

A second-order numerical scheme is developed for the approximation of the forward backward system of stochastic differential equations \[ \begin{aligned} X_t &= x_0+ \int^t_0 V_o(X_u)\,du+ \sum^d_{j=1} \int^t_0 V_j(X_u)\circ dB^j_u,\\ Y_t &= \phi(X_T)+ \int^T_t f(X_u, Y_u, Z_u)\,du- \sum^d_{j=1} \int^T_t Z^j_u dB^j_u,\end{aligned} \] where \(t\in [0,T]\) and \(B\) is a \(d\)-dimensional Brownian motion. The authors introduce a discretization method for the backward equation with asymptotic rate of convergence of order \(n^{-2}\), when \(n\) points in the discretization interval are used. It is then combined with their already established cubature algorithm and tree based branching algorithm to arrive at the numerical scheme. An example is presented that demonstrates a substantial reduction of error when compared to a first-order scheme.