A regression-based Monte Carlo method to solve backward stochastic differential equations (Q2572405)

A numerical method is presented for approximating the solution of the decoupled forward-backward system of stochastic differential equations (FBSDE) \[ S_t= S_0+ \int^t_0 b(s, S_s)\,ds+ \int^t_0\sigma(s, S_s)\,dW_s, \] \[ Y_t= \varphi({\mathbf S})+ \int^T_t f(s, S_s, Y_s,Z_s)\,ds- \int^T_t Z_s dW_s, \] where \({\mathbf S}\) is \(d\)-dimensional, \(Y\) is one-dimensional, \(W\) is a \(q\)-dimensional Brownian motion, and \(b\), \(f\), \(\varphi\) are deterministic. A thorough analysis follows which proves convergence of the method and establishes error estimates. The paper concludes with an examination of results of numerical experiments that are engendered from applications of FBSDEs in finance.