Malliavin calculus for backward stochastic differential equations and application to numerical solutions (Q657705): Difference between revisions

Given a Brownian motion \(W\) and a terminal condition \(\xi\), the authors construct numerical schemes for a general backward stochastic differential equation \[ dY_t=-f(t,Y_t,Z_t)+Z_tdW_t,\;t\in[0,T],\, Y_T=\xi, \] without assuming that the randomness of \(\xi\) or that of the driving coefficient come from the solution of a forward equation. Supposing that \(\xi\) and \(f\) are twice Malliavin differentiable with derivatives which satisfy some integrability conditions, the authors use the Malliavin calculus to show that, for some constant \(K\), \[ \text{E}[|Z_t-Z_s|^p]\leq K|t-s|^{p/2},\, s,t\in[0,T], \] which implies the first result on \(\gamma\)-Hölder continuity (for any \(\gamma<1/2 - 1/p\)) in this generality, without relating the BSDE to a forward one. These regularity properties for \(Z\) are then the basis for the study of different approximation schemes and their convergence rates (one similar to that proposed by \textit{J. Zhang} [``A numerical scheme for BSDEs'', Ann. Appl. Probab. 14, No. 1, 459--488 (2004; Zbl 1056.60067)] for a BSDE related to a forward one, but also another ``implicit'' scheme with a better result on the convergence rate). In the both studied schemes, the integral of \(Z\) is used in each iteration step. In order to avoid necessary discretisation for implementation on computers, a completely discrete numerical scheme is proposed and its convergence rate is determined.