A generalized Haussmann's formula (Q4036135)

Let \(dy_ t=b(t,y)dt+\sigma(t,y)dw_ t\) be a stochastic differential equation of the functional type, \(f(.)\) be a smooth function on \(C[0,T]\). It is shown that under certain conditions of integrability the random variable \(\xi=Ef(y)\) admits the representation \(\xi=E\xi+\int_ 0^ T\varphi_ tdw_ t\) with \(\varphi_ t=E\left[\int_ t^ T\eta(y,ds)\Phi(s,t)\mid{\mathcal F}_ t\right]\sigma(t,y)\) where \(\eta(x,.)\) is the measure from the Riesz representation of \(f'(x)\) and \(\Phi\) is defined as the solution of the linearized SDE. In the present setting the latter can be written as follows: \[ \Phi(t,\tau)=I+\int_ \tau^ t\int_ \tau^ T\eta_ 0(s,y,d\theta)\Phi(\theta,\tau)ds+\int_ \tau^ t\int_ \tau^ T\eta(s,y,d\theta)\Phi(\theta,\tau)dw_ s \] where \(\eta_ 0(s,x,.)\) and \(\eta(s,x,.)\) are the measures from the Riesz representations of \(b_ x(t,x)\) and \(\sigma_ x(t,x)\), respectively. The proof involves some ideas from the Malliavin calculus.