From: Nils Christian Framstad Subject: Re: A Brownian motion question Date: 30 Apr 1999 13:41:35 +0200 Newsgroups: sci.math.research Keywords: Stochastic differential equation fhchen@my-dejanews.com writes: > Hi, > > Does anyone know how to get an explicit expression for the following > value function V(x,t)? > > Consider the Brownian motion X(t) with drift mu and diffusion > coefficient sigma. That is, dX=(mu)dt+(sigma)dW, where W is a > standard Wiener process. X(0)=0. Given a "retirement" value R, find > V(X,t)=E_X [\int_0^\tau {exp(-rt) mu dt}+exp(-r(tau))R] where > tau=inf{t:X(t)=at+b}, b<0. > > Can anyone help me, or does anybody know of a good reference for > this type of problems? Thanks. First a general comment: Problems of this kind are treated in most textbooks on stochastic calculus and applications to boundary value problems and/or optimal control, for example Bernt Øksendal: Stochastic Differential Equations on Springer Universitext. (Disclaimer: I am not the author's cousin, but I am his student.) The tricky part is that most of these references use the Dynkin formula, which requires the stopping time to be integrable. Yours isn't - it is even infinite with positive probability. You may go back to Dynkin's original formula in his book "Markov Processes" (vol II page 132, theorem 5.1 if I remember correctly), which is a corollary to a similar theorem with "discounting" and possibly infinite stopping time, but this book is not easy to read. To solve your particular case: Presuming you have constant coefficients with nonzero sigma, we can get rid of some of them: X(t) = mu t + sigma W(t) and X(t) = at+b iff W(t)= ((a-mu)t + b) /sigma =: -pt + q and tau=inf{t; W(t) + pt = q} Furthermore, solving the integral we get V= mu + (R-mu)/r * E[exp(-r tau)] Then I apply to Borodin and Salminen: Handbook of Brownian Motion (Birkhäuser 1996), formula 2.0.1 p223: E[exp (-r tau)] = exp (pq - |q|*sqrt(2r+p^2) ) The rest is now trivial. nc -- mailto:ncf@math.uio.no http://www.nef.wh.uni-dortmund.de/~eike/poe.html