paper

Note on the (non-)smoothness of discrete time value functions

arXiv:1911.05414

Abstract

We consider the discrete time stopping problem \[ V(t,x) = \sup_τE_{(t,x)}[g(τ, X_τ)],\] where is a random walk. It is well known that the value function is in general not smooth on the boundary of the continuation set . We show that under some conditions is not smooth in the interior of either. More precisely we show that is not differentiable in the component on a dense subset of . As an example we consider the Chow-Robbins game. We give evidence that as well is not smooth and that is not convex, even if is for every .

Note on the (non-)smoothness of discrete time value functions · wovepaper