|
|
Principle Seminar of the Department of Probability Theory, Moscow State University
November 25, 2009 16:45, Moscow, MSU, auditorium 16-24
|
|
|
|
|
|
On the new approach to the optimal stopping problem
E. L. Presman |
|
Abstract:
We consider the optimal stopping problem of a Markov chain with an
infinite horizon and the payoff function g(z). Sonin I.M. for the case
of finit number m of states proposed an algorithm which allows one to
find the value function and the stopping set in no more than 2m steps.
This algorithm on each step considers a new chain with a new state
space, eliminating the states where we know we should not stop. We
discuss the case of an arbitrary state space, which requires a
modification of the algorithm. Consider, as Sonin did, set C_1 where the reward of an immediate
stopping is less than the reward of stopping at the first step. At
this set we know we should not stop. Let g_1(z) be the reward of
stopping at the moment of the first visit to the set complimentary for
C_1. Then under certain assumptions
g_1(z) is greater than g(z) for z from C_1 and is equal to g(z) on the
complimentary set. For the same chain consider the problem with a new
payoff function which is equal to g_1(z). Obviously this problem is
equivalent to the initial one. Let us now add to the set C_1 the
points where the reward of an immediate stopping is less than the
reward of stopping at the first step with payoff function g_1(z). We
get set C_2 and corresponding function g_2(z). Finally we get sequence
of functions g_n(z) and sets C_n which are nondecreasing and converge
correspondingly to the value function and the continuation set of the
initial problem. We show the efficiency of this procedure and possibilities for
generalizations in continuous time.
|
|