Nonlinear Problems: Mathematical Modeling, Analyzing, and Computing for FinanceView this Special Issue
The High Contact Principle with Reward Functions Involving Initial Points
This paper examines a class of general optimal stopping problems in which reward functions depend on initial points. Two points of view on the initial point are introduced: one is to view it as a constant, and the other is to view it as a constant process starting from the point. Based on the two different views, two versions of the generalized high contact principle are derived. Finally, we apply the generalized high contact principle to one example.
The “high contact principle,” first introduced by Samuelson  and McKean  and developed in greater depth by Øksendal  and Brekke and Øksendal , is a very useful tool to verify if a given function is a solution to the following optimal stopping problem: where is a bounded continuously differentiable function in , is an Ito process satisfying the following stochastic differential equation: and are Lipschitz continuous functions with at most linear growth, and is an -dimensional standard Brownian motion.
Generally, it is very hard to solve optimal stopping problem (1) directly. The usual way used in practice to finding a solution is to use the following two-step procedure. First, guess a solution based on some economic principles and common senses and second verify if it is indeed the solution using the high contact principle.
It is noted that the reward function in optimal stopping problem (1) is assumed to be independent of the initial point . That is, the existing high contact principle can only be applied to such a restricted case. However, there are many optimal stopping problems in economics and finance in which the reward functions do depend on the initial points.
More generally, a typical general optimal stopping problem has the following form:
The aim of this paper is to derive a generalized high contact principle which can be used to check if a given function is a solution to (3). We use two approaches for deriving two versions of the generalized high contact principle: one is more practical from applied perspective and the other is more intuitive from theoretical perspective.
This paper is organized as follows. In Section 2, we review the three basic questions of the classical optimal stopping problems (1). Unfortunately, the answers to the same three basic questions for the general optimal stopping problems (3) are negative. Two counterexamples are provided in Section 3. Our main results, the two versions of the generalized high contact principle, are presented in Section 4. The application of the generalized high contact principle to one example is examined in Section 5. In the final section, Section 6, we discuss some remaining questions related to the general optimal stopping problems.
2. The Reward Functions Not Involving Initial Points
Question 1. If the optimal value is known, can we find the continuation region ?
The answer is “yes.” In this case, the continuation region can be defined as follows: Thus, for any , where .
Question 2. If the continuation region is known, can we find the optimal value ?
Again the answer is “yes.” In this case, is the solution to the following Dirichlet problem: where is called regular if a.s. , is the characteristic operator of defined as follows: and and are given in (2).
Question 3. If both the optimal value and the continuation region are not known, can we find them and how?
In this case, the optimal stopping is a problem of free boundary. Therefore, an additional boundary condition is needed to identify the boundary. The usual approach to identifying the extra boundary condition is the “high contact principle.”
3. The Reward Functions Involving Initial Points
In this section, we consider a reward function which depends not only on but also on the initial point . That is, .
Similarly, we ask the same three questions as in Section 2.
Question 4. If the optimal value is known, can we define the continuation region the same as before? In other words, is
Unfortunately, the answer is “no.” Next is such a counterexample.
Example 1. Suppose that
where the reward function is
and denotes 1-dimensional Brownian motion starting at ; is a constant .
Obviously, . If we follow the notion used in the classical case, then the “continuation region” could be defined as follows: That is, However, Therefore, the classical notion of the continuation region cannot be used here. In fact, the continuation region for this example is and the optimal stopping time, accordingly, is
It is noted that this example shows that both the continuation region and the optimal stopping time may depend on initial point . Thus, there does not exist a universal continuation region for every initial point in general.
Question 5. If the continuation region is known, is the optimal value a solution to the following Dirichlet problem for We have
where is the characteristic operator of defined in (7).
Unfortunately, it is not true either. Next is such a counterexample.
Example 2. Suppose that
where the reward function is
and denotes 1-dimensional standard Brownian motion.
Because is a nonnegative martingale, then . The characteristic operator of process is and . Therefore, is not a solution of the above Dirichlet problem.
Question 6. If both the optimal value and the continuation region are not known, can the classical high contact principle be used to solve for ?
Clearly, the classical high contact principle stated in  cannot directly be used when the reward function depends on initial point due to the negative answers to the first two questions (Questions 4 and 5).
To solve the general optimal stopping problem (3), a new high contact principle is needed. We call it the generalized high contact principle which will be stated and proved in next section.
4. The Main Results
We will state and prove our main results in this section. To this end, we fix a domain (the “solvency” set) in . Let where the process satisfies (2).
Let denote the set of all stopping times . Consider the following problem: find and such that where is a bounded continuously differentiable function in . For simplicity we also assume that the diffusion process in (2) satisfies with .
Before stating the generalized high contact principle, let us give a heuristic argument for the optimal stopping problem. The argument is fairly intuitive, informal, and quite constructive.
Given , let us define a new function where denotes the expectation with respect to the law of given that . Obviously, . For function , first of all, we can see that . Next we consider a small number : Accordingly, where the first equality comes from the strong Markov property of the Ito diffusion process. That is, From the stochastic analysis Therefore, we have (approximately) Assuming that is smooth enough, by Taylor’s formula and the properties of Brownian motion process, Consequently after cancellation and letting , we get
In this case, on the one hand, it is similar to the variational inequality in the classical case; this suggests to us that this problem can be solved by the “high contact principle”; on the other hand, it is different from the classical case, because in the reward function is fixed.
Next is our key result which rigorously justifies the above idea.
Theorem 3. Let , be described as at the beginning of this section. For any given initial point , if there exist an open set with -boundary and a function on such that (i) , and (ii) (the second order condition) solves the following free boundary problem: then extending to all of by putting for , one has where ,
Proof. Using the generalized Dynkin formula in  for the diffusion process , we have
for all , where is the first exit time of .
According to conditions (i) and (ii) of , we get and combine it with Theorem 12.16 in ; then, is a superharmonic function. Owing to the condition that , so is a superharmonic majorant of . We know that in (23) is indeed the least superharmonic majorant of . So When , .
To the opposite inequality, we note by condition (ii) that where and . From the definition of the , then When , . Thus, we get (33) for all as well as defined in (34).
It is noted that Theorem 3 tells us how to solve the general optimal stopping problem. First, fix in the reward function and change the initial point of Ito diffusion process from to ; then, the optimal value function becomes a new function as (23) and get the form of in terms of the conditions (i) and (ii); second, find by letting in and the optimal stopping time defined in (34).
So far we have introduced our first point of view on the original initial point , that is, to view it as a parameter or a constant. Next, we will introduce our second point of view on it. That is, instead of viewing it as a constant, we will view it as a process starting from .
To this end, we change the initial point of the Ito diffusion process from to : and define a constant Ito diffusion process as follows: Put them together and we have a new Ito diffusion process in by where with .
Thus, is an Ito diffusion process starting at . Let denote the probability law of and let denote the expectation with respect to . In terms of , (22) can be rewritten as which is a special case of the problem where the optimal stopping time associated with is denoted by .
From the above, we have the second version of the generalized high contact principle.
Theorem 4. Let be described as the above, let be described as in (40), and let be an open set and denote . For any given initial point , if there exist an open set with -boundary and a function on such that (i) , , and (ii) (the second order condition) solves the following free boundary problem: then extending to all of by putting for and letting , one has where
Proof. Because is uniformly elliptic in and by conditions , and of Theorem 1 in , we get where Let , and we get the result.
We will end this section with a comparison of the two versions of the generalized high contact principle. In Theorem 4 we view in the reward function as a degenerate diffusion process and then the problem becomes the classical case with two diffusion processes. This method is much easier to understand theoretically, but it leads to double the number of the dimensions of the domain of the test function and accordingly the continuation region. Generally, the higher the dimension of the domain is, the more difficult it is to guess the solution. Thus, the first version of the generalized high contact principle, Theorem 3, is much easier to use than the second version of the generalized high contact principle, Theorem 4.
5. One Application
In this section, we apply the first version of generalized high contact principle to one example: the perpetual American put.
Example 5 (the perpetual American put). Suppose a stock value process is as follows:
where denotes 1-dimensional standard Brownian motion and , are constant.
The perpetual American put option is a contract signed at time 0 which entitles the buyer to sell one unit of the stock at any time at a price (the exercising price is a function of ). So his discounted reward is if exercising the option at time . To determine such an optimal striking time , we must solve the following optimal stopping problem: where .
It is noted that the reward function depends on the initial point . Thus, the classical high contact principle cannot be used to solve optimal stopping problem (53).
Because no point can be exercised if , First we fix in the reward function and change the initial value of the diffusion process to ; that is, From direct calculation, we get where . Let , and we get the solution where .
6. Some Remaining Problems
We will end the paper with some comments on some remaining questions.
It is well known that it is very hard to solve a typical partial differential equation in general. However, as we know, function is the solution of the following partial differential equation: That is, solving the classical optimal stopping problem (1) provides a method for solving a class of partial differential equations (59).
Moreover, (59) plays a key role in the “high contact principle” since it sets up the connection between optimal stopping problem and free boundary problem. Equation (59) provides a general method for guessing a solution to the classical optima stopping problem (1) since the solution to (1) must be a solution to (59).
Likewise, we also want to know what kind of partial differential equation should satisfy. Unfortunately, we have not succeeded in finding such a PDE yet. Similar to the classical case, there are also two reasons why we are interested in the PDE. One is that it provides a method for guessing a solution candidate to the general optimal stopping problem (3) since the solution to (3) must be a solution to the PDE if it exists. The other one is that solving the general optimal stopping (3) would provide a method for solving a class of such PDEs.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
The authors are indebted to the Canadian SSHRC for financial support and to Zengjing Chen and Huntley Schaller and especially Zhenting Hou, Shige Peng, and Jiezhong Zou. This work was partially supported by the National Natural Science Foundation of China (Grant nos. 71301173 and 71301160), China Postdoctoral Science Foundation funded project (Grant nos. 2012M520419, 2012M520420, and 2013T60186), Beijing Planning Office of Philosophy and Social Science (13JGB018), Collaborative Innovation Project of Economics School, and Program for Innovation Research in Central University of Finance and Economics.
P. A. Samuelson, “Rational theory of warrant pricing,” Industrial Management Review, vol. 6, pp. 13–32, 1965.View at: Google Scholar
H. P. McKean Jr., “Appendix: a free zboundary problem for the heat equation arising from a problem in mathematical economics,” Industrial Management Review, vol. 6, pp. 32–39, 1965.View at: Google Scholar
K. A. Brekke and B. Øksendal, “The high contact principle as a sufficiency condition for optimal stopping,” in Stochastic Models and Option Values (Loen, 1989), D. Lund and B. Øksendal, Eds., vol. 200 of Contrib. Econom. Anal., pp. 187–208, North-Holland, Amsterdam, The Netherlands, 1991.View at: Google Scholar | Zentralblatt MATH | MathSciNet
E. Dynkin, Markov Processes. Vol. II, Springer, New York, NY, USA, 1965.