Utility maximization in incomplete markets for unbounded processes (Q2488492)

The authors are interested in a classical problem in economic theory: expected utility maximization from terminal wealth in a continuous time stochastic security market. They examine a challenging situation where the semimartingale representing the price process can be possibly not locally bounded. The class of ``admissible'' trading strategies \(H^W\) depends here on a random variable \(W\) that controls the losses admitted in trading. It is required that \(W\) is suitable for the market model, i.e. the class of strategies is rich enough for trading purposes, and that \(W\) is compatible with the preferences so that the expected utility of terminal wealth is never equal to \(-\infty\). It is proved that for all loss variables \(W\) contained in a properly identified set, the optimal value on the class \(H^W\) is constant and it coincides with the optimal value of the maximization problem over some larger domain. The existence of a solution in this domain is proved and it is shown that this solution can be presented as a stochastic integral which is a uniformly integrable martingale under the minimax measure. The approach to solve the main problem is based on the dual terminology. The method naturally leads to the selection of a set of sigma martingale measures having finite generalized entropy as the domain of the dual optimisation problem, and of some polar cone as the domain in the primal optimisation problem. The economic relevance and interpretation of both sets are described.