Controlled stochastic differential equations under Poisson uncertainty and with unbounded utility

The present paper is concerned with the optimal control of stochastic differential equations, where uncertainty stems from Poisson processes. Optimal behavior (e.g., optimal consumption) is usually determined by employing the Hamilton-Jacobi-Bellman equation. This requires strong assumptions on the model, such as a bounded utility function and bounded coefficients in the controlled differential equation. The present paper relaxes these assumptions. We show that one can still use the Hamilton-Jacobi-Bellman equation as a necessary criterion for optimality if the utility function and the coefficients are linearly bounded. We also derive sufficiency in a verification theorem without imposing any boundedness condition at all. It is finally shown that, under very mild assumptions, an optimal Markov control is optimal even within the class of general controls.