Sensitivity analysis of the utility maximisation problem with respect to model perturbations (Q1999596): Difference between revisions

The authors consider the expected utility maximization problem and its response to small changes in the market price of risk in a continuous semimartingale setting. Assuming that the preferences of a rational economic agent are modelled by a general utility function, they obtain a second-order expansion of the value function and a first-order approximation of the terminal wealth. Besides the mathematical motivation, the generalization to utilities which are not powers provides financial insight in two directions: first, there are important situations where the base model can still be solved rather explicitly, even with a general utility function, but the approximate model cannot. Second, such analysis provides the general structure of the approximation, in the directions of both the parameter and the wealth, giving additional insight on why the constant relative risk aversion case is so particular and therefore more explicit. Mathematical input of the paper can be characterized as follows: the authors consider perturbations simultaneously in the direction of the state (initial wealth) and the parameter and thus increase the dimensionality of the value function. In order to understand the general structure of the approximation, they need to formulate auxiliary quadratic stochastic control problems and relate the second-order approximations of both primal and dual value functions to these problems. Finally, if the risk-tolerance wealth process exists, they use it as numéraire and change the measure accordingly, to identify solutions to the general quadratic optimization problems above in terms of a Kunita-Watanabe decomposition (of a certain martingale) generated by the perturbation process. The paper contains an abstract version of the theorems that can be potentially applied to other stochastic control problems which are convex in the state variable, but not convex with respect to the parameter.