AGENCY MODELS OF COST VARIANCE INVESTIGATION DECISIONS: A NUMERICAL ANALYSIS

Generally standard cost systems and investigation policies are not tailored to the individual whose performance is being monitored. This suggests that some heuristic is being used to integrate standard cost systems and incentive compensation methods. Simulation and numerical methods are used to explore the economic effects of employing heuristics in determining the appropriate criterion for conducting an investigation of a cost variance and designing the associated incentive scheme within a particular principal-agent model. The heuristic evaluated is based on those utilized in quality control, and allows the principal to design an investigation policy and incentive scheme without explicitly considering the agent's degree of risk-aversion. The most important finding is that the level at which the control limit is set is not nearly as crucial as tailoring the incentive scheme to the particular agent being employed. Also, the opportunity losses suffered by the principal from the failure to tailor the incentive scheme are a significant portion of his expected utility when the optimal policy is used.;A generalization of previous results with respect to optimal investigation regions is explored through analytical methods. In previous models with imperfect conditional information about the agent's action, the optimal investigation policy calls for the investigation of either high or low outcomes (depending on the particular agent), but not both. In order to analytically characterize the region of the outcome triggering an investigation, the agent's utility function had been restricted to members of the hyperbolic absolute risk-aversion (HARA) class. This study relaxes this restriction on the agent's utility function, arriving at characterizations of optimal investigation regions for problems involving a non-HARA risk-averse agent and a risk-neutral principal. Under certain conditions the admission of non-HARA utility functions implies the optimality of two-tailed investigation regions, even when the underlying cost-generating process is not dynamic. This provides some justification for the use of two-tailed policies in practice.;In order to perform this part of the study, numerical solutions must be found for the principal-agent model. Since general numerical procedures for solving this model do not appear in the literature, they are developed and documented as part of this study.