No abstract

No abstract

Decision analysis stands on a foundation of hundreds of years of philosophical and practical thought about uncertainty and decision-making. The accomplishments and promise of the field are impressive, yet it has not become commonplace even in very important decisions. While human nature may pose an ultimate limitation, maintaining clarity of concept and exploiting progress in the realms of scope, skill, and efficiency should lead to more widespread use. A central conceptual distinction is that between normative and descriptive views of decision-making. We discuss the importance of maintaining this distinction in the face of attempts to compromise. The procedures for formulating, eliciting, evaluating, and appraising the decision problem are all experiencing major improvements. The strategy-generation table helps in finding creative alternatives. Decision quality concepts permit us to assure both effectiveness and efficiency in analyzing decision problems. The influence diagram provides new clarity to the conversation between decision-maker and analyst, allowing representations that are both easily understandable and mathematically consistent. The clarity test makes sure we know what we are talking about regardless of what we are saying about it. Direct and indirect values illuminate preferences. Generic risk attitude considerations indicate how to relate corporate risk tolerance to the financial measures of the corporation. Spreadsheet, decision tree, and influence diagram programs speed evaluation. Intelligent decision systems realized in computers offer promise of providing the benefits of decision analysis on a broader scale than ever before. Decision analysis is now poised for a breakthrough in its usefulness to human beings.

This paper considers the maximization of certain equivalent reward generated by a Markov decision process with constant risk sensitivity. First, value iteration is used to optimize possibly time-varying processes of finite duration. Then a policy iteration procedure is developed to find the stationary policy with highest certain equivalent gain for the infinite duration case. A simple example demonstrates both procedures.

No abstract