Abstract:
We propose simple axioms that characterize a generalization of backward induction in which at any node of a decision tree, the decision maker can look forward a fixed number of stages perfectly. Beyond that, the decision maker aggregates continuation values according to a function that captures reasoning under unpredictability. The model is uniquely identified from the decision maker's preference over decision trees. The model allows the decision maker to iteratively revise her future plan, as she moves forward in a decision tree. We analyze a comparative measure of unpredictability aversion, and discuss how a principal may exploit the agent's imperfect foresight.