Sequential Decision Making: From Decision Elicitation to Strategy Identification

Characterizing behavior in sequential problems is often complicated by the presence of multiple decision rules with overlapping predictions. To address this issue, we introduce a new experimental and econometric approach for identifying decision strategies in sequential contexts. This approach consists of eliciting conditional strategies (as opposed to direct choices) and measuring policy adherence via maximum-likelihood estimation (as opposed to counting coincidences). Applying this approach to several common types of sequential problems increases the proportion of uniquely identifiable subjects by up to a third relative to standard methods and yields the following findings. First, in search and stopping problems, decision makers respond less strongly to state and time of the dynamic problem than in problems that do not have a stopping structure. Second, decision rules are often biased toward being more accepting (less demanding) than the optimal policy would predict. Third, the format used to elicit decisions (menu-based choice versus numeric threshold entry) has a significant effect on policy adoption. In addition to identifying decision rules that better fit observed behavior in dynamic choice problems, these results have implications for firms serving customers who face sequential decisions. We use a revenue management example (optimal subscription service pricing) to show that failing to account for the relevant decision rules can reduce firm profits by up to 54%.