Working Papers

• When Are Decisions Improvable? An Evaluation of Diagnostic Methods (with B. Douglas Bernheim, Kirby Nielsen and Charles D. Sprenger). Submitted. [Abstract]
  We evaluate three methods for identifying improvable choices: documenting specific misconceptions (the Characterization Assessment method), gauging confidence in choices (the Decision Confidence method), and showing that specific behavioral patterns in the domain of interest also emerge in a related domain where they are objectively suboptimal (the Pattern Matching method). In experiments involving risky choice, the three methods imply that different choices are improvable and have conflicting implications regarding legitimate risk preferences. We clarify the assumptions underlying each method and reevaluate the evidence on risk-taking in light of their limitations.
• Robust Estimation of Risk Preferences (with Shunto J. Kobayashi). [Abstract]
  Economic models have been successful at rationalizing specific empirical findings, but their ability to concurrently explain multiple behaviors remains limited. In this paper, we illustrate this issue by conducting an experiment with 500 participants that studies two classical behaviors inconsistent with Expected Utility: the common ratio effect and preferences for randomization. The lack of generalizability of leading economic models across these two behaviors calls for the development of new empirical strategies to make predictions. Motivated by this observation, we introduce a novel empirical approach that enables us to predict behavior under risk without leaning on specific decision models. We further demonstrate that this method offers more accurate out-of-sample predictions about behaviors under risk, both inside and outside laboratory settings, than leading economic models and machine learning algorithms.
• Deliberate Randomization under Risk (with Marco Loseto). [Abstract]
  We consider a decision-maker (DM) with convex preferences who faces a set of risky actions and can delegate his choice to a randomization device. Under convexity, the DM's preferences admit a conservative multi-utility representation: each utility generates an evaluation for each action, and actions are ranked according to the lowest evaluation. Building on this multi-utility representation, we characterize the set of optimal actions and propose an efficiency criterion to rank them. Next, we narrow our attention to deliberate randomization for a DM with two utilities. In this case, we show that the DM never needs to select more than two actions with positive probability and study when the desire to randomize reveals information about risk attitude. Finally, we apply our results to games where each player has two actions and two utility functions. We show that incentives to randomize extend to strategic settings and derive a new class of mixed Nash equilibria that we call ``strict" because players strictly prefer randomization. In general, convexity may lead to a multiplicity of mixed Nash equilibria. However, we show that when they exist, only strict equilibria are such that all the mixed actions are efficient.