JOB MARKET PAPER
A Theory of Choice Bracketing under Risk (Draft coming soon)
Extended abstract appears in EC'21
Abstract: Aggregating risks from multiple sources can be complex and demanding, and decision makers often adopt heuristics to simplify this evaluation process. This paper studies two related heuristics: narrow bracketing and correlation neglect. We axiomatize the two heuristics as behavioral deviations from the rational benchmark by relaxing the independence axiom of the expected utility model. Our axiomatization reveals alternatives to the model of narrow bracketing often used to explain evidence. These alternatives (i) allow for narrow bracketing or correlation neglect or both and (ii) allow for different levels of irrationality. We interpret the different sources as different streams of income to explain experimental evidence on violations of first order stochastic dominance. We propose a rationality ranking of decision makers based on their tendency to violate first order stochastic dominance and rank our models accordingly. Then, we interpret one source as background risk and show that narrow bracketing can explain risk aversion over small gambles. Finally, we interpret the different sources as consumption in different periods to explore the effect of narrow bracketing and correlation neglect on the optimal saving behavior. We show that an Epstein-Zin type utility function can emerge because of narrow bracketing. This new utility function achieves separation between time and risk preferences without inducing an unreasonably high timing premium as observed by Epstein, Farhi, and Strzalecki (2014).
Abstract: Aggregating risks from multiple sources can be complex and demanding, and decision makers often adopt heuristics to simplify this evaluation process. This paper studies two related heuristics: narrow bracketing and correlation neglect. We axiomatize the two heuristics as behavioral deviations from the rational benchmark by relaxing the independence axiom of the expected utility model. Our axiomatization reveals alternatives to the model of narrow bracketing often used to explain evidence. These alternatives (i) allow for narrow bracketing or correlation neglect or both and (ii) allow for different levels of irrationality. We interpret the different sources as different streams of income to explain experimental evidence on violations of first order stochastic dominance. We propose a rationality ranking of decision makers based on their tendency to violate first order stochastic dominance and rank our models accordingly. Then, we interpret one source as background risk and show that narrow bracketing can explain risk aversion over small gambles. Finally, we interpret the different sources as consumption in different periods to explore the effect of narrow bracketing and correlation neglect on the optimal saving behavior. We show that an Epstein-Zin type utility function can emerge because of narrow bracketing. This new utility function achieves separation between time and risk preferences without inducing an unreasonably high timing premium as observed by Epstein, Farhi, and Strzalecki (2014).
PUBLICATIONS
Abstract: This paper studies the implementation problem of a mechanism designer with ambiguity averse agents. The mechanism designer, desiring to implement a choice correspondence, can create ambiguity for agents by committing to multiple allocation rules and transfer schemes without revealing which one to use. By extending the cyclical monotonicity condition from choice functions to choice correspondences, we show that the condition can fully characterize implementable choice correspondences. We then study the implementability of choice correspondences in supermodular environments. As an application, we consider a mechanism designer who wants to allocate one object to one of her most desired agents and show that she can strictly benefit from concealing the tie-breaking rules. An intuitive and computationally tractable condition is provided to characterize when the mechanism designer's preference induces an implementable choice correspondence.
A Robust Reference-dependent Model for Rational Speculative Bubbles (with Jie Zheng) Journal of Economic Behavior & Organization, 137: 232-258, 2017.
Abstract: We present a robust model of speculative bubbles by introducing loss-averse reference-dependent preferences by Kőszegi and Rabin (2006) into the framework of Allen et al. (1993), where in equilibrium, asymmetrically-informed rational investors buy overvalued assets, hoping to sell them to less informed agents before the crash occurs. With reference-dependent preferences, the asset price may not necessarily be observable to agents when there is no trade. However, this is never the case with classical preferences, as shown in the paper. Incorporating the classical model as a special case, we generalize the notion of bubbles to allow for the analysis in the case of a silent market with unobservable prices, and our model is able to generate strong bubbles robust to moderate perturbations in parameters without the need for stronger conditions as suggested in previous literature. Assuming for simplicity that dividends can only take on two values, we construct an example of a robust reference-dependent bubble which is not robust in the classical setting, and we also show that the positive results regarding the limit of the bubble size and bubble frequency in the classical setting are preserved in our framework. Our main results and economic implications remain valid in more general settings.
WORKING PAPERS
(Previous title: Optimistic Dynamic Inconsistency)
Abstract: We model a decision maker who anticipates her preference to change in the future and optimistically evaluates each menu according to the best choice that could possibly be made by her future self. We characterize this menu preference, discuss the uniqueness of our representation and propose a comparative measure of optimism. We apply our model to three behavioral biases: naivete about present bias, projection bias and the disjunction effect.
Abstract: We model a decision maker who anticipates her preference to change in the future and optimistically evaluates each menu according to the best choice that could possibly be made by her future self. We characterize this menu preference, discuss the uniqueness of our representation and propose a comparative measure of optimism. We apply our model to three behavioral biases: naivete about present bias, projection bias and the disjunction effect.
Market Design and Walrasian Equilibrium (with Faruk Gul and Wolfgang Pesendorfer) (Update coming soon)
Abstract: We establish the existence of Walrasian equilibrium for economies with many discrete goods and possibly one divisible good. Our goal is not only to study Walrasian equilibria in new settings but also to facilitate the use of market mechanisms in resource allocation problems such as school choice or course selection. We consider all economies with quasilinear gross substitutes preferences but allow agents to have limited quantities of the divisible good (limited transfers economies). We also consider economies without a divisible good (nontransferable utility economies). We show the existence and efficiency of Walrasian equilibrium in limited transfers economies and the existence and efficiency of strong (Walrasian) equilibrium in nontransferable utility economies. Finally, we show that various constraints on minimum and maximum levels of consumption and aggregate constraints of the kind that are relevant for school choice/course selection problems can be accommodated by either incorporating these constraints into individual preferences or by incorporating a suitable production technology into nontransferable utility economies.
Information Aggregation under Ambiguity (with Pëllumb Reshidi and João Thereze) (Update coming soon)Ag
Abstract: We study information aggregation when an observer is ambiguous about the precisions of her information sources. The observer estimates a payoff-relevant state by minimizing quadratic loss according to MaxMin Expected Utility, and updates her beliefs prior by prior, which induces ambiguity regarding the state. We show that this ambiguity does not vanish even if the number of information sources grows indefinitely, and characterize the asymptotic set of posteriors the observer entertains. When information sources are unbiased signals, the observer learns the state correctly. In contrast, when the observer has access only to other agent's guesses, her estimate converges away from the truth with probability one.
WORK in progress
Preference for Simplicity (with Rui Tang)
Abstract: Evaluating and comparing the difficulty of problems are playing an important role in decision making. In this paper, we define and axiomatize the preference for simplicity as a subjective measure for simplicity of problems and introduce the concept of difficulty levels as endogenous outcomes. A prior and an induced preference over finite lotteries of natural numbers characterize the the preference for simplicity. The decision maker transforms each problem into a prospect of difficulty levels and then rank prospects according to the induced preference over lotteries. We then characterize the the preference for simplicity with more structures like the mixture continuous utility representation and the bounded expected utility representation. We also define the comparison of accuracy and variation aversion for the preference for simplicity and relate our models to existing models on ambiguity.
(Old version in 2018)
(Old version in 2018)
Abstract: Besides forgetting (limited memory), a decision maker might have biased memory. For instance, her memory might be cursed in the sense of Eyster and Rabin (2005) and undervalues the informational content of her previous actions. She might also suffer from the choice-supportive bias and tends to overly justify her previous choices. We study the effects of such memory biases on optimal learning in a parsimonious Wald problem and Bayesian persuasion. We show that compared with limited memory, biased memory can lead to more risk-taking and overconfidence. We also show that, while the concavification result holds if the receiver has limited memory, it fails if receiver has biased memory. In this case, the sender can strictly benefit from dynamic persuasion.