Primary: Economic Theory, Decision Theory, Behavioral Economics
Secondary: Financial Economics
Random Choice as Behavioral Optimization
with Faruk Gul and Wolfgang Pesendorfer, Econometrica, Vol. 82, No. 5 (September, 2014), 1873–1912
We develop an extension of Luce’s random choice model to study violations of the weak axiom of revealed preference. We introduce the notion of a stochastic preference and show that it implies the Luce model. Then, to addresses well-known difficulties of the Luce model, we define the attribute rule and establish that the existence of a well-defined stochastic preference over attributes characterizes it. We prove that the set of attribute rules and random utility maximizers are essentially the same. Finally, we show that both the Luce and attribute rules have a unique consistent extension to dynamic problems.
Random Choice and Learning
Updated June 2016. Theoretical Economics, revise and resubmit.
The attraction effect and other decoy effects are often understood as anomalies and modeled as departures from rational inference. We show how these decoy effects may arise from limited sampling and Bayesian updating. Our new model, the Bayesian probit, is parameterized like the classic multinomial probit model: each choice alternative is associated with a Gaussian random variable. Unlike the multinomial probit and any random utility model, however, the Bayesian probit can jointly accommodate similarity effects, decoy effects, and phantom alternative effects. We provide a new definition of revealed similarity from observed choice behavior, and show that it is captured by signal correlation in the Bayesian probit. We show that signal averages capture revealed preference; and that signal precision captures the decision maker's familiarity with the options. This link of parameters to observable choice behavior facilitates measurement and provides a useful tool for discrete choice applications.
Subjective Ambiguity and Preference for Flexibility
Updated January 2016. AEJ: Microeconomics, revise and resubmit.
A preference over menus is monotonic when every menu is at least as good as any of its subsets. The main result is that any numerical representation for a monotonic preference can be written in minimax form. The representation suggests a decision maker who faces uncertainty about her own future tastes and who exhibits an extreme form of ambiguity aversion with respect to this subjective uncertainty. In a setting with a finite number of alternatives, this generalizes Kreps' (1979) seminal characterization of preference for flexibility, and clarifies the consequences of his last axiom, ordinal submodularity. While the remaining axioms are equivalent to the existence of a (weakly) increasing aggregator of second period maximal utilities, ordinal submodularity holds if and only if this aggregator can be taken to be strictly increasing. [Slides]
Random Evolving Lotteries and Intrinsic Preference for information
(with Faruk Gul and Wolfgang Pesendorfer) March 2016.
We introduce random evolving lotteries to study preference for non-instrumental information and history-dependent attitudes to risk-consumption. We provide representation theorems for separable and for non-separable risk-consumption preferences. We characterize information seeking behavior and its opposite, information aversion. We show how our rich set of choice objects allows nuanced attitudes to information, including a preference for savoring the prospect of positive surprises, or a distaste for waiting for news.
Preference Reversal or Limited Sampling? Maybe túngara frogs are rational after all. (NEW)
[Paper][Slides] November 2016
In this paper, we demonstrate that revealed preference analysis and standard welfare analysis can be applied to context-dependent choice data. We propose a new discrete choice estimation framework outside the scope of traditional random utility models, building on the individual learning model of Natenzon (2016). We identify stable (context-independent) preferences from the context-dependent choice data in Lea and Ryan (2015), and offer a new perspective in the debate about the rationality of context-dependent choice. We show that our model, the Bayesian probit, outperforms any random utility specification in goodness of fit and out of sample prediction. We conclude that our model presents a useful alternative to random utility —the current workhorse of discrete choice estimation— for applications where decision makers systematically exhibit context effects, including attraction, compromise, and phantom alternative effects.
[Non-technical summary] Lea and Ryan (Science, Reports, 28 August 2015, p. 964) interpret mate choice data collected from frogs in the laboratory as being incompatible with rational choice models currently used in sexual selection theory. A close look at their data supports the hypothesis that some options offered in the lab are easier to compare than others. If we take into account that some pairs of options are easier to compare, and that frogs operate under conditions of uncertainty, we can restore rationality to túngara frogs.
Work in Progress
Closed-form Choice Probabilities in the Multinomial Probit Model
Similarity and Substitutability in Random Choice [Slides]
Sour Grapes and Optimal Expectations [Slides]