Topics in Choice Theory
Paper Session
Tuesday, Jan. 5, 2021 3:45 PM - 5:45 PM (EST)
- Chair: Kemal Yildiz, Bilkent University
Behavioral Neural Networks
Abstract
We provide an axiomatic foundation for a class of neural-network models applied to decision-making under risk, called neural-network expected utility (NEU) models. Motivated by classic experimental findings, we weaken the independence axiom in a novel way. We show how to use simple neurons, referred to as behavioral neurons, in NEU models to capture behavioral effects, such as the certainty effect and reference dependence. Empirically, we show that some simple NEU model with natural interpretation predicts better than existing theories, such as expected utility theory and cumulative prospect theory out of sample, and that behavioral neurons help improve NEU models’ performance.Stochastic Choice and Social Preferences: Inequity Aversion versus Shame Aversion
Abstract
In social contexts, we can often take deliberately stochastic choice behavior. However, we cannot easily judge the motivations behind observed altruistic/prosocial behavior. By focusing on inequity aversion and shame aversion as a social image concern, we characterize the two additively perturbed utility models, i.e., the sum of (non-)expected utility and a non-linear cost function. We examine how to distinguish stochastic inequity-averse behavior from stochastic shame-averse behavior, and vice versa. Moreover, we show that additively perturbed inequity-averse utility captures the general class of inequity-averse preferences, including both ex-ante and ex-post fairness. Finally, we consider the relationship between our models and random utility.Limit Points of Endogenous Misspecified Learning
Abstract
We study how a misspecified agent learns from endogenous data when their prior belief does not impose restrictions on the distribution of outcomes, but can assign probability 0 to a neighborhood of the true model. We characterize which actions are stable, in the sense that they have a very high probability of being the long-run outcome for some initial beliefs, and which are positively attracting, meaning that they have positive probability of being the long-run outcome for any initial full support belief. Our characterizations are based on two variants of Berk-Nash equilibria: A Berk-Nash equilibrium is uniformly strict if the equilibrium action is a strict best response to all the outcome distributions that minimize the Kullback-Leibler divergence from the truth, and uniform if the action is a best response to all those distributions. Uniformly strict Berk-Nash equilibria are stable, and only uniform Berk-Nash equilibria are stable. All the uniformly strict Berk-Nash equilibria are positively attracting under causation neglect, where the agent believes that their action does not influence the outcome, and under correlation neglect, where the agent believes that the outcome distribution associated with one action does not convey information about the outcomes associated with other actions. In supermodular decision problems, extremal actions are positively attracting if they are uniformly strict Berk-Nash equilibria. We generalize some results to settings where the agent observes a signal before choosing their action.Every Choice Function Is Pro-con Rationalizable
Abstract
We present a new choice model. An agent is endowed with two sets of orderings: pro-orderings and con-orderings. For each choice set, if an alternative is the top-ranked by a pro-ordering (conordering), then this is a pro (con) for choosing that alternative. The alternative with more pros than cons is chosen from each choice set. Each ordering may have a weight reflecting its salience. In this case, each alternative is chosen with a probability proportional to the total weight of its pros and cons. Weshowthateverynuanceoftherichhumanchoicebehaviorcanbecapturedviathisstructured model.OurtechniquerequiresanextensionofFord-FulkersonTheorem,whichmaybeofindependent interest. As an application of our results, we show that every choice rule is plurality-rationalizable.JEL Classifications
- D8 - Information, Knowledge, and Uncertainty
- D9 - Micro-Based Behavioral Economics