Working Papers and Works in Progress
Delegation to a rationally inattentive agent (Job Market Paper) (Draft coming soon)
Abstract. This paper studies delegation to an agent who flexibly chooses what information to acquire. The principal and the agent share the same state-dependent payoff from actions, but only the agent bears the information cost. Restricting discretion changes both learning incentives and subsequent decisions. In a binary-state model with uniformly posterior-separable costs, we decompose the effect of restriction into a learning effect and a distortion effect. With a continuum of actions, full discretion is strictly suboptimal when it induces two interior posteriors. A local restriction produces a first-order learning gain and a higher-order distortion. When full discretion induces no learning, a local curvature comparison gives a sufficient condition for improvement. Under global curvature bounds, this condition is also necessary. In a quadratic-loss utility with a uniform prior and Shannon-entropy information costs, we characterize optimal delegation. An extreme binary menu is optimal at low and intermediate costs, and an optimal hollow menu expands toward full discretion as costs increase. The value of restriction peaks where full discretion stops inducing learning.
Information Representations and Choices in Costly Acquisition of Information with Suha Kim and Jason Tayawa. (Draft coming soon)
Abstract. Do people acquire the same information when asked in accuracy or sampling size terms? Rational inattention models often treat information choice as choosing an accuracy level, yet the same choice can also be represented as choosing a sampling size. Theory treats these representations as equivalent, so revealed information-cost structure should be stable across them. We design a laboratory environment to test whether it is. Subjects choose from the same feasible information set, either by accuracy level or by the corresponding sampling size. We vary priors, rewards, and representation of information and evaluate behavior using the two major classes of models—Prior-Invariant (PI) and Uniformly Posterior-Separable (UPS) costs. Furthermore, we let subjects select the information from their preferred representation. Both classes of models are empirically valid in either representation of information, with PI observed more often in the sampling representation and UPS in the accuracy representation. Behavior is consistent: 60% of subjects are best fit by the same model in either representation. Even when consistent, different representations can reduce overall information. Representation choice is also consistent: 75% of subjects choose a single representation more often, with a majority choosing the sampling-size representation, contrasting with the modeling approach of Rational Inattention. Theoretically equivalent representations need not be behaviorally equivalent. The results invite future study about transporting revealed information costs across elicitation environments.
Paying too much attention with Jason Tayawa and Adnan Mahmood
Abstract. Why do people gather more information than needed? We propose an inattentive agent with distorted beliefs and implementation noise. Such an agent may hold under-responsive beliefs—beliefs that move too little given the signal— and therefore undervalue information and acquire more of it. We show when their choice-frequency data can satisfy the canonical test of rational inattention. Even then, estimated attention-cost functions from choice frequencies can be misspecified, identifying some cost families only up to a bias. In a controlled laboratory setting, we find that participants respond monotonically to rewards and beliefs, as characterized. Implementation is noisy: participants choose the optimal action given beliefs 94% of the time. Beliefs are distorted: 65% are under-responsive. As the model predicts, those with more under-responsive beliefs acquire more information. Estimating costs from choices assigns less than a quarter of rationally inattentive participants to the correct cost family. Distortion and noise embedded in choice frequencies cost 25% of optimal profits. Choices alone obscure over-acquisition and mismeasure attention and its costs.
Abstract. We study a long-run persuasion problem where a long-lived Sender repeatedly interacts with a sequence of short-lived Receivers who may adopt a misspecified model for belief updating. The Sender commits to a stationary information structure, but suspicious Receivers compare it to an uninformative alternative and may switch based on the Bayes factor rule. We characterize when the one-shot Bayesian Persuasion-optimal (BP-optimal) structure remains optimal in the long run despite this switching risk. In particular, when Receivers cannot infer the state from the Sender’s preferred action, they never switch, and the BP-optimal structure maximizes the Sender’s lifetime utility. In contrast, when such inference is possible, full disclosure may outperform BP-optimal. Our findings highlight the strategic challenges of information design when the Receiver’s interpretation of signals evolves over time, offering new insights into persuasion mechanisms in repeated interactions.
When Fairness Bends Rationality: Incorporating Inequity-Aversion in Models of Regretful and Noisy Behavior with Doruk İriş and Alessandro Tavoni. Revise & Resubmit at the International Journal of Game Theory
Abstract. Impulse balance equilibrium (IBE) and quantal response equilibrium (QRE) are prominent examples of tractable equilibrium concepts, which relax the rationality assumption and empirically outperform the Nash equilibrium. This paper introduces equity considerations into IBE and QRE, and assesses the performance of the modified concepts relative to the established ones using experimental data from 12 completely mixed constant and non-constant sum 2x2 games. We rank all concepts in terms of fit to the data for different sets of games and stability over time. Results show that fairness does not uniformly improve behavioral equilibrium models; its explanatory value depends on whether behavior is generated through noisy responses or regret-based impulse balancing, and on whether the game fixes or varies total surplus.
Published and Forthcoming Papers
Subjective Beliefs in International Agreements. with Doruk İriş and Sungwoo Im. Singapore Economic Review, 2022. [Working paper version]