top of page


Social Learning with Heterogeneous Preferences

This paper analyzes a model of social learning allowing for heterogeneous preferences. Heterogeneity of preferences is associated with public amnesia of information: the impact of past signals on public beliefs can be erased over time. The paper characterizes a necessary and sufficient condition for asymptotic learning to exist under any non-trivial signal structure. Furthermore, it also contains a necessary and sufficient condition for asymptotic learning under any unbounded information structure. If such condition is not met, infinitely many players will be arbitrarily close to extracting zero value from observational learning. The failure of information aggregation is associated to an anti-herding phenomenon: agents' actions stop reflecting signals received by previous generations. The paper also shows that the geometric position of the prior has consequences for asymptotic learning analysis. 

DALL·E 2022-10-18 21.38.43 -  A Line of Farmers Gossiping to Each Other, in the style of t

To Bayes or Not To Bayes: Choosing How to Learn (draft soon)

This paper studies the adversarial interaction between a reinforcement learning algorithm and a human. It studies an infinitely repeated zero-sum game with one informed agent (à la Aumann Maschler 1968), in which the uninformed agent chooses a learning rule to use to update its beliefs. The equilibrium learning rule follows a quasi-concavification pattern similar to Lipnowski Ravid 2020. It also features characteristics of the behavioral trait known as reality denial. It is possible to quantify the value of learning rule flexibility. The paper discusses an application to algorithmic pricing.

DALL·E 2022-10-18 21.59.45 - a medieval painting of a robot and a human studying together,

A Puzzle on Informational Bubbles (draft soon)

This paper considers a model in which at every period, an agent chooses another agent to sample from, establishing an information network. There  is a trade-off between interpretability and novelty for the sampling decisions, but novelty plays a bigger role. This means that rational players should always prioritize to seek new, unfamiliar sources of information. The  paper discusses possible sources of information cliques stemming from breaking the standard rationality assumptions.

DALL·E 2022-10-18 21.41.09 - a network of communication, in the style of el greco.png

A Polynomial Algorithm of School Choice with Quotas - with Andrew Postlewaite and Rakesh Vohra (work in progress)

We consider a student-school allocation problem with quotas for different groups of students. We propose an algorithm that is ex-post Pareto efficient and solvable in polynomial time. 

DALL·E 2022-10-18 22.01.19 - a painting of a diverse set of students going into different

All the images were created using Dall-E 2.

bottom of page