A principal seeks to screen an agent based on his demonstrable knowledge of a subject matter, modeled as a binary state. The agent learns about the state through two kinds of opposing verifiable signals, each kind providing evidence in favor of one of the states. A good quality agent has an evidence structure which is more informative than a bad quality one. In a symmetric setting, we show that under the optimal test, regardless of whether the agent can predict the state correctly, he is failed if the amount of evidence he is able to show is below a threshold. Conditional on providing evidence above this threshold, the agent is passed based on a simple True-False test -- i.e., if and only if he gives the correct answer. We see this result as rationalizing a common test structure where test-takers are given credit for giving the correct answer only if they show a minimal amount of data, arguments, or steps, in support of their answer. We prove the results by identifying a connection to the optimal transport problem and leveraging it to show the existence of an appropriate virtual value function. 

A principal seeks to screen an agent based his knowledge of a subject matter, modeled as a state which can take finitely many values. In this note I show that in a symmetric setting, the standard multiple choice question - where the test-taker is awarded full credit if and only if he selects the correct answer - is optimal, if and only if the principal is sufficiently quality-sensitive. 

A privately informed sender with state-independent preferences communicates with an uninformed receiver about a single-dimensional state. The sender can verifiably reveal the state precisely. She can also communicate via cheap messages when not disclosing the state. I show that unravelling occurs, with or without cheap talk, if and only if sender's preferences satisfy a condition slightly weaker than strict quasiconvexity. Moreover, the ability to use cheap talk can prevent unravelling for some prior distribution of the state, if and only if the sender's payoff is "M-shaped". When unravelling does not occur, the model features multiple equilibria. I show that as long as the sender's payoff function does not pool "too many types" together, cheap talk typically expands the set of equilibria, regardless of other parameters. Varying across equilibria, I also show that equilibria that feature more disclosure are worse for the sender, with the disclosure minimizing equilibrium being sender-best. 

Presentations and seminars - European Winter Meeting of the Econometric Society (Dec 2021), Winter School 2021 by the Econometric Society and the Delhi School of Economics (Dec 2021), 16th Annual Conference at the Indian Statistical Institute, Delhi (Dec 2021), , 50th Annual Meeting of the Illinois Economics Association (Oct 2021), theory group seminars / brown bags at UChicago (Nov 20210), Colorado State University (Nov 2021)

Many transactions in the marketplace rely on hard (or verifiable) information about the underlying value of the intended exchange, typically through certification— housing, diamonds, bonds being cases in point. What is the class of Pareto efficient certifications for such scenarios? This paper studies the canonical monopolistic screening problem, and models certification as hard information produced through a test to be flexibly chosen pre-trade. It argues that Pareto efficient tests take a simple form—they produce certification with a partitional structure, often with one or two thresholds. This claim is shown to be true for both the linear trading model and the non-linear pricing model. 

In the one-sided matching problem, objects are allocated to agents based on agent preferences. However, agents may not always know, a priori, their cardinal or even ordinal preferences over the objects, because they do not have enough information. In this context, I try to answer the question: How should a benevolent planner optimally reveal information to the agents to maximize welfare, in an environment where agents have no private information to start with? As a benchmark, I first show that when using any of the standard strategyproof ordinal mechanisms, such as Deferred Acceptance, Serial Dictatorship, Random Priority or Top Trading Cycle, letting each agent know his true ordinal ranking over the objects is almost never a social welfare-maximizing information policy. By way of a partial solution, I then propose a simple signal I call the Object Recommendation (OR) Signal. Under i.i.d. agent priors satisfying a mild regularity condition, I show that, when agents' a priori relative preferences over the objects are "not too strong", the OR Signal, used together with any of the aforementioned standard mechanisms, not only maximizes welfare, but achieves first-best, i.e. the unconstrained maximum total ex-ante welfare.

Presentations and seminars - 32nd Stony Brook International Conference on Game Theory, 2021 (Poster session, Jul 2021);  Graduate Student Conference, Economics department of the Washington University in St. Louis * (Sep 2021), 2021 Midwest Economic Theory conference, Michigan State University in East Lansing * (Sep 2021), 50th annual meeting of the Illinois Economics Association (Oct 2021),  theory group seminars / brown bags at NYU (Mar 2021),  Colorado State University (Feb 2021), Simon Fraser University (Feb 2021),  University of Sao Paolo, Brazil (Nov 2021),  University of Melbourne (Mar 2021),  Indian Statistical Institute  (Jan 2021).
* Could not attend due to Covid related travel restrictions

We explore the consequences of weakening the notion of incentive compatibility from strategy-proofness to ordinal Bayesian incentive compatibility (OBIC) in the random assignment model. If the common prior of the agents is a uniform prior, then a large class of random mechanisms are OBIC with respect to this prior -- this includes the probabilistic serial mechanism. We then introduce a robust version of OBIC: a mechanism is locally robust OBIC if it is OBIC with respect all independent priors in some neighborhood of a given independent prior. We show that every locally robust OBIC mechanism satisfying a mild property called elementary monotonicity is strategy-proof. This leads to a strengthening of the impossibility result in Bogomolnaia and Moulin (2001): if there are at least four agents, there is no locally robust OBIC and ordinally efficient mechanism satisfying equal treatment of equals.