Published Papers

"The Optimal Assortativity of Teams Inside the Firm" (with Ashwin Kambhampati), RAND Journal of Economics, 2022

Abstract

How does a profit-maximizing manager form teams and compensate workers when workers have private information about their productivity and exert hidden effort once in a team? We study a team-production model in which positive assortative matching is both efficient and profit-maximizing under pure adverse selection and pure moral hazard. We show that the interaction of adverse selection and moral hazard can lead to nonassortative matching if complementarities are sufficiently weak. When this is the case, the manager may prefer to delegate matching, allowing workers to sort themselves into teams.  

"Why Informationally Diverse Teams Need Not Form, Even When Efficient" (with Ashwin Kambhampati  and Peng Shao)

Abstract

We introduce a model of team formation in which workers first match and then produce correlated signals about an unknown state. While it is efficient to maximize the number of informationally diverse teams, such teams need not form in equilibrium when output is shared equally. Our analysis identifies the two sources of matching inefficiency: (i) workers may form diverse teams that are beneficial to its members, but force excluded workers to form homogeneous teams, and (ii) even when a diverse team is efficient, a worker may prefer to join a homogeneous team if she can exert less effort than her teammate. We completely characterize each inefficiency. 

Working Papers

"Selling Data"

Abstract

A profit-maximizing monopolist (seller) sells multi-attribute consumer data to a firm (buyer). The seller is uncertain about which consumer characteristic the buyer is interested in forecasting and how much the buyer values information. In order to screen among potential buyers along both margins, the seller chooses a menu of statistics of the data to offer and the price of each statistic. Assuming that the data and unknown characteristics follow an elliptical distribution, I obtain two results. First, I show that the seller optimally offers statistics that are linear combinations of the data. Second, I show that the seller might need to offer a continuum of statistics, and that they are less correlated than they would be if the seller could perfectly discriminate. Every optimal statistic contains information about every variable in the data, and does not include uncorrelated noise.

"Inverse Selection" (with Markus Brunnermeier and Rohit Lamba)

Abstract

AI, big data, and machine learning inverts adverse selection problems; it allows insurers to infer statistical information, thus reversing the information advantage from insuree to insurer. However, standard insurance models assume private information only on the insuree’s side. In this paper, the underlying risk is two-dimensional; one of which is exclusively known to the insuree and the insurer knows the correlation between the two. The insurer faces a new obfuscation-vs-discrimination trade-off, and by controlling the statistical information about the correlation, she can significantly increase her profits, especially if the insurer cannot do the appropriate Bayesian inference.

"Higher Order Information Complementarities and Polarization"

Abstract

I study endogenous network formation in an environment in which individuals want to forecast a stochastic state and it is costly for them to communicate with others to exchange some exogenously observed information. Due to the existence of information complementarities, individuals’ preferences for networks in which they have multiple neighbors cannot be characterized by a linear ranking of the pairwise correlations between their signals. Instead, these complementarities generate a counterintuitive result: for a fixed number of individuals, information structures exist in which all signals are conditionally positively correlated, and these are preferred to a structure in which all signals are conditionally independent. Therefore, it may be that the only strongly stable network consists of two cliques with signals that are highly positively correlated within each clique that generate different beliefs across cliques, even when there are opportunities to exchange information with individuals sharing less correlated signals. Thus, this model exemplifies how homophily and belief polarization can coexist in a rational environment.