I'm a Ph.D. student in Economics at the University of Bonn, with research interest in microeconomic theory (primary), political theory and behavioral economics (secondary). Currently, I am visiting Yale University.
I will join the University of Vienna as an assistant professor in fall 2020.
Contact: +49 151 1083 7212; +1 203 892 2723
Office: 28 Hillhouse Ave, New Haven, CT 06511
Curriculum Vitae (PDF)
Demographic groups in the population pay systematically different attention to politics and acquire different levels of information. This paper studies the effects of heterogeneous attention when a reform may benefit one group at the expense of others (distributive politics). In the benchmark, when the information of voters is exogenous, a median voter theorem holds, and a welfare-enhancing reform is not adopted if it is not preferred by a majority. When information is endogenous, attention shifts election outcomes into a direction that is welfare-improving. Even when a welfare-enhancing reform is not preferred by a majority ex-post, under certain conditions, there are equilibria where the reform is adopted. The key driver of the results is that voters who are more severely affected by a proposed reform will pay more attention, consistent with empirical studies (``issue publics hypothesis,'' Converse (1964)). This information advantage translates into voting power, precluding the majority from exerting its dominance.
This paper studies a large majority election with voters who have heterogeneous, private preferences and exogenous private signals. We show that a Bayesian persuader can implement any state-contingent outcome in some equilibrium by providing additional information. In this setting, without the additional information, a version of the Condorcet Jury Theorem would hold (Feddersen and Pesendorfer, 1997). Persuasion does not require detailed knowledge of the distribution of the voters' exogenous private information and preferences: the same additional information is effective across environments. A numerical example shows that persuasion is effective in elections with as few as 15 voters.
Individuals can often inquire about how their decisions would affect others. When do they stop the inquiry if they prefer one of their options for selfish reasons? We show causal evidence with a laboratory experiment that having a selfishly preferred option, individuals are more likely to continue the inquiry when the dominant information received up to that point suggests that behaving selfishly harms others. In contrast, when the dominant information up to that point suggests that being selfish harms nobody, individuals become more likely to stop acquiring information. Drawing on the Bayesian persuasion model of Kamenica and Gentzkow (2011), we propose a theoretical model showing that this information acquisition strategy can be optimal for a Bayesian agent who values the belief that she does not harm others but attempts to persuade herself to behave self-interestedly. The model predicts that strategic information acquisition motivated by self-interest can reduce the decisions' resulting negative externalities and improve the welfare of the affected others. This prediction was indeed found to be the case in our experiment.
We propose a non-parametric method to elicit probability weights as in Prospect Theory based on this note. This method trades off risky rewards and delayed ones. It has the following two features: first, it does not require functional form assumptions on the utility function, nor elicitation of the utility function; second, it is compatible with both monetary and non-monetary incentives. Requiring only few measurements, it is also easy to implement.
We study voting on a reform for which the implementation is at the discretion of a politician who is biased against the reform. Valuable information about the efficacy of the reform is dispersed among citizens, and by voting, citizens can transmit their information to the politician. When the politician has no discretion, the Condorcet Jury Theorem holds (Feddersen & Pesendorfer, 1998). When the politician has full discretion, in all equilibria, there is no information transmission at all (Battaglini, 2017). We consider a scenario of ``intermediate discretion’’ akin to political referenda. When the majority votes against the reform, the status quo is kept. Otherwise, the politician is bound to implement the reform but can choose the degree of implementation. This creates one-sided signaling incentives for the voters. Even when the politician is only slightly biased against the referendum, there are equilibria with ``oversignaling’’ for which the reform always passes, even in the situations where all citizens and the politician prefer it not to.
We study the design of quality certificates. The certificate holder (sender) tries to convince a receiver of his quality. The receiver benefits from more precise knowledge about the sender’s quality and can costly verify the sender's quality. We show that if the certificate is too selective, in all equilibria, senders of very high quality do not present the certificate; they pool with low-quality types who do not hold a certificate. The high types forgo a signaling opportunity (``understatement’’), but thereby motivate the receiver to costly verify their high quality herself. Conversely, when the certificate is not too selective, there are monotone equilibria where higher types are more likely to present their certificates.
In this note, I study a variant of the canonical binary-state binary-choice social learning model (Bikhchandani et al., 1992). An individual would like to choose an action only in the high state. When making her own decision, she observes previous decision-makers who chose the action. Importantly, the likelihood of observing the action of previous decision-maker depends on the state. I show that when observing the action is more likely in the low state, the individual faces an inference problem: does she observe many actions because the state is high and previous decision-makers had private information about this or because the state is low and previous actions are more visible. In this situation, learning is confounded (Smith and Sorensen, 2000).