Martin Vaeth

Job Market Paper

Rational Voter Learning, Issue Alignment, and Polarization (SSRN)

2024 Best Job Market Paper Award (European Economic Association and UniCredit Foundation)

Abstract: We model electoral competition between two parties when voters can rationally learn about their political positions through flexible information acquisition. Rational voter learning generates polarized and aligned political preferences, even when voters’ true positions are unimodally distributed and independent across policy issues. When parties strategically select their positions, voter and party polarization mutually reinforce each other, and both rise as information costs decline. Because voters learn exclusively about the axis of disagreement between parties, party positions respond to only one dimension of aggregate shocks to voter preferences. We adapt our model to a market setting with horizontally differentiated goods when consumers learn about their product preferences. A reduction in information cost not only increases product differentiation but also leads to higher markups, reducing consumer welfare. These results show how lower information costs can reduce welfare in both political and economic contexts.

Working Papers

Attention and Regret (SSRN)- Revise & Resubmit at Journal of Political Economy

Abstract: This paper explains regret as an optimal self-control mechanism to motivate attention, and so improve decision-making. The model endogenizes the optimal emotions as incentives for an agent who overweights the cost of attention, for example due to temptation or present bias. If ex post the realized state is observable, the model provides a foundation for regret theory, including disproportionate aversion to large regrets. Advancing regret theory, the model explains why regret is stronger than rejoicing and why it is stronger in simpler decision problems. If the realized state is imperfectly observable, the model predicts a combination of regret and disappointment.

Imprecision Attenuates Updating

Abstract: Agents often base decisions on noisy signals, attenuating Bayesian updating toward the prior expectation — a phenomenon well-established in the normal-normal signal-extraction model. We show that this attenuation effect extends to all symmetric, log-concave distributions. By introducing a notion of precision based on likelihood-ratio dominance, we prove that when both the prior and noise are symmetric and log-concave, the posterior mean moves closer to the prior mean as the signal becomes less precise. We discuss  applications to cognitive imprecision, prior precision, and overconfidence.

The Optimal Design of Public Recognition Schemes

Abstract: We study the optimal design of public recognition schemes to incentivize agents who care about their social image – the public’s belief about their private type, such as ability or prosociality. We allow public recognition schemes to take the form of any signal structure, employing an information design approach. When agents are risk neutral over image, we show that one can restrict attention without loss to monotone partitional signals and characterize the optimum. If agents are risk averse over image, it may be optimal to maintain full privacy and not screen the agent – by contrast to monetary incentive schemes where screening remains optimal even under risk aversion.

Work in Progress

Self-Control through Emotions - Guilt and Pride