Job Market Paper:
"Risk Aversion in Information Cascades"
[Draft November 2025, AEA RCT ID: AEARCTR-0014525]
How do risk preferences influence the speed and character of social learning? I consider this question with theoretical modeling and a laboratory experiment, extending classic information cascade settings (Banerjee, 1992; Bikhchandani et al., 1992) by assuming that agents differ in risk preferences and choose between asymmetrically risky options. Theory predictions show that heterogeneity of agents’ risk preferences may slow down convergence speed and increase the frequency of converging to incorrect choices. The experiment shows that preference heterogeneity per se neither slows cascades nor affects the likelihood of convergence to incorrect choices. However, when individuals know the exact risk preference types of others – rather than just their population frequencies – the likelihood of converging to incorrect choices increases significantly. This suggests that detailed information on heterogeneous preferences can undermine social learning, likely because its perceived complexity prompts reliance on behavioral heuristics and introduces imprecision in information processing – a result not previously documented. These insights help explain coordination failures in insurance, financial, and business contexts, where knowledge of others’ preferences may inadvertently distort collective decisions.
Work in Progress:
"Post(-Mongol) Roads to Path Dependence" (joint with Sebastian Ottinger)
Why do cities emerge where they do? This paper exploits a rule-based transport network in Imperial Russia to study the origins of urban centers. The yams postal system, introduced by the Mongols in the thirteenth century and maintained by Muscovy, required relay stations in regular intervals to change horses, creating an infrastructure grid whose spacing reflected logistics rather than geography or pre-existing settlements. We digitize all stations listed in the 1777 Russian Road Guide along a sample of 15 major routes, and divide rays between consecutive stops into 0.5 km cells. In modern satellite data, cells located at the historical interval where horses were changed are about thirty percent brighter today than neighboring cells before or after that range. The effect is robust to first- and second-nature controls, ray fixed effects, and controlling of pre-1800 settlements, and is absent for the later Trans-Siberian Railway. Additional analyses show that subsequent city growth correlates little with geographic endowments, but was amplified by later infrastructure investments, suggesting that administrative accidents – not natural advantages – seeded some of Russia’s urban geography. The findings illustrate how spatial inequality can arise from arbitrary historical coordination points, with lasting consequences for the distribution of economic activity.
"The Role of Rational Inattention in the Ellsberg Paradox"
In this paper I examine how rational inattention affects decisions under ambiguity in the Ellsberg paradox and interacts with ambiguity aversion. I develop a theoretical model in which individuals acquire information sequentially at a cognitive cost, and ambiguity aversion is captured by a concave aggregator. The model is followed by an online experiment in which participants make standard Ellsberg choices, report beliefs for monetary incentives, and can revise their decisions, allowing the effect of rational inattention on decision-making to be identified. A separate task measures willingness to pay to increase or decrease ambiguity under combined risk and ambiguity. Preliminary results suggest that inattention and ambiguity aversion can reinforce each other, reducing learning and increasing reliance on non-ambiguous options.