Working Papers
Misspecification-Aware Sequential Social Learning, Job Market Paper
We study a model of sequential social learning à la Smith and Sørensen (2000), in which agents face misspecified signal distributions and are aware of this misspecification. Agents therefore choose actions according to a criterion that is robust to misspecification. We show that greater concern for misspecification makes the thresholds for taking high and low actions more extreme. We also identify a key trade-off governing the speed of learning: greater concern makes extreme actions more informative, but also less frequent. As a result, the speed at which beliefs converge is non-monotone in the degree of concern for misspecification. A moderate degree of concern can accelerate learning, whereas excessive concern hinders it. We further characterize limiting beliefs and the likelihood of convergence.
Cartel Sustainability under Declining Demand, with Masahiro Nishida
How are cartels sustained despite declining demand and firm exit? Standard repeated game theory predicts that cartels cannot be maintained when firms anticipate exit, as exiting firms have incentives to deviate just before exit without facing punishment. We develop a theoretical model showing that money transfers to exiting firms can sustain cartels through conditional transfers in firms’ final period. Using novel data from Japan’s ready-mixed concrete industry (1993-2004), where cooperatives engage in legally sanctioned cartels, we find evidence that declining future expected demand increases cartel collapse probability, cartels increase prices, and firm exit responds to demand conditions. These results are broadly consistent with our theoretical predictions.
Procurement Auctions with Ambiguity, 2020 (MSc. thesis)
We study how model ambiguity aversion and risk attitudes affect bidding behavior and market outcomes in procurement auctions with limited liability. In many government contracts, bidders face uncertainty not only about competitors’ costs but also about their own future production costs, relying on multiple models without knowing which is correct. I show that ambiguity aversion leads to less aggressive bidding when uncertainty concerns rivals’ costs, while ambiguity over execution costs generates heterogeneous bidding responses due to limited liability. The results offer an explanation for the abnormally low tenders often observed in construction procurement and shed light on how model uncertainty and commitment constraints shape strategic behavior and market efficiency.
Work in Progress
Trustworthy Mechanisms. Designer Incentive Compatibility under Distribution-Preserving Manipulation, with Alfonso Maselli
We develop a new framework to incorporate partial commitment in a mechanism design setting. The designer can strategically manipulate the proposed mechanism as long as she does not alter the implied distribution over outcomes. In this context, a mechanism is deemed credible only if there is no mechanism inducing the same output distribution that improves the designer’s expected payoff. First, we argue that the revelation principle can fail in this setting. We then show that when the preferences of the designer do not depend on the type, then all mechanisms are credible. We focus our analysis on the case when the designer’s preferences depend on the type and argue that in the classical screening problem in a quasi-linear environment, the optimal incentive compatible and individually rational mechanism is not credible, as the designer will have an incentive to swap contracts between types. Finally, we characterize the credibility constraint via optimal transportation methods. We show that a mechanism is credible if and only if the joint distribution it induces over agents’ messages and outcomes is cyclically monotone with respect to the designer’s interim utility.