(As Yet) Unpublished Papers:

NEW!!! Information and the Bandit: The Good, the Bad and the Ugly (with Johannes Hölzemann)

Abstract: We study a game of strategic experimentation in which information arrives through fully revealing, publicly observable, breakdowns. In line with our theoretical predictions, we find that players experiment significantly less and payoffs are lower when actions are hidden. We run a robustness test where we study a game of strategic experimentation in which information arrives through fully revealing, publicly observable, breakthroughs. In the case of breakthroughs, both experimentation and payoffs are higher with hidden actions—even when theory does not predict any difference in equilibrium. We view this as evidence that behavior is systematically affected by the informational environment. Moreover, behavior is consistent with strategic free-riding, as information is a public good and players produce inefficiently little of it.

Racing with a Rearview Mirror: Innovation Lag and Investment Dynamics (with Chantal Marlats and Lucie Ménager)

Abstract: We analyze a dynamic investment model in which short-lived agents sequentially decide how much to invest in a project of uncertain feasibility. The outcome of the project (success/failure) is observed after a fixed lag. We characterize the unique equilibrium and show that, in contrast with the case without lag, the unique equilibrium dynamics is not in thresholds. If the initial belief is relatively high, investment decreases monotonically as agents become more pessimistic about the feasibility of the innovation. Otherwise, investment is not monotone in the public belief: players alternate periods of no investment and periods of positive, decreasing investment. The reason is that the outcome lag creates competition between a player and her immediate predecessors. A player whose predecessors did not invest may find investment attractive even if she is more pessimistic about the technology than her predecessors. We compare the total investment obtained in this equilibrium with that obtained with an alternative reward scheme where a mediator collects all the information about the players' experiences until some deadline, and splits the payoff between all the players who obtained a success before the deadline.

[Full paper to come soon]

Non-Common Priors, Incentives, and Promotions: The Role of Learning (with Matthias Fahn)

Abstract: We analyze a repeated principal-agent setting in which the principal cares about the agent's verifiable effort as well as an extra profit that can be generated only if the agent is talented. The agent is overconfident about his talent and updates beliefs using Bayes' rule. An exploitation contract in which the agent is only compensated for his effort if the extra profit materializes maximizes the principal's profits. In this optimal contract, the agent's principal-expected compensation decreases over time and learning exacerbates his exploitation, unless he has been revealed to be talented. Therefore, the principal's profits may increase with failures, and the agent may only be employed if his perceived talent is sufficiently low. As an application of these results, we analyse a firm's optimal promotion policy, and show that promotion to a new job may optimally be based on the agent being successful in a previous job, even if the agent's talent across jobs is entirely uncorrelated. This provides a novel explanation for the so-called Peter Principle, for which Benson et al., 2019 have recently provided evidence in a setting with verifiable performance and highly confident workers.

Strategic Experimentation with asymmetric safe options (with Kaustav Das and Katharina Schmid)

Abstract: We study a two-player game of strategic experimentation with exponential bandits à la Keller, Rady and Cripps (2005) where the safe-arm payoff is different across players. We show that, as in Das, Klein and Schmid (2020), there exists an equilibrium in cutoff strategies if and only if the difference in safe-arm payoffs is large enough. In the equilibrium in cutoff strategies, the player with the higher safe-arm payoff conducts less experimentation. This feature of the equilibrium offers an explanation for the fact that oftentimes technological innovations are due to startups rather than established market leaders.

Over-and Under-Experimentation in a Patent Race with Private Learning (with Kaustav Das)

Abstract: This paper analyses a two-player game with two-armed exponential bandits. A player experiences publicly observable arrivals by pulling the safe arm. On the other hand, a player operating a good risky arm experiences publicly observable arrivals at an intensity greater than that in the safe arm. In addition, a player pulling the risky arm can also privately learn about its quality. With direct payoff externalities and private learning, we construct a symmetric Markov equilibrium where, depending on the initial optimism about the quality of the risky arm, we can have either too much or too little experimentation.

Work in Its Earlier Gestational Stages:

* Over-Cautious or Trigger-Happy Advisors---When Best to Stop (with Sidartha Gordon)