Contact Information 224-522-9569 Email Umberto

Curriculum Vitae
(PDF 125 KB / 3 pages)

Research Interests: Microeconomic Theory, Dynamic Games, Information Economics

This paper studies an experimentation game between a principal and a biased agent with no commitment and no transfers. We cast our analysis in an exponential bandit model with two actions: a ``risky'' action and a ``safe'' action. We first characterize the unique Markov-perfect equilibrium when the agent's effort choice is observable, and thus learning is symmetric. The agent trades off the chance to generate positive information about the risky action with the risk of producing negative information that makes the principal lean towards the safe action. This tension produces delay in information acquisition in equilibrium. When effort is unobservable, delay can actually hurt the agent. Thus, the agent ends up implementing the principal's optimal experimentation policy, provided that the cost of experimentation for the agent is low.

We analyze social learning and innovation in an overlapping generations model in which available technologies have correlated payoffs. Each generation experiments within a set of policies whose payoffs are initially unknown and drawn from the path of a Brownian motion with drift. Marginal innovation consists in choosing a technology within the convex hull of policies already explored and entails no direct cost. Radical innovation consists in experimenting beyond the frontier of that interval, and entails a cost that increases with the distance from the frontier, and may decrease with the best technology currently available. We study how successive generations alternate between radical and marginal innovation, in a pattern reminiscent of Schumpeterian cycles. Even when the underlying Brownian motion has a positive drift, radical innovation stops in finite time. New generations then fine-tune policies in search of a local optimum, converging to a single technology. Our analysis thus suggests that sustaining  radical innovation in the long-run requires external intervention.

This paper studies the optimal allocation of effort over projects whose undiscounted value is weakly increasing over time. Examples include exploration of new uses for existing medical drugs, and software development. Research is modeled as a sequential search problem with recall in continuous time. By exerting costly effort, a decision maker (DM) controls the drift of the discovery process, which evolves according to a Geometric Brownian Motion.  The project pays off its value--defined as the supremum of the diffusion over the realized sample path--only at the time of stopping. At each instant, the DM decides whether to work, shirk or shut down the project. The paper characterizes the optimal search policy as a function of the value of the project as well as the current discovery. The solution involves cutoff functions. The value of the project identifies two cutoffs, which determine whether the DM is going to work or shirk, and shirk or stop, depending on the level of the current discovery. An increase in the value of the project shifts up both cutoffs, because of the corresponding increase in the DM’s outside option. We find that correlation between present and future discoveries gives rise to a ``value trap’’. The DM shirks initially over projects with a low starting value, until the project’s value reaches a sufficiently high level.

Umberto Garfagnini is a doctoral candidate in Managerial Economics and Strategy at Kellogg School of Management, Northwestern University. Prior to joining Kellogg, Umberto studied Finance at Bocconi University in Milan, Italy. His research interests are in Microeconomic Theory, with applications to dynamic games with learning.