RESEARCH

This paper studies how agents choose to be vague in their proposals in a delegation environment. Two agents compete for the approval of a decision maker to implement a multidimensional action. Based on their knowledge of the consequences of actions, agents propose future actions but can be vague about any dimension. The decision maker, uncertain about the consequences of actions, chooses one agent to act. I show that vagueness on the dimension where one stands closer to the decision maker than his opponent preserves such an advantage, while preciseness undermines it. Vagueness therefore tends to occur on agents' advantageous dimensions.

Learning is crucial to effective organizational decision making, but often it has to be delegated. We examine a dynamic delegation problem where a principal needs to decide on a project with uncertain profitability. A biased agent, who is initially as uninformed as the principal, privately learns about the profitability over time and communicates to the principal. We formulate learning delegation as a dynamic mechanism design problem and characterize the optimal delegation scheme. Our results show that private learning gives rise to the tradeoff between how much information to acquire and how promptly the acquired information is reflected in the decision. We discuss implications on learning delegation for distinct organizations.

We study mediated information tranmission with intrinsic privacy concerns. A receiver faces a decision whose payoff depends on an unknown state. A sender has an informative signal, but prefers the receiver to know as little about it as possible. A platform with an independent informative signal relates  the combined information to the receiver to maximize her payoff. We show that it is possible to convey valuable information without incurring any privacy loss, but impossible to do so via simple restrictions such as prohibiting the platform to condition its messages on the sender's signal, or only disclosing sufficient statistics for the receiver's decision. We show that such restrictions result in Pareto loss in the environment where the tension lies between providing useful information and privacy concerns, but not when the tension lies between providing useful information and incentivizing truthful report of the sender.

We study innovation adoption as social learning with payoff externalities, where mass adoption generates information reflecting the innovation quality and influences non-adopters' payoffs. We provide a general framework for studying equilibrium adoption dynamics when the payoff externality depends on adoption level only. With negative externality, equilibria exhibit different regimes depending on how fast information arrives - faster arrival tends to induce gradual adoption, while slower arrival tends to induce rushes, the possible timing and sizes of which can be determined. With positive externality, adoption is generally gradual except for an initial rush. We discuss equilibrium welfare properties and policy implications. 

It has been demonstrated repeatedly that when experiment subjects do a slot-machine type gambling task, a near miss outcome, which is perceived close to a win, leads to higher estimation of probability of winning in the future even though it does not have a monetary reward. We explain this behavior through a model of experimentation where one agent is learning about two parameters: her ability and whether it matters in the experimentation task. We show that for certain range of beliefs about these parameters, the player exhibits the near miss effect. However, it wears off as the experimentation goes on, which is consistent with experimental evidence. 

[Click here for some new results with (1) bayesian player and (2) alternative modeling in the player's head. Sufficient confidence that her ability matters or sufficient certainty about her ability suffices for the near miss effect!]

I examine Roth (1989)'s model of matching with incomplete information about others' preferences and suggest a revision of the nonexistence proof of a mechanism that always produces a matching stable with respect to true preferences. I also show that for any stable revelation matching mechanism, the outcome is sometimes "truly unstable"; that is, it is unstable with respect to true preferences and, more importantly, the information released by the matching outcome is sufficient for some agents in a blocking pair to realize their ability to block.

This organization economics paper asks the following question: what is the optimal way to experiment when the experimenter only cares about perception? In similar spirit of Holmstrom's signal-jamming model, we consider a model where an agent chooses to costly learn about a state of the world. A principal observes the outcome of learning but not the learning choices taken by the agent. The agent’s payoff solely depends on how much the principal believes that the agent has learned. Contrasting other signal-jamming models, our model explores the trade-off between manipulating the principal’s perception now versus learning now  in order to better manipulate later. 

[This project is currently under consideration for a Chilean national research grant!]

This organization economics paper studies the simple question: how are online advertisements tailored to potential customers? We model online advertising as experimentation by a monopoly seller, who provides information about products and observes a buyer’s reaction to it in order to learn about her preferences. Different from the traditional approach to experimentation, here the outcome of experimentation can be manipulated since the buyer is aware of the seller’s motive. We study the dynamic incentives of the seller to determine how much information to provide in order to learn about the buyer, and of the buyer to determine how much information to reveal. 

[This project has received a research grant from Universidad Diego Portales!]

This information design paper asks the following question: when I have information valuable to you, how should I convey it over time to maximize your engagement, given that I have no idea what your reaction is to the information I revealed in the past? For example, a newspaper with a specific piece of information about a presidential candidate may want to strategize about the timing and extent of the revelation in order to maximize readership. Moreover, to engage a reader for as long as possible, the newspaper may pander to the possible beliefs and demand for information of the reader based her past reading habits. I take as given (1) the total value of information possessed by the newspaper, measured by the change in expected payoffs due to the information (Frankel and Kamenica, 2019), and (2) the cost of attention a consumer needs to pay in each period to obtain the information, and ask what is the maximum number of periods in which a reader is willing to pay attention.