We study the design of approval rules when costly experimentation must be delegated to an agent with misaligned preferences. When the agent has the option of ending experimentation, we show that, in contrast to standard stopping problems, the optimal approval rule must be history-dependent. We characterize optimal rule and show that the approval threshold moves downward over the course of experimentation. The threshold in force at any time depends on the history only via the minimum of previous beliefs and the current belief. We find that private information qualitatively may change the optimal mechanism: an agent can choose a fast-track option in the form of an initially depressed approval threshold. On expiry of the fast track the threshold jumps up, introducing more stringent standards. Our results provide a theoretical foundation for both the loosening of approval standards on longer experimentation paths and fast-track mechanisms.
A famous result in the auction literature is that in a common-value auction, the auctioneer can raise revenue by revealing information about the value of the good being auctioned. In this paper, we ask whether a similar result exists when the auctioneer reveals information about the type of bidders in the auction (e.g., the precision of their information structure). We consider a model in which “type” corresponds to signal precision, or a possible private-value advantage, and study equilibrium in these multi-dimensional settings. We establish the existence of equilibrium and then turn to questions of auction design. Specifically, we demonstrate that the public disclosure of all bidder types decreases revenue compared to the case in which types remain private knowledge. Simple examples show that this decrease can be very large. Our finding speaks to the benefits of keeping bidder’s types hidden as a simple and easily implementable element of auction design.
Economists have often studied how changing the auction format can change bidding strategies and be beneficial for the auctioneer. A specific aspect of auction design concerns information disclosure, in which the auctioneer can change the information available to bidders, thereby altering bidder behavior and possibly increasing the revenue an auction generates. We study common-value second-price auctions where bidders differ in the precision of their information (i.e., they are experts or non-experts) and compare two auction designs, Disclosure and Non-Disclosure. Under Disclosure in which bidders are told if their opponents are experts or not. The theory developed in a companion paper predicts that bidders should decrease their bids when facing an expert, so that Non-Disclosure generates higher revenue. Despite the presence of the winner's curse, we find experimental evidence that Non-Disclosure does generate higher revenue and achieves roughly 50% of the theoretical gains. Looking at individual bidding behavior, the higher revenue generated by non-disclosure appears to be due bidder behavior in line with that predicted by the theory.
We study a principal-agent relationship in which the agent has private information about the future profitability of the relationship or a currently operated project, but is biased in favor of continuing the project. When the principal retains liquidation rights over the relationship or project and must introduce distortions in the liquidation policy itself in order to elicit the agent's private information. The optimal policy consists of a threshold which, if the profitability falls below, triggers liquidation. When the agent reports a higher growth rate of the projects profitability, the optimal threshold will be either decreasing over time and approach the principal’s first-best level (i.e., the distortions from eliciting the agent’s information are temporary) or will be increasing and divergent over time (i.e., liquidation at later times takes place at unboundedly inefficient levels). A simple condition on the relative profitability of the project across agent types tells us when the distortions are temporary or permanent. These results are robust to the use of transfers (e.g., wage payments) provided that a limited liability condition is respected for the agent. They are also robust to the use of direct auditing methods to assess profitability. The model provides a tractable way to analyze contractual distortions in the pretense of private information, and in particular, shows that contracts simultaneously front- and back-loaded across a menu of options in the same principal-agent relationship.
Consider a financial firm whose solvency is perceived to be in danger, so that the market is on the edge of withdrawing investment to it, thus jeopardizing its growth rate. We study the problem of a government that is deciding how to optimally support the firm. Among the tools available to the government are fiscal interventions (a bailout or ongoing financial injections) and the possibility of information disclosure regarding the health of the firm (say, via a stress test). We fully characterize optimal interventions. First, when information disclosure is not available as a policy instrument, the optimal course of action bails out only firms of intermediate size and thereafter injects funds to ensure they remain solvent. We show that these features are maintained even when information disclosure is allowed as an additional instrument. Additionally, if information disclosure (modeled in the spirit of a persuasion game) is chosen at all, it will be tried before any fiscal intervention and subsequent intervention is triggered only after the stress test reveals the firm’s state. We fully characterize the optimal dynamic information structure under some technical conditions and show that the optimal information structure consists of the realization of a signal which reveals the bank’s health to be poor and that the government immediately bails out the firm upon such a signal realization. We explore the relationship between these two policies and show that they are complements. Additionally, we examine when full and immediate information disclosure is optimal and provide a necessary and sufficient condition for the optimality of such a policy.
This paper adds to the existing experimental (and theoretical) common value auction literature in two ways. First we introduce social learning into an auction and demonstrate that such learning can be dysfunctional, i.e., lead to a greater incidence of the winner’s curse. Second, on a methodological level, we introduce some methods not previously used in the laboratory. More precisely, we elicit the entire bid function of bidders each period during the auction rather than eliciting a single bid associated with one particular value (or signal). We demonstrate that the winner’s curse is not uniform over the domain of the bid function - - it is less pronounced for high as opposed to low signals (values).