Abstract: We study the effects of intensifying competition for contracts in the context of U.S. Defense procurement. Conceptually, opening contracts up to bids by more participants leads to lower awarding prices, but may hinder buyers' control over non-contractible characteristics of prospective contractors. Leveraging a regulation that mandates agencies to publicize certain contract opportunities, we document that expanding the set of bidders reduces award prices, but deteriorates post-award performance, resulting in more cost overruns and delays. To further study the scope of this tension, we develop and estimate a model in which the buyer endogenously chooses the intensity of competition, invited sellers decide on auction participation and bidding, and the winner executes the contract ex-post. Model estimates indicate substantial heterogeneity in ex-post performance across contractors, and show that simple adjustments to the current regulation that account for adverse selection could provide 2 percent of savings in procurement spending, or $104 million annually.

Abstract: Attempts to curb undesired behavior through regulation get complicated when agents can adapt to circumvent enforcement. We test a model of enforcement with learning and adaptation, by auditing vendors selling illegal fish in Chile in a randomized controlled trial, and tracking them daily using mystery shoppers. Conducting audits on a predictable schedule and (counter-intuitively) at high frequency is less effective as agents learn to take advantage of loopholes. We observe the specific defensive actions vendors adopt to circumvent fines, and their pattern of adoption over time is consistent with the model of learning. A consumer information campaign proves to be almost as cost-effective at curbing illegal sales, and obviates the need for complex monitoring and policing. The Chilean government subsequently chose to scale up the information campaign.  

Media: Yale Insights, VoxDev, J-PAL [Summary], [Blog], [Case Study]

Abstract: This paper studies regression discontinuity (RD) designs in settings where the assignment variable is mismeasured. In addition to the standard sources of classical measurement error, we allow the assignment variable to be affected by the treatment. We first establish how to recover the RD parameters of interest in this case, given the distributions of measurement error and treatment effects. We then show that these objects can be nonparametrically recovered from the densities of the mismeasured running variable conditional on treatment status. In the absence of mismeasurement, the conditional densities of the running variable for treatment and control units should each be sharply discontinuous at the threshold. The difference between these sharp benchmarks and the observed densities reveals the extent of mismeasurement, and can be used to adjust the RD estimates. Imposing further assumptions, identification is possible even in the presence of “manipulation” of the running variable. We develop a method to estimate treatment effects in this context based on our results. We discuss examples of settings that fit within our framework, and illustrate our method with a particular empirical application on the effect of a policy to increase competition for public procurement contracts.