A principal can contract with one of several agents based on observed output. If she did not hire anyone, she would take a baseline action that is also available to agents and produces some output. Agents know what happens under the baseline action, but the principal does not. In addition, the principal ignores what other actions are available to agents. In this situation, simply giving a contract away for free can make the principal worse off. To avoid this, she auctions off the contract. She cannot pinpoint what agents know about their contract values and evaluates contract and auction pairs according to her payoff in worst-case scenarios. We find that a first-price auction of a full-benefit contract, which pays all of the realized output, is optimal with a payoff guarantee of zero. Any other contract can make the principal worse off, no matter how she sells it. These findings guide those who want to outsource efforts towards improving observable outcomes but face nontrivial opportunity costs of contracting and uncertainty that is hard to quantify.

A principal considers hiring one agent to improve a valuable, observable outcome. Who to hire? How to motivate? This paper studies a decision-making process where the principal designs an incentive contract that pays according to the realized outcome and sells the contract to an agent through an auction. It characterizes the principal's worst-possible payoff of every contract-auction pair, finds the class of maximin optimal contract-auction pairs, and shows that optimal pairs offer the principal a non-negative payoff guarantee. The characterization builds off two assumptions made on a subset of contract-auction pairs: that the contract induces the contracted agent to improve (or at least not worsen) the outcome and that the auction's revenue exceeds a threshold in expectation. The principal can design an auction that attains this revenue guarantee if she knows the expected contract payment of the outcome that would occur absent agent activities or if agents know this quantity. Weak additional assumptions allow the principal to design a contract-auction pair that is maximin optimal and gives her a non-negative payoff guarantee if she has limited liability, if she incurs various additional costs, or if agents cannot be trusted to pay for the incentive contract.

I develop tools to detect irregular assignments of cases to judges and apply them to Ecuador's judicial system. I derive the sharp bounds on the overall, court-specific, and judge-specific probabilities that a case's assignment is inconsistent with existing regulations. The bounds rely on administrative case assignment data and one, or both, of the following assumptions: (i) that certain observed case characteristics do not influence which judge a case should be assigned to, and (ii) that the probability distribution over the judges that each case should be assigned to is known (e.g. uniform, random assignment). I construct a database of all publicly-available case assignments in Ecuador's district courts, with over two million assignments from 2016 to 2020, and I find that 5% of courts and judges account for 43% and 37% of irregular assignments, respectively. Overall, at least 65 thousand assignments, 2.9%, are irregular.

This study examines the relationship between municipal police forces and criminal activity in Mexico during 2008, the year that ended a fifteen year decline in the national homicide rate. Regression discontinuity estimates indicate that a 1 million 2019 dollar municipal police subsidy caused more policing for infractions and misdemeanors, but not for felonies. Surprisingly, the estimates also point to substantial homicide increases, mostly related to organized criminal activity. Our evidence suggests this effect is due to enhanced communications between the Federal Secretariat of Public Security, which is in charge of the Federal Police, and municipal governments. In addition, the subsidy is shown to reduce popular support for the president’s party, the PAN.