Job Market Paper
This paper uses graduate admissions to study dynamic selection under costly evaluation. With a synthetic data protocol to protect applicants’ privacy, I relate acceptance to measures of success, such as job placements. I then estimate structural models of an admissions committee’s decision problem that capture first-round filtering and waitlists, and I develop statistical tests of whether observed decision rules are optimal given time and effort constraints. To determine which if any applicant characteristics are incorrectly valued, I quantify deviation gains from adopting counterfactual decision rules. This framework is applicable to multiple settings where decision-makers face uncertain outcomes and evaluation costs.
Publications
Do the world’s best tennis pros play Nash equilibrium mixed strategies? We answer this question using data on serve direction choices (to the receiver’s left, right or body) from the Match Charting Project. Using a new approach, we test and reject a key implication of a mixed strategy Nash equilibrium: that the probability of winning the service game is identical for all possible serve strategies. We calculate best-response serve strategies by dynamic programming and show that for most elite pro servers, the DP strategy significantly increases their win probability relative to the mixed strategies they actually use.
We show that firms can employ data-driven methods to improve their hiring decisions. Specifically, we use data available to National Football League (NFL) teams prior to the NFL draft to estimate econometric models that predict the future performance of drafted quarterbacks. Since our methods are replicable, stakeholders can use them to improve the draft's efficiency and help it accomplish its mission to promote competitive balance. Furthermore, data-driven methods such as ours can help firms avoid biases against employee characteristics that do not affect future job performance.
Working Papers
This paper studies resource optimization under asymmetric spillovers, which occur when one position’s unobserved true quality disproportionately affects other positions’ observed performances. Because the NFL has vast public data, we use it as a laboratory for our study. Specifically, we estimate a production function equilibrium model of team performance by mapping salary cap spending into expected wins. To account for asymmetric spillovers between quarterbacks and their teammates, we use game-level variation to identify quarterbacks’ marginal productivity, calibrate the model, and calculate optimal positional allocations. We find that most positions were paid near their model-implied optima with a few significant deviations.
Works in Progress
Do outcomes in sequential contests exhibit dynamic dependence? We study this question using point-level data from modern professional tennis and assess how well Klaassen and Magnus’s (2001) iterated feasible generalized least squares (FGLS) estimator performs on dynamic panels. Using data from 1,336 men and women’s clay, grass, and hard-court matches from 2014 to 2017, we reject independence and identical distribution at the 1% level. However, our estimated hot hand effect is smaller than KM’s and insignificant for men, and the receiver’s estimated advantage on important points is also smaller. We then compare FGLS to several dynamic panel estimators, including GMM and bias-corrected fixed effects (BCFE), and show that FGLS estimates are comparable to BCFE and consistently lie between OLS and uncorrected FE. Given FGLS’s computational speed and ability to handle predetermined and time-invariant covariates, we recommend it for applied work with dynamic panel data.