Xiao Yun Qiu 

Abstract: We study how the design of testing institutions, encompassing both the tests themselves and the procedures used to administer them, shapes selection outcomes in environments with multiple criteria and strategic agents.

We model the testing agency as either a set of independent bureaucracies (each test administered separately) or a joint bureaucracy (where test order and personalization can be coordinated). Our mechanism design analysis shows that under a joint bureaucracy, fixed-order sequential mechanisms with stringent tests are optimal for maximizing the probability mass of qualified candidates selected. Furthermore, we demonstrate that personalizing tests through upfront communication, now increasingly feasible via AI and automation, can select all qualified candidates.

Finally, we compare institutional settings and quantify the value of controlling test order, showing that the benefit depends critically on the distribution of testees and the stringency of optimal tests. Our results contribute to the design of robust, efficient, and fair testing systems in both human and AI-mediated environments.

Abstract:  We study optimal allocation mechanisms that use costly signals as screening devices. A social planner seeks to minimize total signaling costs, subject to the implementation of a given monotone allocation rule. In these settings, signal recommendations based on all agents' reports (allowing information leakage) can improve coordination among losers and reduce their signaling costs. However, this comes at the expense of inducing excessive effort from potential winners. We show that when higher types have higher certainty-equivalent signals, the incentive cost of coordination outweighs its benefit. In this case, non-coordination, recommending signals without conditioning on others' reports, is optimal. Finally, we show that such non-coordination mechanisms can be implemented through coarse-ranking contests.

Abstract:  We study how to allocate resources to participants who can strategically misrepresent their deservingness at a cost. A principal assigns item(s) (or money) among multiple agents on the basis of their costly signals. Each agent's signal reflects his private type, absence of misrepresentation, but can be inflated above his true type at a cost. The principal is a social planner who aims to maximize the weighted average of matching efficiency (allocating resources to the most needed) and a utilitarian objective (a measure of participants' welfare ). Strategic misrepresentation introduces novel incentive-compatibility constraints, under which we characterize the optimal mechanism. We apply our characterization to two kinds of public programs, distinguished by resource scarcity, and show that the principal strictly benefits from randomizing the allocations based on costly signals when the population of participants is large enough. Interestingly, in public programs with scarce resources, the format of the optimal mechanism converges to a winner-takes-all contest; however, there is a non-diminishing value of randomizing allocations to middle types as the population of participants grows.

Abstract: The word `overfitting' has ambiguous meaning in different contexts. The goal of this paper is to provide a game-theoretic definition of overfitting in a generic machine learning contest, where each contestant can allocate effort among two actions: model development that   improves  quality  of the model as desired by the contest host, and parameter tuning that only improve the model's fitness to the particular task in contest which is not  the contest host's true objective. We establish the existence of a symmetric monotone pure strategy  equilibrium in this competition game. It also provides  a natural definition for overfitting in this strategic context by comparing a player's equilibrium effort allocation to a single-agent benchmark scenario. Under our definition, contestants with types below certain threshold (low types) always overfit, whereas those above a (maybe) different threshold do not. As the contest reward increases, low types overfit more and we provide empirical evidence for this.  We also show that the equilibrium observed in Kaggle competition data is monotone.

Abstract: This paper studies the algorithmic complexity for computing the pure Nash Equilibrium (PNE) in Tullock Contests. The (possibly heterogeneous) elasticity parameter ri determines whether a contestant i’s cost function is convex, concave or neither. Our core finding is that the domains of ri governs the complexity for solving Tullock contents. 

• When no contestant’s ri lies between (1,2], we can design an efficient algorithm to compute the pure NE; 

• when many ri values fall within (1,2], we prove that determining NE existence can not be solved in polynomial time, assuming the Exponential Time Hypothesis (ETH); 

• when many ri values fall within (1,2], we design a Fully Polynomial-Time Approximation Scheme (FPTAS) to find an ϵ-PNE when an exact PNE exists. 

All our algorithms are efficiently implemented for solving large-scale instances, and computational experiments demonstrate their effectiveness even in complex scenarios. 

Abstract: This paper considers a contract design problem where the agent can allocate effort between long-term investment, which can upgrade productivity in the future but require upfront investment, and short-term gaming behavior, which can boost superficial performance but not real productivity. The designer only cares about productivity but the only available information to her is muddled. I show that the optimal contract is stationary. Monetary incentive per se cannot deter gaming, but a contract with a trial period can effectively separate long-term investment apart from short-term gaming behavior.