Optimal Large-Contest Design

summary

The gist

In this paper, we characterize the extreme points of a class of multidimensional monotone functions and apply this result to large contests to provide a useful representation of optimal allocation…

In short

The paper characterizes extreme points of multidimensional monotone functions to find optimal allocation rules for large contests. It shows that the best contest designer can use a two-threshold scheme, providing a mathematical framework for college admissions and funding schemes under various preference distributions.

Key concepts

Extreme Points of F*
These are the simplest possible solutions (step functions with at most two jumps) for the set of functions satisfying a budget constraint in this context. They represent the fundamental building blocks for understanding optimal allocation rules.
Maximizing a Functional Π
This involves finding an allocation rule that maximizes a specific objective function related to agent performance and effort distribution. The paper uses Bauer’s Maximum Principle to show that the maximizers can be found among these extreme points.
Optimal Allocation Rule q*
This is the best strategy for a contest designer to set, determining who gets a prize based on an agent's effort level. Depending on the preference function, this rule is either a two-threshold scheme or a simpler constant allocation.
Monotone Functions
These are functions where increasing an input leads to an output that never decreases. The paper studies how these functions behave in multidimensional settings, which helps model complex agent preferences and performance metrics.

Terminology used across episodes

This episode discusses

The paper

Optimal Large-Contest Design · Read on arXiv

University of Fribourg

In this paper, we provide a theoretical framework to analyze the optimal design of large contests, such as university admissions and research funding. Under a broad class of distributional preferences, the optimal allocation rules are step functions with at most two steps, thus rationalizing several real-world examples. From a methodological perspective, we leverage the mathematics of extreme points to characterize optimal allocation rules in environments where the distribution of efforts can be endogenous. When the designer wants to maximize aggregate effort, the optimal allocation rule sets a performance threshold and it awards prizes at random above it, provided noise is sufficiently small. When noise is larger, awarding prizes to the best performers becomes optimal.

Transcript

Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.

Ruth: We're looking at the paper: "Optimal Large-Contest Design".

Warren: Here's the gist of the paper. In this paper, we characterize the extreme points of a class of multidimensional monotone functions and apply this result to large contests to provide a useful representation of optimal allocation rules under a broad class of distributional preferences The best the contest designer can do is to set a two-threshold scheme This model provides a framework to rationalize several real-world examples of large contests, such as college and university admissions and funding schemes Characterization of Extreme Points The paper characterizes the extreme points of the set of functions F∗, which is defined as those monotone functions f in F satisfying a budget constraint A result by

Win88: shows that the set of extreme points of F∗ is a convex combination of at most two extreme points of F This means the extreme points are a mixture of at most two indicator functions defined on up-sets Theorem 1 states that the set of extreme points of F∗ is given by f∗(x, µ) = λIA1 + (1 − λ)IA2 for each µ in ∆(X) and some A1 ⊆ A2 In other words, extreme points in F∗ are step functions on X with at most two jumps Maximizing a Functional The paper considers maximizing a functional Π: F∗−→ R defined as Z X Φ (f(x, µ), x) dµ(x) Theorem 2 states that the integral inequality holds for any two functions f, g in F such that f majorizes g if and only if Φ(u, t) is convex in u and supermodular in (u, t) By Bauer’s Maximum Principle, the maximizers of (1) can be found among the extreme points of the feasible set Theorems 1 and 2 yield simple solutions to (1) Large Contest Analysis The paper analyzes large contests where there is a measure-one mass of agents who can choose an effort level x in X =

x, x: Definition 2 specifies that for every x in X and µ in ∆(X), an Allocation Rule q: X × ∆(X)−→

0, 1: is non-decreasing in x and satisfies Z q(x, µ)dµ(x) = k Theorem 3 characterizes the optimal allocation rules based on the monotonicity of π(x) If π(x) is increasing, the optimal Allocation Rule is given by q∗(·, µ) = (1 x > x˜ 0 x < x, ˜) Incomplete Information and Extensions The analysis extends to incomplete information where performance is noisy, leading to Corollary 1 which states that Theorem 3 still holds under incomplete information The paper also considers more general objective functions, such as maximizing Π′ = Z X π'(x) dµ(x) Furthermore, the model can be extended to include diversity constraints by considering agents with observable characteristics θ ∈ Θ The designer’s problem then becomes equivalent to two subproblems for each subset of agents This paper characterizes the set of extreme points of a class of multidimensional monotone functions and applies it to large contests The results provide an economic foundation for two-tier prize schemes and significantly simplify the characterization of equilibria in this class of games with complete information Future research directions include relaxing the assumption that the effort level has compact support and investigating robustness to more general specification of agents’ utility functions The results strongly follow from the additive separable specification of the agents’ utility function in (3) and (8) References

AC24: Yaron Azrieli and Christopher P.

Ruth: Chambers, Success functions in large contests, May 2024, arXiv:2403.07152

AO24: J´erˆome Adda and Marco Ottaviani, Grantmaking, Grading on a Curve, and the Paradox of Relative Evaluation in Nonmarkets.

Warren: The Quarterly Journal of Economics, 139(2):1255–1319, May 2024

Bal23: Ian Ball, Bauer’s maximum principle for quasiconvex functions.

Ruth: arXiv preprint:2305.04893, 2023

Cho54: Gustave Choquet, Theory of capacities.

Title and authors: Ruth: Let's talk about the title of this paper, "Optimal Large-Contest Design," and who wrote it. It’s a very direct title, suggesting it’s focused squarely on figuring out the best way to design contests when you have a lot of agents.

Warren: The authors are Valvassori Bolge, and they thank a lot of people for their help in getting this work done. It seems like it's built on some older ideas from people like Olszewski and others who worked on large contests before.

Iris: I see the references pointing back to work by Azrieli and Chambers, which talks about contest success functions, and they are building on that framework to tackle a specific problem: optimal design by the principal.

Ruth: That's right. The paper is trying to bridge the gap between characterizing these extreme points of monotone functions and actually designing an allocation rule that works best for the designer's goals.

Warren: So, when you look at what they are doing, it seems they’re taking a mathematical property—the extreme points—and translating it into a practical design principle for things like college admissions or funding schemes.

Iris: It moves the discussion from just modeling how agents behave to determining what the best possible structure for the rules actually looks like mathematically.

Ruth: And that's what it is, they are providing an economic foundation for two-tier prize schemes, which is a very concrete application of their theory on extreme points.

Warren: It’s interesting because it suggests that even with complex distributional preferences, there’s a relatively simple structure to the optimal solution.

Iris: But we have to keep in mind what they are actually assuming about the people involved in these contests—that's always a key thing for an economist.

Ruth: True. They focus on agents whose utility functions follow certain separable specifications, which is what allows them to get those simple results, like the two-threshold scheme.

Warren: So it’s less about modeling every single agent’s internal thought process and more about finding the robust structure that works across a broad class of preferences.

Iris: And I wonder how robust those two thresholds are if we introduce noise or incomplete information, because they touch on that in the extensions.

The paper's summary: Ruth: Moving into the core summary of "Optimal Large-Contest Design," the authors explain that their main contribution is characterizing these extreme points of multidimensional monotone functions.

Warren: They show that this characterization leads directly to a representation of optimal allocation rules, and they point out that for contests with complete information, this representation significantly simplifies how we characterize the equilibria.

Iris: Basically, they take the complex problem of finding the best rule and show it’s equivalent to finding these specific combinations of indicator functions.

Ruth: Yes, they state that these extreme points are a convex combination of at most two extreme points of the original set, which means we're dealing with step functions on those performance levels with at most two jumps.

Warren: So, the key takeaway for me is that instead of having an infinite number of possible rules to check, they’ve narrowed the search down to just a couple of fundamental structures.

Iris: The paper then moves into maximizing a functional that measures the total expected utility across all possible effort levels and distributions.

Ruth: They use Bauer's Maximum Principle here, which tells them that the maximizers for this integral inequality can be found among those extreme points they just characterized.

Warren: So they’re linking the optimization problem—finding the best rule—to this characterization of the extreme points of monotone functions.

Iris: And when you look at how they apply it to large contests, they define an allocation rule based on whether the performance level is above or below some threshold, depending on what's increasing or decreasing in terms of the designer's preference function.

Ruth: That specific mapping, Theorem three characterizes these optimal rules based entirely on the monotonicity of that preference function pi over effort levels <ref:2601.19331#pg2>.

Warren: So if the designer prefers higher effort generally, you get one type of rule, and if they prefer lower effort, you get another structure for the allocation.

Iris: And they even show how this framework handles incomplete information in a corollary, meaning their two-threshold idea still holds up even when performance is noisy.

The paper's improvements: Ruth: Now let's talk about what the authors suggest as improvements to this work. They aren't just presenting a static result; they are showing how this model can be extended.

Warren: They suggest extending the model to include more general objective functions, like maximizing pi prime, which is an integral over the performance derivative instead of just the performance itself.

Iris: That opens up the possibility of optimizing for different things than just raw performance outcomes; you could optimize for something else entirely.

Ruth: Right, and they also show how this model can be extended to include diversity constraints by considering agents with observable characteristics theta in addition to their effort levels x.

Warren: So instead of just thinking about effort, the designer can factor in other things about the agent, like demographics or skills.

Iris: That changes the problem structure a bit because instead of one main allocation rule, it becomes equivalent to solving two separate subproblems for each subset of agents with those observable characteristics.

Ruth: That’s a real practical implication; it means you can design rules that balance effort incentives with other agent traits simultaneously.

Warren: It seems like they are moving from a simple one-dimensional decision space to something much richer, which is necessary for modeling more realistic scenarios.

Iris: And they flag a limitation, which is important—the paper relies heavily on the additive separable specification of agents’ utility function in equations three and eight to get these clean results.

Ruth: That's fair. If you move away from that specific way the utility functions are structured, you lose that nice characterization of the extreme points they found so useful for large contests.

Conclusion: Ruth: So to wrap up, "Optimal Large-Contest Design" gives us a mathematical foundation for finding optimal allocation rules by characterizing the extreme points of these functions.

Warren: The main implication is that we can move away from trying to model every possible preference interaction and use this two-threshold scheme as a powerful way to simplify equilibrium analysis in large contests with complete information.

Iris: I just want to emphasize that this structure is most useful when the utility functions have that specific additive separable specification, because if you change that, the nice results about the extreme points don't hold anymore.

Ruth: Exactly. And they showed this framework works even under incomplete information in a corollary, which makes it more flexible for real-world use where things aren't perfectly known.

Warren: It’s a solid result because it connects abstract function characterization to concrete design choices, whether that’s college admissions or funding schemes.

Iris: I think the constraint on the utility functions is the most important caveat—it tells us exactly what kind of economic behavior this simple model can accurately predict.

Ruth: And they’ve shown how this works for various types of selection procedures, from random mapping to deterministic thresholds, giving us a lot of tools.

Warren: It’s a strong tool for designing rules where the designer wants to favor better performers without getting bogged down in the infinite complexity of all possible agent interactions.

Iris: So while it simplifies things dramatically, we still need more research on robustness when those assumptions about the effort level having compact support are relaxed.

Ruth: That sounds like a good direction for future work, pushing the boundaries of where this specific model can be applied.

More episodes

← Home