Robust Tournaments
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.
Ruth: I'm Ruth.
Warren: And I'm Warren, here with Iris.
Ruth: We're looking at the paper: "Robust Tournaments".
Warren: Here's the gist of the paper. We characterize robust tournament design—the prize scheme that maximizes the lowest effort in a rank-order tournament where the distribution of noise is unknown, except for an upper bound, H¯, on its Shannon entropy The robust prize scheme awards positive prizes to all ranks except the last, with a distinct top prize.
Ruth: Asymptotically, the prizes follow the harmonic number sequence and induce an exponential distribution of noise with rate parameter e −H¯.
Warren: The robust prize scheme is highly unequal, especially in small tournaments, but becomes more equitable as the number of participants grows, with the Gini coefficient approaching 1/2.
Ruth: Model Setup and Equilibrium Conditions The model considers tournaments with n ≥ 2 identical, risk neutral agents indexed by i = 1,..., n.
Title and authors: Ruth: To get into what they actually did, the paper "Robust Tournaments" outlines a problem where the principal doesn't know the noise distribution but they have an upper bound on its entropy, let’s call it H bar.
Warren: The core of their work is characterizing a prize schedule that maximizes the minimum effort exerted by players under this uncertainty.
Iris: They show that this robust design awards positive prizes to every rank except the very last one and includes one top prize.
Ruth: This is different from what you might expect because it contrasts with the optimal structure when you *do* know the noise distribution, which awards some number of equal prizes at the top and zero prizes to the remaining ranks.
Warren: The paper says that for intuition, think about how tournament incentives are built by prize differences between adjacent ranks.
Iris: If an agent improves their effort enough to beat someone ranked just above them, the reward is proportional to that difference in prize.
Ruth: The robust scheme sets those differentials following a harmonic sequence as you go down the ranks.
Warren: This is what drives the structure of the prizes across all participants.
Iris: They also show that this robust design awards positive prizes to every rank except the last one with a distinct top prize.
Ruth: So they are showing how to build an incentive system that doesn't rely on knowing the exact noise distribution but still maximizes performance in the worst case scenario.
The paper's summary: Ruth: Now, where they suggest improving things, they focus on how this robustness helps us move away from just a winner-take-all idea.
Warren: They contrast their robust design with a winner take all scheme which gives all the prize budget to the top performer.
Iris: They argue that this is often implausible because executives might not know the conditions workers face or what they’re dealing with daily.
Ruth: So this paper motivates a robust design approach that guarantees a lower bound on performance no matter how uncertain the noise is.
Warren: They fully characterize this robust prize schedule as a unique solution to a convex programming problem.
Iris: This is an improvement because it gives us the actual mathematical structure we need to pursue when we have limited information about the noise.
Ruth: So they are giving us something concrete instead of just saying that being robust is good for uncertain environments.
The paper's improvements: Ruth: So, to wrap up on "Robust Tournaments," these prize structures are designed to maximize the lowest effort your agents exert given only an upper bound on noise entropy.
Warren: They give us a concrete prescription: positive prizes for all ranks except the last one, plus one top prize.
Iris: The key result is that asymptotically, this schedule becomes equitable as the number of participants grows, with a Gini coefficient approaching half.
Ruth: For me, this means we have a simple rule of thumb for designing incentives under uncertainty that isn't overly complicated.
Warren: Iris, what’s your final thought on what these results really mean for someone trying to design a system where they can't control the noise?
Iris: It means that even without knowing the exact distribution of noise, you can still build a structure that prevents catastrophic failure in terms of effort.
Ruth: The idea is that you don't need perfect information to have a reliable minimum outcome.
Warren: So for someone listening who just wants to know what this does for them, it’s about building incentives that are resilient against unknown uncertainties.
Conclusion: Ruth: So we've looked at "Robust Tournaments Mikhail Drugov Dmitry Ryvkin," and what we see is that you can design prize systems that maximize the lowest effort your agents exert even when you don't know exactly what kind of noise they're going to face.
Warren: Right, so the core idea is finding a prize structure that maximizes the minimum effort your agents exert when the noise distribution is unknown but you have an upper bound on its entropy, that H bar.
Iris: That upper bound on entropy means we can design a scheme that guarantees a certain minimum performance regardless of what's actually going wrong with the system.
Ruth: And they find that this robust scheme gives positive prizes to all ranks except the last one, plus one distinct top prize.
Warren: And as you get bigger tournaments, this prize structure starts following a harmonic number sequence.
Iris: That means the prize differences are related in a specific way that scales with the number of participants.
Ruth: It's interesting because they show that this robust scheme actually induces an exponential distribution of noise with a rate parameter e minus H bar.
Warren: And they also point out that this robust prize structure is really unequal, especially when you only have a few participants.
Iris: But as the number of participants grows, it gets more balanced and the Gini coefficient approaches half.
Ruth: So for someone just listening, it means even if you don't know the exact noise in your contest, you can design a prize system that protects against the worst-case scenario.
Warren: That leads us to thinking about how this structure works in practice, though the math is pretty dense.
Iris: The real power here is that it gives us something concrete instead of just saying "be robust."
Ruth: Exactly. It gives us the actual mathematical structure we need when we have limited information about the noise distribution.
Warren: So for someone listening who just wants to know what this does for them, it’s about building incentives that are resilient against unknown uncertainties.
Iris: It means that even without knowing the exact distribution of noise, you can still build a structure that prevents catastrophic failure in terms of effort.
Ruth: The idea is that you don't need perfect information to have a reliable minimum outcome.
Warren: So for someone listening who just wants to know what this does for them, it’s about building incentives that are resilient against unknown uncertainties.
Mikhail Drugov, Dmitry Ryvkin
New Economic School · Universitat Autònoma de Barcelona · Barcelona School of Economics
econ.TH
Submitted: 2025-07-22
Updated: 2026-11-04
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 67/100
The gist: We characterize robust tournament design—the prize scheme that maximizes the lowest effort in a rank-order tournament where the distribution of noise is unknown, except for an upper bound, H¯…
Key concepts
- Robust Tournament Design
- This is about creating a prize structure for competitions where the exact nature of performance noise is unknown, only its upper bound (H_bar). The goal is to design prizes that maximize the lowest effort required by participants, ensuring good outcomes regardless of the true noise distribution.
- Equilibrium Effort Condition
- In this model, agents choose an effort level based on their expected payoff. The equilibrium is found when the marginal probabilities of being ranked r are related to the cost function's derivative. This balance dictates how much effort each agent should exert in a stable tournament setting.
- Harmonic Prize Schedule
- For large tournaments, the optimal prize structure asymptotically follows a harmonic number sequence. This specific distribution of prizes is crucial because it results in an exponential noise distribution and ensures that the overall system remains robust against unknown noise levels.
- Gini Coefficient Convergence
- The Gini coefficient measures inequality in the prize distribution. The paper shows that for large tournaments, this inequality decreases and approaches 1/2. This suggests a stable, controlled level of prize disparity is optimal for robustness as the number of participants grows.
Terminology
Summary
We characterize robust tournament design—the prize scheme that maximizes the lowest effort in a rank-order tournament where the distribution of noise is unknown, except for an upper bound, H¯, on its Shannon entropy The robust prize scheme awards positive prizes to all ranks except the last, with a distinct top prize. p. 0 Robust Tournaments
Mikhail Drugov∗ Dmitry Ryvkin†
This version: July 23, 2025
Abstract
We characterize robust tournament design—the prize scheme that maximizes the
lowest effort in a rank-order tournament where the distribution of noise is unknown,
except for an upper bound, H¯ , on its Shannon entropy. The robust tournament
scheme awards positive prizes to all ranks except the last, with a distinct top prize. The robust tournament
scheme awards positive prizes to all ranks except the last, with a distinct top prize. Asymptotically, the prizes follow the harmonic number sequence and induce an exponential distribution of noise with rate parameter e
−H¯
. Asymptotically, the prizes follow the harmonic number sequence and induce an exponential distribution of noise with rate parameter e
−H¯
. The robust prize scheme
is highly unequal, especially in small tournaments, but becomes more equitable as
the number of participants grows, with the Gini coefficient approaching 1/2.Where the paper says it
Where the paper says it
Where the paper says it
Model Setup and Equilibrium Conditions
The model considers tournaments with n ≥ 2 identical, risk neutral agents indexed by i = 1,..., n. Each agent i chooses effort xi ≥ 0 at a cost c(xi), where c(·) is strictly increasing and strictly convex, with c(0) = c′(0) = 0. p. 4 2 The model and preliminaries
The environment We consider tournaments with n ≥ 2 identical, risk neutral agents
indexed by i = 1, . Each agent i chooses effort xi ≥ 0 at a cost c(xi), where c(·) is
strictly increasing and strictly convex, with c(0) = c
′
(0) = 0. Each agent i chooses effort xi ≥ 0 at a cost c(xi), where c(·) is
strictly increasing and strictly convex, with c(0) = c
′
(0) = 0. The output of agent i is
Yi = xi + εi
, where shocks εi are i. The output of agent i is
Yi = xi + εi
, where shocks εi are i. across agents, drawn from a distribution F(·) with
density f(·). We
assume limited liability with zero outside option, in which case it is always optimal to set
the last prize vn to zero. The equilibrium For a given vector of efforts x = (x1, . The equilibrium For a given vector of efforts x = (x1, . , xn), the expected payoff of
agent i is
π
(i)
(x) = Xn
r=1
p
(i,r)
(x)vr − c(xi),
where p
(i,r)
(x) is agent i’s probability of being ranked r. symmetrized first order condition πx(x
∗
, x∗
) = 0. This gives
X
r=1
βrvr = c
′
(x
∗
), (1)
where coefficients βr = p
(r)
x (x
∗
, x∗
)—the marginal probabilities of being ranked r in
equilibrium—are
βr =
n − 1
r − 1
Z
F(t)
n−r−1
[1 − F(t)]r−2
[n − r − (n − 1)F(t)]f(t) dF(t).Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Robust Design Problem Formulation
The principal solves the problem maxv∈V minm∈M x ∗ [F, v], which is equivalent to max d∈D min m∈M Z 1 0 a(z; d)m(z) dz. The adversary minimizes the equilibrium effort by solving min Z 1 0 a(z; d)m(z) dz s. p. 7 6) are satisfied, and the result holds. The minimization problem For a given d ∈ D, the adversary selects a distribution
of noise that minimizes the equilibrium effort, by solving
min Z 1
0
a(z; d)m(z) dz s. H[f] = −
R
f(t) log f(t) dt, is bounded above by some H¯ ∈ R. 4 Notice that we can write
the entropy using the inverse quantile density, in the form H[f] = −
R 1
0
log m(z) dz. This is a well-behaved continuous programming problem for which the Lagrangian approach produces the unique5
solution (see Appendix A for details)
m∗
(z; d) = exp
−H¯ +
Z 1
0
log a(z
′
; d) dz
′
1
a(z; d)
. (9)
The constant multiplying 1
a(z;d)
in (9) is the Lagrange multiplier associated with the
constraint in (8), which is binding.Where the paper says it
Where the paper says it
Where the paper says it
Asymptotic Results for Large Tournaments
For large tournaments, the asymptotic robust prize schedule is defined as d∞ r = 1 (n − 1)r, r = 1,..., n − 1. This prize schedule is asymptotically optimal for problem (11), in the sense that limn→∞[W(d∗)−W(d∞)] = 0. p. 3 In Section 3, we introduce
the robust tournament design problem and characterize its solution. Section 4 presents
our main results on the properties of robust optimal prizes, intuition, a detailed discussion
of small tournaments (with n = 3 and 4 agents), large tournaments (for which we identify
an asymptotically robust prize schedule), and prize inequality. the distribution of noise so as to minimize the equilibrium effort, for a given d. The
resulting optimal m(·), therefore, assigns lower density to regions where a(z; d) is larger,
trying to avoid noise realizations that are incentivized by positive prize differentials. 7This is not a coincidence because f
B(z; n − r, r) is the pdf of order statistic (n − r : n − 1) from the
uniform distribution on [0, 1]. We will use this connection in the next section to derive the asymptotic
robust prize schedule for large n.Where the paper says it
Where the paper says it
Where the paper says it
Prize Inequality Characteristics
The Gini coefficient of the optimal prize schedule decreases with tournament size and approaches 1/2 asymptotically. Proposition 3 states that for prizes v∞ in (14), the Gini coefficient is equal to 1/2 for any n. p. 0 Asymptotically, the prizes follow the harmonic number sequence and induce an exponential distribution of noise with rate parameter e
−H¯
. The robust prize scheme
is highly unequal, especially in small tournaments, but becomes more equitable as
the number of participants grows, with the Gini coefficient approaching 1/2. , prizes become more equitable,
as n increases. The asymptotically optimal prize schedule has the Gini coefficient of 1/2,
which corresponds to a substantial but not extreme level of inequality.Where the paper says it
Where the paper says it
Small Tournament Specifics
For n = 3, the robust tournament awards distinct prizes v∗1 > v∗2 > v∗3 = 0. For n = 4, the robust tournament awards two distinct (positive) prizes, with v∗1 > v∗2 = v∗3 > v∗4 = 0. p. 0 Robust Tournaments
Mikhail Drugov∗ Dmitry Ryvkin†
This version: July 23, 2025
Abstract
We characterize robust tournament design—the prize scheme that maximizes the
lowest effort in a rank-order tournament where the distribution of noise is unknown,
except for an upper bound, H¯ , on its Shannon entropy. The robust tournament
scheme awards positive prizes to all ranks except the last, with a distinct top prize. Because of the entropy constraint, it is very costly
for the adversary to have regions of low density in the (hypothetical) noise distribution. Besides, when the number of agents is small, adjacent order statistics overlap, and having
a zero prize differential does not completely shut down incentives in the corresponding
region.Where the paper says it
Where the paper says it
Conclusion and Implications
Robust tournament design shows that prize sharing can arise as optimal response to limited information on the part of the designer. The harmonic prize structure offers a simple rule of thumb for designing incentives under uncertainty. p. 1 An important
challenge in designing such mechanisms is that the principal often does not know the
characteristics of the noise, which can vary widely across settings. In this paper, we
study robust tournament design, where the prize structure is chosen to maximize the
minimum effort exerted by players, given only an upper bound on the entropy of the
noise distribution. Significant prize inequality is observed across sectors, evidenced by the profiles of CEO compensation (Song et al. , 2019) or
prize profiles in professional sports such as tennis (Gilsdorf and Sukhatme, 2008) or poker
(Levitt and Miles, 2014).Where the paper says it
Where the paper says it
Proofs
The proof of Proposition 2 shows that W(d∗)−W(d∞) is bounded above by a sequence that converges to zero. The proof of Proposition 3 shows that the Gini coefficient equals 1/2 for prizes v∞ in (14). p. 0 Asymptotically, the prizes follow the harmonic number sequence and induce an exponential distribution of noise with rate parameter e
−H¯
. The robust prize scheme
is highly unequal, especially in small tournaments, but becomes more equitable as
the number of participants grows, with the Gini coefficient approaching 1/2. Lemma 2 The optimal Lagrange multiplier in problem (11) is µ
∗ = 1. Proof Consider a version of problem (11) where the budget constraint is parameterized
as Pn−1
r=1 rdr = b, with b > 0. Lemma 2 The optimal Lagrange multiplier in problem (11) is µ
∗ = 1. Proof Consider a version of problem (11) where the budget constraint is parameterized
as Pn−1
r=1 rdr = b, with b > 0. 7This is not a coincidence because f
B(z; n − r, r) is the pdf of order statistic (n − r : n − 1) from the
uniform distribution on [0, 1]. We will use this connection in the next section to derive the asymptotic
robust prize schedule for large n. Lemma 2 The optimal Lagrange multiplier in problem (11) is µ
∗ = 1. Proof Consider a version of problem (11) where the budget constraint is parameterized
as Pn−1
r=1 rdr = b, with b > 0. Asymptotically, the prizes follow the harmonic number sequence and induce an exponential distribution of noise with rate parameter e
−H¯
. The robust prize scheme
is highly unequal, especially in small tournaments, but becomes more equitable as
the number of participants grows, with the Gini coefficient approaching 1/2. Asymptotically, the prizes follow the harmonic number sequence and induce an exponential distribution of noise with rate parameter e
−H¯
. The robust prize scheme
is highly unequal, especially in small tournaments, but becomes more equitable as
the number of participants grows, with the Gini coefficient approaching 1/2. Lemma 2 The optimal Lagrange multiplier in problem (11) is µ
∗ = 1. Proof Consider a version of problem (11) where the budget constraint is parameterized
as Pn−1
r=1 rdr = b, with b > 0.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
References
Bergemann, Dirk, and Stephen Morris. 2005. “Robust mechanism design. p. 3 The asymptotically optimal prize schedule has the Gini coefficient of 1/2,
which corresponds to a substantial but not extreme level of inequality. Related literature Our paper contributes to a growing literature on robust mechanism
design, which seeks to develop mechanisms that perform well under limited or ambiguous
information. , Bergemann and Morris, 2005), available actions (e. , Carroll, 2015),
and other aspects of the environment (see Carroll, 2019, for a review). , Bergemann and Morris, 2005), available actions (e. , Carroll, 2015),
and other aspects of the environment (see Carroll, 2019, for a review). (3)
Since c(·) is strictly convex, (2) and (3) imply that the impact of the distribution of noise
on the equilibrium effort is captured entirely by function m(·). Moreover, the left-hand
side of (2) is a linear functional of m(·), which we further rewrite as
Xn−1
r=1
Br(vr − vr+1) = Z 1
0
a(z; d)m(z) dz, (4)
6 The robust prize scheme
is highly unequal, especially in small tournaments, but becomes more equitable as
the number of participants grows, with the Gini coefficient approaching 1/2. Keywords: robust tournament design, max-min, prize allocation
JEL codes: C72, D82, D86, M52
∗New Economic School, Universitat Autònoma de Barcelona and Barcelona School of Economics
(BSE), and CEPR, mdrugov@nes. The asymptotically optimal prize schedule has the Gini coefficient of 1/2,
which corresponds to a substantial but not extreme level of inequality. Related literature Our paper contributes to a growing literature on robust mechanism
design, which seeks to develop mechanisms that perform well under limited or ambiguous
information. The robust prize scheme
is highly unequal, especially in small tournaments, but becomes more equitable as
the number of participants grows, with the Gini coefficient approaching 1/2. Keywords: robust tournament design, max-min, prize allocation
JEL codes: C72, D82, D86, M52
∗New Economic School, Universitat Autònoma de Barcelona and Barcelona School of Economics
(BSE), and CEPR, mdrugov@nes. , Carroll, 2015),
and other aspects of the environment (see Carroll, 2019, for a review). For example, in the
auction literature, Che (2022) and Suzdaltsev (2022) assume that while the distribution
of values is common knowledge among the bidders, the auction designer has only partial
information about it, such as a moment. , Carroll, 2015),
and other aspects of the environment (see Carroll, 2019, for a review). For example, in the
auction literature, Che (2022) and Suzdaltsev (2022) assume that while the distribution
of values is common knowledge among the bidders, the auction designer has only partial
information about it, such as a moment. , too little or too much inequality in prizes can be
detrimental (for a review see, e. , Chung, DeVaro and Fung, 2025). , 2019) or
prize profiles in professional sports such as tennis (Gilsdorf and Sukhatme, 2008) or poker
(Levitt and Miles, 2014). There is, however, evidence that the impact of pay inequality on
firm performance is hump-shaped, i. There is, however, evidence that the impact of pay inequality on
firm performance is hump-shaped, i. , too little or too much inequality in prizes can be
detrimental (for a review see, e. , Drugov and Ryvkin, 2020b), endogenous noise (Kim, Krishna and
Ryvkin, 2025), and settings with career concerns (e. , Correa and Yildirim, 2024). , Moldovanu and Sela, 2001), tournaments with exogenous noise (e. , Drugov and Ryvkin, 2020b), endogenous noise (Kim, Krishna and
Ryvkin, 2025), and settings with career concerns (e. The robust prize scheme
is highly unequal, especially in small tournaments, but becomes more equitable as
the number of participants grows, with the Gini coefficient approaching 1/2. Keywords: robust tournament design, max-min, prize allocation
JEL codes: C72, D82, D86, M52
∗New Economic School, Universitat Autònoma de Barcelona and Barcelona School of Economics
(BSE), and CEPR, mdrugov@nes. If that happens, the reward is
proportional to the difference between prizes for these ranks. Because the equilibrium is
symmetric, output is effectively ranked the same as the order statistics of noise. The equilibrium For a given vector of efforts x = (x1, . , xn), the expected payoff of
agent i is
π
(i)
(x) = Xn
r=1
p
(i,r)
(x)vr − c(xi),
where p
(i,r)
(x) is agent i’s probability of being ranked r. Robust Tournaments
Mikhail Drugov∗ Dmitry Ryvkin†
This version: July 23, 2025
Abstract
We characterize robust tournament design—the prize scheme that maximizes the
lowest effort in a rank-order tournament where the distribution of noise is unknown,
except for an upper bound, H¯ , on its Shannon entropy. The robust tournament
scheme awards positive prizes to all ranks except the last, with a distinct top prize. Robust Tournaments
Mikhail Drugov∗ Dmitry Ryvkin†
This version: July 23, 2025
Abstract
We characterize robust tournament design—the prize scheme that maximizes the
lowest effort in a rank-order tournament where the distribution of noise is unknown,
except for an upper bound, H¯ , on its Shannon entropy. The robust tournament
scheme awards positive prizes to all ranks except the last, with a distinct top prize. Robust Tournaments
Mikhail Drugov∗ Dmitry Ryvkin†
This version: July 23, 2025
Abstract
We characterize robust tournament design—the prize scheme that maximizes the
lowest effort in a rank-order tournament where the distribution of noise is unknown,
except for an upper bound, H¯ , on its Shannon entropy. The robust tournament
scheme awards positive prizes to all ranks except the last, with a distinct top prize. The robust prize scheme
is highly unequal, especially in small tournaments, but becomes more equitable as
the number of participants grows, with the Gini coefficient approaching 1/2. Keywords: robust tournament design, max-min, prize allocation
JEL codes: C72, D82, D86, M52
∗New Economic School, Universitat Autònoma de Barcelona and Barcelona School of Economics
(BSE), and CEPR, mdrugov@nes. Robust Tournaments
Mikhail Drugov∗ Dmitry Ryvkin†
This version: July 23, 2025
Abstract
We characterize robust tournament design—the prize scheme that maximizes the
lowest effort in a rank-order tournament where the distribution of noise is unknown,
except for an upper bound, H¯ , on its Shannon entropy. The robust tournament
scheme awards positive prizes to all ranks except the last, with a distinct top prize. The robust prize scheme
is highly unequal, especially in small tournaments, but becomes more equitable as
the number of participants grows, with the Gini coefficient approaching 1/2. Keywords: robust tournament design, max-min, prize allocation
JEL codes: C72, D82, D86, M52
∗New Economic School, Universitat Autònoma de Barcelona and Barcelona School of Economics
(BSE), and CEPR, mdrugov@nes. Therefore, the conditions of Sion’s Minimax Theorem (see, e. , Ekeland and
Témam, 1999, Ch. Hence, for a fixed F, the effort-maximizing
prize allocation is one where dr is positive at the “right” place in order to activate the
corresponding part of the distribution. 8 However, in problem (8) the adversary chooses
5Here and below, uniqueness of solutions to variational problems is understood in the L
1
-a. The impact of prize allocation on effort has been studied using various contest models, including complete-information all-pay auctions (e. , Fang, Noe and Strack, 2020),
incomplete-information settings (e. 1 Introduction
Tournaments are used extensively to incentivize effort in environments where output is
noisy but relative performance is observable. Examples include innovation races, the
allocation of bonuses and promotions in firms, sales contests, and sports. Section 5 concludes. 1A natural interpretation of the robust tournament design approach is that of an ambiguity averse
principal with the Maxmin Expected Utility preferences (Gilboa and Schmeidler, 1989), whose set of
possible priors about the distribution of noise is the set of all distributions with entropy at most H¯ . Section 5 concludes. 1A natural interpretation of the robust tournament design approach is that of an ambiguity averse
principal with the Maxmin Expected Utility preferences (Gilboa and Schmeidler, 1989), whose set of
possible priors about the distribution of noise is the set of all distributions with entropy at most H¯ . The robust prize scheme
is highly unequal, especially in small tournaments, but becomes more equitable as
the number of participants grows, with the Gini coefficient approaching 1/2. Keywords: robust tournament design, max-min, prize allocation
JEL codes: C72, D82, D86, M52
∗New Economic School, Universitat Autònoma de Barcelona and Barcelona School of Economics
(BSE), and CEPR, mdrugov@nes. , 2019) or
prize profiles in professional sports such as tennis (Gilsdorf and Sukhatme, 2008) or poker
(Levitt and Miles, 2014). There is, however, evidence that the impact of pay inequality on
firm performance is hump-shaped, i. An important
challenge in designing such mechanisms is that the principal often does not know the
characteristics of the noise, which can vary widely across settings. In this paper, we
study robust tournament design, where the prize structure is chosen to maximize the
minimum effort exerted by players, given only an upper bound on the entropy of the
noise distribution. , Moldovanu and Sela, 2001), tournaments with exogenous noise (e. , Drugov and Ryvkin, 2020b), endogenous noise (Kim, Krishna and
Ryvkin, 2025), and settings with career concerns (e. A central question in tournament design is how to allocate prizes across ranks to maximize effort. A long-standing debate concerns whether a winner-take-all (WTA) scheme,
whereby the entire prize budget is awarded to the top performer, is optimal, or effort is
better incentivized by awarding prizes more broadly. Significant prize inequality is observed across sectors, evidenced by the profiles of CEO compensation (Song et al. , 2019) or
prize profiles in professional sports such as tennis (Gilsdorf and Sukhatme, 2008) or poker
(Levitt and Miles, 2014). , 2019) or
prize profiles in professional sports such as tennis (Gilsdorf and Sukhatme, 2008) or poker
(Levitt and Miles, 2014). There is, however, evidence that the impact of pay inequality on
firm performance is hump-shaped, i. The robust prize scheme
is highly unequal, especially in small tournaments, but becomes more equitable as
the number of participants grows, with the Gini coefficient approaching 1/2. Keywords: robust tournament design, max-min, prize allocation
JEL codes: C72, D82, D86, M52
∗New Economic School, Universitat Autònoma de Barcelona and Barcelona School of Economics
(BSE), and CEPR, mdrugov@nes. , Fang, Noe and Strack, 2020),
incomplete-information settings (e. , Moldovanu and Sela, 2001), tournaments with exogenous noise (e. 1 Introduction
Tournaments are used extensively to incentivize effort in environments where output is
noisy but relative performance is observable. Examples include innovation races, the
allocation of bonuses and promotions in firms, sales contests, and sports. The robust prize scheme
is highly unequal, especially in small tournaments, but becomes more equitable as
the number of participants grows, with the Gini coefficient approaching 1/2. Keywords: robust tournament design, max-min, prize allocation
JEL codes: C72, D82, D86, M52
∗New Economic School, Universitat Autònoma de Barcelona and Barcelona School of Economics
(BSE), and CEPR, mdrugov@nes. In order to be ranked r or higher, one needs to surpass the agent currently ranked
r, and the reward for this is the prize differential vr − vr+1. Quantile representation Let F
−1
(z) = inf{t : F(t) ≥ z} denote the quantile function
of noise, and let m(z) = f(F
−1
(z)) denote the inverse quantile density (Parzen, 1979). For intuition, note
that tournament incentives are generated by prize differentials between adjacent ranks. Indeed, a marginal increase in effort raises the agent’s earnings to the extent it helps
the agent surpass a competitor who is ranked just above. The robust prize scheme
is highly unequal, especially in small tournaments, but becomes more equitable as
the number of participants grows, with the Gini coefficient approaching 1/2. Keywords: robust tournament design, max-min, prize allocation
JEL codes: C72, D82, D86, M52
∗New Economic School, Universitat Autònoma de Barcelona and Barcelona School of Economics
(BSE), and CEPR, mdrugov@nes. A long-standing debate concerns whether a winner-take-all (WTA) scheme,
whereby the entire prize budget is awarded to the top performer, is optimal, or effort is
better incentivized by awarding prizes more broadly. Significant prize inequality is observed across sectors, evidenced by the profiles of CEO compensation (Song et al. A long-standing debate concerns whether a winner-take-all (WTA) scheme,
whereby the entire prize budget is awarded to the top performer, is optimal, or effort is
better incentivized by awarding prizes more broadly. Significant prize inequality is observed across sectors, evidenced by the profiles of CEO compensation (Song et al. The robust prize scheme
is highly unequal, especially in small tournaments, but becomes more equitable as
the number of participants grows, with the Gini coefficient approaching 1/2. Keywords: robust tournament design, max-min, prize allocation
JEL codes: C72, D82, D86, M52
∗New Economic School, Universitat Autònoma de Barcelona and Barcelona School of Economics
(BSE), and CEPR, mdrugov@nes. , Carroll, 2015),
and other aspects of the environment (see Carroll, 2019, for a review). For example, in the
auction literature, Che (2022) and Suzdaltsev (2022) assume that while the distribution
of values is common knowledge among the bidders, the auction designer has only partial
information about it, such as a moment. (7)
Problem (7) can be interpreted as representing a zero-sum game between the principal
choosing d and an adversary choosing m. For such games, under certain conditions, a
version of the minimax theorem implies that the value of the problem is independent of
the order of the max and min operators. The robust prize scheme
is highly unequal, especially in small tournaments, but becomes more equitable as
the number of participants grows, with the Gini coefficient approaching 1/2. Keywords: robust tournament design, max-min, prize allocation
JEL codes: C72, D82, D86, M52
∗New Economic School, Universitat Autònoma de Barcelona and Barcelona School of Economics
(BSE), and CEPR, mdrugov@nes. Examples include innovation races, the
allocation of bonuses and promotions in firms, sales contests, and sports. An important
challenge in designing such mechanisms is that the principal often does not know the
characteristics of the noise, which can vary widely across settings.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
- Robust Prize Allocation System for Uncertainty
The system can design incentive structures for AI agents in environments where performance noise is unknown but bounded by a Shannon entropy H bar. This allows the AI to select a prize scheme that maximizes lowest effort, ensuring guaranteed performance regardless of the true noise distribution, as described by the robust tournament scheme awards positive prizes to all ranks except the last, with a distinct top prize.
- Adaptive Prize Sharing for Large Scale Systems
For large tournaments or complex AI agent teams (large n), the system can implement a prize schedule that asymptotically follows a harmonic number sequence. This structure ensures that as the number of participants grows, the Gini coefficient approaching 1/2,
promoting a substantial but controlled level of inequality to safeguard against adversarial noise exploitation.
- Noise Distribution Learning from Effort Profiles
The system can infer characteristics about the underlying noise distribution F by analyzing equilibrium effort profiles derived from different prize allocations. This is achieved because the impact of the distribution of noise on the equilibrium effort is captured entirely by function m(cdot)
and the resulting limit distribution of noise is exponential,
allowing the AI to estimate H bar from observed performance metrics.
- Small Tournament Strategy Optimization
In small agent groups (n=3 or n=4), the system can switch to a near winner-take-all strategy, as shown by results like "For n = 3, the robust tournament awards distinct prizes v1 > v2 > v3 = 0." This ensures that for smaller teams, maximum incentive is focused on the top performer even under uncertainty.
- Risk Management via Prize Differential Tuning
The system can dynamically adjust prize differentials (d r) based on a pre-defined set of incentives to minimize equilibrium effort against an adversarial noise distribution. This involves solving the maximization problem: max d in D min m in M Z 1 0 a(z; d)m(z) dz,
effectively choosing which regions of noise to incentivize or suppress.
Abstract
We characterize robust tournament design---the prize scheme that maximizes the lowest equilibrium effort in a rank-order tournament where the distribution of noise is unknown, except for an upper bound, H bar, on its Shannon entropy. The robust tournament awards positive prizes to all ranks except the last, with a distinct top prize. Robust prizes spread marginal incentives as evenly as possible across performance quantiles, minimizing the Kullback--Leibler divergence from the uniform distribution to the induced distribution of performance cutoffs. Asymptotically, robust prizes follow the harmonic number sequence, while the worst-case distribution of noise converges to an exponential distribution with rate e 1-H bar. The prize scheme is highly unequal, especially in small tournaments, but becomes more equitable as the number of agents grows, with the Gini coefficient approaching 1/2. In the large tournament model, logarithmic prizes and an exponential worst-case distribution also arise under a bound on expected noise above the lower endpoint of its support, and more generally under bounds on a class of means of the quantile hazard.