pith. sign in

arxiv: 2605.19302 · v1 · pith:MY73L7ICnew · submitted 2026-05-19 · 💻 cs.GT

Distributionally Robust Games via Coherent Risk Measures

Bharat Gangwani , Arunesh Sinha This is my paper

Pith reviewed 2026-05-20 02:49 UTC · model grok-4.3

classification 💻 cs.GT
keywords distributionally robust gamescoherent risk measuresequilibrium existencePPAD-completenessdata-driven gamesambiguity setsrisk-aware equilibria
0
0 comments X

The pith

Coherent risk measures define distributionally robust games that admit equilibria in data-driven settings with finite payoff samples.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that when players in a strategic game face uncertainty about payoff distributions learned from finite data samples, coherent risk measures can be used to define distributionally robust versions of the game. Prior existence theorems for such robust games then guarantee that equilibria exist for several common ambiguity sets constructed from the samples. The resulting games turn out to be continuous rather than finite matrix games, which changes their structure, and equilibrium computation is shown to be PPAD-complete in general while remaining in PPAD for specific measures such as Conditional Value-at-Risk. Bounds are derived on the utility penalty a player incurs by being risk-averse, and multilinear complementarity programs are given as solution methods. This provides a way to incorporate calibrated risk attitudes directly into equilibrium analysis instead of assuming known probability distributions.

Core claim

By treating coherent utility measures as a primitive of player preferences and substituting them into distributionally robust payoff functions whose ambiguity sets are defined from finite samples, the paper shows existence of equilibria follows from prior results, the games are inherently continuous, a bound holds on expected-utility loss from risk aversion, and equilibrium computation is PPAD-complete in general with membership in PPAD and multilinear complementarity formulations for several concrete coherent measures.

What carries the argument

Coherent risk measures (such as Mean-semideviation and Conditional Value-at-Risk) substituted into the payoffs of distributionally robust games whose ambiguity sets are built from finite samples.

If this is right

  • Equilibria exist for various ambiguity sets in data-driven games once coherent risk measures define the robust payoffs.
  • The games are continuous rather than finite matrix games, which fundamentally changes equilibrium structure and blocks direct use of standard correlated-equilibrium notions.
  • A bound holds on the loss in expected utility a player suffers by adopting a risk-averse attitude.
  • Equilibrium computation is PPAD-complete in general and lies in PPAD for several specific coherent utility measure games.
  • Multilinear complementarity programs provide a practical formulation for computing equilibria in these games.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same substitution technique could be tested on repeated or dynamic games to see whether risk attitudes produce qualitatively different long-run behavior.
  • Numerical out-of-sample robustness observed in the experiments suggests the equilibria may remain stable when new samples arrive after the ambiguity sets are fixed.
  • The continuity of the games implies that standard discrete solution concepts may need continuous analogs when risk measures are present.

Load-bearing premise

That prior existence theorems for distributionally robust equilibria continue to apply after coherent risk measures replace ordinary expected utility and ambiguity sets are formed from finite samples.

What would settle it

A concrete finite-sample game instance using a coherent risk measure such as CVaR for which no equilibrium exists, or a polynomial-time algorithm for the general case, would refute the existence and complexity claims.

Figures

Figures reproduced from arXiv: 2605.19302 by Arunesh Sinha, Bharat Gangwani.

Figure 1
Figure 1. Figure 1: The payoff for DRG (left) and mean-semideviation game (right). [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: MSD-Coordination game payoffs with ˆp = 0.6 and different risk aversion levels. · − ·− lines mark the mixed Nash equilibria whenever either player plays xi = 0.5 and ⋆ denote the pure Nash equilibria of the game. Notice that the equilibria set shrinks in size and becomes more robust to variance and skew in the emprical distribution at higher γs. demonstrate similar properties. In this example, our true pay… view at source ↗
Figure 3
Figure 3. Figure 3: OOS-mean payoff ±1 s.d. of the two players as a function of γs in the Small-K Prisoner’s Dilemma. OOS performance increases with γs due to greater robustness to sampling variation. we find equilibrium strategies x ∗ , y ∗ in a closed-form which are plotted in [PITH_FULL_IMAGE:figures/full_fig_p017_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Player 2’s strategy y ∗ 1 and player 1’s risk-adjusted payoff ρ1(x ∗ , y ∗ ), VaR, z ∗ 1 , and payoff variance V[u1(x ∗ , y ∗ | ξ)] as a function of α and γc in equilibrium. Player 1 experiences a regime change at α = 0.5. V[u1(x ∗ , y ∗ | ξ)] monotonically decreases in γc. these settings, this paper introduces tractable formulations of one-shot, simultaneous move DRG by using coherent utility measures wit… view at source ↗
read the original abstract

We study strategic interaction in data-driven games where players face uncertainty about payoff distributions inferred from finite samples. To model calibrated attitudes toward such uncertainty, we formulate distributionally robust games with a special focus on coherent utility (risk) measures, including Mean-semideviation and Conditional Value-at-Risk. This framework treats risk sensitivity as a primitive feature of player preferences while retaining a formal connection to distributional robustness. We make a number of contributions that are enumerated next. (1) We use prior results for the existence of distributionally robust equilibria to show the existence of equilibria in data-driven settings for various ambiguity sets, and (2) show that these games are inherently continuous, rather than finite matrix games, which fundamentally alters equilibrium structure and precludes direct extensions of standard correlated equilibrium notions. (3) We bound the loss in expected utility that a player can expect from being risk-averse. (4) We further characterize the computational complexity of equilibrium computation, proving PPAD-completeness in general and PPAD membership for several coherent utility measure games. (5) We present multilinear complementarity program formulations for several coherent utility measure games. (6) Numerical experiments reveal the robustness and out of sample performance of the game solutions. Our results unify risk-theoretic modeling and equilibrium analysis, providing a principled foundation for risk-aware strategic decision-making in data-driven environments.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The manuscript studies distributionally robust games in data-driven settings where players employ coherent risk measures (Mean-semideviation and Conditional Value-at-Risk) to express calibrated attitudes toward uncertainty in payoff distributions inferred from finite samples. It claims to (1) establish existence of equilibria by invoking prior DRO equilibrium results after substitution of coherent risk measures and construction of ambiguity sets from samples, (2) show that the resulting games are inherently continuous (altering equilibrium structure and precluding direct extension of correlated equilibrium notions), (3) bound the loss in expected utility from risk aversion, (4) prove PPAD-completeness of equilibrium computation in general with PPAD membership for several coherent utility games, (5) give multilinear complementarity program formulations, and (6) present numerical experiments on robustness and out-of-sample performance.

Significance. If the existence claim and complexity results hold after verification of the requisite regularity conditions, the work would usefully connect coherent risk measures with equilibrium analysis, supplying both theoretical foundations and concrete computational tools (formulations and complexity classification) for risk-aware strategic decision-making under distributional uncertainty. The numerical component adds practical grounding.

major comments (1)
  1. [Abstract, contribution (1)] Abstract, contribution (1): The existence of equilibria in the data-driven setting is obtained by direct substitution of coherent risk measures into prior DRO equilibrium theorems together with finite-sample ambiguity sets. Prior theorems of this type require the effective (risk-adjusted) payoff to remain continuous or upper semi-continuous in the joint mixed-strategy profile for every distribution in the ambiguity set, as well as suitable compactness of the ambiguity set in a compatible topology. The manuscript provides no explicit verification that these conditions are inherited when the underlying payoff is only measurable with respect to an empirical measure supported on finitely many points; this step is load-bearing for the central existence claim.
minor comments (2)
  1. The specific ambiguity sets (e.g., Wasserstein balls or moment-based sets) employed for the existence and complexity results should be stated explicitly in the main text with references to their convexity and compactness properties.
  2. Notation for the risk-adjusted payoff functions and the mapping from empirical samples to ambiguity sets could be introduced earlier and used consistently to improve readability of the formulations in Section 4.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful and constructive review. The major comment correctly identifies a point where the manuscript's invocation of prior results would benefit from more explicit verification of the requisite regularity conditions. We address this below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract, contribution (1)] Abstract, contribution (1): The existence of equilibria in the data-driven setting is obtained by direct substitution of coherent risk measures into prior DRO equilibrium theorems together with finite-sample ambiguity sets. Prior theorems of this type require the effective (risk-adjusted) payoff to remain continuous or upper semi-continuous in the joint mixed-strategy profile for every distribution in the ambiguity set, as well as suitable compactness of the ambiguity set in a compatible topology. The manuscript provides no explicit verification that these conditions are inherited when the underlying payoff is only measurable with respect to an empirical measure supported on finitely many points; this step is load-bearing for the central existence claim.

    Authors: We thank the referee for highlighting this important technical point. The existence claim in contribution (1) does rely on direct substitution into prior DRO equilibrium results, and the manuscript does not contain an explicit paragraph verifying that continuity (or upper semi-continuity) of the risk-adjusted payoffs and compactness of the ambiguity sets are preserved under finite-sample constructions. In the data-driven setting the ambiguity sets are balls (e.g., Wasserstein or moment-based) centered at the empirical measure, which is supported on finitely many points; such sets are compact in the weak topology. Coherent risk measures such as CVaR and mean-semideviation are continuous with respect to weak convergence on compact supports, and the finite-support structure ensures that the map from mixed-strategy profiles to the induced payoff distributions is continuous. Consequently the effective (risk-adjusted) payoffs inherit the required regularity. We agree that this reasoning should be stated explicitly rather than left implicit and will add a dedicated verification paragraph (with appropriate citations to continuity properties of coherent risk measures) in the revised manuscript. revision: yes

Circularity Check

0 steps flagged

Existence via external prior results; no reduction to self-inputs by construction

full rationale

The paper's central existence claim (contribution 1) invokes prior results on distributionally robust equilibria after substituting coherent risk measures and defining ambiguity sets from samples. This is an application of external theorems rather than a self-definitional loop or a fitted parameter renamed as a prediction. No equations or steps in the abstract reduce the claimed equilibria to the paper's own fitted values or unverified self-citations by construction. Computational complexity (PPAD) and formulation contributions are separate and do not rely on the existence step for their validity. The derivation remains self-contained against the cited external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Central claims rest on prior existence theorems for distributionally robust equilibria and standard assumptions from game theory and coherent risk measures; no new free parameters or invented entities are introduced in the abstract.

axioms (1)
  • domain assumption Prior results establish existence of distributionally robust equilibria for the ambiguity sets considered
    Abstract contribution (1) explicitly invokes these prior results to obtain existence in the data-driven case.

pith-pipeline@v0.9.0 · 5762 in / 1224 out tokens · 44123 ms · 2026-05-20T02:49:39.424903+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

34 extracted references · 34 canonical work pages

  1. [1]

    Mathematical finance , volume=

    Coherent measures of risk , author=. Mathematical finance , volume=. 1999 , publisher=

    work page 1999

  2. [2]

    Mathematics of operations research , volume=

    Optimization of convex risk functions , author=. Mathematics of operations research , volume=. 2006 , publisher=

    work page 2006

  3. [3]

    SIAM Journal on Computing , volume=

    On the complexity of Nash equilibria and other fixed points , author=. SIAM Journal on Computing , volume=. 2010 , publisher=

    work page 2010

  4. [4]

    2022 , publisher=

    Fearnley, John and Goldberg, Paul and Hollender, Alexandros and Savani, Rahul , journal=. 2022 , publisher=

    work page 2022

  5. [5]

    Journal of mathematical Economics , volume=

    Subjectivity and correlation in randomized strategies , author=. Journal of mathematical Economics , volume=. 1974 , publisher=

    work page 1974

  6. [6]

    Learning Theory and Kernel Machines: 16th Annual Conference on Learning Theory and 7th Kernel Workshop, COLT/Kernel 2003, Washington, DC, USA, August 24-27, 2003

    A general class of no-regret learning algorithms and game-theoretic equilibria , author=. Learning Theory and Kernel Machines: 16th Annual Conference on Learning Theory and 7th Kernel Workshop, COLT/Kernel 2003, Washington, DC, USA, August 24-27, 2003. Proceedings , pages=. 2003 , organization=

    work page 2003

  7. [7]

    The Annals of Probability , volume=

    The Tight Constant in the Dvoretzky-Kiefer-Wolfowitz Inequality , author=. The Annals of Probability , volume=. 1990 , publisher=

    work page 1990

  8. [8]

    A Data-Driven Distributionally Robust Game Using Wasserstein Distance

    Peng, Guanze and Zhang, Tao and Zhu, Quanyan. A Data-Driven Distributionally Robust Game Using Wasserstein Distance. Decision and Game Theory for Security. 2020

    work page 2020

  9. [9]

    Computational Optimization and Applications , author=

    Interfaces to path 3.0: Design, implementation and usage , volume=. Computational Optimization and Applications , author=. 1999 , month=. doi:10.1023/a:1008636318275 , number=

    work page doi:10.1023/a:1008636318275 1999

  10. [10]

    Existence of Correlated Equilibria , urldate =

    Sergiu Hart and David Schmeidler , journal =. Existence of Correlated Equilibria , urldate =

    work page

  11. [11]

    Proceedings of the 24th ACM Conference on Economics and Computation , pages=

    The Computational Complexity of Multi-player Concave Games and Kakutani Fixed Points , author=. Proceedings of the 24th ACM Conference on Economics and Computation , pages=

    work page

  12. [12]

    2015 , eprint=

    Distributionally Robust Game Theory , author=. 2015 , eprint=

    work page 2015

  13. [13]

    Proceedings of the 11th EAI International Conference on Performance Evaluation Methodologies and Tools , pages =

    Bauso, Dario and Gao, Jian and Tembine, Hamidou , title =. Proceedings of the 11th EAI International Conference on Performance Evaluation Methodologies and Tools , pages =. 2017 , isbn =. doi:10.1145/3150928.3150950 , abstract =

    work page doi:10.1145/3150928.3150950 2017

  14. [14]

    Distributionally robust chance-constrained games: existence and characterization of Nash equilibrium

    Singh, Vikas Vikram and Jouini, Oualid and Lisser, Abdel. Distributionally robust chance-constrained games: existence and characterization of Nash equilibrium. Optim. Lett

    work page

  15. [15]

    7 (1) , author=

    Portfolio Selection, the journal of finance. 7 (1) , author=. N , volume=

    work page

  16. [16]

    The review of financial studies , volume=

    Measuring systemic risk , author=. The review of financial studies , volume=. 2017 , publisher=

    work page 2017

  17. [17]

    2024 , publisher=

    Risk-averse optimization and control , author=. 2024 , publisher=

    work page 2024

  18. [18]

    The Eleventh International Conference on Learning Representations , year=

    Risk-aware reinforcement learning with coherent risk measures and non-linear function approximation , author=. The Eleventh International Conference on Learning Representations , year=

    work page

  19. [19]

    Robust game theory

    Aghassi, Michele and Bertsimas, Dimitris. Robust game theory. Math. Program

    work page

  20. [20]

    2025 , eprint=

    Strategically Robust Game Theory via Optimal Transport , author=. 2025 , eprint=

    work page 2025

  21. [21]

    Games and Economic Behavior , volume=

    Correlated equilibria in continuous games: Characterization and computation , author=. Games and Economic Behavior , volume=. 2011 , publisher=

    work page 2011

  22. [22]

    Distributionally robust equilibrium for continuous games: Nash and Stackelberg models

    Liu, Yongchao and Xu, Huifu and Yang, Shu-Jung Sunny and Zhang, Jin. Distributionally robust equilibrium for continuous games: Nash and Stackelberg models. Eur. J. Oper. Res

    work page

  23. [23]

    2022 , eprint=

    Computing the optimal distributionally-robust strategy to commit to , author=. 2022 , eprint=

    work page 2022

  24. [24]

    2025 , eprint=

    Bayesian Distributionally Robust Nash Equilibrium and Its Application , author=. 2025 , eprint=

    work page 2025

  25. [25]

    Journal of Machine Learning Research , volume=

    Variance-based regularization with convex objectives , author=. Journal of Machine Learning Research , volume=

    work page

  26. [26]

    Journal of Artificial Intelligence Research , volume=

    Empirical game theoretic analysis: A survey , author=. Journal of Artificial Intelligence Research , volume=

    work page

  27. [27]

    Finance and Stochastics , volume=

    Coherent and convex monetary risk measures for unbounded cadlag processes , author=. Finance and Stochastics , volume=. 2006 , publisher=

    work page 2006

  28. [28]

    Mathematics of Operations Research , volume=

    Robust dynamic programming , author=. Mathematics of Operations Research , volume=. 2005 , publisher=

    work page 2005

  29. [29]

    2023 , eprint=

    Double Pessimism is Provably Efficient for Distributionally Robust Offline Reinforcement Learning: Generic Algorithm and Robust Partial Coverage , author=. 2023 , eprint=

    work page 2023

  30. [30]

    Rectangularity and Duality of Distributionally Robust Markov Decision Processes: Y. Li, A. Shapiro , author=. Mathematical Programming , pages=. 2025 , publisher=

    work page 2025

  31. [31]

    Mathematical programming , volume=

    Risk-averse dynamic programming for Markov decision processes , author=. Mathematical programming , volume=. 2010 , publisher=

    work page 2010

  32. [32]

    Journal of Intelligent & Fuzzy Systems , volume=

    Distributionally robust games with an application to supply chain , author=. Journal of Intelligent & Fuzzy Systems , volume=. 2017 , publisher=

    work page 2017

  33. [33]

    Transportation Research Part B: Methodological , volume=

    Beyond normality: A cross moment-stochastic user equilibrium model , author=. Transportation Research Part B: Methodological , volume=. 2015 , publisher=

    work page 2015

  34. [34]

    Convergence Analysis for Distributionally Robust Optimization and Equilibrium Problems , author=. Math. Oper. Res. , year=

    work page