Contracting a crowd of heterogeneous agents

Guillermo Alonso Alvarez, Erhan Bayraktar, Ibrahim Ekren · 2025

Read Contracting a crowd of heterogeneous agents by Guillermo Alonso Alvarez, Erhan Bayraktar, Ibrahim Ekren (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-09415.txt.

Introduction

Guillermo Alonso Alvarez, Erhan Bayraktar, and Ibrahim Ekren solve a contract design problem for a large crowd of differing agents. Their efforts spill onto each other through a network, in a paper from 2025. Think of teams where one worker's output lifts or drags a neighbor's. A single contract template must still coax the right effort from each person.

Working in a tractable linear quadratic setup, the authors solve the finite crowd problem exactly. They solve its endless crowd limit exactly as well. The limit contract, applied to a large real sample, stays valid and loses almost nothing as the crowd grows. Results stay stable when the interaction pattern is slightly off. Examples show how network position should shape incentives: bigger spillovers earn stronger terms.

Catalog record

CREATORGuillermo Alonso Alvarez, Erhan Bayraktar, Ibrahim Ekren
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE78 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-09415

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Contracting a crowd of heterogeneous agents Thanks: E. Bayraktar is partially supported by the NSF grant DMS-2507940, and the Susan M. Smith Chair. I. Ekren is partially supported by the NSF grant DMS-2406240. Guillermo Alonso Alvarez, Erhan Bayraktar, Ibrahim Ekren Abstract We study a principal-agent model involving a large population of heterogeneously interacting agents. By extending the existing methods, we find the optimal contracts assuming a continuum of agents, and show that, when the number of agents is sufficiently large, the optimal contracts for the problem with a continuum of agents are near-optimal for the finite agents problem. We make comparative statistics and provide numerical simulations to analyze how the agents’ connectivity affects the principal’s value, the effort of the agents, and the optimal contracts. 1 Introduction Principal-agent models describe the problem of finding optimal incentives in delegated tasks. A principal, usually playing the role of a higher authority, delegates a project to an agent who, in exchange, receives a payment from the principal (i.e., the contract). In many situations, the nature of the task prevents the principal from observing the effort that the agent is putting into the project. This lack of transparency makes finding an optimal contract a challenging problem. The asymmetry of information between the two parties, known as moral hazard, is a well-studied phenomenon by economists. The first paper on this matter in continuous time is written by Holmström and Milgrom [ HM87 ] , who show that, when the agent maximizes the expected exponential utility of his payoff, it is optimal for the principal to offer a linear contract. Two decades after the finding of Holmström and Milgrom, Sannikov [ San08 ] proposed a new model in continuous time where the contracts consist of a continuous incentive and a terminal lump-sum payment transacted at a time established by the principal. In addition to the extensive modeling contributions, Sannikov introduced a new method to approach principal-agent problems based on reducing them to a stochastic control framework, allowing the use of dynamic programming techniques. Thereafter, by combining Sannikov’s approach and the use of backward stochastic differential equations (BSDE), Cvitanic-Possamaï-Touzi [ CPT18 ] provided a general method connecting principal-agent problems to stochastic control theory. The approach of Cvitanic et al. has set the basis for multiple extensions. We highlight the work by Elie and Possamaï [ ÉP19 ] in which the authors study a contracting problem between a Principal and multiple agents in competition. On the other hand, mean-field game theory studies stochastic games involving a large number of agents. The main idea is that, when the number of agents is large, the individual actions of every agent have a negligible impact on the stochastic environment. This fact justifies heuristically that the N -player game can be approximated only by considering the interactions between a representative agent and the population distribution. The previous argument was rigorously proved by Lasry and Lions [ LL06a , LL06b , LL07 ] , and Huang, Malhamé and Caines [ HMC06 , HCM07 ] independently by using partial differential equations (PDEs). Since then, the theory of mean-field games has flourished, producing a large number of theoretical and applied results. A crucial limitation of standard mean-field games is that it is assumed that the agents are symmetric. This situation does not hold in many realistic applications. To address this issue, Parise and Ozdaglar [ PO19 ] , and Caines and Huang [ CH21 ] introduced the notion of graphon mean-field games. In their model, a continuum of heterogeneous agents interact via a graphon, a function G:[0,1]^{2}\mapsto\mathbb{R} representing the connectivity of the agents. The articles [ ACL22 ] , [ BW22 ] , [ BWZ23 ] , [ BCW23 ] , [ PO23 ] , [ BK24 ] , [ BZ25 ] , [ CCP24 ] , [ CdFP25 ] , [ BHK25 ] extend the results from Caines and Huang [ CH21 ] , and Parise and Ozdaglar [ PO19 ] . We emphasize that working with a continuum of state and noise processes is technically challenging, and most of the results from mean-field games cannot be generalized directly to graphon mean-field games. Building on this idea, Lacker and Soret [ LS22 ] developed a model in which, instead of working with a continuum of state processes, they consider a representative agent with a distributional type and a population distribution representing the joint law of the agents’ state and type. The main advantage of their formulation is to reduce a graphon mean-field game to a standard mean-field game. The first paper connecting principal-agent problems with mean-field games is authored by Élie, Mastrolia, and Possamaï [ ÉMP19 ] . The authors study a contracting problem between a principal and infinitely many homogeneous agents in competition via a standard mean-field game. Employing a weak-formulation approach similar to Carmona and Lacker [ CL15 ] , they reduce the principal’s problem to a mean-field control problem from which they can apply dynamic programming techniques, for example [ PW18 ] , [ BCP18 ] . We mention the following contributions extending the model by Élie et al. [ BCZ19 ] , [ EHMP20 ] , [ BZ21 ] , [ Dje23 ] , [ DL23 ] , [ MZ25 ] , [ ABDJ22 ] . We note that in all the works mentioned, it is assumed that the agents interact homogeneously. The latter is crucial, as the previous works are built upon mean-field game theory. In most practical applications, agents interact in a heterogeneous manner. From this perspective, it is essential to consider the agents’ connectivity when designing optimal incentives, which may need to be tailored to the specific characteristics of each agent. In this scenario, we derive a model with a principal and multiple agents in charge of heterogeneously dependent projects. We describe the interdependence of the agents through an interaction function G:[0,1]^{2}\mapsto\mathbb{R} indicating the influence of the agents on each other. We present two versions of the model: a version with N agents and a version with a continuum of agents. In each case, we compute the respective optimal contracts and the equilibrium response efforts of the agents (Theorem 1 , Theorem 2 ). Furthermore, we establish that the N agents problem converges to the problem with a continuum of agents as N\rightarrow\infty in the following sense: i) The agents’ Nash equilibrium efforts converge as N\rightarrow\infty . Theorem 6 . ii) The optimal contracts and the principal’s value converge as N\rightarrow\infty . Theorem 7 . iii) The optimal contract for a continuum of agents generates a near-optimal contract for the N agents problem when N is large enough. By near-optimal, we mean that a) the contract satisfies the agents’ reservation constraints; b) the contract induces a unique Nash equilibrium effort in the N agents problem; c) the contract is \mathcal{O}(1/N) -optimal from the principal’s viewpoint. Theorem 8 . Moreover, we show that the optimal contracts are designed to incentivize more influential agents to exert greater effort (Theorem 3 ). In addition, we show that the optimal contracts and agents’ optimal efforts are stable to perturbations in the agents’ connectivity (Theorem 4 , Theorem 5 ). To solve the version of our model with a continuum of agents

Read the complete text