Welfare and Beyond in Multi-Agent Contracts

TL;DR

This work investigates welfare-focused contract design in multi-agent settings, revealing that for value functions in the XOS and submodular classes, the optimal principal utility approximates social welfare within a constant factor under feasibility constraints. It develops a suite of polynomial-time algorithms that either approximate welfare directly (notably a 2+o(1) guarantee for symmetric XOS) or translate utility-based guarantees into welfare guarantees with bounded loss, and it extends these results to generalized feasibility constraints including budgeted welfare and transfer limits. The paper also establishes tight upper bounds on welfare-utility gaps (e.g., 16+o(1) for sXOS and 5 for submodular) and fundamental lower bounds showing gaps can be unbounded beyond XOS, underscoring the boundary of constant-factor guarantees. Beyond welfare, it provides constant-factor approximations for maximizing the project value under feasibility and transfer constraints, with results extending to symmetric and budgeted settings. Collectively, the results demonstrate that contract-based coordination can achieve near-optimal social welfare in meaningful function classes while clarifying the computational limits and offering practical approximation methods for welfare and value objectives.

Abstract

A principal delegates a project to a team $S$ from a pool of $n$ agents. The project's value if all agents in $S$ exert costly effort is $f(S)$. To incentivize the agents to participate, the principal assigns each agent $i\in S$ a share $ρ_i\in [0,1]$ of the project's final value (i.e., designs $n$ linear contracts). The shares must be feasible -- their sum should not exceed $1$. It is well-understood how to design these contracts to maximize the principal's own expected utility, but what if the goal is to coordinate the agents toward maximizing social welfare? We initiate a systematic study of multi-agent contract design with objectives beyond principal's utility, including welfare maximization, for various classes of value functions $f$. Our exploration reveals an arguably surprising fact: If $f$ is up to XOS in the complement-free hierarchy of functions, then the optimal principal's utility is a constant-fraction of the optimal welfare. This is in stark contrast to the much larger welfare-utility gaps in auction design, and no longer holds above XOS in the hierarchy, where the gap can be unbounded. A constant bound on the welfare-utility gap immediately implies that existing algorithms for designing contracts with approximately-optimal principal's utility also guarantee approximately-optimal welfare. The downside of reducing welfare to utility is the loss of large constants. To obtain better guarantees, we develop polynomial-time algorithms directly for welfare, for different classes of value functions. These include a tight $2$-approximation to the optimal welfare for symmetric XOS functions. Finally, we extend our analysis beyond welfare to the project's value under general feasibility constraints. Our results immediately translate to budgeted welfare and utility.

Theorems & Definitions (81)

• Definition 2.1: Welfare-utility gap
• Theorem 3.1
• Remark 1
• Lemma 3.1
• proof : Proof of \ref{['3']}
• proof : Proof of \ref{['sXOS:lem:algorithm']}
• Proposition 3.1
• Lemma 3.2
• proof
• proof : Proof of \ref{['submodular:UB']}