Combinatorial Contract Design: Recent Progress and Emerging Frontiers

TL;DR

This survey addresses the computational challenges of combinatorial contracts in principal–agent settings, where action spaces and team configurations can be exponentially large. It develops an algorithmic framework built on value and demand oracles, the notion of critical values, and structured reward classes (e.g., GS, ultra, submodular, XOS) to map tractability frontiers across three settings: (i) single-agent multi-actions, (ii) multi-agent binary-actions, and (iii) multi-agent multi-actions. The results include poly-time algorithms for optimal contracts in GS, ultra, and certain supermodular reward regimes; FPTAS for monotone rewards with demand oracle access; and constant-factor approximations in multi-agent binary-action and multi-agent multi-action settings, alongside hardness results showing NP-hardness or inapproximability for submodular and XOS rewards under standard oracle models. The work identifies key structural notions (e.g., critical values, pseudo-dimension, subset stability) that govern both computation and learnability, and outlines open problems such as extending frontiers to XOS rewards, richer outcomes, budgets, and learning under partial information, with relevance to online platforms, AI delegation, and distributed protocols.

Abstract

Contract theory studies how a principal can incentivize agents to exert costly, unobservable effort through performance-based payments. While classical economic models provide elegant characterizations of optimal solutions, modern applications, ranging from online labor markets and healthcare to AI delegation and blockchain protocols, call for an algorithmic perspective. The challenge is no longer only which contracts induce desired behavior, but whether such contracts can be computed efficiently. This viewpoint has given rise to \emph{algorithmic contract design}, paralleling the rise of algorithmic mechanism design two decades ago. This article focuses on \emph{combinatorial contracts}, an emerging frontier within algorithmic contract design, where agents may choose among exponentially many combinations of actions, or where multiple agents must work together as a team, and the challenge lies in selecting the right composition. These models capture a wide variety of real-world contracting environments, from hospitals coordinating physicians across treatment protocols to firms hiring teams of engineers for interdependent tasks. We review three combinatorial settings: (i) a single agent choosing multiple actions, (ii) multiple agents with binary actions, and (iii) multiple agents each selecting multiple actions. For each, we highlight structural insights, algorithmic techniques, and complexity barriers. Results include tractable cases such as gross substitutes reward functions, hardness results, and approximation guarantees under value- and demand-oracle access. By charting these advances, the article maps the emerging landscape of combinatorial contract design, and highlights fundamental open questions and promising directions for future work.

Figures (5)

• Figure 1: The agent's and principal's utilities as a function of $\alpha$ for an additive reward function, with $f({1}) = 0.3$, $f({2}) = 0.2$, and $f({3}) = 0.5$, and action costs $c1 = c2 = 0.1, c3 = 0.4$. Left: Upper envelope of the agent's utility. Right: The principal's utility for the agent's best response.
• Figure 2: Enumerate CV$(L,R)$
• Figure 3: Upper envelope for Example \ref{['ex:gs']}, with GS $f$
• Figure 4: Three 2-action instances. In all examples $c(1)=0.012, c(2)=0.02$, and they differ only in $f$. Left: $f(\emptyset) = 0, f(\{1\}) = 2, f(\{2\})=3, f(\{1,2\})=4$. Middle: $f(\emptyset) = 0, f(\{1\}) = 2, f(\{2\})=2, f(\{1,2\})=4$. Right: $f(\emptyset) = 0, f(\{1\}) = 1, f(\{2\})=1, f(\{1,2\})=4$.
• Figure 5: An example of an XOS reward function $f$ as a function of $|S|$, and the induced principal's utility $g$ under the best contract incentivizing $S$.

Theorems & Definitions (41)

• Definition 2.1
• Theorem 2.2: GS characterization PaesLeme17
• Lemma 3.1: Monotonicity DuttingRT19DuettingEFK21
• Example 3.2: 3-action instance with additive reward DuttingFT24-survey
• Proposition 3.3: enumerating all critical values DuettingFG23
• proof : Proof sketch
• Theorem 3.4: Optimal contract algorithm DuettingEFK21DuettingFG23
• proof
• Example 3.5: 3-action instance with GS reward
• Lemma 3.6