Incentive Design with Spillovers

A principal uses payments conditioned on stochastic outcomes of a team project to elicit costly effort from the team members. We develop a multi-agent generalization of a classic first-order approach to contract optimization by leveraging methods from network games. The main results characterize the optimal allocation of incentive pay across agents and outcomes. Incentive optimality requires equalizing, across agents, a product of (i) individual productivity (ii) organizational centrality and (iii) responsiveness to monetary incentives. We specialize the model to explore several applied questions, including whether compensation should reward individual ability or collaborativeness and how the strength of complementarities shapes pay dispersion.

💡 Research Summary

The paper studies optimal incentive contracts in a team setting where the outcomes of a joint project are stochastic and only observable at the level of the project’s final results. A principal can commit to non‑negative payments to each agent that are contingent on the realized outcome, but cannot condition on individual effort or on the unobservable team performance. Each agent chooses a costly effort level, incurring a convex cost, and enjoys a concave, increasing utility of money. The team’s performance is a smooth, strictly increasing function of all agents’ efforts, and the probability distribution over observable outcomes is a twice‑differentiable function of this performance.

The authors extend the classic Holmström (1979) single‑agent moral‑hazard model to a multi‑agent environment by exploiting the structure of the Hessian matrix of the performance function. The off‑diagonal entries of the Hessian capture how a marginal change in one agent’s effort affects the marginal returns to another agent’s effort; these cross‑partials are interpreted as “spillover effects.” By normalizing rows (or columns) of the Hessian they construct a “spillover network” and define each agent’s centrality as a measure of how strongly the agent is linked, directly or indirectly, to productive peers.

The core theoretical contribution is a balance condition that characterizes optimal contracts when the binding incentive constraints are local. For every agent i who receives a positive payment, the product

  ( marginal productivity ) × ( network centrality ) × ( marginal utility of money )

must be equal across all such agents. Formally,

  ∂Y/∂aᵢ · cᵢ · uᵢ′(·) = λ

for some constant λ common to all incentivized agents. Here ∂Y/∂aᵢ is the partial derivative of team performance with respect to i’s effort, cᵢ is the centrality derived from the Hessian, and uᵢ′ is the derivative of the agent’s utility function. This condition simultaneously accounts for (i) the direct contribution of effort, (ii) the indirect effect of that effort on peers through spillovers, and (iii) the agent’s responsiveness to monetary incentives. The result holds under limited liability and risk‑averse agents, and it requires no parametric assumptions about the production function beyond smoothness.

Two concrete applications illustrate the condition’s implications. First, when each agent’s effort complements others according to a single scalar parameter βᵢ, the centrality reduces to a monotone function of βᵢ. The optimal pay formula then blends an agent’s own productivity with his complementarity parameter. As βᵢ rises, the optimal contract shifts from rewarding “productivity‑driven” agents to rewarding “complementarity‑driven” agents, sometimes dramatically. Numerical examples show that ignoring spillovers can forfeit a large fraction of the principal’s potential profit.

Second, the authors examine a setting where agents have identical standalone productivities but are embedded in an arbitrary complementarity network. By solving for the equilibrium centralities of the spillover network, they obtain closed‑form expressions for optimal payments. Interestingly, centrality is not monotone in the complementarity parameters: strengthening complementarities can sometimes reduce an agent’s network position, leading to lower optimal pay. This non‑monotonicity stems from the fact that the spillover network itself depends on the equilibrium contract, creating a feedback loop between incentives and network structure.

Robustness is addressed by analyzing measurement error in the estimated spillover matrix. The paper links the principal’s ability to improve profits despite noisy estimates to the spectral gap of the spillover network (the difference between the first and second eigenvalues). When the gap is large—i.e., the network is well‑connected—small perturbations do not change the ranking of agents’ (productivity × centrality) scores, and the suggested reallocation of incentives remains profit‑enhancing. Conversely, when the gap is small (the network is nearly partitioned), modest estimation errors can flip the ranking, potentially leading to harmful reallocation. This connection opens a bridge to econometric literature on network identification and perturbation theory.

The framework is further extended to Cobb–Douglas and CES production functions, which capture team‑level complementarities rather than pairwise ones. In these cases the balance condition still yields tractable solutions, and the authors show that pay dispersion is governed by the elasticity of substitution: high substitutability concentrates pay on high‑productivity agents, whereas strong complementarity induces a more equal distribution of compensation.

The paper situates its contributions within the contract theory and network games literatures, noting that while first‑order approaches are standard in single‑agent moral hazard, their interaction with spillovers has been largely unexplored. It also contrasts its non‑parametric, continuous‑action setting with recent algorithmic contract‑theory work that deals with discrete actions and combinatorial optimization.

In conclusion, the authors provide a clear, analytically tractable characterization of optimal team contracts that internalizes both direct productivity and indirect spillover effects. The balance condition offers a practical rule: allocate incentive pay proportionally to the product of marginal productivity, network centrality, and marginal utility of money. The paper also offers guidance on when this rule is robust to estimation error and how production technology (substitutability vs. complementarity) shapes the optimal dispersion of compensation. Future research directions include extending the analysis to multi‑stage contracts, heterogeneous risk preferences, and empirical estimation of spillover networks in real organizations.