Network Interventions: Targeting Agents or Targeting Links?

1. Problem Statement

The paper addresses the design of optimal interventions by a social planner in a network game characterized by strategic complementarities and endogenous link formation.

• Context: Agents (e.g., firms in R&D, students in education) choose an action level (effort/investment) and simultaneously choose the intensity of links (collaborations/friendships) with other agents.
• The Planner's Goal: Maximize a welfare function (typically increasing in agents' actions) subject to a budget constraint.
• The Dilemma: The planner can subsidize actions (individual effort) or links (connections between agents). The core question is: Under what conditions is it optimal to subsidize links, and when is it strictly better to focus solely on subsidizing actions?
• Key Distinction: Unlike prior literature that often assumes exogenous networks, this model allows agents to endogenously form links in response to subsidies, creating a complex feedback loop between link formation and action levels.

2. Methodology and Model Setup

The Model

• Agents: A set $N = \{1, \dots, n\}$.
• Choices:
  • Action $a_i \geq 0$.
  • Link intensity $g_{ij} \geq 0$ with every other agent $j$.
  • The effective link strength is $G_{ij} = g_{ij} + g_{ji}$.
• Utility Function:
  $$U_i(a, g) = b_i a_i + \rho \sum_{j \neq i} G_{ij} a_i a_j - \frac{c_i a_i^2}{2} + \sum_{j \neq i} s_{ij} G_{ij} - \sum_{j \neq i} \frac{f_{ij} g_{ij}^2}{2}$$
  • $b_i$: Standalone incentive for action.
  • $\rho$: Strength of strategic complementarities (spillovers).
  • $s_{ij}$: Baseline link incentive (intrinsic value of the link). This is the critical parameter.
  • $f_{ij}$: Cost of forming links (quadratic).
• Planner's Intervention:
  • Subsidy $\beta_i$ per unit of action $a_i$.
  • Subsidy $\sigma_{ij}$ per unit of link strength $G_{ij}$.
  • Budget Constraint: $\sum \beta_i a_i + \sum \sigma_{ij} G_{ij} \leq B$.

Analytical Approach

1. Equilibrium Characterization: The authors derive the Nash Equilibrium $(a^*, g^*)$ using first-order conditions.
   • Actions satisfy a linear system: $(C_S - \rho G_S) a^*_S = (b + \beta)_S$.
   • Link intensities satisfy: $g^*_{ij} = \frac{\rho a^*_i a^*_j + s_{ij} + \sigma_{ij}}{f_{ij}}$.
2. Sensitivity Analysis (Lemma 1): The authors establish a relationship between the marginal welfare effects of action subsidies ($\beta$) and link subsidies ($\sigma$). They show that the spillover dynamics are governed by the same matrix structure, allowing for a direct comparison of the "cost-effectiveness" of each tool.
3. First-Order Conditions: Assuming differentiability of the equilibrium selection and a binding budget constraint, the authors derive necessary conditions for optimality.

3. Key Contributions and Results

The paper's primary contribution is a sharp characterization of the optimal intervention structure based on the baseline link incentive ($s_{ij}$).

Main Result (Theorem 1)

The optimal intervention strategy depends entirely on the sign of $s_{ij}$:

1. Case A: Non-Negative Baseline Incentive ($s_{ij} \geq 0$)

   • Result: The optimal policy sets link subsidies to zero ($\sigma^*_{ij} = 0$).
   • Mechanism: Even though links generate positive externalities, agents already have an intrinsic motivation to form them. Subsidizing actions is strictly cheaper for the planner to induce proportional spillovers. Subsidizing links would be redundant and inefficient because the planner can "free ride" on the agents' intrinsic motivation to connect.
   • Implication: In settings like education (where social ties are naturally desired), the planner should only subsidize study effort, not the formation of friendships.

2. Case B: Negative Baseline Incentive ($s_{ij} < 0$)

   • Result: If the planner subsidizes the actions of two connected agents ($\beta^*_i > 0, \beta^*_j > 0$) and they form a link ($G^*_{ij} > 0$), the planner must also subsidize the link ($\sigma^*_{ij} > 0$).
   • Mechanism: When $s_{ij} < 0$, agents underinvest in links because the intrinsic cost outweighs the benefit. The planner must directly subsidize the link to overcome this distortion. Without the link subsidy, the agents would not form the link even if their actions were high, breaking the spillover chain.
   • Implication: In settings like R&D collaborations (where collaboration is costly and not inherently "fun"), the planner must subsidize the collaboration itself to make it viable.

Comparison with Exogenous Networks (Theorem 2)

The authors contrast their findings with a benchmark model where links are exogenous (chosen before the game without considering payoffs).

• Exogenous Case: Optimal link subsidies are high when actions are high (strong spillovers).
• Endogenous Case: Optimal link subsidies are high when actions are low (induced by negative $s_{ij}$).
• Significance: The endogeneity of links reverses the standard intuition. In endogenous settings, subsidizing links is a tool to create connectivity where it wouldn't exist, whereas in exogenous settings, it is a tool to amplify existing high-value networks.

Generalization (Theorem 3)

The results hold beyond the linear-quadratic framework. The authors define conditions on the curvature of cost and utility functions:

• No Link Subsidy ($s_{ij} \geq 0$): Requires link costs to be super-quadratic (growing very fast) and spillovers to be sub-linear (diminishing returns).
• Link Subsidy Required ($s_{ij} < 0$): Requires link costs to be sub-quadratic (not too steep) and spillovers to be super-linear (increasing returns).

4. Significance and Applications

• Policy Design: The paper provides a clear rule of thumb for policymakers.
  • Education: Since students naturally want social connections ($s_{ij} \geq 0$), policies should focus on individual performance incentives (e.g., scholarships, tutoring) rather than programs designed solely to force students to network.
  • R&D/Innovation: Since collaboration is often costly and risky ($s_{ij} < 0$), policies must directly subsidize the collaborative mechanism (e.g., joint venture grants, cluster subsidies) alongside individual firm R&D tax credits.
• Theoretical Insight: The paper resolves the trade-off between "targeting agents" vs. "targeting links." It demonstrates that the relative cost of the two instruments is not just about the price of the subsidy, but about the marginal cost of inducing equilibrium changes. When agents naturally form links, the marginal cost of inducing a link via subsidy is high (because you are paying for something they would do anyway); when they don't, the marginal cost is low.
• Robustness: The findings are robust to the specific welfare function and the specific costs of effort, relying primarily on the sign of the intrinsic link value.

Summary Conclusion

Dasaratha and Shah demonstrate that in network games with endogenous link formation, the optimal intervention strategy is dictated by the intrinsic value of the link. If links are intrinsically valuable, the planner should ignore them and subsidize actions. If links are intrinsically costly (negative value), the planner must subsidize the links themselves to unlock the benefits of network spillovers. This finding fundamentally reverses the logic found in models with exogenous networks.