A Theory of Network Games Part 1: Utility Representations

We demonstrate that a ubiquitous feature of network games, bilateral strategic interactions, is equivalent to having player utilities that are additively separable across opponents. We distinguish two formal notions of bilateral strategic interactions. Opponent independence means that player i’s preferences over opponent j’s action do not depend on what other opponents do. Strategic independence means that how opponent j’s choice influences i’s preference between any two actions does not depend on what other opponents do. If i’s preferences jointly satisfy both conditions, then we can represent her preferences over strategy profiles using an additively separable utility. If i’s preferences satisfy only strategic independence, then we can still represent her preferences over just her own actions using an additively separable utility. Common utilities based on a linear aggregate of opponent actions satisfy strategic independence and are therefore strategically equivalent to additively separable utilities–in fact, we can assume a utility that is linear in opponent actions.

💡 Research Summary

The paper investigates the foundational preference structure that underlies the widely used utility specifications in network games. The authors formalize the intuitive notion that “strategic interactions are bilateral” by introducing two precise properties: opponent independence and strategic independence. Opponent independence requires that a player’s ranking of a particular opponent’s action does not depend on the actions of any other opponents. Strategic independence requires that when an opponent changes his action, the direction in which this change shifts the player’s preference between any two of her own actions is the same regardless of what the remaining opponents do. The authors encode these ideas through a set of four possible preference comparisons; violating any of them destroys the possibility of an additively separable representation.

The first main result (Theorem 1) shows that if a player’s preferences satisfy both opponent and strategic independence, then her utility can be written as a sum of pairwise interaction terms:
(u_i(s)=\sum_{j\neq i} g_{ij}(s_i,s_j)).
Each term (g_{ij}) captures the bilateral effect of opponent (j) on player (i). This representation is cardinal—unique up to a positive affine transformation.

If only strategic independence holds, Theorem 2 establishes that for each fixed profile of opponents’ actions, the player’s preferences over her own actions can still be represented by an additively separable utility of the same form. This requires modest technical conditions (rich action sets, continuity, completeness) but does not guarantee separability over the full profile space.

The paper then connects these abstract conditions to the concrete utilities that dominate the network‑games literature. The classic linear‑quadratic specification of Ballester et al. (2006) and its extensions—where a player’s payoff depends on a linear aggregate of neighbors’ actions—satisfy strategic independence. Consequently, any model that assumes a linear aggregate can be transformed without loss into an exactly additively separable form, justifying the prevalent use of such utilities for equilibrium analysis and comparative statics.

A further contribution is the introduction of constant rate of substitution (CRS) across opponents. If a player’s marginal effect of increasing opponent (j)’s action is a constant multiple of the effect of increasing opponent (k)’s action, then Theorem 3 shows that the utility can be rewritten in a fully linear fashion:
(\tilde u_i(s)=\tilde b(s_i)+\tilde\gamma(s_i)\sum_{j\neq i} g_{ij}s_j).
Thus, even utilities that involve a strictly monotone non‑linear function of the aggregate (e.g., (u_i(s)=b_i s_i - s_i^2 + s_i\ln\sum_{j} g_{ij}s_j)) are strategically equivalent to a linear‑in‑neighbors form. The authors caution, however, that while the strategic incentives are identical, welfare implications may differ because the original non‑linear utility can reverse the sign of the player’s preference for higher neighbor actions.

The final technical section introduces the notion of a balanced sequence of strategy profiles. Two sequences are balanced if they differ only in the focal player’s action while each opponent‑action pair appears the same number of times in both sequences. The authors prove that a player’s preferences admit an additively separable representation if and only if the total utility across any balanced pair of sequences is equal. Violations can be detected with as few as four comparisons, but under the paper’s richer technical assumptions, checking short sequences suffices. This characterization links the abstract independence properties to a concrete testable condition.

Overall, the paper provides a rigorous justification for the ubiquitous linear‑aggregate utilities in network games, showing they are exactly the utilities that satisfy bilateral strategic interaction. It offers a clear empirical test for whether observed preferences conform to these assumptions, simplifies utility estimation, and opens the door to systematically exploring richer yet tractable utility families (e.g., those with CRS or other symmetry properties). The work bridges game‑theoretic modeling with consumer‑theory results on additive separability, extending classic Debreu‑type theorems to the strategic setting of network games.