Communication networks in markets

GAICIL Insights

A market is only as strong as its least informed members.¹ That sounds like a slogan. Communication networks in markets, by Edoardo Gallo, proves it is a theorem, and then confirms it in the lab. The paper builds a dynamic model of bargaining in decentralized markets where buyers and sellers learn about past deals through their social network.

The theory yields a unique equilibrium in which every individual gets the same outcome regardless of network position. That outcome depends on the peripheral, least connected individuals. More precisely, the equilibrium division depends on the number and strength of connections of the least connected members of each group. Testable predictions follow: networks with high density, or with low variability in connections across individuals, let their members obtain better deals.

A laboratory experiment across 4 treatments that vary the network confirms the theory. Subjects converge to an outcome independent of network position, and they fare better in dense networks or networks with low variability in connections. The design has an elegant trick: creating a communication network in the lab amounts to designing a communication protocol among workstations, sidestepping the difficulty of creating real social relations. Random assignment gives unambiguous identification of the causal link from network structure to market outcome, something field data rarely allow.

Turn the headline result around and it becomes a radical claim about where to invest in information. Improving the information of peripheral members raises the outcome for everyone in the group. The best-connected member gains nothing from more connections at the top. The gains live at the edges, and they are shared across the whole group.

The application to AI is direct. Artificial intelligence is a periphery-connecting machine. AI tools can aggregate information about past transactions, prices and terms and deliver it to anyone at low cost. In the model's language, AI raises the effective connectivity of peripheral individuals. The theory implies those improvements lift outcomes for the entire group. Price-transparency tools, AI-assisted negotiation and market information services may therefore create gains well beyond their direct users. The farmer with a phone-based price assistant improves the bargaining position of every farmer in the village.

The flip side matters as well. When one side of a market is better connected, it can extract better terms. AI that democratizes access to market information can shrink such asymmetries. Tools available only to the already well connected could widen them. The design and accessibility of AI information services shape the distribution of gains from trade, which makes affordability a question of market structure as much as of consumer convenience.

Policy implications follow cleanly. Rules facilitating the sharing of transaction data, while protecting privacy and confidentiality where needed, strengthen peripheral participants. Competition among providers of market information and AI tools keeps those services accessible and affordable. Because improvements at the periphery lift outcomes across the network, the social value of such policies may be large, far larger than counting direct users would suggest.

The author notes that network theorists have only recently begun to study such questions. A natural next step is to compare AI-mediated information networks, where algorithms aggregate and redistribute information, with human communication networks in their effects on bargaining. An algorithm that summarizes every past deal for every participant would, in the model's terms, make the network maximally dense.

Read as a whole, market efficiency depends on the architecture of information flows. Dense, evenly connected networks produce better outcomes, and the least connected participants set the result. In an economy where AI increasingly carries information, technologies that reach the periphery offer a path to markets that work better for everyone. The smartest information investment, on this evidence, targets those who currently know the least.