Co-adaptation: From MARL Non-stationarity to the Dynamics of Economic Adjustment

Economic agents adapt their behavior to experience generated by interactions with other market participants. When several agents update their behavioral rules simultaneously, these adaptation processes become interdependent. We refer to this mechanism as co-adaptation. The paper develops a formal representation that separates the continuation of an agent's own rule updates from cross-agent effects on subsequent updates. In a two-agent system, mutual influence can increase the cumulative effect of an initial change and prolong the resulting adjustment process. A simple model of price adjustment between two firms shows how these effects depend on adjustment speed and the response to a competitor's price change. With more agents, interdependencies between adaptation processes form a directed network. This perspective links economic adjustment to multi-agent reinforcement learning, where learning by one agent changes the environment experienced by others, and clarifies when such non-stationarity is part of economic dynamics rather than only a computational difficulty.