Congestion Pricing and Learning in Traffic Network Games

A stochastic model describing the learning process and adaptive behavior of finitely many users in a congested traffic network with parallel links is used to prove convergence almost surely toward an efficient equilibrium for a related game. To prove this result, we assume that the social planner charges on every route the marginal cost pricing without knowing what is the efficient equilibrium. The result is a dynamic version of Pigou’s solution, where the implementation is made in a decentralized way and the information about players gathered by the social planner is minimal. Our result and setting may be extended to the general case of negative externalities.