Learning agents in an artificial power exchange: Tacit collusion, Market power and efficiency of two double-auction mechanisms

This paper addresses the widely debated issue of comparing the relative efficiency of two doubleauction mechanisms for power exchanges. An agent-based modeling approach is adopted to simulate a double-auction power exchange. Two standard pricing rules are considered and compared: “discriminatory” and “uniform”. Computational experiments, characterized by different inelastic demand level, reproduce oligopolistic competitions on both quantity and price between learning sellers/producers. Two reinforcement learning algorithms are considered “Marimon and McGrattan” and “Q-learning” in an attempt to simulate different behavioral assumptions. In particular, with the former algorithm greedy sellers optimizing their instantaneous rewards on a tick by tick basis are simulated, while with the latter inter-temporal optimizing sellers which exhibit tacit collusive behavior are modeled. Results have been interpreted by means of game-theoretical solutions and performance metrics. Nash equilibria in pure strategies are considered for monitoring the correct learning of competitive solutions and sellers’ joint profit maximizing for monitoring the emergence of tacit collusion among the sellers. Furthermore, the difference between payments to suppliers and total generation costs have been estimated for measuring the degree of market inefficiency. Results show that tacit collusive behaviors are penalized by the discriminatory auction mechanism in a low demand scenario. Lower profits for sellers/producers are determined compared to the uniform mechanism. Conversely, in a high demand scenario the differences result negligible.