Duopolistic competition in an electricity markets with heterogeneous cost functions

In this paper the compelling issue of efficiency of electricity markets has been studied by means of an artificial power exchange based on the agent-based approach. In particular, two common market-clearing rules, i.e., discriminatory and uniform, have been compared with respect to efficiency outcomes. Computational experiments have been performed, where two heterogeneous competing sellers face an inelastic and constant demand within a repeated auction framework. Each seller is endowed with a limited production capacity, a specific cost function (linear and non linear) and learning capabilities. The seller's decision-making process has been modeled according to different reinforcement learning algorithms, namely, Marimon and McGrattan and Q-learning algorithms, which can be implemented under the same behavioral hypothesis, i.e., game-structure independence. Two different levels of demand are considered. A high-demand situation where overall demand is greater than the capacity of the greatest producer, and a low-demand situation where overall demand is less than the capacity of the smallest seller. Results are presented according to the occurrence of Nash equilibria and Pareto optima in the long-run behavior of the learning processes and to the profits achieved by the sellers. Computational experiments lead to the conclusions that the discriminatory auction mechanism tends to increase competitive behavior