An interconnection game between mobile network operators: Hidden information forecasting using expert advice fusion

In this article, we deal with the resolution of a dynamic game of interconnection between mobile network operators (MNOs) sharing their scarce resources and mobile virtual network operators (MVNOs) lacking the infrastructure but trying to bargain to send as much traffic as possible on the MNOs’ networks. Besides consumers’ perception of the MVNO’s brand is based on a word-of-mouth process; the consumers’ cheating probability being an exogeneous parameter. The difficulty arises from the fact that the MVNOs ignore the unused bandwidth volumes on both MNOs’ networks and how these latter allocate it between the competitive MVNOs. In fact, to increase their revenues, the MVNOs need to learn as fast as possible the hidden information. Five learning strategies have been tested: (i) tit for tat, (ii) fictitious play, (iii) external regret minimization, (iv) repeated exploration–exploitation, (v) internal regret minimization. We check numerically that in general, solely the internal regret minimization strategy reaches a correlated equilibrium. Besides MVNOs’ power relation and consumers’ cheating probability play a central role in the learning rate and the resulting organization. Indeed either one of the MVNOs learns faster, capturing a larger part of the market while the other has no choice but to follow (leader/follower), or an implicit cooperation occurs i.e., both MVNOs learn and increase their performances simultaneously (peers).