Learning Market Equilibria Using Performative Prediction: Balancing Efficiency and Privacy

International audience; We consider a peer-to-peer electricity marketmodeled as a network game, where End Users (EUs) minimizetheir cost by computing their demand and generation whilesatisfying a set of local and coupling constraints. Their nominaldemand constitutes sensitive information, that they might wantto keep private. We prove that the network game admits aunique Variational Equilibrium, which depends on the privateinformation of all the EUs. A data aggregator is introduced,which aims to learn the EUs’ private information. The EUsmight have incentives to report biased and noisy readings topreserve their privacy, which creates shifts in their strategies.Relying on performative prediction, we define a decision-dependentgame G^stoch to couple the network game with adata market. Two variants of the Repeated Stochastic GradientMethod (RSGM) are proposed to compute the PerformativelyStable Equilibrium solution of G^stoch, that outperform RSGMwith respect to efficiency gap minimization, privacy preservation,and convergence rates in numerical simulations.