MIT researchers' new equilibrium concept achieves social optimality with internal transfers

A fundamental limitation of Nash equilibrium is that individual rationality can lead to outcomes arbitrarily far from the social optimum. In a new paper on arXiv, MIT researchers Mingyang Liu, Gabriele Farina, and Asuman Ozdaglar propose a way out: let players make internal transfers before the game begins. Their first concept, Self-Enforcing Transfer Equilibrium (SETE), allows players to commit to nonnegative peer-to-peer payments that are only triggered if the recipient deviates from a prescribed strategy. Because transfers are budget-balanced (the sum of all payments is zero), no external subsidies are needed. For polymatrix games—a broad class of multiplayer interactions—any stationary point of social welfare, including the global social optimum, can be supported as a SETE. Crucially, the authors provide a polynomial-time algorithm and a decentralized learning dynamic to compute such equilibria, making the concept computationally tractable.

However, SETE has a subtle limitation: it guarantees a Nash equilibrium only in the agent normal form of the augmented game, not in the original extensive form. To fix this, the team introduces Mediated Self-Enforcing Transfer Equilibrium (M-SETE), where a mediator makes binding offers for both the payment schedule and the prescribed strategies. With this added enforcement, an M-SETE becomes a Nash equilibrium of the augmented game itself, for any finite game. Social optimality can still be achieved while preserving budget balance and independent play on the equilibrium path. The work thus shows that carefully designed internal transfers—either backed by self-enforcement or mediation—can rescue the welfare properties of decentralised multi-agent systems without sacrificing computation or autonomy.