Abstract

This talk explores how feedback between learning agents and their environment shapes collective behaviour and market dynamics. In the first part, I present a decentralized multi-agent reinforcement learning framework where agents develop cooperation through participation incentives, even without explicit coordination or shared information. In the second part, I discuss performative market making—how financial models can become self-fulfilling by influencing the very markets they aim to describe. Together, these studies reveal how stochastic control and reinforcement learning intertwine in systems where learning agents not only adapt to, but also create, their environment. Co-authors: Stefan Roesch (PhD student at KCL ), Prof. Odinaldo Rodrigues (KCL), Dr. Yali Du (KCL), Charalampos Kleitsikas (PhD student at KCL ), Prof. Carmine Ventre (KCL).