Network Formation with Adaptive Agents

In this paper, a reinforcement learning version of the connections game first analysed by Jackson and Wolinsky is presented and compared with benchmark results of fully informed and rational players. Using an agent-based simulation approach, the main nding is that the pattern of reinforcement learning process is similar, but does not fully converge to the benchmark results. Before these optimal results can be discovered in a learning process, agents often get locked in a state of random switching or early lock-in.