A structural model of segregation in social networks

The main challenges in estimating strategic network formation models are the pres-ence of multiple equilibria, and the fact that the number of possible network config-urations increases exponentially with the number of players. I propose a dynamic model of strategic network formation with heterogeneous players, which converges to a unique stationary equilibrium. Hence, the structural preference parameters can be estimated using a single network observation. In addition, the model provides the first equilibrium micro-foundation of exponential random graphs. Because of the curse of dimensionality, the likelihood is computationally intractable. Therefore, I propose a Bayesian estimation strategy that samples from the posterior, interleaving parameter and network simulations, without evaluating the likelihood. I prove that the proposed algorithm converges to the correct posterior distribution. A mean-field analysis shows that the algorithm converges fast for practical applications. Estimation is tested with artificial and Add Health data, confirming evidence of homophily in high schools