A Dynamic Bayesian Network Approach to the Interbank Market

ABSTRACT

We propose a Bayesian dynamic interbank network model with probit links that simultaneously incorporates three underlying mechanisms for interbank trading: dynamic activity indices for overall market confidence, bank‐specific latent variables for individual banks' fitness as borrowers or lenders, and pairwise covariates characterizing past trading relationships. Correspondingly, a computationally efficient Gibbs sampling algorithm is developed for sampling model parameters from posterior distributions that handles constrained parameters to achieve identification in a model with latent components. By applying our Bayesian approach to the e‐MID interbank market, we obtain empirically useful estimates of latent parameters that not only forecast trading linkages, but also provide pricing information about the interest rates banks charge each other. New model‐based proxies of network topology change and relationship lending are proposed to investigate their impact on relevant economic variables and the price of liquidity. The model is also extended to allow logit links as a flexible alternative.