We propose a novel methodology for dynamic econometric modelling of large financial networks subject to persistence, structural changes and sparsity. We estimate bivariate dynamic Tobit-type models for each pair of banks, allowing for deterministic or stochastic time-varying parameters, and then aggregate across all bank pairs. To tackle the high dimensionality of the model, we construct a few lagged variables that efficiently summarize the position of a bank pair in the network. We propose a simple and computationally easy kernel-based local maximum likelihood estimator of the time-varying parameters of the model and establish its asymptotic properties. We then apply the model to the time series of daily overnight money market network in the UK 2003-2012. The results show that our model can successfully accommodate the numerous structural breaks arising from changes to the monetary framework and captures well the dynamics of the interbank lending relationships in this period.
