A System-Wide Approach to Measure Connectivity in the Financial Sector

Abstract

We introduce and estimate a model that leverages a system-wide approach to identify systemically important financial institutions. Our Debiased Lasso penalized Vector Auto-regressive (DLVAR) framework, based on formal Granger Causality tests for large multivariate time series, explicitly allows for the possibility of connectivity amongst all institutions under consideration: this is in sharp contrast with most extant measures of systemic risk that, either explicitly or implicitly, estimate such connections using pairwise relationships between institutions. In addition, our method corrects for shrinkage bias and provides formal uncertainty quantification of estimated network edges, hence overcoming two key limitations of recently proposed lasso and sparsity regularized VAR models. Using simulations we show that our approach can provide considerable improvement over extant measures. We estimate our model for large financial institutions in the U.S. and show its usefulness in detecting systemically stressful periods and institutions.

Keywords: Systemic Risk, Financial Networks, Lasso, Vector Autoregression

JEL Classification: C58, G01, G29