Contingent linear financial networks

This chapter develops a methodology to estimate hidden linear networks when only an aggregate outcome is observed. It implements the methodology to estimate financial networks among US financial institutions. After controlling for aggregate macroeconomic shocks, the authors find the data is explained more accurately with a mixture of different networks, where each network corresponds to a different transmission mechanism. The multiple network identification is implemented using an EM algorithm and the authors develop a systematic methodology to find the optimal number of networks. Credit default swap rates are the observable variable and the authors show that more than one network is needed to understand the data. The estimation of different networks allows the implementation of a macro-prudential and robust approach to the systemic risk in the financial system. The chapter shows that if the monetary authority estimates a single network, it will underestimate the systemic risk of the top 10 banks in the US by 71%.