We develop a framework to analyse the credit default swap (CDS) market as a network of risk transfers among counterparties. From a theoretical perspective, we introduce the notion of flow-of-risk and provide sufficient conditions for a bow-tie network architecture to endogenously emerge as a result of intermediation. This architecture shows three distinct sets of counterparties: (i) Ultimate Risk Sellers (URS), (ii) Dealers (indirectly connected to each other), (iii) Ultimate Risk Buyers (URB). We show that the probability of widespread distress due to counterparty risk is higher in a bow-tie architecture than in more fragmented network structures. Empirically, we analyse a unique global dataset of bilateral CDS exposures on major sovereign and financial reference entities in 2011–2014. We find the presence of a bow-tie network architecture consistently across both reference entities and time, and that the flow-of-risk originates from a large number of URSs (e.g. hedge funds) and ends up in a few leading URBs, most of which are non-banks (in particular asset managers). Finally, the analysis of the CDS portfolio composition of the URBs shows a high level of concentration: in particular, the top URBs often show large exposures to potentially correlated reference entities.

This paper provides a unique snapshot of the asset exposures of EU banks to shadow banking entities within the global financial system. Drawing on a rich and novel dataset, we show that 60% of the EU banks’ exposures are towards non-EU entities, particularly US-domiciled shadow banking entities. We assess the degree of concentration across different types of shadow banking counterparties. We show that while banks’ exposures are diversified at the individual level, this diversification leads to high overlap across different types of shadow banking entities, with consequent systemic risk. We also examine how bank- and country-level characteristics relate to the exposures of EU banks to shadow banking entities. Our results emphasise the importance of monitoring these cross-border and cross-sector exposures and closing remaining data gaps.