Authors: Vasilis Hatzopoulos and Giulia Iori
Status: Working Papers
Abstract: In this paper we examine the temporal evolution of the e-Mid interbank market transactions and quantify topological changes of the resulting credit network at or near events that where considered pivotal in the 2007-2008 credit crisis. The main question we address is whether banks behaviour regarding the choice of counter parties in a trade changed before and during the subprime crisis. In particular, using a network based entropy measure, we assess the level of randomness in the weights distribution across the links of the credit network. We interpret this randomness as a proxy of the level of trust among credit institution. Simple analysis of fundamental properties of the time-ordered set of networks defined over non- overlapping maintenance periods indicate a shrinking market size. The number of nodes (banks) present per maintenance period, the number of edges, edge density and average degree in the system continued to shrink at a roughly constant rate. In order to compare the evolution of network metrics when the underlying network size changes over time it becomes crucial to define appropriate network null models against which the statistical significance of such metrics can be assess. Given the directed and weighted nature of our connections we construct a randomised ensemble of networks using the edge swap procedure, but conserving the vertex in-out strength sequence rather then the in-out degree sequence. We compare the entropy measure in the real networks with the one calculated in reshuffled networks and show that the interbank market moved at the beginning of the subprime crisis to a less random structure, with trading concentrated to a few selected counter parties, leading to a poor liquidity flow. Trust was only recovered after Lehman defaults, following the intervention of central banks around the world to inject liquidity into the banking system.