This paper proposes a new methodology based on non-negative matrix factor- ization to detect communities and to identify Systemically Important Financial In- stitutions in the interbank network as well as within communities. The method is specifically designed for directed weighted networks and it is able to take into account exposures on both sides of banks' balance sheets, distinguishing between Systemically Important Borrowers and Lenders. Using interbank transactions data from the e-Mid platform, we show that the systemic importance associated with Italian banks decreased during the 2007-2009 financial crisis while the opposite happened for foreign institutions. We also show that, as the transactions volume grew, the number of communities rose as well. The contrary happened during the crisis phase. Moreover results indicate that, during financial crisis, banks strongly operate into non overlapping communities with few institutions playing the role of SIFIs. On the contrary during business as usual times banks act in several and overlapping modules.
Financial networks community detection systemic risk