To see the forest - A study on the directed spanning arborescence framework used in systemic risk analysis based on financial networks

Absztrakt (kivonat)

This paper explores the possibilities of using the directed spanning arborescence method to further improve the analysis of financial networks and the estimation of systemic risk between financial institutions. The networks of several banks from different countries are created from the pairwise Granger causality connections of these institutions based on the stock returns and volatility time series, where the resulting graph is weighted and directed. The usual network characteristics, such as centrality measures and In/Out connections are supplemented by the ones derived from the directed spanning arborescence of the network. By limiting the number of edges to only include the most important ones, the characteristics of these graphs increase the information content of the analysis and reduce noise. A regression framework is used to estimate both the general connection between firm performance and the attributes of the network, as well as the out of sample prediction. Comparison of the models including only traditional network measures and the ones including the spanning arborescence characteristics as well show that a more complex framework, where a simplification of the network takes place, can better estimate systemic risk and serve as a better indicator for firm losses under financial distress.