Network Representations of Interconnections and Contagion

Project Start:01/2016
Researchers:Nils Bertschinger, Aleksey Kolokolov, Roberto Panzica, Loriana Pelizzon, Zorka Simon, Tatiana von Landesberger
Area: Financial Markets, Systemic Risk Lab
Funded by:LOEWE

This project was part of the team project "Systemic Risk and Network Connectivity".

Topic & Objectives

We aim to show the interconnections among companies through visualization tools. We analyze two different sources of interconnectedness: the first one is relative to the sells among firms, the second one is built by using econometrics methodologies as Granger Causality. Analyzing and comparing different sources of interconnectedness is useful to understand the possible way of the risk spreading. Network measures are used to detect the difference between the two kind of networks previously defined, and to comprehend if these measures can be in some way compared with the moments of the returns distribution for that companies. We define methods for detecting communities inside the networks for the contagion mechanism comprehension. New measures and representation tools for treating topology similiarity among networks have been developed by looking at the overlapping edges and nodes. Tree, Sankey and cluster diagrams support our analysis in order to understand in which way the nodes aggregate themselves respectively in group, communities and finally in a unique complex system: the network.

The network estimation based on the causality methodology is used to understand how the risk spreads across assets returns. We assume that the idiosyncratic shocks move according to the channel defined by the network based on the Granger causality. We investigate if a relation by indegree centrality and stock returns exists and if the risk factor based on the indegree explains the idiosyncratic volatility puzzle. The puzzle consists of observing empirically a negative relation between portfolios sorted by idiosyncratic volatility at the previous month and the expected stock returns.

Key Findings

  • Portfolios sorted by increasing indegree computed on the network based on Granger causality test have lower expected returns, not related to idiosyncratic volatility.
  • Empirical evidence indicates that stocks with higher idiosyncratic volatility have the lower exposition on the indegree risk factor.


Related Working Papers

228Roberto PanzicaIdiosyncratic Volatility Puzzle: The Role of Assets' Interconnections2018 Financial Markets, Systemic Risk Lab Idiosyncratic volatility puzzle; Networks; Expected Returns; Granger Causality