Empirical Analyses of Networks in Finance
The recent global financial crisis has triggered a huge interest in the use of network concepts and network tools to better understand how instabilities can propagate through the financial system. The literature is today quite vast, covering both theoretical and empirical aspects. This review concentrates on empirical work, and associated methodologies, concerned with the evaluation of the fragility and resilience of financial and credit markets.
The first part of the review examines the literature on systemic risk that arise from banks mutual exposures. These exposures stem primarily from interbank lending and derivative positions, but also, indirectly, from common holdings of other asset classes, that can lead to common shocks in instances of fire sales, and from widespread non-performing loans to the real sector during period of economic downturns. We survey (a) studies that characterize the structure of national interbank networks, in some cases using a multiplex representations, (b) studies that introduce novel methods to quantify systemic risk and identify systemically important institutions, such as via stress test scenarios, (c) studies that assess which regulatory measures can help mitigate the propagation of contagion and distress in the financial system, and (d) studies that explore which location advantages may arise from holding privileged positions in the interbank network, such as via preferential lending relationships, or because of occupying a more central node, and if such advantages can provide an early indication of the build up of systemic risk.
The second part of the review is dedicated to the analysis of indirect networks, specifically (e) proximity based network, i.e. networks obtained starting from a proximity measure sometime filtered with a network filtering methodology, (f) association network, i.e. networks where a link between two financial actors is set if a statistical test again a null hypothesis is rejected, and (g) statistically validated networks, i.e. event or relationship networks where a subset of links is selected according to a statistical validation associated with the rejection of a random null hypothesis. The need for a joint consideration of direct and indirect channels of contagion is briefly discussed.