(Re-) Creating Banking Networks

6 June 2019
Versione stampabile

Venue: Seminar room, first floor, Department of Economics and Management, via Inama 5

Time: 2 PM

Speaker:

Abstract

In the aftermath of the financial crisis,  capturing the level of interconnectedness across financial institutions and developing risk monitoring and mitigating tools became a priority for regulators. In particular, systemic risk and contagion through network interconnectedness were considered the biggest threat for the collapse of the entire financial system.
 
The entire interbank network is often unobservable, and the mechanism behind network formation and construction are often not clear. Still, marginal distributions of assets and liabilities can be obtained from regulatory filings or credit registers, promoting a stream of research which focuses on methods that aim to "fill in the blanks", i.e., to reconstruct the network. Early attempts suggested to employ maximum entropy methods as they are easy to implement, despite not being capable of replicating certain stylized facts, such as network sparsity and core-periphery structure, where core members are strongly connected to each other, while periphery members establish just few relationships with core members and not among each other.  Alternative methods based on copulas, bootstrap and iterative algorithms have then subsequently developed and compared, yet come with their own limitations.
 
In this paper, we suggest a method to create parsimonious networks. Previous work by Ben Craig and Goetz von Peter argues that establishing and maintaining links between banks is costly; networks with fewer links should therefore be preferable. We build on this, but extend it in two aspects: First, we refine the model by introducing declining marginal costs for establishing new links. Establishing links and everything that comes with it (setting up contracts, monitoring, etc.) follows a learning curve, and eventually turns into a cheap routine exercise. Secondly, we propose a new method to find optimal networks. The optimization problem underlying the network creation model is, from an optimization point of view, tough and challenging. We therefore suggest a numerical procedure that combines concepts for a related problem in logistics with non-deterministic optimization principles. We find that the resulting networks exhibit stylized facts close to real-world networks.