Reconstruction of financial network for robust estimation of systemic risk

TL;DR

This paper introduces a message-passing algorithm to accurately reconstruct interbank credit networks, improving systemic risk estimation by capturing realistic network structures and providing upper bounds on contagion risk.

Contribution

The authors develop an efficient algorithm that explores network structures beyond maximum entropy assumptions, enabling more reliable systemic risk assessments and regulatory insights.

Findings

More accurate risk estimations with realistic network structures

Algorithm produces upper bounds for contagion risk

Effective control of information requirements for regulators

Abstract

In this paper we estimate the propagation of liquidity shocks through interbank markets when the information about the underlying credit network is incomplete. We show that techniques such as Maximum Entropy currently used to reconstruct credit networks severely underestimate the risk of contagion by assuming a trivial (fully connected) topology, a type of network structure which can be very different from the one empirically observed. We propose an efficient message-passing algorithm to explore the space of possible network structures, and show that a correct estimation of the network degree of connectedness leads to more reliable estimations for systemic risk. Such algorithm is also able to produce maximally fragile structures, providing a practical upper bound for the risk of contagion when the actual network structure is unknown. We test our algorithm on ensembles of synthetic data encoding some features of real financial networks (sparsity and heterogeneity), finding that more accurate estimations of risk can be achieved. Finally we find that this algorithm can be used to control the amount of information regulators need to require from banks in order to sufficiently constrain the reconstruction of financial networks.