Central Bank Policy Experiments Using a Multilayer Network and an Embedded Agent-Based Model of the Interbank Market

Abstract

Why did interbank markets freeze during the crisis despite an unprecedented level of policy intervention? This paper seeks to answer that question by combining a realistic, multilayer network structure with an Agent-Based Model of the interbank market. The multilayer network consists of two types of interdependencies: direct exposures arising due to borrowing/lending behaviour of banks on the interbank market and indirect exposures in which banks are linked due to commonly-held securities in overlapping portfolios. Embedded in both layers of the simulated network, the Agent-Based Model allows a large number of heterogenous banks to form adaptive expectations in the form of simple heuristics which they use to set interbank borrowing/lending volumes, set interbank rates and sell off their securities to meet obligations to creditors. The heuristics are constructed to reflect banks' funding liquidity, counterparty and market liquidity risk, all of which were seen as pivotal in driving interbank market tensions during the crisis. The model is allowed to run over a fixed number of iterations featuring a crisis period (represented by falling market liquidity and asset prices) after which the results are presented as two policy experiments: (ii) The baseline model in which the central bank maintains normal bank refinancing conditions and (ii) With a short delay, the central bank introduces a policy of full allotment in which additional liquidity is injected with the aim of preventing bank insolvencies. Our simulation results show that while the new policy is able to avert a widespread breakdown of the banking sector, the delay between the onset of the crisis and the first intervention allowed interbank market risks to develop and feed back on one another, resulting in lower trading activity relative to the pre-crisis situation. This is broadly in line with a number of empirical findings associated to the crisis.

Keywords: Interbank network, Agent-Based Computational Economics, financial crisis, financial stability, financial contagion, central banking

JEL Classification: C63, G01, E03