Identifying Systemic Risk in Interbank Markets by Applying Network Theory

Zhuoran Xu

Research output: ThesisDoctoral Thesis

Abstract

Risk assessment on interbank networks has drawn attention from researchers since the 2007 Subprime mortgage crisis.The lack of data for interbank transactions, which are usually not disclosed unless required by regulatory bodies, is one of the most critical difficulties to this research. A remedy to this issue is the dense reconstruction of interbank networks by using balance sheet data. The Maximum-Entropy estimation has been adopted by literature, however, this method produces networks with unrealistic properties: too dense in terms of having too many links. One alternative is sparse reconstruction that proposed by literature recently. This thesis applies the Message-Passing algorithm, which is extensively applied in Thermodynamics or Computer Science, and is suggested by Mastromatteo et al. [2012] for application in network reconstruction. Dense networks and sparse networks are reconstructed from Statistics on Depository Institutions data provided by Federal Deposit Insurance Corporation, and are compared by performance in both network properties and contagion simulations. The popular contagion mechanisms proposed by Furfine [2003] and the model of liquidity dry-up contagion proposed by Malherbe [2014] are adopted and compared in contagion simulations. Results show that dense networks and sparse networks perform differently in network properties and in contagions triggered by single-bank failures, while for contagions triggered by multiple-bank failures, both types of networks perform similarly. Furfine’s mechanism fail to predict some bank failures via the credit risk contagion on liquidity side, while these failures can be simulated by the liquidity dry-up model via fire-sale and marking-to-market effect. Both mechanisms overestimate the losses before the crisis, yet this signals the instability of the banking system, while the liquidity dry-up model proposes an explanation for why the banking system did not fail before the crisis, regarding to whether the equilibrium of high liquidity will shift to the self-fulfilling liquidity dry-up equilibrium. Implications on regulation are given.
LanguageEnglish
QualificationPh.D.
Awarding Institution
  • University of Bath
Supervisors/Advisors
  • Krause, Andreas, Supervisor
  • Giansante, Simone, Supervisor
Award date23 Apr 2016
StatusUnpublished - 25 Mar 2016

Fingerprint

Circuit theory
Message passing
Insurance
Risk assessment
Computer science
Sales
Fires
Entropy
Deposits
Statistics
Thermodynamics
Industry

Keywords

  • Systemic risk
  • Liquidity dry-up
  • Interbank network

Cite this

Identifying Systemic Risk in Interbank Markets by Applying Network Theory. / Xu, Zhuoran.

2016. 188 p.

Research output: ThesisDoctoral Thesis

@phdthesis{f15753eed81949b999172d3b81817d9d,
title = "Identifying Systemic Risk in Interbank Markets by Applying Network Theory",
abstract = "Risk assessment on interbank networks has drawn attention from researchers since the 2007 Subprime mortgage crisis.The lack of data for interbank transactions, which are usually not disclosed unless required by regulatory bodies, is one of the most critical difficulties to this research. A remedy to this issue is the dense reconstruction of interbank networks by using balance sheet data. The Maximum-Entropy estimation has been adopted by literature, however, this method produces networks with unrealistic properties: too dense in terms of having too many links. One alternative is sparse reconstruction that proposed by literature recently. This thesis applies the Message-Passing algorithm, which is extensively applied in Thermodynamics or Computer Science, and is suggested by Mastromatteo et al. [2012] for application in network reconstruction. Dense networks and sparse networks are reconstructed from Statistics on Depository Institutions data provided by Federal Deposit Insurance Corporation, and are compared by performance in both network properties and contagion simulations. The popular contagion mechanisms proposed by Furfine [2003] and the model of liquidity dry-up contagion proposed by Malherbe [2014] are adopted and compared in contagion simulations. Results show that dense networks and sparse networks perform differently in network properties and in contagions triggered by single-bank failures, while for contagions triggered by multiple-bank failures, both types of networks perform similarly. Furfine’s mechanism fail to predict some bank failures via the credit risk contagion on liquidity side, while these failures can be simulated by the liquidity dry-up model via fire-sale and marking-to-market effect. Both mechanisms overestimate the losses before the crisis, yet this signals the instability of the banking system, while the liquidity dry-up model proposes an explanation for why the banking system did not fail before the crisis, regarding to whether the equilibrium of high liquidity will shift to the self-fulfilling liquidity dry-up equilibrium. Implications on regulation are given.",
keywords = "Systemic risk, Liquidity dry-up, Interbank network",
author = "Zhuoran Xu",
year = "2016",
month = "3",
day = "25",
language = "English",
school = "University of Bath",

}

TY - THES

T1 - Identifying Systemic Risk in Interbank Markets by Applying Network Theory

AU - Xu,Zhuoran

PY - 2016/3/25

Y1 - 2016/3/25

N2 - Risk assessment on interbank networks has drawn attention from researchers since the 2007 Subprime mortgage crisis.The lack of data for interbank transactions, which are usually not disclosed unless required by regulatory bodies, is one of the most critical difficulties to this research. A remedy to this issue is the dense reconstruction of interbank networks by using balance sheet data. The Maximum-Entropy estimation has been adopted by literature, however, this method produces networks with unrealistic properties: too dense in terms of having too many links. One alternative is sparse reconstruction that proposed by literature recently. This thesis applies the Message-Passing algorithm, which is extensively applied in Thermodynamics or Computer Science, and is suggested by Mastromatteo et al. [2012] for application in network reconstruction. Dense networks and sparse networks are reconstructed from Statistics on Depository Institutions data provided by Federal Deposit Insurance Corporation, and are compared by performance in both network properties and contagion simulations. The popular contagion mechanisms proposed by Furfine [2003] and the model of liquidity dry-up contagion proposed by Malherbe [2014] are adopted and compared in contagion simulations. Results show that dense networks and sparse networks perform differently in network properties and in contagions triggered by single-bank failures, while for contagions triggered by multiple-bank failures, both types of networks perform similarly. Furfine’s mechanism fail to predict some bank failures via the credit risk contagion on liquidity side, while these failures can be simulated by the liquidity dry-up model via fire-sale and marking-to-market effect. Both mechanisms overestimate the losses before the crisis, yet this signals the instability of the banking system, while the liquidity dry-up model proposes an explanation for why the banking system did not fail before the crisis, regarding to whether the equilibrium of high liquidity will shift to the self-fulfilling liquidity dry-up equilibrium. Implications on regulation are given.

AB - Risk assessment on interbank networks has drawn attention from researchers since the 2007 Subprime mortgage crisis.The lack of data for interbank transactions, which are usually not disclosed unless required by regulatory bodies, is one of the most critical difficulties to this research. A remedy to this issue is the dense reconstruction of interbank networks by using balance sheet data. The Maximum-Entropy estimation has been adopted by literature, however, this method produces networks with unrealistic properties: too dense in terms of having too many links. One alternative is sparse reconstruction that proposed by literature recently. This thesis applies the Message-Passing algorithm, which is extensively applied in Thermodynamics or Computer Science, and is suggested by Mastromatteo et al. [2012] for application in network reconstruction. Dense networks and sparse networks are reconstructed from Statistics on Depository Institutions data provided by Federal Deposit Insurance Corporation, and are compared by performance in both network properties and contagion simulations. The popular contagion mechanisms proposed by Furfine [2003] and the model of liquidity dry-up contagion proposed by Malherbe [2014] are adopted and compared in contagion simulations. Results show that dense networks and sparse networks perform differently in network properties and in contagions triggered by single-bank failures, while for contagions triggered by multiple-bank failures, both types of networks perform similarly. Furfine’s mechanism fail to predict some bank failures via the credit risk contagion on liquidity side, while these failures can be simulated by the liquidity dry-up model via fire-sale and marking-to-market effect. Both mechanisms overestimate the losses before the crisis, yet this signals the instability of the banking system, while the liquidity dry-up model proposes an explanation for why the banking system did not fail before the crisis, regarding to whether the equilibrium of high liquidity will shift to the self-fulfilling liquidity dry-up equilibrium. Implications on regulation are given.

KW - Systemic risk

KW - Liquidity dry-up

KW - Interbank network

M3 - Doctoral Thesis

ER -