Discussion Tomohiro Ota Bank of England * The views expressed in this presentation are mine and not necessarily those of the Bank of England. Mapping the network studies Network Data Contagion mechanism Metrics Gross and Kok Implied exposure of banks and sovs (estimated from CDS premia) Estimated from dynamic Spill-over potential interaction of banks’ CDS (average impulse premia using MCS-GVAR response) Denbee, Julliard, Li and Yuan O/N interbank loans Strategic interaction of (estimated from banks’ liquidity holding payment data) (between linked banks) Indegree/outdegree Katz-Bonacich centrality Giansante and Markose RBI’s bilateral Loss of capital exposure data of FIs propagates through network (marginal contagion) Eigen-pair method (right/left Eigenvector) Marginal contagion model: shocks propagates through solvent banks. E.g. Battiston’s et.al. (2012) DebtRank and Ota’s (2013) Chained Merton model Gross and Kok: Main findings • Banks’ and sovereigns’ interconnectedness is estimated by dynamic interaction of CDS premia (with MCS-GVAR) • The impulse response matrix provides the size of amplification (and key players, I guess?) • Implied networks became more dense during the crisis • The spill-over effect from banks to sovereigns was strong in 2008, but the direction went opposite in 2012 • The spill-over effect is regionally heterogeneous (South EA countries are linked tighter) Yuan et.al.: Main findings • Studying the strategic behaviour of liquidity holding (as NE, based on Ballester et.al. 2006) • Katz-Bonacich centrality, characterising the NE, identifies key players in the market • Testing the theory model to estimate indeterminate parameters and an ‘amplification’ factor of agg. liquidity • Before LB crisis, the banks’ liq holding showed strategic complementarity • After the introduction of QE, a bank’ liquidity holding reduces the counterparties’ liquidity holdings Giansante and Markose: Main findings • Modelling higher-order ‘Infections’ of loss of capital through a given network structure • The maximum Eigenvalue of ‘relative exposure’ matrix gives the size of amplification potential • The column-sum and row-sum of corresponding Eigenvector identifies key ‘offenders’ and ‘victims’ • The Pigouvian tax using Eigenvector centrality reduces Maximum Eigenvalue (i.e. amplification) by internalising the negative externality of the infection A question from a central bank economist: Advantages and disadvantages of the methods? Network Data Contagion mechanism Metrics Gross and Kok Implied exposure of banks and sovs (estimated from CDS premia) Estimated from dynamic Spill-over potential interaction of banks’ CDS (average impulse premia using MCS-GVAR response) Denbee, Julliard, Li and Yuan O/N interbank loans Strategic interaction of (estimated from banks’ liquidity holding payment data) (between linked banks) Indegree/outdegree Katz-Bonacich centrality Giansante and Markose RBI’s bilateral Loss of capital exposure data of FIs propagates through network (marginal contagion) Eigen-pair method (right/left Eigenvector) A question from a central bank economist: how should we model systemic liquidity risk? • Credit exposures have been (will be) collateralised significantly (e.g. collateralisation of OTC derivatives) UK interbank bilateral exposures by instrument (net of collateral) Langfield, Liu and Ota (2014) • … but where is the collateral coming from? A question from a central bank economist: how should we model systemic liquidity risk? • Modelling liquidity contagion would be more complicated than credit risk contagion – Credit risk is defined on balance sheets – Liquidity risk is more behavioural – Higher degree of freedom in liquidity management models A question from a central bank economist: how should we model systemic liquidity risk? • Theoretical approaches – Considering actions on a network – Endogenous network formation is an essential part of systemic liquidity risk – No workhorse model (yet) – Could be too complicated to solve A question from a central bank economist: how should we model systemic liquidity risk? • Price-based approaches – Can skip the complicated propagation channels and focus on the results – We don’t have decent network time series data – Do we have market-traded prices indicating liquidity stress of a bank / firm? – We need to be aware of what we are estimating A question from a central bank economist: how should we model systemic liquidity risk? • Agent-based modelling approach – Powerful tool to model the complex nature of liquidity risk – How many assumptions do we need? – Potential risk of being a ‘black box’ • ABM for simulating real life • ABM to understand structure – Do we need consensus of assumptions? – Can we define social optimum and/or 1st best? Discussions of Denbee, Julliard, Li and Yuan “Network risk and key players: A structural analysis of interbank liquidity” Tomohiro Ota Bank of England Summary • Studying the strategic behaviour of liquidity holding (as NE, based on Ballester et.al. 2006) • Seeing the impact of an idiosyncratic shock to aggregate liquidity shock • Social optimum can be obtained • Proposing systemic risk metrics consistent with the theory • Testing the model empirically Findings • A bank holds more liquidity when the counterparties increase their liq holding if: – Holding liquid assets promises borrowing from connected counterparties • Katz-Bonacich centrality, characterising the NE, identifies key players in the market • Before LB crisis, the banks’ liq holding showed strategic complementarity • During the crisis, the network density decreased and banks are less dependant • After the introduction of QE, a bank’ liquidity holding reduces the counterparties’ liquidity holdings Questions and comments • Does a larger liquid asset holding increase borrowing capability? – Is leverage stack applicable to consider the issue • How to define liquidity for empirical tests – Reserve plus ‘collateral holding’ • Reserves are ‘on average’ affected by monetary policy – The collateral holding may not be a strategic choice – Hidden links between the banks and BOE – Tiering: some trades are not for the banks Discussions of Giansante and Markose “Multi-Agent Financial Network models for systemic risk monitoring and design of Pigou tax for SIFIs” Tomohiro Ota Bank of England Two main approaches to financial contagion • Assume specific transmission channels of financial stress to test the size of systemic risk (for a given network structure) • Estimating transmission channels from market data – Diebold and Yilmaz (2011): – Billio et. al. (2013): – Perasan et.al. (2006): variance decomposition Granger causality GVAR Questions and comments • Does this measure the unobservable interdependence between Fis (and sovereigns)? • Is it intuitive that network became dense after the crisis? • Is MCS-GVAR better than the other ways?