University of Groningen

Credit and liquidity risk of banks in stress conditionsEnd, Willem Adrianus van den

IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite fromit. Please check the document version below.

Document VersionPublisher's PDF, also known as Version of record

Publication date:2011

Link to publication in University of Groningen/UMCG research database

Citation for published version (APA):End, W. A. V. D. (2011). Credit and liquidity risk of banks in stress conditions: analyses from a macroperspective. Groningen: University of Groningen, SOM research school.

CopyrightOther than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of theauthor(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).

Take-down policyIf you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediatelyand investigate your claim.

Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons thenumber of authors shown on this cover page is limited to 10 maximum.

Download date: 31-12-2019

Credit and liquidity risk of banks in stress conditions

Analyses from a macro perspective

ISBN 978-90-9026295-6

© W.A. van den End, 2011

Alle rechten voorbehouden. Niets uit deze uitgave mag worden verveelvoudigd, opgeslagen in een

geautomatiseerd gegevensbestand, of openbaar gemaakt in enige vorm of op enige wijze, hetzij elektronisch,

mechanisch of door fotokopieën, opnamen, of op enige andere manier, zonder voorafgaande schriftelijke

toestemming van de auteur.

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or

transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise

without the written permission of the author.

Druk: Nextprint

RIJKSUNIVERSITEIT GRONINGEN

Credit and liquidity risk of banks in stress conditions

Analyses from a macro perspective

Proefschrift

ter verkrijging van het doctoraat in de Economie en bedrijfskunde

aan de Rijksuniversiteit Groningen op gezag van de

Rector Magnificus, dr. E. Sterken, in het openbaar te verdedigen op

donderdag 27 oktober 2011 om 11.00 uur

door

Willem Adrianus van den End geboren op 12 maart 1970

te IJsselmuiden

Promotor: Prof. dr. J. de Haan Beoordelingscommissie: Prof. dr. S.C.W. Eijffinger

Prof. dr. S. Gerlach Prof. dr. M. Koetter

Contents

1 Introduction 1

1.1 Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Research questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.3 Research approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.4 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.4.1 Bank behaviour during the crisis . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.4.2 Macro stress-testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.4.3 Policy responses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

Part I: Bank behaviour during the crisis

2 When liquidity risk becomes a systemic issue: Empirical evidence of bank behaviour 9

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.1.1 Dimensions of liquidity risk . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.1.2 Modelling bank behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.1.3 Contribution to the literature . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.2 Data and trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.2.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.2.2 Trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.3 Empirical measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.3.1 Instruments used to react . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.3.2 Size of reactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.3.3 Dependence of reactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.3.4 Herding: aggregate number of reactions . . . . . . . . . . . . . . . . . . . . . 20

2.3.5 Herding: number of reactions by instrument . . . . . . . . . . . . . . . . . . . 21

2.3.6 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

Appendix 2.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

Appendix 2.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

Appendix 2.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

3 Banks’ responses to funding liquidity shocks:

Lending adjustment, liquidity hoarding and fire sales 31

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

3.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3.3 Data and stylized facts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.4.1 Response of lending . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.4.2 Liquidity hoarding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

3.4.3 Fire sales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

3.5 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

3.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

Appendix 3.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

Part II: Macro stress-testing

4 Macro stress-testing methods 53

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

4.2 Micro stress-testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

4.3 Bottom-up macro stress-testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

4.4 Bottom-up stress-testing in the crisis: three approaches . . . . . . . . . . . . . . . . . . 56

4.5 Top-down macro stress-testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

4.5.1 Modelling the macro-micro link . . . . . . . . . . . . . . . . . . . . . . . . . . 58

4.5.2 Integrated models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

4.6 Considerations on the use of stress-tests . . . . . . . . . . . . . . . . . . . . . . . . . 60

4.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

5 Modelling scenario analysis and macro stress-testing 63

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

5.2 Scenario building . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

5.3 Credit risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

5.3.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

5.3.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

5.3.3 Estimation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

5.4 Interest rate risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

5.4.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

5.4.2 Data and estimation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

5.5 Simulation of scenario effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

5.5.1 Deterministic scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

5.5.2 Stochastic simulation of credit risk in base scenario . . . . . . . . . . . . . . . 75

5.5.3 Stochastic simulation of credit risk in stress scenarios . . . . . . . . . . . . . . 78

5.5.4 Stochastic simulation of interest rate risk in stress scenarios . . . . . . . . . . . 80

5.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

Appendix 5.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

6 Liquidity Stress-Tester: A model for stress-testing banks’ liquidity risk 85

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

6.2 Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

6.3 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

6.3.1 Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

6.3.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

6.3.3 First round effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

6.3.4 Banks’ response to scenario (mitigating actions) . . . . . . . . . . . . . . . . . 93

6.3.5 Second round effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

6.3.6 Impact different scenario rounds . . . . . . . . . . . . . . . . . . . . . . . . . 98

6.3.7 Influence of alternative distributional assumptions . . . . . . . . . . . . . . . 100

6.3.8 Parameter sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

6.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

6.4.1 Banking crisis scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

6.4.2 Credit crisis scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

6.4.3 Impact scenario length and market conditions . . . . . . . . . . . . . . . . . 105

6.4.4 Back-test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

6.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

Appendix 6.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

Appendix 6.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

Appendix 6.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

7 Liquidity Stress-Tester: Do Basel III and unconventional monetary policy work? 113

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

7.1.1 Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

7.1.2 Contribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

7.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

7.2.1 Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

7.2.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

7.2.3 Initial liquidity ratios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

7.2.4 First round effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

7.2.5 Mitigating actions by banks . . . . . . . . . . . . . . . . . . . . . . . . . . 121

7.2.6 Second round effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

7.2.7 Central bank reaction function . . . . . . . . . . . . . . . . . . . . . . . . . 125

7.2.8 Parameter sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

7.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

7.3.1 Credit crisis scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

7.3.2 Wholesale and retail bank scenarios . . . . . . . . . . . . . . . . . . . . . . 130

7.3.3 The impact on credit supply . . . . . . . . . . . . . . . . . . . . . . . . . . 131

7.3.4 The influence of central bank interventions . . . . . . . . . . . . . . . . . . 132

7.3.5 Effects of adjusting to the new liquidity regulation . . . . . . . . . . . . . . . 135

7.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

Appendix 7.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

Appendix 7.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

Part III: Policy responses

8 Crisis measures and limiting possible distortions 143

8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

8.2 Crisis measures to address market failure . . . . . . . . . . . . . . . . . . . . . . . . 144

8.3 Proper design of support policies essential but complicated . . . . . . . . . . . . . . 146

8.4 Market conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

8.4.1 Impact on level playing field financial sector . . . . . . . . . . . . . . . . . . 147

8.4.2 Distortionary effects on markets and business models . . . . . . . . . . . . . 149

8.4.3 Cross-border effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151

8.5 External effects, negative impact on confidence . . . . . . . . . . . . . . . . . . . . 152

8.5.1 Investor confidence in supported institutions . . . . . . . . . . . . . . . . . . 152

8.5.2 Confidence in the creditworthiness of governments . . . . . . . . . . . . . . . 153

8. 5.3 Risks for the central bank . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154

8.6 Longer-term distortions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

8.6.1 Moral hazard among management . . . . . . . . . . . . . . . . . . . . . . . 155

8.6.2 Moral hazard among investors . . . . . . . . . . . . . . . . . . . . . . . . . 156

8.6.3 Moral hazard from deposit insurance . . . . . . . . . . . . . . . . . . . . . . 157

8.6.4 Moral hazard from extremely low interest rates . . . . . . . . . . . . . . . . 157

8.7 Policy instruments to limit distortions . . . . . . . . . . . . . . . . . . . . . . . . . 158

8.7.1 Market-compatible conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 158

8.7.2 International harmonisation . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

8.7.3 The importance of providing clarity . . . . . . . . . . . . . . . . . . . . . . . 159

8.7.4 Relation between government and management . . . . . . . . . . . . . . . . 159

8.7.5 Involvement of the private sector . . . . . . . . . . . . . . . . . . . . . . . . 159

8.7.6 Temporary nature of support and exit . . . . . . . . . . . . . . . . . . . . . . 160

8.7.7 Prudential supervision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

8.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

9 Macro-effects of higher capital and liquidity requirements for banks:

Empirical evidence for the Netherlands 163

9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163

9.2 New regulatory standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164

9.3 Channels of effects during the transitional phase . . . . . . . . . . . . . . . . . . . . 166

9.3.1 Direct consequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166

9.3.2 Effect on lending via interest rate channel . . . . . . . . . . . . . . . . . . . 166

9.3.3 Effect on credit supply via bank capital channel . . . . . . . . . . . . . . . . 167

9.3.4 Influence on risk behaviour of banks . . . . . . . . . . . . . . . . . . . . . . 168

9.3.5 Broader effects of liquidity requirements . . . . . . . . . . . . . . . . . . . . 168

9. 3.6 Impact on financial markets . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

9.4 Effects during the transitional phase: model outcomes for the Netherlands . . . . . . . 169

9.4.1 Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

9.4.2 Satellite models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

9.4.3 Simulation outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173

9.4.4 Simulations using macro-econometric model . . . . . . . . . . . . . . . . . . 175

9.4.5 Time series analysis using a VAR model . . . . . . . . . . . . . . . . . . . . 176

9.4.6 International perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178

9.5 Effects in a new steady state with higher buffers . . . . . . . . . . . . . . . . . . . . 180

9.5.1 Higher buffer requirements: costs and benefits . . . . . . . . . . . . . . . . . 180

9.5.2 More stable economic development . . . . . . . . . . . . . . . . . . . . . . 184

9.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185

Appendix 9.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186

Appendix 9.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187

10 Summary and conclusions 189

10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189

10.2 Bank behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190

10.3 Macro stress-testing models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191

10.4 Policy responses to the crisis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193

References 195

Samenvatting (Summary in Dutch) 209

List of Figures

2.1 Development of balance sheet items . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.2 Size: average relative change balance sheet adjustment . . . . . . . . . . . . . . . . . . 18

2.3 Size: median relative change balance sheet adjustment . . . . . . . . . . . . . . . . . . 18

2.4 Correlation of item changes across banks . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.5 Number of banks, extreme balance sheet adjustments . . . . . . . . . . . . . . . . . . . 21

2.6 Number of banks, direction balance sheet adjustments . . . . . . . . . . . . . . . . . . 21

2.7 Number of extreme balance sheet adjustments . . . . . . . . . . . . . . . . . . . . . . 22

2.8 Direction of balance sheet adjustments . . . . . . . . . . . . . . . . . . . . . . . . . 22

2.9 Factor analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.1 Stylized bank balance sheet: Possible asset side responses . . . . . . . . . . . . . . . . 32

3.2 Development of model variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

3.3 Adjustment of lending, sample of 17 banks . . . . . . . . . . . . . . . . . . . . . . . . 41

3.4 Liquidity hoarding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

3.5 Fire sales , sample of 17 banks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

3.6 Fire sales and solvency, sample of 17 banks . . . . . . . . . . . . . . . . . . . . . . . 47

3.7 Credit default swap spreads Dutch banks . . . . . . . . . . . . . . . . . . . . . . . . . 50

4.1 Dimensions of stress-testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

5.1 Stress-testing framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

5.2 Outcomes deterministic scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

5.3 Robustness checks AR and VAR models . . . . . . . . . . . . . . . . . . . . . . . . . 77

5.4 Outcomes stochastic scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

6.1 Flow chart of Liquidity Stress-Tester . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

6.2 Systemic effects through contagion channels . . . . . . . . . . . . . . . . . . . . . . . 91

6.3 Frequency distributions of risk aversion indicators . . . . . . . . . . . . . . . . . . . . 96

6.4 Relationships between model parameters . . . . . . . . . . . . . . . . . . . . . . . . . 97

6.5 Distribution of buffers after scenario rounds . . . . . . . . . . . . . . . . . . . . . . . 99

6.6 Bank-testing the scenario outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . 107

7.1 Model framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

7.2 Effect asset purchases central bank . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

7.3 Impact on credit supply . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

7.4 Central bank interventions, mitigating influence . . . . . . . . . . . . . . . . . . . . 134

7.5 Impact of exit from extended central bank operations . . . . . . . . . . . . . . . . . . 134

7.6 Influence of stronger liquidity profiles . . . . . . . . . . . . . . . . . . . . . . . . . 136

8.1 Market shares Dutch deposit market . . . . . . . . . . . . . . . . . . . . . . . . . . 148

8.2 Money market rate and trading volume (euro area) . . . . . . . . . . . . . . . . . . . 151

8.3 Stock prices supported vs. non-supported financial institutions, worldwide . . . . . . 153

8.4 CDS premium supported vs. non-supported financial institutions, worldwide . . . . . 153

8.5 Dependence between banks and governments . . . . . . . . . . . . . . . . . . . . . . 154

9.1 Effects of new supervisory standards . . . . . . . . . . . . . . . . . . . . . . . . . . 166

9.2 Balance sheet adjustments, percentage point higher TCE/RWA ratio . . . . . . . . . . 174

9.3 Impact on loan spread of rising liquidity ratio, percentage point higher TCE/RWA ratio . 175

9.4 Real GDP impact increase TCE / RWA ratio, structural model . . . . . . . . . . . . . 176

9.5 Real GDP impact increase liquidity ratio, structural model . . . . . . . . . . . . . . . 176

9.6 Real GDP impact increase TCE / RWA ratio, VAR model . . . . . . . . . . . . . . . 178

9.7 Probability of systemic crisis at various buffer levels . . . . . . . . . . . . . . . . . . 182

9.8 Solvency and liquidity of Dutch banks in historical perspective, 1990-2009 . . . . . . 184

List of Tables

2.1 Pecking order of balance sheet adjustments . . . . . . . . . . . . . . . . . . . . . . . . 17

3.1 Summary statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

5.1 Estimation result of regression of default rate (equation 5.1) . . . . . . . . . . . . . . . 70

5.2 Estimation result of regression of LLP (equations 5.5 - 7.7) . . . . . . . . . . . . . . . 71

5.3 Estimation result of regression of Net Interest Income (equation 5.8) . . . . . . . . . . . 73

6.1 Correlation between changes of buffer and balance sheet items . . . . . . . . . . . . . 94

6.2 Parameter sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

6.3 Outcomes scenario simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

6.4 Parameter sensitivity credit crisis scenario . . . . . . . . . . . . . . . . . . . . . . . 106

7.1 Parameter sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

7.2 Outcomes at different reaction triggers . . . . . . . . . . . . . . . . . . . . . . . . . 128

7.3 Outcomes scenario runs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

8.1 Government support to banks and central banks’ balance sheet . . . . . . . . . . . . 144

8.2 Policy measures during the crisis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

8.3 Distortions and mitigating instruments . . . . . . . . . . . . . . . . . . . . . . . . . 147

8.4 Composition collateral pledged at Eurosystem . . . . . . . . . . . . . . . . . . . . . 155

9.1 Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

9.2 Estimation outcomes satellite model for balance sheet adjustments . . . . . . . . . . . 172

9.3 Estimation outcomes satellite model for loan spread . . . . . . . . . . . . . . . . . . 173

9.4 Equity of banks versus other enterprises in the Netherlands, 2000-2008 . . . . . . . . 183

1

Chapter 1

Introduction

This thesis brings together research on credit and liquidity risks of banks in stress conditions. It

investigates banks’ reactions to those risks, presents macro stress-testing models and analyses policy

measures to contain the risks during the 2007-2009 financial crisis. Banks are important financial

intermediaries because of their risk and maturity transformation function. This inherently exposes

them to credit risk, which is basically the risk of default on loans, and liquidity risk, i.e. the risk of

funding drying up and reduced tradability of assets. The recent financial crisis has shown that both

credit and liquidity risk can assume a systemic dimension in times of stress and can undermine

financial stability and economic growth. A thorough understanding of the transmission channels

through which credit and liquidity risks affect the banking sector is needed to analyse and address the

causes of the crisis. This goes beyond an analysis of traditional measures that are based on balance

sheet information, such as non-performing loans, liquidity and capital ratios. Rather a macro

perspective is needed, which views the banking sector in relation to the economic environment and

other financial sectors. Due to the crisis, authorities responsible for financial stability have realised this

and have been adopting a ‘macroprudential’ approach, which aims at supervising the financial system

as a whole, in the context of the environment (De Larosière Report, 2009). The research in this thesis

also takes a macro perspective, by concentrating on the features of credit and liquidity risk, and the

interaction between both risk factors, which have the potential to move the financial system into the

tail of the loss distribution.

1.1 Context

The context of this thesis can be explained by the role of banks in the financial cycle. The cycle is

driven by the various components of the financial system that are mutually dependent and interact

through various transmission channels. In the economic and market environment of banks, debts may

rise and asset prices can become overvalued. Banks themselves may contribute to such financial

imbalances, by excessive lending and extended financial market exposures. Generally, these financial

activities are pro-cyclical due to the financial accelerator effect (Bernanke et al., 1996). This effect

amplifies and propagates shocks to the economy and works through endogenous developments in

credit markets. Key mechanisms for the pro-cyclicality are the external finance premium and the net

worth of borrowers, which both fluctuate with the financial cycle. Shocks can also have magnifying

effects on the economy through banks’ balance sheets. This channel may arise from changes in

2

monetary and regulatory policy or capital losses, which may force banks to adjust the cost and supply

of credit. Next to that, the recent crisis has raised the importance of the liquidity channel, which could

reinforce the traditional bank balance sheet channel or create additional transmission channels. In

literature on the liquidity channel, high leverage ratios and maturity mismatches in banks’ balance

sheets are considered determinants of the propagation of funding liquidity shocks to bank lending and

the real economy (see, for instance, Gauthier et al., 2010). The potential for pro-cyclical reactions by

banks has grown over the last decade, due to their increased reliance on wholesale funding and the

practise of mark-to-market valuation of assets. These factors act as mechanisms that transmit high

market volatility to banks’ balance sheets, through swings in the prices of their assets and in the

availability of liquidity, as was underscored by the recent crisis (FSF, 2008).

The exposures of banks on other sectors of the economy - in particular if they are the result of

a pro-cyclical spiral - involve risks, of which credit risk and liquidity risk are usually dominant. Credit

risk can become manifest through defaults on loans or downgradings of bonds outstanding on the asset

side of banks’ balance sheets. If the exposures have to be written-down, the solvency position of banks

will be affected. In the credit crisis, banks had to write-off over USD 1,600 billion worldwide, mainly

on credit exposures (IMF, 2010b). This major shock destabilised the sector and necessitated

government injections of capital into a number of banks. Liquidity risks can become manifest in a

drying up of funding sources (‘funding liquidity’), for instance related to a decline of retail deposits.

Another manifestation of liquidity risk is an evaporation of liquidity on financial markets (‘market

liquidity’), leading to reduced tradability of assets and mark-to-market losses on bank exposures.

Declining funding and market liquidity can culminate in liquidity problems and may impair the

intermediary function of banks. This liquidity channel was an important mechanism in the recent crisis

(IMF, 2008b). Credit and liquidity risks both assumed a systemic dimension, because financial

imbalances were built up on a large scale, which caused massive credit losses and serious liquidity

drains in the downturn.

If banks react to emerging risks by reducing the liquidity supply to wholesale counterparties or

the credit supply to companies and households, adverse feedback effects on financial markets or the

economy emerge. Such effects can reinforce the downturn in markets or the economy. In the recent

financial crisis the economy in the euro area was sheltered to some extent, since European banks

reacted foremost by withdrawing their exposures in wholesale markets and less so by reducing credit

supply to the real economy (Giannone et al., 2011). Banks’ reactions may have systemic repercussions

within the banking sector too, for instance if interbank credit exposures are adjusted. In the 2007-2009

crisis, the strong risk aversion among banks indeed led to a freeze of the interbank market, which was

evident by a drop in trading volumes and a hike of credit spreads, in particular for funding with

medium and long-term maturities (Cassola et al., 2010). This was the main reason for central banks to

extend their intermediary role in the market, by supplying additional liquidity through unconventional

measures.

3

The credit crisis was a unique tail event, driven by adverse developments in credit and

liquidity risk that will probably not recur in the same form. However, the episode has provided

valuable data on the role of bank behaviour and adverse feedback loops, two important channels

through which risks can assume a systemic dimension. A deeper understanding of these dynamics

helps to assess the relevant risks from a macro perspective, which is crucial information for central

banks and financial regulators to take adequate measures to contain a financial crisis or to prevent the

next one. An important policy response after the 2007-2009 crisis has been the new supervisory

framework for banks (“Basel III”). It aims to enhance the shock absorption capacity of banks and

encourages countercyclical behaviour (BCBS, 2010a).

1.2 Research questions

The thesis concentrates on three closely related research questions that emerge from the developments

in the banking sector as presented in the previous section.

1. How did banks adjust their credit and liquidity risk management during the crisis and how do

empirical estimates of banks’ reactions relate to the behavioural assumptions as generally used in

the theoretical literature? The crisis has stimulated new research on phenomena that can explain

departures from the efficient market hypothesis (EMH).1 The validity of the EMH has been

questioned due to the excessive behaviour of market participants that contributed to the bubble

preceding the crisis and the subsequent crash. Departures from the EMH relate, for instance, to

herding, fire sales, leveraging and risk taking, externalities and liquidity spirals (Brunnermeier,

2001). Furthermore, leveraging and deleveraging, driven by behavioural incentives, have been

important amplifying forces in the crisis (Adrian and Shin, 2010). However, most research in this

field is theoretical and lacks an empirical underpinning, although the crisis provides a rich set of

data on such phenomena.

2. How can the impact of tail events on banks that involve credit and liquidity risk, and banks’

reactions to those risks, be modelled? Tail risks are typically characterised by correlation break-

downs, non-linear developments and feedback effects. Such elements provide another argument to

depart from the EMH, which assumes representative agents and the system being in equilibrium.

To capture the dynamics that are typical for extreme situations, non-standard analytical

1 The efficient market hypothesis (EMH) assumes that market participants are fully rational and make sensible decisions based on all information available in the market (Ross, 2005). The EMH implies that agents understand the underlying model structure and the distribution of shocks. In fact, the literature has recognised that perfectly efficient markets are a theoretically useful benchmark rather than an accurate characterisation of reality.

4

frameworks that are not yet widely accepted in economics can be useful (Trichet, 2011).

Disequilibrium models provide for such a framework, for instance heterogeneous agent models

that focus on the interactions of bounded rational agents (Hommes, 2006). These models describe

the behaviour of investors that may follow rule of thumb strategies and they are mostly concerned

with financial market applications. A different, more policy oriented, approach to model tail

events is represented by stress-testing models. They provide a framework to analyse the impact of

tail events on banks and their reactions to stress events. The latter is usually not captured in

traditional stress-testing models, but requires a more advanced approach that includes feedback

effects of banks’ management actions on the economy and/or the financial system.

3. How should the policy responses to the eruption of credit and liquidity risks during the 2007-2009

crisis be assessed, both with regard to the possible distortionary effects on behavioural incentives

and the impact on the economy? In the crisis there were no blueprints available for central banks,

governments and supervisors to optimally respond to the rapidly evolving threats to financial

stability. They had to act under great uncertainties and were in unchartered territory. Although

policymakers were aware of possible undesired effects of the measures they took and the new

regulation they developed, the extent of the market distortions and economic costs were uncertain.

Research on these side-effects is scarce, although such analysis is very useful to guide

policymakers in the design and calibration of their responses to (future) financial crises.

1.3 Research approach

In this thesis the research questions are addressed by analytical instruments that are not (yet) part of

the standard paradigm of economic modelling. The first research question is addressed by analysing

the credit and liquidity management of banks empirically from a bottom-up perspective. This means

that we use granular data to capture the variations at the portfolio or bank level and differing responses

of banks. In particular, we use firm-specific data of Dutch banks, derived from a unique data source on

assets and liabilities available at De Nederlandsche Bank (DNB). Based on these micro observations

we investigate general trends in bank behaviour, by specifying indicators and time series models,

which capture the responses of banks to stress situations. A multi-equation time series approach (panel

vector autoregressive (VAR) model) is used to take into account the dynamic interrelations among

instruments of bank liquidity management. The empirical techniques map the micro information to the

level of the banking system. So they capture both the time dimension (‘pro-cyclicality’) and cross-

sectional dimension (‘dependencies’) of systemic risk (Borio, 2006). We analyse concrete

manifestations of these dimensions, in particular the size and direction of balance sheet adjustments,

herding, liquidity hoarding and fire sales. Moreover, the empirical approach provides insight in the

5

interactions that pose a risk to the financial system and the economy, such as correlated balance sheet

adjustments and the linkage between funding liquidity risk and bank lending.

The second research question is addressed by modelling the interactions between tail events in the

external environment on the one hand and banks’ credit and liquidity risk on the other in a stress-

testing framework. This provides for methodologies to map stress in the macro-environment into

indicators that can be used to estimate the impact of stress scenarios on the balance sheets of banks,

their responses to stress and the related second round effects in the financial system and the economy.

In the literature those steps are covered by two types of models; i) those that establish the link between

macro variables and micro risk drivers and ii) integrated models that include liquidity risk and

feedback effects (Foglia, 2009). The methods used in this thesis enclose these two strands and are

operationalised by a suite of models, such as reduced form satellite models (that specify a particular

relationship between a balance sheet indicator and macroeconomic variables by a single equation),

VAR models and calibrated simulation tools. This eclectic approach is motivated by the absence of a

fully fledged model that integrates all the potential interlinkages. For that reason stress-testing

frameworks are not single coherent economic models, but typically a combination of separate modules

(ECB, 2010d). Moreover, according to Knight (1921), there is no distribution of probabilities of

extreme events, which implies that models are prone to large parameter and model uncertainties. In

combination with the scarcity of data on tail events, this makes standard regression techniques less

appropriate for modelling credit risk and even less so for modelling liquidity risk in tail situations.

Hence we use several, most fairly basic, simulation and stress-testing techniques to analyse credit and

liquidity risks and the dynamics related to banks’ responses to stress events.

The modelling approach has some limitations. First, there is not yet a generally accepted

analytical framework for macro stress-testing, implying that simulation techniques have subjective

elements, for instance regarding the assumptions concerning the behaviour of banks. This well-known

problem in behavioural economics is obviated to some extent by keeping the behavioural assumptions

simple and by embedding them in a plausible story, based on the empirical insights from the analysis

conducted in the context of the first research question. Second, there is fundamental uncertainty with

regard to risks in tail situations, implying that the model outcomes are also surrounded by large

uncertainties. This caveat is taken into account by the use of scenario analyses and loss distributions,

including quantifications of losses in the extreme tail. Despite its limitations, the stress-testing

framework enhances the understanding on how credit and liquidity risks may evolve in crises, how

they affect banks and what the possible feedback effects of banks’ reactions are on the financial

system and the economy. For that reason, stress-testing has evolved as a crisis management instrument

for supervisors and central banks, to shape crisis measures and restore market confidence.

6

This leads us to the third research question, i.e. the assessment of policy responses to the crisis. First,

this is addressed by analysing the short-term crisis measures taken by central banks and governments

in 2007-2009, with a particular attention to the undesired side-effects on the incentives of market

participants and the functioning of the financial sector. We investigate the empirical evidence of

potential distortionary effects as found in market prices, trading volumes and market shares of

financial institutions. Second, the effects of longer-term measures taken by regulators in response to

the crisis are analysed, in particular the macroeconomic effects of Basel III. The new regulatory

standards for capital and liquidity will affect bank behaviour and lending and thereby the economy,

both in the transitional phase and in the new steady state. The (admittedly uncertain) effects on lending

in the transition phase are simulated with reduced form satellite models and DNB’s structural

macroeconomic model of the Dutch economy. The effects in the steady state are even more uncertain,

given the as yet unknown possible adjustments of banks’ business models. Therefore, we rely on

simulation outcomes and conceptual insights from the literature to assess possible effects in the new

steady state.

1.4 Outline

The outline of the thesis is as follows. Chapters 2 and 3 deal with the first research question on bank

behaviour during the crisis. The second research question, on modelling the impact of shocks in credit

and liquidity risk on banks and their reactions to those risks in a stress-testing framework, is covered

in Chapters 4 to 7. The third research question, which focuses on the responses of policymakers, is

addressed in Chapters 8 and 9.

1.4.1 Bank behaviour during the crisis

Chapter 2 provides empirical evidence of behavioural responses by banks in the recent crisis. This is

based on aggregate indicators of macroprudential risk, constructed from firm-specific balance sheet

data. The indicators provide tools to empirically test the concepts of macroprudential risks, i.e. the

time dimension and the cross-sectional dimension, as described by Borio (2006). The empirical

measures of size and herding show that balance sheet adjustments have been pro-cyclical in the crisis,

while responses became increasingly dependent across banks and concentrated on certain market

segments. The analysis shows that while banks usually follow a pecking order in their balance sheet

adjustments (by making larger adjustments to the most liquid balance sheet items compared to less

liquid items), in the crisis they were more inclined to a static response (by reacting with instruments

proportional to the composition of their balance sheet). This suggests that banks have less room for a

pecking order in their liquidity risk management during stressed circumstances. The chapter concludes

7

that the behavioural indicators are useful tools for monetary and macroprudential analyses and argues

that they can contribute to the micro foundations of financial stability models.

Chapter 3 extends the empirical evidence on bank behaviour to the relationship between

funding liquidity and adjustments on the asset side of banks’ balance sheets. This liquidity channel of

financial transmission focuses on lending adjustments, liquidity hoarding and fire sales. These

behavioural concepts from the literature on disequilibrium models are empirically tested by a panel

VAR model using data of Dutch banks. The model takes the endogeneity between asset side

adjustments and funding into account. Orthogonalized impulse responses show that in reaction to

shocks in money market spreads and repo funding, banks reduce wholesale lending in favour of retail

lending. Moreover, a shock to wholesale funding costs urges banks to hoard liquidity through

accumulating liquid bond holdings and central bank reserves. Another insight is that fire sales of

equity holdings are more likely to be triggered by constraints in funding liquidity than by constraints

in banks’ solvency position.

1.4.2 Macro stress-testing

Chapter 4 gives an overview of stress-testing methods for banks, based on the literature and policy

practise, focusing on macro stress-testing. It distinguishes micro from macro stress-testing and bottom-

up from top-down approaches. The chapter assesses the different use of bottom-up macro stress-tests

by authorities in Europe and the US during the crisis. The overview of top-down stress-testing

approaches covers the range of modelling approaches developed by central banks to link macro

variables to micro risk drivers in bank portfolios (mainly applied to credit risk) and integrated models

that include liquidity risk and feedback effects within the financial sector.

Chapters 5, 6 and 7 present several applications of macro stress-testing models for credit and

liquidity risk. Chapter 5 presents a tool kit for scenario analysis and macro stress-testing of credit risk.

First, macroeconomic scenarios are simulated by a structural macroeconomic model. The scenarios are

then mapped in banks’ loan portfolios, by estimating reduced form satellite models that link the

exogenous shocks in the macro variables to micro drivers of bank risk, i.e. credit quality indicators and

an interest income measure. To capture the different responses of banks and portfolios in stress

situations, we use disaggregated data for a panel of individual banks and a break-down of domestic

and foreign portfolios. We further explore the variation in the credit loss distribution by estimating

both the probability of default (PD) and the loss given default (LGD) in bank loan books by using

nonlinear specifications. An important contribution to the literature is that we propose an additional

alternative approach for scenario simulation, based on a vector autoregressive (VAR) model. It allows

for simultaneous changes in the macro variables and portfolio drivers of bank loans and changing

correlations between them in stress situations. The stochastic VAR simulations generate loss

distributions that provide insight in the extent of possible extreme losses.

8

Chapter 6 presents a stress-testing model for liquidity risks of banks. It focuses on both market

and funding liquidity risk and combines the multiple dimensions of liquidity risk into a quantitative

measure of banks’ liquidity position. It takes into account the first and second round (feedback) effects

of shocks, induced by reactions of heterogeneous banks, and reputation effects. The impact on

liquidity buffers and the probability of a liquidity shortfall is simulated by a Monte Carlo approach.

An application to Dutch banks illustrates that the second round effects in specific scenarios could have

more impact than the first round effects and hit all types of banks, indicating systemic risk.

Chapter 7 extends the liquidity stress-testing model, by incorporating the proposed Basel III

liquidity regulation, unconventional monetary policy measures and credit supply effects. In the

extended model, banks react according to the Basel III standards, endogenising liquidity risk. The

model shows how banks’ reactions interact with extended refinancing operations and asset purchases

by the central bank. The results indicate that Basel III limits liquidity tail risk, in particular if it leads

to a higher quality of liquid asset holdings. The flip side of increased bond holdings by banks is that

monetary policy conducted through asset purchases gets more influence on banks’ balance sheets

relative to refinancing operations.

1.4.3 Policy responses

Chapter 8 analyses the (temporary) financial crisis measures taken between 2007 and 2009 by central

banks and governments, including the potential distortionary effects on behavioural incentives and

market functioning. We assess the effects of the policy interventions on the level playing field between

supported and non-supported institutions, on the capital flows between market segments and on

investor confidence. The chapter shows that in the longer term, interventions by governments and

central banks may lead to excessive risk taking and moral hazard problems. The main policy

conclusion is that such negative side effects can be limited in the design of the support measures

(market compatibility) and in the exit strategy (timely withdrawal).

Chapter 9 analyses the long-term measures taken by regulators to strengthen the banking

sector. In particular the effects of Basel III on the economy are simulated, both in the transitional

phase and in the new steady state. Outcomes of satellite models for lending and loan spreads are used

as input in DNB’s structural macroeconomic model. The results of the exercise indicate that the

negative impact on Gross Domestic Product (GDP) during the transitional phase to higher capital and

liquidity buffers will be limited to a few tenths of a percent. Another conclusion is that a sufficiently

long transitional period will limit the costs in the early years, because it will give banks more scope to

adapt. Insights from the literature indicate that once banks have adapted to the new standards, the

benefits of a more solid financial system will outweigh the disadvantages, since higher buffers make a

financial crisis in the future both less likely and less deep, while economic growth will be more stable

in normal times.

Chapter 10 concludes.

9

Chapter 2

When liquidity risk becomes a systemic issue:

Empirical evidence of bank behaviour

2.1 Introduction

2.1.1 Dimensions of liquidity risk

The 2007-2009 crisis showed that liquidity risk stemming from collective reactions by market

participants can exacerbate financial instability.2 Liquidity hoarding by funding constrained banks

added to the tense liquidity situation in financial markets, underscoring the strong link between banks’

funding risk (the ability to raise cash to fund asset holdings, see Matz and Neu (2007) and Drehmann

and Nikolau (2010)) and market liquidity (the ability to convert assets into cash at a given price at

short notice). Through this channel liquidity risk led to solvency problems and banks had to write off

illiquid assets. These developments have induced policymakers to focus on the interactions between

funding and market liquidity risk and related systemic risk, as part of the macroprudential approach

(De Larosière Report, 2009). Getting a better grip on such dynamics requires an understanding of

firms’ behaviour on a micro level in relation to macro-financial developments. In practice, liquidity

risk is either analysed, managed and regulated from the perspective of banks’ funding positions (e.g.

by supervisors) or on the level of the financial system as a whole (by central banks). However, recent

events have underscored that systemic risk can originate at the nexus of funding and market liquidity

and is influenced by market participants who react to market-wide shocks.

The relevance of behavioural reactions of market participants for financial stability is

recognised in literature describing the macroprudential approach. According to Borio (2006), this

approach focuses on the financial system as a whole, including the underlying correlations.

Dependencies relate to similar investments and risk management strategies of financial institutions

that have common exposures. This cross-sectional dimension is measured by the correlation between

institutions’ balance sheets and by the marginal contribution of each institution to total systemic risk

(Borio et al., 2010). Next to this cross-sectional dimension of systemic risk, the macroprudential

approach distinguishes the time dimension. This concerns how risks evolve over time (which can be

measured by macroeconomic variables like credit growth (BIS, 2009c)) and whether pro-cyclicality

plays a role. Pro-cyclicality is caused by collective behaviour of financial institutions that reinforces

the interaction between the financial system and the real economy. In the literature these feedback

mechanisms are attributed to increasing risk tolerance, overextension of balance sheets and high

leverage during an expansion, which are reversed in a downturn. The increased link between market 2 This chapter is a revised version of Tabbae and Van den End (2011).

10

and funding liquidity is another driving factor (e.g. through increased use of collateral in secured

financing).

2.1.2 Modelling bank behaviour

Endogenous cycle models, where risk is endogenous with respect to collective behaviour of market

participants, are still primitive, with very limited behavioural content (Borio and Drehmann, 2009).

This also holds for macro stress-testing models that are used by central banks and supervisory

authorities to simulate shocks to the system as a whole. Even in the most sophisticated stress-testing

models, the behaviour of financial institutions is included by rules of thumb rather than through

empirical estimations. Responses are usually assumed to be triggered by shocks that lead to a

declining solvency ratio of banks below a certain threshold level. This default risk can be caused by a

drying up of market liquidity which depresses the value of banks’ assets, as in Cifuentes et al. (2005).

This triggers fire sales of assets, depresses market prices and induces further sales. In the financial

sector model of the Bank of England, behavioural responses are related to funding liquidity risks of

banks (Aikman et al., 2009). In this model, funding strains increase the default risk of banks, which at

a certain stress level resort to fire sales of assets. This leads to liquidity feedbacks through depressed

market prices of assets. In the Liquidity Stress-Tester model described in Chapters 6 and 7, banks’

responses are triggered by a certain decline of the liquidity buffer. The subsequent second round

effects are mechanically determined by the number and size of reacting banks and the similarity of

their reactions.

Stress-testing models often lack empirical foundations of bank behaviour. For this,

information on the effects of management actions on the stability of the financial system and the

economy is required, based on balance sheet data and market indicators in extreme situations. The

recent crisis provides a rich set of such data, which helps to assess behavioural responses by banks and

their contribution to system-wide liquidity stress.

2.1.3 Contribution to the literature

This chapter contributes to the literature by exploring data on bank behaviour in the crisis, with the

focus on liquidity risk. Thereby, it differs from the approach of Berger and Bouwman (2009), who

focus on the liquidity creation by banks. They present four measures of liquidity creation, based on the

category and maturity of assets and liabilities of US banks. From a utility perspective, the measures

judge the creation of illiquid assets out of liquid liabilities by banks as positive, since it reflects the

ability of customers to obtain liquidity from banks. In contrast to this liquidity creation theory, we take

a liquidity risk perspective. This implies that liquidity creation that goes with an increasing maturity

mismatch on banks’ balance sheets is judged negatively.

Our approach is based on a unique dataset from the Dutch supervisory liquidity report, which

comprises a detailed break-down of liquid assets and liabilities, including cash in- and outflows. Since

11

the dataset is collected from banks, it mainly provides information on funding liquidity risk. We first

analyse the type of instruments used by Dutch banks in response to the crisis. Statistical tests show

that while banks usually follow a pecking order in their balance sheet adjustments (by making larger

adjustments to the most liquid balance sheet items compared to less liquid items), in the crisis they

were more inclined to a static response (by reacting with instruments proportional to the composition

of their balance sheet). This suggests that banks have less room for a pecking order in their liquidity

risk management during stressed circumstances.

Next, we construct aggregate measures of bank behaviour with data of individual banks. This

follows the macroprudential approach. Herein, risks to the financial system can either be measured by

aggregate balance sheet indicators (IMF, 2008a), market prices, or by composite indicators. All of

these have their limitations in assessing common exposures and interactions (see Borio and Drehmann,

2009). We go a step further than traditional balance sheet indicators, by defining measures for

behavioural reactions and testing them empirically. The main drivers of systemic risk are measured,

i.e. the time dimension and the cross-sectional dimension. The time dimension is quantified by

indicators of size and direction of balance sheet adjustments. The cross-sectional dimension by

indicators of the dependence of behaviour across banks. Herding is seen as having both a time and

cross-sectional dimension; as a point in time indicator it reflects the commonality of exposures, while

the time series of the herding indicator reflects whether bank’s reactions are pro-cyclical. By analysing

the measures for size and herding also on an instrument by instrument basis, we assess the similarity

of reactions and the concentration of trades in particular market segments, indicative of the cross-

sectional dimension of macroprudential risk. We show that the indicators are robust to different

specifications and distributions of the data. Applied to Dutch banks, the measures illustrate that the

size and number of responses (time dimension) and the dependence (cross-sectional dimension)

substantially changed in the crisis.

The structure of the chapter is as follows. Section 2.2 describes the data and developments of

liquid assets and liabilities during the crisis. In Section 2.3, measures of bank behaviour are

constructed and statistical tests on the type of instruments used to react are performed. Section 2.4

concludes.

2.2 Data and trends

2.2.1 Data

Our analysis of bank behaviour is based on a unique dataset from the Dutch supervisory liquidity

report. It covers a detailed break-down of liquid assets and liabilities including cash in- and outflows

of banks. The report includes on and off-balance sheet items for all Dutch banks (85 on average,

including subsidiaries of foreign banks) with a rather detailed break-down per item (average

12

granularity of around 7 items per bank). The report contains end of month data, which are available for

the 2003m10 - 2009m3 period. Appendix 2.1 provides a detailed overview of the items in the report.

According to the supervisory requirements, actual liquidity of a bank must exceed required liquidity,

at both a one week and a one month horizon. Actual liquidity is defined as the stock of liquid assets

(weighted for haircuts) and recognised cash inflows (weighted for their liquidity value) during the test

period. Required liquidity is defined as the assumed calls on contingent liquidity lines, assumed

withdrawals of deposits, drying up of wholesale funding and liabilities due to derivatives. In this way,

the liquidity report comprises a combined stock and cash flow approach, in which respect it is a

forward looking concept. The weights (wi) of the assumed haircuts on liquid assets and run-off rates of

liabilities are presented in last two columns of the table in Appendix 2.1. In the report, the weights are

fixed values (DNB, 2003) and reflect a mix of a bank specific and market wide scenario. The values of

wi are based on best practices of values of haircuts on liquid assets and run-off rates of liabilities of the

industry and rating agencies. This differs from the liquidity weights attached to assets and liabilities in

the study of Berger and Bouwman (2009), in which three classes of weights are used that have more or

less arbitrary values (-0.5, 0, 0.5).

The various balance sheet and cash flow items in the prudential report are assumed to reflect

the instruments (i) which banks (b) use in the liquidity risk management in response to shocks. The

instruments are expressed in gross amounts (Ibi). To enhance the economic interpretation we define

coherent groups (g) of instruments and the sum of item amounts per group as Ibig. The first column of

the table in Appendix 2.1 provides the group classification (items not classified were deemed to be

irrelevant for the analysis in this chapter). Figure 2.1 shows the time series of the instrument groups.

13

Figure 2.1. Development of balance sheet items

2.2.2 Trends

In the remainder of this chapter we distinguish three stages of the crisis, in line with the description by

González-Páramo (2009). In each phase, market and funding liquidity risks had different dimensions

to which the banks reacted in their liquidity management actions. In the first stage (August 2007 until

the demise of Bear Stearns in March 2008), market liquidity dried up, which interacted with the

funding liquidity of banks. The second stage (between March 2008 and the failure of Lehman in

September 2008) was characterised by increased counterparty risks, due to concerns about bank

failures. In the third phase (October 2008 until March 2009) liquidity stress was accompanied by

solvency problems of banks, on the back of the economic recession. Since March 2009, financial

markets have shown a recovery.

14

Several trends in the behaviour of banks appear from the graphs in Figure 2.1. The most

obvious response of banks to the crisis was a marked reduction of secured lending and borrowing

(repo, reverse repo and securities lending transactions, Figure 2.1, panels B and D). These exposures

grew rapidly before the crisis, when booming asset prices stirred up credit supply that was

collateralised by tradable securities. The sharp decline since the crisis reflects the acute stress on

markets, in particular the drying up of the interbank market, which urged banks to shrink their balance

sheets and to deleverage. Secured positions were reduced in particular in the second stage of the crisis,

when counterparty risks mounted and only collateral of the highest quality was accepted in repo

transactions. These developments contrast to those of unsecured wholesale exposures, which still grew

until Autumn 2008 and only modestly declined in the third stage of the crisis (Figure 2.1 panels A and

D).

One would expect that unsecured positions are more sensitive to market turbulence and risk

aversion than secured positions. However, the data suggest that secured financing was more pro-

cyclical instead, due to its dependence on market values of collateral, of which lower quality assets

that faced large mark-to-market haircuts became unacceptable in repo transactions at some stage.

Moreover, the financing of hedge funds by banks was reduced and this business is usually conducted

against tradable securities as collateral, while institutional investors became reluctant with regard to

securities borrowing and lending (also to banks) for repo transactions. These factors caused a

worldwide decline of activities in the repo market (ICMA, 2009). In contrast to that, unsecured

wholesale lending to non-banks expanded, since these clients drew on their credit lines more actively

(such as strained investors that had to meet margin requirements). Retail credit supply also held up

relatively well; the outstanding amount increased in each stage of the crisis (Figure 2.1 panel B).

Trading portfolios were scaled back in the crisis; fixed income investment in particular in the first

phase and equity holdings primarily since the failure of Lehman (Figure 2.1 panels A and B).

Another observation is that banks changed their funding structure to assure themselves of

liquidity. The decline of (primarily secured) wholesale funding was partly compensated by an increase

of fixed-term retail deposits (Figure 2.1, panel D), which is a more stable source of liquidity than

demand deposits. Liquidity was also assured by an increased reliance on central bank facilities (Figure

2.1 panels A and C). Due to the drying up of the interbank market, the ECB increased its intermediary

function on a large scale. In particular in the second stage of the crisis, banks increasingly borrowed

from the central bank in refinancing operations. This secured the funding needs of banks that were

rationed from the private market and supported the monetary transmission through the banking sector,

since impaired access to funding liquidity can affect credit supply through the bank lending channel

(ECB, 2009c). Besides, banks stepped up their use of the ECB deposit facility, in particular in the later

stages of the crisis. Banks increased their demand for central bank reserves due to a high preference

for risk free liquidity buffers in an environment of increased counterparty risk.

15

2.3 Empirical measures

In this section, the trends in the behaviour of banks are described by empirical measures, based on

firm-specific balance sheet data. First, we assess the type of instruments used, by testing whether

banks reacted according to a pecking order. Second, empirical metrics are constructed to describe the

macroprudential dimensions of behaviour. Measures of the size of reactions are defined to reflect the

time dimension. Measures of the dependence of banks’ responses reflect the cross-sectional

dimension, while the number of reactions (i.e. herding) covers both dimensions. The measures are

applied to different market segments to explain the behaviour of banks in more detail.

2.3.1 Instruments used to react

The influence of behaviour on markets depends on banks’ risk management strategies. One possible

strategy is that banks react according to a pecking order in adjusting the items on their balance sheet.

The pecking order theory developed by Myers and Majluf (1984), applies to the capital structure of

non-financial firms. The theory predicts a strict preference of corporate finance, in which new

investments are financed by internal funds first, then by low risk debt and hybrid securities and by

equity as last resort. This order is explained by the role of asymmetric information, whereby outside

investors have less knowledge on the firm than insiders (owners and/or managers) and demand a risk

premium in their financing. Applying the pecking order theory to the liquidity risk management by

banks, we assume that banks in the first place react by adjusting their most liquid assets and liabilities

and only in the second place by changing less liquid balance sheet items. This assumption is based on

the fact that the most liquid items can more easily be converted to cash at short notice with lower

haircuts and costs than less liquid items. This reflects the tradability on liquid markets (for instance in

the case of government bonds), which reduces asymmetric information and thereby the costs of selling

the liquid assets.

Our version of the pecking order hypothesis is tested empirically by classifying the assets and

liabilities of the banks in our sample according to the month weights in the liquidity report (as

presented in the last column of the table in Appendix 2.1). The weights reflect the liquidity value of

the individual balance sheet items. In Table 2.1 below, we constructed four bands of weights (wb, with

steps of 20) and classified the assets and liabilities accordingly. For each band, the relative month-on-

month change of item values is given, in absolute terms (ii I/I∆ ), averaged over the total number of

banks and months. The pecking order assumption is summarised as the null hypothesis that the relative

average change of items in a higher weighting band (wb) exceeds the change in a lower weighting

band (wb-20),

16

20

200

−

−∆>

∆

wb,i

bw,i

wb,i

bw,i

I

|I

I

|I:H (2.1)

The outcomes in Table 2.1 indicate that over the whole sample period for all banks, the pecking order

hypothesis is confirmed; the relative change ratios of liquid assets and liabilities (in the higher

weighting band) are larger than the changes of less liquid items (in the lower weighting bands).3 Only

in case of the liabilities of small and medium sized banks, the pecking order is less obvious (the

relative change ratio is larger in the 0-20 band than in the 20-40 band in case of small banks and the

relative change ratio is larger in the 20-40 band than in the 40-60 band in case of medium sized

banks). This suggests that small banks more frequently follow a mechanical response rule in which the

balance sheet composition determines the availability and use of liquidity instruments (a reflection of

banks’ specialisation and presence in certain market segments). Table 2.1 also indicates that banks

were more inclined to a static response in the crisis than before. In the 2007m7-2009m3 period, the

difference between the relative changes in the 80-100 and the 60-80 bands for assets (80-100 versus

40-60 band for liabilities) is much smaller than the same differences in 2003m10-2009m3. This

suggests that banks have less room to follow a pecking order in their liquidity risk management in

stressed circumstances. It implies that banks’ responses in crisis periods may have more material

effects on the economy, since a static response rule means that banks are more inclined to adjust their

(less liquid) retail lending and deposits than in normal market conditions.

3 In some weighting bands there are no observations according to the weights presented in the liquidity report in the table in Appendix 2.1. In this table, assets with a 100% weight and liabilities in the 0-20 and 80-100 weighting bands dominate.

17

2.3.2 Size of reactions

Besides the type of instruments which banks use to react, market stability is also influenced by the size

of their transactions. This could indicate the build up (or unwinding) of imbalanced exposures and

leverage, both being potential drivers of pro-cyclicality. Thereby the size of the balance sheet

adjustments represents the time dimension of systemic risk. Size is approximated by Sr,t (weighted

average relative monthly change of balance sheet item groups (I ig), averaged across item groups and

banks) and by Sm,t (median of relative monthly item changes, per bank, per item group),

xn

I

/xn

I

Sig

bt,ig

bb ig

bt,ig

t,r ++=

∑∑∑∑∆ (2.2)

∆∆=

)n()x(

)n()x(

)()(

)()(

bt,ig

bt,ig

bt,ig

bt,ig

t,mI

I....

I

ImedianS

11

11 (2.3)

with x being the total number of balance sheet item groups (ig) and n the total number of banks (b).

Figures 2.2 and 2.3 show that in the years before the crisis, Dutch banks expanded their balance

18

sheets, with Sr,t being positive until the start of the crisis and Sm,t until the end of the second crisis

stage. 4 This concurs with the finding of Berger and Bouwman (2008) that financial crises tend to be

preceded by an abnormal liquidity creation by banks, which then decreases during crises. The

difference between Sr,t and Sm,t indicates that large banks (which dominate Sr,t since this is a weighted

measure) had expanded their balance sheets more vigorously before the crisis than smaller banks and

started to unwind their exposures at an earlier stage. The downward trend indicates the substantial

balance sheet adjustments conducted by the banks. Both measures fell rapidly, even being two

standard deviations below average in the third stage of the crisis. This indicates the asymmetric nature

of banks’ behaviour; being more intense in busts than in booms. The deleveraging process seems to

have run its course since March 2009, the end of the third crisis stage. Large banks were most

advanced in deleveraging their balance sheets, as indicated by the upturn of Sr,t in Figure 2.2.

-0,04

-0,03

-0,02

-0,01

0,00

0,01

0,02

0,03

2004 2005 2006 2007 2008 2009

Figure 2.2. Size: average relative change balance sheet adjustmentSize measure (Sr), 6 month moving average

av + 1*stdev

av - 1*stdev

I II III

av - 2*stdev

-0,01

-0,01

0,00

0,00

0,00

0,01

0,01

2004 2005 2006 2007 2008 2009

Figure 2.3. Size: median relative change balance sheet adjustmentSize measure (Sm), month moving average

av + 1*stdev

av - 1*stdev

I II III

av - 2*stdev

av + 2*stdev

To assess potential distortions due to behavioural actions in more detail, the size measure is applied to

different market segments. This is done by constructing Sr,t(i) for each group of balance sheet items

separately, by dropping the summation sign Σig from the numerators and x from the denominators of

the ratio in equation 2.2. The figures in Appendix 2.2 show that in the crisis, typical balance sheet

adjustments took place in various market segments (based on the item groups as defined in Section

2.2). The main factor behind deleveraging was secured lending and borrowing, with Sr,t(i) being close

to three standard deviations below average at the peak of the crisis in Autumn 2008 (see panels A.5

and A.11 in Appendix 2.2).

4 The value change of a balance sheets item is corrected for changes of related market prices, to reflect volume changes and to get a better indication of deliberate portfolio adjustments. This is only done for exposures held for trading and available for sales, since other exposures are not classified as mark-to-market.

19

2.3.3 Dependence of reactions

The cross-sectional dimension of systemic risk is measured by the dependence of bank behaviour. The

dependence measure Dc,t is determined by the correlation of balance sheet changes across the banks. It

follows the few studies in which causalities between loan portfolios of banks are analysed (like

Nakagawa and Uchida (2007) for Japanese banks and Jain and Gupta (1987) for US banks). Dc,t is

based on the relative changes Sr,t(i) of each balance sheet item group i, j…x, per bank.5 Bilateral

correlations between Sr,t(i) of a bank (b1) and Sr,t(i) of the other banks (b2..n) are calculated per month.

Dc,t is then constructed by averaging the bilateral correlations in the matrix below, for each month

(excluding the diagonal elements),

Bank 1 Bank 2 Bank 3 ….. Bank n

Bank 1 Cor (Sr (b1,i..x) ,Sr (b2,i..x)) Cor (Sr (b1,i..x) ,Sr (b3,i..x)) ….. Cor (Sr (b1,i..x) ,Sr (bn,i..x))Bank 2 Cor (Sr (b2,i..x) ,Sr (b1,i..x)) Cor (Sr (b2,i..x) ,Sr (b3,i..x)) ….. Cor (Sr (b2,i..x) ,Sr (bn,i..x))Bank 3 Cor (Sr (b3,i..x) ,Sr (b1,i..x)) Cor (Sr (b3,i..x) ,Sr (b2,i..x)) ….. Cor (Sr (b3,i..x) ,Sr (bn,i..x))

….. ….. ….. ….. ….. …..Bank n Cor (Sr (bn,i..x) ,Sr (b1,i..x)) Cor (Sr (bn,i..x) ,Sr (b2,i..x)) Cor (Sr (bn,i..x) ,Sr (b3,i..x)) …..

Dc,t measures the commonality of balance sheet adjustment across banks. It shows to what extent

banks adjust their balance sheets together. The measure reflects collective behaviour that could disrupt

the functioning of markets. Figure 2.4 shows that Dc,t peaked in the crisis, after accelerating in the first

stage, when banks hoarded liquidity and market liquidity dried up. The dependence measure confirms

an increased correlation between banks’ behaviour in the crisis period.

0,15

0,20

0,25

0,30

0,35

2003 2004 2005 2006 2007 2008 2009

Figure 2.4. Correlation of item changes across banksDependence indicator D c,t , 6 months moving average

I II III

5 To have the relative change of each balance sheet item separately, Σig is dropped from the numerators and x from the denominators of the ratio in equation 2.2.

20

2.3.4 Herding: aggregate number of reactions

The number of reacting banks has a time as well as a cross-sectional macroprudential dimension. As a

point in time indicator it reflects the commonality of exposures, while measured over time herding

reflects whether bank’s reactions are pro-cyclical. In the literature, measures of herding usually apply

to financial market trades and not to behaviour of banks (applied to banks, herding is usually

associated to bank runs by depositors, like in Allen and Gale, 1998). Based on market indicators of

herding, we reconstruct a metric to assess whether there has been herding by our sample of banks.

This statistical indicator resembles the herding measure of Lakonishok, Shleifer and Vishney (LSV,

1992), as applied by Bikhchandani and Sharme (2001) amongst others. The LSV measure is defined as

the number of investors who buy or sell a stock, relative to the average number of investors trading in

the stock market.

Compared to stock market trades, a distinction between positive and negative adjustments in

the balance sheets of banks is not obvious, since the direction of a management action will be

determined by the nature of a balance sheet item (for instance, it can be expected that a balance sheet

item j will be reduced in a crisis, while another balance sheet item k will be expanded, depending on

whether an item is an asset or liability, a stable or unstable source of liquidity etc.). For the purpose of

an aggregate herding measure, we define Hb,t as the total number of banks that make one or more

extreme positive and negative changes to their balance sheet items, measured by value changes of

items that exceed a threshold z,

( )yfH ib

t,b ∑= for yi > z (2.4)

with yi for each balance sheet item being the ratio n

I

/I b

bt,i

bt,i

∑ ∆∆

, with z the cut-off point at the

5% tail of the distribution of this ratio and f(yi) = 1 if yi > z for any item i per bank. Figure 2.5

indicates that in the crisis the number of banks with an extreme response (measured by Hb,t) was much

higher than in the years before, pointing at increased herding.6 The break-down of Hb,t in Figure 2.6

shows that this was driven by increased downward adjustments of balance sheet items, while before

the crisis the number of banks with extreme upward adjustments was substantially higher. It indicates

that the directional change of herding shifted from extension to contraction of balance sheets and

reflects that in the 2005-2007 period there was an increased number of extreme actions to extend

balance sheets (for instance, to high yielding exposures), followed by an intense unwinding in the

crisis. 6 In Figures 2.4 and 2.5, Hb is based on the 10% largest items in terms of value, to filter for large relative changes that are related to a small denominator value of the ratio.

21

20

30

40

50

2003 2004 2005 2006 2007 2008 2009

Figure 2.5. Number of banks, extreme balance sheet adjustments Herding indicator Hb

(6 month moving average, 10% largest items in sample)

I II III

10

15

20

25

30

2003 2004 2005 2006 2007 2008 2009

Positive adjustment Negative adjustment

Figure 2.6. Number of banks, direction balance sheet adjustments Number of banks with postive, negative adjustments (6 month moving average, 10% largest items in sample)

I II III

2.3.5 Herding: number of reactions by instrument

While equation 2.4 summarises across yi to construct an aggregate measure of herding, in fact herding

should be analysed on an instrument by instrument basis to assess the similarity of reactions and the

concentration of trades in particular market segments. This may reflect the build up of common

exposures by banks (or vice versa concerted withdrawals from markets), indicative of the cross-

sectional dimension of macroprudential risk. For this reason, Hb,t(i) is constructed for each balance

sheet item (i) separately, as in equation 2.5,

( )yf)i(H ib

t,b ∑= for yi > z(i) (2.5)

with yi being similar as in the previous section and z(i) the cut-off point at the 5% tail of the

distribution of ratio Hb,t(i). The figures in Appendix 2.3 show that in the crisis, herding was

concentrated in particular market segments. The first observation is that, although the size of equity

and bond holdings was reduced during the crisis (Sr,t(i) in panels A.2 and A.6 in Appendix 2.2), this

was not due to increased herding; the number of banks making extreme negative adjustment to their

investment portfolios remained low (Hb(i) in panels B.2 and B.6, Appendix 2.3). This indicates that

the reduction of bond and equity holdings was concentrated at a limited number of banks. A second

observation is that an increasing number of banks experienced substantial changes in their deposit

funding base, due to volatile in- and outflows (see panels B.9-10 and B.13-14 in Appendix 2.3). This

reflects the increased mobility of deposits and the scramble for this funding source by a growing

number of banks, looking for alternative funding sources.

In the crisis there was an increasing number of banks that experienced an extreme substitution

between demand and fixed-term deposits; in the first and second stages of the crisis there were more

banks with very large increases in fixed-term retail deposits on balance (and decreases in retail

22

demand deposits), while in the third stage this trend reversed (compare panels B.10 and B.14).

Possibly, the banks tried to compensate the outflow in wholesale fixed-term deposits, which was

extremely strong in the third stage (panel A.9). The search for alternative sources of funding is also

evident by the increased number of banks that has issued debt securities since 2008 (partly supported

by government guarantees, panel B.8). A final observation is the increased intermediary role of the

ECB in the crisis, indicated by a growing number of banks that upwardly adjusted their deposit

holdings at the central bank (see Hb(i) in panel B.1). It expresses the increased preference for central

bank reserves, which acted as a precautionary liquidity buffer for banks that were increasingly risk

averse.

2.3.6 Robustness

To assess the robustness of the measures to different specifications, we construct alternative measures

for herding and dependence (Sr and Sm are yet two alternatives for size). As an alternative measure for

herding, we include in H the multiple swings of balance sheet items on account of one bank, to

construct Hi,t,

( )yfH ii

t,i ∑= for yi > z(i) (2.6)

with yi and z(i) being similar as in equation 2.5. Comparing Figures 2.7 and 2.5 indicates that both Hb,t

and Hi,t peaked in the crisis, while the break-down of Hi,t in Figure 2.8 confirms that the direction of

herding changed from positive to negative in the crisis.

50

100

150

200

2003 2004 2005 2006 2007 2008 2009

Figure 2.7. Number of extreme balance sheet adjustments Herding indicator H i

(6 month moving average, 10% largest items in sample)I II III

25

50

75

100

2003 2004 2005 2006 2007 2008 2009

Positive adjustment Negative adjustment

Figure 2.8. Direction of balance sheet adjustments Number of postive, negative adjustments(6 month moving average, 10% largest items)

I II III

23

As an alternative dependence measure we extract the correlation structure between balance sheet item

changes by factor analysis. We define Df,t as the common driver of the relative changes Sr,t(i), of

balance sheet item groups i, j…x, across banks (based on correlations of item changes between banks).

Df,t is determined by the eigenvalue of the first factor of the factor analysis, indicating the variance that

is explained by this factor. The first factor is loaded on 18 banks and explains 49% of the total

variance. Comparing Figures 2.4 and 2.9 shows that both Dc,t and Df,t peaked in the first and second

stages of the crisis. Dc,t and Df,t accelerated in the beginning of the crisis, when banks hoarded

liquidity, confirming an increased correlation between banks’ behavioural reactions in that period.

Hence, we conclude that the metrics are robust to different herding and dependence measures.

10

15

20

25

30

2003 2004 2005 2006 2007 2008 2009

Figure 2.9. Factor analysisDependence indicator of common behaviour D f

Eigen value 1st factor, 6 month moving averageI II III

Another conclusion from the measures for herding and dependence under different specifications is

that before the crisis they all pointed to risks building up over time and across institutions. The

measures peaked in the crisis, sometimes with different leads and lags. This underscores the relevance

of using several indicators of macroprudential liquidity risk at the same time.

Robustness is further tested more formally by reconstructing the indicators with a new perturbed

dataset. The simulated data distribution is used to reconstruct the measures for size, dependence and

herding. Robustness is then tested by comparing the measures obtained with the original data and the

perturbed data by the Kolmogorov-Smirnov (KS) test for statistical equivalence.7 In most cases the

7 The Kolmogorov-Smirnov (KS) test is a nonparametric test for the equality of continuous, one dimensional probability distributions. The test statistic quantifies a distance between the empirical distribution functions of two samples. Under the null hypothesis, it is assumed that the samples are drawn from the same distribution. The KS test is useful to compare two samples, as it is sensitive to differences in both location and shape of the empirical cumulative distribution functions of the two samples.

24

equivalence holds from which we conclude that the measures of size, dependence and herding are

robust to distributions of the underlying data.8

2.4 Conclusions

The behaviour of banks can be described by rather simple indicators constructed from firm-specific

balance sheet data. Although they are descriptive in nature, the measures identify trends in banks’

behaviour that convey forward looking information on market-wide developments. A key insight from

the analysis is that while banks usually follow a pecking order in their balance sheet adjustments (by

making larger adjustments to the most liquid balance sheet items compared to less liquid items),

during the crisis banks were more inclined to a static response. This suggests that they have less room

to follow a pecking order in their liquidity risk management in stressed circumstances. It implies that

banks’ responses in crises may have more material effects on the economy, since a static response rule

means that banks are more likely to adjust their (less liquid) retail lending and deposits than under

normal market conditions. A sufficient stock of liquid buffers could prevent that banks are forced to

such detrimental static responses, which lends support to the initiatives of the Basel Committee to

tighten liquidity regulation for banks (BCBS, 2009c).

The measures for size and the number of extreme balance sheet adjustments gauge the time

dimension of macroprudential risk, and indicators of the dependence and concentration of reactions

capture the cross-sectoral dimension. The measures are robust to different specifications and

distributions of the data. Applied to Dutch banks, the measures show that the number, size and

similarity of responses substantially changed during the crisis, on certain market segments in

particular. They also indicate that the nature of banks’ behaviour is asymmetric; being more intense in

busts than in booms. Furthermore, during the crisis the deleveraging of large banks started earlier, was

more intense and more advanced than the deleveraging of smaller banks.

Given these findings, the indicators are useful for macroprudential analysis, for instance with

regard to monitoring frameworks. Our analysis underscores the relevance of using several indicators

of liquidity risk at the same time, given the different leads and lags of the measures with systemic risk.

The empirical results also provide useful information for financial stability models. A better

understanding of banks’ behaviour helps to improve the micro foundations of such models, especially

with regard to the behavioural assumptions of heterogeneous institutions. Finally, the empirical

8 Let FI

o be the cumulative distribution function (CDF) of the original item values Ibi. F

Ip is the CDF under

perturbation, obtained by adding a disturbance to the original data, i.e. FIp = FI

o + ε, where ε is generated from a normal distribution with mean 0 and standard deviation equal to a factor α times the standard deviation of the original distribution FI

o (α is chosen to be: 10%, 15% and 20%). FIp is used to reconstruct the measures for size,

dependence and herding based on the new distribution FIp. To test robustness, the measures obtained with the

original data and the perturbed data are compared using the KS test. The test results indicate that in most cases the equivalence holds at α = 10%.

25

findings in our study are relevant to understand the role of banks in monetary transmission and to

assess the potential demand for central bank finance in stress situations. The measures explain

developments of financial intermediation channels (wholesale and retail, unsecured, secured etc.)

along the cross-sectional and time dimensions. They also shed more light on the size and number of

banks that rely on central bank financing.

26

Appendix 2.1 Liquidity values of assets and liabilities

Source: DNB (2003)

The values in columns WEEK and MONTH represent haircuts on assets and run-off rates of liabilities. For the liquidity test for the full month, a distinction is made between non-scheduled items and scheduled items. In contrast to non-scheduled items, scheduled items are included on the basis of their possible or probable due dates. For the liquidity test for the first week, scheduled items are only included if they are explicitly taken into account in day-to-day liquidity management (treasury operations). In the following table, scheduled items are indicated by the letter M. GROUP ASSETS M WEEK MONTH

Banknotes/coins 100 100 Receivables from central banks (including ECB) 1 1Demand deposits 100 100 1 2Amounts receivable M 100 100 1 3Receivables in respect of reverse repos M 100 100 1 4Receivables in the form of securities or tier 2 eligible assets

M d* d*

Collection documents 1Available on demand 100 100 2Receivable

M 100 100

Readily marketable debt instruments/ECB eligible assets Issued by public authorities and central banks 2 1ECB tier 1 and tier 2 eligible assets 95** 95** 2 2ECB tier 2 eligible assets, deposited 85** 85** 2 3ECB tier 2 eligible assets, not deposited 85 85 2 4Other readily marketable debt instruments, Zone A 95 95 2 5Other readily marketable debt instruments, Zone B

70 70

Issued by credit institutions 2 1ECB tier 1 eligible assets 90** 90** 2 2ECB tier 2 eligible assets, deposited 80** 80** 2 3Other debt instruments qualifying under the CAD (Capital Adequacy

Directive) 90 90

2 4Other liquid debt instruments

70 70

Issued by other institutions 2 1ECB tier 1 eligible assets 90** 90** 2 2ECB tier 2 eligible assets, deposited 80** 80** 2 3Other debt instruments qualifying under the CAD (Capital Adequacy

Directive) 90 90

2 4Other liquid debt instruments

70 70

Amounts receivable Branches and banking subsidiaries not included in the report 3 1Demand deposits 50 100 3 2Amounts receivable in respect of securities transactions M) 100 100 3Other amounts receivable

M 100 90

Other credit institutions 3 1Demand deposits 50 100 3 2Amounts receivable in respect of securities transactions M) 100 100 3 3Other amounts receivable M 100 90 Public authorities 3 1Demand deposits 50 100 3 2Amounts receivable in respect of securities transactions M) 100 100 3 3Other amounts receivable

M 100 90

Other professional money market players 3 1Demand deposits 50 100 3 2Amounts receivable in respect of securities transactions M) 100 100 3 3Other amounts receivable

M 100 90

Other counterparties 1Demand deposits 0 0

2Amounts receivable in respect of securities transactions M)

100

90

27

4 3Other amounts receivable, including premature redemptions M

50

40

Receivables in respect of repo and reverse repo transactions Reverse repo transactions (other than with central banks) 5 1Receivables in respect of bonds M 100 100 5 2Receivables in respect of shares

M 100 100

Repo transactions (other than with central banks) 5 1Receivables in the form of bonds M 90/d*/** 90/d*/** 5 2Receivables in the form of shares

M 70 70

Securities lending/borrowing transactions 5 1

Securities stock on account of securities lending/borrrowing transactions

100

100

5 2Securities receivable on account of securities lending/borrowing transactions

M

100

100

Other securities and gold 6 1Other liquid shares 70 70 6 2Unmarketable shares 0 0 2 3Unmarketable bonds M 100 100 4Gold

90 90

Official standby facilities 14 1Official standby facilities received

100 100

14 Receivables in respect of derivatives M *** ***

Total

LIABILITIES M WEEK MONTH Moneys borrowed from central banks 7 1Overdrafts (payable within one week) 100 100 7 2Other amounts owed

M 100 100

Debt instruments issued by the bank itself 8 1Issued debt securities M 100 100 8 2Subordinated liabilities

M 100 100

Deposits and fixed-term loans Branches and banking subsidiaries not included in the report 9 1Amounts owed in respect of securities transactions M) 100 100 9 2Deposits and other funding – fixed maturity

M 100 90

Other credit institutions 9 1Amounts owed in respect of securities transactions M) 100 100 9 2Deposits and other funding – fixed maturity

M 100 90

Other professional money market players 9 1Amounts owed in respect of securities transactions M) 100 100 9 2Deposits and other funding – fixed maturity – plus interest payable

M 100 90

Other counterparties 1Amounts owed in respect of securities transactions M) 100 100

10 10

23Deposits and other funding – fixed maturity – plus interest payable Fixed-term savings deposits

M M

50 20

40 20

Liabilities in respect of repo and reverse repo transactions Repo transactions other than with central banks

11 1Amounts owed in respect of bonds M 100 100 11 2Amounts owed in respect of shares

M 100 100

Reverse repo transactions other than with central banks 11 1Amounts owed in the form of bonds M 100 100 11 2Amounts owed in the form of shares

M 100 100

Securities lending/borrowing transactions 11 1Negative securities stock on account of securities lending/borrowing

transactions

100

100 11 2Securities to be delivered on account of securities

lending/borrowing transactions

M

100

100 Credit balances and other moneys borrowed with an indefinite

effective term

Branches and banking subsidiaries not included in the report

28

12 1Current account balances and other demand deposits

50 100

12 Other credit institutions 12 1Balances on vostro accounts of banks 50 50 12 2Other demand deposits

50 100

Other professional money market players

12 1Demand deposits

50 100

LIABILITIES (continued) M WEEK MONTH Savings accounts

13 1Savings accounts without a fixed-term

2.5 10

Other 13 1Demand deposits and other liabilities 5 20 13 2Other amounts due and to be accounted for, including the balance of

forward transactions and amounts due in respect of social and provident funds

5

20

Official standby facilities 14 1Official standby facilities granted

100 100

Liabilities in respect of derivatives 14 1Known liabilities in respect of derivatives M *** *** 14 2Unknown liabilities in respect of derivatives

*** ***

Other contingent liabilities and irrevocable credit facilities 14 1Unused irrevocable credit facilities, including underwriting of issues 2.5 10 14 2Bills accepted M 100 100 14 3Credit-substitute guarantees 2.5 10 14 4Non-credit-substitute guarantees 1.25 5 14 5Other off-balance-sheet liabilities 1.25 5

Total

M = Scheduled item. M) = Settlement due within one week or open-ended, including first week or as scheduled. * = Less applicable discount. ** = Either at stated percentage or at percentages applicable for ECB/ESCB collateral purposes. *** = Calculated amount for the period concerned. 90/d*/** = 90% OR: less applicable discount (provided the method is consistently applied).

29

Appendix 2.2 Relative Size measure Sr,t(i), 6 month moving average

30

Appendix 2.3 Herding measure Hb,t(i): number of banks with positive, or negative adjustments, 6 month moving average

31

Chapter 3

Banks’ responses to funding liquidity shocks:

Lending adjustment, liquidity hoarding and fire sales

3.1 Introduction

The financial crisis of 2007-2009 has shown that if wholesale funding dries up, banks face huge

funding liquidity problems.9 The freeze of wholesale funding markets was an essential characteristic

of the crisis (IMF, 2010b). In particular, the part of wholesale funding that is linked to asset markets,

i.e. repo funding, issuance of securities and asset-backed finance, was hit hard. This activated the

liquidity channel of financial transmission through which funding liquidity shocks are propagated to

bank lending and the real economy (BCBS, 2011). Evidence on the role of financial markets in the

liquidity channel remains scarce. This chapter contributes to fill this gap by analysing empirically how

banks adjust to a funding liquidity shock originating from financial market volatility, using data on

Dutch banks during the financial crisis.

We focus on three types of adjustment on the asset side of the bank balance sheet: (1) reduced

lending, (2) liquidity hoarding, and (3) fire sales. Figure 3.1 shows a stylized bank balance sheet

illustrating these three types of responses. If a bank is confronted with a negative shock in wholesale

funding (depicted by a downward pointing arrow), it has the following options. First, it can cut down

lending, either retail or wholesale. Second, it can sell securities from its investment portfolio, which is

known as ‘fire sales’ if the bank is under pressure to do so. Third, it can borrow from the central bank

(thus bringing down its net claims position). If the bank fears that its future access to liquidity is

uncertain, it may even borrow extra from the central bank and hold these funds as a buffer in deposit at

the central bank. Liquidity buffers could also be strengthened by holding more highly liquid bonds.

These two precautionary saving measures can be classified as ‘liquidity hoarding’ (denoted by the

arrows within parentheses in Figure 3.1).

9 This chapter is a revised version of De Haan and Van den End (2011).

32

Figure 3.1. Stylized bank balance sheet: Possible asset side responses

Net claims on Central Bank ↓(↑) Retail deposits

Retail credit ↓ Wholesale borrowing ↓

Wholesale credit ↓ Capital

Securities holdings ↓

- of which: Liquid securities holdings (↑)

Note: A downward (upward) pointing arrow denotes a decrease (increase).

Aspects of the above mentioned three behavioural responses to funding liquidity shocks have been

addressed in the recent literature, both empirically and theoretically. Theoretically, Diamond and

Rajan (2005) stress the interaction and reinforcing effects of banks’ liquidity shortages and solvency

problems. They explain how aggregate liquidity shortages can emerge and force banks to prematurely

foreclose otherwise profitable loans, which can result in banks facing sizable losses that will restrain

future lending. Empirically, the response of bank lending to funding shocks has been examined mostly

by means of single equation models. For example, Ivashina and Scharfstein (2010) find that a greater

volatility of deposits and draws on committed credit lines prompt banks to reduce lending. Cornett et

al. (2010) find that US banks with more stable funding sources were better able to continue lending

during the crisis. They also find that liquidity hoarding is mostly related to the proportion of illiquid

assets and the presence of unused off-balance sheet loan commitments on the bank balance sheet.

Acharya et al.’s (2008) theoretical study relates liquidity hoarding to so-called ‘predatory behaviour’,

aimed at the exploitation of urgent funding needs of other market participants. They show that banks

with surplus liquidity have an incentive to strategically underprovide liquidity to other banks, to be

able to benefit from the latter’s forced fire sales of assets against low liquidation prices. Similarly,

Diamond and Rajan (2009) show that the expectation of distressed banks being forced to sell assets in

the future at fire-sale prices drives healthy banks to hoard liquid funds so as to allow them to take

advantage of future investment opportunities. Fire sales as such are mostly captured in theoretical

models (e.g. Cifuentes et al., 2005) or in simulation models of central banks (e.g. Aikman et al., 2009).

These models consider both liquidity and capital constraints as triggers of fire sales, without

specifying which constraint is the most binding.

To the best of our knowledge, the link between fire sales on the one hand, and liquidity and

capital constraints on the other, has not been examined empirically. Hence, another purpose of this

chapter is to examine the effects of both liquidity and capital constraints on fire sales. For theorists as

well as regulators, it is important to know the relative importance of the bank liquidity and bank

capital channel as a driver of adjustments on the asset side of banks’ balance sheet. We employ a

multi-equation framework instead of a single-equation framework, thus taking into account the

33

dynamic interrelations among instruments of bank liquidity management. To investigate bank liquidity

management strategies in more detail, this chapter uses disaggregated balance sheet data. The multi-

equation approach has been used before. Spindt and Tarfan (1980), for example, model US banks’

liquid assets and liabilities as a system of equations. In their model, liabilities are qualified as (weakly)

exogenous and assets as endogenous, based on the idea that banks can determine their investment and

lending strategies, while the availability of funding is predominantly given. We adopt similar

assumptions in this chapter. However, there are several differences between their and our approach.

Spindt and Tarfan estimate separate models for five large US money-center banks and then average

the coefficients. In contrast, we estimate a multi-equation model while pooling our sample of banks, so

that the model represents the banks’ average behaviour. Further, we use a panel Vector Auto-

Regressive (p-VAR) model, which takes into account the heterogeneity between individual banks by

allowing for fixed effects. An advantage of VAR models is that they can be used to generate

orthogonalized impulse-response functions, identifying the impact of an isolated shock to one variable

to all the other variables in the system.

Our VAR model is estimated using monthly data of 17 of the largest Dutch banks over the

period January 2004 to April 2010. This period encompasses the run-up to and subsequent unwinding

of the financial crisis. We find that banks respond to an asset market driven funding shock in several

ways. First, banks reduce lending, especially wholesale lending. Second, banks hoard liquidity, in the

form of liquid bonds and central bank reserves. Third, banks conduct fire sales of securities, especially

equity. Finally, our results suggest that fire sales are triggered by liquidity constraints rather than by

solvency constraints.

The structure of the chapter is as follows. Section 3.2 introduces the model. Section 3.3

describes the data and some stylized facts. Section 3.4 discusses the results. Section 3.5 presents

several robustness checks, after which Section 3.6 concludes.

3.2 Model

We use a panel-VAR model, which treats all variables in the system as endogenous and allows for

unobserved individual heterogeneity by including fixed effects. The model reads as follows:

itit

ti

it

t

Y

X)L(BA

Y

Xε+

+=

(3.1)

where Xt is a vector containing one market funding cost variable for each month t and Yit is a vector

holding a set of balance sheet variables for each bank i and month t. In Section 3.4 the variables which

are included in the respective models are specified. Ai is a matrix of bank-specific fixed effects, B(L) is

34

a matrix polynomial in the lag operator whose order is 3 according to Akaike’s information criterion.

itε is the error term. The coefficients of the p-VAR model are estimated by system Generalised

Method of Moments (GMM), using lags of the model variables as instruments.10 GMM is widely used

in the absence of strictly exogenous variables or instruments, see for instance Doytch and Uctum

(2011). System GMM has one set of instruments to deal with endogeneity of regressors and another

set to deal with the correlation between lagged dependent variables and the error term. The fixed

effects are eliminated by expressing all variables as deviation from their means. Since the fixed effects

are correlated with the regressors as a result of the inclusion of lags of the dependent variables,

ordinary mean-differencing (i.e. expressing all variables as deviations from their full sample period’s

means) as commonly used to eliminate fixed effects would create biased coefficients. To avoid this

problem, forward mean-differencing, also known as ‘Helmert’ transformation’, is used instead (cf.

Arellano and Bover, 1995). This procedure removes only the forward mean, i.e. the mean of all future

observations available in the sample and preserves the orthogonality between transformed variables

and lagged regressors, so that the lagged regressors can be used as valid instruments for estimating the

coefficients by system GMM.

The model variables are chosen for their relevance with respect to our three behavioural

hypotheses under consideration (see Appendix 3.1 for the definitions of the variables). On the liability

side, we distinguish retail funding (RETDEP), secured wholesale funding by repurchase agreements

(REPO) and securities funding (SECUR). Next to these balance sheet variables, we include a market

funding cost variable, proxied by the spread on the money market swap rate (SPR). SPR is the cost of

unsecured interbank funding and is usually considered to be an important determinant of banks’

deposit and lending rates. The model is used to simulate banks’ responses on the asset side of their

balance sheets to shocks in the above mentioned funding variables. Thereby, three types of responses

are considered: (1) bank lending, (2) liquidity hoarding and (3) fire sales.

For bank lending, we consider two main categories, wholesale lending (WSCR) and retail

lending (RETCR). Liquidity hoarding is captured by the asset side variables of highly liquid bonds

(BONDL) and net claims on the Central Bank (NCCB). Both can act as liquidity buffer in times of

stress. For fire sales, we consider investments in less liquid bonds (BONDI) and equity investments

(EQ), under the assumption that under stressful market conditions banks prefer to sell their least liquid

bonds (BONDI) first, while holding on to their highly liquid bonds (BONDL) for precautionary

(liquidity hoarding) reasons.

Two remarks should be made as to the scope of the model. First, the causality between market

liquidity and funding liquidity is not explored in the paper. Our focus is on the causality running from

funding liquidity to bank assets. Second, contagion effects between individual banks are not studied

10 For more details we refer to Love and Zicchino (2006), whose Stata code we gratefully used for the estimation.

35

explicitly in this paper. However, several of the model variables (for example, WSCR and REPO)

partly measure how much a particular bank lends to c.q. borrows from all other banks. Hence, spill-

over effects are captured implicitly by the panel VAR model’s coefficient estimates.

To examine banks’ responses to funding liquidity shocks, we use impulse-response functions

that are derived from the p-VAR model. The shocks are orthogonalized, so that the response of one

variable to a shock in another variable can be interpreted as the reaction of the former variable to the

innovations in the latter, while holding all other shocks equal to zero. To orthogonalize the shocks it is

necessary to decompose the residuals. The decomposition is conducted by imposing a particular

ordering of the variables in the system and attributing any correlation between the residuals of any two

elements to the variable that comes first in the ordering. This procedure is known as the Choleski

decomposition. The identifying assumption is that variables that come earlier in the ordering affect the

following variables contemporaneously, as well as with lags, while the variables that come later affect

the previous variables only with lags. In other words, the variables that appear earlier in the ordering

are more exogenous than the ones that appear later (or, more formally, in the short run the former are

weakly exogenous with respect to the latter). We will perform robustness checks to test the sensitivity

of the outcomes for changes in the ordering of the variables.

For our model specifications, we generally adopt the following principles with respect to the

ordering of the variables. First, we assume that shocks in the cost of wholesale funding have an

immediate effect on the balance sheet variables and that the funding cost responds to the balance sheet

shocks with a lag. Second, we assume that bank liabilities respond more quickly than bank assets. This

assumption reflects the fact that funding depends on market conditions that are often outside the

banks’ direct control, while banks’ asset management in principle is at their own discretion.11 Third,

we assume that wholesale instruments (assets as well as liabilities) respond more quickly than retail

items. This takes into account that wholesale instruments usually have shorter maturities than retail

instruments and therefore can be more easily adjusted. Fourth, we assume that liquid balance sheet

items with a short maturity adjust more quickly than less liquid and longer-term items.

Since the impulse-response functions are constructed from the model’s estimated coefficients,

the latter’s standard errors need to be taken into account. We calculate the standard errors and generate

confidence intervals of the impulse response functions using Monte Carlo simulations. This is

conducted by taking random draws of the model’s coefficients, using the estimated coefficients and

their variance–covariance matrix. We take 1,000 draws. The 5th and 95th percentiles of the resulting

distribution are used for the 90% confidence intervals of the impulse-responses.

11 Access to funding may depend on banks’ risk management strategies as well, but most likely with a lag.

36

3.3 Data and stylized facts

We use monthly data on liquid assets and liabilities of Dutch banks over the period January 2004 to

April 2010. This period encompasses both the pre-crisis and the crisis period. Our variables of interest

are summed up and defined in Appendix 3.1. All balance sheet variables are scaled by total assets. The

forward mean-differencing transformation contributes to the stationarity of the model variables. Panel

unit root tests indicate that all series are stationary.12 The variables for securities holdings (BONDL,

BONDI and EQ) have been deflated by a relevant market price index13, since we are interested in

deliberate portfolio adjustments net of revaluation effects.

The data source of the bank variables is De Nederlandsche Bank’s (DNB) prudential liquidity

report (DNB, 2003). This unique data source contains end-of-month data on liquid assets and

liabilities for all Dutch banks (including branches and subsidiaries of foreign banks) under

supervision, with a detailed break-down per balance sheet item. Not every item is reported by all

banks, since small banks do not have exposures in all categories. For that reason we use data of 17

banks whose average size during the sample period, measured by total assets, falls above the 80th

percentile of the full sample’s distribution.14 We also use a sub-sample of the 5 largest banks. These

top-5 banks (ING, ABN-Amro, Rabobank, SNS and Fortis-Netherlands, until its merger with ABN-

Amro mid-2010) represent around 85% of total assets in the sector. The 17 institutions consist of the

top-5 banks, 9 smaller domestic banks and 3 subsidiaries of foreign banks, together accounting for

around 95% of the sector.

The asset side of the balance sheets is dominated by retail and wholesale loans (Table 3.1). On

the liability side of the balance sheet, retail borrowing accounts for only a small portion (on average

10% to 15%). This is due to the relatively limited retail savings market in the Netherlands, where

banks have to compete with pension funds and insurers (DNB, 2010b). Our two samples mostly differ

with regard to their reliance on asset market related wholesale funding. The largest 5 banks are more

dependent on the repo market, with a share of secured wholesale borrowing (REPO) in total funding

twice as high compared to the average of 17 banks. The smaller banks are relatively more dependent

on the issuance of securities (bonds, commercial paper, certificates of deposits, including asset-back

securities), as reflected in the average share of SECUR of 32.6% for the whole sample of 17 banks

versus 22.0% for the top-5 banks.

12 The Levin, Liu and Chu t-test and the Fisher Chi-square-test, respectively, indicate that the null hypotheses of a common unit root process and individual unit root processes can be rejected. 13 BONDL and BONDI have been deflated by the FTSE EURO index of AAA rated corporate bonds. EQ has been deflated by the MSCI worldwide stock index. 14 The total number of banks under supervision at the end of March 2010 was 81.

37

Table 3.1. Summary statistics, January 2004 – April 2010

All 17 banks

Top-5 banks

________________________ _________________________ Mean Median Standard

deviation

Mean Median Standard

deviation

Assets

NCCB 0.011 0.004 0.037 0.007 0.004 0.025

BONDL 0.062 0.046 0.057 0.066 0.048 0.044

BONDI 0.063 0.016 0.089 0.077 0.082 0.061

EQ 0.010 0.000 0.018 0.441 0.356 0.224

RETCR 0.492 0.549 0.299 0.441 0.356 0.224

WSCR 0.328 0.234 0.286 0.341 0.340 0.200

Liabilities

SECUR 0.326 0.233 0.292 0.220 0.164 0.158

REPO 0.111 0.000 0.196 0.225 0.225 0.166

RETDEP 0.106 0.049 0.126 0.154 0.153 0.089

Capital

TIER1 0.192 0.139 0.197 0.129 0.102 0.060

Financial market

SPR 31.2 6.3 38.2 31.2 6.3 38.2

CDS 54.2 16.2 60.7 54.2 16.2 60.7

Note: Variable definitions are given in Appendix 3.1.

Before estimating the model, we first describe some stylized facts for our selected model variables.

The money market spread clearly depicts the pre-crisis period with a constant and low spread, and the

crisis-period beginning in August 2007 with a surging spread. This reflects the drying up of the

unsecured interbank market15 (Figure 3.2, panel E). The reliance on secured wholesale funding by

Dutch banks varied substantially between these two periods. In the years before the crisis, the use of

secured wholesale funding relative to retail funding almost doubled (Figure 3.2, panel A). The benign

market conditions and the development of new financial instruments (such as asset-backed securities)

15 Unsecured wholesale funding is captured by the cost variable SPR in the model. Unsecured wholesale funding itself is not among our model variables (therefore not shown in the figure). Besides, it was fairly stable during the crisis period, since the strong decline of interbank borrowing and fixed-term deposits was compensated by the growth of demand deposits from other professional money market participants.

38

stirred banks to expand their wholesale funding rapidly between 2003 and 2007. This trend was driven

by the strong growth of secured wholesale transactions. The boom in asset prices in the run up to the

crisis boosted financing that was collateralised by tradable securities, particularly repo transactions.

During the crisis, this trend reversed dramatically. This illustrates the sensitivity of wholesale funding,

repos in particular, for stress in financial markets. Secured wholesale funding declined strongly

relatively to retail funding, also because banks increased reliance on retail deposits in their search for

more stable sources of funding (ECB, 2009d). The issuance of securities fell somewhat back in 2008

but has recovered since 2009, which partly reflects the increased securitisation of assets pledged as

collateral at the central bank.

Figure 3.2. Development of model variables

0

0,05

0,1

0,15

0,2

0,25

0,3

0,35

0,4

2004 2005 2006 2007 2008 2009 2010

SECUR REPO RETDEP

Panel A. Funding liquidity(Ratio of total assets, monthly sample averages of 17 banks)

-0,02

0

0,02

0,04

0,06

0,08

0,1

0,12

2004 2005 2006 2007 2008 2009 2010

NCCB BONDL

Panel C. Highly liquid assets(Ratio of total assets, monthly sample averages of 17 banks)

0

0,02

0,04

0,06

0,08

0,1

2004 2005 2006 2007 2008 2009 2010

EQ BONDI

Panel D. Less liquid securities holdings(Ratio of total assets, monthly sample averages of 17 banks)

0

0,1

0,2

0,3

0,4

0,5

0,6

2004 2005 2006 2007 2008 2009 2010

WSCR RETCR

Panel B. Loans outstanding(Ratio of total assets, monthly sample averages of 17 banks)

0

30

60

90

120

150

180

2004 2005 2006 2007 2008 2009 2010

SPR

Panel E. Money market spread(Basis points)

Adjustments to lending were concentrated in the wholesale loan portfolio (WSCR) of the banks. In

terms of total assets it fell from around 35% mid-2007 to 25% in 2010 (Figure 3.2, panel B). Retail

lending (RETCR) was more stable. It even increased in 2007 and 2008 and has decreased slightly since

2009.

39

Liquidity hoarding by Dutch banks was evident by the increased amount of deposits and

collateral pledged at the central bank. This outpaced central bank borrowing and as a result net claims

on the central bank increased (NCCB; Figure 3.2, panel C). The rising share of highly liquid bonds in

the total bond portfolio also indicates that Dutch banks hoarded liquidity in the crisis. The share of

liquid bonds doubled between 2007 and 2010 to nearly 10%. Holdings of less liquid bonds also

increased between end-2008 and the beginning of 2010 (BONDI in panel D). This development partly

relates to securitisation of loans. Since banks during the crisis were no longer able to place securitised

assets in the market, they retained these (asset-backed) securities on their balance sheets for later use

as collateral when borrowing from the central bank.

Figure 3.2, panel D, shows the development of the bond and equity portfolios, adjusted for

revaluations. The decline of equity and bond portfolios between mid-2007 and mid-2008 reflects an

active scaling down of these exposures, possibly reflecting fire sales.

3.4 Results

In this section four p-VAR models are estimated. The first three are designed to capture three types of

bank asset reallocation after a shock in funding liquidity, i.e. (1) a cut in lending, (2) liquidity

hoarding, and (3) fire-sales. As an encore, a fourth model is estimated designed to test a fourth

hypothesis, i.e. whether fire sales are triggered by solvency constraints.

Results are discussed for the sample of 17 banks and for the sub-sample of the 5 largest banks.

However, we only display the results for the sub-sample of 5 banks if those are materially different

from those of the full sample of 17 banks.

3.4.1 Response of lending

In the bank lending model, the variables in vectors X and Y of model (1) are:

[ ] 'SPR REPO RETDEP WSCR RETCR

For bank lending we consider two main categories, wholesale lending (WSCR) and retail lending

(RETCR). By also taking into account two main funding sources, secured wholesale borrowing

(REPO) and retail deposits (RETDEP), we model credit management in relation to funding liquidity.

With the inclusion of the money market spread (SPR), the model incorporates the price of bank

funding, which also determines bank lending rates. Hence, the model captures both credit demand and

credit supply effects. The price of funding is assumed to affect credit demand. When the interbank

spread (SPR) rises, banks pass the increased funding costs on to their customers by raising lending

40

rates. As a consequence, the demand for credit may fall according to the traditional interest rate

channel of monetary transmission.16 Credit supply effects are assumed to originate from changes in the

available volume of wholesale funding, i.c. repo and securities funding. When banks are rationed on

the funding market, they have less means to support their asset side activities. As a consequence they

may curtail lending according to the liquidity channel of financial transmission. We allow retail

deposits to be immediately affected by the stress in the wholesale funding market, while any feedback

effect is assumed to occur only with a lag. The response variable of interest is bank lending, which is

split into wholesale lending and retail lending, of which wholesale lending comes first. Robustness

checks indicate that changes in the ordering of the variables have no substantial effect on the results.

From the impulse responses (Figure 3.3) it appears that wholesale lending (WSCR) reacts

significantly and positively to a shock in secured wholesale funding (REPO) and significantly and

negatively to a shock in the money market spread (SPR). This applies both to the sample of 17 banks

and the sub-sample of the 5 largest banks, and is in line with the experience in the 2007-2009 financial

crisis that wholesale loans were most vulnerable to funding liquidity risk (ECB, 2010a). A sudden rise

of interbank spreads and/or constraints in repo funding urge banks to adjust their asset side quickly,

both in terms of size and in terms of risk. It is plausible, and evident from the data (Figure 3.2, panel

B), that banks realise this adjustment by changing their wholesale lending rather than their retail

lending, since in general the former has a shorter maturity and a higher risk profile than the latter. This

outcome is consistent with the theoretical framework of Huang and Ratnovski (2010), who show that

negative market signals are an incentive for wholesale financiers to withdraw from lending, especially

short-term interbank lending. Liedorp et al. (2010) establish the channel of contagion running from

wholesale funding to interbank lending empirically.

16 As a robustness check, we also try an alternative control variable for credit demand effects, i.e. real GDP growth (see Section 3.5).

41

42

The impulse responses show a significantly negative response of retail lending (RETCR) to a shock in

wholesale lending (WSCR); see Figure 3.3, panel E. This suggests that, after a shock in the repo

market, banks reduce the share of wholesale loans in their loan book in favour of (lower risk) retail

loans. This substitution effect is weaker for the top-5 banks, which could reflect the fact that the

largest banks have a more diversified asset side and therefore more flexibility to adjust their balance

sheets. For both groups of banks, retail lending (RETCR) shows a brief but significantly positive

response to a shock in retail deposits (RETDEP; Figure 3.3, panel D). This reflects the linkage

between both retail items in the asset and liability management of banks. By matching retail lending

with retail deposits, banks limit the retail funding gap and thereby their dependence on volatile

wholesale markets for funding. Under volatile market conditions, banks shift their funding to more

stable retail deposits, as is shown by the significant positive response of RETDEP to a shock in SPR

(Figure 3.3, panel H). This response is only borderline significant for the top-5 banks, which again

underlines that these banks have access to a wider range of non-retail funding possibilities than

smaller banks.

3.4.2 Liquidity hoarding

The variables in the model for liquidity hoarding are:

[ ] 'SPR REPO SECUR BONDL NCCB

Liquidity hoarding is captured by highly liquid bonds (BONDL) and net claims on the central bank

(NCCB). Both can act as liquidity buffer in times of stress. By relating these two variables to both

REPO and issued securities (SECUR) the link between liquidity hoarding and market dependent

funding sources can be investigated. The variable SPR takes into account the influence of funding

costs on the unsecured interbank market. The funding variables come first in the ordering of the p-

VAR. By implication of the ordering, the money market spread has an immediate effect on repo

borrowing and the issuance of securities, while any feedback effects are assumed to occur only with a

lag. The response variable of interest is liquidity holdings, which is split into highly liquid bonds and

net claims on the central bank.

The impulse responses (Figure 3.4) indicate that liquidity hoarding is evident in response to a

shock in repo funding. For both samples of banks BONDL shows a (short) significant and negative

reaction to a shock in secured borrowing (REPO; see Figure 3.4, panels B and H), indicating that a

disruption in the secured funding market is followed by an accumulation of highly liquid assets. This

is in accordance with the experience during the crisis, when at some point only high-quality collateral

was accepted for repo transactions, which stimulated the hoarding of such assets. There is also

43

significant evidence of liquid bond hoarding in response to an upward shock in the money market

spread (SPR) by the top-5 banks (Figure 3.4, panel G). We find no empirical evidence for feedback

effects running from liquidity hoarding to the money market spread; the response of SPR to a shock in

BONDL is not significant (result not shown in the figure).

The sample of 17 banks also accumulates central bank reserves (NCCB) in response to a shock

in the money market spread (SPR), see Figure 3.4, panel D. Hence, the price of funding liquidity

appears to be an incentive for precautionary savings at the central bank. This is in line with the

theoretical model of Gale and Yorulmazer (2011), according to which the price of liquidity is an

incentive to hoard liquidity. For the top-5 banks, NCCB does not respond significantly to a shock in

SPR (Figure 3.4, panel G). This could be related to the tiering of the interbank market during the crisis,

as a result of which large banks in general paid lower spreads on unsecured interbank borrowing than

small banks. Liquidity hoarding in the form of increasing claims on the central bank (NCCB) is also

visible in response to a (negative) shock in secured funding (REPO) for the sample of 17 bank; see

Figure 3.4, panel E.

The 5 largest banks also seem to be less dependent on the central bank in case of a shock to

repo funding; the impulse response in panel K of Figure 3.4 is not significant. This is not in

accordance with the liquidity hoarding hypothesis, which assumes a negative response of NCCB to a

shock in REPO (e.g. a decline in repo funding urges banks to hoard central bank reserves, as is the

case for the whole sample of banks, see panel E). However, it should be noted that the variable NCCB

is the difference between central bank deposits and borrowings, which implies that a change in NCCB

could also reflect a change in central bank borrowing. This could explain the positive response of

NCCB to SECUR (which is borderline significant for the top-5 banks, see panel L), since a shut-down

of the primary market for securities issuance may stimulate banks to step up their borrowing from the

central bank (lowering NCCB) by using asset-backed securities as collateral in refinancing operations.

During the crisis, such securities were partly created for the purpose of collateralised borrowing at the

central bank. The large banks in the Netherlands are particularly active in this field.

44

45

We note that the results on liquidity hoarding are relatively sensitive to the ordering of the variables,

especially with respect to the responses of NCCB to a shock in SPR.

3.4.3 Fire sales

The variables in the model for fire sales are:

[ ] 'SPR REPO SECUR BONDI EQ

The first three variables are identical to the ones in the liquidity hoarding model specification. The

response variable of interest is securities holdings, which is split into less liquid bonds (BONDI) and

equity investments (EQ). We include BONDI instead of BONDL (which we used in the liquidity

hoarding model), assuming that under stressful market conditions banks prefer to sell their least liquid

bonds (BONDI) first, while holding on to their highly liquid bonds (BONDL) for precautionary

reasons.

The impulse responses in Figure 3.5 do not show a significant response of investment

portfolios to shocks in securities issued (SECUR; Figure 3.5, panels B and D), but the significant

positive response of equity holdings to a shock in secured wholesale funding (REPO; panel C) is

consistent with the occurrence of fire sales (this result is robust to changing the ordering of the

variables in the VAR, while the sample of the 5 largest banks shows a similar impulse response). The

positive relation between equity holdings and secured funding could also reflect the use of equities in

repos and securities lending transactions. When these activities are buoyant, banks equity holdings are

useful as collateral, while these become less useful when the secured funding market collapses and

only high-quality bonds are accepted as collateral. The significant negative response of EQ to an

upward shock in the money market spread (SPR) confirms the risk of fire sales after a shock in

wholesale funding (panel F). This finding is in line with the results of Nyborg and Östberg (2010), that

tightness in the interbank market for liquidity leads banks to pull-back liquidity, by selling equity

portfolios, among other things. They conclude that this could be either due to direct sales of equity

holdings by banks or to sales by other stock market investors that are confronted by a reduced liquidity

supply of banks.

46

Surprisingly, there is a negative response of less liquid bond holdings (BONDI) to a shock in secured

wholesale funding (Figure 3.5; panel A), while there is no significant ‘fire sales effect’ for bonds in

response to a shock in the funding spread (panel E). The same result - not shown in the figure - is

found when less liquid bonds (BONDI) are replaced by highly liquid bonds (BONDL), suggesting that

banks do not distinguish between liquid and illiquid bonds when they adjust their bond portfolio in

response to a funding shock. Apparently, the liquidity hoarding motive (i.e. an increase of liquid bond

holdings after a negative funding shock, cf. Section 3.4.2) dominates the fire sale motive with regard

47

to bond portfolios. A reason for this dominance could be the additional liquidity supplied by the

central bank during the crisis, which enabled banks to obtain funding against liquid and less liquid

bonds as collateral. By these liquidity operations, central banks aimed to prevent costly fire sales of

assets in financial markets by banks with a strong reliance on wholesale funding (ECB, 2010a).

As pointed out in Section 3.1, theoretical models (e.g. Cifuentes et al., 2005) and simulation

models (e.g. Aikman et al., 2009) are not clear about the issue whether liquidity or capital constraints

are the main trigger for fire sales. Therefore, we also estimate a fourth model that relates securities

holdings to both bank capital and the money market spread:

[ 1 ] 'SPR TIER BONDI EQ

Variable TIER1 is the ratio of Tier 1-capital to risk-weighted assets. If solvency constraints trigger fire

sales of assets, there should be a significantly positive response of BONDI and EQ to a shock in TIER1

(meaning that a deteriorating solvency position urges a bank to offload its investment holdings and

vice versa). Figure 3.6 shows that such a relationship is not evident, neither for bonds nor for equity,

while the impulse response of EQ is negative and significant with regard to a shock in the money

market spread. A similar result is found for the sample of the 5 largest banks. From this we conclude

that fire sales of equity are more likely to be triggered by funding liquidity constraints than by

solvency constraints.

48

Summing up, we find that in times of stress on the wholesale funding markets, banks reduce lending,

particularly wholesale lending, hoard liquidity in the form of liquid bonds and sell off part of their

investment portfolio, especially equity. We also find that fire sales are more likely to be triggered by

funding liquidity constraints than by solvency constraints.

3.5 Robustness

In this section we present some robustness tests.17 First, we re-estimate the models for the 12 smaller

banks from our sample of 17 banks, i.e. excluding the 5 largest banks. Concerning the lending model

the only notable difference is that retail credit does not significantly respond to shocks in SPR and

REPO, which suggests that credit supply by the smaller banks is less sensitive to developments in

wholesale funding markets. The impulse responses for the liquidity hoarding model are in line with the

findings for the whole sample. The response of central bank reserves (NCCB) to repo funding (REPO)

and funding cost (SPR) shocks is even stronger for the 12 banks than for the whole sample, suggesting

that the smaller banks are more dependent on the central bank for liquidity. With regard to the fire

sales model, the response of equity holdings (EQ) to a shock in the money market spread (SPR) and

secured wholesale funding (REPO) is not found to be significant for the smaller banks (compared to

the significant response for the whole sample of banks). One explanation for this difference could be

that the smaller banks in the Netherlands hold less equity in their trading books and more equity in the

form of participations that can be sold less easily. A shock to the capital ratio has no significant effect

on equity or bond holdings, similar to the result for the whole sample of banks.

Second, we re-estimate the models for a sub-period representing the financial crisis (June 2007

to the last month in the dataset, April 2010). The impulse responses for the lending model show some

notable differences. The response of wholesale lending (WSCR) to a shock in the money market spread

(SPR) and secured wholesale funding (REPO) is stronger for the crisis period. This can be explained

by the strong adverse shocks to the wholesale funding market during the crisis. At the peak of the

crisis in September/October 2008, the money market rate increased by more than 2 standard deviations

in one month, while repo funding of Dutch banks dropped by almost 1 standard deviation on average

for two months in a row. For comparison: all impulse response functions show a 1 standard deviation

shock during one single month. The results of the liquidity hoarding and fire sales models are almost

similar for both sample periods.

Third, we test the robustness of the model results for bank lending for the choice of the control

variable with respect to credit demand effects. Specifically, we replace the interest rate spread (SPR),

17 For reasons of space, the results are not presented in figures or tables, and are available on request.

49

which is included in the original model specification in Section 3.4.1, by real GDP growth (quarter-on-

quarter change). Economic growth can be considered to be another driver of credit demand, alternative

to the interest rate spread (SPR). The impulse responses indicate that a shock in GDP growth has no

significant effects on either retail or wholesale lending, in contrast to the significant effects of a shock

in SPR to lending (see Section 3.4.1). However, controlling for credit demand by GDP growth instead

of SPR does not materially change the impulse responses of retail and wholesale lending to a shock in

repo funding or retail deposits (which we interpreted as credit supply effects in Section 3.2). This

suggests that the credit supply effects are robust to different variables that control for credit demand.

Fourth, we introduce a variable to the VAR specifications to test the influence of the default

risk of the banks. This risk is reflected in the credit default swap spread (CDS, see Figure 3.7)18 which

replaces the money market spread variable (SPR) in the model specifications. In all models, CDS is

included as the first variable, assuming that market prices are more exogenous to banks than their own

balance sheets. In general, the results are similar to those of the original model specifications including

SPR. A notable difference in the model for bank lending is that the response of wholesale credit to a

shock in CDS is not significant for the whole sample of banks, while it is significantly negative if SPR

is included instead of CDS (see Section 3.4.1). This suggests that wholesale lending is to a larger

extent driven by liquidity risk than by banks’ default risk. A similar conclusion can be drawn with

regard to the response of equity holdings in the fire sales model, which is significant for a shock in

SPR (see Section 3.4.3), but not significant for a shock in CDS (this difference is specifically due to

the largest 5 banks). This is in line with the result found in Section 3.4.3 that liquidity constraints

rather than solvency constraints seem to trigger sales of equity holdings. A difference in the liquidity

hoarding model is that the response of net central bank reserves (NCCB) to a shock in secured

wholesale borrowing (REPO) is no longer significant. There is a significantly positive response of

NCCB to CDS, though, suggesting that stress in financial markets (reflected in a higher CDS) goes in

tandem with increased demand for central bank reserves (reflected in an increase of NCCB).

18 For the 12 smaller banks CDS spreads are not available. Therefore, for those banks we use the average CDS spreads of the five largest banks.

50

0

100

200

300

400

05 06 07 08 09 10

ABNAmro Fortis Rabo ING bank

Figure 3.7. Credit default swap spreads Dutch banksMontly averages, basis points

3.6 Conclusions

This chapter provides empirical evidence on banks’ responses to funding liquidity shocks, using data

of seventeen of the largest Dutch banks over the period January 2004 to April 2010. The dynamic

interrelations among instruments of bank liquidity management are modelled in a panel Vector

Autoregressive (p-VAR) framework. Orthogonalized impulse responses reveal that banks respond to a

negative funding liquidity shock in a number of ways. First, banks reduce lending, especially

wholesale lending. Wholesale loans are most vulnerable to funding liquidity risk and banks adjust

their wholesale lending rather than their retail lending. Second, banks hoard liquidity, in the form of

liquid bonds and central bank reserves. A disruption of the secured funding market is followed by an

accumulation of highly liquid assets and the price of funding liquidity appears to be an incentive for

precautionary savings at the central bank. Third, banks conduct fire sales of securities, especially

equity. With regard to bond holdings, the liquidity hoarding motive seems to dominate the fire sale

motive when the central bank supplies additional liquidity during the crisis, enabling banks to obtain

funding against bonds as collateral. We also find that fire sales are triggered by liquidity constraints

rather than by solvency constraints.

These results have two important policy implications. First, the results suggest that extended

liquidity operations by the central bank can effectively complement the market when it fails to

function in a crisis. Central bank deposits provide banks with a precautionary liquidity buffer, while

the additional liquidity supply of the central bank enable banks to obtain funding against collateral (of

less liquid bonds such as asset-back securities) that is otherwise not eligible for private repo

transactions. By doing this, the central bank can prevent costly fire sales by banks with too much

reliance on wholesale funding. This underlines that a flexible collateral framework of central banks,

51

which can be broadened in times of stress, is an important safeguard against banks’ responses that

destabilise financial markets. Second, the results support the proposal by the Basel Committee to

enhance the quantity and quality of liquidity buffers of banks and to reduce their maturity mismatches

(BCBS, 2010c). Our results show that shocks to wholesale funding can induce major adjustments of

banks’ balance sheets that can be costly for the economy and destabilise financial markets. It may be

assumed that banks with more and higher-quality liquid buffers respond less strongly to shocks in

financial markets. Our results particularly support the requirement of the Basel Committee to increase

liquidity buffers mostly for banks with a strong reliance on wholesale funding.

52

Appendix 3.1

Variable names and definitions

Name Definition

Assets 1

NCCB Net claims on central bank (deposits minus borrowing)

BONDL Liquid bond holdings (Tier 1 assets according to previous ECB list)

BONDI Less liquid bond holdings (non-Tier 1 assets according to previous ECB list)

EQ Equity portfolio

RETCR Retail credit (households and companies)

WSCR Wholesale credit (secured and unsecured, professional counterparties)

Liabilities 1

SECUR Securities issued (bonds, CPs, CDs, etc.)

REPO Secured wholesale borrowing (repos and securities borrowing)

RETDEP Retail deposits (households and smaller companies)

Capital

TIER1 Ratio of Tier 1 capital to risk-weighted assets

Financial

markets

SPR Money market spread (Euribor 3 month rate minus EONIA swap index), in basis

points

CDS Credit default swap spread, in basis points 1 Ratios to total assets.

53

Chapter 4

Macro stress-testing methods

4.1 Introduction

By way of introducing part II of the thesis, this chapter provides an overview of stress-testing methods

for banks, based on the literature and policy practise, focussing on macro stress-testing. Since the end

of the 1990s, stress-testing has been increasingly used by financial institutions and supervisory

authorities. It is a tool to quantify the impact of extreme, but plausible shocks in the financial-

economic environment on an institution or the financial system as a whole in a forward looking

manner. Stress-tests are a welcome addition to risk measures that focus on isolated risks and compute

losses by assuming normally distributed risk factors and historic correlations. Stress-tests, on the other

hand, focus on tail risks that may relate to a simultaneous realisation of risk factors with historic

correlations breaking down (Haldane, 2009).

Stress-testing methods can be identified along two dimensions; micro versus macro and

bottom-up versus top-down (Figure 4.1). The main difference between micro and macro stress-testing

is that the former is conducted by individual institutions as part of their risk management, while the

latter is usually applied by central banks and supervisors to assess the resilience of the financial sector

as a whole. There is a range of possible applications between pure micro and classical macro stress-

tests. For instance, vertical macro stress-tests - which is a supervisory tool tailored to individual

institutions - can be distinguished from horizontal tests, which aim at strengthening the solvency of the

banking sector (see Figure 4.1). The second dimension, i.e. bottom-up versus top-down, characterises

the level at which a stress-test is conducted, i.c. the financial institutions in a bottom-up test, or the

central bank or supervisor in a top-down test. The classification sketched in Figure 4.1 holds for all

types of risks that are subject of a stress-test, although the different methods have been mainly applied

to credit risk. The methods for liquidity risk are less advanced. Hence, this overview pays most

attention to stress-testing of credit risk. After a short introduction of micro stress-testing in Section 4.2,

the remainder of the chapter concentrates on macro stress-testing. Section 4.3 describes the bottom-up

approach, which Section 4.4 extends with three applications by policymakers in the crisis. Section 4.5

elaborates on top-down approaches, including integrated models for macro stress-testing. Section 4.6

discusses some issues with regard to the use of stress-testing results and Section 4.7 concludes.

54

4.2 Micro stress-testing

Micro stress-testing has become an increasingly important risk management tool for financial

institutions. It can be used to assess portfolio risks, set risk limits and guide the planning of capital

resources within an institution. Supervisory frameworks have promoted micro stress-testing. Basel II

requires banks to perform stress-tests using their internal models (see column I in Figure 4.1). In this

framework, stress-tests for credit risk are prescribed to reflect stress conditions in probabilities of

default (PDs) - the main parameter in banks’ credit risk models - and to assess the impact of an

economic recession on capital requirements (‘cyclicality stress-test’). More generally, Basel II requires

that banks conduct rigorous, forward-looking stress tests to identify possible events or changes in

market conditions that could adversely impact a bank’s risk profile and capital adequacy (BCBS,

2006).

Liquidity stress-testing by banks has been less advanced. A survey by the ECB shows that

some banks use stress-tests to quantify their liquidity risk tolerance as expressed in survival periods or

limit systems (ECB, 2008a). However, the underlying methodologies are highly heterogeneous across

the institutions in the sample and based on subjective assumptions. To enhance liquidity stress-testing,

in 2008 the Basel Committee has issued principles on which banks should base their internal stress

analyses and related quantification of the appropriate liquidity buffer and maturity profile (BCBS,

2008a). Compared to the guidelines for credit risk, the BCBS recommendations for liquidity risk are

less explicit on the model parameters that should be stressed. The Committee of European Banking

Supervisors (CEBS19) has also issued guidelines on liquidity stress-testing, which are more precise on

the assumptions concerning the impact of stress on the components of the liquidity buffer (CEBS,

2009a).

Banks can conduct stress-tests by using sensitivity analyses or scenario analyses. The former

considers the impact of a single shock and is usually performed to assess the stability of model

parameters, such as correlations and volatilities. In Basel II it is a way of assessing the robustness of

parameters in credit risk models, such as PD or loss given default (LGD). Scenario analysis considers

the impact of a combination of two or more assumed shocks on risk portfolios or on the whole

institution. To be meaningful for risk management, the scenarios should be tailored to the specific

exposures and vulnerabilities of the individual bank.

The recent crisis has revealed several shortcomings of micro stress-tests. They were calibrated

on normal market conditions instead of on tail events, scenarios were often too mild and

underestimated the liquidity risks of new financial instruments (Senior Supervisors Group, 2008).

Furthermore, exposures were usually not stress-tested on the level of the entire organisation,

neglecting correlations and risk concentrations across portfolios. To address such failures, the Basel

19 The CEBS has been succeeded by the European Banking Authority (EBA) in 2011.

55

Committee has published additional guidelines for micro stress-tests by banks (BCBS, 2009a). One of

the purposes of the guidelines is to raise banks’ awareness of the link between funding and asset

markets in crisis and the related consequences for liquidity and credit risk.

4.3 Bottom-up macro stress-testing

Macro stress-testing is used by central banks and supervisors to quantify the link between

macroeconomic variables and the health of either a single financial institution or the financial sector as

a whole (ECB, 2006). The practise was spurred by the introduction of macro stress-tests as a part of

Financial System Assessment Programs, conducted by the IMF (Jones et al., 2004). In most countries,

authorities apply bottom-up stress-tests as a regular health check of the financial sector. In that

approach, the central bank or supervisor designs scenarios that the institutions subsequently apply in

their internal models. There are several ways to design a scenario in a bottom-up stress-test, for

instance with a structural macroeconomic model of the central bank (this method is applied in Chapter

5). Meaningful stress scenarios should represent extreme realisations of underlying risk factors that

may strain the financial system. Most bottom-up macro stress-tests focus on credit risk as the main

source of risk for banks and use multi-year scenarios to capture downturns in the business cycle,

which represent the main risk driver. Market risk is assessed over much shorter time horizons. The

difference in the relevant horizons makes a joint treatment of market and credit risk in stress-testing

frameworks challenging (Boss et al., 2006). Compared to credit and market risk, stress-testing

scenarios usually pay less attention to liquidity risk. It is usually confined to mechanical run-off

scenarios, which assume sudden withdrawals of funding (see, for instance, Čihák, 2007). Lastly,

scenarios could be more directly tailored to the shock absorption capacity of banks. This could focus

on the profitability of banks, taking into account that it is their future capacity to absorb shocks

(Coffinet et al., 2009). Or scenarios could be defined by asking the question what shock size would

deplete the buffers of a bank. This method of reverse stress-testing translates this impact into a

scenario, which is the other way around as commonly applied in stress-tests. A disadvantage is that

reverse stress-testing is hard to apply to scenarios with multi-factor shocks, since the correlation

between risk factors complicates the translation from the impact into a scenario.

To guide the calculation of stress-test outcomes by the institutions, the central bank or

supervisor usually translates the simulated macro shocks in default and loss ratios via reduced form

satellite models. These link the macroeconomic variables to the portfolio drivers at the bank level.

This macro-micro issue is at the heart of most stress-testing methods (Van Lelyveld, 2009). In bottom-

up tests, the macro-micro issue can be dealt with by aligning the stress-test to the individual risk

profiles of institutions that participate in the test. Discussions between authorities and institutions on

tail risks are a valuable part of the bottom-up approach (Tarullo, 2010). It relates the stress-tests to the

56

‘real world’ and gives it credibility that is much more difficult to achieve with a top-down approach

(see Section 4.5).

As the final step in the process, the stress-test outcomes of the individual banks are aggregated

to the level of the system, to judge whether the banking sector as a whole is able to withstand adverse

shocks, or whether there are weak spots that could destabilise the sector. The quantification of such

contagion risks is usually not part of bottom-up stress-tests. Neither are feedback effects from the

banking sector to the real economy commonly part of those exercises. As an approximation, the co-

ordinating authority could collect qualitative information on possible management actions and second

round effects from the bottom-up stress-test outcomes (DNB, 2007). However, the assessment of

possible feed-back effects is complicated by the fact that banks’ behaviour in stress situations is

difficult to predict.

4.4 Bottom-up stress-testing in the crisis: three approaches

Traditionally, bottom-up stress-tests aim at assessing the resilience of the financial system to

exogenous shocks (column IV in Figure 4.1). In that sense, it is a monitoring device for central banks

with a responsibility for financial stability. However, since 2008 bottom-up macro stress-tests have

evolved as an instrument to design crisis measures for banks and restore market confidence by

increasing the transparency on risk exposures (ECB, 2010b). The use of stress-tests changed the crisis

management of authorities from re-active to pro-active, by providing a tool to prepare banks for future

shocks. In the crisis, bottom-up stress-tests have been applied either ‘vertically’, as a tool aimed at

specific risk exposures or business models (column II in Figure 4.1), or ‘horizontally’, as a sector-wide

exercise based on uniform scenarios (column III in Figure 4.1). This classification can be illustrated by

the stress-testing approaches applied in the US, UK and the EU during the crisis.

In 2009, the US authorities performed a ‘horizontal’ stress-test to determine banks’ capital

requirements under forward-looking scenarios (Board of Governors of the Federal Reserve System,

2009). To ensure similar treatment of the institutions, the macro stress scenarios were translated in

default rates on loans that were similar for all banks. Equal capital ratio targets for the participating

banks indicated how much additional capital would be needed. Individual bank’s results were

published to provide clarity to the markets and thereby make the stress scenario itself less likely to

occur. Crucial for the success of this strategy was the US government’s advance announcement that it

would replenish possible capital shortfalls in case banks would not be able to raise the required capital

in the market.

The Financial Services Authority (FSA) in the UK adopted a different approach, by applying

bottom-up stress-tests in a ‘vertical’ way to individual banks on different occasions in 2008 and 2009

(FSA, 2009). The tests were regarded as a regular feature of the supervisory practice instead of a one-

57

time comprehensive crisis measure, like in the US. Furthermore, the FSA used the stress-tests to

determine the amount of capital needed by specific institutions and the extent to which the authorities

had to insure certain portfolio risks. Unlike the US supervisory authorities, the FSA adopted tailor

made scenarios to individual banks, to reflect the differences in loan quality.

The CEBS coordinated a ‘regional’ macro stress-test to the banking system in the EU, for the

first time in 2010 (CEBS, 2009b). Based on common scenarios, the test was applied by local

supervisors to a number of large, cross-border European banks in 20 member states. The outcomes

provided the authorities with insight into the stability of the banking sector at a regional level. Also,

the use of common scenarios and principles has been conducive to the convergence of macro stress-

testing methods in the EU. The CEBS did not publish the results of the test by institution, as the

exercise was not meant to determine the capital requirements of individual institutions (thereby it was

a classical stress-test, see Figure 4.1). In Europe, supervisory measures and recapitalisation of

institutions remain the responsibility of the national authorities. The European exercise was the first

attempt to capture cross-border risks in a macro stress-test. Cross-border risks are usually neglected in

stress-tests, partly because of their complex nature, with liquidity risk as a key factor in transmitting

shocks cross-border. Although liquidity risk was not part of the CEBS stress-test, the test shed some

light on cross-border risks by applying the scenarios to the banks on a group-wide level, including

subsidiaries and branches operating in other countries.

Based on the experiences with macro stress-tests in the crisis, the ECB (2010b) identifies three

conditions for the success of bottom-up macro stress-tests: i) clear and synchronised communication

of the outcomes, ii) high level of disclosure, iii) complementarities with other policy actions for

institutions that do not pass the test. The three approaches applied in the crisis met these conditions to

a varying extent, whereby the US approach was deemed to be most successful mitigating the crisis

(Véron, 2010).

4.5 Top-down macro stress-testing

In top-down stress-testing methods the central bank or supervisor simulates the impact of adverse

shocks to financial institutions or sectors by in-house models (see the lower end of column IV in

Figure 4.1). Compared to bottom-up stress-testing, the use of in-house models improves the

comparability of outcomes between institutions and provides for a greater flexibility in applying

different scenarios, while some models also allow for quantification of the second-round effects in the

economy and financial markets. To capture the wide array of financial sector risks, a range of

modelling approaches has been developed over the last decade, although they are still mainly confined

to the credit risk of banks. According to the various stages of the stress-testing process, we distinguish

two strands of models; i) models that establish the link between macro variables and micro risk

58

drivers, mainly for credit risk and ii) integrated models that include liquidity risk and feedback effects

within the financial sector.

4.5.1 Modelling the macro-micro link

Foglia (2009) provides an overview of models that link macro variables to micro risk drivers of bank

portfolios. In many cases, macroeconomic models do not include financial sector variables and

therefore satellite models are commonly used in stress-testing exercises. These models are usually

reduced form satellites for credit risk and map exogenous macro shocks into measures of banks’ asset

quality. The models differ with regard to the measures of credit quality, level of aggregation and

estimation methodology. Two types of (micro) credit risk measures can be distinguished: i) indicators

of loan performance, such as non-performing loans and loan loss provisions and ii) default rates of

household and/or corporate sectors. The choice of the measures and the level of aggregation in the

models mostly depend on data availability, for instance related to the existence of a credit register in a

country.

Some studies use loan data of a panel of individual banks, to control for the individual bank

characteristics that affect credit risk and banks’ different sensitivities to macroeconomic developments

(Lehmann and Manz, 2006; see also Chapter 5). Other models estimate credit losses on an industry

level, taking into account inter-sector correlations (Düllmann and Erdelmeier, 2009). These

correlations are explained by macro variables that represent the systemic risk component and capture

contagion effects (Fiori et al., 2009). The studies underline that there could be hidden risks due to

unobserved correlation between risks across sectors. Some credit risk models that estimate default

rates of corporate borrowers use market-based measures of credit risk, such as Moodys-KMV

expected default frequencies (Åsberg and Shahnazarian, 2008), while others use ratings-based

measures (Bank of Japan, 2007). Most of the credit risk models use non-linear specifications, such as

logit and probit transformations, to take into account that nonlinearities are important when shocks are

large (Foglia, 2009).

A sophisticated extension to credit risk models is the portfolio approach, in which sectoral

default frequencies are combined with default probabilities of individual borrowers to simulate the

overall portfolio credit loss distribution (Boss el al., 2006). A further refinement of the portfolio stress

testing method is the incorporation of bank stability measures proposed by Goodhart and Segoviano

(2009). They define the banking system as a portfolio of institutions and estimate stability measures,

including the distress dependence among the banks in a system.

Credit risk models usually do not take into account feedback effects from credit risk to the

macro economy that may relate to shocks to bank’s credit exposures that affect lending. Feedback

effects do play a role in vector autoregressive (VAR) models that include both macroeconomic

variables and measures of default risk in a system of equations, as in Åsberg and Shahnazarian (2008)

and Aspachs et al. (2006). The latter study shows that shocks to banks’ default probabilities and equity

59

values can impact economic growth. A similar approach is applied by De Graeve et al. (2008), who

integrate a rating model that measures the probability of distress at the bank level into a

macroeconomic VAR model. The main disadvantage of the VAR approaches is that they include the

feedback effects in a non-structural way and thereby do not explain the complex interactions and

transmission channels of financial sector shocks to the real economy.

4.5.2 Integrated models

Models that integrate different satellite models provide a more structural approach to simulate

feedback effects within the financial system. These models draw on theoretical work on modelling

systemic financial crises. Allen and Gale (2000) explore the spread of contagion in a banking network

and Cifuentes et al. (2005) examine how defaults across the network are amplified by asset price

effects. The credit crisis has clearly shown the need to assess risks in a systemic perspective, taking

into account the possible interlinkages between different risk factors and contagion risks within the

financial system. Traditional macro stress-testing methods usually do not include those systemic

effects, such as the collapse of the interbank money market and other wholesale markets and the

importance of feedback effects between market liquidity and funding liquidity risks of banks.

One of the earliest integrated stress-testing models is the Systemic Risk Monitor (SRM) of the

Oesterreichische Nationalbank (Boss et al., 2006), which integrates satellite models of credit and

market risk with a network model to evaluate the probability of bank default. In the SRM, shocks to

credit and market risk exposures may trigger defaults of banks and this leads to interbank contagion

effects in a network model that is build on a matrix of bilateral interbank exposures. A similar

integrated framework is the RAMSI model of Bank of England (Alessandri et al., 2009), which

consists of a suit of models: a Bayesian VAR model to simulate macroeconomic scenarios, satellite

models for credit and market risk and net interest income, an interbank network model and an asset

price function to simulate fire sales of assets (market liquidity risk). Both the SRM and RAMSI

models do not allow for feedback effects from the banks to the real economy.

The RAMSI model is extended by Aikman et al. (2009) with feedback effects resulting from

liquidity risk. Funding liquidity risk on the liability side of banks’ balance sheets is included by

relating funding costs and market access to a banks’ credit rating and confidence effects. In the stress-

testing model of Bank of Canada, funding liquidity risk is modelled as an endogenous outcome of the

interaction between market liquidity risk, solvency risk and the funding structure of banks (Gauthier et

al., 2010). Spill-over effects occur due to the network effects among banks. The interaction between

credit and liquidity risk is also modelled by Hui and Wong (2008) in a stress-testing framework, where

negative asset price shocks affect banks’ liquidity risk through different channels. The shocks raise

banks’ default risk and induce deposit outflows, they depress the marketability of assets and increase

the risk of draw downs on contingent liabilities. In the framework, the linkage between the market and

credit risk of banks is established by a Merton-type model, while the liquidity risk of individual banks

60

is quantified by Monte Carlo simulations. The Liquidity Stress-Tester model described in Chapters 6

and 7 relates to this strand of the literature. It also simulates the liquidity position of banks for

different scenarios, taking into account the interaction between market and funding liquidity risks of

banks in a macro stress-testing framework.

A pitfall of integrated models for macro stress-testing is that the complexity of the model

structure makes the causal linkages and final results less transparent. Thereby the models may violate

a basic rule in macro stress-testing, i.e. that models must be kept sufficiently straightforward,

transparent and flexible to use and easy to communicate to policymakers and the public (Kwast et al.,

2010). On the other hand, integrated models provide a more complete picture of the possible impact of

tail events, by taking into account multiple transmission channels and feedback effects.

4.6 Considerations on the use of stress-tests

Micro and macro stress-tests have been increasingly used to determine capital ratios and liquidity

buffers of banks. This contributes to a forward looking risk management by institutions and a pro-

active approach of supervisors. Nonetheless, an important caveat with regard to the use of stress-tests

is the considerable uncertainty surrounding the test outcomes. In the first place, the choice of a

scenario is basically subjective, while scenarios represent only one possible unfavourable state of the

world in an otherwise uncertain future. Breuer et al. (2009) have developed a method to find scenarios

that are both plausible and extreme. The uncertainty about the realization of risks is captured by a risk

factor distribution that is estimated from historical data. However, this statistical approach neglects

possible scenarios that are not in the historic set of data.

Secondly, stress-testing models have even more limitations than models that are used for other

purposes, for instance structural macroeconomic models to forecast inflation. Those models generally

are local approximations of equilibrium relationships and for this reason they are less suitable for

assessing the effects of large shocks (Foglia, 2009). Moreover, data on tail events are scarcely

available and in stress situations historic correlations may break down, due to changing behaviour of

agents and non-linear adjustment processes. Hence, macro stress-testing models have to cope with

large parameter and model uncertainties. In essence this relates to Knightian uncertainty, i.e. there is

no distribution of probabilities of extreme events (Knight, 1921). Even if a range of extreme scenarios

is simulated, the underlying probability distribution is subjective and thereby possibly a wrong

representation of the future.

For these reasons, the outcomes of stress-tests should not be taken as a precise quantification

of potential losses, but rather as an indication of the likely impact of tail events. In this vein, they

should prepare banks and authorities for possible extreme events. Moreover, the stress-tests could

61

comfort financial markets if they are accompanied by the publication of underlying exposures, which

improves the transparency on the risks that banks face.

4.7 Conclusions

Methods for macro stress-testing are still in a developing stage. Bottom-up stress-testing has recently

evolved as a valuable tool for crisis management in the hands of authorities. This has brought macro

stress-testing in the realm of prudential supervision. Top-down models have made important advances

to improve the macro-micro link by using disaggregated data. Moreover, the development of

integrated models enables to simulate feedback effects within the banking sector, although modelling

second round effects to the real economy remains an important field for further research.

The outcomes of macro stress-tests are inherently uncertain due to the subjectivity of the

scenario choice and model risk. Therefore it is advisable to combine both bottom-up and top-down

methods in policymaking. Both approaches complement each other in a valuable way and using them

along side each other allows for a cross-check and provides a range of outcomes which helps to

quantify the margins of uncertainty.

62

63

Chapter 5

Modelling scenario analysis and macro stress-testing

5.1 Introduction

Scenario analysis and macro stress-testing are key instruments in most monitoring frameworks for

financial stability.20 They enable a quantitative and forward looking assessment of the resilience of the

financial system to exogenous shocks. Most authorities with a responsibility for financial stability

publish the outcomes of macro stress-tests in their financial stability reports on a regular basis. For

instance, De Nederlandsche Bank (DNB) reports the outcomes in its Overview of Financial Stability

(OFS). This practise intends to raise the awareness of market participants to downside risks and to

encourage them to be forward looking in their risk management. This provides an important

contribution to macroprudential supervision, which aims at the resilience of the financial sector as a

whole.

This chapter describes a tool kit for scenario analysis and top-down macro stress-testing. The

methodology is applied to the Dutch banking sector. First we simulate macroeconomic scenarios by a

structural macroeconomic model, in line with the common practise of macro stress-testing by central

banks and the IMF Financial Assessment Programs (FSAP). The macro scenarios are mapped in

banks’ credit and interest rate risk, by estimating reduced form satellite models that link the exogenous

shocks in the macro variables to micro drivers of bank risk, i.e. credit quality indicators and an interest

income measure. In this step, we use disaggregated data consisting of a panel of individual banks and

a break-down of domestic and foreign portfolios, to capture the different responses of banks and

portfolios in stress situations. This addresses a major shortcoming of macro stress-testing models

based on aggregate data, which may conceal significant variation at the portfolio or bank level. We

further explore the variation in the credit loss distribution by estimating both the probability of default

(PD) and the loss given default (LGD) in bank loan books. This is done by using nonlinear

specifications, since ignoring nonlinearities in the relationship between macro variables and credit risk

can lead to a substantial underestimation of risk, particularly when considering large shocks

(Drehmann et al., 2006).

In addition, we propose an alternative approach for scenario simulation, based on a vector

autoregressive (VAR) model. It allows for simultaneous changes in the macro variables and portfolio

drivers of bank loans and changing correlations between them in stress situations. The stochastic VAR

simulations generate loss distributions that provide insight in the extreme losses. The idea of

measuring the impact of shocks in terms of an overall system wide loss distribution of the banking

sector builds on Sorge and Virolainen (2006).

20 This chapter is a revised version of Hoeberichts, Tabbae and Van den End (2006).

64

This rest of this chapter is organised as follows. Section 5.1 presents a general framework and

an approach for modelling macro scenarios. Section 5.2 deals with methods that could be used to map

these macro scenarios to the portfolios of banks. Sections 5.3 and 5.4 explain the models for stress-

testing of credit risk and interest rate risk that can be used for simulating the first round impact of

shocks. In Section 5.5 this is done with a deterministic and a stochastic model. Section 5.6 concludes

and identifies areas of possible further research.

5.2 Scenario building

Figure 5.1 presents a stylised framework for macro stress-testing. The process begins with the

selection of extreme but plausible shocks. These can be univariate shocks in single risk factors such as

an isolated decline of equity prices. Univariate shocks can be combined into multivariate scenarios, in

which various (macro) risk factors change. For instance, in one of the scenarios that we use, a

depreciation of the dollar exchange rate is combined with falling GDP and rising interest rates.

Multivariate scenarios are more realistic than univariate shocks (sensitivity tests), since in stress

situations risk factors usually interact. Scenarios can be developed through a number of methods

(Hoggarth et al., 2005). First, they can be designed with a structural macroeconomic model, which

generates projections of macro variables, sometimes as deviations from a base line scenario. These

scenarios can be based on historic events (e.g., the 1998 emerging market crisis) or on hypothetical

assumptions. Analysing the economic reasoning behind those scenarios is an important element in

using them for financial stability policy purposes. Structural macro models help to achieve this and

contribute to the consistency of the generated paths of macro variables. Hypothetical scenarios might

seem less plausible than historic scenarios, but they can be forward looking and sufficiently flexible to

formulate events that could significantly affect the financial sector. Macro models have their

shortcomings for stress-testing since overshooting and spill-over effects of financial prices, which are

typical for stress situations, are taken into account by adding assumptions, because they are generally

not part of the model itself. In addition, the estimated parameters of the models may not be stable in

stress situations. Second, scenarios can be developed by a probabilistic method, in which shocks are

based on stochastic simulations of macro variables (Drehmann, 2005), which can be extended to

include tail dependence between macro-financial variables (Boss et al., 2006). The tail outcomes of

the probabilistic simulations present extreme scenarios. As an alternative, a scenario could be based on

the tail outcomes of distributions of financial sector losses. In this so-called ‘reverse engineered’

approach, the change in the (macro) risk factors that corresponds to these losses determine the

scenario. Third, transition matrices of credit ratings can be used to stress-test credit portfolios of

banks. Herein, adverse scenarios are presented by probabilities of a rating downgrade that are due in a

recession (Peura and Jokivuolle, 2003). Finally, scenarios can be simulated by a VAR model, wherein

65

a set of macro variables is affected by the initial shock and the vector process is used to project the

stress scenario (Hoggarth, 2005). VAR models are more flexible than structural macro models, but do

not provide for an economic foundation structure of a stress scenario.

In this chapter we combine the first approach (by designing multi-year scenarios with a

macroeconomic model) with the last approach (by using a VAR model to simulate stochastic stress

scenarios and changing correlations between risk factors). First, a base scenario for the Dutch

economy is projected which is based on DNB’s macro model MORKMON (Van Els, 2005). This

model is also used for the projections of economic growth and inflation over a horizon of one to three

years. For the financial stability assessment, next to the base scenario some alternative, hypothetical

scenarios are developed with the NIGEM model.21 MORKMON and NIGEM are large-scale structural

econometric models. The alternative scenarios are specified by assuming a set of initial shocks. These

shocks are used as exogenous input in NIGEM. The interactions between the initial shocks and the

other macroeconomic variables over the scenario horizon follow from the model. Monetary policy

rates are assumed to remain constant in this analysis, so monetary rules like the Taylor rule are

excluded. The type and size of the shocks are based on extreme percentiles of time series and

fundamental imbalances, e.g. the overvaluation of house prices or exchange rates. These ‘realisations’

are the starting point for the construction of the hypothetical scenarios which are further based on an

economic assessment (expert opinion) of how risk factors could evolve in the future. Besides, the

specification of the scenarios depends on their potential impact on the Dutch financial sector. This is

done by tailoring the scenarios to the main risk exposures of the financial institutions, on both sides of

the balance sheet. The Dutch financial sector faces international risk factors, owing to its large cross-

border exposures. The main risk exposures of the Dutch banks - which are the subject in this chapter -

are related to (international) interest rate and credit risk. Hence, these factors feature prominently in

our scenario analysis.

Source: Bank of England

21 World model of the National Institute of Economic and Social Reearch (http://www.niesr.ac.uk).

Initial shock

Interaction of risk drivers in scenario

Process for borrower defaults

Impact on P&L and capital

Feedback effects on economy, markets

Mapping defaults into losses

Figure 5.1. Stress-testing framework

Earnings financial sector

66

Usually macro stress-testing is based on comparing outcomes under the base scenario and those under

alternative scenarios. For example, in previous issues of the DNB’s Overview of Financial Stability

(OFS) two alternative scenarios, the Malaise scenario and the Global correction scenario, were

formulated (see Appendix 5.1 and DNB, 2005). These scenarios are centred around two diverging

trends of interest rates. The most important financial stability risk in the Malaise scenario concerns the

implications of falling interest rates and a flattening yield curve for financial institutions. In the Global

correction scenario, both credit and market risks are adversely affected by increasing interest rates. By

presenting these two scenarios, the OFS tries to encourage market participants to take into account

both directions of possible interest rate shocks in their risk management.

The mapping of a macro scenario to the portfolios of the financial sector is the third step in the

stress-testing framework (the boxes in the middle of Figure 5.1). For this, most central banks follow a

top-down approach, i.e. use inhouse models. These are usually reduced form satellite models,

specified for estimating credit risk of banks. Such models relate the position of borrowers to macro

variables (‘Process for borrower defaults’ in Figure 5.1) and relate borrowers’ defaults to losses of the

financial sector (‘Impact on P&L and capital’ in Figure 5.1). This last step could also be performed

directly (see arrow in Figure 5.1 pointing from ‘Interaction of risk drives’ to ‘Mapping defaults into

losses’), without explicitly estimating the process for borrower defaults. By this, the first round effects

of shocks on the financial sector are estimated. Modelling second round effects (i.e. feedback effects

on the economy and the financial markets, see the far right panel in Figure 5.1) is more complex and

remains an issue that is yet in its early stage of development (see Chapter 4). This chapter follows the

mainstream literature and focuses on first round effects only.

5.3 Credit risk

For stress-testing the credit risk of the banking sector, we have developed reduced-form satellite

models. The models are mainly developed to quantitatively underpin the scenario analysis, by

quantifying the first round effects of shocks.

5.3.1 Model

In modelling credit risk, we use two basic equations. In equation 5.1, the relationship between

borrower defaults and some key macro variables is estimated (process for borrower defaults). In

equation 5.2, the default rate together with some macro variables are used to explain loan loss

provisions (LLP) and map defaults into losses. This procedure is given by the following equations:

( ) ttttt )RSRL(GDPeDefaultrat υββαλ +−++= 21 (5.1)

67

ttttit,i

)eDefaultrat(RLGDPeffectsfixedCRED

LLP ηλβββλ ++++=

321 (5.2)

Defaultratet is the number of defaults relative to the population of firms. GDPt stands for real GDP

growth, RLt for the long-term interest rate, RSt for the short-term interest rate and RLt - RSt for the term

spread. (LLP/CRED)i,t is the ratio between LLP and loans outstanding of bank i. The regressors have

been chosen out of various macro variables because their parameter estimates provide for a good fit

and have the expected sign; see Section 5.3.3. (We have also estimated the equations including real

effective exchange rate, unemployment, house prices, stock prices and oil prices.). By using different

constant terms in equation 5.2 (fixed effectsi)22, the structural differences in the level of provisions for

each bank is taken into account. This is done to include bank specific characteristics, which in other

studies are included through bank specific control variables (e.g. in Bikker and Hu, 2002). While the

inclusion of such micro data provides insight into the underlying bank fundamentals, for our purpose

of linking bank’s balance sheets to macroeconomic scenarios we can restrict the model to the limited

number of key variables that drive macro scenarios. The parameters represent the marginal effects

(assumed to be uniform across banks) of macro variables on loan loss provisions, allowing for bank-

specific intercepts. In the equations, non-linear functions of Defaultratet and (LLP/CRED)i,t – the logit

– are used to extend the domain of the dependent variable to negative values and to take into account

the possible non-linear relationships between the macro variables and LLP. Non-linearities are likely

in stress situations as shocks could lead to extreme outcomes in credit losses (credit risk is not

normally distributed). Several other studies on stress-testing models take non-linearities into account

by a logit transformed provision ratio (Lehmann and Manz (2006) and Bundesbank (2006)); others

include squares and cubes of macro variables (Drehmann et al, 2005). The logit is defined as:

−=

t

tt X

XX

1ln)(λ (5.3)

where Xt is Defaultratet as defined in equation 5.1 and (LLP/CRED)i,t as defined in equation 5.2. Using

these two basic equations the Loss Given Default (LGD) can be derived implicitly, by using the

identity:

EL (Expected Loss) = PD (Probability of Default) * LGD * EAD (Exposure at Default) (5.4)

In terms of our model equations, EL/EAD is approximated by (LLP/CRED)i,t and PD by Defaultratet.

In most macro stress-testing models, LGD is assumed to remain constant. By estimating equation 5.1

22 We have tried random effects estimation as well. This gives a near-singular matrix as a result of too few cross-section observations in the foreign loans model. For domestic and total loans, estimation with random effects gives parameter estimates very similar to those obtained with fixed effects.

68

and 5.2 and then including )CRED/LLP( and Defaulrate in identity 5.4, LGD can be estimated

implicitly. This allows for determining the impact of stress scenarios on LGD, next to the impact on

Defaults (Defaultratet) and losses ((LLP/CRED)i,t). While both Defaultratet and (LLP/CRED)i,t are

regressed on the same macro variables, the impact of the macro variables on default risk is isolated

from their impact on losses. The implicit method is different from (and less efficient than) modelling

LGD separately as is done by, for instance, Coleman et al. (2005) who explain the LGD out of the

loan-to-value ratio and the age of the loan. Their method requires detailed information on individual

loans, which is not available for the Netherlands. Several studies show quite different outcomes of

LGDs in downturns, dependent on the underlying portfolio and the region (Frye, 2000, Altman et al.,

2004, Trück et al., 2005).

Besides LGD, we also take into account the typical risk factors that drive the credit risk of different

bank portfolios. This is done by specifying different model versions for (LLP/CRED)i,t and

Defaultratet for the domestic and foreign loan books, next to the total loan book, as in equations 5.5-

5.7.23

ttttit,i

)NL_Defaulrate(NL_RLNL_GDPeffectsfixeddom_CRED

dom_LLP ηλβββλ ++++=

321 (5.5)

ttttit,i

)world_Defaulrate(NL_RLEU_GDPeffectsfixedfor_CRED

for_LLP ηλβββλ ++++=

321 (5.6)

ttttit,i

)world_Defaulrate(NL_RLEU_GDPeffectsfixedtotal_CRED

total_LLP ηλβββλ ++++=

321 (5.7)

Defaultrate_NL and Defaultrate_world are the failure rate of domestic businesses, resp. the default rate

on global corporate bonds. As in equation 5.1, they are explained by GDPt and RLt - RSt, which in case

of Defaultrate_NL are the growth rate of domestic GDP and Dutch interest rates and in case of

Defaultrate_world the growth rate of US GDP and US interest rates. LLP_dom (LLP_for) stands for

LLP related to the domestic (foreign) portfolios and LLP_total for LLP related to the consolidated

portfolio of the banks. CRED_dom (CRED_for) stands for loans outstanding in the domestic (foreign)

portfolios and CRED_total for loans outstanding in the consolidated portfolio of the banks. The

domestic loan book of the Dutch banks is dominated by retail (mostly mortgage) loans, while the

foreign and the consolidated portfolios are dominated by wholesale exposures. To fit these different

risk profiles of the portfolios, the credit quality of domestic loans is explained by domestic risk factors

(GDP_NL and Defaultrate_NL in equation 5.5) and the credit quality of the foreign and consolidated

23 The model estimations for LLP on foreign loans are based on data of the three banks that have foreign exposures outstanding.

69

loan portfolios by global risk factors (GDP_EU and Defaultrate_world in equation 5.6). With respect

to the consolidated loan portfolio of the banks, equation 5.7 includes credit risks from foreign

branches or subsidiaries abroad. By this group-wide approach, cross-border risks are taken into

account to some extent.

5.3.2 Data

The modelling of the credit risk of Dutch banks is restricted by the availability of data. As suggested

by Jones et al. (2004), we use LLP as a reference value for credit quality (LLP being the additive

provisions), since sufficiently long series of non-performing loans (NPL) are not available. A

disadvantage of using LLP is that it is an accounting concept, which does not necessarily reflect

default risk. Another limitation is that no sectoral break-down of credit portfolios of Dutch banks is

available. To allow for the different risk profiles of portfolios, we distinguish between domestic,

foreign and total loans. Defaultrate_NL is determined by bankruptcies in the domestic corporate sector

over the number of registered Dutch companies, and Defaultrate_world is determined by worldwide

defaults on corporate bonds over the number of bonds outstanding worldwide. RL is the ten years

government bond yield, whereas RS is the three months risk free rate. The source of the bank specific

data (LLP, CRED and the macroeconomic variables) is DNB. Sources for Defaultrate_NL and

Defaultrate_world are Statistics Netherlands, and Standard&Poors, respectively.

The credit risk models are estimated with annual data, covering the 1990-2004 period, as

quarterly data are only available since 1998 and annual data are not available before 1990. By using

annual data more business cycles are included. The number of observations is increased by including

cross-sectional data in the estimations of equations 5.5-5.7. We use a panel data set of the largest five

banks in the Netherlands, which represent approximately 85% of the banking sector’s total assets.

5.3.3 Estimation results

Table 5.1 shows the estimation results of equations 5.1, with Defaultrate_NL and Defaultrate_world as

dependent variables. The parameter estimates of GDP and RL - RS are both significant with the

expected (negative) sign. A decreasing term spread either means that short term rates increase, e.g.

through tightening monetary and financial conditions, or that long term rates decrease, which may

reflect a subdued outlook for inflation and the business cycle. Both could raise credit risk. Besides, the

yield curve is an indicator for the business cycle. A flattening curve might point to decelerating

economic growth, which again could raise credit risk.

70

Table 5.1. Estimation result of regressions of default rate (equation 5.1)

Defaultrate_NL Defaultrate_world

Constant -4.47*** -3.17***

(-59.0) (-9.3)

GDP_US -0.27**

(-2.6)

GDP_NL(-1) -0.10***

(-4.8)

GDP_NL (-2) -0.06**

(-2.3)

RL_NL – RS_NL -0.04*

(-1.8)

RL_US(-1) – RS_US(-1) -0.26**

(-2.3)

Observations 15 15Adjusted R-squared 0.85 0.58

SEE 0.09 0.54

DW statistic 1.17 1.02

Prob (F-statistic) 0.00 0.01

t-statistic in parentheses, ***, **, * denote statistical significance at 1, 5, 10% confidence level

OLS, sample period 1990 2004, annual data

Table 5.2 shows the results of the panel regressions of equations 5.5-5.7. Herein, Default rate has been

included as an explanatory variable. The estimation results show that it contributes significantly to

explaining (LLP/CRED)i,t, both in the case of domestic, foreign and consolidated exposures. The

parameter estimates for GDP and one year lagged RL also have the expected sign and are significant

most of the times. In all three equations, interest rates are leading on (LLP/CRED)i,t as could be

expected, since interest rates usually lead the business cycle and hence credit quality. The different

size of the fixed effects (not reported) suggests that the Dutch banks have a different sensitivity to

macroeconomic developments, owing to their typical risk profiles.

71

Table 5.2. Estimation result of regressions of LLP (equations 5.5 - 5.7)

LLP_dom LLP_for LLP_total

GDP_EU -0.03 -0.08***(-1.2) (-7.2)

GDP_EU(-1) -0.13*** -0.09***(-3.4) (-6.5)

GDP_NL -0.06*(-1.8)

RL_NL(-1) 0.16*** 0.21*** 0.14***(6.7) (6.0) (10.2)

Defaultrate_NL 1.02***(4.3)

Defaultrate_world 0.41*** 0.17***(5.4) (5.8)

Observations 70 45 70Adjusted R-squared 0.97 0.92 0.99SEE 0.48 0.33 0.37DW statistic 1.72 1.37 1.46Prob (F-statistic) 0.00 0.00 0.00

GLS (cross section weights), sample period 1990 2004, annual data

t-statistic in parentheses, ***, **, * denote statistical significance at 1, 5, 10% confidence levelFixed effects not reported. White Heteroskedastic-consistent standard errors and covariance

5.4 Interest rate risk

5.4.1 Model

Macro stress-testing models are usually confined to credit risk (ECB, 2006). However, interest rate

risk is another important source of banks’ profitability and capital base and, hence, their stability.

Changes in interest rates affect earnings by changing net interest income and other interest sensitive

income and expenses (earnings perspective). Changes in interest rates also affect the underlying value

of the bank’s assets, liabilities and off-balance sheet instruments, because the present value of future

cash flows changes when interest rates change (economic-value perspective). Since the economic-

value perspective takes into account the potential impact of interest rate changes on the present value

of all future cash flows, this would be the preferred measure for interest rate risk (Basel Committee on

Banking Supervision, 2004). However, as most banking assets (i.e. loans) are held to maturity the

economic-value perspective is less relevant in practise. Besides, there are data limitations involved in

modelling the value effect of interest rate risk as long time series of banks’ balance sheets at economic

value are not available (only since 2005 banks have to report (parts of) their balance sheets at fair

value, following the International Financial Reporting Standards, IFRS).

For macro stress-testing only a few central banks have modelled interest rate risk, as part of

modelling the banking sector’s profitability. The Bank of England explains interest income out of

GDP growth, in a reduced form fashion (Bunn et al., 2005). A new model developed at the Bank of

England also takes into account the interest rate effects on the economic value of a bank (Drehmann et

72

al., 2006). Banque de France specifies a reduced form equation for interest income based on a panel

dataset of French banks (De Bandt and Oung, 2004). The yield spread, its volatility, lending growth

and the cost of risk are used as explanatory variables. We employ a similar model for the growth rate

of the net interest income of Dutch banks (in equation 5.8),

ttttttt,i RSRLGDPGDPGDPeffectsfixed)IncomeInterestNet(Ln µβββββ ++++++=∆ −−− 51423121 (5.8)

in which the growth rate of net interest income is explained by the (lagged) growth rate of real GDP,

the lag of the long-term interest rate and the current short-term rate in the euro area. The lags have

been chosen so as to yield the best fit. When GDP increases, we expect NetInterestIncomei,t to

increase through an expanding supply of loans (volume effect). RS is representative for the banks’ cost

of funding, which on average is attracted at short terms. Hence, a rise of RS lowers a bank’s interest

rate margin and reduces NetInterestIncomei,t. RL is representative for the banks’ lending rate, since on

average banks loans are issued on longer terms (80% of the loans of Dutch banks has a maturity of

five years or more). Hence, an increase of RLt increases NetInterestIncomei,t. The lag of RL is used

rather than the current rate because the change of the market interest rate will not immediately affect

the interest rate that banks receive on their outstanding loans. The parameters of RS and RL capture the

price effect on net interest income.24

5.4.2 Data and estimation results

The interest rate risk model has been estimated with annual data, covering the 1994-2005 period,

including cross-sectional data for NetInterestIncomei,t (based on the same set of banks as in the credit

risk model). The estimation results are summarized in Table 5.3. We find significant parameter

estimates with the expected signs for all parameters, except for one-year lagged GDP. The combined

effect of GDPt, GDPt-1 and GDPt-2 is larger than the effect of RLt and RSt, which implies that the

volume effect is more important than the price effect. The adjusted R-squared of the interest rate

model is lower than those of the credit risk models, which could be expected since the explanatory

power of models with variables in terms of changes (as in the interest rate model) is usually lower than

that of models specified in levels (or ratios as in the credit risk models).

24 Including the term spread of interest rates rather than RS and RL separately implies a restriction that the parameters of RS and RL are the same. Testing this restriction reveals that it does not hold.

73

Table 5.3. Estimation result of regressions of Net Interest Income (equation 5.8)

∆ NetInterestIncome

GDP_NL 0.02**

(3.3)

GDP_NL(-1) -0.01**

(-2.3)

GDP_NL (-2) 0.03***

(6.3)

RK_NL(-1) -0.03***

(-3.8)

RL_NL(-1) 0.01**

(2.3)

Observations 59

Adjusted R-squared 0.25

SEE 0.09

DW statistic 2.15

Prob (F-statistic) 0.00

GLS (cross section weights), sample period 1990 2005, annual data

t-statistic in parentheses, ***, **, * denote statistical significance at 1, 5, 10% confidence level

Fixed effects not reported. White Heteroskedastic-consistent standard errors and covariance

5.5 Simulation of scenario effects

After having designed the scenarios and specified the stress-testing models, both can be linked by

simulating the impact on financial sector exposures. First, we conduct deterministic scenario analysis,

by using the average macro variables as projected by NiGEM as input in the stress-testing model. This

assumes no uncertainty about the forecasted macroeconomic variables. The advantage of this approach

is that the results are easy to understand and provide insight in the quantitative links between macro

variables and financial soundness indicators. However, the deterministic analyses only generate one

future path of outcomes without allowing for uncertainty in the projections. However, uncertainty is

inherent to hypothetical scenarios, even more so if they have a multi-year horizon. We therefore

perform stochastic scenario analysis as well, to generate probability distributions of the impact on the

financial sector. This yields a more complete description of the scenario outcomes. The tails of the

distributions also provide insight in the likelihood of extreme losses which is relevant from a financial

stability perspective. The deterministic and the stochastic scenario analyses have been based on the

Malaise scenario and the Global correction scenario from DNB’s OFS (see Appendix 5.1). We are

able to distinguish between the impact on domestic exposures and foreign exposures.

5.5.1 Deterministic scenarios

In the deterministic scenarios for credit risk, the deviation of the macro variables from the base line

(following from the MORKMON and NiGEM model simulations) are input in equations 5.1-5.2 and

74

5.8. By this, the impact on credit risk (Default rate, LGD and LLP) and interest income can be

projected over one to three years horizons. These are point estimations since they result from using the

path of macro variables as projected by the NiGEM model as input in the stress-testing model.

Panels A-F of Figure 5.2 show the projected log-transformations of changes in Default rate,

LGD and LLP/CRED, of which the three year results are cumulative effects. The figures clearly show

that the Global correction scenario has the largest impact on credit risk, raising LLP/CRED on average

by 65 to 92%. In this scenario, both the rise in interest rates and the decline of GDP adversely affect

credit risk, whereas in the Malaise scenario the decline of interest rates compensates for the subdued

business cycle effect. The impact over time of the Global correction scenario illustrates the benefit of

using a multi-year horizon. The impact on foreign exposures is more frontloaded than on domestic

exposures, which are substantially affected only after three years. A possible explanation for this is

that movements of international risk factors take time to affect the domestic economy and hence the

credit quality of domestic loans. The wholesale exposures, which dominate the foreign loans, are more

directly affected by international risk factors. The average total loss in the Global correction scenario

in the three years period is EUR 2.2 billion, which equals around ¼ of one years’ profits and 2.5% of

total own funds of the Dutch banking sector (2005 data25). The total capital ratio of the banking sector

declines from 11.5 to 11.2% ceterus paribus. This relatively low impact is partly related to the low

base levels of PDs and LLPs in 2004. As could be expected, the Malaise scenario has most adverse

consequences for the domestic loan book, which is more dependent on developments in the euro area

than the consolidated book, which is internationally diversified.

Figure 5.2. Outcomes deterministic scenarios

-20%

0%

20%

40%

60%

80%

PD LGD LLP/CRED

1 year 3 years

0,1 bn

1.3 bn

Panel A. Global correction and domestic credit portfoliopercentage change, deviations from base scenario

0%

20%

40%

60%

80%

100%

PD LGD LLP/CRED

1 year 3 years

Panel B. Global correction and foreign credit portfoliopercentage change, deviations from base scenario

0.7 bn

2.0 bn

0%

10%

20%

30%

PD LGD LLP/CRED

1 year 3 years

0.1 bn

0.2 bn

Panel D. Malaise and domestic credit portfoliopercentage change, deviations from base scenario

0%

10%

20%

30%

PD LGD LLP/CRED

1 year 3 years

Panel E. Malaise and foreign credit portfoliopercentage change, deviations from base scenario

0.1 bn0.2 bn

0%

10%

20%

30%

PD LGD LLP/CRED

1 year 3 years

Panel F. Malaise and total credit portfoliopercentage change, deviations from base scenario

0.2 bn0.2 bn

0%

20%

40%

60%

80%

PD LGD LLP/CRED

1 year 3 years

Panel C. Global correction and total credit portfoliopercentage change, deviations from base scenario

0.6 bn

2.2 bn

25 The sum of the separate outcomes for losses on domestic and foreign loans is not equal to the outcome for the total loan book since by estimating two equations a larger part of the variance is explained. Besides, the log transformation does not allow for a simple summation of the model outcomes.

75

Panel G and H of Figure 5.2 show the simulated effects on interest income. The Global correction

scenario appears to have a more negative impact on interest income than the Malaise scenario,

although in the former the (relatively sharp) decline of GDP is accompanied by a widening term

spread of interest rates. In the Malaise scenario, both the change of interest rates (through a tightening

term spread) and GDP have an adverse impact on interest income. Even so, the Global correction

scenario has the most negative impact due to the dominating influence of the volume effect over the

price effect.

-3

-2

-1

0

1 yr 1-3 yr cum.

Panel H. Global correction and interest incomeEUR bn, deviations from base scenario

-3

-2

-1

0

1 yr 1-3 yr cum.

Panel G. Malaise and interest incomeEUR bn, deviations from base scenario

5.5.2 Stochastic simulation of credit risk in base scenario

The stochastic scenario analysis follows Sorge and Virolainen (2006), who simulate default rates over

time by generating macroeconomic shocks to the system. The model for credit risk that governs the

joint evolution of Default rate, LLP/CRED and the associated macroeconomic shocks is given by

equations 5.1 and 5.2 and a set of univariate autoregressive equations of order 2 (AR(2)), to estimate

the macroeconomic variables:

tttt xkxkkx ε+++= −− 22110 (5.9) where k0..n are the regression coefficients to be estimated for the macroeconomic factors (xt) used in

equations 5.1 and 5.2. Equation 5.9 is estimated with GDP growth rates and interest rate levels which

are non-stationary according to unit roots tests. As an alternative, we also tried a multivariate

specification by modelling the macroeconomic factors (xt) as a Vector Autoregression (VAR) model.

The VAR (2) model takes into account the correlations between the macro variables.

tttt XKXKKX ε+++= −− 22110 (5.10)

where Xt is a vector of macroeconomic variables, K0..n is a vector of coefficients to be estimated and εt

is a vector of innovations.

76

The system of equations 5.1, 5.2 and 5.9 (or 5.10) is completed by a vector of innovations, E, and a

variance-covariance matrix of errors, ∑:

∑∑∑

∑∑∑

∑∑∑

=∑∑

=εηευε

εηηυη

ευηυυ

εηυ

,,

,,

,,

,),0(N~E

With this system of equations the future paths of the macroeconomic variables, Default rate and

LLP/CRED can be simulated with a Monte Carlo method. The simulations are carried out by taking

random draws of variables Zt+s ~ N(0,1). These are transformed into correlated innovations in the

macroeconomic factors, Default rate and LLP/CRED by Et+s = A’ Z t+s, where A’ results from the

Cholesky decomposition26 ∑ = AA’. The simulated error terms and the initial values of the

macroeconomic variables are then used to derive the corresponding values of the macroeconomic

variables (xt+s), Defaultratet+s and (LLP/CRED)t+s by using equations 5.1, 5.2 and 5.9 (or 5.10). With

these outcomes and the information on outstanding exposures of the banking sector, distributions of

credit losses can be determined. Figures 5.3a-b show the probability distributions of losses over a one

and a three years horizon (the horizons have no material influence since all the simulations are based

on a normal distribution). These results are conditional on the bank exposures in the last year of the

data series (2004) and present a stochastic base scenario for the next years. The stochastic base

scenario differs from the deterministic base scenario as projected with the macroeconomic models

since it takes into account the uncertainty around the average future paths of macro variables. Like a

typical distribution of credit risk, the simulated distributions of losses are skewed to the right, due to

the correlation structure of the innovations.27 It is striking that the loss distributions for domestic

exposures are more skewed than the distribution for total exposures. This can be explained by the

volatility of domestic GDP, which is larger than the volatility of euro area GDP (GDP is the most

important driver in the model).

Panels A and B of Figure 5.3 show that the outcomes of the simulations based on the AR-model (5.9)

and the VAR-model (5.10) are very similar. This indicates that the specification of the model which

generates the macro factors has no material influence, from which we may conclude that the statistical

objections (such as the non-stationarity of the data) are neither materially important. The robustness of

the model has also been explicitly tested for by using alternative specifications of the variance-

26 The results are not dependent on the order of the error terms in matrix ∑ since the simulated innovations are applied to the corresponding equations 5.1, 5.2 and 5.9 (or 5.10). 27 All elements in the variance-covariance matrix of errors ∑ are positive, except for the correlation between the error terms of the interest rate and GDP (however, the parameter estimates for interest rate and GDP have an opposite sign in equation 5.2, by which the effect of the negative correlation between the error terms has the same direction on LLP/CRED).

77

covariance matrix of errors ∑.28 This is done by applying impulse responses of one standard deviation

to the error terms of the macroeconomic factors (ε) and to the covariance between these error terms in

∑. Panels C and D show that the system of equations is fairly robust to different specifications of ∑;

the probability distribution of losses on total loans is hardly affected. The impulse responses have a

larger impact on domestic loan losses (the average expected loss changes by around 6%) which is

most pronounced for extreme losses (the 1% tail outcome changes by nearly 20%). We also check for

the sensitivity of the outcomes for a different starting year from which the scenario estimations

proceed. To this end, the system of equations has been re-estimated for total and domestic loans with a

cut-off date of the data series at 2001, which we assume to be the new starting year of the simulations.

The resulting probability distributions of losses are subsequently applied to the exposures outstanding

at end-2001. Panels E and F indicate that the shape of the loss distributions does not change much if

another starting year is used, which again illustrates that the model is fairly robust.

28 Implicitely, ∑ also changes due to the specification in an AR, respectively VAR mode.

78

Since LLP/CRED and Default rate are separately modelled, the probability distribution of LGD is also

dependent on the simulated macroeconomic variables (implicitly via identity 5.4). This comes close to

the ‘integrated approach’ for macro stress-testing, in which all parameters are functions of a vector X

of macroeconomic variables, which evolve over time following an autoregressive stochastic process,

and can be summarised into a value-at-risk measure (Sorge and Virolainen, 2006):

{ })x();x(LGD);x(PD;)x(Ef)xx~/y~(VaR ttttttttttt ∑=≥++ 11 (5.11)

where xx~/y~ 1t1t ≥++ represents the uncertain future realisation of the aggregate credit loss (ỹt+1) for the

financial system in the event of a simulated macroeconomic stress scenario (i.e. conditional on a tail

realisation of xx~ 1t ≥+ ). The difference with a ‘fully integrated approach’ is that the effect on market

prices is not taken into account since our model is a default mode (with losses stemming from

counterparty defaults) and not a mark-to-market framework (with changes in portfolio values

associated with changes in credit quality), such as, for example, the model of Drehmann (2005).

5.5.3 Stochastic simulation of credit risk in stress scenarios

To simulate the hypothetical stress scenarios, the future values of the macroeconomic factors as

projected by the NiGEM model are included to re-estimate the VAR model (5.10). The resulting new

error terms (εt) change the corresponding elements in the variance-covariance matrix ∑. Next, Monte

Carlo simulations are carried out by taking random draws of Zt+s ~ N(0, σstress / σbase)29 for the

innovations in the macro variables used in the VAR model, and Zt+s ~ N(0,1) for the innovations used

in equations 5.1 and 5.2 in the model. Loss distributions for the assumed stress scenario can then be

determined with the simulated paths for macroeconomic variables (xt+s), Defaultratet+s and

(LLP/CRED)t+s, as in the base scenario. To simulate the Malaise and Global correction scenarios, the

VAR model is re-estimated, including equations for GDP, RL and RL-RS. This approach differs from

Sorge and Virolainen (2006), who only simulate single-factor shocks in GDP and the interest rate. We

apply multi-factor simulations by taking into account simultaneous changes in the macroeconomic

variables and their interactions. The interactions are taken into account in the variance-covariance

matrix ∑, including the change in the correlations, which result from re-estimating the VAR for the

macroeconomic factors. This is an important advantage of our approach since in stress situations the

historical correlations between risk factors can change.30

29 σstress results from the error terms of the re-estimated VAR model and σbase from the error terms of the original VAR . By the ratio σstress / σbase, the innovations in the stressed macro factor are normalised by the respective standard deviation of the error terms in the base scenario. 30 Maximum stress could be simulated by assuming full correlation between the macro variables (i.e. interest rate and GDP growth). However, the model does not allow for a high (absolute) covariance that would be imposed in the variance-covariance matrix, since then it is no longer positive definite which, in turn, precludes a Cholesky

79

The stochastic simulations have been applied to the domestic, foreign and total portfolios of

the banks, by using equations 5.5-5.7 in the simulations. As in the deterministic scenarios, the impact

of the Global correction is larger than the Malaise scenario (Figure 5.4, panels A-F). In the latter, the

probability distribution of credit losses is close to the base scenario, since the decline of interest rates

compensates for the subdued business cycle effect. Over the three years horizon in the Global

correction scenario, the average credit loss on the total loan portfolio of the Dutch banks increases by

EUR 1.9 billion compared to the base scenario.31 For the foreign and domestic portfolios, the figures

are EUR 0.5 and 0.7 billion, respectively. This is less than indicated by the deterministic scenarios, but

the outcomes of the deterministic and stochastic simulations are not fully comparable since they are

based on different model structures (in the stochastic scenarios the macroeconomic variables are

estimated by VAR models). The benefit of stochastic simulations is that they provide insight in the

possible extreme outcomes in the right tail of the probability distributions. Compared to the base line,

in the Global correction scenario the tail outcomes are much larger since the probability distribution is

flatter.32 This is most pronounced for the foreign portfolios where the loss amount at the 99th percentile

more than doubles from around EUR 0.8 billion in the base scenario to around EUR 1.8 billion in the

stress scenario (3 years horizon, panel B). This compares with an increase of extreme losses in the

total loan portfolio from around EUR 7.1 billion in the base scenario to around EUR 9.5 billion in the

stress scenario (99th percentile, 3 years horizon, panel C). The relatively strong flattening of the

distribution of foreign loans losses indicates the sensitivity of this portfolio to shocks in international

risk factors, in particular of Defaultrate_world, for which the coefficient in equation 5.6 (explaining

LLP_for) is much larger than in equation 5.7 (explaining LLP_total). Besides, the covariance between

GDP and the interest rate which results after Cholesky decomposition is larger in case the system of

equations is estimated with LLP_for than with LLP_total. The stronger interaction between the macro

variables leads to more widely dispersed simulation outcomes.

decomposition. The maximum absolute covariance that could be imposed equalises to a correlation coefficient of 0.77. Simulations that we have conducted with imposed (limited) higher covariances between interest rate and GDP growth indicate that a higher covariance does not lead to significant higher outcomes for loan loss provisions. 31 The sensitivity of the outcomes to different specifications of the variance-covariance matrix ∑ is limited; including impulse responses of one standard deviation to the error terms of the macroeconomic factors (ε) and to the covariance between these error terms in ∑ changes the estimated average losses by less than 1% and the one percent extreme loss by less than 5%. 32 The kurtosis of the probability distribution for the base scenario is -1.00 and for the global correction scenario -1.52 (3 years horizon).

80

5.5.4 Stochastic simulation of interest rate risk in stress scenarios

Stochastic simulations can also be applied to net interest rate income, by using equations 5.8, the VAR

model (5.10) and a vector of innovations E, with ),0(N~E ∑

=εµ , in the same way as we used the

system of equations 5.1, 5.2 and 5.10 and vector E for credit risk. For this, the simulated error terms in

E and the initial values of the macroeconomic variables are used to derive the corresponding values of

the macroeconomic variables (xt+s) and net interest income (NetInterestIncomet+s). With these

outcomes and the total net interest income of the banking sector, distributions of the change in net

interest income can be determined. Next, distributions can be determined for the assumed stress

scenarios. Like in the stochastic simulations for credit risk, the future paths of the macroeconomic

variables in the Malaise and Global correction scenarios (projected by the NiGEM model) are included

to re-estimate the Vector Autoregression model 5.10, following the multi-factor approach. The

stochastic simulations have been applied to the net interest income of the banks.

Panels G and H show the probability distributions of the change of net interest income over a

3 years horizon for each scenario. In the stress scenarios, the distributions shift to the left which means

that the banks’ income growth turns out lower than in the base scenario. Compared to the base

scenario, the growth of net interest income is EUR 1.5 billion lower in the Malaise scenario and EUR

0.9 billion lower in the Correction scenario, which is significant from an economic perspective.33 Note

33 The outcomes of the simulations based on the AR-model (5.9) and the VAR-model (5.10) are of comparable magnitude (average loss of EUR 0.8 (AR) vs. 1.5 billion (VAR) in the Malaise scenario and EUR 1.2 (AR) vs.

81

that in the deterministic approach, the adverse effect of the Global correction scenario is larger than

the Malaise scenario, which again illustrates that the stochastic and deterministic simulations my lead

to different outcomes. It is worth mentioning that the distribution of credit risk (e.g. for the base

scenario of domestic loan losses in Figure 5.4, panel A) is flatter than the distribution of interest

income (kurtosis -0.94 for LLP/CRED versus -0.48 for interest income, three years base scenario).

From this it follows that the extreme outcomes of changes in interest income in the stress scenarios are

closer to the tail outcomes of the base scenario than in the case of credit risk (interest income changes

at the 1% percentile (left tail) of the interest income distribution are just EUR 10 to 30 million worse

than the outcomes at the 1% percentile in the base scenario). The relatively fat-tailed distributions of

credit risk correspond to the nature of credit risk, which is usually driven by a limited number of large

defaults.

0

2

4

6

8

-25 -19 -13 -7 -1 5 11 17 23

Base scenario Global correction

Panel G. Global correction and interest incomeProbability distribution of change of net interest income, horizon 3 years (cum)

change of net interest income (EUR bn)

per

cent

age

0

2

4

6

8

-25 -19 -13 -7 -1 5 11 17 23

Base scenario Malaise scenario

Panel H. Malaise and interest incomeProbability distribution of change of net interest income, horizon 3 years (cum)

change of net interest income (EUR bn)

perc

enta

ge

5.6 Conclusions

Scenario analysis is an important tool for assessing the possible impact of low-probability events and

extreme shocks. Macro models help to structure the scenario analyses. To map macro scenarios to the

portfolios of the banking sector, we develop macro stress-testing models, which quantify the first

round effects of shocks to credit and interest rate risk. Compared to the base line, the worst scenario

for credit risk results in an average loss on the total loan portfolio of Dutch banks of around EUR 2

billion and for interest income of nearly EUR 3 billion, which represents just around 5% of banks’

own funds in total. By including credit risk and interest rate risk, two important sources of risk in the

0.8 billion (VAR) in the Correction scenario, while the 1% tail losses per scenario are almost similar in both specifications). This underlines the robustness of the model for different specifications, including changes in the variance-covariance matrix of errors ∑, which is influenced by the specifications in an AR vs. VAR model.

82

loan portfolios of banks are modelled. In Chapters 6 and 7, a macro stress-testing model is applied to

liquidity risk.

The contributions of our approach are the inclusion of loss given default, next to probability of

default (PD) and expected loss (by LLP), and the separate modelling of credit risk in domestic and

foreign portfolios. Hereby, cross-border risks are taken into account to some extent, which is usually a

missing dimension in macro stress-testing models. Another important contribution of this chapter is

the multi-factor approach in applying deterministic and stochastic simulations of the macro scenarios.

This approach takes into account simultaneous changes in the macroeconomic variables and their

interactions. Moreover, the stochastic simulations allow for the changing correlations between risk

factors which is typical for stress situations. To some extent this gives in to the objection that stress-

testing models are based on statistical relationships that are assumed to remain constant, which might

not be the case in stress. A remaining challenge is the modelling of second round effects. Our

approach, like most stress-testing models, is confined to quantifying the first round effects of shocks.

Estimating second round effects would require more complex models than the reduced form equations

that are standard practice in macro stress-testing, since to analyse feedback effects, the interlinkages

between the economy and the financial sector should be modelled. This is an important area for further

research.

83

Appendix 5.1 Scenarios OFS

(source: DNB, 2005)

Base scenario

The base scenario is based on the MORKMON estimates from the DNB Quarterly Bulletin of June

2005. In conformity with the expectations of most market participants, this foresees continuing high

oil prices, a gradual increase in international bond yields and a steady depreciation of the US dollar.

As a result, growth in the euro area will lag behind, but the economic recovery in the Netherlands will

gather momentum and become more broadly-based. Global balance of payment imbalances will be

reduced in a gradual and orderly manner.

Malaise scenario

This scenario centres on the loss of consumer and producer confidence in the euro area, either due to

continuing high unemployment or the stagnation of European integration. Weak producer confidence

and the high oil prices depress the propensity to investment, while low consumer confidence and high

unemployment produce a negative consumption shock and stagnation of house prices. The slump in

demand causes intra-European trade to stagnate, pushing the Netherlands into recession. In this

scenario European long-term interest rates fall sharply, by 150 basis points over three years, leading to

a substantial flattening of the yield curve. This trend is reinforced by hedging behaviour of

institutional investors. In the case of rising inflation, due to knock-on effects of the high oil price,

interest rates would fall less sharply. The most important financial stability risk in the ‘Malaise

scenario’ concerns the implications of falling interest rates and a flattening yield curve for financial

institutions, notably life insurance companies and pension funds.

Global correction scenario

This scenario revolves around a correction of the Global balance of payments imbalances by a series

of sharp shocks. Loss of confidence among investors and/or abrupt adjustments in the reserve

management of Asian central banks put capital flows to the US under pressure, triggering a sharp

adjustment of the US dollar and US interest rates. The assumption is that the trade-weighted dollar

depreciates in the first quarter of this scenario by 40% with US bond yields rising in three years by

250 basis points, sparking a sharp steepening of the yield curve. This has negative repercussions for

the US asset markets such as the stock market and the housing market. Though the global imbalances

are considerably reduced (the US current account deficit drops almost 3.5% of GDP within three

years), this scenario contains various financial stability risks. Owing to the traditional correlation

between US and European long-term interest rates, it is assumed that European bond yields also

increase and that the European yield curve steepens, so that corrections also occur in the European

stock markets and housing markets. The Dutch economy would be hard hit by negative wealth effects

84

and exports would decline. In this scenario the sharp rise in risk-free interest rates brings the search for

yield to an abrupt halt, causing a worldwide repricing of risk premiums. This process could possibly

be reinforced by the role played by hedge funds in less liquid market segments. A shift will occur

towards liquid and less risky instruments (money market paper, deposits etc.). In other words the

scenario assumes a flight to liquidity so that bonds are sold and interest rates rise. But it is also

conceivable that a ‘global correction’ triggers a flight to quality whereby investors flee to seemingly

safe (e.g. European) government bonds. In that case a rise of European bond yields is less likely.

85

Chapter 6

Liquidity Stress-Tester: A model for stress-testing banks’ liquidity risk

6.1 Introduction

The recent financial crisis has underscored the need to explicitly take into account liquidity risk in

stress-testing frameworks.34 The manifestation of liquidity risk can rapidly move the system into the

tail of the loss distribution through bank runs, the drying up of market liquidity or doubts of

counterparties about banks’ liquidity conditions. In these situations, liquidity can evaporate making a

bank subject to multiple possible equilibria with very different levels of liquidity supply (Banque de

France, 2008). Liquidity risk is not only a source of banks’ funding risk (the ability to raise cash to

fund the assets), but also has a strong link to market liquidity (the ability to convert assets into cash at

a given price). The originate-to-distribute model has made banks increasingly dependent on market

liquidity to secure funding by issuing securities on wholesale markets and by trading credits. As a

result, banks have become more vulnerable to macroeconomic and financial shocks that may engender

liquidity risk.

Various regulatory initiatives in response to the credit crisis have highlighted that banks’

stress-testing practices usually do not incorporate liquidity risk scenarios sufficiently (FSF, 2008).

Banks often underestimate the severity of market-wide stress, such as the disruption of several key

funding markets simultaneously (e.g. repo and securitisation markets). Moreover, banks do not

systematically consider second-order effects that can amplify losses. These can be caused by

idiosyncratic reputation effects and/or collective responses of market participants, leading to

disturbing (endogenous) effects on markets. Banks have insufficient incentives to insure themselves

against such risks. This is because holding liquidity buffers is costly and may create a competitive

disadvantage (FSA, 2007). Besides, liquidity stresses have a very low probability and market

participants could have the perception that central banks will intervene to provide liquidity in stressed

markets.

Macro stress-testing, i.e. testing the financial system as a whole, is an instrument of central

banks and supervisory authorities to assess the impact of market-wide scenarios and possible second

round effects. Such tests with regard to liquidity risk can enhance the insight in the systemic

dimensions of liquidity risk. These exercises can also contribute to market participants’ awareness of

systemic risks. However, liquidity risk is not included in most macro stress-testing models. A main

reason for this is that the multiple dimensions of liquidity risk make quantification difficult (IMF,

34 This chapter is a revised version of Van den End (2010a).

86

2008b). This could also explain the large variation in the extent to which supervisors prescribe limits

on liquidity risk and insurance that banks should hold (BCBS, 2008b).

This chapter presents a stress-testing model which focuses on both market and funding

liquidity risk of banks. Multiple dimensions of liquidity risk are combined into a quantitative measure.

Section 6.2 describes related models by reviewing the literature. Section 6.3 outlines the model

framework of Liquidity Stress-Tester and explains the model structure for the first and second round

effects of shocks to banks’ liquidity. It also provides a parameter sensitivity analysis Section 6.4

presents model simulations for Dutch banks as an illustration, including an anecdotal back test.

Section 6.5 concludes.

6.2 Literature

Our study relates to models of financial intermediation by banks in transmitting and amplifying

shocks. For instance, liquidity risk plays a role in the interaction and contagion between banks in the

interbank market. Upper (2006) presents a survey of interbank contagion models, concentrating on

interbank loans. This channel of contagion is operative when banks become insolvent due to defaults

by their (interbank) counterparties. Contagion may also take the form of deposit withdrawals due to

fears that banks will not be able to meet their liabilities because of losses incurred on their (interbank)

exposures. Upper sees scope for improvements in the specification of the scenarios leading to

contagion. He concludes that a fundamental shortcoming is the absence of behavioural foundations of

the interbank contagion models, which results in the assumption that banks do not react to shocks (i.e.

absence of optimising banks). Adrian and Shin (2008b) add to this that domino models do not take

sufficient account of how prices change. Related to interbank contagion studies is literature that

analyses payment and settlement systems as a potential source of liquidity shocks and contagion

between banks (see, for instance, Leinonen and Soramäki, 2005). Some studies in this field also pay

attention to behavioural reactions (e.g. Bech et al., 2008, Ledruth, 2007).

Recent work provides some more guidance on how micro foundations could be introduced

into financial sector models. In agent based simulation models, market dynamics are driven by

bounded rational, heterogeneous agents using rule of thumbs strategies (Hommes, 2006). A bounded

rational agent behaves under uncertainty and responds to feedbacks in interaction with other agents.

Applied to the financial system, the model of Goodhart et al. (2006) is based on both heterogeneous

banks and households (investors) and operates through endogenous feedback mechanisms, both

amongst banks, investors and between the real and financial sectors. Liquidity indirectly plays a role

through the credit supply of banks to other banks and consumers, while default is endogenous within

the system. A drawback of their model is the simplification of the economy to only banks and

consumers. Furthermore, the authors recognise the challenge of their approach to reflect reality.

87

Aspachs et al. (2006) have calibrated the Goodhart model to values of several banking systems by

using the probability of default of banks as a measure of financial fragility.

Another strand of models links the banking sector to asset markets, which differs from earlier

studies that view liquidity shortages as stemming from the bank’s liability side, due to depositor runs

(e.g. Allen and Gale, 2000) or withdrawals of interbank deposits (Freixas et al., 2000). Von Peter

(2004) relates banks and asset prices in a simple monetary macroeconomic model in which asset

prices affect the banking system indirectly through debtors’ defaults. Asset price movements that are

driven by market liquidity can also lead to endogenous changes in banks’ balance sheets through a

financial accelerator (Adrian and Shin, 2008a). Cifuentes et al. (2005) examine how defaults across

the interbank network are amplified by asset price effects. Herein, market liquidity drives the market

value of banks’ assets which in a downturn can induce sales of assets, depressing prices and inducing

further sales. Nier et al. (2008) apply the same mechanism to an interbank network in which contagion

is dependent on the connectivity, concentration and tiering in the banking sector. In this framework,

the default dynamics with liquidity effects are simulated, including second round defaults of banks.

These result from shocks to the assets of banks, rather than to the liabilities. The model of Diamond

and Rajan (2005) also focuses on the bank’s asset side and shows that a shrinking common pool of

liquidity exacerbates aggregate liquidity shortages. Boss et al. (2006) have developed a system in

which models for market and credit risk are brought together and connected to an interbank network

module. This is similar to the framework developed by Alessandri et al. (2009), which also takes into

account asset-side feedbacks induced by behavioural responses of heterogeneous banks. These two

models are used for stress-testing by the Oesterreichische Nationalbank and the Bank of England,

respectively. Off-balance contingencies are not covered in these models. Feedback effects arising from

market and funding liquidity risk are also (still) missing in most macro stress-testing models of central

banks. Such effects are featuring in models with margin-constrained traders, as in Brunnermeier and

Pedersen (2007). They model two ‘liquidity spirals’, one in which market illiquidity increases funding

constraints through higher margins, and one in which shocks to traders funding contributes to market

illiquidity due to reduced trading positions.

Our approach relates to the last strand of work, but while the study of Brunnermeier and

Pedersen (2009) is mainly conceptual in nature, our model is based on a more mechanical algorithm to

make it operational for simulations with real data. In this respect, the Liquidity Stress-Tester belongs

to the class of simulation models of central banks that are used to quantify the impact of shocks on the

stability of the financial system. The value added of our approach is the focus on the liquidity risk of

banks, taking into account the first and second round (feedback) effects of shocks, including price

effects on markets, induced by behavioural reactions of heterogeneous banks and idiosyncratic

reputation effects. The model centres on the liquidity position of banks and their related risk

management reactions. The contagion channels through which the banks are affected (e.g. the

interbank network, asset markets) are not explicitly modelled. Instead, contagion results from the

88

effects of banks’ reactions on prices and volumes in the markets where other banks are exposed to, as

described in the next section.

6.3 Model

6.3.1 Framework

In stylised form the Liquidity Stress-Tester model can be represented by Figure 6.1. Banks’ liquidity

positions are modelled in three stages: after the first round effects of a scenario, after the mitigating

actions of the banks, and after the second round effects. In each stage, the model generates

distributions of liquidity buffers by bank, including tail outcomes and probabilities of a liquidity

shortfall. The model is driven by Monte Carlo simulations of univariate shocks to market and funding

liquidity risk factors, which are combined into a multifactor scenario. For instance, a credit market

scenario can be assumed to include rising credit spreads, falling market prices of structured credit

securities (market liquidity) and reduced liquidity in the primary markets for debt issues (funding

liquidity). The model is flexible to choose any plausible set of shock events. This deterministic

approach of scenario building is based on economic judgement and historical experiences of

confluences of events that are likely to lead to a banking liquidity crisis. In the model, the scenario

horizon is set at 1 month but the model is flexible to extend it (as an example, Section 6.4.2 presents

outcomes at a horizon of 6 months).

A scenario is uniformly applied to individual banks by weighting the banks’ liquid asset and

liability items (i) that would be affected by the scenario with stress weights (wi). For instance, in case

of the credit market scenario, weights would be attached to banks’ tradable credit portfolios, collateral

values and wholesale funding liabilities. The weights (wi) stand for haircuts in the case of liquid assets

(reflecting reduced liquidity values or mark-to-market losses) and run-off rates in the case of liabilities

(reflecting the drying up of funding). The size of the weights wi differs per balance sheet item

according to the varying sensitivity of assets and liabilities to liquidity stress (see Section 6.3.2).

In the model, a scenario is assumed to unroll in two rounds. In the first round, the initial

effects of shocks to banks’ market and funding liquidity risks are modelled (stage 1 of the model,

represented by the first line of the flow chart in Figure 6.1). This is done by multiplying the liquid

asset and liability items that are affected in the first round of the scenario by the stress weights (wi).

The resulting loss of liquidity is then subtracted from a banks’ initial liquidity buffer. The outcome is

given by ‘liquidity buffer (1)’ in the figure and is in fact a distribution of buffer outcomes per bank,

following from the simulated market and funding liquidity risk events (i.e. the simulated stress

weights, wi).

89

Figure 6.1. Flow chart of Liquidity Stress-Tester

Scenario 1st round effects Liquidity buffer (1) STAGE 1

Threshold?

STAGE 2

Liquidity buffer (2) mitigate 1st round ef Reactions by bank

Loss of reputation

STAGE 3Collect. behaviour?

Liquidity buffer (3) 2nd round effects

The second stage of the model entails the mitigating actions of the banks in response to the shocks in

the first round of the scenario. Their responses are assumed to be triggered if the decline of the

liquidity buffer due to the initial shocks breaches a predefined threshold, which reflects a significant

impact (the threshold for reactions is derived in Section 6.3.4). The reacting banks take mitigating

measures to mobilise liquidity and restore their liquidity buffers (resulting in the improved ‘liquidity

buffer (2)’ at the end of the second line of the flow chart). The type of measures by which the banks

react (i.e. the markets in which they operate) are defined beforehand as part of the deterministic

scenario. The reactions of banks set in train the second round effects of the scenario (stage 3 in the

model). One part of the second round effect is the idiosyncratic risk of the reacting bank. It faces a

reputational risk, since it might be perceived to be in trouble by conducting measures to restore its

liquidity buffer (signalling effect). The other part of the second round effect is systemic risk. This

relates to collective reactions by banks that could lead to wider disturbing effects in the banking sector

or on financial markets. Both the idiosyncratic and systemic second round effects of a scenario

determine the final liquidity buffer (‘liquidity buffer (3)’ at the end of the third line of the flow chart).

Liquidity Stress-Tester takes into account that systemic risk turns out to be larger if i) more

banks react, since collective reactions are more disturbing, ii) reactions are more similar, taking into

account possible distortions by ‘crowded trades’ and iii) reacting banks are larger, since reactions by

sizable banks are more likely to cause market-wide instability. For instance, in the case of a credit

market scenario, if large banks would collectively react to the initial shocks by withdrawing interbank

credit lines and by fire sales of certain assets, dislocations in the unsecured interbank markets and

distressed market prices in particular market segments are likely. In the model, both the idiosyncratic

loss of reputation and the wider systemic effects have an impact on banks’ liquidity buffers through

90

additional haircuts on liquid assets and withdrawals of liquid liabilities (i.e. the second round effects

further increase the stress weights, wi of the affected balance sheet items).

The systemic second round effects embody contagion within the banking sector as well as

interactions between markets and banks. The contagion results from the effects of banks’ reactions on

the prices and volumes in the markets where the banking sector is exposed to (possible market stress

caused by other developments is included in the model as an exogenous variable). For instance, if

banks would react to restore their liquidity position by cutting credit lines to other banks, the banking

sector experiences reduced funding liquidity in the interbank market (this type of second round effect

is depicted in the upper left panel of Figure 6.2). In the model, the effect on interbank exposures does

not operate directly through mutual balance sheet linkages as in traditional interbank contagion

models, but indirectly through a reduced liquidity in the interbank market as a whole (reflected in a

stress weight (wi) applied to interbank liabilities). This is assumed to be an aggregate effect; the model

does not specify whether it relates to increased borrowing rates or reduced credit supply. The same

applies to contagion through interlinkages between markets and banks. For instance, if banks would

react to restore their liquidity position by fire sales of stock portfolios, which could be a reaction

defined in the scenario, the banking sector as a whole is affected by reduced mark-to-market values of

stocks. This shows up in a stress weight (wi) applied to the stock portfolios of banks (this type of

second round effect is depicted in the upper right panel of Figure 6.2). Possible interactions between

market and funding liquidity, as explored in IMF (2008b), are also taken into account in the model

framework. For example, funding pressures due to margin calls may lead to reduced liquidity

provision by banks to investors. This will strain the trading activity on financial markets and give rise

to falling market prices, affecting the banks with exposures to the pressed tradable securities. In the

model this is accounted for by applying a stress weights (wi) to the affected securities holdings of the

banks (see lower left panel of Figure 6.2). Contagion can also run from liquidity shocks on markets to

funding liquidity, as depicted in the lower right panel of Figure 6.2. This could be the case if banks are

forced to sell tradable securities in response to an initial shock, engendering falling market prices and

reduced collateral values. The latter will strain the funding possibilities of banks in the repo market. In

the model this is reflected in a stress weight (wi) applied to repo funding lines.

91

Figure 6.2. Systemic effects through contagion channels

Interbank contagion Contagion through asset markets

Reacting banks Total banking sector Reacting banks Total banking sectorinterbank interbank fire sales value stock

lending ↓ funding ↓ stocks ↓ portfolios ↓

From funding to market liquidity From market to funding liquidity

Reacting banks Total banking sector Reacting banks Total banking sectorliquidity margin calls value tradable fire sales MtM loss securedprovision ↓ securities ↓ securities ↓ collateral ↓ funding ↓

6.3.2 Data

Although Liquidity Stress-Tester is a top-down model, it is run with bank level data. We use the

liquidity positions (both liquid stocks or non-calendar items and cash flows or calendar items) of the

Dutch banks, that are available from De Nederlandsche Bank’s (DNB’s) (2003) liquidity report. It

contains end of month data, which are available since 2003. Data are provided by all Dutch banks (85

on average, including branches and subsidiaries of foreign banks) and cover liquid assets and

liabilities, scheduled payments and on and off-balance sheet items, with a detailed break-down per

balance sheet item. Appendix 2.1 in Chapter 2 provides an overview of the items in the report. Not all

items are reported by all banks, since most do not have exposures to all categories. The average

granularity reported per bank is around 7 items; the large banks report more items, owing to their more

diversified businesses. The top 5 banks, which cover around 85% of the Dutch sector, have an average

granularity of 54 items.

The baseline is a going concern situation, as reflected in unweighted liquid assets and

liabilities. This assumes that liabilities can be fully refinanced and that the liquidity value of assets is

100%, i.e. the weights (wi) are 0. The weights are taken from DNB’s liquidity report (DNB, 2003). In

the report, the actual liquidity of a bank must exceed the required liquidity, at both a one week and a 1

month horizon. By this, the report tends to focus not only on the very short term, but also on the more

structural liquidity position of banks. In the report, actual liquidity is defined as the stock of liquid

assets (weighted for haircuts) and the cash inflow (weighted for their liquidity value) during the test

92

period. Required liquidity is defined as the assumed calls on contingent liquidity lines, assumed

withdrawals of deposits, drying up of wholesale funding and liabilities due to derivatives. In this way,

the liquidity report comprises a combined stock and cash flow approach. The weights (wi) applied to

the liquid assets and liabilities in the DNB report represent a mix of a firm specific and market wide

scenario and are based on best practices and values of haircuts on liquid assets and withdrawal or run-

off rates of liabilities typically used by the industry and rating agencies.35 This makes them a useful

point of departure for our model. The parameterisation of the run-off rates, either based on best

practices or historical data, is a weakness in most liquidity stress-testing models of banks. This is

because data of stress situations are scarcely available and in times of stress the assumed elasticities

may behave differently. As a consequence, banks may underestimate the stability of their funding

base. By applying a stochastic approach, Liquidity Stress-Tester takes into account this uncertainty of

the model parameters.

6.3.3 First round effects

In Liquidity Stress-Tester the fixed weights of DNB’s liquidity report are assumed to be 0.1% tail

events (wi ≈ 3σ).36 The scenario impact of the first round effect on an item i is determined by simulated

weights (w_sim1,i). These are based on Monte Carlo simulations by taking random draws from a log-

normal distribution Log-N (0,1), scaled by (3

LCR wi ), so that wi sim1 ~ Log-N (µ,σ2). The use of a log-

normal distribution is motivated by the typical non-linear features of extreme liquidity stress events.

The log-normal distribution, which is skewed to the right, captures this feature. Its asymmetric shape

fits well on financial market data in particular in high volatility regimes. For that reason the log-

normality of asset returns plays an important role in theory of risk management and asset pricing

models. Besides, the log-normal distribution is bounded below by 0 which is also due for the

simulated weights in our model. As an upper bound, the weights are truncated at w_sim1,i ≤ 100 in the

simulations, since haircuts and withdrawal rates cannot exceed 100%. This procedure delivers a log-

normal distribution of weights which is bounded below by 0 and truncated at the top by 100. The

liquidity buffer in the baseline situation (normal market conditions), B0, is

∑=

−=nc

1i

b

i,calnon

b

0IB (6.1)

b being the individual bank and Inon-cal, i the amount of available assets of non-calendar items (the stock

items of liquid assets 1 .. nc). By this, the buffer consists of deposits at the central bank, securities that

35 In the model, the weights of DNB’s liquidity report that apply to a horizon of 1 month are used. The liquidity model of Standard & Poor’s (2007) is based on a standard set of assumptions, i.e. a spectrum of asset haircuts and liability run-off rates, that were established after a review of bank balance sheets, industry, S&P data and dialogue with risk managers. 36 In the model simulations this assumption could be changed according to other insights.

93

can be turned into cash at short notice, ECB eligible collateral, interbank assets available on demand

and receivables from other professional money market players available on demand. B0 provides

counterbalancing capacity to liquidity scenarios in which liquidity values of the stock of assets could

decline and a drain of liquidity could occur due to decreasing net outflows of liquidity. This means

that the scenario effects could be felt through both deteriorating liquid stocks and flows. The first

round effect (E1) of the scenario is determined by,

i,i

bi

b sim_wIE 11 ∑= (6.2)

I i being the amount of all liquid (non-calendar and calendar) asset and liability items. The liquidity

buffer after the first round impact of the scenario, B1, is,

b

1

b

0

b

1EBB −= (6.3)

6.3.4 Banks’ response to scenario (mitigating actions)

Banks that are affected seriously by the first round effects of the scenario are assumed to restore their

liquidity buffer to the initial level (B0). Banks may take actions to safeguard their stability and/or to

meet liquidity risk criteria of supervisors and rating agencies. In the model, the trigger for a bank’s

reaction is a decline of its original liquidity buffer that exceeds a threshold θ. By this, reactions are

triggered by a significantly large impact of the first round of the scenario (as reflected in the simulated

buffer B1). The trigger q (0, 1) is based on a probability condition (probit),

with q =

>

otherwise0B

Eif1

b

o

b

1 θ

The latent variable θ can be seen as a ‘rule measure’ which banks follow due to self imposed liquidity

risk controls or regulatory requirements. The rule is operationalised by assuming that large value

change of balance sheet items reflect banks’ intentional responses to a buffer decline. The rule variable

θ can then be derived from the average correlation between value changes of balance sheet items and

declines of liquidity buffers one month lagged:

)I

II,

B

BB(Correl

b

0t,i

b

0t,i

b

1t,i

b

1t

b

1t

b

0t

=

==

−=

−==−− , conditioned by

0B

BBb

1t

b

1t

b

0t <−

−=

−==

The lag controls for the influence of possible endogeneity in the relationship between the buffers and

the balance sheet items. In an empirical application for the Dutch banking sector the correlation

94

coefficient has been computed with monthly data of 85 Dutch banks, over the 2003 - 2007 period.37

Table 6.1 shows that only substantial declines of the liquidity buffer (from 40%) lead to significant

changes of balance sheet items in the next period. This only indicates whether a bank would react and

not the direction of the response.38 Smaller declines are probably (passively) absorbed by the buffers

of the banks. Based on this outcome, a rule variable θ equal to 40% is used as a uniform trigger for

each banks’ reaction.

Table 6.1.Correlation between relative change of buffer (B), lagged relative change of balance sheet items (I)Spearman correlation coefficient

Buffer change (%) obs Correl

0 - 10 25453 -0,0001510 - 20 8303 0.0028520 - 30 3437 0.0021830 - 40 1892 -0.0016340 - 50 1134 -0.05615 *50 - 60 767 0.0175660 - 70 681 0.07847 **= 70 617 0.0291

***, **, * significant at 1%, 5%, 10% confidence levelBased on 85 Dutch banks and around 7 items per bank on averageSource: own calculations based on DNB liquidity report.

The type of instruments (items i, amounting I) which banks use to react is specified beforehand in the

design of the second round of the scenario, based on judgement of the set of instruments that will most

likely be used in a particular scenario. For instance, banks can use securities eligible for repo with

central banks, draw on liquidity lines from other banks, sell liquid securities, such as government

bonds or asset backed securities, or rely on unsecured funding in the (money) markets. The choice of

instruments may be determined by internal rules or contingency funding plans that sometimes

prescribe different sets of measures for various scenarios. Regulators promote the linkage of stress-

tests to contingency funding plans (FSF, 2008).

In the model, the extent to which banks use particular instruments to restore the liquidity

buffer is assumed to be (mechanically) determined by the relative importance of items on the balance

37 The assumption that the change of I i reflects balance sheet adjustments is quite strong as changes of I i could also be caused by exogenous price movements. However, very large changes of I i are more likely to be caused by portfolio adjustments since extreme price effects within one month can be considered quite rare. Moreover, banks do not value all the balance sheet items on a mark-to-market basis. Bt=0 ≠ B0 and Bt=1 ≠ B1, as the former are the actual monthly buffers, whereas the latter are the buffer in each stage of the model simulations. 38 Hence the sign of the correlation coefficients cannot be interpreted straightforward since the value of items can either increase or decrease in reaction to declines of the buffer, depending on the type of crisis, the nature of the balance sheet item and the response of the individual bank. For instance, to generate liquidity a bank can either sell tradable securities (value of asset item decreases) or issue additional securities (value of liability item increases), or substitute some assets of liabilities with other items.

95

sheet (

∑i

bi

bi

I

I ), reflecting a bank’s specialisation and presence in certain markets.39 Since in liquidity

crises, time is usually very short and banks often do not have the opportunity to change their strategy

(e.g. by diversifying funding or spreading risk). The size of the transactions that a bank conducts with

instrument i is expressed by b

iRI ,

)I

I()BB(RI

i

bi

bibbb

i∑

−= 10 (6.4)

Since B1 ≤ B0, by definition b

iRI is positive. This does not imply anything about the direction of the

transaction (e.g. buying or selling) but it indicates the (absolute) size of the transaction that is needed

to generate liquidity ( b

iRI is a size factor). Hence, the liquidity buffer after the mitigating actions (B2)

of a bank is equal to,

)sim_w(RIBB i,i

bi

bb112 100−+= ∑ (6.5)

with B2 > B1, but B2 < B0, since the buffer cannot be fully restored due to the market disturbances in

the first round of the scenario (as reflected in w_sim1,i). In an extreme stress situation, financial

markets may be gridlocked completely due to the drying up of liquidity. Such an extreme case is

represented by w_sim1,i = 100, implying that banks have no possibility to enter a particular market

segment to raise additional liquidity. In the case of the repo markets this could mean that certain

collateral of banks may be useless.

6.3.5 Second round effects

The behavioural reactions of the banks can have wider disturbing (endogenous) effects on markets,

feeding back on the banks. This will be manifested in additional haircuts on liquid assets and

withdrawals of liquid liabilities in the market segments where banks react, as reflected in w_sim2,i

(with w_sim1,i ≤ w_sim2,i ≤ 100). The feedback effects are larger if more banks react (∑b

q) and if

reactions are similar, which is expressed by the sum of reactions by a particular instrument (∑b

b

iRI ).

This summation is divided by the total amount of reactions )RI(i b

b

i∑∑ to get the ratio that indicates

the similarity of reactions (

∑∑

∑

i b

bi

b

bi

RI

RI ). In the case of deep and liquid markets (e.g. the government bond

market) where discretionary transactions will have little effects, w_sim2,i is smaller than in the case of

39 The model does not specify the conditions (e.g. credit spreads) at which funding is attracted.

96

illiquid market segments. Such differences will already be reflected in w_sim1,i from which w_sim2,i is

derived,

∑=

∑∑∑

∑+

b

b

s)RI

RI(

i,i, q

q

sim_wsim_w

i b

b

i

b

b

i

1

12 (6.6)

Since b

iRI indicates the size of the transaction that is conducted to generate liquidity, higher values of

b

iRI imply a higher liquidity demand, which will adversely affect the availability of liquidity in market

segments in which the banks operate. By including RIi in equation 6.6, large transactions have more

impact on markets than small transactions. This implicitly means that reactions by large banks induce

stronger second round effects than reactions by small banks.40 Variable s is a state variable which

represents the exogenous market conditions. Equation 6.6 has parallels with the asset price function

used by Alessandri et al., 2009 and Nier el al., 2008. In their models, the price of banking assets is a

decreasing function of the amount of assets sold by banks, while the price also depends on market

liquidity.

More in particular, the state variable s represents an indicator of exogenous market stress. The

ranges of this variable are derived from standardised distributions of risk aversion indicators. For this

the implied stock price volatility (VIX index) and the US corporate bond spreads (Baa) were used as

proxies. Figures 6.3a and 6.3b show standardised frequency distributions of these series. To determine

a range of s for use in the model, we assume that normal market conditions are reflected by -1 ≤ s ≤ 1

(which according to a standardised distribution of risk indicators, represents 2/3 of market conditions)

and severe market stress by s = 3 (i.e. 0.5% of adverse market situations). s could be even higher, as

panels A and B in Figure 6.3 indicate. For the purpose of measuring liquidity stress in the model, the

restriction s ≥ 1 applies. The risk aversion indicators could be used to conduct periodic runs with

Liquidity Stress-Tester in which changing market conditions play a role.

Figure 6.3. Frequency distribution of risk aversion indicators

0%

10%

20%

30%

40%

50%

-3 -2 -1 0 1 2 3 More

Panel B. Frequency distribution of implied volatilityNormalised value of S&P500 stock price volatility (VIX index), daily data period 1986-2007

Source of VIX: Chicago Board Option Exchange

0%

10%

20%

30%

40%

50%

-3 -2 -1 0 1 2 3 More

Panel A. Frequency distribution of credit spreadsNormalised value of Moodys Baa average credit spreads on corporate bonds, daily data period 1986-2007

Source credit spreads: US Federal Reserve 40 By running Liquidity Stress-Tester with a limited sample of banks (in this chapter the Dutch banks) it is implicitly assumed that the reactions of this sample are representative for the (global) banking system as a whole.

97

In the model, the market conditions contribute to the severity of the second round effects: the higher is

s, the stronger are the effects of the number and the similarity of banks’ reactions. In that respect, the

fall-out of the market stress (reflected in s) differs for each market segment. We do not model

feedback effects running from banks’ reactions to s, assuming that the market stress represents an

exogenous shock that drives the reactions by banks. Endogenising variable s, by making it dependent

on banks’ reactions, would complicate the model by introducing a circular reference. Conducting

periodic model runs with a time variant value of s will solve this to some extent, since market wide

risk aversion indicators will reflect the influence of banks’ reactions on market conditions.

Panels A and B in Figure 6.4 illustrate the relationship between w_sim2,i and w_sim1,i and its

dependence on the number of reacting banks (∑b

q ), the similarity of reactions (

∑∑

∑

i b

bi

b

bi

RI

RI ) and the level

of market stress (s). It is assumed that the similarity of reactions has a stronger effect on markets than

the number of reacting banks (see the exponential relationship in panel B). The intuition behind is that

the similarity of reactions points to crowded trades in markets which cause a drying up of market

liquidity.

Banks that react in order to restore their liquidity buffer face a reputation risk in the financial markets.

While applying sensible measures ought to strengthen a banks’ financial position and comfort

counterparties, the adverse signalling effect of the transactions could reverberate on the conditions that

banks face in the markets. This could translate in even more (idiosyncratic) haircuts on liquid assets

and withdrawals of liquid liabilities, as reflected in w_sim*2,i (with w_sim2,i ≤ w_sim*2,i ≤ 100). The

reputation effect will depend on the market conditions (s) driving the second round effects, since

particularly in stressed circumstances the signalling effect of reactions will adversely feedback on a

98

bank (the stigma associated with accessing central bank standing facilities in the recent crisis is

illustrative).41 In functional form, the reputation risk is expressed by,

ssim_wsim_w i,*

i, 22 = (6.7)

Next, the additional impact of the (systemic and idiosyncratic) second round effects on banks is

determined by E2,

))sim_wsim_w()RII((E i,i

i,bi

bi

b122 ∑ −+= (6.8)

with w_sim2,i being replaced by w_sim*2,i in case a reacting bank also faces reputation risk. The

liquidity buffer after the second round effects (B3) is,

b

2

b

2

b

3EBB −= (6.9)

6.3.6 Impact different scenario rounds

The stylised balance sheet in Appendix 6.1 shows how the model works in a simplified one bank

situation. A hypothetical scenario is assumed to affect all liquid assets and liabilities of the bank

through fixed in stead of simulated stressed weights. Furthermore, it is assumed that the first round

effect of the scenario leads to a decline of the initial liquidity buffer that exceeds the threshold θ and

that the bank reacts with all instruments available at its disposal (i.e. asset items 1 and 2 and liability

items 1 and 2 on the stylised balance sheet). This example shows that the mitigating actions of the

bank improves its liquidity buffer (to B2), although it remains below the initial level (B0). The second

round effects reduce the buffer further (to B3), below the level after the first round shock (B1).

In the stochastic mode of the model, each round of a scenario has its typical effect on the

distribution of buffer outcomes. Simulations with real bank data show that the first round effect leads

to a shift of the distribution to the left (B1), while the mitigating actions shift the distribution (B2) back

towards B0 and cause a peakening of the shape (panel B in Figure 6.5)42. If a bank does not react

because θ<0

1

B

E , the distributions of B1 and B2 coincide. This is the case with bank AH in panel A of

Figure 6.5. The second round effect shifts the distribution (B3) to the left again and causes a flattening

of the distribution. The average of this final distribution (3B ) is substantially smaller than 1B , which

indicates that the second round effects outweigh the initial shock. Such an outcome is conceptually

explained by Nikolaou (2009). For a bank which does not face a reputation risk the second round

41 Equation 6.7 has been calibrated on the actual outcomes of the individual banks and on the share of the reputational effect in the total second round effect (see Section 6.4). If s = 1 (the downside restriction for s), than the mitigating reaction of a bank will not be counteracted by adverse reputational effects and will improve a banks’ liquidity position by definition. 42 The parameters of these simulations are equal to those applied in Section 6.4.

99

effect is limited. This shows up in a more limited shift of the distribution to the left compared to the

reacting bank (see the most left distributions in panels A and B in Figure 6.5).

Figure 6.5. Distribution of buffers after scenario rounds

Panel A. Distribution of buffers after each scenario round Panel B. Distribution of buffers after each scenario roundFor bank AH as illustration, buffers normalised by B0 (θ=0.4, s =1.5) For bank AU as illustration, buffers normalised by B0 (θ=0.4, s =1.5)

0

20

40

60

80

100

120

140

-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0

Buffer after 1st roundBuffer after reactionsBuffer after 2nd round

Fre

quen

cy

0

10

20

30

40

50

60

70

0.84 0.86 0.88 0.90 0.92 0.94 0.96 0.98 1.00

B uffer after 1st roundB uffer after reac tionsB uffer after 2nd round

Fre

que

ncy

0

100

200

300

400

500

0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1

LogN (0,1) LogN(0,0.5) LogN(0,0.25)LogN(0,1.25) LogN(0,1.5) LogN(0,2)

Panel C. Buffer after 1st scenario round, LogN distributionDistribution with different scale parameters (sigma), buffer of representative bank, normalised by B0 (θ=0.4, s=1.5)

0

50

100

150

0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1

LogN (0,1) LogN(0,0.5) LogN(0,0.25)LogN(0,1.25) LogN(0,1.5) LogN(0,2)

Panel D. Buffer after 2nd scenario round, LogN distributionDistribution with different scale parameters (sigma), buffer of representative bank, normalised by B0 (θ=0.4, s=1.5)

0

25

50

75

100

125

0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1

LogN (0,1) Weibull(1,1)

Pareto(1,1.5) Gamma(1,5)

Panel F. Buffer after 2nd scenario round, other distributionsBuffer of representative bank, normalised by B0 (θ=0.4, s=1.5), with LogN, Weibull, Pareto and Gamma distributions

0

100

200

300

0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1

LogN (0,1) Weibull(1,1)

Pareto(1,1.5) Gamma(1,5)

Panel E. Buffer after 1st scenario round, other distributionsBuffer of representative bank, normalised by B0 (θ=0.4, s=1.5), with LogN, Weibull, Pareto and Gamma distributions

100

6.3.7 Influence of alternative distributional assumptions

As explained in Section 6.3.3 the simulated weights are based on a log-normal distribution Log-N

(0,1). The choice of a probability distribution function for non-negative random variables is motivated

by the need to produce weights that are bounded below by 0. This lower bound explains the location

parameter µ =0, which reflects normal market conditions (no stress). The scale parameter σ =1 is used

to scale the weights by (3

LCR wi ). To test the sensitivity of the model for alternative distributional

assumptions, the scale parameter (σ) is varied between 0.25 and 2. Panels C and D in Figure 6.5 show

the resulting distributions of the liquidity buffer of a representative bank after the first and second

rounds of a hypothetical scenario. It appears that the simulation outcomes are quite robust to different

values of the scale parameter, in particular with regard to the first round effects presented in panel C.

The scale parameter has more influence on the second round effects as presented in panel D. A higher

value of σ leads to a flattening of the buffer distribution, which becomes very pronounced at σ =2.

Next to the log-normal distribution, other distributions as well exhibit the features that are

desirable for our model, such as being skewed to the right. Hence, as another sensitivity test, different

distributional forms are used to generate buffer outcomes of a hypothetical scenario. Panels E and F in

Figure 6.5 show the liquidity buffers of a representative bank after the first and second rounds of a

hypothetical scenario, based on the log-normal, Gamma, Weibull and Pareto distributions. It appears

that the first round effects presented in panel E are quite robust to different distributional forms (only

the Gamma distribution generates outcomes that are located more to the left). The distributional form

has more substantial influence on the second round effects as presented in panel F (of course the

location and shape of the distributions depend on the choice of the moments for each distribution). The

simulation outcomes based on the log-normal and Weibull distributions are quite similar, but the use

of the Gamma and Pareto distributions changes the outcomes significantly. This highlights the

importance of the choice for a distributional form, which for our model is motivated in Section 6.3.3.

In extreme value theory (EVT) the Gumbel, Frechet and Weibull distributions are used (Poon et al.,

2004). This class of distributions provides for a non-degenerate limit as n → ∞, which is the desired

feature for estimates of tail values beyond a certain cut-off point. EVT focuses on the tail of the

distribution and only uses data from the tail area to model that part of the distribution. Thereby it

differs from our approach, as the Liquidity Stress-Tester model simulates the full range of buffer

outcomes.

6.3.8 Parameter sensitivity

Based on the same stylised balance sheet in Appendix 6.1, this section exposes the sensitivity of the

outcomes for changing the model parameters. In the base line situation, the level of market stress (s) is

101

set at 1.5, the number of reacting banks (∑b

q ) at 2, the similarity of reactions (

∑∑

∑

i b

bi

b

bi

RI

RI ) at 0.05 and

the scenario horizon at 1 month. Table 6.2 shows the impact of changing each parameter in isolation

on the banks’ liquidity buffer, in terms of deviations of the final buffer (B3) from the initial buffer (B0).

At first sight the model outcomes look relatively sensitive to changes of s (the buffer declines by

nearly 2/3 if s = 3) and less to changes in the number of reacting banks and the similarity of reactions

(the sensitivity analysis affirms that the latter has a stronger effect on markets than the number of

reacting banks). As explained in Section 6.3.5, s reinforces the effects of the number of reacting banks

and the similarity of reactions and these factors can hardly be assessed in isolation. Following from

equation 6.7, the impact of reputational risk (due when banks respond to a scenario by mitigating

actions) also depends on the level of market stress. Table 6.2 shows that reputational risk could

severely impact on banks in stressed markets. The model outcomes are also quite sensitive to

lengthening the scenario horizon; the final buffer declines by ¼ if the horizon is lengthened from 1 to

12 months (which includes the total run-off schedule of liability 1, which is a calendar item).

102

6.4 Results

This section describes model outcomes by simulating a hypothetical scenario (a ‘classical’ banking

crisis), and an historical scenario (the recent credit market crisis). These scenarios are run with July

2007 data of all banks in the Netherlands (including subsidiaries of foreign banks).43 The model

outcomes are based on 500 Monte Carlo simulations. In first instance we assume θ = 0.4 (the critical

threshold determined in Section 6.3.4), s = 1.5 (the middle of the range determined in Section 6.3.544)

and a horizon of one month (typically used in banks’ liquidity stress-tests). These values are used in

the simulations, but can be adjusted to other circumstances (as illustrated in Section 6.4.3).

Experimenting with the parameter values enhances the insight in the sensitivity of the model outcomes

for banks’ reactions, the level of market stress and the length of the scenario horizon.

6.4.1 Banking crisis scenario

The first round of the hypothetical banking crisis scenario seizes at the liability side of banks’ balance

sheets. It assumes a public crisis of confidence affecting the banking sector, which could result from

massive misselling of a financial product in the retail market. This scenario leads to a withdrawal of

non-bank deposits and other funding by professional money market players, other institutional

investors and corporates and by withdrawals of savings deposits by households. These first round

effects are simulated by stressing the weights of the affected deposits and funding sources (through

w_sim1,i). These weights determine the first round effect (E1) according to equation 6.2 and the

liquidity buffer (B1) according to equation 6.3. Table 6.3 shows the average outcomes for all banks.

On average, the first round effect erases 8% of the initial liquidity buffer. Some small banks would be

faced with a negative liquidity buffer after the first round of the scenario.

Table 6.3 shows that for 30 banks the decline of the liquidity buffer exceeds the threshold θ =

0.4 which triggers them to restore their liquidity buffer to the initial level (B0).45 The reactions mitigate

the first round effect of the scenario on the sector as a whole to around 7% on average (B2 being 0.5%

smaller than B0). Panel A.3 in Appendix 6.2 indicates that smaller banks tend to react relatively more

than large banks, which indicates that an outflow of deposits would foremost bring small banks in a

critical liquidity position.

In the second round of the scenario, it is assumed that banks react to the funding pressures by

drawing upon credit lines in the unsecured interbank market. These mitigating actions can be an

important source of feedback effects among banks. They are interdependent via interbank liquidity

43 In July 2007 the number of banks included in the data was 82 (on average, since 2003, 85 banks have been included in the dataset). 44 Note that at mid December 2007 during a height of the credit crisis, s was around 1 based on corporate bond spreads and around 0.5 based on implied stock price volatility. 45 The table reports the averages of the simulated buffers, whereas the reactions are triggered by extreme downward changes in the simulated sample of buffers.

103

promises and widespread use of these lines will lead to contagion of liquidity risk. The feedback

effects (w_sim2,i) are simulated by stressing the weights of the unsecured interbank assets and

liabilities. Next to these systemic second round effects, the banks which react by drawing upon

liquidity promises of counterparties face a reputation risk since their actions could be perceived as a

sign of weakness. In the model simulations this translates into additional (idiosyncratic) stress on the

weights (w_sim*2,i) according to equation 6.7. Both the reputational risk and the systemic (second

round) effects on the markets have an impact on the liquidity buffers of the banks (E2) according to

equation 6.8 and on the final liquidity buffer (B3) according to equation 6.9. Table 6.3 shows that due

to the second round effects the banks additionally loose 6% of their initial liquidity buffers on average

(including the effects of mitigation actions). Table 6.3 also shows the 5% and 1% tail outcomes of the

final liquidity buffer and the probability of a liquidity shortfall (i.e. B3 < 0). Insight in the extreme tail

outcomes is particularly relevant for financial stability analysis which assesses the resilience of the

system to extreme, but plausible shocks. In the 5% (1%) tail the liquidity buffer declines by 26%

(32%) on average. Out of the total sample, 25 banks have a probability larger than 0% to end up with a

liquidity shortage. These are mostly small banks which explains that the (by the initial liquidity buffer)

weighted average probability of a liquidity shortfall is limited to 0.5%. The latter is an indicator of the

liquidity risk of the financial system as a whole. Panel A.5 in Appendix 6.2 indicates a significant

negative correlation between the shortfall probability and size of banks, which affirms that small

banks are most vulnerable to a ‘classical’ banking crisis scenario.

6.4.2 Credit crisis scenario

The first round of the credit crisis scenario seizes at the asset side of banks’ balance sheets. It is

designed by assuming declining values of banks’ tradable credit portfolios, due to uncertainties about

the asset valuations which cause a drying up of market liquidity. The falling collateral values lead to

higher margin requirements on banks’ derivative positions. These first round effects are simulated by

stressing the weights of the credit portfolios and margin requirements (through w_sim1,i). Table 6.3

shows the average outcome for all banks. The first round effect erases 13% of the initial liquidity

buffer, with a maximum of 92% for the bank that is most severely affected. Although most banks

would be affected by the scenario (i.e. b

0

b

1BB < ), the liquidity buffers of the affected banks remain in

surplus in all cases. The banks that are not affected at this stage of the scenario are mostly small

branches of foreign banks. They can count on liquidity support from the head office and probably

therefore do not hold eligible collateral. A break-down of the sample by bank size and funding

structure indicates that banks with a more diversified funding profile are relatively more vulnerable to

the first round of the scenario (see panel A.8 in Appendix 6.3; panel A.7 indicates that there is no

significant correlation with bank size). Although a more diversified funding profile in general

improves banks’ resilience to liquidity shocks, the fact that the recent crisis has been most felt in

104

international financial markets has raised the vulnerability of banks that rely on wholesale funding,

next to retail deposits. This underscores that liquidity risk management should identify and measure

the full range of liquidity risks which banks face.

Table 6.3 shows that in case of 33 banks, the decline of the liquidity buffer exceeds the

threshold θ = 0.4 which triggers them to restore their liquidity buffer to the initial level (B0). The

reactions mitigate the first round effect of the scenario on the sector as a whole to around 3% on

average (B2 being 3% smaller than B0). Panels A.9 and A.10 in Appendix 6.3 indicate that larger banks

with a more diversified funding structure tend to react relatively more than smaller banks, which

relates to the stronger first round impact on the former group. According to the model (equation 6.6),

the responses of the large banks potentially have a relatively strong impact on the markets. If the

threshold θ is doubled to 0.8 only 13 banks respond to the first round impact. Table 6.4 shows that this

limits the second round effects of the scenario, indicating the models’ sensitivity to behavioural

reactions. In particular, the tail outcomes of the buffers are more favourable if fewer banks react.

The second round of the scenario designed by assuming that the market illiquidity spills over

into strained funding liquidity of the banks. Like in the recent credit crisis, we assume that the

difficulties to roll-over asset backed commercial paper (ABCP) imply an increased probability that off

balance liquidity facilities are drawn. This looming liquidity need induces banks to hoard liquidity.

Moreover, higher perceived counterparty risks induce banks to withdraw their promised credit lines.

This contributes to dislocations in the unsecured interbank market. The increased counterparty risk

among banks worsens their access to funding in the bond and commercial paper markets. Moreover,

collective actions of banks (e.g. fire sales of assets) in response to the first round effect of the scenario

could further disrupt credit and stock markets and raise margin calls. These second round effects

(w_sim2,i) are simulated by further stressing the weights of the credit portfolios and margin

requirements (on top of the first round effects) and by stressing the weights of the equity portfolios,

unsecured interbank assets and liabilities, capital market liabilities and off balance liquidity

commitments. The reputation risk of the reacting banks translates into additional (idiosyncratic) stress

on the weights (w_sim*2,i) according to equation 6.7. Table 6.3 shows that the second round effects of

the scenario have a larger impact than the first round effects; the banks additionally loose 26% of their

initial liquidity buffers on average (including the effects of mitigation actions). A breakdown of the

total second round effect indicates that more than half of the second round effects on the banks which

react is caused by the idiosyncratic reputational effects. Several banks loose over 100% of their initial

liquidity buffer which means that they become illiquid. Table 6.3 also shows the 5% and 1% tail

outcomes of the final liquidity buffer for each bank and the probability of a liquidity shortfall (i.e. B3 <

0). In the 5% (1%) tail the liquidity buffer declines by 68% (83%) on average. Out of the total sample,

33 banks have a probability larger than 0% to end up with a liquidity shortage. Panels A.11 and A.12

in Appendix 6.3 indicate no significant correlation between the shortfall probability and size or

funding diversification of banks, indicative of the systemic dimension of the second round effects, that

105

affect all types of banks. This underscores that policy initiatives to enhance banks’ liquidity buffers

can contribute to prevent financial stability risks.

6.4.3 Impact scenario length and market conditions

The recent liquidity crisis has been more prolonged than most banks assume in their liquidity stress-

tests (FSF, 2008). These are typically based on a one to two months horizon. The same applies to

liquidity frameworks of supervisors, like DNB’s liquidity report. Our model allows for lengthening the

stress horizon, by including the recognised cash inflows and outflows that fall due after one month as

well in the simulations. The weights of assets and liabilities should also be changed according to the

prolonged horizon, but this turned out to be impossible as information on appropriate weights for

longer horizons is lacking. This implies that the simulation outcomes probably underestimate the full

impact of a prolonged horizon. To illustrate the sensitivity of the liquidity buffers for prolonged

liquidity stress, we ran the credit crisis scenario at a 6-months horizon. Table 6.4 shows that

lengthening the stress period has a substantial impact on the scenario outcomes, partly because

liabilities falling due after one month exceeds cash inflows. At a 6-months horizon, the final average

buffer turns out to be more than 100% lower compared to a 1-month horizon and the 1% tail outcome

almost 150% lower. The latter indicates that a prolonged stress horizon has a relatively large impact

on the extreme (tail) outcomes.

To illustrate the sensitivity of the model outcomes to changing market conditions, the credit

crisis scenario has also been run with parameter value s = 2.0 in stead of s = 1.5 (s = 2.0 represents

2.5% of adverse market situations according a standardised distribution of risk indicators). Table 6.4

shows that such an increase of market wide stress has a comparable impact as lengthening the scenario

horizon. The relatively high probability of a liquidity shortfall indicates that the outcomes are quite

106

sensitive to changing market conditions; raising the level of s has a relatively large impact on the

extreme (tail) outcomes. This is conforming the intuition that extreme market conditions can affect the

liquidity risk profile of banks.

6.4.4 Back-test

As back-test, Figure 6.6 compares the outcomes of the credit crisis scenario to the actual change of the

average liquidity buffer of the Dutch banks since July 2007, when the crisis began to unfold. The

actual outcomes are quite close to the first round effects of the scenario (excluding mitigating actions),

but are substantially smaller than the buffers modelled after the second round. This could indicate that

the assumptions in Liquidity Stress-Tester are inappropriate, for instance the assumptions that the

weights in DNB’s liquidity report resemble 0.1% tail events or that the threshold θ for mitigating

reactions is 0.4. Another explanation could be that some functional relationships in the model fail to

reflect reality, for instance in case of the second round effects.

It could also be the case that the designed scenario is an imperfect replication of the recent

crisis. This is amongst others characterised by a re-intermediation of assets by banks which are not

able to fund those in the markets. Returning assets could be classified by banks as liquid items on their

balance sheets, which may distort the actual liquidity position of banks if market liquidity for such

assets has dried up. In case of the Dutch banks, this has not been a relevant factor since the off balance

items are being consolidated in the balance sheet and hence recur in the DNB liquidity report. The

difference between the actual and the model outcomes could also indicate that the extent of the market

stress is not fully reflected in banks’ balance sheets due to valuation issues.

However, the most likely explanation of the differences between the simulation outcomes and

actual developments is the expanded liquidity operations of central banks in the money market, which

have enabled banks to liquefy eligible collateral (against certain haircuts) for which the market had

107

seized up. By doing so, central banks addressed a market failure, by breaking the loop between market

and funding liquidity risk and preventing further market distress (Nikolaou, 2009). In terms of our

model, this implies that the value of certain collateral does not fully reflect the second round effects of

the market turmoil (which have come to the fore in reduced liquidity and fallen mark-to-market

values, in particular for structured credit securities which, in some cases, is eligible collateral for

central bank borrowing). The simulation outcomes on the other hand, are dominated by the adverse

second round effects on the liquidity buffers (in the scenario, the central bank facilities are only

included implicitly and partially, i.e. for the banks which react through pledging collateral at the

central bank).

In the next chapter, the Liquidity Stress-Tester model is extended with a reaction function of

the central bank. With that extended model, the mitigating effects of additional liquidity supply and

asset purchases by the central bank on the second round effects of a scenario are simulated.

-70%

-60%

-50%

-40%

-30%

-20%

-10%

0%

1month 6 months

Actual outcome Scenario, 1st round Scenario, 2nd round

Figure 6.6. Back-testing the scenario outcomesChange of liquidity buffer since July 2007 (monthly data, average Dutch banks). Model parameters: θ=0.4; s =1.5

6.5 Conclusions

Liquidity Stress-Tester is an instrument to simulate the impact on banks of shocks to market and

funding liquidity. It takes into account the important drivers of liquidity stress, i.e. on and off balance

sheet contingencies, feedback effects induced by collective reactions of heterogeneous banks and

idiosyncratic reputation effects. Contagion results from the effects of banks’ reactions on prices and

volumes in the markets where the banks are exposed to. The model contributes to understand the

influence on liquidity risk of collective reactions by banks, the level of market stress and the length of

the scenario horizon. These factors have been main drivers of the recent financial crisis. Liquidity

Stress-Tester could be used by central banks to stress-test the liquidity risk at the level of the financial

108

system. In this chapter the model has been applied to Dutch banks, but it could also be applied to other

countries’ banking systems, provided that data for liquid assets and liabilities are available on an

individual bank level. The parameters of the model (such as the weights and the threshold for

reactions) can be tailored to a local banking sector, according to the insights of the supervisor or

central bank.

The model outcomes lend support to policy initiatives to enhance the liquidity buffers and

liquidity risk management at banks, as recently proposed by the Basel Committee and the FSF (FSF,

2008). A sufficient level of liquidity buffers limits idiosyncratic risks to a bank, by providing

counterbalancing funding capacity to weather a liquidity crisis. Moreover, buffers are important to

reduce the risk of collective reactions by banks and thereby to prevent the risk of amplifying effects

and instability of the financial system as a whole. Admittedly, this should be considered in conjunction

with the cost of holding higher liquidity buffers, also on the macro level of the financial system. To

assess such equilibrium effects one would perhaps need a more stylised model of the financial system.

Holding liquidity buffers should be part of sound liquidity risk management, which identifies

and measures the full range of liquidity risks, including the interaction between market and funding

liquidity and potential feedbacks on banks’ reputation related to signalling effects or flawed external

communication. Furthermore, to fully grasp the liquidity risk of a bank, stress-tests should cover the

group-wide liquidity exposures on a consolidated basis, including the risks of multi-currency

exposures, complex instruments and off balance sheet contingencies. These factors are included in

DNB’s liquidity report which has proven to be useful for Dutch banks and the supervisor, particularly

during the recent market turmoil. Based on the features of the liquidity report, Liquidity Stress-Tester

provides a tool to evaluate the importance of the various risk factors for banks’ liquidity positions in

different scenarios.

109

Appendix 6.1 Stylised balance sheet bank Y (base line)

110

Appendix 6.2 Impact banking crisis scenario

-40%

-30%

-20%

-10%

0%

0% 5% 10% 15% 20% 25% 30% 35%

Panel A.1 Bank size and 1st round impactImpact (E1 / Bo), share in total sectors' assets (x-axis)

Corr 0.06

-40%

-30%

-20%

-10%

0%

0 20 40 60 80

Panel A.2 Funding diversification and 1st round impactImpact (E1 / Bo), kurtosis of funding structure (x-axis)

Corr -0.05

High diversification Low diversification

0

1

0% 5% 10% 15% 20% 25% 30% 35%

Panel A.3 Bank size and reactionTrigger for reaction (0,1), share in total sectors' assets (x-axis)

Corr -0.15*

* significant at 10% confidence

0

1

0 20 40 60 80

Panel A.4 Funding diversification and reactionTrigger for reaction (0,1), kurtosis of funding structure (x-axis)

Corr -0.13*

High diversification

Low diversification

* significant at 10% confidence level

0%

25%

50%

75%

100%

0% 5% 10% 15% 20% 25% 30% 35%

Panel A.5 Bank size and shortfall probabilityProbability B3<0, share in total sectors' assets (x-axis)

Corr -0.11*

* significant at 10% confidence

0%

25%

50%

75%

100%

0 20 40 60 80

Panel A.6 Funding diversification and shortfall probabilityProbability B3<0, kurtosis of funding structure (x-axis)

Corr -0.06

High diversification

Low diversification

Notes 1) The banking crisis scenario assumes a withdrawal of non-bank deposits and other funding by professional money market players, other institutional investors and corporates and by withdrawals of savings deposits by households. 2) Panels A.1, A.3 and A.5 show the relationship between bank size on the horizontal axis and the impact of the first scenario round (A.1), the probability of reaction (A.3) and the probability of B3 < 0 (A.5) on the vertical axis. In panels A.2, A.4 and A.6, the diversification of funding is on the horizontal axis (as approximated by the kurtosis of the liability structure of banks’ balance sheets).

111

Appendix 6.3 Impact credit crisis scenario

-40%

-30%

-20%

-10%

0%

0% 5% 10% 15% 20% 25% 30% 35%

Panel A.7 Bank size and 1st round impactImpact (E1 / Bo), share in total sectors' assets (x-axis)

Corr -0.1

-40%

-30%

-20%

-10%

0%

0 10 20 30 40 50 60 70 80

Panel A.8 Funding diversification and 1st round impactImpact (E1 / Bo), kurtosis of funding structure (x-axis)

Corr +0.31**

High diversification Low diversification

** significant at 5% confidence level

0

1

0% 5% 10% 15% 20% 25% 30% 35%

Panel A.9 Bank size and reactionTrigger for reaction (0,1), share in total sectors' assets (x-axis)

Corr +0.24**

** significant at 5% confidence level

0

1

0 10 20 30 40 50 60 70 80

Panel A.10 Funding diversification and reactionTrigger for reaction (0,1), kurtosis of funding structure (x-axis)

Corr -0.27**

High diversification

Low diversification

** significant at 5% confidence level

0%

25%

50%

75%

100%

0% 5% 10% 15% 20% 25% 30% 35%

Panel A.11 Bank size and shortfall probabilityProbability B3<0, share in total sectors' assets (x-axis)

Corr -0.05

0%

25%

50%

75%

100%

0 10 20 30 40 50 60 70 80

Panel A.12 Funding diversification and shortfall probabilityProbability B3<0, kurtosis of funding structure (x-axis)

Corr -0.10

High diversification

Low diversification

Notes 1) The credit crisis scenario assumes declining values of banks’ tradable credit portfolios, due to uncertainties about the asset valuations which cause a drying up of market liquidity. The falling collateral values lead to higher margin requirements on banks’ derivative positions. 2) Panels A.7, A.9 and A.11 show the relationship between bank size on the horizontal axis and the impact of the first scenario round (A.7), the probability of reaction (A.9) and the probability of B3 < 0 (A.11) on the vertical axis. In panels A.8, A.10 and A.12, the diversification of funding is on the horizontal axis (as approximated by the kurtosis of the liability structure of banks’ balance sheets).

112

113

Chapter 7

Liquidity Stress-Tester: Do Basel III and unconventional monetary policy

work?

7.1 Introduction

7.1.1 Context

This chapter extends the Liquidity-Stress-Testing model described in Chapter 6.46 The extended

framework incorporates the new Basel III liquidity regulation, unconventional monetary policy and

credit supply effects. In the model, banks react according to the Basel III standards, endogenising

liquidity risk. It shows how banks’ reactions interact with extended refinancing operations and asset

purchases by the central bank.

The recent literature has added to the understanding of liquidity risk in relation to the

behaviour of market participants. Several studies link the drying up of market liquidity to

precautionary hoarding. Uncertainty is found to be a main reason for this. In the model of

Eisenschmidt and Tapking (2009), banks refrain from lending to other banks due to uncertainty about

their own liquidity needs, while Acharya et al. (2009) argue that pessimistic expectations lead to a

freeze in the interbank market. Liedorp et al. (2010) show that interbank contagion primarily runs

through funding exposures. Acharya et al. (2008) show that banks with surplus liquidity have an

incentive to underprovide liquidity to benefit from fire sales of assets, if there is imperfect competition

in the interbank market. Such predatory behaviour can lead to a decrease of market liquidity and a

freeze in the interbank market. The interaction between funding liquidity and market liquidity is

further explored by Brunnermeier and Pedersen (2009), who model liquidity spirals. These are due to

market illiquidity that increases funding constraints as a result of higher margins and to funding

shocks that hit traders and induce them to reduce trading positions, which adds to market illiquidity.

Cornett et al. (2010) provide empirical evidence on the relationship between liquidity profiles of US

banks and adjustments of their loan books in the crisis. They show that banks with more stable sources

of funding were better able to continue lending.

Fernando at al. (2008) demonstrate that market collapse can be an endogenous phenomenon,

depending on the commonality in the liquidity needs of market participants. This fits in the

endogenous cycle view of instability, where risk is endogenous with respect to collective behaviour of

market participants. Their reactions amplify shocks and aggravate a liquidity crisis. Endogenous cycle

models, where risk is endogenous with respect to collective behaviour of market participants, are still

primitive, with very limited behavioural content (Borio and Drehmann, 2009). This also holds for 46 This chapter is a revised version of Van den End (2010b).

114

macro stress-testing models that are used by central banks and supervisory authorities to simulate

shocks to the system as a whole (see Chapter 4). Even in the most sophisticated stress-testing models,

the behaviour of financial institutions is included by rules of thumb rather than through empirical

estimations. This resembles the assumed behaviour in agent-based simulation models in which

heterogeneous, bounded rational agents use rule of thumb strategies (Hommes, 2006). In macro stress-

testing models, responses are usually assumed to be triggered by shocks that lead to a declining

solvency ratio of banks below a certain threshold level. This default risk can be caused by a drying up

of market liquidity which depresses the value of banks’ assets, as in Cifuentes et al. (2005). This

triggers fire sales of assets, depressing market prices and inducing further sales. Default risk is also a

trigger for portfolio adjustments by banks in the stress-testing framework of the Oesterreichische

Nationalbank (Boss et al., 2006). In a recent version of the Bank of England’s RAMSI model,

behavioural responses are related to funding liquidity risks of banks (Aikman et al., 2009). Funding

strains increase the default risk of banks, which may at a certain stress level resort to fire sales of

assets. This leads to liquidity feedbacks through depressed market prices of assets. Gauthier et al.

(2010a) model funding liquidity risk as an endogenous outcome of the interaction between market

liquidity risk, solvency risk, and the funding structure of banks. Spill-over effects occur due to the

network effects among banks. In the Liquidity Stress-Tester model of Van den End (2010a), second

round feedback effects are determined by the number and size of reacting banks and the similarity of

reactions. Contagion results from the effects of balance sheet adjustments on prices and volumes in the

markets and funding channels where banks are exposed to.

7.1.2 Contribution

The Basel III liquidity regulation is the basic principle in this chapter.47 Key part of the proposed

regulation is two minimum standards for the funding liquidity risk of banks (BCBS, 2009b): the

Liquidity Coverage Ratio (LCR) and the Net Stable Funding Ratio (NSFR), see also Chapter 9. The

LCR ensures that banks have a sufficient liquidity buffer to survive an acute stress scenario lasting for

one month. It is defined as the stock of unencumbered high quality liquid assets, divided by net cash

outflows over a 30-day time period. The ratio should be at least 100% at all times. The numerator

includes cash, central bank reserves, high quality sovereign bonds and a proportion of high quality

corporate bonds and/or covered bonds; these ‘level 2 assets’ are capped at 40% of the total stock of

liquid assets and receive appropriate haircuts (BCBS, 2010b). Net cash outflows are determined by the

total expected cash outflows minus total expected cash inflows and reflect the net amount of funding

that may disappear within the 30 days under a stress scenario (see Section 7.2.2 and Appendix 7.1).

The NSFR is defined as the available amount of stable funding, divided by the required

amount of stable funding. This longer-term structural ratio must be greater than 100%. The NSFR

47 We refrain from discussing whether a price-based (through taxing) or a quantity approach (through buffers) is most optimal, as in Perotti and Suarez (2011).

115

creates additional incentives for banks to address liquidity mismatches by funding their activities by

more stable sources of funding. Stable funding is defined as equity and liability financing that is

expected to be available over a one year horizon under stress conditions, according to the Available

Stable Funding factors (ASF, see Appendix 7.1). Central bank borrowing is not considered stable

funding, in order not to create incentives for banks to rely on it. Required funding is determined by the

liquidity characteristics of a bank’s assets and contingent exposures, as reflected in the Required

Stable Funding factors (RSF, see Appendix 7.1).

Besides, the extended Liquidity Stress-Tester model includes a central bank reaction function.

Central banks have responded to the increased systemic risk since 2007 with extended liquidity supply

through unconventional monetary policy measures. This has supported the functioning of the money

market and averted a collapse of the banking system. These features are incorporated in the extended

model framework.

We contribute to the literature on liquidity risk and bank behaviour by providing a model

framework that links funding liquidity risk in a stochastic approach to regulation and central bank

operations. The model provides a tool for scenario analyses. As far as we know, there is no other

model that combines the Basel III regulation and monetary policy in a stress-testing framework.

Monte Carlo simulations produce the Basel III liquidity ratios after the first and second round effects

of a stress scenario. Since the model is an empirical algorithm that is driven by real data of banks’

liquidity positions, it can be applied to all banking systems that comply with Basel III. In the model,

banks react according to the proposed new liquidity standards, by assuming that they are a binding

incentive to behaviour. Behavioural responses have wider effects, through reputation risk and market-

wide disturbances, by which liquidity risk is endogenous. We simulate the first and second round

shock effects on the funding liquidity of banks and generate measures of systemic risk under various

stress scenarios. This process enables a quantification of some wider consequences of the proposed

liquidity regulation, for instance the impact on credit supply. This impact is found to be limited in the

model simulations of liquidity stress scenarios. Another result is that second round effects and tail

risks of a stress scenario are substantially lower if banks would adjust to Basel III by holding a higher

quality of liquid assets. In particular a narrowly defined liquidity buffer - consisting of high quality

government bonds - makes a big difference in limiting the tail risks of banks. The changing behaviour

of banks, triggered by the new liquidity regulation, interacts with monetary policy measures.

Simulations with the model show that as a consequence of larger bond holdings, monetary policy

conducted through asset purchases influences banks more compared to central bank refinancing

operations. We also simulate the consequences of an exit from extended refinancing operations on

banks’ funding liquidity. The outcomes indicate that the liquidity ratios of banks actually improve

compared to the pre-exit situation, if alternative stable funding is available.

The structure of the rest of the chapter is as follows. Section 7.2 outlines the model framework

and explains its structure for the first and second round effects of shocks to banks’ liquidity risk and

116

the reaction function of the central bank. Sensitivity analyses are performed to test the robustness of

the model to parameter changes. Section 7.3 presents model simulations for Dutch banks as an

illustration of the impact of different scenarios and the influence of Basel III and central bank

interventions. Section 7.4 concludes.

7.2 Model

7.2.1 Framework

The extended version of Liquidity Stress-Tester is a four stage algorithm, as represented by Figure 7.1

in stylised form. It models banks’ liquidity profiles after the first round effects of a stress scenario (t1),

after the mitigating actions of the banks (t2), after the second round effects (t3) and after the central

bank reaction (t4). In each stage, the model generates distributions of the LCR by individual banks

(including tail outcomes and the probability of breaching a certain LCR level). The scenario horizon is

one month, equal to the assumed stress horizon in the LCR.

Figure 7.1. Model framework

LCRt0, NSFRt0

Scenario 1st round effects LCRt1 STAGE 1

if ∆LCR > θ Λ LCRt1 < 100%

STAGE 2

LCRt2, NSFRt2 mitigate 1st round effects Reactions by bank

Loss of reputation

STAGE 3 Collect. Behaviour

LCRt3 2nd round effects

STAGE 4

LCRt4 Reaction by Central Bank

In the initial stage, LCRt0 and NSFRt0 are based on available balance sheet and cash flow information

of a bank. At t1 the first round effects of scenario shocks are simulated (first row in Figure 7.1). This

is conducted through Monte Carlo simulations of market and liquidity risk events, which are combined

in a multi-factor scenario. A scenario is designed as a set of shocks to banks’ liquid asset and liability

117

items (i) and is uniformly applied to all banks. In the model, any consistent combination of shocks can

be chosen. For instance, a credit market scenario is designed by assuming shocks to tradable credit

portfolios, collateral values and margin calls as first round effects. Shocks are reflected in stressed

weights of balance sheet and cash flow items (wi), with wi being the haircut or shocked inflow rate in

case of liquid assets and withdrawal rate in case of liabilities. The weights are based on the weighting

factors proposed by the Basel Committee for determining the LCR. The liquidity ratio declines

through changes in the weighting factors, which reflect reduced inflows, higher outflows and

additional haircuts on assets and thereby reduce LCRt1 compared to LCRt0 (see Figure 7.1). We

assume that the scenario shocks do not affect the weighting factors of the NSFR, because it is a

structural mismatch measure based on fixed weights for available and required funding.

In the second stage (t2), the mitigating measures by banks in response to the scenario are

simulated. Banks react if LCRt1 falls by more than a threshold θ and is below the supervisory

requirement of 100%. Banks aim to move the ratio back to its initial level, assuming this is the desired

steady state ratio. Banks are assumed to react by ‘within exposure’ adjustments, conducted through

shortening the maturity of assets and lengthening the maturity of liabilities and ‘across exposure’

adjustments, conducted through increasing liquid assets (stable liabilities) and reducing illiquid assets

(volatile liabilities). This reaction rule reflects the incentives of the LCR and NSFR. Hence, the

responses by banks improve both supervisory ratios (LCRt2 and NSFRt2).

The second round effects in stage 3 (t3) are shaped by systemic market-wide effects and

idiosyncratic reputation effects (see Figure 7.1). The systemic effects reflect the market disturbances

due to the reactions of banks in stage 2. In the model, these disturbances are larger if: i) more banks

react, ii) reactions are more similar and iii) the reacting banks are larger. Following from the reaction

rule, the market-wide effects are largest on illiquid asset markets (where assets are sold) and markets

for stable sources of funding (where banks scramble for funding). The model assumes that responding

to the scenario has negative repercussions for the bank involved. It faces reputation risk, as it might be

perceived to be in trouble by conducting measures to restore its liquidity ratio (signalling effect). Both

the possible loss of reputation and the wider effects on markets have an impact on LCRt3, through

additional haircuts on liquid assets and net outflows of liquidity, reflected in further stressed weights

(wi) of the items. The NSFR remains unchanged, since the items on the balance sheet do not change in

t3 and the stressed weights have no impact on this structural mismatch ratio.

The fourth stage of the framework becomes effective when the central bank reaction rule is

activated (see Figure 7.1). In that case, the central bank intervenes in the market to mitigate the second

round effects of the crisis scenario. The framework allows for asset purchases by the central bank and

extended refinancing operations. Both alleviate the stress on markets and banks. In the model,

extended refinancing operations are approximated by a substitution of private wholesale funding for

central bank borrowing. This reflects the larger intermediary role of the central bank owing to a

widened access to its refinancing operations. Asset purchases are modelled by adjusting the price

118

function of illiquid assets, to reflect that central bank interventions provide a floor for the asset values

on banks’ balance sheets. This can be applied to specific asset categories, taking into account that asset

purchase programs usually target specific stressed markets. The central bank measures improve LCRt4

of banks that rely on central bank credit and/or have exposures to supported asset classes.

7.2.2 Data

The Liquidity Stress-Tester model is calibrated on data of the liquidity positions of Dutch banks that

are available from the DNB´s supervisory liquidity report (DNB, 2003; see the data descriptions in

Chapters 2 and 6). From the data in the report, for each bank b we define vectors with dimension i x 1

that include values (I) of liquid asset and liability items (i), specified according to the maturity

structure that is relevant for the LCR and NSFR. Vector ILCR,t0 contains the balance sheet items and

contractual payments with maturity up to 1 month, vector INSFR_ST,t0 the items with maturity up to 12

months and vector INSFR_LT,t0 the items with maturity longer than 12 months. We map the data in the

liquidity report to the item classification of the Basel Committee to construct the LCR and NSFR (see

Appendix 7.1). For some items this mapping is imperfect, since the DNB report is not yet adjusted to

the Basel definitions.48 Since equity is also defined as stable funding, this item is added to vectors

INSFR_ST,t0 and INSFR_LT,t0. Data for equity is taken from DNB’s internal report on balance sheet statistics.

The data in the vectors are the item values measured at one point in time, period t0 (for instance, end

December 2009).

<

<

<

=

mI

..

mI

mI

I

t0i,

t0,

t0,

tLCR,0

1

1

1

2

1

<

<<

=

mI

..

mI

mI

I

t0,i

t02,

t0,

tNSFR_ST,0

12

12

121

≥

≥

≥

=

mI

..

mI

mI

I

t0,i

t0,

t0,

tNSFR_LT,0

12

12

12

2

1

The balance sheet items are weighed by haircuts or inflow rates of assets and run-off rates of liabilities

as proposed by the Basel Committee (see Appendix 7.149). These fixed weights reflect a mix of a bank

48 For most items the DNB and Basel Committee classifications are similar, in a limited number of cases we re-classify the exposures from the DNB report to the Basel classification on the basis of supervisory expert judgement. For some items this leads to an imperfect mapping and imprecise calculations of the LCR and NSFR. In particular this holds for internal securitisations (classified as bond holding in the DNB report, but not classified as liquid assets in the LCR) and deposits of corporates (no clear seperation between SME and non-SME in the DNB report, while the Basel classification takes this difference into account). Furthermore, the total item values in the liquidity report do not match precisely with the item values on the balance sheet, which could distort the calculation of the NSFR, which for some items can be based on balance sheet information. In cases where the mapping leads to a serious distortion of the liquidity ratios (e.g. for banks with large securitised asset holdings) we leave those banks out of the simulations. Moreover, the scenarios are presented in terms of deviations from baseline (i.e. from LCRt0 and NSFRt0), which reduces the influence of the imperfect mapping. 49 The table in Appendix 7.1 includes the liquidity value factors of assets; the haircut is equal to 100% minus that factor. In the model simulations we apply the weighting factors according to the decision of the overseeing body (GHOS) of the Basel Committee in July 2010 (BCBS, 2010b); these factors are slightly different for some balance sheet items compared to the weighting factors in Appendix 7.1, which originate from BCBS, 2009b.

119

specific and market wide scenario, which makes them a useful point of departure for our model. The

weights can easily be adjusted in the model, if the Basel Committee would opt for different values. In

the initial stage t0, liquid assets and liabilities are multiplied with those regulatory weights. The

parameterisation of the haircuts, inflow and run-off rates, either based on best practices or historical

data, is a weakness in most liquidity risk models of banks. This is because data of stress situations are

scarcely available and in times of stress the assumed elasticities may be different. As a consequence, a

bank may be too optimistic about the liquidity of assets and the stability of its funding. To account for

this uncertainty the model is based on a stochastic approach, by using Monte Carlo simulations of

haircuts, inflow and run-off rates.

The weights (haircuts, inflow and run-off rates) of the LCR at t0 are included in the i x 1

dimensional vector WLCR. The available and required stable funding factors of the NSFR (ASF, RSF in

Appendix 7.1) are included in vectors WNSFR_ST and WNSFR_LT. The former contains the ASF and RSF

factors that are applicable to the balance sheet items with maturity up to 12 months and the latter the

factors that belong to the items with a maturity longer than 12 months.

=

LCRi

LCR

LCR

LCR

w

..

w

w

W2

1

=

<

<

<

mRSF,ASFi

mRSF,ASF

mRSF,ASF

NSFR_ST

w

..

w

w

W

12

122

121

=

≥

≥

≥

mRSF,ASFi

mRSF,ASF

mRSF,ASF

NSFR_LT

w

..

w

w

W

12

122

121

7.2.3 Initial liquidity ratios

LCRbt0 is determined as,

)inflowsCashoutflowsCash(

AssetsLiquidLCR b

t0bt0

bt0b

t0 −= (7.1)

)w(IbAssetsLiquid LCRjb

t0,jj

t0 −= ∑ 1 (7.2)

LCRb

t0,LCRbt0 W'IinflowsCash = (7.3)

LCRb

t0,LCRbt0 W'IoutflowsCash = (7.4)

Liquid assets ∑I j in equation 7.2 is the weighted sum of assets 1, 2 .. j defined by the Basel Committee

as the stock of high quality liquid assets and level 2 liquid assets (up to 40% of the total buffer, BCBS,

2010b). This buffer contains assets measured in stressed conditions, as reflected by the haircuts wj LCR.

120

Vector ILCR,t0 in equation 7.3 contains the outstanding balances of contractual receivables, that are

expected to flow in according to WLCR over 1 month. Vector ILCR,t0 in equation 7.4 contains the

outstanding balances of various categories of liabilities that are expected to run-off or be draw down

according to WLCR over 1 month.

NSFRbt0 is determined as,

bt0

bt0

t0fundingstableRequired

fundingstableAvailableNSFRb = (7.5)

)W'I()W'I(bfundingstableAvailable LT_NSFRb

t0,LT_NSFRST_NSFRb

t0,ST_NSFRt0 += (7.6)

)W'I()W'I(fundingstablequiredRe LT_NSFRb

t0,LT_NSFRST_NSFRb

t0,ST_NSFRbt0 += (7.7)

Vector INSFR_ST in equation 7.6 contains the liability items with maturity up to 12 months and INSFR_LT in

equation 7.7 the asset items with maturity longer than 12 months. WNSFR_ST and WNSFR_LT are the vectors

with ASF and RSF factors as defined in Section 7.2.2.

7.2.4 First round effects

In the model, the fixed weighting factors of LCRt0 are assumed to be 0.1% tail events (wi LCR ≈ 3 σ).

The scenario impact of the first round effect on an item i at t1 is determined by simulated weights (wi

sim1). These are based on Monte Carlo simulations by taking random draws from a log-normal

distribution Log-N (0,1), scaled by (3

LCR wi ), so that wi sim1 ~ Log-N (µ,σ2). The use of a log-normal

distribution is motivated in Chapter 6. The distribution is bounded below by 0, which fits with the

simulated weights in our model. As an upper bound, the weights are conditioned by (wi LCR + wi sim1) ≤

100%, since haircuts and withdrawal rates cannot exceed 100%.

The first round effects have an impact on the liquidity position of bank b, which is modelled

through additional haircuts on assets, reduced inflows and higher outflows (reflected in wi sim1), on top

of the regulatory weighting factors (wi LCR), affecting LCRbt1 and its subcomponents,

)inflowsCashoutflowsCash(

AssetsLiquidLCRb

bt1

bt1

bt1

t1 −= (7.8)

)ww(IAssetsLiquid sim1jLCRjb

t0,jj

bt1 −−= ∑ 1 (7.9)

121

)W'I(inflowsCashinflowsCash sim1b

t0,LCRbt0

bt1 −= (7.10)

)W'I(outflowsCashoutflowsCash sim1b

t0,LCRbt0

bt1 += (7.11)

Wsim1 in equations 7.10-7.11 is the i x 1 dimensional vector of simulated effects of the first round of

the scenario. The scenario shocks do not change the available and required stable funding factors

(ASF, RSF). These reflect the structural mismatch on the balance sheet and are assumed not to be

influenced by a stress scenario that lasts for the shorter horizon of the LCR (i.e. one month). Hence,

NFSRt1 = NSFRt0.

7.2.5 Mitigating actions by banks

In the model, a bank b reacts if its LCRbt1 falls more than the threshold θ compared to LCRbt0 in one of

the simulations and if LCRbt1 is below the supervisory requirement of 100%.50 It presumes that as long

as LCRbt1 remains above 100% a bank has no need to react, because it has excess liquidity buffers

above the supervisory minimum requirement that can be used to absorb the scenario shocks. A

reacting bank tries to restore its liquidity ratio by raising additional liquid assets and improve the

stability of funding, up to the level of its initial liquidity ratio. This reaction rule reflects the liquidity

hoarding and the scramble for stable sources of funding by banks in the crisis (ECB, 2009d). The

resulting new item values I i,t2 (elements of vectors ILCR,t2 and INSFR,t2) result from adjusting the initial

item values I i,t0,

])R)(S()w(E[II t2t2sim1ibt1

bt0,i

bt2,i λλ −+−+= 11 (7.12)

)outflowscashNett1outflowscashNet()assetsLiquidt0assetsLiquid(bE t0t1t1 −+−= (7.13)

LT_NSFRST_NSFR

i

bi

bi

t2 Iiandliabilityi,IiandassetiifI

IS ∈=∈=+=

∑ (7.14)

LT_NSFRST_NSFR

i

bi

bi

t2 Iiandssetai,IiandiabilityliifI

IS ∈=∈=−=

∑ (7.15)

assetiifw

RmRSFi

t2 =−

= <100

50 12 (7.16)

liabilityiifw

RmASFi

t2 =−

=<

100

5012 (7.17)

50 The threshold should reflect a substantial decline of the LCR; the precise threshold value could be based on supervisory intervention levels.

122

Ebt1 in equation 7.13 is the amount of liquidity lost due to the first round of the scenario, through

haircuts on asset holdings, additional outflows and reduced inflows. It represents the liquidity needed

to bring the LCR back to its assumed steady state level and thereby determines the amount of

mitigating actions by a bank.51 The benefits of the mitigating measures depend on the market

disturbances in the first round of the scenario (reflected in wi sim1 in equation 7.12). In an extreme stress

situation, funding markets could dry up, leaving banks with no possibility to enter a market segment i

to raise liquidity. This is the case if wi sim1 = 100%.

The last part of equation 7.12 is the reaction rule, which determines the type of instruments

(items i, amounting I) which banks use to react. The rule is based on the specialisation of a bank (S)

and a regulatory component (R). Variable S ranges from [-1, 1] and drives the ‘within exposure’

reactions, which are conducted through shortening the maturity of assets and lengthening the maturity

of liabilities, to the extent of the share of an item on the balance sheet. Variable R ranges from [-0.5,

0.5] and reflects the ‘across exposure’ reactions. It assumes that banks substitute different balance

sheet items, through increasing liquid assets and stable funding and reducing illiquid assets and short-

term liabilities. Parameter λ [0, 1] in equation 7.12 is a behavioural parameter that weights both parts

of the reaction rule. If the horizon of the scenario is short, it is more likely that reactions will be static

(λ > 0.5 and S dominates), while a longer period of time provides banks more leeway to steer their

balance sheets (λ < 0.5 and R dominates).

In equation 7.14 the maturity structure is shortened by increasing the amount of assets falling

due within one year and in equation 7.15 by reducing assets with a maturity longer than one year (vice

versa for liabilities).52 The idea behind this part of the reaction rule is that in liquidity crises, time is

usually short, leaving banks little opportunity to change their strategy (e.g. by diversifying funding or

spreading risk). Such a ‘static’ reaction by banks in crises is confirmed by empirical research (Tabbae

and Van den End, 2011) and is also applied in the model of Aikman et al. (2009).

We assume that banks change their balance sheet according to the available and required

stable funding factors (ASF, RSF) in equations 7.16-7.17. The value 50 in the equations represents the

mid point of stable funding factors, which range from [0,100]. It acts as a pivoting factor that

determines the extent to which banks adjust the balance sheet items, implying that banks substitute

illiquid assets (wRSF > 50) for liquid assets (wRSF < 50) and volatile funding (wASF < 50) for stable

sources of funding (wASF > 50). This ‘regulatory imposed pecking order’ reflects reactions by banks in

liquidity crises, when liquid assets are hoarded and volatile funding substituted with stable sources

such as retail savings. Cornett et al. (2010) provide empirical evidence of such reactions in the recent

51 In the model, liquidity lost on the asset side determines the mitigating actions taken with assets, and liquidity lost on the liability side determines the measures taken by liabilities. The model imposes the restriction that a bank does not expand its total balance sheet by the mitigating actions; i.e. the sum of elements in I i,t2 is smaller or equal to the sum of elements in I i,t0. This is applied by reducing the assets and liabilities proportionally if the balance sheet total at t2 would exceed the total at t0. 52 A restriction in the model is that a bank with no exposure in a certain market does not enter this market.

123

crisis. Since the mitigating actions of a bank change its balance sheet composition in the direction of a

lower maturity mismatch (reflected in vectors ILCR,t2 and INSFR,t2), both supervisory ratios improve

compared to the ratios at t1 (with LCRt2 > LCRt1 and NSFRt2 > NSFRt1).

7.2.6 Second round effects

The reactions of banks have wider disturbing (endogenous) effects on markets that feed back on the

banks. This will crystallise in additional haircuts on liquid assets in the markets where banks react

and/or cause additional withdrawals and reduce the availability of liquid funding. For instance, if

many banks try to lengthen their funding profile, term funding will become scarcer. Such effects are

reflected in wi sim2, an element of the i x 1 dimensional vector of simulated second round effects,

))R)(S(n

n(react

sim1isim2it3t3syst

react

nwwωλλ −+

=1

(7.18)

∑∑

∑=

i

bi

b

bi

bt3

I

I

S (7.19)

100

12mRSF,ASFit3

wR

<= (7.20)

The multiplication factor to the first round effects (wi sim1), which determines the second round effects

(wi sim2), depends on the number of reacting banks (nreact) and increases if nreact exceeds a hurdle level

nsyst, which ranges from [0, n]. This reflects that if the number of banks responding becomes too high,

it may disturb the financial system, incorporating a channel of contagion between banks. The second

round effects further depend on the size and similarity of the transactions (St3 in equation 7.19), which

is specified as the reaction with an individual balance sheet item relative to the total balance sheet

adjustment, aggregated over all banks. It is a measure of concentrated trades on specific market

segments. The empirical findings in Chapter 2 show that the size of reactions, the number of reacting

banks and the concentration of reactions on market segments go in tandem with elevated levels of

market stress. The regulatory component Rt3 in equation 7.20 is equal to the available and required

stable funding factors (ASF, RSF), for assets and liabilities. Rt3 differs slightly from Rt2. The intuition

behind this is that second round effects are largest on illiquid asset markets with a high RSF factor

(where assets are sold) and markets for stable funding with a high ASF factor (where banks scramble

for funding). Second round effects are more moderate in highly liquid markets (e.g. AAA government

bond markets), where asset sales have little disturbing effects and in short-term funding markets,

where banks reduce their demand for funding and cash-long banks increase their loan supply.

State variable ω represents the exogenous level of market stress that determines the

availability of market liquidity. It is derived from standardised distributions of risk aversion indicators

(see Chapter 6 for a detailed description of s which is equal to ω). Variable ω also magnifies the

124

effects on market liquidity of fire sales of assets and scramble for long-term funding (reflected in Rt3 in

equation 7.20).

Banks that react to restore their liquidity position face a reputation risk in financial markets.

While applying sensible measures ought to strengthen a banks’ financial position and comfort

counterparties, the adverse signalling effect of active trades could feed back on the funding conditions

of banks. This could translate in even more (idiosyncratic) haircuts on assets and withdrawals of

funding, as reflected in wi simR,

ϖRtsimi

bsimRi Sww 32= (7.21)

bt0,i

bt2,iR

t3I

IS

∆+= 1

(7.22)

The reputation effect depends on the relative size of reactions conducted by a bank (bt0,i

bt2,i

I

I∆) and the

market conditions (ω).53 Particularly in stressed circumstances the signalling effect of reactions will

adversely feed back on a bank (the stigma associated with accessing central bank standing facilities in

the recent crisis is illustrative).

The systemic and idiosyncratic second round effects both have an impact on LCRt3 through additional

haircuts on assets (the numerator) and net outflows (the denominator), as determined by wi sim2 (which

is replaced by wi simR if a reacting bank faces a reputation risk),

)inflowsCashoutflowsCash(

AssetsLiquidLCRb

bt3

bt3

bt3

t3 −= (7.23)

)ww(IAssetsLiquid sim2jLCRjb

t2,jj

bt3 −−= ∑ 1 (7.24)

)W'I(inflowsCashinflowsCash sim2b

t2,LCRbt2

bt3 −= (7.25)

)W'I(outflowsCashoutflowsCash sim2b

t2,LCRbt2

bt3 += (7.26)

Like in case of the first scenario round, the NSFR remains unchanged since the items on the balance

sheet do not change in t3, neither do the available and required stable funding factors (ASF, RSF).

Therefore NFSRt3 = NSFRt2.

53 If ω = 1 (the downside restriction for ω), the mitigating reaction of a bank will not be counteracted by adverse reputational effects and will by definition improve a bank’s liquidity position.

125

7.2.7 Central bank reaction function

The systemic nature of the 2007-2010 liquidity crisis prompted central banks to extend their liquidity

supply to banks and to introduce asset purchase programs. In the model, those features are captured by

a central bank reaction function which, if switched on in the fourth stage, mitigates the second round

effects of a scenario. The model includes a central bank reaction function for refinancing operations

and one for asset purchases.

Extended refinancing operations are modelled by assuming that banks borrow a larger amount

from the central bank (backed by collateral), while wholesale funding is proportionally lower,

reflecting a shut down of the capital market for bank funding. This assumption is applied by changing

the elements in vectors ILCR,t0, INSFR_ST,t0 and INSFR_LT,t0. Since the roll-off percentage of central bank

financing in the LCR (25% for secured funding backed by assets not included in the stock of liquid

assets) is lower than the run-off rate of wholesale funding (100% for debt securities issued on the

private market), a greater reliance on the central bank reduces a bank’s vulnerability to stress

scenarios. In terms of the model, this means that if the central bank reaction is switched on the

liquidity outflow in the denominator of the LCR decreases and hence LCRt4 > LCRt3.

The mitigating influence of asset purchases by the central bank is modelled by a price function

of illiquid assets. Cifuentes et al. (2005) specify an inverse demand curve which also features in

models of Bank of England (Aikman et al., 2009) and Bank of Canada (Gauthier et al., 2010b),

nq

epα−

= (7.27)

where p is the price of an illiquid asset and qn the fraction of illiquid assets sold by n banks on the

market. The maximum price p = 1 occurs when sales are zero. Parameter α is a positive constant and is

calibrated by assuming that p falls by around 50% if all illiquid assets of the banks are sold (as in

Cifuentes et al., 2005). The corresponding value of α is 0.8. If qn = 0.2 the haircut on market prices is

15% (Figure 7.2). This equals, for instance, the haircut wi LCR on level 2 bonds (BCBS, 2010b), which

reflects a three standard deviation shock on these assets (see Section 7.2.4). We extend equation 7.27

with asset purchases by the central bank (qCB) that counters the price effects of asset sales and shift the

demand curve up to the right (see Figure 7.2).

)CB

qn

q(ep

−−=

α (7.28)

If asset purchases equal banks’ fire sales (qCB = qn) there is no price effect and p = 1. The mitigating

effect of central bank interventions crystallise in lower second round weights for bonds. For instance,

if the fraction of level 2 bonds sold would be 20% (qn = 0.2) and the central bank purchases half this

126

amount (qCB = 0.1) than wi sim2 for those bonds is limited at 7.5% in stead of 15%. This mitigates the

shock to a 1.5 standard deviation event. The central bank could target specific market segments. For

instance, the reaction function can be framed such that wi sim2 is only limited for covered bonds. The

limited haircuts support the value of assets included in the numerator of LCR and hence LCRt4 >

LCRt3. Banks with exposures to the supported asset classes will benefit more than other banks from

the asset purchases.

0,6

0,7

0,8

0,9

1,0

0,2 0,25 0,3 0,35 0,4 0,45 0,5

Figure 7.2. Effect asset purchases Central BankInverse demand curve for bond prices

(y axis: price p of illiquid assets, x axis: fraction q n of fire sales)

q CB = 0.1

q CB = 0

q CB = 0.25

q CB = q n

7.2.8 Parameter sensitivity

The calibration of λ (behavioural parameter), nsyst (threshold for number of reacting banks) and the

level of market stress (ω) can be based on economic intuition, guided by sensitivity analyses. As an

illustration we test the sensitivity of the model outcomes to changing the parameter values for a

stylised bank (bank Y in Appendix 7.2). A hypothetical scenario is assumed to affect all liquid assets

and liabilities, through fixed instead of stochastically simulated weights. Furthermore, it is assumed

that the first round effect of the scenario leads to a decline of the initial liquidity ratio that exceeds the

threshold θ and that the bank reacts with all instruments available at its disposal (i.e. liquid assets,

credit, term funding, savings and equity, as presented on the stylised balance sheet). In the base line

situation, the level of market stress (ω) is set at 1.5, the number of reacting banks (nreact) at 10 and the

behavioural parameter λ at 0.5.

127

Table 7.1. Parameter sensitivityImpact on Liquidity Coverage Ratio (LCR) of bank Y

(Resulting LCRt3 as percentage of LCRt3 in base line)

Changing the parameters (in isolation)

ω 1 1.5 2 2.5 3Impact 21% 0% -14% -22% -28%

n react 5 10 15 20 25Impact 40% 0% -24% -33% -43%

n syst 5 10 15 20 25Impact -28% 0% 21% 31% 37%

λ 0 0.33 0.5 0.66 1Impact 18% 5% 0% -6% -19%

Reputation (Y/N) N* Y**(ω =1) Y**(ω =1.5) Y**(ω =2) Y**(ω=3)Impact 0% 20% -27% -35% -53%

Note: the shaded cells present the parameter values in the baseline situation.N*: for the purpose of the sensitivity analysis in the base line situation,it is assumed that there is no reputational risk (i.e. w i simR = w i sim2 )

Y**: to isolate the impact of reputation risk, ω is only changed in Equation 7.21.

Table 7.1 shows the impact of changing each parameter in isolation, in terms of deviations from LCRt3

in the baseline situation. The model outcomes are quite sensitive to changes of ω (LCRt3 is 28% lower

if ω = 3). This is in line with the intuition that extreme market conditions can severely affect on the

liquidity risk of banks. Following from equation 7.21, the impact of reputational risk also depends on

the level of market stress. Table 7.1 shows that reputation risk could severely impact on banks in

stressed markets. In a tail situation with ω = 3, the LCR is more than halved compared to the baseline

in which reputational risk is assumed to be absent and ω = 1.5. Both the number of reacting banks and

hurdle level nsyst have the expected impact on LCRt3, indicating the models’ sensitivity to behavioural

reactions. Increasing the number of reacting banks exacerbates the second round effects of a scenario

(as reflected in equation 7.18) and reduces LCRt3. This effect is stronger at a lower value of nsyst. It

implies that a higher susceptibility to collective responses leads to larger second round effects if more

banks react.

Shifting behavioural parameter λ has the expected influence on the LCR. If the reaction rule

does not follow the new regulation at all (i.e. λ = 1) LCRt3 is almost 20% lower than under the reaction

rule in the baseline situation (λ = 0.5). If a bank only reacts according to the regulatory incentive (λ =

0), the LCR is almost 20% higher than in the baseline. This is because the regulatory variable R

promotes that a bank changes its balance sheet composition from less liquid to more liquid assets (and

from volatile to stable funding), which improves the liquidity ratio.

The stylised example assumes that the bank reacts to the scenario because the decline of LCRt1

compared to LCRt0 exceeds threshold θ = 25%, with LCRt1 < 100%. To illustrate the sensitivity of the

model outcomes to changing threshold θ, a crisis scenario was run (for all Dutch banks in the sample)

128

with a lower and a higher threshold. Table 7.2 shows that a higher threshold level improves the final

outcome of the LCR. Although banks’ reactions to restore the liquidity profile mitigate the first round

impact of a scenario, if more banks would react, the disturbing influence of collective responses

dominate the mitigating effects. Table 7.2 shows that if θ = 5%, five more banks would react

compared to the baseline situation (θ = 25%), while 16 banks less would react if θ is raised to 75%.

The reduced number of collective responses limits the second round effects of the scenario, with

LCRt3 on average being 8 percentage points higher than in the baseline situation.

Table 7.2. Outcomes at different reaction triggers (θ )Deviation from outcomes with θ =0.25. Ratios in percentage points (sectoral averages)Baseline parameters: θ=25%, n syst =10, λ=0.5; ω=1.5, n =10,000 simulations

Reaction thresh. Reaction thresh.θ = 5% θ = 75%

stage 1

LCR after 1st round (LCRt1) 0.0 0.0

stage 2

Number of reactions: n react 5 -16

LCR after reaction (LCRt2) 0.1 -2.2

NSFR after reaction (NSFRt2) 0.4 -9.0

stage 3

LCR after 2nd round (LCRt3) -9.4 8.0

LCRt3 5% tail 0.8 5.2

LCRt3 1% tail 1.2 4.1

7.3 Results

This section describes the simulation outcomes of three scenarios: a replication of the recent credit

crisis, a wholesale and a retail bank run. The scenarios are simulated with end-2009 data of all 85

banks in the Netherlands (including subsidiaries of foreign banks).54 The model outcomes are based on

10,000 Monte Carlo simulations. We assume λ = 0.5 (the behavioural parameter), θ = 25% (threshold

for reactions), nsyst = 10 (i.e. hurdle number of reacting banks), ω = 1.5 (the middle of the range

determined in Section 7.2.655). Initially no central bank interventions are assumed. Section 7.3.4

54 By running the model with a sample of banks in one country it is assumed that the second round effects of scenarios are caused by shocks in the local banking sector. The reactions of banks in other countries are not taken into account. 55 Note that during the peak of the crisis, in October 2008, ω was around 5, both based on corporate bond spreads and implied stock price volatility. In that respect the presented model results of the replicated credit scenario may be an underestimation of the second round effects.

129

presents the results including interventions and Section 7.3.5 the results if banks would have a stronger

initial position in line with the new liquidity regulation.

Table 7.3. Outcomes scenario runsDeviation from initial ratios (LCRt0, NSFRt0) in percentage points (sectoral averages)

λ=0.5; θ=25%, n syst =10, ω=1.5, n=10,000 simulations, no CB intervention

Credit crisis Wholesale run Retail run

stage 1

LCR after 1st round (LCRt1) -16.3 -4.7 -1.1

stage 2

Number of reactions, n react 26 12 20

LCR after reaction (LCRt2) -10.0 -4.7 -0.8

NSFR after reaction (NSFRt2) 196.8 176.9 176.4

stage 3

LCR after 2nd round (LCRt3) -48.1 -39.8 -46.2

LCRt3 5% tail -57.9 -60.4 -61.6

LCRt3 1% tail -59.9 -62.8 -63.1

Scenario shocks

Shock size (average w sim1 ) 17.1 15.5 14.7

Shock size (average w sim2 ) 32.7 25.6 31.5

7.3.1 Credit crisis scenario

The credit crisis scenario in first instance affects the asset side of banks’ balance sheets. It assumes

declining values of tradable credit portfolios, strains in repo markets and higher margin requirements

on banks’ derivative positions. Similar to the recent crisis, we assume difficulties to roll-over asset

backed commercial paper (ABCP) and drawings on committed liquidity facilities of banks. Increased

counterparty risks among banks worsens their access to funding in bond and commercial paper

markets. These first round effects are simulated by stressing the weights of credit portfolios, margin

requirements, liquidity facilities and securities issued (through wsim1, see Table 7.3).

Table 7.3 shows the weighted average outcomes for the whole sample of banks.56 The first

round effect erases over 16 percentage points of the initial LCR, with a maximum of around 100

percentage points for the bank that is most severely affected. Although most banks would be hit by the

scenario (i.e. LCRt1 < LCRt0), banks that are not affected at this stage are mostly small branches of

foreign banks. They can count on liquidity support from the head office and probably therefore do not

hold eligible collateral (which explains why the LCR of some bank is nil). 56 If the denominator of the LCR equals 0 because outflow = inflow, than the LCR is fixed at 100%.

130

Table 7.3 shows that in case of 26 banks, the decline of LCRt1 compared to LCRt0 exceeds the

threshold θ = 25% and LCRt1 < 100%. This triggers them to restore their liquidity ratio to the initial

level (LCRt0). The reactions mitigate the first round effect of the scenario on the sector as a whole to

10 percent points on average (the difference between LCRt2 and LCRt0). We find that larger banks

with a more diversified funding structure tend to react relatively more than smaller banks, which

relates to the higher probability that large universal banks are hit by the credit crisis scenario.

According to the model (equation 7.19), the responses of large banks potentially have a relatively

strong impact on markets.

The collective responses of banks (e.g. fire sales of assets, acquiring stable funding sources) in

response to the first round shocks of the scenario further disturb financial markets. As a consequence,

in the second round of the scenario, market illiquidity spills over into strained funding liquidity. These

second round effects (wsim2, see Table 7.3) are simulated by further stressing the haircuts of assets and

run-off of liabilities on top of the first round effects, according to equations 7.18-7.20. The reputation

risk of the reacting banks translates into additional (idiosyncratic) stress on the weights (wsimR)

according to equation 7.21. On the other hand, banks that react improve their liquidity profile, by

increasing liquid asset holdings and stable funding, while reducing illiquid assets and volatile funding.

All in all, reputation and systemic (second round) effects have an impact on the final liquidity ratio

(LCRt3). Table 7.3 shows that on average LCRt3 is 38 percent points lower than the ratio after the

mitigating actions (48.1 minus 10.0 percentage points). It also shows the 5% and 1% tail outcomes. In

the 5% (1%) tail the LCR declines by 58 (60) percentage points on average. Insight in the extreme tail

outcomes is particularly relevant for financial stability analysis, which assesses the resilience of the

system to extreme, but plausible shocks. We find no significant correlation between the shortfall

probability and size or funding diversification of banks, indicative of the systemic dimension of the

second round effects, which affect all types of banks.

7.3.2 Wholesale and retail bank scenarios

The bank run scenarios simulate funding liquidity crises, seizing at the liability side of banks’ balance

sheets. The wholesale run assumes that banks and other professional money market parties withdraw

unsecured demand deposits from other banks and do not roll-over their unsecured fixed-term deposits.

The retail run scenario assumes a run on retail saving accounts, through withdrawal of demand

deposits and no roll-over of fixed-term deposits. The first round effects of the scenarios are simulated

by stressing the weights of the affected demand and fixed-term deposits (through wsim1, see Table 7.3).

These weights determine the amount of liquidity lost (Ebt1) due to the first round of the scenarios,

according to equation 7.13 and determine LCRt1 according to equation 7.8.

Table 7.3 shows that on average the first round effects are larger in the wholesale scenario

(LCRt1 declines almost 5 percent points) than in the retail scenario. Besides, the scenarios have a

different distributional impact. We find that the wholesale scenario has a larger impact on big banks

131

than small banks; while less diversified banks are most susceptible for the retail scenario. Less

diversified banks are usually more dependent on savings and therefore vulnerable to a retail run. In the

retail run scenario there are 20 (mostly smaller) banks where the decline of LCRt1 compared to LCRt0

exceeds threshold θ = 25%, while LCRt1 < 100%. This triggers them to restore their liquidity ratio to

the initial level (LCRt0). In the wholesale scenario the number of reacting banks is smaller. The

reactions barely mitigate the first round effect of the scenarios (LCRt2 ≈ LCRt1), which indicates that

the market disruptions of the first round effects reduce the opportunities for banks to raise additional

liquidity. The mitigating actions markedly improve the NSFR. This is caused by the regulatory

component in the reaction rule, which stimulates banks to reduce their funding mismatch.

In the second round of the scenarios, the mitigating actions lead to further stress on asset

values (fire sales raise the haircuts on assets) and funding (liquidity hoarding by counterparties

increases the run-off rates of funding). Next to these systemic second round effects, reacting banks

face a reputation risk. On average, LCRt3 declines 46.2 percentage points in the retail scenario,

compared 39.8 percentage points in the wholesale scenario. The effects in the former scenario are

larger because more banks react (20 versus 12 in the wholesale scenario), leading to more widespread

market disturbances and reputational effects than in the latter scenario. As a consequence, the 5%

(1%) tail risks are also larger in the retail scenario; the LCR is 61.6 (63.1) percentage points lower

than the initial ratio on average.

7.3.3 The impact on credit supply

The model allows for analysing the impact of liquidity stress and the reactions by banks on credit

supply. The credit effect depends on the reaction function in equation 7.12, which is driven by ‘within

exposure’ (St2) and ‘across exposure’ adjustments (regulatory variable Rt2), as well as by parameter λ.

Within exposure adjustments in the case of loans imply according to equation 7.16 that outstandings

with a maturity up to one year are increased, while longer term loans are reduced. Across exposure

adjustments imply that banks increase the most liquid assets (items with wRSF < 50), while reducing

less liquid assets. Assume that the RSF for residential mortgages and other loans with a maturity up to

one year (with a 35% or better risk weight under Basel II’s standardised approach for credit risk) is 65

and the RSF is 50 for loans to non-financial corporate clients (BCBS, 2010b). The resulting

unweighted average RSF for the retail and corporate credits (57.5) is close to 50, the pivoting factor

that determines the extent to which banks increase or reduce individual balance sheet items in

equations 7.16-7.17. This implies that banks only slightly reduce their loan books in response to

shocks.

The impact on credit supply is illustrated by the credit crisis scenario at various values of

parameter λ (see Figure 7.3). In all cases, total credit supply contracts limitedly, with the contraction in

132

long-term loans exceeding that of short-term loans.57 This follows from both components of the

reaction rule; St2 causes a shortening of the maturity of assets (e.g. an increase of short-term loans and

decrease of long-term loans) and Rt2 an increase of liquid assets at the expense of illiquid assets (i.c.

longer-term loans). Simulation outcomes of the credit crisis scenario with λ = 0.5 show that long-term

loans would be reduced by 0.6%, while short-term loans are barely affected (see Figure 7.3). The latter

owes to the ‘within exposure’ adjustment, which implies that long-term loans are substituted for short-

term loans. If this rule dominates (λ = 0.75) than short-term credit even increase. The regulatory part

of the reaction rule has a downward effect on both short- and long-term loans, since wNSFR_ST and

wNSFR_LT are both higher than 50. Of course this outcome depends on the calibration of the RSF for

loans; if it would be lower than assumed in the simulation, there would be less downward pressures on

credit supply.

-0,8%

-0,6%

-0,4%

-0,2%

0,0%

0,2%

0,4%

λ=0.25 Baseline (λ=0.5) λ=0.75

maturity < 1 yr maturity > 1 yr

Figure 7.3. Impact on credit supplyPercentage change (credit crisis scenario)

7.3.4 The influence of central bank interventions

To illustrate the effects of central bank interventions, several scenarios were run with the central bank

reaction function switched on. Extended refinancing operations are simulated by assuming that

wholesale funding partly becomes unavailable and that banks substitute it with borrowings from the

central bank. In another simulation we assume an exit from the extended central bank borrowings,

through reducing them again and increase alternative funding proportionally.58

57 In the wholesale and retail run scenarios, the impact on aggregate credit supply is negligible, since the model simulations indicate that large banks do not breach the reaction threshold and hence do not adjust balance sheets. 58 The increase of central bank borrowing is simulated by assuming that wholesale funding is halved and that this loss of funding is made up by central bank borrowing. The exit is simulated by assuming that banks substitute the additional central bank funding again with wholesale funding, or in an alternative simulation, with retail savings.

133

To simulate the mitigating effect of an asset purchase program, we assume that the central

bank stabilises the bond market by purchasing government and non-financial bonds. For instance, in

the credit crisis scenario the haircut on bonds due to the second round effects (wsim2) is 26% on

average, which reflects a nine standard deviation shock.59 According to the demand function in

equation 7.27, this impact would occur if 35% of the bond holdings of banks is sold (qn = 0.35). If the

central bank counters the impact and aims to normalise the situation in the bond market to a one

standard deviation shock (i.e. wsim2 ≈ 3%), it should buy around 90% of the bonds sold on the market

(qCB = 0.30). Besides, we simulate that the asset purchases target a specific market segment, by

assuming the central banks tries to normalise the situation on the market for high quality government

bonds (to a one standard deviation shock). This mimics the Securities Market Program (SMP) of the

ECB, which aims at distressed sovereign bond markets. The central bank interventions limit the

haircuts on bonds, supporting the value of liquid assets and improve the LCR.

Simulations were run for extended refinancing operations and asset purchases together and

separately. The outcomes indicate that central bank interventions substantially mitigate the second

round effects of a stress scenario. In the credit crisis scenario, extended refinancing operations together

with asset purchases result in LCRt4 being on average nearly 6 percentage points higher compared to

LCRt3 in the scenario excluding central bank interventions (see Figure 7.4). The interventions limit tail

risk in particular, as indicated by the higher outcome of the 5 and 1% tail ratios. Tail risks are

relatively more limited by asset purchases than by extended refinancing operations. This is explained

by the fact that for most banks, bond holdings are an important item on the balance sheet, implying

that limiting the volatility of bond prices reduces the risk of small probability, high impact losses of

banks. The tail risks primarily relate to riskier bonds, such as issued by corporates. High quality

government bond holdings have less tail risk. This is indicated by the outcomes of simulating only

government bond purchases by the central bank, which have less influence on the tail impact (see

Figure 7.4).

59 This haircut (wsim2) results from the scenario simulation in Section 7.3.1; it is the average haircut on bonds included in the denominator of the LCR, weighted by the total holdings of these bonds. Compared to the weighted average haircut on bonds in the initial situation (wLCR ≈ 9%) - which is assumed to be a three standard deviation shock in Section 7.2.4 - the 26% average price fall is equal to a nine standard deviation shock.

134

0

2

4

6

8

10

Refinancing& asset

purchases

Assetpurchases

Assetpurchases

govermt

Extendedrefinancing

Extendedrefinancing

(λ=0.75)

LCR after 2nd round LCR 5% tail LCR 1% tail

Figure 7.4. Central Bank interventions, mitigating influencePercentage point differences compared to outcomes excluding Central

Bank interventions (credit crisis scenario), q n =0.35 and q CB =0.30

0

1

2

3

Extendedrefinancing,

pre-exit

Exit:wholesale

funding

Exit: retailfunding

LCR after 2nd round LCR 5% tail LCR 1% tail

Figure 7.5. Impact of exit from extended Central Bank refinancing Percentage point difference compared to outcomes excluding Central Bank interventions (credit crisis scenario)

The outcome that asset purchase programs are more effective in limiting second round effects than

extended refinancing operations can also be explained by the reactions of banks after the first round of

the scenario. Based on the regulatory incentive in the reaction rule, banks increase their bond holdings

(i.e. liquid assets with a low RSF factor as reflected in wNSFR_ST ) and reduce their reliance on central

bank funding (i.e. liabilities with a low ASF factor as reflected in wNSFR_ST ). This is an intended effect

of the proposed Basel liquidity regulation, which aims at strengthening the capacity of banks to

withstand shocks independently from the central bank. The model reflects this feature in the reaction

rule. As a consequence, asset purchases would get more influence on the adjusted balance sheets than

refinancing operations. This is further illustrated by simulating the credit crisis scenario with a lower

weight of the regulatory component in the reaction rule (λ = 0.75 in stead of λ = 0.5). This implies that

banks’ responses to the scenario are more based upon ‘within exposure’ adjustments (St2 dominates the

reaction rule) and less on regulatory incentives (Rt2). Under that assumption, extended refinancing

operations are more effective than asset purchases. Central bank financing lifts the average LCRt4 6

percentage points compared to the scenario excluding interventions (see Figure 7.4). This result

illustrates the interaction of Basel III with monetary policy. Banks that do not fully adapt to Basel III,

by having less stable funding and fewer liquid assets than they would have when following the

regulation, are more dependent on central bank borrowing in times of stress. The flip side is that banks

with high liquid asset holdings are more dependent on developments in bond markets. For monetary

policy this implies that central banks increasingly may have to resort to asset purchases in stead of

refinancing operations to influence banks’ liquidity positions.

The simulation of the exit from extended central bank refinancing operations shows that the

consequences of an exit for funding liquidity risk depend on the way in which banks substitute central

bank borrowings with alternative funding sources. If banks would increase wholesale funding again by

issuing unsecured debt securities, the average LCR falls back to its pre-intervention level. However, if

135

banks are able to compensate the reduced central bank borrowing with increased retail savings, the

LCR actually improves compared to the pre-exit situation (see Figure 7.5). This reflects that retail

savings are more stable than central bank funding, as reflected in the LCR.

7.3.5 Effects of adjusting to the new liquidity regulation

The new liquidity regulation stimulates that banks increase their liquid buffers (through the LCR) and

reduce their funding mismatch (through the NSFR). This should make the sector more resilient to

adverse shocks to market and funding liquidity. It will reduce the risk of behavioural responses by

banks and related disturbing effects on markets. Moreover, moral hazard risks of banks counting on

the central bank are also reduced. To simulate the benefits of the regulatory incentives, we run the

credit crisis scenario with the presupposition that banks had adjusted their liquidity position before the

scenario in line with Basel III. First, we assume that banks hold more liquid assets (+25%), while other

assets, for instance loans, are proportionally lower. The second assumption is that banks substitute

level 2 liquid assets (corporate bonds, covered bonds) with high-quality government bonds. This gives

an indication of the difference that a narrow buffer definition – which only allows for high quality

assets – can make compared to a broad definition, in terms of stress resilience. Thirdly, we assume that

banks partly substitute wholesale funding for retail savings, which are a more stable funding source

with a lower run-off rate. In the fourth case, we assume that banks lengthened the maturity of their

wholesale funding, by assuming that debt securities maturing within one year were partly replaced by

longer term securities. In the fifth case, banks to some extent substitute unsecured with secured

borrowing, by reducing interbank deposits and increasing repos. Borrowing in the unsecured market

has a negative effect on the LCR through a 100% run-off rate, while repos with high quality collateral

have a 0% run-off rate.60

60 The assumptions are applied by changing the elements in vectors ILCR,t0, INSFR_ST,t0 and INSFR_LT,t0. The first assumption implies that liquid asset holdings - the numerator in the LCR - are 25% higher and that this amount proportionally reduces other assets. The second assumption implies that high quality government bond holdings are 25% higher, while level 2 bonds are reduced by the same amount. Under the third condition, debt securities issued are 25% lower, while retail savings are increased by the same amount. The fourth assumption implies that debt securities maturing within one year are reduced by 25% and longer term securities increased by the same amount. The fifth condition implies that unsecured funding is reduced by 25% and secured borrowing increased by the same amount.

136

0

2

4

6

8

10

12

Liquidasset ↑

Qualityliquid

asset↑

retail ↑whole-sale ↓

Longwhole-sale ↑

secur ↑unsec↓

LCR after 1st round LCR after 2nd round LCR 5% tail LCR 1% tail

Figure 7.6. Influence of stronger liquidity profilesPercentage point difference compared to outcomes with initial liquidity profiles

Figure 7.6 shows that the stress scenario has less impact on banks if liquidity profiles are stronger in

the initial situation. The scenario impact would be most mitigated if banks improve their stock of

liquid asset holdings. In case banks have 25% more liquid assets, the LCR after the first round of the

scenario is over 12 percentage points higher on average, while a higher quality of liquid assets is most

effective to limit the tail risks. Substituting riskier bonds (for instance, corporate bonds) with high

quality bond holdings, limits the price volatility of bond portfolios and thereby the probability of

extreme losses. It indicates that a narrowly defined liquidity buffer, which limits the numerator of the

LCR to high quality sovereign bonds, makes a big difference in containing systemic risk, compared to

a broadly defined buffer (the assumption in the baseline). Stronger liquid asset buffers contribute more

to stress resilience than substituting wholesale for retail funding, unsecured for secured funding or

prolonging the maturity of wholesale funding (see Figure 7.6). The main reason for this is that a higher

level and quality of liquidity buffers enhance the capacity to absorb the first round effects of a stress

scenario, which reduces the risk of collective reactions and related disturbing market effects. The

outcomes illustrate that the incentives in Basel III, to enhance liquidity buffers through the LCR, are

an important contribution to reduce systemic risk.

7.4 Conclusions

The Liquidity Stress-Tester model is an instrument to simulate the impact of shocks to market and

funding liquidity on banks. It takes into account the important drivers of liquidity stress, i.e. on and off

balance sheet contingencies, feedback effects induced by collective reactions of heterogeneous banks

and idiosyncratic reputation effects. Contagion results from the effects of balance sheet adjustments on

prices and volumes in the markets where the banks are exposed to. The model contributes to

137

understand the influence of reactions by banks, which are endogenously driven by liquidity shocks and

liquidity regulation. It also allows for analysing the impact of liquidity stress and the reactions by

banks on credit supply. The impact of liquidity stress scenarios on lending seems limited, with the

contraction in long-term loans exceeding that of short-term loans.

The model explores the interactions with central bank interventions, by including a central

bank reaction function in the model. This measures the influence of extended refinancing operations

and asset purchase programs on banks’ liquidity positions, as well as the interaction with liquidity

regulation. One result is that second round effects and tail risks of a stress scenario are substantially

lower if banks would adjust to Basel III, by holding a higher stock of liquid assets. In particular a

narrowly defined liquidity buffer (consisting of high-quality government bonds) makes a big

difference in limiting the tail risks of banks. The flip side of larger liquid bond holdings is that

monetary policy conducted through asset purchases gets more influence on banks relative to extended

refinancing operations. Simulations of the phasing-out of the extended refinancing operations indicate

that the consequences for funding liquidity risk depend on the nature of the alternative funding sources

that are available to substitute for central bank borrowing.

The model outcomes lend support to policy initiatives to enhance the liquidity buffers and

reduce liquidity mismatches of banks, as proposed by the Basel Committee (BCBS, 2009b, 2010b). A

sufficient level of high quality liquid assets limits the idiosyncratic risks to a bank, by providing

counterbalancing funding capacity to weather a liquidity crisis. Moreover, stronger liquidity profiles

are important to reduce the risk of collective reactions by banks and thereby to prevent second round

effects and instability of the financial system as a whole. The model provides a tool to evaluate the

effects of behavioural changes on banks’ liquidity positions and systemic liquidity risk under different

scenarios. While the model is applied to Dutch banks in this chapter, it is applicable to other countries’

banking systems as well, since it is based on balance sheet data and parameters that are, or will

become, commonly available as a result of Basel III.

138

Appendix 7.1

Source: BCBS, 2009b

139

140

141

Appendix 7.2

Source: author’s calculations

142

143

Chapter 8

Crisis measures and limiting possible distortions

8.1 Introduction

This chapter analyses the financial crisis measures taken between 2007 and 2009 by central banks and

governments.61 The crisis, which started in 2007 as a problem in the US mortgage market and

subsequently spread around the world, severely affected financial markets and institutions.

Evaporating liquidity, falling asset prices, excessive debt positions and soaring losses at financial

institutions reinforced each other. These factors undermined confidence within the financial sector and

disrupted the functioning of various markets simultaneously, affecting the system’s core: the interbank

market. Even solid financial institutions became vulnerable in the autumn of 2008 as market

confidence collapsed. The creditworthiness of some countries with large primary borrowing

requirements came under pressure as well. To restore confidence and safeguard system stability,

governments and central banks instigated far-reaching interventions (see Table 8.1 for a brief

overview of the size and nature of the interventions; a more complete overview is provided by Stolz

and Wedow, 2010).

This episode underlined that financial stability cannot be taken for granted. Information

asymmetries and wrong incentives gave rise to financial imbalances that emerged during the crisis.

Market failures occurred, forcing central banks and governments to intervene in order to preserve

financial stability. In doing so, they also supported the economy, which depends on a well-functioning

financial system. Of course, there is a risk that interference with the operation of market forces – even

in times of crisis – produces distortionary effects that may cause inefficiencies and possible

imbalances in the long run. This chapter analyses the various distortions arising from interventions in

the financial sector and examines whether they actually occurred in the 2007-2009 period. Where

possible, such assessment is based on empirical evidence. Central banks and governments were well

aware of possible distortionary effects when they took extraordinary measures during the crisis, and

sought to limit these effects as much as possible in putting together their support packages. We discuss

the conditions for support, and review the various policy options of central banks and governments to

limit distortionary effects.

Market failure and how the authorities responded to it are briefly discussed in Section 8.2. In

Section 8.3 it is emphasised that a proper design of support policies is essential but complicated. This

is followed by an analysis of possible distortionary effects of public interventions in the financial

sector, in particular on market conditions (Section 8.4) and on investor confidence in financial

61 This chapter is a revised version of Van den End, Verkaart and Van Dijkhuizen (2009).

144

institutions and support-providing authorities (Section 8.5). In the longer term, interventions by

governments and central banks may lead to excessive risk taking and thus create ‘moral hazard’

among market participants and other stakeholders, as discussed in Section 8.6. Finally, Section 8.7

discusses the policy instruments that may limit the onset of possible distortionary effects.

Table 8.1. Government support to banks and central banks' balance sheetsMeasures taken in 2007-2009 crisis. EUR bn, unless stated otherwise.

US Euro area UKGovernment support banks¹ Capital injected 369 187 83

Asset relief 385 407 385 Debt guaranteed 274 498 152

Total in % GDP 9.3 16.6 36.6

Central bank balance sheet expansion2

(USD / EUR / GBP bn) Securities purchases 818 18 172

Open market operations & 462 243 18 special credit facilities

Total in % GDP 8.8 4.0 13.0

1 Euro area: sum of support in DE, FR, ES, NL, BE and IE. Position as at mid-2010.2 Increase of assets between July 2007 - October 2009.

Sources: Bloomberg, BIS, ECB, BoE, Fed and DNB-calculations.

8.2 Crisis measures to address market failure

The support measures initiated during the crisis have overcome market failures in different ways

(Table 8.2). Limited insight into risks and heightened uncertainty made banks reluctant to lend to each

other. As a result, the interbank market became gridlocked. Therefore, central banks worldwide

extended their refinancing operations by providing more short-term loans to banks, by easing the

collateral requirements and, where necessary, by introducing new liquidity facilities (IMF, 2010a). For

example, the Eurosystem and the Bank of Japan fully allotted the bids submitted by banks at a fixed

interest rate in their liquidity operations. The Eurosystem also introduced a new longer-term liquidity

facility, allowing banks to refinance for a one-year period (Stark, 2009). Furthermore, central banks

sharply cut their policy rates and purchased debt securities, in order to support lending to companies

and households.

The crisis of confidence led to an increased risk of customers withdrawing their deposits from

banks. In response to the increased risk of bank runs, most governments have extended their deposit

insurance schemes in autumn 2008. They raised the amount of guaranteed deposits, abolished

145

depositors’ own risk and – in some instances – provided a blanket guarantee on all deposits (Germany)

or even all banking liabilities (Ireland). Also, because of the evaporation of liquidity in financial

markets, banks had difficulty raising long-term funding. As a result, their typical maturity mismatch

between assets and liabilities increased rapidly and made banks extremely vulnerable to financial

market shocks. To limit this risk, governments introduced guarantee schemes on bank debt in the final

quarter of 2008, which supported the issuance of medium-term debt securities. At that time, sharply

deteriorating market sentiment gave rise to doubts about the solidity of financial institutions. This

uncertainty was fuelled by the downward spiral of assets prices and declining mark-to-market values

of bank assets. Most financial institutions thereby lost access to the equity market, although there was

a great need for new capital to replenish substantial write-downs on loan portfolios and to remove

uncertainty over the adequacy of their capital buffers. In this environment, governments had to step in

providing capital support from the autumn of 2008. For instance, in the Netherlands ING, Aegon and

SNSReaal received capital support (for an overview of national rescue measures see ECB, 2009a). In

the UK, capital support was provided, for instance, by guaranteeing equity issues made by banks (with

the government buying shares in RBS and Lloyds as private investors kept to the sidelines). The

conduct of macro stress-tests by the authorities in Europe, the US and the UK made an important

contribution to improving the solvency of banks and supporting market confidence (see Chapter 4).

Information asymmetries and heightened uncertainty had their strongest impact on securitised

assets. The evaporation of liquidity in the markets for these structured products severely distorted their

pricing. As a result, they became illiquid or could not be realistically valued. This caused great

uncertainty over the solidity of financial institutions having such assets on their balance sheets. To

remove this uncertainty, some governments have taken over the risks of toxic assets by introducing

guarantee schemes, asset swaps (e.g. with regard to ING’s US Alt-A mortgage portfolio) or bad bank

schemes from early 2009. The advantage of a bad bank is that it clears a bank’s balance sheet from

toxic assets, allowing it to focus on its actual banking business again.

The crisis measures were ultimately aimed at preserving the intermediation function of the

banking sector. The support of banks’ funding liquidity and capital positions had to prevent a credit

crunch and a worsening of the economic situation. Research on the economic impact of the crisis

measures shows that they indeed have been effective. The cross-country study by Laeven and

Valencia (2011) finds that the recapitalizations of banks, concluded between September 2008 and

March 2009, had a significant positive effect on the growth performance of credit dependent firms.

The other crisis measures (guarantees, asset purchases and liquidity support), were found to have no

significant effect individually, but all these measures together turned out to have a joint significantly

positive effect on the performance of credit dependent firms.

146

8.3 Proper design of support policies essential but complicated

In crisis management, the design of government support schemes is essential in order to mitigate the

risk of new distortions (Table 8.3). The authorities were aware of this when they put together their

extraordinary measures. Determining the most appropriate design is rendered more difficult, however,

because governments and central banks are faced with uncertainties in the middle of a crisis. First,

with their unconventional measures, central banks stepped out of their traditional ‘comfort zone’: the

interest rate pass-through to the economy changed under the influence of the crisis, whilst liquidity

injections led to a liquidity oversupply in the money market, causing a change in the way in which

central banks steer short-term interest rates (Heider et al., 2009). Furthermore, new instruments were

wielded, in particular by the Federal Reserve (Fed), the effectiveness of which was not clear in

advance (Sarkar, 2009). Second, the distinction between functioning and non-functioning market

segments was not evident, which made it difficult to choose the proper form of intervention. The

147

system went through phases of heightened stress and recovery with highly volatile market prices. That

is why central banks extended liquidity in the money market as a whole and why governments aimed

their support in first instance at direct capital support instead of specific toxic asset solutions. A factor

at play here was that, in a crisis, the distinction between solvent and insolvent institutions is not

always clear. Such an assessment is complicated by the high degree of uncertainty over balance sheet

positions, which involves the risk that support is given to institutions that will ultimately prove not to

be solvent. Third, in a crisis it is difficult to determine the proper conditions for support. Volatile

market prices are not a proper yardstick in this respect. Besides, blueprints for specific solutions are

mostly not available, whereas a crisis calls for immediate and direct action. All of this makes it

conceivable that official interventions entail distortionary effects. The following sections contain a

detailed discussion of a range of possible distortionary effects.

Table 8.3. Distortions and mitigating instruments

Distortionary effects Mitigating instruments

Market conditions Market-compatible and harmonised conditions for support - uneven playing field - adequate support conditions (price, instrument, governance) - distortion of international capital flows - harmonisation of national programmes - financial protectionism - no territorial discrimination

- equal treatment of foreign subsidiaries/branches - crowding out non-supported markets - purchase of debt securities at market prices

External effects / confidence Authorities to supplement market forces and to act clearly - uncertainty over outcome of support - clarity on details of support policies - limited market access for institutions - clarity on position of private financiers - uncertainty over public influence on firms - government at arm’s length from business management - government creditworthiness - budgetary consolidation, multilateral initiatives - central bank independence - limiting financial risks or ex-ante government guarantees

Moral hazard Disciplining mechanisms - stakeholders (management, - private sector involvement share, bond and deposit holders) - temporary nature of support, incentives for timely and smooth exit

- prudential supervision - search for yield - timely increase of interest rate

short-term

long-term

8.4 Market conditions

8.4.1 Impact on level playing field financial sector

Although government support is provided as much as possible on market-compatible conditions to

preserve a level playing field (see also Section 8.7), such support may nevertheless distort competition

between financial institutions. Beck et al. (2010) distinguishes two ways in which state aid can have

negative repercussions for competition. First, it can reduce the private marginal costs of banking

activities below their true social costs. Second, it can encourage socially undesirable and excessive

risk taking. Both ways create inefficiencies in the banking sector with regard to the allocation of

resources by banks and across banks. Due to the distortionary effects on competition, it is conceivable

148

that non-supported institutions also apply for support. This would unintentionally frustrate the

functioning of market forces. Competitive conditions could be distorted further if foreign subsidiaries

or branches are not provided with the same measure of support as domestic institutions. For instance,

in the US, foreign financial institutions were not eligible for all US support programmes, which put

them at a competitive disadvantage.

In the retail savings market, the level playing field could be distorted by state supported

institutions that offer relatively high deposit rates and benefit from a more stable image. In addition,

extension of the maximum coverage under deposit insurance schemes may influence competitive

conditions in the banking sector. Figure 8.1 suggests that state support of individual Dutch banks did

not give rise to seriously distorted market conditions. Market shares in the retail savings market

changed only little on balance, with government interventions apparently helping to stabilise the

decline in saving deposits at a number of institutions during the stressful market conditions in the

autumn of 2008. Increased competition for deposits primarily resulted from substitution effects, due to

banks (supported as well as non-supported) trying to substitute wholesale funding with more stable

retail deposits. Extended deposit insurance coverage facilitated this. As a result, some banks

aggressively offered high deposit rates to the public, and banks with a relatively high risk profile were

able to attract deposits at high interest rates. It is conceivable, however, that a bank’s growing demand

for retail funding does not fit in with its business model, and extended deposit insurance coverage may

therefore lead to inefficiencies in the financial sector (DNB, 2010a). The same risk is associated with

central banks’ more accommodative money market policies, as banks benefit from high amounts of

relatively cheap funding and therefore have fewer incentives to adjust their business model.

40

45

50

55

60

Jan 08 May 08 Sep 08 Jan 09 May 09 Sep 09

Supported banks Non-supported banks

Figure 8.1. Market shares Dutch deposit market¹Percentage of outstanding amount

1) Inte res t ra te bearing depo s its o f ho us eho lds and firms . So urce : DNB.

149

8.4.2 Distortionary effects on markets and business models

Interventions by governments and central banks may also affect the relative performance of markets.

In market segments that receive official support, financial intermediation is likely to recover more

quickly, because it enhances the willingness of investors to take exposures. After all, state support

ensures a certain degree of liquidity and/or guarantees a certain price level. This could impact

negatively on non-supported market segments as investors withdraw from them. This mechanism was

visible in various market segments during the 2007-2009 crisis episode. In the UK, funding guarantees

crowded out non-guaranteed debt, while in the euro area, the issues of both asset classes were more on

a par (Panetta et al., 2009). The US markets for commercial paper and mortgage-backed securities also

picked up thanks to the Fed’s asset purchase programmes. Negative effects could be seen in the US

markets for car and credit card loans and for commercial real estate, which initially received no

support and where capital was withdrawn (IMF, 2009a). The Eurosystem’s covered bond purchase

programme, which the IMF deemed to be effective (IMF, 2009b), also had potential side effects.

Purchases were carried out in a specific market segment, thereby influencing relative prices between

market segments. It appeared to be difficult to distribute the purchases neutrally over institutions and

countries, since in some euro area countries, the covered bond market is more developed than in

others, and some institutions issue more covered bonds than others. Also, the conduct and financing

structure of banks may unwittingly be influenced, because the covered bond purchases render just one

of the many funding opportunities for banks more attractive. All in all, full neutrality was not always

feasible in asset purchase programmes, which could cause distortions at the micro level.

The mechanism whereby recovery of supported markets could go at the expense of non-

supported markets also applies to the easing of collateral requirements for central banks’ liquidity

operations. Assets that are added to the list of eligible collateral are deemed to have a higher liquidity

and thus an eligibility premium vis-à-vis debt securities that are ineligible. Such a premium will be

high in particular for illiquid assets such as bank loans, even though this is difficult to quantify

because of the absence of a market price (ECB, 2007). This distortion of market prices may crowd out

ineligible assets.

Due in part to full or high allotment in their liquidity operations since October 2008, central

banks fully or partially took over the lending and borrowing activities in the interbank money market.

Additionally, low money market rates made banks less willing to lend to each other. Consequently,

volumes in the euro area interbank overnight market (EONIA) fell to very low levels (Figure 8.2). To

mitigate the risk that central bank operations are distortionary, the crisis measures should be phased

out as soon as markets become self-relient. A complication is that the market functioning to some

extent is endogenous with regard to the central bank’s policy stance. If market participants expect a

continuation of the extended liquidity supply by the central bank they have little incentive to adjust

their business model or trade with each other. To prevent this, it should be clear in advance that

150

support is only provided temporarily and as much as possible on conditions that should ensure that

support becomes unattractive as soon as the market recovers (see Section 8.7).

An accommodative monetary policy and low interest rates may also distort the functioning of

financial institutions. It puts pressure on the business model of money market funds, as these invest

exclusively in short-term debt securities (IMF, 2010b). If money market rates are very low, the yields

on these short-term debt securities barely exceed the costs of money market funds, making it no longer

interesting for investors to invest in them. In the US, where money market funds invest heavily in

Commercial Paper (CP), problems at money market funds also put pressure on the CP market. To

counter such pressure, the Fed introduced asset purchase programmes to support both the CP market

and money market funds.

Extremely low short-term interest rates can also distort the functioning of the repo market. In

this market, parties lend each other financial securities for short periods of time, against payment of a

money market rate. With money market rates at a very low level, little if any costs are involved if the

securities are not delivered on time (ICMA ERC, 2010). As a result, there was an increasing incidence

of failed repo settlements, and parties stopped lending securities to each other in the last quarter of

2008. To support the repo market, a penalty for non-delivery of the securities was instituted in the US.

This led to negative interest rates on the repo market, so that the party who lends the securities in

exchange for cash needs to repay less if the other party delivers late.

The low short-term interest rates also led to declining long-term interest rates from mid-2007,

partly as a result of the asset purchase programmes of central banks. The decline of interest rate led to

a deterioration of the balance sheets of pension funds and insurance corporations. Not only did their

investments generate less income, the present value of pension funds’ liabilities also increased and, in

the event of a longer period of low interest rates, the public may be less interested in the long-term

savings products that are offered by insurance companies and pension funds.

Finally, if long-term yields would decline further, with short-term rates having hardly any

scope to come down any further, the yield curve could flatten. This could put banks’ business model

under pressure, as banks are using short-term funding to finance their longer-term lending operations.

Actually, in 2007-2009 the yield curve steepened, because short-term (policy-driven) interest rates had

fallen much more than long-term rates.

151

10000

30000

50000

70000

90000

110000

jun-07 dec-07 jun-08 dec-08 jun-09

0

1

2

3

4

5

Volume 20 days moving average EONIA rate (rh-axis)

Figure 8.2. Money market rate and trading volume (euro area)Daily data; mln euro

Source: ECB.

8.4.3 Cross-border effects

Support measures may also lead to a redirection of cross-border financial flows, with capital flowing

out of markets where no government guarantees or asset purchase programmes exist. In 2008-2009 for

instance, emerging countries had to compete for financing with state-guaranteed debtors in industrial

countries. It was feared that the capital inflows to emerging countries would decline strongly, due in

part to crowding-out effects of government support in developed countries, besides overall risk

aversion (Financial Times, 2009). Such cross-border effects may lead to higher volatility in financial

markets and undermine their integration. In the course of 2009, capital flows into emerging markets

picked up again, in an environment of increased risk tolerance in financial markets. In some emerging

countries, strong capital inflows even caused abundant domestic liquidity, credit expansion and rising

asset prices.

Country-specific differences in support conditions may also generate capital flows. National

interests to support the economy may translate into targets for domestic lending. In France, the

Netherlands and the UK, for instance, state-supported banks committed themselves to keep up

domestic lending. Such conditions reinforce the home bias of financial institutions and threatens to

distort the internal market in Europe, at the expense of an efficient international allocation of credit

and economic growth. Macro figures from the Bank for International Settlements (BIS) confirm that

cross-border lending declined strongly. At the height of the crisis – in the fourth quarter of 2008 –

global lending fell by more than 5% and even by 12% compared with banks in emerging countries

(BIS, 2009a). In 2009, this type of financing remained under pressure, due in part to the deleveraging

process of banks in developed countries. Hoggarth et al. (2010) conclude on the basis of BIS data and

information from market participants that the reversal in cross-border credit flows was concentrated in

banks’ ‘non-core’ markets. This may be driven by strategic choices, made by banks as part of an

152

overall balance sheet restructuring and risk mitigation package. Research by De Haas and Van Horen

(2011) shows that the sensitivity of cross-border lending to shocks cannot be generalised, as they find

that during the crisis banks continued to lend more to countries that are geographically close, where

they are integrated into a network of domestic co-lenders, and where they had gained experience by

building relationships with (repeat) borrowers.

In addition, central banks’ asset purchase programmes and easier collateral requirements,

aimed by definition at their own currency areas, could also have contributed to a withdrawal of capital

from non-supported foreign financial markets. Finally, internal market distortions are possible as a

result of crisis-driven prudential supervisory requirements that are imposed nationally, such as more

stringent requirements for the liquidity management of foreign subsidiaries by the host country

supervisor, or the unilateral imposition of higher capital ratio requirements. To prevent a ‘race to the

top’, such regulations should preferably set up in an international framework, i.e. Basel III (see

Chapter 9).

8.5 External effects, negative impact on confidence

8.5.1 Investor confidence in supported institutions

Government support may damage confidence in financial markets if it is accompanied by uncertainty

over the implementation and duration of the support measures. For example, investors may be

uncertain about the government influence on the management of a supported institution. Or they may

be deterred by the prospect of profit dilution, resulting from the state’s generally preferential stake in

the institution’s capital. This may give rise to a negative spiral of successive provision of support and

withdrawal of private capital, which can, at worst, make full government control necessary.

This effect was reflected in the share prices of a group of 38 large financial institutions from

nine countries, of which 24 institutions received government capital injections. Following a

temporarily favourable impact of government support in October and November 2008, on average the

share prices of supported institutions showed a more unfavourable development than those of non-

supported institutions (see Figure 8.3). The event study by King (2009), based on share prices of 52

large banks in six countries that received capital injections, emergency loans or asset support, shows a

similar result. He suggests this indicates that government support was viewed by investors as a

negative signal of a bank’s health.

The positive effect on CDS premiums lasted longer than on share prices, consistent with the

aim of governments to limit the risk of default to the benefit of creditors and CDS protection sellers (at

the expense of shareholders). As appears from the CDS premiums of the same set of 38 institutions,

the positive influence of government support faded gradually in the course of 2009 (see Figure 8.4). In

June 2009, the cumulative difference between the risk premiums of supported versus non-supported

153

institutions became negative. This suggests that, nine months after the first government injections,

bond holders as well begun to regard government influence as a relative drawback, possibly due to the

reduced financial flexibility of government supported institutions. The study by Panetta et al. (2009),

on a different data set, confirms the initial positive influence on CDS premiums of supported

institutions. Moreover they show that larger support programmes were associated with bigger

reductions in CDS premia.

Hybrid loans (subordinated debt) were also adversely affected by government interventions.

These loans were subject to sizeable declines of market value as investors feared that government

ownership stakes would be at the expense of subordinated debt (in 2009, Standard & Poors

downgraded its ratings for hybrid loans issued by state-supported banks). Investors increasingly

realised that these loans are, in fact, risk-bearing capital. The fact that a number of financial

institutions, in line with contractual terms but against market expectations, defaulted on their

subordinated debt obligations played a role as well.

-50

-40

-30

-20

-10

0

Oct 08 Dec 08 Feb 09 Apr 09 Jun 09

Non-supported Supported

Figure 8.3. Stock price non-supported vs. supported financial institutions, worldwidePercentage of cumulative change stock price (monthly average)

Source: Datastream, own calculations, data of 38 large banks and insurance companies worldwide.

-10

15

40

65

90

Oct 08 Dec 08 Feb 09 Apr 09 Jun 09

Non-supported Supported

Figure 8.4. CDS premium non-supported vs. supported financial institutions, worldwidePercentage of cumulative change stock price (monthly average)

Source: Datastream, own calculations, data of 38 large banks and insurance companies worldwide.

8.5.2 Confidence in the creditworthiness of governments

Another side effect of government support is that it could affect the creditworthiness of support-

providing governments. Partly as a result of large-scale support, such as capital injections into banks,

the sovereign debt of some countries increased rapidly. ‘Bad bank’ schemes, in which the authorities

take over assets from banks, may raise public debts even more, since the amounts needed to take over

assets will be a multiple of capital injections. This explained the reticence on the part of governments

to adopt such a solution in the 2007-2009 crisis. Another reason why bad bank schemes were less

appropriate was the scale and nature of the troubled asset problem in the recent crisis. It concerned the

worldwide banking sector and involved complex financial products. As a consequence, the prospects

for selling the assets in a later stage were poor, also because they were not easy to value. This was

different compared to other banking crises, for instance in Sweden in the 1990s, where a bad bank

154

solution turned out to be successful, because it was less difficult to get a clear view of the scale and

nature of the troubled asset problem.

In assessing the financial position of governments, market participants also weigh any

contingent obligations of governments, e.g. arising from guarantees. Indeed, government interventions

were accompanied by an increase in sovereign risk spreads (in West-European countries by about 50

to, occasionally, more than 100 basis points between October 2008 and mid 2009). The financial

position of governments was increasingly associated with that of the banking sector (and vice versa).

This was evidenced by the decline in the risk spread differential between the banking sector and the

government from October 2008 (when governments began to intervene) to May 2009, when the

market prices of financial institutions had recovered to some extent. The decline in the spread

differential was clearly visible in, e.g., the UK, Belgium, Spain and Ireland (Figure 8.5). This signifies

that the support measures translated indirectly into rising costs for the taxpayer. In order to limit the

negative impact of support on government financing costs, most governments have embarked on

large-scale budgetary consolidation since 2009.

-100 -75 -50 -25 0

UK

BEL

SP

IRE

FR

DE

IT

NL

AUT

Figure 8.5. Dependence between banks and governmentsChange of difference between average CDS spread large banks in country and CDS spread government between May 2009 and September 2008, basis points

Source: ow n calculations based on data f rom Datastream.

8.5.3 Risks for the central bank

Intervention by central banks in financial markets and banking systems can weaken their balance

sheets. The involvement of the central bank in crisis management may also reduce their independence

vis-à-vis the government. This is illustrated by the finding of Artha and De Haan (2011) that banking

crises significantly increase the likelihood of a central bank governor turnover. In the 2007-2009 crisis

the risk profile of central bank balance sheets was affected in various ways. The provision of much

larger quantities of liquidity for longer periods of time (in the case of the Eurosystem, up to one year)

155

exposed the central bank more than before to credit risk, even though it is secured by collateral. The

easing of collateral requirements in combination with the diminished liquidity of all but the safest

government bonds added to the risk burden of central banks. For the Eurosystem this is illustrated by

the fact that the collateral pledged by banks in 2009 consisted for 12% of (relatively risk-free)

sovereign debt, as against 21% in 2006 (Table 8.4). Moreover, the purchase of private sector bonds led

to increased financial risks on central bank balance sheets. High-quality sovereign debt, also

purchased by several central banks, carries fewer financial risks, but the risk of political interference

remains, as the funding conditions of governments can become dependent on the central bank

interventions. Partly on that account, in first instance the Eurosystem did not buy sovereign debt, but

only covered bonds issued by banks (this strategy was changed in May 2010 with the introduction of

the Securities Market Program, see González-Páramo, 2010).

Table 8.4. Composition collateral pledged at EurosystemPercentage of total collateral pledged, averages per year

2006 2007 2008 2008 2009

Government bonds 21 15 10 9 12Bonds issued by banks 31 32 28 27 27

Asset-backed securities (ABS) 11 16 28 30 23Covered bonds 18 14 11 11 13Non-tradeable assets 4 10 12 12 15

Explanation: selected categories; does not sum up to 100%.Source : ECB

8.6 Longer-term distortions

Distortionary effects from support measures in the longer term could arise in particular from excessive

risk taking (‘moral hazard’). Crisis measures may result in moral hazard among the various

stakeholders of supported institutions: management, shareholders, bondholders and depositors. Moral

hazard may also be created by extremely low interest rates. In designing their support measures, the

authorities have sought to limit possible distortionary effects as much as possible, although challenges

remained.

8.6.1 Moral hazard among management

A lesson learned from the financial crisis is that variable remuneration structures and short-termism on

the part of shareholders may create the wrong incentives for the management (FSF, 2008). It can entail

unjustifiable credit, market and reputation risk. Institutions that threatened to come into the danger

zone because of this, in some instances called for government support, with consequences for sitting

board members. However, government interventions may stimulate excessive risk taking as well, as

156

the disciplining influence of the market diminishes (in contrast to that, concentrated ownership by

private shareholders may even reduce a bank’s risk profile, especially at low levels of supervisory

control and shareholders protection rights, Shehzad et al., 2010). As soon as the government acquires

an ownership stake, private shareholders lose some of their influence. The incentive to monitor

institutions fully disappears in the event of nationalisation. If an institution ceases to be quoted on the

stock exchange, the market signal of it disappears. Market discipline also diminishes in the case of

guarantees on bank funding. The related premium for the credit risk of banks is dictated by the

government and no longer by the market. In schemes for toxic assets, moral hazard may arise from

information asymmetries. The management has more information on the quality of the assets than the

state (which has taken over the risks of the assets), but has fewer incentives for a sound management

of the assets. Market discipline also diminishes by the central banks’ virtually unlimited provision of

liquidity at a fixed low interest rate. After all, all banks can obtain central bank liquidity at the same

interest rate, whereas in the private market, banks with more risky operations would have to pay more

for liquidity. This may diminish the incentive of weak banks to clean up their balance sheets.

8.6.2 Moral hazard among investors

Government support may encourage institutions to take on excessive risk, with the upward potential

benefiting shareholders and the downward risks being for the government’s account (‘risk shifting

behaviour’). The incentive for risk taking exists as long as the expected proceeds exceed the costs of

support. At the height of a crisis, when institutions are highly risk averse and private financiers remain

on the sideline, moral hazard is not the government’s prime concern. Shareholders of financial

institutions would then shy away: being the risk-bearing parties, they would suffer most (between the

onset of the crisis and early 2009, on average three-quarters of the market value of financial

institutions had evaporated worldwide). At that stage, governments treated shareholders and

bondholders with gloves so as not to cut off financial institutions from the capital market.

Governments were the most reticent vis-à-vis bondholders, as the losses they sustained from the

bankruptcy of Lehman Brothers and the restructuring of Washington Mutual in 2008 had strongly

intensified market turbulence and cut off systemic banks from liquidity (Brunnermeier, 2008). By

sparing bondholders in their support actions, governments implicitly took over the downward risk

from them. This may create moral hazard as debt financiers lack the incentive to carry out a thorough

risk assessment of financial institutions. It gives them the opportunity of free riding on the

government. Therefore, the governments should take measures that would reduce such moral hazard

among shareholders and bondholders, by providing greater clarity ex ante about their rights, see

Section 8.7.3.

157

8.6.3 Moral hazard from deposit insurance

Research shows that extensive deposit insurance coverage is a source of moral hazard, encouraging

financial institutions to enter into risky exposures (Demirgüç-Kunt and Detragiache, 2000). The

reasoning behind this is that it gives depositors fewer incentives to monitor banks, assuming that the

risk is (fully or partially) borne by the state. Generally, however, retail depositors neither have the

expertise nor the resources to monitor the financial position of banks. In addition, market discipline of

depositors works differently in a crisis. Evaporating confidence in a bank may create a bank run,

enforcing abrupt adjustments.

Extended deposit insurance coverage may create moral hazard among banks, as they engage in

risky exposures assuming that the downward risk for their retail funding will be covered by the state

via the deposit insurance scheme. This may preserve already weakened business models, e.g. in the

case of investment banks that try to continue their business by attracting guaranteed savings deposits

at a high interest rate. To realise a positive interest rate spread, they may engage in risky exposures.

The resulting risks to stability appear from a recent study, which shows that a high deposit rate is

correlated with a high probability of distress among banks (Čihák and Poghosyan, 2009).

8.6.4 Moral hazard from extremely low interest rates

Very low (short-term) interest rates in general encourage a search for yield and lead to risk taking by

market participants (BIS, 2009c). Low interest rates may also stimulate the demand for credit and

boost the economy. These factors may create new financial imbalances, certainly if market

participants assume that central banks and governments will relieve the pain once the bubble bursts.

But even if the prospect of new financial excesses is subdued after a crisis, higher interest rates could

be necessary to further unwind the financial imbalances that caused the crisis. In a low interest rate

environment, weak debtors can be evergreened by banks who continue to roll over loans to insolvent

companies, as was the case with regard to zombie firms in Japan in the 1990s (BIS, 2010a). To

prevent such distortions, it is important not to take accommodating policy measures too far and ensure

that accommodative policies are unwound in a timely fashion (Agur and Demertzis, 2010).

In addition, a longer period of low interest rates may incite governments to take on more debt

than is good for them. Japan, for example, where interest rates have been extremely low for more than

a decade, has meanwhile built up a debt-to-GDP ratio of 180%, the highest of the developed world. In

the debt build-up phase, this may lead to a misallocation of funds. Also, the government budget will

thereupon become more sensitive to a rise in (long-term) interest rates as soon as the economy

recovers. To limit this risk, the government’s consolidation process should move in parallel with

economic recovery. Finally, governments with a high debt-to-GDP ratio have less room to absorb

future setbacks, which is another reason to pursue prudent fiscal policies after a crisis.

158

8.7 Policy instruments to limit distortions

8.7.1 Market-compatible conditions

To maintain a level playing field in the financial sector, state support should be based on market-

compatible conditions (IMF, 2009c). Whilst these conditions should not frustrate the effectiveness of

support schemes, they should also discourage improper use of them. One way to do this is to base the

pricing of support on an institution’s longer-term risk profile plus a surcharge, which would render the

support prohibitive under normal circumstances, but not in a crisis. Also, additional conditions may be

imposed on directors of state-supported institutions, such as remuneration restrictions or dismissal in

case of poor performance. Such conditions reduce the risk that institutions which do not need state

support would be at a disadvantage (and competitive conditions would be distorted) or would also use

state support (and so frustrate the functioning of markets). Adherence to market compatibility is the

basic principle of the directives for capital support, funding guarantees and toxic asset solutions, as

instituted by the Eurosystem and the European Commission (during the crisis, market compatibility

has been ensured by setting price conditions that would prevail on normally functioning markets in the

longer term, EC, 2009b). To maintain a level playing field, the Commission also requires that state-

supported institutions that are non-viable or have received a certain measure of state aid, should

implement a restructuring plan (EC, 2009a). Beck el al. (2010) are critical towards enforcing strong

competition rules on supported banks. Since support enables banks to conduct their economic function

and implies a positive externality for its competitors, bank bailouts do not necessarily require

compensation for competitors. Related to this, the authors mention that imposing balance sheet

reductions on banks may affect other sectors in the economy through credit constraints, while

divestments by banks can lead to renewed downward price spirals in asset markets. It underlines that

the specific characteristics of banking need to be taken into account in the design of the conditions for

support.

8.7.2 International harmonisation

European directives contributed towards the international harmonisation of support conditions (EC,

2009b). This was important to prevent disorderly cross-border capital flows and distortion of

competition in the internal market. The directives confine themselves to the more unequivocal

parameters of support programmes such as premium, duration and instrument. This gave countries

some room for manoeuvre, e.g. with respect to the institutional setup of bank funding guarantee

schemes (individual vs. collective) and the structuring of toxic asset solutions. Stipulations against

territorial discrimination in support conditions were not harmonised, which - with the benefit of

hindsight - was an omission. After all, equal provision of support to foreign subsidiaries and branches

can prevent distortion of the international playing field. All of this underlines the importance of

159

monitoring of the cross-border effects of support, as has been done by the European Commission (EC,

2009b).

8.7.3 The importance of providing clarity

To prevent negative confidence effects, the objective and design of support policies should be clear

(OECD, 2009). This provides investors something to hold on to and can prevent a downward spiral of

market prices. Furthermore, the position of private financiers following the provision of state support

should be clear, e.g. by respecting the seniority of bondholders. To provide clarity about the position

of shareholders in case of government intervention, it may be necessary for the government to have

more ex ante instruments at its disposal to curtail shareholder rights (e.g., to restrict voting rights

during future state interventions). This would allow the government to act more rapidly, which would

benefit the credibility of the support measures. On the other hand, the effects of government

interventions on investors’ share holdings are likely to be factored into prices and market participants’

behaviour. Due to this, financial institutions could be faced with higher financing costs.

The credibility of toxic asset solutions benefits from clarity on the valuation of the assets (e.g.

by having them valued by a third, independent party), on the effectiveness of the solution and on a

possible restructuring of the assets. These are lessons from the Scandinavian crisis of the early 1990s

(Honkapohja, 2009). The governments of Norway and Sweden intervened heavily in troubled banks

by a bad asset resolution scheme, government guarantees, capital injections and nationalisations.

Several years after the crisis, the net costs for the taxpayers were estimated to be limited, thanks to

sales of government stakes in banks and sales of bad assets that were restructured by an asset

management company (De Haan et al., 2009).

8.7.4 Relation between government and management

To create as little uncertainty among investors as possible and not to obstruct a commercially viable

business operation, the government should remain at arm’s length from the day-to-day business of

supported financial institutions. Therefore, participation in subordinated loans or preferential non-

voting shares is preferable over holding common stock or over full government control. The

government can also remain at arm’s length by moving its interests to a management company. One

such example is UK Financial Investments, which manages the British government’s shares in RBS,

Lloyds, HSBC, Northern Rock and Bradford & Bingley (UKFI, 2009). This structure limits the

political influence on the institutions and contributes to preserving the value of the investments.

8.7.5 Involvement of the private sector

Moral hazard among directors of supported institutions can be limited by having them to share in the

risks (ECB, 2009b). In toxic asset insurance schemes, this can be done by having the institutions bear

the ‘first loss’ on the assets (e.g. as in the UK and US schemes), or by having the state and the

160

institution share any losses proportionally (e.g., as in the ING Alt-A guarantee scheme, see Ministerie

van Financiën, 2009). This would still give the management an incentive for responsible management

and a proper unwinding of risk assets. Risk sharing is also important in other forms of government

support. To divide the costs among various stakeholders, the European Commission envisaged to pass

on a larger share of the costs of government support to private financiers (EC, 2009a). Last but not

least, moral hazard among directors may be limited by remuneration structures that provide incentives

to pay heed to the long-term position of an institution, and by dismissal in case of poor performance.

8.7.6 Temporary nature of support and exit

In order not to give the wrong incentives to stakeholders of government-supported institutions, it

should be clearly communicated that the support provided is of a temporary nature and is only

intended for exceptional circumstances. Support should be withdrawn as soon as markets are able to

operate under their own steam (IMF, 2009c). This requires flexibility and smooth exit policies in the

support programmes to prevent the safety nets from being cast for too long. If support is provided at

arm’s length (e.g. government guarantees), a more flexible response can be given to changing market

conditions than in the case of a strong involvement on the part of the government (e.g. through

shareholdership). A smooth exit is promoted by an incentive towards accelerated buyout of state

holdings (e.g. by using the government’s progressive profit sharing as an incentive) and/or by a low

premium for repayment. It should be noted that, in buying out the government’s stake, an institution

must continue to comply with supervisory capital requirements. A smooth exit has also been built into

the design of funding guarantees, by setting an end date for guaranteed issues, a maximum maturity

for loans to be guaranteed and a premium rate that makes the guarantees unattractive in the event of

market recovery (ECB, 2008b). To prevent that deposit insurance is relied on too strongly and for too

long, it is preferable not to issue full guarantees but to provide limited coverage only. Moral hazard

may also be diminished by a system with ex ante funding and risk-weighted premia (BCBS and IADI,

2009). In the design of the deposit insurance scheme, the confidence of market participants in banks

should be carefully taken into account.

Also, central bank interventions must be unwound as and when market conditions allow it.

The Fed designed various support facilities in such a manner that they become inoperative as soon as

the market segment in question recovers. The market recovery in the course of 2009, for instance,

made the fee rates for the Term Securities Lending Facility (TSLF) so unattractive that the facility was

de facto no longer used. Also, as regards other central bank measures, the challenge is not so much of

an operational nature – liquidity can be absorbed quite simply and policy rates can be raised – but it is

more a question of the right timing. The unwinding of monetary stimulus and support to the financial

sector should be subject to the soundness of financial institutions, the recovery of financial markets,

macroeconomic developments and the risks to price stability (OECD, 2009). If central banks were to

unwind their policies too early, they could harm economic recovery, but if support measures are

161

phased out too late, it could lead to inflationary risks and permanent behavioural distortions (IMF,

2010a). Apart from that, the exit policies of various support schemes should be harmonised. The IMF

has outlined a possible time path in which, first, liquidity support is unwound, then guarantee schemes

and finally government’s ownership stakes and toxic asset schemes (IMF, 2009c). This should

facilitate a transition that would support financial stability.

8.7.7 Prudential supervision

Prudential supervision may also play a role in reducing moral hazard. Supervisory authorities may be

able to determine whether corporate decisions are based on rational economic grounds or are

influenced by moral hazard. For instance, with regard to the possibility of high deposit rates offered by

state-supported banks and their impact on risk propensity and profitability. A more objective manner

of limiting moral hazard is to prescribe adequate liquidity and solvency buffers. A greater proportion

of the insurance costs of excessive risk taking is thus allocated to the institutions and their

shareholders, reducing systemic risks. This macroprudential angle plays a significant role in raising

the capital requirements for banks, through Basel III (see Chapter 9).

8.8 Conclusions

Central banks and governments conducted unprecedented crisis measures to preserve financial

stability in the 2007-2009 crisis. The measures were taken rapidly and preventively, with respect to

individual institutions (capital injections, asset solutions) as well as more generally (guarantee

schemes, liquidity operations). They reduced the default risks among financial institutions and thus

helped to safeguard financial stability. However, the interventions had distortionary effects, because

the rapid unfolding of the crisis and market failures complicated the proper design of support policies.

Setting conditions for support cannot always prevent that the level playing field between supported

and non-supported institutions is affected and that undesirable shifts in capital flows occur. Moreover,

it cannot be ruled out that interventions damage the confidence of market participants, also with regard

to the financial strength of support-providing governments and central banks. To reduce such negative

side effects of support, it is important that the support policies are market compatible and

unambiguous and are timely withdrawn through an appropriate exit strategy.

162

163

Chapter 9

Macro-effects of higher capital and liquidity requirements for banks:

Empirical evidence for the Netherlands

9.1 Introduction

The credit crisis has demonstrated that the ability of banks to absorb shocks needs to be

strengthened.62 The crisis could assume such serious proportions, because the exposure of banks was

too high and too risky in relation to their capital reserves. As a result, they had too little capacity to

absorb the losses on their market and loans exposures. Banks were forced to respond by reducing their

high-risk positions. Liquidity buffers held by banks were also generally inadequate, making them

vulnerable when market liquidity dried up. Against this backdrop, investors lost confidence at the

height of the crisis in the autumn of 2008, and governments and central banks had to step in (see

Chapter 8).

To prevent a repetition of such problems, the Basel Committee developed a raft of measures to

strengthen the banking system (i.e. Basel III). The intention is to improve the resilience of individual

banks whilst also improving the stability of the financial system as a whole. The measures will have an

impact on the financial position of banks, which will be required to raise and hold more and better-

quality capital, thereby affecting their funding costs. The asset side of banks’ balance sheets will also

change, for example, because banks need to hold more liquid assets.

The influence of such developments on the functioning of banks will also have

macroeconomic consequences, both in the transitional phase and in the new ‘steady state’ in the longer

term. Precisely how these effects will manifest themselves is difficult to predict, and depends among

other things on how banks behave, on supply and demand in the money and capital markets (e.g. the

demand for long-term bank bonds) and the strategy of other financial institutions, which could take

over part of the maturity transformation role of banks. The challenge is to phase in the new capital and

liquidity requirements in such a way that lending is not unnecessarily constrained and that economic

recovery is not stifled. This chapter examines the macroeconomic effects of the stricter supervisory

standards for the Netherlands.

The design of the new standards is discussed further in Section 9.2. The channels through

which those standards could impact on bank behaviour, lending and the economy during the

transitional phase form the subject of Section 9.3. Section 9.4 quantifies the potential macro-effects of

the transitional process for the Netherlands. Section 9.5 focuses on the new steady state, in which

banks meet the new standards and have adapted their business models. The chapter only addresses the 62 This chapter is a revised version of Berben, Bierut, Kakes and Van den End (2010).

164

effects of the new quantitative standards, and does not consider the potential interaction with new

qualitative supervisory standards.

9.2 New regulatory standards

A key objective of Basel III is to raise the quality and amount of capital held by banks (see Box 9.1 for

details). For example, a higher proportion of capital must consist of core equity, especially paid-up

share capital and retained earnings. The capital requirements to cover market risk, resecuritisation and

counterparty credit risk (mainly interbank exposures) will also be raised. An additional leverage ratio

will be introduced to control the relationship between banks’ assets and capital reserves. In order to

constrain pro-cyclical behaviour by banks, a higher target capital ratio will be introduced over and

above the minimum capital requirement. Restrictions on profit distribution will encourage banks to

build up this additional buffer in good times, so that they can fall back on it during periods of

economic downturn. The buffer may also be temporarily increased if lending in a given country

accelerates to exceptionally high levels. Forward looking provisioning for non-performing loans could

also help prevent pro-cyclical activity.

The Basel Committee and the Committee of European Banking Supervisors (CEBS63) also

developed internationally harmonised standards for liquidity risk. The Liquidity Coverage Ratio

(LCR) is intended to enable a bank to survive a severe stress scenario lasting one month. To make this

possible, the liquid assets held must be sufficient to cover the assumed net cash outflow. The

composition of the assets is important, and notably the proportion of assets that is highly marketable or

can serve as collateral for central bank borrowing, or that can be rapidly turned into cash in some other

way. The Net Stable Funding Ratio (NSFR) was developed to reduce banks’ maturity mismatch.

Under this standard, longer-term bank lending must be covered by long-term stable funding, such as

retail savings and wholesale funding with a term to maturity of more than one year.

In addition to the quantitative capital and liquidity standards developed by the Basel

Committee, several other initiatives, most of them more qualitative in nature, have been considered in

order to limit the systemic risks of large, complex financial institutions. These include greater

independence for group entities and measures to facilitate an orderly break-up of financial institutions,

for example by making it mandatory for institutions to formulate a ‘living will’. Policymakers also

considered the introduction of restrictions on the extent and nature of banking activities and a bank

tax. These initiatives may provide added value, but ideally should not distract from the core need for a

fundamental strengthening of the financial system in terms of capital and liquidity requirements. The

new regulatory framework will be phased in gradually in order to prevent overly abrupt changes in the

63 The CEBS has been succeeded by the European Banking Authority (EBA) in 2011.

165

sector. The intention is that most of the quantitative standards will be implemented from 2013

onwards. The leverage ratio and the NSFR will be introduced after a longer transition period.

Box 9.1 New Basel Committee measures

In the wake of the financial crisis, the Basel Committee on Banking Supervision announced a

comprehensive raft of measures aimed at strengthening the banking sector. The measures supplement

and reinforce the Basel II Capital Accord for banks, which was introduced in 2008. Basel II introduced

important new elements to banking supervision compared with the first Accord from 1988. A more

risk-oriented approach was deliberately chosen in Basel II, in which banks are required to hold more

capital for high-risk activities than for activities with a low level of risk. The Basel II rules for the

‘banking book’ incorporate capital requirements to cover the credit risk on lending and securitisation

operations. These activities represent the lion’s share of bank balance sheets. Basel II has stuck to the

rules introduced since 1996 to deal with the trading book and market risk. In addition, in contrast to

the old Basel I, Basel II incorporates capital requirements to cover operational risk.

The Basel III capital standards were agreed in September 2010 by the group of Governors and

Heads of Supervision (GHOS), which oversees the Basel Committee. Among other reforms, the

GHOS proposed a strengthened definition of capital; calibrated requirements for minimum capital

ratios and for a new capital conservation buffer; and specified a transition path for the new standards

The definition of capital was strengthened to improve the quality of capital, through a tightening up of

the admission criteria for capital instruments which do not form part of core capital. The introduction

of new tax allowances will ensure that capital elements of insufficient quality are deducted from total

capital. Basel III entails a new minimum requirement for the quantity of capital, i.e. the amount of

capital that an institution needs to hold in order to be regarded as viable by the markets, of 4.5%

(common core equity ratio). In addition, banks must maintain a buffer over and above the minimum,

the aim being to enable them to survive a period of stress without falling below the minimum capital

requirement. This buffer will take two forms. First, the capital conservation buffer of 2.5%, will

encourage banks to grow towards a target capital ratio above the minimum; as long as this target has

not been reached, profit distributions such as dividends will be reduced. A second element of the

buffer is linked to the growth in national credit: a bank must allow its buffer to increase whenever

credit growth is excessive. Banks will be able to use both buffer elements during bad years. The

existing capital requirements to cover market risk and risks related to complex financial products are

also being raised; during the crisis it became apparent that banks had suffered particularly large losses

on these activities.

The risk-weighted capital requirements are being supplemented with a non-risk-weighted

capital criterion, known as the leverage ratio. This ratio, which compares unweighted total assets to

capital held, is intended to set a limit to the build-up of excessive debt positions, one of the causes of

166

the crisis. The proposed leverage ratio of 3 percent is intended to serve as a ‘back-stop’ for the risk-

weighted requirements and thus to limit the growth of the balance sheet.

Basel III entails major progress in the area of liquidity supervision. The Liquidity Coverage

Ratio (LCR) is intended to enable institutions to survive a severe stress scenario during a period of one

month. To this end, the liquid assets held must be sufficient to cover the presumed net cash outflow. In

addition, the Net Stable Funding Ratio (NSFR) was been developed to reduce the maturity mismatch

of banks. Under this measure, longer-term bank lending must be covered by long-term, stable funding

such as savings and wholesale finance with a term of more than one year.

9.3 Channels of effects during the transitional phase

9.3.1 Direct consequences

During the transitional phase, the higher regulatory standards will impact on banks’ balance sheets and

profitability in various ways (see Figure 9.1). In order to achieve the higher capital ratio, banks will

either have to raise more equity or retain more profits. Another option would be to reduce the size of

the balance sheet by selling assets or reducing risk-weighted assets. In order to meet the liquidity

standards, banks will have to limit their maturity mismatch, for example, by enhancing the liquidity

profile of assets or extending the term to maturity of funding. Taken together, these measures would

increase the cost of funding and reduce interest income; all things being equal, this will lead to a fall in

profitability.

9.3.2 Effect on lending via interest rate channel

In order to maintain profitability, banks will seek to counter the dip in their profits by raising the

interest rates charged on loans and reducing the rates paid on deposits where possible, i.e. increasing

the ‘lending wedge’; see Figure 9.1. This means that the interest rate channel, traditionally the main

167

monetary transmission channel, will be influenced by the new regulatory standards. The cost of capital

will increase for households and businesses, while demand for credit will fall (via price rationing).

According to Elliot (2009), an increase of the capital ratio from 4 percent to 10 percent will raise the

interest rate permanently by around 50 basis points, because the costs of higher capital buffers will be

distributed among shareholders, lenders and borrowers. Barell et al. (2009) also model the lending

wedge and use it as input for a structural macro-model to calculate the impact of higher regulatory

standards on the British economy. In their approach, a one percentage point increase in the capital and

liquidity ratios of banks would push up the costs of capital for households and businesses by just under

1 percent. This, they argue, would have a negative impact on investment and consumption. Scott and

Vlček (2011) use a DSGE model to simulate a scenario in which capital requirements increase 2

percentage points in two years. The simulations show a peak rise of lending spreads by 120 basis

points in the euro area and by 130 basis points in the US, implying a peak decline in output of 0.6

percent in the euro area and 0.5 percent in the US.

9.3.3 Effect on credit supply via bank capital channel

The higher liquidity and capital requirements will limit the excess capital and liquidity of banks. This

could lead to adjustments on the assets side of the balance sheet, such as a reduction in the risk grade

of assets or restrictions on the availability of credit. This may be caused by constraints on funding, via

the bank lending channel, or on equity via the bank capital channel (ECB, 2009c). This latter channel

can become active if regulators or the market impose higher capital requirements. This reduces the

surplus between available capital and required capital, forcing banks to de-risk their balance sheet if

they are unable to compensate for this by retaining earnings or raising capital externally. Empirical

research confirms that a falling capital surplus prompts banks to modify their behaviour (Alfon et al.,

2004). The bank capital channel is described in the literature in relation to monetary transmission, with

undercapitalised banks being more sensitive to a rise in interest rates because they have fewer funding

alternatives (Peek and Rosengren, 1995) and are more vulnerable to a reduction in the interest rate

margin (Van den Heuvel, 2002). Other studies view the bank capital channel in relation to

deleveraging and the risk of credit rationing.64 This manifests itself in a reduction in lending capacity

or credit limits. Studies carried out for the US and the UK suggest that a one percentage point

reduction in surplus capital corresponds with a decline in lending of between 0.1 and 2.5 percent

(Bayoumi and Melander, 2008; Berrospide and Edge, 2009; Francis and Osborne, 2009). The supply

of credit could also be restricted through modification of lending standards such as collateral

requirements and covenants. Since the state of banks’ balance sheets can have an impact both via the

interest rate channel and the bank lending and capital channels, the effects on lending will overlap.

64 For the US, see e.g. Bayoumi and Melander (2008) and Berrospide and Edge (2009), and for the UK Francis and Osborne (2009).

168

9.3.4 Influence on risk behaviour of banks

Supervisory standards also influence banks’ risk behaviour. In principle, the new requirements are

intended to mitigate the risks that banks are inclined to take on the back of limited financial liability of

bank owners and deposit insurance. A counter veiling influence to the propensity of banks to take risk

is their fear of losing charter value if they should fail. Theoretical and empirical research does not

produce uniform findings on the impact of higher capital requirements on risk behaviour (see Stolz,

2002, for a detailed literature review). On the one hand, banks with higher buffers have less scope for

high-risk lending, especially if the capital requirements match the risk profile of the loans (in other

words, are risk-weighted). On the other hand, higher capital requirements provide an incentive for

banks to compensate for the costs involved by taking risky positions, especially if the risk weights

imposed by the regulator are not in line with the actual risks. As regards the transmission to the

economy, increasing the risk-weighted capital requirements can be accompanied by a relative change

in risk premiums between sectors in the economy, affecting the demand for credit from those sectors

(Figure 9.1).

9.3.5 Broader effects of liquidity requirements

Stricter liquidity requirements could also have negative effects on lending. The LCR increases the

need of high-grade government bonds and other liquid assets, and this can crowd out lending. Apart

from changes on the assets side of the balance sheet, the LCR and the NSFR will encourage banks to

reduce their maturity mismatch by raising more stable funding, for example in the form of longer-term

bonds or retail deposits. This will increase the funding costs for banks, which they may pass on to

customers through a higher lending wedge. The effects of this on banks and the economy cannot

simply be added together with those of the higher capital standards, because they are communicating

vessels. An increase in liquid assets could be accompanied by a decrease in more risky assets such as

loans, thereby improving the capital ratio. Conversely, a stronger capital position will lead to a higher

NSFR and thus to a reduction in liquidity risk. The extent of this offset depends on the structure of the

balance sheet, how binding the new standards are and on the response of individual banks. This

presents a challenge for regulators who should consider capital and liquidity requirements in a holistic

way to assess the influence on the banking sector and the economy.

Finally, there is an interaction between the liquidity standards and the monetary operations of

central banks. Extending funding maturities could lead to higher bids in the long-term refinancing

operations of central banks. This has implications for the implementation of monetary policy, among

other things through changes in the relationship between the demand for liquidity in short and longer-

term tender operations. The reduced demand for short-term funding is likely to result in reduced

activity in the short end of the money market. Increasing demand for term liquidity could result in a

steeper money-market curve. These factors could have consequences for (the transmission of)

monetary policy and the intermediary target variable (currently the overnight rate EONIA).

169

9.3.6 Impact on financial markets

The stronger regulatory framework will also affect the financial markets during the transitional phase.

Banks will be forced to turn more to equity markets to strengthen their capital position. It is estimated

that the 20 biggest banks in the euro zone will have to raise approximately EUR 115 billion in Tier 1

capital in order to meet a two percentage point increase in the capital ratio (BIS, 2010b). The liquidity

requirements could also have major consequences for the fixed-income market, though it is difficult to

be precise here (Figure 9.1). On the one hand, there will be greater demand for government bonds

from banks as they seek to increase their liquidity buffers. This will put downward pressure on bond

yields. The asset purchasing programmes of the US Federal Reserve give an indication of the interest

rate effects of large-scale bond purchases; a study by the New York Fed estimates that asset purchases

(USD 1,800 billion) have pushed down ten-year Treasury yields by 50 basis points (Sack, 2009). On

the other hand, the requirement to reduce the maturity mismatch (via the NSFR) will create greater

demand from banks for long-term funding. This additional funding demand could drive up yields on

bank bonds. As regards monetary transmission, this means that the effect of monetary policy on long-

term interest rates during the transitional phase will become (temporarily) less predictable.

The impact of the new supervisory requirements on financial markets in the eurozone could be

limited, especially if a sufficiently long transitional period is adopted. The euro area equity markets

(total amount outstanding EUR 3,500 billion, of which almost EUR 400 billion has been issued by

banks) and the government bond markets (total amount outstanding EUR 5,500 billion) appear to be

deep enough to absorb the demand for additional issues of equity and debt securities (source of the

data: ECB, 2010c).

9.4 Effects during the transitional phase: model outcomes for the Netherlands

In order to assess the impact on the lending wedge, lending and economic growth in the Netherlands

during the transition period, a number of modelling methods were used. This provides a more

complete picture of the effects, while a multiple-model approach is justified because of the uncertainty

that surrounds the outcomes. First, simple regression techniques (‘satellite models’) are used to

explain developments in balance sheet variables and the lending wedge out of liquidity and capital

ratios, macroeconomic and bank-specific variables. The outcomes are then used in the macro-

econometric model of De Nederlandsche Bank (DNB) to simulate the impact on Gross Domestic

Product (GDP). Finally, a Vector Autoregression (VAR) model is used to estimate the relationship

between macroeconomic variables and bank variables. The macro-effects during the transitional phase

depend on how far the standards are raised and on the length of the period over which they are

implemented. For this reason, the outcomes are presented in the form of scenarios.

170

9.4.1 Scenarios

The scenarios on which the simulations of the macro-effects of the new capital and liquidity standards

are based assume different levels of the standards and different implementation periods (Table 9.1).

They serve as examples of simulated macro-effects of each percentage point increase in the target

capital ratio. For the capital ratio, the scenarios use the core capital relative to the risk-weighted assets

(RWA). Core capital is defined as Tangible Common Equity (TCE), consisting of ordinary share

capital and retained profits, reflecting the fact that the new standards are aimed at this highest-quality

capital because it offers the best guarantee that a bank will be able to absorb any losses.

The liquidity scenarios assume that the relationship between liquid assets (consisting mainly

of cash and government bonds) and total assets increases by 25 percent. This reflects a change in the

LCR. Changes in the NSFR are estimated on the basis of the assumption in the scenarios that banks

will extend the maturity of their wholesale funding by one year (compared with the present average of

around six years for Dutch banks). This will increase funding costs. For the implementation periods, it

is assumed that the higher standards will be phased in gradually over a period of two, four or six years.

This means that banks will seek to achieve higher capital and liquidity ratios and will do so at the end

of these periods (in practise, the trajectory in which the new standards will be implemented by the

banks will depend both on the regulatory requirement and market pressure).

Table 9.1. Scenarios

Capital scenarios TCE/RWA ratio increases per percentage point

Liquidity scenarios Liquid assets / total asset ratio increases 25%

Maturity wholesale fundingincreases by 1 year

Implementation period Changes implemented within 2 yearsall scenarios within 4 years

within 6 years

9.4.2 Satellite models

Simple regression techniques were used in the form of satellite models in order to estimate the

relationship between capital and liquidity ratios on the one hand and balance sheet items (such as total

assets, loans and capital reserves) and the lending wedge on the other. These relationships give an

impression of the speed with which banks adjust their balance sheets and loan rates in order to meet

the higher capital and liquidity targets. The estimates are based on historical data from the five largest

Dutch banks, which together represent around 90 percent of the sector.

171

The satellite model for assets and capital assumes that changes in banks’ assets and liabilities

are determined by movements in the surplus capital ratio (the model specification is based on Francis

and Osborne, 2009). The surplus ratio Zi,t is defined as the ratio between the current capital ratio (ki,t)

with a one-period lag, and the target ratio (k* i,t) for bank i, measured in percentage points,

−= − 1100 1

t,i

t,it,i

*k

k*Z (9.1)

The target ratio (k* i,t) is approximated by the long-term average capital ratio and the external rating of

the banks (Rati,t), as a proxy for the market demand for capitalisation,65

t,i

n1t t,i

t,i Rat

kn

1

*k∑ =

= (9.2)

The parameter (β), that relates assets and liabilities to the surplus capital ratio, ensues from the

following panel regression model,

t,is4

1ss

2

1jjtj,3jt

2

1jj,2jt

2

1jj,1t,i

t,i

t,iQRLINFGDPZ

C

Aερδδδβα ++++++=

∑∑∑∑==

−−=

−=

(9.3)

where Ai,t represents the total assets, risk-weighted assets and loans of bank i and Ci,t represents the

total capital and the Tier 1 capital. The balance sheet items are included as percentage changes. The

explanatory variables, in addition to the surplus capital ratio Zi,t, are the percentage change in real

Gross Domestic Product (GDP), inflation (INF) and long-term interest rates (RL), as an approximation

of interest rates on loans and quarterly dummies (Qs).

The estimation outcomes of equation 9.3 are shown in Table 9.2. The coefficients of Zi,t are

generally significant or almost significant, and have the expected sign. This implies that banks will

reduce their assets in response to a falling surplus ratio (caused either by a reduction in the existing

capital reserve in the numerator of Zi,t, or by a higher target capital in the denominator of Zi,t). The

coefficient of lending growth is not significant, indicating the reluctance of banks to reduce their loan

book in the event of a falling capital surplus. The significant or almost significant positive sign of the

coefficient of total assets and risk-weighted assets suggests that Dutch banks prefer to reduce assets

65 The rating was determined by converting the Moodys ‘letter rating’ for each bank into a numerical value (where AAA is equal to 19 and C to 1). The numbers were then divided by the long-term average rating of Dutch banks. The result is a ratio of around 1.

172

other than loans. The significant negative sign of the coefficient of total capital and Tier 1 capital

shows that banks respond to a fall in the surplus ratio Zi,t by raising additional capital.

Table 9.2. Estimation outcomes satellite model for balance sheet adjustmentsBased on 1998q1-2009q4 period and panel of 5 large Dutch banks

Growth in: Loans Assets RWA BIS cap Tier 1 cap

Zi,t 0.07 0.23** 0.21** -0.13 -0.17*(1.12) (2.51) (2.32) (-1.60) (-1.62)

GDPt-1 0.91* 2.55*** 2.14** 1,24 1.89*(1.76) (2.74) (2.36) (1.44) ' (1.71)

GDPt-2 0.79 1.14 2.23** 0.83 -0.5(1.10) (1.19) (2.37) (0.93) (-0.43)

INFt-1 -1.16* -1.91 -2.45* -1.02 -2.55*(-1.73) (-1.42) (-1.84) (-0.80) (-1.78)

INFt-2 -0.96 -2.59* -3.20** -2.85** (-0.52)(-1.28) (-1.77) (-2.23) (-2.09) (-0.31)

RLt-1 -0.02* -0.06*** -0,02 -0.04** -0,02(-1.66) (-2.91) (-1.16) (-2.14) (-0.72)

RLt-2 0.02* 0.06*** 0,03 0.04** 0.02(1.85) (3.04) (1.47) (2.31) (0.67)

Constant 0.02 0.03 -0.02 0.02 0.04(0.75) (1.06) (-0.83) (0.84) (0.72)

R2 0.09 0.14 0.16 0.06 0.05Prob (F stat) 0.05 0.00 0.00 0.03 0.36DW stat 1.98 2.09 2.10 1.98 2.33

***, **,* significant at 1%, 5%, 10% confidence level, t-values between bracketsQuarterly dummies not reported.

In the satellite model for the lending wedge it is assumed that banks will increase the lending wedge –

the difference between interest rates charged on loans and rates paid on deposits – in order to

compensate for rising funding costs and declining revenues due to the stricter capital and liquidity

requirements. The relationships between the lending wedge (WEDGEi,t) and the capital and liquidity

ratios are estimated as follows (based on Barell et al., 2009),

t,i1t51t41t31t21t,i STDPROVLIQCAP)RSRL(WEDGE εδδδδδα +++++−+= −−−− (9.4)

Here, (RL-RS) is the difference between ten-year interest rates on government bonds and three-month

EURIBOR. CAP is the Tier 1/RWA ratio and LIQ the liquid assets/total assets ratio. PROV are the

provisions for non-performing loans as a percentage of total loans and STD are the bank lending

standards (the net percentage of banks that tighten up their lending criteria).

Table 9.3 shows the estimations results of equation 9.4, with a breakdown of the lending

wedge for loans to companies (WEDGE_comp) and households (WEDGE_hh). The estimated

173

coefficients are significant and have the expected sign. The sign for LIQ and CAP is positive, which

means that higher ratios are associated with a higher lending wedge. Changes in the lending wedge in

response to higher capital and liquidity requirements are then simulated on the basis of the estimated

coefficients for CAP and LIQ ( 2~δ and 3

~δ ) and the costs of liquidity associated with an assumed

extension of wholesale funding maturity and the loss of profit because loans are substituted by liquid

assets.

Table 9.3. Estimation outcomes satellite model for lending wedgeBased on 1998q1-2009q4 period and panel of 5 large Dutch banks

WEDGE_total WEDGE_comp WEDGE_hh

RL - RS 0.45*** 0.47*** 0.23***(14.91) (7.78) (4.25)

RL 0.64**** 0.54****(11.85) (3.06)

CAPt-1 22.18*** 10.88*** 18.66***(6.57) (3.19) (2.55)

LIQt-1 5.34*** 3.71*** 6.54***(4.82) (3.52) (2.79)

PROVt-1 0.00*** 0.00 0.00(5.65) (1.26) (0.71)

STDt-1 -0.00* -0.00*** -0.00*(-2.23) (-6.12) (-0.63)

Constant -2.99*** -3.49*** -4.62***(-6.18) (-5.68) (-4.24)

R20.98 0.94 0.77

Prob (F stat) 0.00 0.00 0.00DW stat 1.98 1.70 1.62

***, **,* significant at 1%, 5%, 10% confidence level, t-values between brackets

9.4.3 Simulation outcomes

To simulate changes in balance sheet items in response to the higher target capital ratio, it is assumed

that banks will respond to changes in their surplus capital, which falls as the target ratio rises. The

simulation outcomes indicate that banks compensate two-thirds of their declining capital surplus by

reducing (the level of risk of) assets and the other third by raising additional capital (see Figure 9.2, in

which assets and liabilities are modified on the basis of the relationship with the surplus capital ratio

as modelled in equations 9.1 - 9.3; the figure shows the outcomes for each percentage point increase in

the target capital ratio). Banks will mainly raise core capital in the form of ordinary shares, because

under the new requirements it will be mainly this category of capital that is lacking.

174

-8%

-5%

-3%

0%

3%

5%

0 4 8 12 16 20 24 28 32Total assets LoansRisk weighted assets Tier 1 capitalCore capital

Figure 9.2. Balance sheet adjustments, percentage point higher TCE/RWA ratio Deviation from baseline, time on x-axis in quarters, implementation period 4 years

In scenarios involving higher capital requirements, the change in lending is limited. One reason for

this is the historically low elasticity between lending and the capital ratios of banks, which is one of

the determining factors in the satellite model. In scenarios where the capital ratio requirement rises by

several percentage points, the volume of lending by Dutch banks would be between 3 and 6 percent

lower at the end of the implementation period than in the baseline scenario. The reduction in total

assets compared with the baseline is more than three times as large. This difference can be explained

by the pecking order applied by banks in adjusting their balance sheets in response to shocks in the

capital ratio. This means that loans are adjusted after other asset items have been modified, such as

trading book and real estate exposures. This is in line with experiences during the recent crisis, when

lending by Dutch banks continued to grow slightly on an annualised basis despite the sharpest

downturn in the economy for 80 years and despite the pressure on the capital position of the banks.

In addition to the capital requirements, the new higher liquidity requirements will have an

impact on bank balance sheets. The model assumes that banks will respond by substituting loans for

liquid assets, so that the lion’s share of the balance sheet adjustment is accounted for by loans (in a

scenario where the ratio of liquid assets to total assets rises by 25 percent, lending falls by around 4

percent compared with the baseline level). However, this is a rather conservative assumption: during

the crisis, lending held up reasonably well, partly thanks to the crisis measures taken by governments

and central banks. It is plausible that part of the balance sheet adjustment will be achieved through an

increase in the lending wedge. The wedge increases if banks pass on the fall in return (due to the

substitution of less liquid assets for liquid assets) and the raising of longer-term funding to their

175

customers (see equation 9.4).66 The simulation outcomes show that the wedge could increase by

several tens of basis points due to the higher liquidity requirements, depending on the scenario

considered (Figure 9.3). The effect on the wedge of a two percentage point increase in the target

capital ratio is of the same order. As expected, a longer implementation period limits the increase in

the wedge during the early years of the scenario, although this has virtually no impact on the ultimate

outcomes.

0

5

10

15

20

25

30

35

40

0 4 8 12 16 20 24 28 32

+25%, 2yr +25%,4yr +25%, 6 yr

Figure 9.3. Impact on loan spread of rising liquidity ratioDeviation from baseline, time on x-axis in quarters

9.4.4 Simulations using macro-econometric model

The effect of a higher lending wedge on the Dutch economy is simulated using DNB’s new structural

macro-econometric model Delfi (DNB, 2011). In this model, changes in the lending wedge influence

the cost of finance and therefore investment and consumption. The model simulations show that

scenarios for higher capital and liquidity requirements involving changes in the lending wedge have a

limited effect on real GDP. The negative cumulative deviation from the baseline scenario per

percentage point increase in the capital ratio is around 0.05 percent, and an increase in the liquidity

ratio of 25 percent would mean that real GDP was around 0.1 percent lower (Figures 9.4 - 9.5). This

means that GDP would be lower because of temporarily slower growth; it does not mean that potential

economic growth would be lower (the volume effect would be permanent as long as the interest rate

margin would not fall). It should be noted that the simulated GDP effects are surrounded by

uncertainties. The effects of the scenarios cannot simply be added together, because higher capital and

liquidity requirements partially offset each other. For example, substitution of illiquid for liquid assets

would lower the average risk level of the assets. As a result, a bank would need to hold less capital 66 For the purposes of this study, the analysis is deliberately limited to the partial effects of the capital and liquidity scenarios; the analysis abstracts from structural factors which can influence the lending wedge, such as the degree of concentration in the banking sector.

176

(this effect was not included in the analysis in order to maintain a clear view of the individual effects).

One limitation of the macro-model is that it mainly simulates the effects on demand for credit (via the

price of credit), and not so much the effects on the supply of credit. Model outcomes for other

countries suggest that supply constraints – insofar as they lead to stricter lending conditions (other

than the interest mark-up) – could have an additional negative effect on GDP.67 Certain sectors, such

as small and medium-sized enterprises, which have virtually no access to other external funding

sources, are particularly susceptible to restrictions in the availability of bank lending. Moreover, the

calculations for the Netherlands take no account of international spill-over effects: if banks worldwide

adjust their capital and liquidity positions, national economies will also be influenced from abroad.

The figures show that a longer implementation period mitigates the impact in the short term.

Moreover, an implementation period of six years reduces the cumulative effect slightly. A monetary

policy response could also mitigate the impact on real GDP (the outcomes shown here assume no

policy response). Calculations using DNB’s macro-model show that the GDP effect in the Netherlands

would be reduced by roughly one fourth if interest rates react mechanically to developments in the

economy. The calculated effects could also be an overestimation if banks have already anticipated the

higher standards. If capital buffers have already been strengthened and funding profiles adjusted under

pressure from the markets, smaller adjustments would be needed in the future in response to the actual

implementation of the higher regulatory standards.

-0,20%

-0,15%

-0,10%

-0,05%

0,00%

0 4 8 12 16 20 24 28 32

in 2 years in 4 years in 6 years

Figure 9.4. Real GDP impact increase TCE/TWA ratio, structural model Deviation from baseline, impact of 1 percent point increase capital ratio, time on x-axis in quarters

-0,20%

-0,15%

-0,10%

-0,05%

0,00%

0 4 8 12 16 20 24 28 32

in 2 years in 4 years in 6 years

Figure 9.5. Real GDP impact increase liquidity ratio, structural model Deviation from baseline, impact 25% increase of liquidity ratio, time on x-axis in quarters

9.4.5 Time series analysis using a VAR model

An alternative method for simulating the GDP effects of higher capital requirements is a time series

analysis using a Vector Autoregression (VAR) model. A VAR model describes the dynamic of

variables based on their historical relationships. The model estimated here is based on short-term

67 According to the BIS (2010b), an increase in the target capital ratio of one percentage point could lower GDP via that channel by around a further 0.16 percent compared with the baseline scenario (median outcome of several model calculations). The supply effects are expressed in the time series analysis in the next section.

177

relationships, because this proved to be the most robust model.68 The following model was estimated

to simulate the scenarios for an increase in capital requirements for the Netherlands,

Xt = Γ (L) Xt-n + µ + εt (9.5)

where Γ (L) is the matrix of estimated parameters, L the lag operator (the model incorporates two

lags), vector Xt = ( log(GDPt), log(INFt), SPRt, log(LOANt), CAPt, STDt) and µ is a vector with

constants. The model therefore contains the standard variables of VAR models from the monetary

transmission literature, namely real GDP (GDP), inflation measured in terms of the GDP deflator

(INF) and an interest rate (the lending wedge SPR). The supply variables included in the model are

total bank lending (LOAN), the surplus bank capital (CAP) measured as available capital less required

capital as a percentage of risk-weighted assets, the bank lending wedge (SPR) and the net lending

standards (STD). 69 The model was estimated with log(GDP), log(LOAN) and log(INF) as quarter-on-

quarter changes and SPR, CAP and STD in levels. SPR was calculated as the difference between the

average interest rate on loans to businesses and the three-month money market rate. CAP was based on

core capital and risk-weighted assets calculated on the basis of banks’ balance sheet data. The model

was estimated using quarterly data for the period 1990q1-2009q4.

The variables LOAN, CAP, SPR and STD say something about the interaction between the

economy and the banking sector and align with the ‘credit view’ of monetary transmission in which

credit supply constraints play a role. For example, a shortage of capital can urge banks to deleverage.

A change in the lending wedge or lending criteria could also impose constraints on lending. These

transmission channels mean that a falling capital surplus has an impact on lending and on GDP in the

VAR simulations. The effect of the bank lending standards says something about the bank credit

supply constraints, which are translated into stricter lending criteria.

The outcomes of equation 9.5 are presented in the form of ‘impulse response functions’ in

Appendices 9.1 and 9.2.70 Although few responses are statistically significant, the growth in lending is

found to respond positively to an increase in the capital surplus (CAP); see panel A.4 in Appendix 9.1.

The net lending standards rise and are tightened up, possibly because banks with high capital buffers

are more critical in accepting loans (panel A.6). The lending wedge barely responds to a shock

movement in CAP (panel A.3). Not shown in the Appendices is that the capital surplus of the banks

responds positively to an upward movement in GDP growth and lending wedge (which contributes to

higher profits) and negatively to a positive shock in inflation and net lending standards (the latter

68 By way of alternative, a Vector Error Correction Model (VECM) was also estimated, but this produced very volatile outcomes (both upwards and downwards) and was left out of consideration for substantive and statistical reasons. 69 Data on lending standards are only available from 2003. We have therefore back-forecast them based on a model with the interest charged on loans to corporates and the credit spread on corporate bonds. 70 The impulse responses are based on Cholesky decomposition, whereby the inverse of the Cholesky factor of the covariance matrix of residues was used to make the shocks orthogonal.

178

effect is shown in panel A.11 in Appendix 9.2). A positive shock in lending standards means there is a

net tightening of banks’ loan criteria. As panels A.7 and A.10 in Appendix 9.2 show, this has a

negative impact on GDP and lending, which provides an indication of the effect of supply constraints.

Model simulations of the scenarios show that each percentage point increase in the capital

requirements reduces real GDP by between 0.1 and 0.3 percent compared with the baseline scenario

(Figure 9.6). These estimates are higher than the simulation outcomes using the macro-econometric

model, because the VAR approach takes credit supply effects into account. The maximum negative

effect manifests itself after two to three years. GDP thereafter returns to the baseline level, although

this is due mainly to the statistical properties of time series models, in which shock effects disappear

over time. The model outcomes show that a short implementation period leads to a relatively sharp fall

in real GDP, concentrated in the first two years of the scenario. In scenarios with a longer transitional

period, the negative effects on the economy are spread out over more years. This outcome is in line

with Scott and Vlček (2011), who conclude from simulations with a DSGE model that extending the

implementation horizon for higher capital requirements from 2 to 4 years reduces the peak effect on

growth by approximately one-third.

-0,5%

-0,4%

-0,3%

-0,2%

-0,1%

0,0%

0,1%

0,2%

0 4 8 12 16 20 24 28

in 2 years in 4 years in 6 years

Figure 9.6. Real GDP impact increase TCE/TWA ratio, VAR modelDeviation from baseline, impact of 1 percent point increase capital ratio, time on x-axis in quarters

9.4.6 International perspective

The economic effects calculated for the Netherlands are in line with outcomes for other countries as

studied by the Basel Committee’s Macroeconomic Assessment Group (MAG) (BIS, 2010b). The

MAG calculated that lending wedges would rise by 15 basis points if the target capital ratio were to

increase by one percentage point, while lending volumes would fall by 1.4 percent compared with the

baseline scenario (outcomes as the median of different model outcomes for several countries, with a

four-year implementation period). The negative impact on real GDP is limited: a one percentage point

increase in the target capital ratio has a negative impact of between 0.07 and 0.31 percent compared

179

with the baseline scenario, with a median of just below 0.2 percent (model outcomes in different

countries with a four-year implementation period after 18 quarters, calculated using structural macro-

models). This includes international spill-over effects which could occur in the event of simultaneous

implementation of the new standards in multiple countries. These effects are not included in the

outcomes for the Netherlands. A scenario where the liquidity ratio rises by 25 percent would lead to an

increase of 14 basis points in the lending wedge and would depress real GDP by 0.08 percent

compared with the baseline scenario (median of model outcomes in different countries with a four-

year implementation period, after 18 quarters). The calculated outcomes show some divergence across

countries due to differences in methods and assumptions used and the different starting positions of

the banking sector in the various countries.

The macro-impact as calculated by the MAG is substantially lower than that calculated by the

Institute of International Finance (IIF, 2010), an organisation which represents financial institutions.

The IIF estimates that the new regulatory standards, assuming an increase in capital requirements of

two percentage points, could have a negative output effect of between 1.9 percent (Japan) and 4.3

percent (euro area) compared with the baseline scenario. Lending rates could increase by more than

130 basis points in the euro area. The differences compared with the outcomes as calculated by the

MAG are due to a number of factors:

° Capital scenario. The MAG includes only an increase in capital and liquidity ratios in its scenario,

whereas the IIF also allows for other national reforms which banks could face, such as restrictions

on their size and activities and the introduction of a bank tax.

° Baseline scenario. The IIF assumes low retention of profits by banks in the coming years, so that

large amounts of capital will have to be raised. The MAG implicitly assumes a return to higher,

historically long-term profit retention. The IIF also assumes a rising return on equity (ROE), while

the MAG generally assumes no change.

° Methodological differences. The IIF uses an approach in which GDP is a direct function of

lending. By contrast, the MAG approach is based on models used by central banks and the IMF,

which allow for the complex interactions between financial and economic variables (including

alternative sources of funding).

° Responses. The IIF assumes that banks will only achieve the NSFR by extending wholesale

funding maturities. This places heavy demands on capital markets and causes interest rate spreads

to rise much more strongly than in the MAG simulations. In practice, however, banks will also

adapt in other ways, such as by raising more retail funding or shortening the maturities of assets.

180

9.5 Effects in a new steady state with higher buffers

This section discusses the situation after all the changes following the introduction of the higher

minimum requirements have settled down (the new ‘steady state’). There is much more uncertainty

about this than about the transitional phase, especially when it comes to quantifying the ultimate

effects. Nonetheless, it is possible to say something about the most likely directions of change for a

number of factors. It is, for example, quite plausible that the increasing lending wedges during the

transitional periods – see Sections 9.3 and 9.4 – will become permanent, as compensation for the

higher funding costs faced by banks. On the plus side, the higher buffers will make a future financial

crisis less likely and less profound, while economic growth will be more stable. This section looks at

the costs and benefits of higher buffers in the new steady state.

9.5.1 Higher buffer requirements: costs and benefits

The costs of maintaining higher capital and liquidity buffers in the new steady state are to some extent

comparable with the costs during the transitional period. Higher capital buffers with a sizeable equity

component will drive up funding costs, since the return on equity (ROE) demanded by equity investors

will be relatively high. The same applies for liquidity risk: a reduction in the maturity mismatch

implies a lower profit margin on average for the traditional banking business, while the increased

exposure to high-quality liquid assets will dent profits. Logically, these costs will be passed on to

borrowers in the form of a higher lending wedge.

However, these cost increases can be kept limited: over the longer term there are also other

options that can be used to the full in the new situation. For example, companies could substitute bank

borrowing for alternative means of financing, via other financial intermediaries, or raise funds directly

in the capital market (although this is less of an option for smaller companies). Banks could also adapt

their behaviour and business models. The new regulatory standards could provide an incentive for

banks to lower their cost base by adopting a different business strategy, or offering products with a

more stable and fee-based income stream. This would reduce the risk-weighted assets and the funding

requirement, thus making the new standards less binding. Furthermore, higher capital and liquidity

buffers could also have a mitigating impact on equity costs; this would create scope for limiting the

increase in the lending wedge. Higher buffers reduce risk and shareholders will therefore be prepared

to accept lower returns.71 All in all, it is likely that the costs of the new minimum requirements as

estimated in the preceding sections will form an upper limit to the economic costs in a new steady

state. Angelini et al. (2011) conclude from simulations with a suite of models applied to multiple

71 This would be consistent with the Modigliani-Miller conditions and implies that the funding costs do not rise if more equity is held (Box 9.2). See also Miller (1995), who argues that the (expected) return on equity and borrowed funds should be comparable if a correction is applied for the risk profile.

181

countries that each percentage point increase in the capital ratio causes a median 0.09 percent decline

in the level of steady state output, relative to the baseline. The impact of the new liquidity regulation is

of a similar order of magnitude, at 0.08 percent. Additionally, a shift towards a somewhat less risky

funding structure for banks would not be abnormal from an historical perspective: a few decades ago,

banks held substantially more equity on average than they do nowadays (see Box 9.2).

It is not just the costs that will become apparent in the new steady state, but also the benefits of

the new capital and liquidity requirements. For example, a financial crisis would probably be less

likely and would cause less economic damage if it should occur. Viewed over a long period, countries

are hit by a banking crisis on average once every 20 to 25 years. While this is not very frequent, given

the potential damage it is worth taking steps to counter it. In a recent IMF study, the costs of a

financial crisis for taxpayers are estimated at around 15 percent of GDP, while the loss of national

income is even higher, at 20 percent.72 Bearing in mind the possibility that the economy could end up

on a permanently lower growth trend after a severe crisis, the costs could escalate far beyond this.

Several studies estimate the cumulative loss of welfare in such cases at between 60 and more than 100

percent of GDP.73

Higher buffers reduce the probability of crises and the amount of damage they cause. The

extent to which this is related to regulatory standards is difficult to quantify precisely: crises are

generally highly diverse and clustered in time. Examples include the Asian crisis at the end of the

1990s and the recent global credit crisis. Using various models, the Basel Committee recently

estimated the relationship between capital levels and the risk of a systemic crisis, in which a number of

measures were also included to reduce the liquidity risk (BCBS, 2010a). Figure 9.7 summarises some

of the outcomes of this study: if equity increases from 6 percent to 8 percent of the risk-weighted

assets, the risk of a crisis is more than halved. A further increase to over 10 percent reduces the risk to

less than one percent. If the liquidity risk is also reduced, in addition to the higher capital buffer, the

probability of a crisis declines much more quickly. The outcomes presented here are based on

reduced-form models, but prove to be robust if different model types and liquidity risk criteria are

used.74 There is ultimately a trade-off between the costs of additional capital and liquidity buffers and

the economic benefits they produce: the probability of a crisis at certain capital and liquidity levels is

so low that the costs of a further tightening of the regulatory requirements dominate.

72 See Laeven and Valencia (2008). The fiscal costs were calculated over the five years following the outbreak of the crisis. The loss of national income relates to the deviation of real GDP from the extrapolated trend over the three years following a crisis. Both figures are averages; the crises studied contain substantial upward and downward outliers. See also Reinhart and Rogoff (2009), for a historical overview of financial crises. 73 This is the present value of permanent losses. See e.g. Boyd et al. (2005) and Haldane (2010). 74 In addition to reduced-form models, similar calculations have been made using portfolio models and stress-testing models. Alternative ways of reducing liquidity risk are an increase in deposit funding and a more balanced liquidity profile of assets and liabilities. See BCBS (2010a).

182

Figure 9.7. Probability of systemic crisis at various buffer levelsPercentage

Probability of crisis at various levels of tangible common equity and liquid assets, simulated with reduced form models.

Source: BCBS

0%

2%

4%

6%

6 7 8 9 10 11 12 13 14 15

no change in liquidity

12.5% increase in liquid assets

50% increase in liquid assets

Box 9.2 Minimum regulatory requirements and the funding structure of banks

A sizeable body of literature deals with the factors that influence the financial structure of businesses.

An important starting point was the influential article by Modigliani and Miller (1958), in which they

demonstrate that – under specific conditions – the financial structure is irrelevant for the value of a

business. This implies that funding costs do not rise when the share of equity in the balance sheet is

increased. In practice, however, there are several reasons why the Modigliani-Miller conditions do not

apply and financial structure – including equity versus debt and liquid versus illiquid balance sheet

items – does make a difference.

For example, the degree of leverage is influenced by the tax system: in most countries interest

payments are tax-deductible, which means that borrowed capital is treated more favourably than

equity. Other factors have the opposite effect, constraining the leverage. For example, businesses with

a high risk profile are required to hold a relatively high capital buffer. Governance aspects may also

co-determine the financial structure – shareholders are after all the owners of the firm – but it is

ambiguous whether this will increase or reduce the leverage.

The financial structure of banks differs markedly from that of other enterprises (see Table 9.4): banks

have much more debt, which is moreover highly liquid, whereas bank lending is relatively illiquid.

183

This imbalanced financial structure is a direct result of the traditional function of banks as

intermediaries, in which they meet a social need for liquidity and maturity transformation. The

mismatch of banks’ balance sheets is facilitated by safety net schemes, such as the deposit guarantee

scheme and central bank facilities to resolve temporary liquidity problems. These safety nets reduce

the financial risks for equity investors and therefore encourage more leverage and a greater maturity

mismatch. Banks are moreover under supervision: this further reduces the risk of financial problems

and gives equity investors less reason to demand a solvency buffer. On the other hand, the regulator

sets minimum requirements for solvency and liquidity.

Table 9.4. Equity of banks versus other enterprises in the Netherlands, 2000-2008

Type of business Equity

Non-financial enterprises 44% Insurers Life 14% Non-life 26% Banks 3%

Note: the table shows equity as a percentage of the balance sheet total of Dutch enterprises, averaged over the years 2000-2008, as published by Statistics Netherlands (CBS) (finances of large companies) and DNB (balance sheets of insurers and banks, based on supervisory reports). For life insurers, technical provisions held at the risk of the policyholder have been deducted from the balance sheet total; if this is not done, equity would amount to 10 percent. Source: CBS, DNB

All in all, banks have much stronger incentives than other businesses to maintain a risky financial

structure with a relatively high proportion of debt with a short maturity. In a new steady state, this will

still be the case. Nonetheless, Figure 9.8 shows that a few decades ago Dutch banks were financed

much more with equity and also had more liquid assets on their balance sheets. This shift is an

international phenomenon which can be attributed partly to the growing national and international

competition between banks.75 It also illustrates that a return to a less risky funding structure would not

be abnormal from an historical perspective.

75 See e.g. Berger et al. (1995) and Greenspan (2010) for discussions of the changed financial structure of the US banking system, which is comparable with the picture outlined here for the Netherlands.

184

9.5.2 More stable economic development

The stricter standards also offer benefits even when there is no crisis. For example, banks with higher

capital and liquidity buffers are better able to support businesses and households in bad times. Buffers

enhance the capacity of banks to absorb losses and uphold lending during a downturn. Various studies

show that better capitalised and more liquid banks are generally more willing to lend (see, for instance,

Jiménez et al., 2010). In a booming economy, the stricter regulatory framework gives banks an

incentive to reduce risk. This can help prevent banks from feeding excessive asset price developments,

thereby moderating business cycle fluctuations and reducing volatility in financial markets.

The Basel Committee proposals devote special attention to reducing pro-cyclical effects. The

new regulatory framework encourages banks to build up an extra capital conservation buffer in good

times by means of restrictions on profit distributions (including dividend payouts, share repurchase

programmes and bonus payments to staff). The restrictions will be designed in such a way that they

become stricter as the capital ratio approaches the minimum level, whereas at around the target value

they will lose most of their binding capacity, so as to avoid ‘cliff effects’. As the restrictions will be

linked to profitability, banks will be encouraged to build up extra capital when they are in the best

position to do so. In bad times – when the banks are making losses – they will be able to address these

buffers and may therefore move less quickly to tighten up their lending criteria.

A fixed target value will be set for the capital conservation buffer, but this could be raised

temporarily depending on the macroeconomic conditions. The ratio between total credit and GDP is an

185

important criterion here; empirical research shows this to be a key leading indicator for financial

imbalances and crises (Drehmann et al., 2010).

The intention is that this mechanism should prevent excesses during good times and help

banks maintain their lending in bad times and therefore to stabilise economic growth. Research on the

benefits of counter-cyclical capital buffers suggests that this regime does indeed reduce the volatility

of GDP. Several studies suggest that the standard deviation of GDP reduces by around a fifth

compared with a baseline scenario in which there is no counter-cyclical buffer (BCBS, 2010a).

9.6 Conclusions

Several national and international model calculations indicate that the negative impact on real GDP

during the transitional phase to higher capital and liquidity buffers will be limited to a few tenths of a

percent. Lending wedges are likely to be permanently higher, but the impact of this on credit volumes

will be limited to a few percent because banks will have more options to adapt to the new

requirements. The model outcomes for the Netherlands are in line with research for other countries. A

sufficiently long transitional period will help limit the costs in the early years, because it will give

banks more scope to adapt. It will also make it easier for markets to absorb the additional demand for

capital and liquidity. Once the banks have adapted to the new situation, the benefits of a more solid

financial system will outweigh the disadvantages. The higher buffers will make a financial crisis in the

future both less likely and less deep. Furthermore, economic growth will be more stable, because the

new regulatory standards will make the banks’ reactions less pro-cyclical. The BCBS (2010a)

accordingly concludes that, all in all, the benefits of stricter capital and liquidity requirements far

outweigh the costs.

186

Appendix 9.1 Impulse responses of shock in target capital ratio

(response to 1 standard deviation Cholesky innovation ± 2 standard errors)

187

Appendix 9.2 Impulse responses of shock in bank lending standards

(response to 1 standard deviation Cholesky innovation ± 2 standard errors)

188

189

Chapter 10

Summary and conclusions

10.1 Introduction

The 2007-2009 financial crisis has provided a rich set of data that may help us to understand the

dynamics of credit and liquidity risk in stress conditions. It sheds light on the role of banks’ behaviour

in the emergence of second round effects in the financial system and the economy. By using these

data, this thesis bridges the gap between the theoretical literature on behaviour in financial markets

and policy oriented analyses of credit and liquidity risks as, for instance, reported on in financial

stability reports.

In line with the macroprudential approach, which examines the stability of the financial

system as a whole, the empirical analysis in this thesis is conducted from a macro perspective. This

first principle of the analysis is the focus on extreme events and dynamics that can give rise to

systemic dimensions of credit and liquidity risks in the banking sector. To grasp how such risks may

unfold, a thorough understanding of banks’ behaviour on a micro level in relation to macro-financial

developments is needed. This calls for the use of granular data to capture the variations at the portfolio

or bank level and the differing responses of institutions. This is taken as the second principle

throughout the thesis. Furthermore, the crisis has underscored that systemic risk can originate through

the interaction of credit and liquidity risk and, more in particular, at the nexus between funding and

market liquidity. Taking that into account, a multi-dimensional and eclectic approach is applied as a

third principle in the analyses. This is operationalised by a suite of models in which the various risk

factors, which determine credit and liquidity risk, are dynamically related. Another feature of credit

and liquidity risk in tail events is the inherent uncertainty with regard to the first round impact of

adverse shocks and the second round effects that could occur in the financial system or the economy.

Hence, as a fourth principle in the thesis it is recognized that the model results are inherently

uncertain. This is taken into account by the use of scenario analyses and the presentation of model

outcomes in terms of loss distributions. Furthermore, uncertainty was a main issue for governments

and central banks when they designed their crisis measures between 2007 and 2009. This uncertainty

is accounted for in the assessment of the crisis measures. The four principles (the macro perspective,

use of granular data, a multi-dimensional approach, recognizing uncertainty) are key building blocks

throughout the thesis and provide the tools to analyse the three research questions:

190

1. How did banks adjust their credit and liquidity risk management during the crisis and how do

empirical estimates of banks’ reactions relate to the behavioural assumptions as generally used in

the theoretical literature?

2. How can the impact of tail events on banks that involve credit and liquidity risk, and banks’

reactions to those risks, be modelled?

3. How should the policy responses to the eruption of credit and liquidity risks during the 2007-2009

crisis be assessed, both with regard to the possible distortionary effects on behavioural incentives

and the impact on the economy?

10.2 Bank behaviour

The first research question is addressed by constructing indicators and time series models that analyse

banks’ reactions empirically. These are based on firm-specific data of Dutch banks, derived from a

unique data source on assets and liabilities available at De Nederlandsche Bank (DNB). In Chapter 2

these micro observations are used to construct indicators, which describe general trends in bank

behaviour. These measures capture both the time dimension (‘pro-cyclicality’) and the cross-sectional

dimension (‘dependencies’) of systemic risk. Although they are descriptive in nature, the measures

identify trends in behaviour that convey forward looking information on market-wide developments.

Chapter 3 extends the empirical analysis by modelling bank behaviour in a panel Vector

Autoregressive (VAR) framework, based on the same source of data of Dutch banks. Both chapters

concentrate on several main assumptions in the literature on banks behaviour in crises, like leveraging,

herding, the pecking order in balance sheet adjustments, and on the relation between funding liquidity

and lending, liquidity hoarding and fire sales.

The following conclusions can be drawn. First, the result concerning leveraging behaviour

indicates that balance sheet adjustments of the banks tend to be asymmetric; deleveraging was more

intense in the bust (mid 2007 to early 2009) than leveraging was in the preceding boom. Furthermore,

during the crisis the deleveraging of large banks started earlier, was more intense and more advanced

than the deleveraging of smaller banks. Second, the results on herding behaviour show that the number

and similarity of banks’ reactions substantially increased in the crisis. This confirms the herding

assumption in the literature and reflects the pro-cyclical nature of bank behaviour. Third, the results on

the pecking order of balance sheet adjustments confirm that banks usually follow a pecking order (by

making larger adjustments to the most liquid balance sheet items compared to less liquid items), but

that they are more inclined to a static response in crises. Fourth, the outcomes of the panel VAR model

for the link between shocks to funding liquidity and lending show that banks react by reducing

wholesale lending in response to shocks to money market spreads and repo funding. Fifth, the VAR

simulations show that in response to wholesale funding shocks banks hoard liquidity through

191

accumulating liquid bond holdings and increasing their reliance on the central bank. Finally, the model

results show that fire sales of equity holdings are more likely to be triggered by constraints in funding

liquidity than by constraints in the solvency position of banks.

The empirical results contribute to our understanding of the role of banks in the transmission

of adverse shocks to the financial system and the economy. The findings can also help to improve the

micro foundations of financial stability models, especially with regard to the behavioural assumptions

of heterogeneous institutions. And the results provide useful insights for macroprudential monitoring

frameworks.

10.3 Macro stress-testing models

The second research question concerns modelling the impact on banks of tail events that involve credit

and liquidity risk and banks’ reactions to those risks. The question is addressed in a stress-testing

framework. This provides for methodologies to map tail events in the macro-environment into

indicators that can be used to estimate the implications for banks’ balance sheets, their responses to

stress situations and the related second round effects in the financial system and the economy. The

framework is operationalised by a suite of models, such as reduced form satellite models, vector

autoregressive (VAR) models and calibrated simulation tools. This eclectic approach is motivated by

the existence of fundamental uncertainty with regard to risks in tail situations and by the absence of a

fully fledged model that integrates all the potential interlinkages.

The overview of macro stress-testing methods in Chapter 4 is followed by several applications

of models for stress-testing credit and liquidity risk in Chapters 5, 6 and 7. Chapter 5 concentrates on

scenario analysis and macro stress-testing of credit risk. More in particular, a multi-factor approach is

used to simulate deterministic and stochastic scenarios, which take into account simultaneous changes

in macro variables and changing correlations between risk factors. Both are typical features of stress

situations. The link between the macro variables and micro risk drivers of banks’ portfolios is

established in reduced form satellite models. These models are estimated using disaggregated data of

individual banks and portfolio break-downs, to capture the different responses of banks and portfolio

sensitivities in stress situations. The variation in credit loss distributions is explored by estimating both

the probability of default and the loss given default in bank loan books. This is based on nonlinear

specifications, since ignoring nonlinearities in the relationship between macro variables and credit risk

can lead to a substantial underestimation of risk, particularly when considering large shocks. The

outcomes of the stochastic scenario simulations are presented in terms of loss distributions. They

provide insight in the extreme losses that banks could face in stress events.

Chapter 6 presents the framework of a stress-testing model for liquidity risk of banks, which is

extended in Chapter 7 with the new liquidity regulation of Basel III and the possible interactions with

192

monetary policy operations and credit supply by banks. The liquidity stress-testing model basically is

a calibrated simulation tool which combines the multiple dimensions of liquidity risk (funding and

market liquidity risk) into a quantitative measure of banks’ liquidity position. The model takes into

account the first and second round (feedback) effects of shocks, induced by reactions of heterogeneous

banks. Moreover, it allows for simulating the impact of reputation risk on banks that react in order to

restore their deteriorated liquidity position. A Monte Carlo approach is used to simulate the impact of

stress scenarios on the liquidity buffers of banks, including the probability of a liquidity shortfall. The

main outcome of the model simulations is that the second round effects in specific scenarios could

have more impact than the first round effects and hit all types of banks. This confirms that shocks to

the liquidity position of banks entail systemic risk through the behavioural responses of individual

banks. While the addition of credit supply effects in Chapter 7 is an attempt to link liquidity risk to

credit risk, a more complete integration of credit and liquidity risk is an area for future work. It

requires deeper analysis of the potential linkages and interactions between credit quality on the one

hand and funding and market liquidity on the other hand. Such research will, amongst other issues,

have to deal with the different nature of the risks (for instance with regard to the driving forces and the

time horizon) and with the behaviour of counterparties that could link liquidity to credit risk.

The extended version of the model assumes that the new liquidity regulation proposed by

Basel III is a binding constraint for banks’ behaviour. This is operationalised by the rule that banks

restore their liquidity ratio through raising additional liquid assets and improving the stability of

funding, as a reflection of the liquidity hoarding and the scramble for stable funding sources by banks

during the crisis. The model simulations provide quantifications of the potential wider effects of the

new liquidity regulation, for instance with regard to the impact on credit supply. This impact is found

to be limited in the model simulations of liquidity stress scenarios. Another result is that second round

effects and tail risks of a stress scenario are substantially lower if banks would adjust to Basel III by

holding a higher quality of liquid assets. In particular a narrowly defined liquidity buffer - made up by

high quality government bonds - makes a big difference in limiting the tail risks of banks. The flip side

of larger bond holdings is that monetary policy conducted through asset purchases has more impact on

banks’ behaviour, relative to central bank refinancing operations. We also simulate the consequences

of an exit from extended refinancing operations on banks’ funding liquidity. The outcomes indicate

that the liquidity ratios of banks actually improve compared to the pre-exit situation, if alternative

stable funding is available.

193

10.4 Policy responses to the crisis

The third research question, on the policy responses to the credit and liquidity risks of banks in the

crisis, is first addressed by assessing the short-term crisis measures taken by central banks and

governments in 2007-2009 (in Chapter 8) and secondly by analysing the effects of longer-term

measures taken by regulators, in particular the macroeconomic effects of Basel III (in Chapter 9).

To preserve financial stability and limit the impact of financial stress on the economy, central

banks and governments had to respond swiftly and under great uncertainty to the eruption of risks in

the banking sector. Although the measures were successful as they contributed to safeguarding

financial stability, they also had distortionary effects. Potential distortions relate to the level playing

field between supported and non-supported institutions and the capital flows between market

segments, including cross-border flows. However, the influence of government interventions here

appears to be hard to disentangle from market incentives. We do find some evidence of a short-term

favourable effect of government support in the market prices of financial institutions, but this faded

away several months after the interventions took place and even turned into a negative effect. This

concurs with the longer term disadvantages of the interventions by governments and central banks,

which may create wrong incentives with regard to risk taking and moral hazard of market participants.

The main policy conclusion of the analysis is that such negative side effects can be limited through

appropriate design of the support measures (market compatibility) and of the exit strategy (timely

withdrawal).

Regulation also plays an important role in influencing the behaviour of banks, both in the

transitional phase and in the new steady state in which the banks comply with the new regulation. The

new capital and liquidity standards of Basel III will affect the intermediation function of banks, and

thereby credit supply and economic growth. Simulation outcomes of reduced form satellite models and

DNB’s structural macroeconomic model indicate that the negative impact of Basel III on real GDP

will be limited to a few tenths of a percent during the transitional phase. Another result is that a

sufficiently long transitional period will limit the costs in the early years, because it gives banks more

scope to adapt. Moreover, once the banks have adapted to the new standards, the benefits of a more

solid financial system will outweigh the disadvantages. Although a precise quantification of the

benefits is complicated because of the uncertain effects of Basel III on the strategies of banks and their

clients, it seems obvious that higher buffers of banks make a financial crisis in the future both less

likely and less deep, while economic growth would be more stable in normal times.

194

195

References

Acharya, V. and C. Merrouche (2009), Precautionary Hoarding of Liquidity and Inter-Bank Markets:

Evidence from the Sub-prime Crisis, NYU Working Paper, 09-018.

Acharya, V., D. Gromb and T. Yorulmazer (2008), Imperfect Competition in the Interbank Market for

Liquidity as a Rationale for Central Banking, Centre for Economic Policy Research Discussion

Paper 6984.

Adrian, T. and H.S. Shin (2008a), Liquidity and Financial Cycles, BIS Working Paper 256.

Adrian, T. and H.S. Shin (2008b), Liquidity and Financial Contagion, Banque de France, Financial

Stability Review 11, 1-7.

Adrian, T. and H.S. Shin (2010), Liquidity and Leverage, Journal of Financial Intermediation, 19 (3),

418-437.

Agur, I. and M. Demertzis (2010), A Model of Monetary Policy and Bank Risk Taking, European

Banking Center Discussion Paper, 2010-06S, CentER Discussion Paper Series, 2010-30S.

Aikman, D., P. Alessandri, B. Eklund, P. Gai, S. Kapadia, E. Martin, N. Mora, G. Sterne and M.

Willison (2009), Funding Liquidity Risk in a Quantitative Model of Systemic Stability, Bank of

England Working Paper 372.

Alessandri, P., Gai, P., Kapadia, S., Mora, N. and C. Puhr (2009), Towards a Framework for

Quantifying Systemic Stability, International Journal of Central Banking 5 (3), 47-82.

Alfon, I., I. Argimon and P. Bascunana-Ambros (2004), What Determines How Much Capital Is Held

by UK Banks and Building Societies, FSA Occasional Paper 22.

Allen, F. and D. Gale (1998), Optimal Financial Crises, Journal of Finance 53, 1245-1283.

Allen, F. and D. Gale (2000), Financial Contagion, Journal of Political Economy 108 (1), 1-33.

Altman, E.I., A. Resti and A. Sironi (2004), Default Recovery Rates in Credit Risk Modelling: A

Review of the Literature and Empirical Evidence, Economic Notes 33 (2), 183-208.

Angelini, P., L Clerc, V. Cúrdia, L. Gambacorta, A. Gerali, A. Locarno, R. Motto, W. Roeger, S. Van

den Heuvel and J. Vlček (2011), Basel III: Long-Term Impact on Economic Performance and

Fluctuations, Federal Reserve Bank of New York Staff Reports 485.

Arellano, M. and O. Bover (1995), Another Look at the Instrumental Variable Estimation of Error

Component Models, Journal of Econometrics 68, 29–51.

Artha, I.K .D. S and J. de Haan (2011), Financial Crises and the Dismissal of Central Bank Governors,

New Evidence, Working Paper.

Åsberg, P. and H. Shahnazarian (2008), Macroeconomic Impact on Expected Default Frequency,

Sveriges Riksbank Working Paper 219.

Aspachs, O., C. Goodhart, M. Segoviano, D. Tsomocos and L. Zicchino (2006), Searching for a

Metric for Financial Stability, LSE Financial Markets Group Special Paper Series 16.

196

Bank of Japan (2007), The Framework for Macro Stress-Testing of Credit Risk: Incorporating

Transition in Borrower Classifications, Financial System Report, September 2007.

Banque de France (2008), Overview, Financial Stability Review 11 (special issue), I–IV.

Barell, R., E.P. Davis, T. Fic, D. Holland, S. Kirby and I. Liadze (2009), Optimal Regulation of Bank

Capital and Liquidity: How to Calibrate New International Standards, FSA Occasional Paper 38.

Bayoumi, T. and O. Melander (2008), Credit Matters: Empirical Evidence on US Macro-Financial

Linkages, IMF Working Paper 08/169.

BCBS, Basel Committee on Banking Supervision (2004), Principles for the Management and

Supervision of Interest Rate Risk.

BCBS, Basel Committee on Banking Supervision (2005), Guidance on Paragraph 468 of the

Framework Document.

BCBS, Basel Committee on Banking Supervision (2006), Basel II: International Convergence of

Capital Measurement and Capital Standards: A Revised Framework.

BCBS, Basel Committee on Banking Supervision (2008a), Liquidity Risk: Management and

Supervisory Challenges.

BCBS, Basel Committee on Banking Supervision (2008b), Principles for Sound Liquidity Risk

Management and Supervision.

BCBS, Basel Committee on Banking Supervision (2009a), Principles for Sound Stress Testing

Practices and Supervision.

BCBS, Basel Committee on Banking Supervision (2009b), International Framework for Liquidity

Risk Measurement, Standards and Monitoring, Consultative Document.

BCBS, Basel Committee on Banking Supervision (2009c), Strengthening the Resilience of the

Banking Sector, Consultative Document.

BCBS, Basel Committee on Banking Supervision (2010a), An Assessment of the Long-Term

Economic Impact of the New Regulatory Framework.

BCBS, Basel Committee on Banking Supervision (2010b), Appendix.

BCBS, Basel Committee on Banking Supervision (2010c), Basel III: A Global Regulatory Framework

for More Resilient Banks and Banking Systems.

BCBS, Basel Committee on Banking Supervision (2011), The Transmission Channels between the

Financial and Real Sectors: A Critical Survey of the Literature, Working Paper 18.

BCBS, Basel Committee on Banking Supervision and IADI, International Association of Deposit

Insurers (2009), Core Principles for Effective Deposit Insurance Systems.

Bech, M., J.T.E. Chapmanz and R. Garratt (2008), Which Bank Is the Central Bank? An Application

of Markov Theory to the Canadian Large Value Transfer System, Federal Reserve Bank of New

York, Staff Report 356.

Beck, T., D. Coyle, M. Dewatripont, X. Freixas and P. Seabright (2010), Bailing Out the Banks:

Reconciling Stability and Competition, CEPR, London.

197

Berben, R.P., B. Bierut, J.I. Kakes and J.W. Van den End (2010), Macro-effects of Higher Capital and

Liquidity Requirements for Banks: Empirical Evidence for the Netherlands, DNB Occasional

Studies 8 (3).

Berger, A.N. and C.H.S. Bouwman (2008), Financial Crises and Bank Liquidity Creation, Working

Paper.

Berger, A.N. and C.H.S. Bouwman (2009), Bank Liquidity Creation, The Review of Financial Studies,

22 (9), 3779-3837.

Berger, A.N., J. Herring and G.P. Szegö (1995), The Role of Capital in Financial Institutions, Journal

of Banking and Finance 19, 393-430.

Bernanke, B., M. Gertler and S. Gilchrist (1996), The Financial Accelerator and the Flight to Quality,

Review of Economics and Statistics, 78 (1), 1-15.

Berrospide, J.M. and R.M. Edge (2009), Linkages between the Financial and Real Sectors: Some

Lessons from the Subprime Crisis, manuscript.

Bikhchandani S. and S. Sharme (2001), Herd Behaviour in Financial Markets, IMF Staff Paper 47, 3.

Bikker. J.A. and H. Hu (2002), Cyclical Patterns in Profits, Provisioning and Lending of Banks and

Procyclicality of the New Basel Capital Requirements, Banca Nazionale del Lavoro Quarterly

Review 55, 143-157.

BIS, Bank for International Settlements (2009a), Statistical Commentary on Provisional Locational

and Consolidated International Banking Statistics at end-December 2008.

BIS, Bank for International Settlements (2009b), 79th Annual Report.

BIS, Bank for International Settlements (2009c), Monetary Policy and the Risk-taking Channel,

Quarterly Review, December 2009.

BIS, Bank for International Settlements (2010a), 80th Annual report.

BIS, Bank for International Settlements (2010b), Assessing the Macroeconomic Impact of the

Transition to Stronger Capital and Liquidity Requirements, Final Report Macroeconomic

Assessment Group.

Blaschke, W., M.T. Jones, G. Majnoni and M. Peria (2001), Stress Testing of Financial Systems: an

Overview of Issues, Methodologies and FSAP Experiences, International Monetary Fund.

Board of Governors of the Federal Reserve System (2009), The Supervisory Capital Assessment

Program: Overview of Results.

Borio, C.E.V and M. Drehmann (2009), Towards an Operational Framework for Financial Stability:

"Fuzzy" Measurement and its Consequences, BIS Working Papers 284.

Borio, C.E.V. (2006), The Macroprudential Approach to Regulation and Supervision: Where Do We

Stand?, Paper prepared for the special volume celebrating the 20th anniversary of Kredittilsynet.

Borio, C.E.V. (2009), Implementing the Macroprudential Approach to Financial Regulation and

Supervision, Financial Stability Review, September 2009, Banque de France.

198

Borio, C.E.V., N.A. Tarashev and K. Tsatsaronis (2010), Attributing Systemic Risk to Individual

Institutions, BIS Working Papers 308.

Boss, M., G. Krenn, C. Puhr and M. Summer (2006), Systemic Risk Monitor: A Model for Systemic

Risk Analysis and Stress Testing of Banking Systems, Financial Stability Report 11, OeNB, 83–

95.

Boyd, J, S. Kwak and B. Smith (2005), The Real Output Losses Associated with Modern Banking

Crises, Journal of Money, Credit and Banking 37, 977-999.

Bracoud, F. (2002), Sequential Models of Bertrand Competition for Deposits and Loans under

Asymmetric Information, Keele Economics Research Papers 15.

Breuer, T., M. Jandacka, K. Rheinberger and M. Summer (2009), How to Find Plausible, Severe, and

Useful Stress Scenarios, International Journal of Central Banking 5 (3), 205-224.

Brunnermeier, M.K. (2001), Asset Pricing under Asymmetric Information: Bubbles, Crashes,

Technical Analysis, and Herding, Oxford University Press.

Brunnermeier, M.K. and L.H. Pederssen (2009), Market Liquidity and Funding Liquidity, Review of

Financial Studies 22 (6), 2201-2238.

Brunnermeier, M.K. (2008), Deciphering the 2007-08 Liquidity and Credit Crunch, Journal of

Economic Perspectives 23 (1), 77-100.

Bundesbank (2006), The Deutsche Bundesbank’s Stress-Testing Approach: Scenarios, Methodology

and Other Issues, mimeo.

Caballero, R. (2008), Moral Hazard Misconception, Financial Times, 14 July 2008.

Cassola, N., C. Holthausen and M. L. Duca (2010), The 2007/2009 Turmoil: a Challenge for the

Integration of the Euro Area Money Market?, ECB mimeo.

CEBS, Committee of European Banking Supervisors (2009a), Guidelines on Liquidity Buffers &

Survival Periods.

CEBS, Committee of European Banking Supervisors (2009b), Aggregate Outcome of the 2010 EU-

Wide Stress Test Exercise Coordinated by CEBS in Cooperation with the ECB, summary report.

Cecchetti, S.G., M. Kohler and C. Upper (2009), Financial Crises and Economic Activity, paper

presented at the Federal Reserve Bank of Kansas City’s symposium at Jackson Hole, Wyoming,

August 2009.

Cifuentes, R. G. Ferrucci and H. S. Shin (2005), Liquidity Risk and Contagion, Journal of the

European Economic Association 3, 556-66.

Čihák, M. (2007), Introduction to Applied Stress-Testing, IMF Working Paper, 07/59.

Čihák, M. and T. Poghosyan (2009), Distress in European Banks: An Analysis Based on a New Data

Set, IMF Working Paper 09/9.

Coffinet, J., S. Lin and C. Martin (2009), Stress Testing French Banks’ Income Subcomponents,

Banque de France Working Paper 242.

199

Coleman, N., N. Esho, I. Sellathurai and N. Thavabalan (2005), Stress Testing Housing Loan

Portfolios: A Regulatory Case Study, APRA Working Paper 2005-01.

Cornett, M.M., J. J. McNutt and P. E. Strahan (2010), Liquidity Risk Management and Credit Supply

in the Financial Crisis, Working paper, Bentley University.

De Bandt, O. and V. Oung (2004), Assessment of Stress Tests Conducted on the French Banking

System, Banque de France Financial Stability Review, November 2004.

De Graeve, F, T. Kick and M. Koetter (2008), Monetary Policy and Financial (In)Stability: An

Integrated Micro-Macro Approach, Journal of Financial Stability 4, 205-231.

De Haan, J, J. E. Sturm and E. Zandberg (2009), The Impact of Financial and Economic Crises on

Economic Freedom, in: Economic Freedom of the World, 2009 Annual Report.

De Haan, L. and J.W. Van den End (2011), Banks’ Responses to Funding Liquidity Shocks: Lending

Adjustment, Liquidity Hoarding and Fire Sales, DNB Working Paper 293.

De Haas, R. and N. Van Horen (2011), Running for the Exit: International Banks and Crisis

Transmission, DNB Working Paper 279.

De Larosière Report (2009), Report of the High-level Group on Financial Supervision in the EU.

Demirgüç-Kunt. A. and E. Detragiache (2000), Does Deposit Insurance Increase Banking System

Stability?, IMF Working Paper 00/3.

Diamond, D.W. and R.G. Rajan (2005), Liquidity Shortages and Banking Crises, Journal of Finance

60 (2), 615-647.

Diamond, D.W. and R.G. Rajan (2009), The Credit Crisis: Conjectures about Causes and Remedies,

American Economic Review 99 (2), 606-610.

DNB, De Nederlandsche Bank (2003), Credit System Supervision Manual, articles 44-47, 2.3

Liquidity Risk.

DNB, De Nederlandsche Bank (2005), Overview of Financial Stability in the Netherlands 3,

December 2005.

DNB, De Nederlandsche Bank (2007), Overview of Financial Stability in the Netherlands 5, March

2007.

DNB, De Nederlandsche Bank (2010a), Closer Supervisory Attention to Business Models and

Strategies, Quarterly Bulletin, September 2010.

DNB, De Nederlandsche Bank (2010b), Overview of Financial Stability in the Netherlands 12,

November 2010.

DNB, De Nederlandsche Bank (2011), DELFI: DNB’s Macroeconomic Policy Model of the

Netherlands, DNB Occasional Studies 9 (1).

Doytch, R. and N. Uctum (2011), Does the Worldwide Shift of FDI from Manufacturing to Services

Accelerate Economic Growth? A GMM Estimation Study, Journal of International Money and

Finance, 30, 410-427.

200

Drehmann, M, A. Patton and S. Sorensen (2005), Corporate Defaults and Large Macroeconomic

Shocks, Bank of England, mimeo.

Drehmann, M. (2005), A Market Based Macro Stress Test for the Corporate Credit Exposures of UK

Banks, Bank of England, draft paper.

Drehmann, M. and Nikolaou (2010), Funding Liquidity Risk: Definition and Measurement, BIS

Working Papers 316.

Drehmann, M., C. Borio, L. Gambacorta, G. Jiménes and C. Trucharte (2010), Countercyclical Capital

Buffers: Exploring options, BIS Working papers 317.

Drehmann, M., G. Hoggarth, A. Logan and L. Zicchino (2004), Macro Stress Testing UK Banks,

mimeo, Bank of England.

Drehmann, M., S. Sorensen and Stringa (2006), Integrating Credit and Interest Rate Risk: Theoretical

Framework and Application, mimeo, Bank of England.

Düllmann, K. and M. Erdelmeier (2009), Stress Testing German Banks in a Downturn in the

Automobile Industry, Deutsche Bundesbank Discussion Paper, Series 2, Banking and Financial

Studies 02/2009.

ECB, European Central Bank (2006), Macro Stress-Testing Practices, Financial Stability Review, June

2006.

ECB, European Central Bank (2007), The Collateral Frameworks of the Federal Reserve System, the

Bank of Japan and the Eurosystem, ECB Monthly Bulletin, October 2007.

ECB, European Central Bank (2008a), EU banks’ Liquidity Stress Testing and Contingency Funding

Plans.

ECB, European Central Bank (2008b), Recommendations of the Governing Council of the European

Central Bank on Government Guarantees for Bank Debt.

ECB, European Central Bank (2009a), National Rescue Measures in Response to the Current

Financial Crisis, Legal Working Paper 8.

ECB, European Central Bank (2009b), Guiding Principles for Bank Asset Support Schemes.

ECB, European Central Bank (2009c), Monetary Policy and Loan Supply in the Euro Area, Monthly

Bulletin, October 2009.

ECB, European Central Bank (2009d), EU Banks’ Funding Structures and Policies.

ECB, European Central Bank (2010a), Report on Deleveraging in the EU Banking Sector.

ECB, European Central Bank (2010b), Stress-Testing Banks in the Crisis, Special Feature, Financial

Stability Review, December 2010.

ECB, European Central Bank (2010c), Monthly Bulletin, December 2010.

ECB, European Central Bank (2010d), Financial Stability Review, June 2010.

Eisenschmidt, J. and Tapking (2009), Liquidity Risk Premia in the Unsecured Interbank Money

Market, ECB Working Paper 1024.

201

Elliot, D.J. (2009), Quantifying the Effects on Lending of Increased Capital Requirements, Briefing

Paper 7, Brookings Institution.

Enders, W. (1995), Applied Econometric Time Series, John Wiley & Sons.

EC, European Commission (2009a), 23.07.2009 Commission Communication the Return to Viability

and the Assessment of Restructuring Measures in the Financial Sector in the Current Crisis under

the State Aid Rules.

EC, European Commission (2009b), DG Competition's Review of Guarantee and Recapitalisation

Schemes in the Financial Sector in the Current Crisis.

Fernando, C.S., R.J. Herring and A. Subrahmanyam (2008), Common Liquidity Shocks and Market

Collapse: Lessons from the Market for Peps, Journal of Banking and Finance 32, 1625-35.

Financial Times (2009), Zoellick warns of fallout from bank guarantees, 27 January 2009.

Fiori, R., A. Foglia and S. Iannotti (2009), Beyond Macroeconomic Risk: the Role of Contagion in the

Italian Corporate Default Correlation, Bocconi University CAREFIN Working Paper 2.

Foglia, A. (2009), Stress-testing Credit Risk: a Survey of Authorities’ Approaches, International

Journal of Central Banking 5 (3), 10-45.

Francis, W. and M. Osborne (2009), Bank Regulation, Capital and Credit Supply: Measuring the

Impact of Prudential Standards, FSA Occasional Paper 36.

Freixas, X., B.M. Parigi and J. C. Rochet (2000), Systemic Risk, Interbank Relations, and Liquidity

Provision by the Central Bank, Journal of Money, Credit and Banking 32, 611–38.

Frye, J. (2000), Depressing Recoveries, Policy Studies, Emerging Issues Series S&R-2000-8, Federal

Reserve bank of Chicago.

FSA, Financial Services Authority (2006), Risk Outlook 2006.

FSA, Financial Services Authority (2007), Review of the Liquidity Requirements for Banks and

Building Societies, Discussion Paper 07 (07), London.

FSA, Financial Services Authority (2009), FSA Statement of its Use of Stress Tests.

FSF, Financial Stability Forum (2008), Enhancing Market and Institutional Resilience, Report of the

Financial Stability Forum.

G20, Group of 20 (2009), Declaration to Strengthening the Financial System, London Summit, 2 April

2009.

Giannone, D., M. Lenza, H. Pill and L. Reichlin (2011), Non-Standard Monetary Policy Measures and

Monetary Developments, ECB Working Paper 1290.

Gale, D. and T. Yorulmazer (2011), Liquidity Hoarding, Federal Reserve Bank of New York Staff

Report 488.

Gauthier, C, Z. He and M. Soussi (2010), Understanding Systemic Risk: The Trade-Offs between

Capital, Short-Term Funding and Liquid Asset Holdings, Bank of Canada Working Paper 2010-

29.

202

Gauthier, C., A. Lehar and M. Soussi (2010), Macroprudential Regulation and Systemic Capital

Requirements, Bank of Canada Working Paper 2010-4.

González-Páramo, J.M. (2010), The European Central Bank and the Policy of Enhanced Credit

Support, Speech for Conference organised by Cámara de Comercio de Málaga and University of

Málaga, Málaga, 18 June 2010.

González-Páramo, J.M. (2009), Liquidity, Funding and Solvency: Policy Responses and Lessons,

Speech delivered at the Universidad de Alcalá de Henares, Madrid, 16 January 2009.

Goodhart, C. and M.A. Segoviano (2009), Banking Stability Measures, IMF Working Paper 09/4.

Goodhart, C., P. Sunirand and D. Tsomocos (2006), A Model to Analyse Financial Fragility,

Economic Theory 27, 107–42.

Greenspan, A. (2010), The Crisis, mimeo, http://www.brookings.edu/~/media/files/

programs/es/bpea/2010_spring_bpea_papers/ spring2010_greenspan.pdf.

Haldane, A.G. (2009), Why Banks Failed the Stress Test, Speech by Mr Andrew G Haldane, at the

Marcus-Evans Conference on Stress-Testing, London, 9-10 February 2009.

Haldane, A.G. (2010), The $100 Billion Question, Comments at the Institute of Regulation & Risk,

Hong Kong, 30 March 2010.

Heider, F., M. Hoerova and C. Holthausen (2009), Liquidity Hoarding and Interbank Market Spreads

The Role of Counterparty Risk, ECB Working Paper 1126.

Hoeberichts, M.M., M. Tabbae and J.W. Van den End (2006), Modelling Scenario Analysis and

Macro Stress-Testing, DNB Working Paper 119.

Hoggarth, G., L. Mahadeva and J. Martin (2010), Understanding International Bank Capital Flows

During the Recent Financial Crisis, Financial Stability Paper 8, Bank of England.

Hoggarth, G., S. Sorensen and L. Zicchino (2005), Stress Tests of UK Banks Using a VAR Approach,

Bank of England Working Paper 282.

Hommes, C. H. (2006), Heterogeneous Agent Models in Economics and Finance, in Tesfatsion, L. and

K. L. Judd (eds.), Handbook of Computational Economics 1 (2), 1109-1186, Elsevier.

Honkapohja, S. (2009), The 1990’s Financial Crises in Nordic Countries, Bank of Finland Research

Discussion Papers 5.

Huang, R. and L. Ratnovski (2010), The Dark Side of Bank Wholesale Funding, IMF Working Paper

10/170.

Hui, C.H. and E. Wong (2008), A Liquidity Risk Stress-Testing Framework with Interactions between

Market and Credit Risks, Hong Kong Monetary Authority Working Paper 6.

ICMA, International Capital Market Association (2009), Survey of European Repo Market Reflects

Deleveraging by Banks.

203

ICMA, International Capital Market Association, ERC, European Repo Council (2010), White Paper

on the Operation of the European Repo Market, July 2010.

IIF, Institute of International Finance (2010), Interim Report on the Cumulative Impact on the Global

Economy of Proposed Changes in the Banking Regulatory Framework.

IMF, International Monetary Fund (2000), Macroprudential Indicators of Financial System Soundness,

Occasional Paper 192.

IMF, International Monetary Fund (2008a), Financial Soundness Indicators.

IMF, International Monetary Fund (2008b), Global Financial Stability Report, April 2008.

IMF, International Monetary Fund (2009a), Global Financial Stability Report, Market Update, January

2009.

IMF, International Monetary Fund (2009b), Global Financial Stability Report, October 2009.

IMF, International Monetary Fund (2009c), Crisis-Related Measures in the Financial System and

Sovereign Balance Sheet Risks, July 2009.

IMF, International Monetary Fund (2010a), Exiting from Monetary Crisis Intervention Measures,

Background Paper.

IMF, International Monetary Fund (2010b), Global Financial Stability Report, October 2010.

IMF, International Monetary Fund (2010c), Central Banking Lessons form the Crisis.

IMF, International Monetary Fund (2010d), Understanding Financial Interconnectedness, Staff Paper.

Ivashina, V. and D. Scharfstein (2010), Bank Lending During the Financial Crisis of 2008, Journal of

Financial Economics 97, 319-338.

Jain, A.K. and S. Gupta (1987), Some Evidence on Herding Behaviour of U.S. Banks, Journal of

Money, Credit and Banking 19(1), 78-89.

Jiménez, G., J. L. Peydro, S. Ongena and J. Saurina (2010), Credit Availability: Identifying Balance

Sheet Channels with Loan Applications and Granted Loans, Banco De España, Working Paper

1030.

Jones, M.T., P. Hilbers and G. Slack (2004), Stress Testing Financial Systems: What to Do When the

Governor Calls, IMF Working Paper 04/127.

Kashyap, A. K. and R.G. Rajan (2008), Rethinking Capital Regulation, conference paper Jackson

Hole.

King, M.R. (2009), Time to Buy or Just Buying Time? The Market Reaction to Bank Rescue

Packages, BIS Working Papers 288.

Knight (1921), Risk, Uncertainty and Profit, Houghton Mifflin Company.

Kwast, M.L., S. Holden, D. Jurcevic and I. Van Lelyveld (2010), Norges Bank Stress Testing of

Credit Risks, Report of an External Review Panel.

Laeven, L. and F. Valencia (2008), Systemic Banking Crises: A New Database, IMF Working Paper

09/224.

204

Laeven, L. and F. Valencia (2011), The Real Effects of Financial Sector Interventions During Crises,

IMF Working Paper, 11/45.

Lakonishok, J., A. Shleifer and R.W. Vishny (1992), The Impact of Institutional Trading on Stock

Prices, Journal of Financial Economics 32, 23-43.

Large, A. (2005), A Framework for Financial Stability, Financial Stability Review June 2005, Bank of

England.

Ledruth, E. (2007), Simulating Retaliation in Payment Systems: Can Banks Control Their Exposure to

a Failing Participant?, DNB Working Paper 133.

Lehmann, H. and M. Manz (2006), The Exposure of Swiss Banks to Macroeconomic Shocks - An

Empirical Investigation, Swiss National Bank Working Paper 4.

Leinonen, H. (ed.) (2005), Liquidity, Risks and Speed in Payment and Settlement Systems – A

Simulation Approach, Bank of Finland Studies, 31, 9–15.

Liedorp, F.R., L. Medema, M. Koetter, R.H. Koning and I. van Lelyveld (2010). Peer Monitoring or

Contagion? Interbank Market Exposure and Bank Risk, DNB Working Paper 248.

Love, I. and L. Zicchino (2006), Financial Development and Dynamic Investment Behaviour:

Evidence from Panel VAR, Quarterly Review of Economics and Finance 46, 190–210.

Matz, L.M. and P. Neu (2007), Liquidity Risk: Measurement and Management, Wiley Finance.

Miller, M.H. (1995), Do the M&M Propositions Apply to Banks?, Journal of Banking and Finance 19,

483-489.

Ministerie van Financiën (2009), Staat verleent back-up faciliteit aan ING, persbericht.

Modigliani, F. and M.H. Miller (1958), The Cost of Capital, Corporation Finance and the Theory of

Investment, American Economic Review 48 (3), 261–297.

Myers, S. and N.S. Majluf (1984), Corporate Financing and Investment Decisions When Firms Have

Information Investors Do Not Have, Journal of Financial Economics 13, 187-221.

Nakagawa, H. and R. Uchida (2007), Herd Behaviour by Japanese Banks After Financial Deregulation

in the 1980s, Center on Japanese Economy and Business Working Paper 257.

Nier, E., J. Yang, T. Yorulmazer and A. Alentorn (2008), Network Models and Financial Stability,

Bank of England, Working Paper 346.

Nikolaou, K. (2009), Liquidity (risk) Concepts, Definitions and Interactions, ECB Working Paper

1008.

Nyborg, K.G. and P. Östberg (2010), Money and Liquidity in Financial Markets, Centre for Economic

Policy Research, Discussion Paper 7905.

OECD, Organisation for Economic Co-operation and Development (2009), The Financial Crisis:

Reform and Exit Strategies.

205

Panetta, F., T. Faeh, G. Grande, C. Ho, M. King, A. Levy, M. Signorretti, M. Taboga and A. Zaghini

(2009), An Assessment of Financial Sector Rescue Programmes, BIS Papers 48.

Peek, J. and E.S. Rosengren (1995), The Capital Crunch: Neither a Borrower Nor a Lender Be, Journal

of Money, Credit, and Banking 27 (3), 625-638.

Perotti, E. and J. Suarez (2011), A Pigovian Approach to Liquidity Regulation, DNB Working Paper

291.

Peura, S. and E. Jokivuolle (2003), Simulation-based Stress Testing of Banks’ Regulatory Capital

Adequacy, Bank of Finland Discussion Paper Series 4.

Poon, S.H., Rockinger, M. and J. Tawn (2004), Extreme-Value Dependence Measures in Financial

Markets: Diagnostics, Models and Financial implications, The Review of Financial Studies 17 (2),

581–610.

Reinhart, C. and K. Rogoff (2009), The Aftermath of Financial Crises, American Economic Review

99, 466-472.

Reinhart, C. and K. Rogoff (2009), This Time Is Different: Eight Centuries of Financial Folly,

Princeton University Press.

Ross, S. (2005), Neoclassical Finance, Princeton University Press.

Sack, B.P. (2009), The Fed’s Expanded Balance Sheet, Remarks at the Money Marketeers of New

York University, New York City, December.

Sarkar, A. (2009), Liquidity Risk, Credit Risk, and the Federal Reserve’s Responses to the Crisis,

Federal Reserve Bank of New York Staff Reports 389.

Scott, S. and J. Vlček (2011), Macroeconomic Costs of Higher Bank Capital and Liquidity

Requirements, IMF Working Paper 11/103.

Segoviano, M. (2006), Conditional Probability of Default Methodology, Financial Markets Group

Discussion Paper 558. London School of Economics.

Senior Supervisors Group (2008), Observations and Risk Management Practises during the Recent

Market Turbulence.

Shehzad, C. T., J. De Haan and B. Scholtens (2010), The Impact of Bank Ownership Concentration on

Impaired Loans and Capital Adequacy, Journal of Banking and Finance 34, 399–408.

Sorge, M. (2004), Stress-testing Financial Systems: An Overview of Current Methodologies, BIS

Working Papers 165.

Sorge, M. and K. Virolainen (2006), A Comparative Analysis of Macro Stress-testing Methodologies

with Application to Finland, Journal of Financial Stability 2, 113-151.

Spindt, P.A. and V. Tarfan (1980), Liquidity Adjustment Behaviour of Large Money Center Banks,

Journal of Money, Credit and Banking 12 (2), 198-208.

Standard and Poor’s (2007), Liquidity Risk Analysis: Canadian Banks, Ratings Direct.

206

Stark, J. (2009), Monetary Policy Before, During and After the Financial Crisis, Speech, University

Tübingen, Tübingen, 9 November 2009.

Stolz, S.M. (2002), Bank Capital and Risk-Taking. The Impact of Capital Regulation, Charter Value

and the Business Cycle, Kiel Institute for World Economics, Springer.

Stolz, S.M. and M. Wedow (2010), Extraordinary Measures in Extraordinary Times – Public Measures

in Support of the Financial Sector in the EU and the United States, ECB Occasional Paper 117.

Tabbae, M. and J.W. Van den End (2011), When Liquidity Risk Becomes a Macroprudential Issue:

Empirical Evidence of Bank Behaviour, Journal of Financial Stability, forthcoming.

Tarullo, D.K. (2010), Lessons from the Crisis Stress Tests. Speech before the Federal Reserve

International Research Forum on Monetary Policy, Washington, D.C.

Trichet, J.C. (2011), Intellectual Challenges to Financial Stability Analysis in the Era of

Macroprudential Oversight, Speech for the panel discussion on Global imbalances and financial

stability at the Eurofi G20 High Level Seminar, Paris, 18 February 2011.

Trück, S., S. Harpaintner and S.T. Rachev (2005), A Note on Forecasting Aggregate Rates with

Macroeconomic Variables, Technical Report, University of California.

UKFI, UK Financial Investments Limited (2009), Framework Document.

Upper, C. (2006), Contagion Due to Interbank Exposures: What Do We Know, Why Do We Know It

and What Should We Know?, mimeo, Bank for International Settlements.

Van den End, J.W. (2010a), Liquidity Stress-Tester: A Model for Stress-testing Banks’ Liquidity Risk,

CESifo Economic Studies 56 (1), 38-69.

Van den End, J.W. (2010b), Liquidity Stress-Tester: Do Basel III and Unconventional Monetary

Policy Work?, DNB Working Paper 269.

Van den End, J.W., S. Verkaart and A. Van Dijkhuizen (2009), Distortionary Effects of Anti-crisis

Measures and How to Limit Them, DNB Occasional Studies 7 (3).

Van den Heuvel, S. (2002), Does Bank Capital Matter for Monetary Transmission?, Fed New York

Economic Policy Review 8 (1), 259-265.

Van Els, P.J.A. (2005), MORKMON: A Macroeconomic Model of the Netherlands’ Economy

(Chapter 14), in Gabriël Fagan and Julian Morgan (eds), 2005, Econometric Models of the Euro-

area Central Banks, Edward Elgar, Cheltenham, 244-258.

Van Lelyveld, I. (2009), Stress Testing - The Link between Macro and Micro, International Journal of

Central Banking 5 (3), 1-7.

Véron, N. (2010), Some Thoughts on EU Financial Reforms, Paper Prepared for the Brookings-HBF

Seminar: In the Wake of the Crisis: Macroeconomic Dilemmas and Financial Regulation

Challenges for Europe, America and the World.

Virolainen, K. (2004), Macro Stress Testing with Macroeconomic Credit Risk Model for Finland,

Bank of Finland Discussion Paper Series 18.

207

Von Peter, G. (2004), Asset Prices and Banking Distress: A Macroeconomic Approach, BIS Working

Papers 167.

Wilson, T. (1997), Portfolio Credit Risk, Risk 10, 9-10.

208

209

Samenvatting (Summary in Dutch)

Inleiding

Door de financiële crisis van 2007-2009 is een rijkdom aan data beschikbaar gekomen. Dit helpt de

dynamiek van krediet- en liquiditeitrisico in stressomstandigheden beter te begrijpen. Het werpt licht

op de rol van bankgedrag bij tweede ronde-effecten in het financiële systeem en de economie. Door

gebruik van deze data slaat dit proefschrift een brug tussen de theoretische literatuur over gedrag op

financiële markten en beleidsgeoriënteerde analyses van krediet- en liquiditeitrisico’s, zoals

gepubliceerd in financiële stabiliteitsrapporten.

Overeenkomstig de macroprudentiële benadering, die de stabiliteit van het financiële system

als geheel beziet, zijn de empirische analyses in dit proefschrift verricht vanuit een macroperspectief.

Het eerste principe in de analyses is de focus op extreme gebeurtenissen en ontwikkelingen in krediet-

en liquiditeitrisico die een systeemdimensie in de bankensector kunnen hebben. Om te doorgronden

hoe zulke risico’s ontstaan is een goed begrip nodig van bankgedrag op microniveau in relatie tot

macrofinanciële ontwikkelingen. Daarbij moeten detaildata worden gebruikt, om verschillen in

portefeuilles van de banken en de uiteenlopende reacties van de instellingen te kunnen vangen.

Gebruik van detaildata wordt in het proefschrift als tweede principe gehanteerd. Verder heeft de crisis

onderstreept dat systeemrisico kan ontstaan door de wisselwerking tussen krediet- en liquiditeitrisico

en, in het bijzonder, op het snijvlak van funding- en marktliquiditeit. Om daarmee rekening te houden

wordt een meerdimensionale en eclectische benadering gebruikt als derde principe in de analyses. Het

principe wordt geoperationaliseerd door een set van modellen, waarin de verschillende risicofactoren

die krediet- en liquiditeitrisico beïnvloeden dynamisch met elkaar in verband staan. Een ander

kenmerk van krediet- en liquiditeitrisico bij staartgebeurtenissen is inherente onzekerheid over de

eerste ronde-effecten van negatieve schokken en over de tweede ronde-effecten die kunnen optreden

in het financiële systeem of de economie. Om die reden wordt als vierde principe in dit proefschrift

erkend dat de modelresultaten inherent onzeker zijn. Hiermee wordt rekening gehouden door de

toepassing van scenarioanalyses en de presentatie van de modeluitkomsten in termen van

verliesverdelingen. Onzekerheid was ook een belangrijk issue voor overheden en centrale banken bij

de vormgeving van crisismaatregelen tussen 2007 en 2009. Deze onzekerheid wordt betrokken bij de

beoordeling van de crisismaatregelen. De vier principes (het macroperspectief, gebruik van detaildata,

een meerdimensionale benadering, erkenning van onzekerheid) zijn hoofdbestanddelen in dit

proefschrift en geven handvaten om de drie onderzoeksvragen te analyseren:

210

1. Hoe hebben banken hun krediet- en liquiditeitrisicomanagement aangepast tijdens de crisis en hoe

verhouden empirische schattingen van de reacties van banken zich tot de veronderstellingen over

gedrag uit de theoretische literatuur?

2. Hoe kan de impact van staartgebeurtenissen op banken, waarbij krediet- en liquiditeitrisico zijn

betrokken, worden gemodelleerd?

3. Hoe moeten de beleidsreacties op de uitbarsting van krediet- en liquiditeitrisico tijdens de crisis

van 2007-2009 worden beoordeeld, zowel met betrekking tot mogelijke verstorende effecten als

met betrekking tot gedragsprikkels en de invloed van de maatregelen op de economie?

Bankgedrag

Voor de eerste onderzoeksvraag worden indicatoren en tijdreeksmodellen geconstrueerd om het

gedrag van banken empirisch te analyseren. Deze zijn gebaseerd op instellingsspecifieke data van

Nederlandse banken, afkomstig uit een unieke databron met activa en passiva van De Nederlandsche

Bank (DNB). In hoofdstuk 2 worden deze micro-observaties gebruikt om indicatoren te construeren

die trends in het gedrag van banken beschrijven. De maatstaven omvatten zowel de tijdsdimensie

(‘procycliciteit’) als de cross-sectionele dimensie (‘afhankelijkheden’) van systeemrisico. Hoewel ze

beschrijvend van aard zijn, identificeren de maatstaven trends in bankgedrag die vooruitblikkende

informatie verschaffen over marktbrede ontwikkelingen. Hoofdstuk 3 breidt de empirische analyse uit

door het bankgedrag te modelleren in een panel Vector Autoregressie (VAR) raamwerk, gebaseerd op

dezelfde dataset van Nederlandse banken. Beide hoofdstukken concentreren zich op enkele belangrijke

veronderstellingen in de literatuur over gedrag van banken in crises, zoals met betrekking tot

hefboomwerking, kuddegedrag, de pikorde bij balansaanpassingen en de relatie tussen

fundingliquiditeit en kredietverlening en tussen het opsparen van liquiditeit en gedwongen verkopen

van activa.

De volgende conclusies kunnen worden getrokken. Ten eerste, het resultaat met betrekking tot

hefboomwerking laat zien dat balansaanpassingen van banken asymmetrisch zijn: de balansverkorting

tijdens de neergang (tussen medio 2007 en begin 2009) was een intensiever proces dan de

balansuitbreiding in de voorafgaande opgang. Verder begon de balansverkorting van grote banken

tijdens de crisis eerder en was dit proces intensiever en verder voortgeschreden dan bij kleinere

banken. Ten tweede, het resultaat met betrekking tot kuddegedrag toont dat het aantal en de

gelijksoortigheid van de bankreacties in de crisis substantieel toenamen. Dit bevestigt de

veronderstellingen over kuddegedrag in de literatuur en weerspiegelt het procyclische karakter van

bankgedrag. Ten derde, het resultaat over de pikorde bevestigt dat banken gebruikelijk een pikorde

toepassen bij balansaanpassingen (door hun liquide balansposten meer aan te passen dan hun minder

liquide posten), maar dat ze in crises een meer statische reactie vertonen. Ten vierde laten de

211

uitkomsten van het panel VAR model voor de relatie tussen schokken in fundingliquiditeit en

kredietverlening zien dat banken reageren door hun wholesale leningen aan te passen in reactie op

schokken in de geldmarktspread en repo financiering. Ten vijfde blijkt uit de VAR simulaties dat in

antwoord op schokken in wholesale funding de banken liquiditeit opsparen door het accumuleren van

liquide obligaties en een toenemende afhankelijkheid van de centrale bank. Ten slotte tonen de

modelresultaten aan dat gedwongen verkopen van aandelenportefeuilles eerder worden veroorzaakt

door beperkingen in fundingliquiditeit dan door solvabiliteitsbeperkingen bij de banken.

De empirische resultaten leiden tot een beter begrip van de rol die banken spelen bij de

transmissie van negatieve schokken naar het financiële systeem en de economie. De uitkomsten

kunnen helpen om de microfundering van financiële stabiliteitsmodellen te verbeteren, in het

bijzonder met betrekking tot de veronderstellingen over het gedrag van heterogene instellingen.

Bovendien leveren de resultaten waardevolle inzichten op voor macroprudentiële monitoring

raamwerken.

Macro-stresstestmodellen

De tweede onderzoeksvraag richt zich op het modelleren van de invloed van staartgebeurtenissen op

banken, waarbij zich krediet- en liquiditeitrisico’s voordoen, en de reacties van banken op deze

risico’s. De vraag wordt benaderd in een stresstestraamwerk. Dit verschaft methoden om

staartgebeurtenissen in de macro-omgeving te vertalen in indicatoren, die kunnen worden gebruikt om

de gevolgen voor de bankbalansen, de reacties van banken op stressomstandigheden en hieraan

gerelateerde tweede ronde-effecten in het financiële systeem en de economie in te schatten. Het

raamwerk wordt geoperationaliseerd door een set van modellen, zoals herleide vorm satellietmodellen,

VAR-modellen en gekalibreerde simulatie-instrumenten. Deze eclectische benadering wordt

gemotiveerd door de fundamentele onzekerheid met betrekking tot staartgebeurtenissen en de

afwezigheid van een alles omvattend model waarin alle mogelijke verbanden zijn geïntegreerd.

Het overzicht van macro-stresstestmethoden in hoofdstuk 4 wordt gevolgd door enkele

toepassingen van stresstestmodellen voor krediet- en liquiditeitrisico in hoofdstukken 5, 6 en 7.

Hoofdstuk 5 concentreert zich op scenarioanalyse en macro-stresstesten van kredietrisico. Meer in het

bijzonder wordt een meer-factoren benadering gebruikt om deterministische en stochastische

scenario’s te simuleren, die rekening houden met de gelijktijdige verandering in macrovariabelen en

veranderende correlaties tussen risicofactoren. Beide elementen zijn kenmerkend voor stresssituaties.

Het verband tussen macrovariabelen en micro-risicoparameters van bankportefeuilles wordt gelegd in

herleide vorm satellietmodellen. Deze modellen worden geschat op basis van gedesaggregeerde data

van individuele banken en portefeuille-uitsplitsingen, om de verschillen in de reacties van banken en

in de gevoeligheid van portefeuilles in stressomstandigheden te kunnen analyseren. De

212

verliesverdelingen van kredietrisico in het leningenboek van de banken worden onderzocht door zowel

de kans op wanbetaling als het eventuele verlies bij wanbetaling te schatten. Dit is gebaseerd op niet-

lineaire specificaties, omdat risico’s aanzienlijk kunnen worden onderschat als voorbij wordt gegaan

aan de niet-lineairiteiten in de relatie tussen macrovariabelen en kredietrisico, met name bij grote

schokken. De uitkomsten van de stochastische scenariosimulaties worden gepresenteerd in termen van

verliesverdelingen. Deze bieden inzicht in de extreme verliezen waarmee banken kunnen worden

geconfronteerd in stressomstandigheden.

Hoofdstuk 6 presenteert een raamwerk van een stresstestmodel voor liquiditeitrisico van

banken, wat in hoofdstuk 7 wordt uitgebreid met de nieuwe liquiditeitregels van Bazel III en de

mogelijke wisselwerking met de monetaire beleidsoperaties en kredietverlening door banken. Het

liquiditeit-stresstestmodel is in essentie een gekalibreerd simulatie-instrument waarin meerdere

dimensies van liquiditeitrisico worden gecombineerd (funding- en marktliquiditeitrisico) tot een

kwantitatieve maatstaf voor de liquiditeitpositie van banken. Het model houdt rekening met de eerste

en tweede ronde- (terugkoppelings)effecten van schokken, ingegeven door de reacties van heterogene

banken. Bovendien kunnen simulaties worden uitgevoerd voor de invloed van reputatierisico op

banken die reageren om hun liquiditeitpositie te herstellen. Een Monte Carlo benadering wordt

toegepast om de invloed van stressscenario’s op de liquiditeitbuffers van de banken te simuleren,

inclusief de waarschijnlijkheid van een liquiditeitstekort. De belangrijkste uitkomst van de

modelsimulaties is dat de tweede ronde-effecten in bepaalde scenario’s meer invloed kunnen hebben

dan de eerste ronde-effecten en alle typen banken raken. Dit bevestigt dat schokken op de

liquiditeitpositie van de banken gepaard gaan met systeemrisico, wat samenhangt met gedragsreacties

van banken. Hoewel de toevoeging van het effect op kredietverlening in hoofdstuk 7 een poging is om

liquiditeit- en kredietrisico met elkaar te verbinden, is een meer complete integratie van krediet- en

liquiditeitrisico een veld voor toekomstig onderzoek. Het vergt een diepere analyse van de mogelijke

relaties en wisselwerkingen tussen kredietkwaliteit aan de ene kant en funding- en marktliquiditeit aan

de andere kant. Dergelijk onderzoek zal zich onder andere moeten buigen over het verschillende

karakter van de risico’s (met name de drijvende krachten en de tijdshorizon) en over het gedrag van

tegenpartijen dat een verband tussen de liquiditeit- en kredietrisico kan aanbrengen.

De uitgebreide versie van het model veronderstelt dat de nieuwe liquiditeitregels, zoals

voorgesteld door Bazel III, een knellende factor zijn voor het bankgedrag. Dit wordt operationeel

gemaakt door de regel dat banken hun liquiditeitratio herstellen door extra liquide activa te verwerven

en de stabiliteit van hun financiering verbeteren, als weerspiegeling van het opsparen van liquiditeit en

het zoeken naar stabiele financiering door banken tijdens de crisis. De modelsimulaties kwantificeren

de potentieel bredere effecten van de nieuwe liquiditeitregels, bijvoorbeeld met betrekking tot de

invloed op het kredietaanbod. Uit modelsimulaties van liquiditeitstressscenario’s blijkt deze invloed

beperkt te zijn. Een ander resultaat is dat tweede ronde-effecten en staartrisico’s van een stressscenario

beduidend lager zijn als banken zich zouden aanpassen aan Bazel III door het aanhouden van

213

hoogwaardiger liquide activa. Met name een nauw gedefinieerde liquiditeitbuffer - bestaande uit

hoogwaardig staatspapier - maakt een groot verschil in het beperken van staartrisico’s voor de banken.

Een gevolg van grotere staatsobligatieportefeuilles is dat monetair beleid uitgevoerd via activa-

aankopen meer invloed heeft op bankgedrag dan herfinancieringsoperaties door de centrale bank. We

hebben ook de consequenties op de fundingliquiditeit van de banken gesimuleerd van een uitfasering

van de uitgebreide herfinancieringsoperaties door centrale banken. De uitkomsten indiceren dat de

liquiditeitratio’s van de banken verbeteren ten opzichte van de pre-exit situatie indien alternatieve

stabiele financiering beschikbaar is.

Beleidsreacties op de crisis

De derde onderzoeksvraag, over de reacties van beleidsmakers op de krediet- en liquiditeitrisico’s van

banken tijdens de crisis, is in de eerste plaats benaderd door beoordeling van de korte termijn

crisismaatregelen van centrale banken en overheden tussen 2007 en 2009 (in hoofdstuk 8) en in de

tweede plaats door het analyseren van de effecten van lange termijn maatregelen zoals genomen door

regelgevers, met name de macro-economische effecten van Bazel III (in hoofdstuk 9).

Om de financiële stabiliteit te beschermen en de invloed van financiële stress op de economie

te beperken, hebben centrale banken en overheden snel en onder grote onzekerheid moeten reageren

op de uitbarsting van risico’s in de bankensector. Hoewel de maatregelen gunstig uitwerkten op de

financiële stabiliteit, gingen ze gepaard met verstorende effecten. Deze hebben betrekking op het

gelijke speelveld tussen gesteunde en niet-gesteunde instellingen en op de kapitaalstromen tussen

marktsegmenten, inclusief grensoverschrijdende stromen. De invloed van de overheidsingrepen is

echter moeilijk te onderscheiden van de invloed van marktprikkels. We vinden bewijs voor een

gunstig korte-termijn effect van de overheidssteun in de marktprijzen van financiële instellingen, maar

dit verdwijnt enkele maanden nadat de interventies hebben plaatsgevonden en slaat dan zelfs om in

een negatief effect. Dit sluit aan bij de lange termijn nadelen die gepaard gaan met interventies door

overheden en centrale banken, welke verkeerde prikkels kunnen geven voor het nemen van risico en

gedrag van marktpartijen. De belangrijkste beleidsconclusie van de analyse is dat zulke negatieve

bijeffecten kunnen worden beperkt door het juiste ontwerp van de steunmaatregelen (marktconform)

en van de exitstrategie (tijdige terugtrekking).

Regulering heeft ook een belangrijke invloed op het gedrag van banken, zowel in de

transitiefase als in de nieuwe evenwichtssituatie waarin banken voldoen aan de nieuwe regelgeving.

De nieuwe kapitaal- en liquiditeitstandaarden van Bazel III zullen de intermediatiefunctie van banken

beïnvloeden en daarmee het kredietaanbod en de economische groei. Simulatie-uitkomsten van

herleide vorm satellietmodellen en van DNB’s structurele macro-economische model indiceren echter

dat de negatieve invloed van Bazel III op het reële bruto binnenlands product (bbp) beperkt zal zijn tot

214

enkele tienden van een procent gedurende de transitiefase. Een ander resultaat is dat een voldoende

lange transitieperiode de kosten in de komende jaren zal beperken, ook omdat het de banken meer

ruimte geeft om zich aan te passen. Als de banken zich eenmaal hebben aangepast aan de nieuwe

standaarden, zullen de voordelen van een meer solide financiële systeem de nadelen waarschijnlijk

overtreffen. Hoewel een precieze kwantificering van de voordelen gecompliceerd is vanwege de

onzekere effecten van Bazel III op de strategie van banken en van hun klanten, lijkt het duidelijk dat

hogere buffers van banken een financiële crisis in de toekomst zowel minder waarschijnlijk als minder

diep maken, terwijl in normale omstandigheden de economische groei stabieler zal zijn.