The Future of Payment Systems (Routledge International Studies in Money and Banking)

The Future of Payment Systems Payment systems are the circulation system for modern monetary economies, ensuring money...

Author: Haldane et al

343 downloads 1590 Views 1MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

The Future of Payment Systems

Payment systems are the circulation system for modern monetary economies, ensuring money for spending and saving moves to the right person in the right place at the right time. When these circulation systems break down, monetary economies themselves run the risk of seizing up. This is why robust payment systems are considered so important by central banks and policymakers throughout the world. This volume draws on wide-ranging contributions from prominent international experts, discussing some of the most pressing issues facing policymakers and practitioners in the field of payment systems today. Because payment systems have been with us for at least as long as money itself, many of the questions raised in this book are timeless. Improvements in information technology mean, however, that answers to these questions are unlikely to be timeless. This book tackles issues regarding the form payment systems might take in the future, the risks associated with this evolution, the techniques being deployed to assess these risks and the implications these risks have for the respective roles of the public and private sector. Based on a conference, ‘The Future of Payment Systems’, organised by the Bank of England, this book will make fascinating reading for practitioners and policymakers in the field of payment systems, as well as students and researchers engaged with the economics of payments and central banking policy. Andrew G. Haldane is Head of Systemic Risk Assessment at the Bank of England. Stephen Millard is a Senior Economist at the Bank of England. Victoria Saporta is a Senior Economist at the Bank of England.

Routledge international studies in money and banking

1 Private Banking in Europe Lynn Bicker 2 Bank Deregulation and Monetary Order George Selgin 3 Money in Islam A study in Islamic political economy Masudul Alam Choudhury 4 The Future of European Financial Centres Kirsten Bindemann 5 Payment Systems in Global Perspective Maxwell J. Fry, Isaak Kilato, Sandra Roger, Krzysztof Senderowicz, David Sheppard, Francisco Solis and John Trundle 6 What is Money? John Smithin 7 Finance A characteristics approach Edited by David Blake

8 Organisational Change and Retail Finance An ethnographic perspective Richard Harper, Dave Randall and Mark Rouncefield 9 The History of the Bundesbank Lessons for the European Central Bank Jakob de Haan 10 The Euro A challenge and opportunity for financial markets Published on behalf of Société Universitaire Européenne de Recherches Financières (SUERF) Edited by Michael Artis, Axel Weber and Elizabeth Hennessy 11 Central Banking in Eastern Europe Edited by Nigel Healey and Barry Harrison 12 Money, Credit and Prices Stability Paul Dalziel

13 Monetary Policy, Capital Flows and Exchange Rates Essays in memory of Maxwell Fry Edited by William Allen and David Dickinson 14 Adapting to Financial Globalisation Published on behalf of Société Universitaire Européenne de Recherches Financières (SUERF) Edited by Morten Balling, Eduard H. Hochreiter and Elizabeth Hennessy 15 Monetary Macroeconomics A new approach Alvaro Cencini 16 Monetary Stability in Europe Stefan Collignon 17 Technology and Finance Challenges for financial markets, business strategies and policy makers Published on behalf of Société Universitaire Européenne de Recherches Financières (SUERF) Edited by Morten Balling, Frank Lierman, and Andrew Mullineux 18 Monetary Unions Theory, history, public choice Edited by Forrest H. Capie and Geoffrey E. Wood 19 HRM and Occupational Health and Safety Carol Boyd

20 Central Banking Systems Compared The ECB, the pre-Euro Bundesbank and the Federal Reserve System Emmanuel Apel 21 A History of Monetary Unions John Chown 22 Dollarization Lessons from Europe and the Americas Edited by Louis-Philippe Rochon and Mario Seccareccia 23 Islamic Economics and Finance: A Glossary, 2nd Edition Muhammad Akram Khan 24 Financial Market Risk Measurement and analysis Cornelis A. Los 25 Financial Geography A banker’s view Risto Laulajainen 26 Money Doctors The experience of international financial advising 1850–2000 Edited by Marc Flandreau 27 Exchange Rate Dynamics A new open economy macroeconomics perspective Edited by Jean-Oliver Hairault and Thepthida Sopraseuth 28 Fixing Financial Crises in the 21st Century Edited by Andrew G. Haldane

29 Monetary Policy and Unemployment The U.S., Euro-area and Japan Edited by Willi Semmler

37 The Structure of Financial Regulation Edited by David G. Mayes and Geoffrey E. Wood

30 Exchange Rates, Capital Flows and Policy Edited by Peter Sinclair, Rebecca Driver and Christoph Thoenissen

38 Monetary Policy in Central Europe Miroslav Beblav´y

31 Great Architects of International Finance The Bretton Woods era Anthony M. Endres 32 The Means to Prosperity Fiscal policy reconsidered Edited by Per Gunnar Berglund and Matias Vernengo 33 Competition and Profitability in European Financial Services Strategic, systemic and policy issues Edited by Morten Balling, Frank Lierman and Andy Mullineux 34 Tax Systems and Tax Reforms in South and East Asia Edited by Luigi Bernardi, Angela Fraschini and Parthasarathi Shome 35 Institutional Change in the Payments System and Monetary Policy Edited by Stefan W. Schmitz and Geoffrey E. Wood 36 The Lender of Last Resort Edited by F.H. Capie and G.E. Wood

39 Money and Payments in Theory and Practice Sergio Rossi 40 Open Market Operations and Financial Markets Edited by David G. Mayes and Jan Toporowski 41 Banking in Central and Eastern Europe 1980–2006 A comprehensive analysis of banking sector transformation in the former Soviet Union, Czechoslovakia, East Germany, Yugoslavia, Belarus, Bulgaria, Croatia, the Czech Republic, Hungary, Kazakhstan, Poland, Romania, the Russian Federation, Serbia and Montenegro, Slovakia, Ukraine and Uzbekistan Stephan Barisitz 42 Debt, Risk and Liquidity in Futures Markets Edited by Barry A. Goss 43 The Future of Payment Systems Edited by Andrew G. Haldane, Stephen Millard and Victoria Saporta

The Future of Payment Systems

Edited by Andrew G. Haldane, Stephen Millard and Victoria Saporta

First published 2008 by Routledge 2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN Simultaneously published in the USA and Canada by Routledge 270 Madison Ave, New York, NY 10016 Routledge is an imprint of the Taylor & Francis Group, an informa business This edition published in the Taylor & Francis e-Library, 2008. “To purchase your own copy of this or any of Taylor & Francis or Routledge’s collection of thousands of eBooks please go to www.eBookstore.tandf.co.uk.” © 2008 Selection and editorial matter, The Governor and Company of the Bank of England; individual chapters, the contributors All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library Library of Congress Cataloging in Publication Data A catalog record for this book has been requested ISBN 0-203-94014-8 Master e-book ISBN

ISBN10: 0-415-43860-8 (hbk) ISBN10: 0-203-94014-8 (ebk) ISBN13: 978-0-415-43860-5 (hbk) ISBN13: 978-0-203-94014-3 (ebk)

Contents

List of figures List of tables Notes on contributors Foreword Acknowledgements General introduction: the future of payment systems

x xii xiii xv xvi 1

ANDREW G. HALDANE, STEPHEN MILLARD AND VICTORIA SAPORTA

PART I

Payment systems and public policy 1 Central banks and payment systems: past, present and future

13

15

STEPHEN MILLARD AND VICTORIA SAPORTA

2 The role of a central bank in payment systems

45

EDWARD J. GREEN

3 Some challenges for research in payments

57

EDWARD J. GREEN

4 Payment economics and the role of central banks

68

JEFFREY LACKER

PART II

New approaches to modelling payments 5 New models of old (?) payment questions RICARDO CAVALCANTI AND NEIL WALLACE

73 75

viii

Contents

6 Optimal settlement rules for payment systems

87

BENJAMIN LESTER, STEPHEN MILLARD AND MATTHEW WILLISON

7 The microstructure of money

100

JAMES McANDREWS

PART III

Current payment policy issues 8 Wholesale payments: questioning the market-failure hypothesis

117

119

GEORGE SELGIN

9 Central bank intraday collateral policy and implications for tiering in RTGS payment systems

138

JOHN P. JACKSON AND MARK J. MANNING

10 Central banks’ interest calculating conventions: deviating from the intraday/overnight status quo

160

GEORGE SPEIGHT, MATTHEW WILLISON, MORTEN BECH AND JING YANG

11 How should we regulate banks’ liquidity?

175

JEAN-CHARLES ROCHET

PART IV

Policy perspectives on the future of payments

187

12 The diffusion of real-time gross settlement

189

MORTEN L. BECH

13 E-settlement: soon a reality?

206

HARRY LEINONEN

14 Real-time liquidity management in a globally-connected market

230

RICHARD PATTINSON

15 Will central banking survive electronic money? STEFAN W. SCHMITZ

233

Contents 16 Payment systems and central banks: where are we now and where will e-payments take us?

ix 255

CHARLES FREEDMAN

Index

262

Figures

1.1 6.1 6.2 6.3 6.4 6.5 6.6 7.1 7.2 9.1 9.2 9.3 9.4 9.5 9.6 9.7 10.1 10.2 12.1 12.2 12.3 12.4 12.5 12.6 12.7 12.8 12.9 13.1 13.2

Stylised models of intervention in payment systems DNS equilibrium RTGS equilibrium ( = 0) RTGS equilibrium ( = 0.005) RTGS equilibrium ( = 0.05) Existence of DNS and RTGS equilibria Welfare Intraday pattern of activity: total value and volume of federal funds traded Fifth and 95th percentiles of the average federal funds rate minus the target rate Payment flows when C accesses the system via bank A A time-line for actions High degree of internalisation Low default probability The impact of imperfect monitoring The impact of tiering risk The impact of tiering risk with high payment value Change in shape of yield curve An alternative change in the shape of the yield curve Value of transfers originated on Fedwire Adoption of RTGS in Europe – 1995 Adoption of RTGS in Europe – 2005 Adoption of RTGS in Asia – 2005 Adoption of RTGS in Africa – 2005 Adoption of RTGS in South America Adoption of RTGS in Central America – 2005 S-curve and adopter groups Adoption of RTGS in central banking Direct interbank communication in a network-based infrastructure The common account number space

28 94 94 95 95 97 98 106 108 143 144 148 149 152 153 155 162 171 191 192 193 194 196 197 197 199 199 208 209

Figures xi 13.3 E-settlement is part of the credit transfer circle, which provides efficient electronic communications between participants in a payment 13.4 The digital e-settlement stamp is part of the payment message 13.5 The digital encrypted stamp with central bank cover will follow the payment message through the network 13.6 E-settlement stamps are produced by e-settlement modules, which are closely integrated with banks’ payment systems 13.7 A dedicated interbank network connects all banks and the central bank with each other for payment processing 13.8 The increasing market failure gap 14.1 Links between world market infrastructure

210 213 213 214 215 222 230

Tables

1.1 1.A1 1.A2 1.A3 7.1 10.1 10.2 10.3 10.4 12.1 12.2

Ranking models of intervention 31 G10 models for intervention in large-value payment systems 37 G10 models for intervention in ACHs 38 G10 models for intervention in the embedded payment systems of securities settlement systems 39 Regression results 109–10 Pay-offs when central bank charges and remunerates at end of day only 167 Equilibrium strategies when the central bank charges and remunerates at end of day only 167 Pay-offs when central bank charges and remunerates at midday and at end of day 169 Equilibrium strategies when the central bank charges and remunerates at midday and at end of day 169 Imported RTGS systems 201 Liquidity savings features 203

Contributors

Morten Bech. Economist, Federal Reserve Bank of New York. Prior to that, Dr Bech worked for the Danish Central Bank where he helped design the KRONOS RTGS system. Dr Bech has written extensively on payment system issues in central bank publications and academic journals. Dr Bech was a visitor at the BoE in the summer of 2006. Ricardo de Cavalcanti. Associate Professor of Economics, Graduate School of Economics, Getulio Vargas Foundation, Rio de Janeiro, Brazil. Charles Freedman. Currently Scholar in Residence in the Department of Economics at Carleton University and consultant to the IMF and central banks. He worked at the Bank of Canada from 1974 to 2003, serving as Deputy Governor from 1988 to 2003. Edward J. Green. Professor of Economics, The Pennsylvania State University. John Jackson. Economist, Systemic Risk Reduction Division, Bank of England. Jeffrey Lacker. President of the Federal Reserve Bank of Richmond. Harry Leinonen. Adviser to the Board of the Bank of Finland and the Finnish representative in the Eurosystem Payment and Settlement System Committee; Finnish representative in the EU Commission; Government Expert Group and Market Group on Payment Issues. Benjamin Lester. University of Pennsylvania. Benjamin is currently finishing his PhD in Economics at the University of Pennsylvania. His work focuses on the macroeconomic implications of issues in money and banking. Mark Manning. Senior manager, Systemic Risk Reduction Division, Bank of England. James J. McAndrews. Vice President and Head of the Money and Payment Studies Function in the Research and Statistics Group of the Federal Reserve Bank of New York. He received a PhD in economics from the University of Iowa.

xiv

Contributors

Richard Pattinson. Head of Regulatory and Industry Issues, Global Payments, Barclays Bank. Holds a number of external positions in the industry primarily concerned with payments and settlements and the management of payment system liquidity including: Chairman, CHAPS Clearing Company Limited (UK); Deputy Chairman, SWIFT (UK) Limited (UK); Director, Voca Limited (UK); Director, EBA Association (France); Director, CLS Group Holding AG (Switzerland); Director, CLS Bank International (USA); Member Bank of England Money Market Liaison Group; Member UK Market Advisory Committee. Jean-Charles Rochet. Professor of Economics and Mathematics at Toulouse School of Economics (Toulouse University) and Research Director at Institut D’Economie Industrielle, Toulouse, France. George Selgin. Professor of Economics at the University of Georgia’s Terry College of Business. His latest book, Good Money: Private Enterprise and the Beginnings of Modern Coinage, is forthcoming from the University of Michigan Press. Stefan W. Smitz. Currently at Oesterreichische Nationalbank. Co-editor of Institutional Change in the Payments System and Monetary Policy, Routledge (with Geoffrey E. Wood) and Carl Menger and the Evolution of Payment Systems: From Barter to Electronic Money, Edward Elgar (with M. Latzer). George Speight. Senior Manager, Systemic Risk Reduction Division, Bank of England. Neil Wallace. Professor of Economics, The Pennsylvania State University. Matthew Willison. Economist, Systemic Risk Assessment Division, Bank of England. Jing Yang. Senior Economist, International Finance Division, Bank of England.

Foreword

The Bank of England has two core purposes – monetary stability and financial stability. Payment systems are the mechanism by which money is transferred to enact both real transactions (such as buying bread) and financial transactions (such as buying bonds). So robust payment systems are integral to both of the Bank’s core purposes. The same is true in central banks around the world. Indeed, in many central banks their payment system role predated the formalisation of their financial and monetary stability objectives. The role of central banks in payment systems has, however, changed significantly over the past decade. Doubtless it is set to change further over the next decade, not least due to the impact of advances in information technology. But what form will this change take? And what are the risks – for policymakers, for payment system operators, for the public at large – associated with this change? These are among the most topical and involved questions facing central banks today. With these questions in mind, the Bank of England hosted a two-day conference on 19 and 20 May 2005 with the title ‘The Future of Payment Systems’. The conference aimed to draw together the views of academics, payment system practitioners and policymakers on payment system issues. All too rarely have attempts been made to integrate the distinct perspectives of these three parties. This volume brings together in one place these contributions, as a first step towards such a synthesis. You will not be surprised to hear that the volume is long on questions and short on answers. That is in the nature of conferences, perhaps especially successful ones. But I hope, nonetheless, you find it useful as a contribution towards understanding the likely future course of payment systems, which has important implications for us all. Sir John Gieve Deputy Governor for Financial Stability Bank of England

Acknowledgements

This volume brings together the papers from a conference on ‘The Future of Payment Systems’ that was held at the Bank of England on 19–20 May 2005. We would first like to thank all the contributors to this volume for their help in preparing papers on topics of our choosing rather than theirs, presenting them at our conference and then revising them for this volume. We would like also to thank everyone who attended and contributed to the conference, as well as all those who made it happen. In particular, we would like to thank the discussants at this conference whose comments led to substantial improvements in all of the papers: Martin Andersson, Morten Bech, Xavier Freixas, Charles Goodhart, Charles Kahn, Nobuhiro Kiyotaki, Thorsten Koeppl, John Mohr, John Moore, Erlend Nier, Will Roberds, Matthew Willison and Randall Wright. And, in particular, we would also like to thank Francesca Desquesnes whose efforts ensured the event ran without a hitch. Many people have commented on the various chapters in this book and we thank them all. Roy Clive, Raxita Dodia, Elizabeth Hughes, Sandra Mills and Julie Pickering have done sterling work in helping pull the manuscript together; and the help of Thomas Sutton and Terry Clague at Routledge has been invaluable at various stages of the project. Of course, all remaining errors and omissions are ours. The views expressed in the chapters in this book are those of the authors and do not necessarily reflect those of the Bank of England, the Federal Reserve System, the Federal Reserve Bank of New York, the Federal Reserve Bank of Richmond, the Oesterreichische Nationalbank, Barclays Bank or the Bank of Finland. Finally, the authors and publishers would like to thank the following for granting permission to reproduce material in this work: the Bank of Finland, Barclays Bank, the Federal Reserve Banks of New York and Richmond, Charles Freedman, Ed Green, Jean-Charles Rochet, Stefan Schmitz and Neil Wallace and Ricardo Cavalcanti. The chapter by George Selgin was reprinted from the International Review of Law and Economics, Vol. 24, No. 3, Selgin, G, ‘Wholesale payments: questioning the market-failure hypothesis’, pages 333–350, Copyright (2004) with permission from Elsevier to whom we give our thanks. Every effort has been made to contact copyright holders for their permission to reprint material in this book. The publishers would be grateful to hear from

Acknowledgements

xvii

any copyright holder who is not here acknowledged and will undertake to rectify any errors or omissions in future editions of the book. Andrew G. Haldane Stephen Millard Victoria Saporta Bank of England

General introduction The future of payment systems Andrew G. Haldane, Stephen Millard and Victoria Saporta

A conundrum At the heart of the study of payment systems lies a contradiction. Mere mention of the words ‘payment systems’ to an economist tends to conjure up images of an obscure and rather technical sub-discipline – or perhaps even backwater – of the profession. This backwater is believed to be inhabited by a small and rather reclusive set of fanatics. This tribe uses tools and a language of their own and spends its time studying issues that are well outside the mainstream. Yet, for the public at large, ‘payment systems’ are part and parcel of their everyday lives. The use of cash, credit and debit cards and electronic money transfers to enact payments and transfers is a practical and straightforward task. Payment systems are unwittingly used by almost everyone, probably several times a day, every day of the week. The tools and the language used to describe these instruments, while distinct, are well understood by almost everyone. So payment systems are obscure yet commonplace, highly technical yet understood by everyone. How do we resolve this conundrum? It was this question which prompted the Bank of England to host an international conference on ‘The Future of Payment Systems’ on 19 and 20 May 2005 in London. This volume collates together the main contributions from that conference. Unlike the conference itself, the volume is organised into four blocks. Part I considers the intersection between payment systems and public policy. The chapters trace the anthropological origins of payment systems: How and why they came into being and how their evolution has been, and is being, shaped by public policy? Historically, central banks and payment systems have been inextricably linked. But technology is reshaping those historical relationships in important ways. Part II of the volume considers some of the methodological advances which have recently been made in the study of payment systems: What models and empirical methods have been used to analyse payment system behaviour? This is a rapidly evolving – though at present rather diffuse – area of the economics profession. Part III of the volume illustrates how some of these approaches can be used to address a number of topical payment system issues, for example, the role of central bank intraday liquidity policy and regulatory liquidity requirements in reducing payment system risks.

2

A.G. Haldane et al.

Finally, having traced the origins of the species, in Part IV of the volume we plot the possible course of payment systems in the future – hence the title of this volume, The Future of Payment Systems. Parts I, II and III of the volume are natural precursors to Part IV because an understanding of the fundamentals of payment instruments is essential when predicting how technology might shape those future fundamentals. The chapters in Part IV sketch some alternative – and in some cases quite radical – visions of payment systems of the future. This volume does not profess to contain all of the answers, nor to solve completely the conundrum. But it may provide, we hope, some clues on what the key issues and questions might be, now and in the future. To that end, the remainder of this introduction considers a few key generic payment system themes which emerge from the volume; it then places the chapters from the volume in the context of these themes and the wider literature on payments.

The economics of payments and payment systems To develop and understand the economics of payments, we first need some definitions and methodology. Perhaps a good – if not entirely uncontroversial – place to start is with a working definition of ‘payments’ and ‘payment systems’. A ‘payment’ is a transfer of monetary value. So a ‘payment system’ is no more than an organized arrangement for transferring value between its participants. So defined, it is clear that payment systems are fundamental to the functioning of all monetary economies. If money is the lifeblood of modern monetary economies, payment systems are the circulation system. Failures in this circulation system risk a seizing up in the real and financial transactions they support, with potentially significant welfare costs. Because value need not be embodied in monetary assets but also in real goods, it could reasonably be argued that payment systems predate the existence of money itself. Certainly, payment systems predate the emergence of central banks, the latter which in many cases emerged during the twentieth century. Some barter economies had quite sophisticated and hence organized exchange of value systems in place, which legitimately could be termed a primitive payment system. But as first commodity money replaced goods, fiat money displaced commodity money, and finally commercial bank (or ‘inside’) money replaced central bank (or ‘outside’) money as the media of exchange, payment systems have increasingly involved monetary transfers routed through financial institutions. And this monetary evolution, in turn, resulted in central and commercial banks increasingly taking centre stage in the design and operation of payment systems (see Part I). These definitions, and this evolution, of payment systems make clear why the study of payments is many-faceted. In particular, the economics of payments can be thought to cover at least the following sub-disciplines and questions: •

Foundations of money and payment systems: Why have payments or money in the first place? What fundamental frictions in the economy do money and

General introduction

•

•

3

payment systems help mitigate? And what can this welfare-theoretic approach to money and payments tell us about the future evolution of payment systems? Payments and the macroeconomy: Where do payments fit within the wider macroeconomy? In particular, what are the macroeconomic benefits of wellfunctioning payment systems, measured in terms of output, employment and inflation? The industrial organization of payments: What is a suitable industrial structure for payments from a societal perspective? In particular, what and how large are the market failures embedded in the payments industry? And how best should these be tackled? How do we design payment systems to provide incentives for appropriate behaviour by system participants from a social welfare perspective?

These questions are distinct and so too have been the analytical frameworks used to tackle them. The chapters in the volume seek to address some of these different questions using often quite different analytical approaches (see, for example, Part II). To date, there has been no grand synthesis of these different approaches. Green’s chapters in this volume argue compellingly against us expecting such a grand synthesis any time soon, in part because the questions being posed of payment systems are so deep and broad, ranging from the microeconomics of money, through the macroeconomics of payments to the industrial organization of banking and finance. The chapters in the volume hopefully give a flavour for such depth and breadth. And, as such, they may hopefully serve as a staging post towards a synthesis of these various strands. The foundations of money and payment systems Kahn and Roberds (2006) suggest that the foundations of payment systems rest on their ability to deal with two fundamental problems inherent in an economy: a mismatch between the times that different agents wish to trade and limited enforcement of privately-issued debt obligations (‘inside money’, such as deposits with commercial banks). They point to the existence of two types of payment system: ‘store-of-value’ systems based on the ability of agents to verify the asset being used to effect payment, and ‘account-based’ systems based on the ability of agents to verify the identities of the account holders. Most payment systems that we observe in the real world feature elements of store-of-value and account-based systems and nearly all rely on the transfer of inside money. The chapter by Cavalcanti and Wallace in this volume uses a model based on these imperfections to investigate the usefulness of inside money as a medium of exchange. In particular, they assume that individuals cannot commit to future actions and to some extent their histories are not known (there is ‘imperfect monitoring’). Specifically, some people are not monitored at all and others are perfectly monitored. The perfectly-monitored agents are able to issue inside money. Cavalcanti and Wallace use this model to show that private money can

4

A.G. Haldane et al.

be useful and that, in the absence of private money, central bank money and central bank lending through the discount window can deliver welfare benefits over and above what can be accomplished with a private interbank money market. They argue that the deep insights gained from this framework can be used to help shape answers to questions about the future design of payment systems, though their existing framework is perhaps as yet too primitive to deliver sharp and robust policy insights. He et al. (2005) develop a search-theoretic model of a payment system in which the introduction of cash benefits the economy for the same reasons as in the chapter by Wallace and Cavalcanti; that is, it solves the problems of lack of commitment and the inability to monitor your trading partners. But they note that the use of cash itself carries costs; it needs to be verified and can be stolen. In their model, they allow agents to guard against the risk of theft by depositing their cash in banks and making payments from their bank account to other agents’ bank accounts by a payment system they call ‘cheques’. They find that the introduction of a payment system expands the range of parameter values consistent with there being an equilibrium in which money is accepted as a medium of exchange – in other words, that the presence of a payment system enables money to solve the imperfections discussed by Kahn and Roberds (2006). But the payment system in He et al. (2005) is risk free. Two more imperfections that need to be considered in a model of payment systems are the risk that the members of the system default on their obligations – credit risk – and the risk that the system itself breaks down – operational risk. Lester (2005) extends the He et al. model to allow for credit risk while Millard and Willison (2006) extend the model to allow for operational risk in payment systems. In both cases, the authors show how one might begin to quantify the welfare benefits of reducing the risks brought about by these frictions. Payments and the macroeconomy The approaches discussed above begin to introduce inside money – and payment systems that enable the transfer of inside money – into a model of the economy in a way that is intellectually rigorous, showing that it delivers welfare gains by mitigating some of the fundamental frictions that may exist in the real world, such as the lack of a double coincidence of wants. This work is essentially grounded in the microeconomics of money. Relatively little work has, however, been done to illustrate the potential links between the design and operation of payment systems and the macroeconomy – for example, how payment systems can improve welfare through their effects on standard macroeconomic variables such as output and inflation. Cifuentes and Willison (2006) suggest that payment systems can benefit the economy by reducing the liquidity needed to make payments, thereby allowing banks to reallocate their resources to less liquid assets which have a higher expected return. Millard et al. (2006) use a standard macroeconomic modelling


5

approach to argue that, if banks require liquid assets to make payments through a payment system, then central banks can improve welfare by creating such liquid assets at the beginning of the day and withdrawing them at the end of the day. If this activity is near-costless – as would be expected in a pure fiat money world – it follows that the operation of the payment system will only affect the intraday money stock and not macroeconomic magnitudes, such as output and inflation. The chapters by Green in this volume make essentially the same point, arguing that central banks are uniquely placed to offer free, short-term credit against illiquid collateral to other financial intermediaries and that doing so improves welfare in the economy, without adversely affecting monetary control and hence inflation and output. Such a liquidity policy by the central bank may, of course, have implications for the efficiency with which banks manage their own liquidity. The chapter by Bech et al. in this volume argues that, if the central bank were to charge interest based on settlement banks’ balances with them more frequently than once a day, the settlement banks would be given an incentive to monitor more closely their customers’ use of intraday credit; this, in turn, would be positive for financial stability since intraday credit risk would be reduced, or at least internalized to some degree in banks’ decision-making.1 But if banks’ customers did not insist their payments be made by a particular time, the banks would tend to delay payments so they could earn interest on positive balances for longer; this would be negative for financial stability since an operational incident would then result in more payments not having been made. Either way, there would be no implications for the ability of the central bank to carry out monetary policy. The industrial organization of payments Perhaps the largest body of literature on the economics of payments has focussed on its industrial organization. Within this, analysis has typically centred on the potential externalities involved in the payments industry and costeffective ways of mitigating the market failures associated with them. Payment systems are an example of a network industry in which the welfare of existing members increases each time a new member joins. In addition, there are likely to be economies of scale in the provision of payment services. Both these features are likely to imply a tendency towards concentration and a lack of competition. These potential externalities are discussed extensively and rigorously in the chapters by Millard and Saporta, Green and Lacker. But the scale of these problems, and hence the appropriate solution to them, remain vexed questions. The chapters by Green and Lacker argue that the importance of many of these externalities may have been overstated historically. Hence they favour minimalist interventions by the public sector to offset these frictions. Allen et al. (2006) and Schanz (2006) discuss one such non-interventionist strategy. If a payment system is mutually-owned by its users, the monopoly problem can, to an extent, be mitigated.

6

A.G. Haldane et al.

A second set of externalities arise from the way in which payment systems can generate risks for the wider financial system – so-called systemic risks. These can arise either because payment systems can act as a channel through which financial contagion is propagated between institutions or because such systems act as a potential single point of failure. The mitigation of these systemic risks is the focus of a number of the chapters in the volume. One of the key tools typically used to mitigate such systemic risks is the design and operation of payment systems including, crucially, the rules of the game for settlement among participants in these systems. Payment systems have historically settled using what is called deferred net settlement (DNS). Payments are collected together over the course of a day, the net amount that each bank owes or is owed is calculated, and then net amounts are settled between participants. As the volume and value of payments increased dramatically throughout the 1980s and 1990s, central banks became worried about the risks inherent in such netting systems. In particular, if one participant failed during the day, all processed payment orders could be unwound with the consequent risk of other participants failing to be able to meet their obligations, thereby generating contagious default. In response, over the past decade there has been a large increase in the number of large-value payment systems in the world that employ real-time gross settlement (RTGS), where payments are paid in full as soon as they are submitted to the system. The diffusion of RTGS systems across the world’s economies is discussed by Bech in his chapter. But RTGS requires banks to hold a much larger amount of the settlement asset to make these gross payments. As Bech and Garratt (2003) show using a simple game-theoretic model, this creates incentives for banks to delay payments in the hope of obtaining liquidity from incoming payments which they can then use to make their outgoing payments. The net result is a ‘bad’ equilibrium where all banks may delay payments and no banks save liquidity by sodoing. McAndrews and Rajan (2000) present some empirical evidence that suggests this may happen in practice. In his chapter for this volume, McAndrews calls this approach to understanding the effects of payment system design on payment outcomes the ‘market microstructure of money’. More generally, he discusses how the methodological insights from the market microstructure literature on securities markets can be used and adapted to address a range of topical payment system issues.

Key issues in payments today Armed with this toolkit, how best might we apply it to key policy issues in payments (Part III)? As Millard and Saporta discuss, historically, central banks’ relationship with payment systems has extended beyond providing their liabilities as the ultimate settlement asset, to include owning, operating and regulating certain payment systems. But which of these roles, if any, should central banks play?


7

Norman et al. (2006) show how the need for an ultimate settlement asset for payments is inextricably linked to the evolution of central banking. Indeed, Millard (2006) argues that providing the ultimate settlement asset is what leads a central bank to care about monetary and financial stability. This foundational approach does not necessarily suggest, however, a clear additional role for the central bank in the ownership, operation and/or regulation of payment systems. The nature and extent of such intervention depends critically on how large systemic risk externalities are believed to be. This, in turn, requires an appreciation of the ways in which different risks within a system – credit, liquidity, operational, etc. – might interact in the face of changes in payment system design.2 There appears to be relatively little consensus, however, on either the precise scale or the source of systemic risks within many payment systems. This is an issue taken up by several chapters in the volume. For example, Selgin argues that a bank is only exposed to the risk of another bank failing to meet a net obligation in a DNS system if it credits customer accounts before settlement occurs. But that is not something banks are required to do. Indeed, such customer credits can typically be reversed in the event of settlement not taking place. In this way, Selgin argues that the credit risk in DNS systems is not as severe as is often assumed and hence that RTGS systems might be a sledgehammer approach to cracking a nut. Moreover, although RTGS systems all but eliminate credit risk, they do so at the expense of requiring much higher levels of liquidity than DNS systems, and in this way potentially aggravate liquidity risk. The chapter by Lester et al. in this volume presents a search-theoretic framework to examine the trade-off between cost and risk inherent in the choice between RTGS and DNS payment systems. It shows that when the costs of settling payments on an RTGS basis are high, only a DNS equilibrium can exist. For intermediate values of costs either settlement rule may hold in equilibrium and there is nothing to suggest that agents in the economy will necessarily use the welfare superior mode of settlement. The results from the chapter thus support a role for central banks in encouraging the move from DNS to RTGS, as they did during the 1990s and as documented by Bech in his chapter in this volume. As stressed in the chapters by Green and Bech et al. in this volume, one design feature that can help to reduce the liquidity costs and hence improve the trade-off is for central banks to provide intraday liquidity at low cost. For example, in the United Kingdom and the Euro Area (as well as in a number of other countries) the central banks provide liquidity to the members of their large-value payment systems at a zero intraday interest rate against high-quality collateral. An alternative (or additional) way of improving the risk-efficiency trade-off is by designing payment systems that combine the liquidity-savings features of DNS with the finality offered by RTGS: so-called ‘hybrid’ systems. McAndrews and Trundle (2001) discuss the development of these systems and how they improve the risk-efficiency trade-off. One way the banks themselves can economize on liquidity costs is through the use of correspondent banks, which process payments on behalf of indirect

8

A.G. Haldane et al.

system participants. The chapter by Jackson and Manning in this volume argues that such ‘tiering’ may bring welfare benefits for the economy as a whole for three reasons: correspondent banks are likely to monitor their customers better than the central bank; correspondent banks can internalize payments, reducing further the collateral needed by the second-tier banks; and the opportunity cost of collateral is likely to be lower for the correspondent banks than it is for the second-tier banks. But they argue that tiering may also introduce costs into the economy since correspondent banks may not have sufficient incentive to monitor their customers, internalized payments might be subject to greater legal risk, and operational or financial problems at a settlement bank will disrupt not only its own payments, but also those of its customers. More generally, the move to RTGS has increased the demand for liquidity by financial firms and hence the potential for liquidity risk materializing in the banking system. This is one of the factors that has resulted in liquidity risk becoming an issue of increased importance for banking regulators over recent years.3 The chapter by Rochet in this volume looks at why it is necessary to regulate banks’ liquidity holdings and how regulators might go about doing this. He argues that liquidity regulation for banks can be justified by two different motives: limiting the risk and the impact of individual bank failures, and limiting the need for massive liquidity injections by the central bank in case of a macroeconomic shock. In his view, a simple form of ‘stock liquidity requirement’ can cover the objective of protecting small depositors. But he suggests that there is a need additionally for a second type of liquidity requirement, based on some ex ante indices of exposure to macroeconomic shocks by individual banks. This systemic liquidity requirement would limit the need for an ex post liquidity injection by the central bank in response to a systemic shock. This would be a radical departure from existing regulatory practices, but perhaps not an implausible one should liquidity risks continue to escalate.

The future of payment systems This book is called The Future of Payment Systems so this introduction would not be complete without some discussion of where payment systems might be headed. Several of the chapters in the volume address this issue (see Part IV). Drawing on Kahn and Roberds (2006), it seems likely that payment systems in the future will still fall into either ‘account-based’ or ‘store-of-value-based’ categories. Rosenblat (1999) argues that future retail payments are likely to be made using one or other of two things: a debit card (which would act in the same way as a credit card when the user was overdrawn) and a cash card (an anonymous stored-value card). The former represents an account-based system; the latter a store-of-value-based system. One interesting open question is whether cash cards will replace cash completely. Rosenblat argues that they will because if such cards require user authorization, there would be no incentive for anyone to steal them (unlike cash itself). One alternative form that the ‘cash substitute’ could take is units of mobile phone airtime; the technology already exists for


9

these to be transferred anonymously from one mobile phone to another and mobile phones are ubiquitous, helping deal with the network externality problem that issuers of cash cards have faced in the past. The only remaining issue is whether non-banks (i.e. the mobile phone companies) would be allowed by the public authorities to enable this: that is, issue their own money. In terms of wholesale payments, we can again consider the possibilities of different ‘store-of-value-based’ and ‘account-based’ systems. The e-settlement vision put forward by Leinonen in his chapter in this volume is one particularly radical version of a ‘store-of-value-based’ system. Under this system, wholesale payments are made directly between banks using encrypted bytes of information – central bank e-money – with no central processing. Leinonen argues that the efficiency gains from moving to such a system are large, but that the network externalities and increasing returns to scale operating in the payments industry and the lack of competition among banks act as barriers against the adoption of this more efficient technology. Nonetheless, he maintains that the new, and more efficient, technology of network-based e-settlement will eventually triumph over the centralized systems that we have today. In terms of ‘account-based’ systems, wholesale payments are already made in this way across accounts held at the central bank. One question here is whether these systems could be merged into one large payment system in which wholesale payments can be made between banks anywhere in the world. In his chapter in this volume, Pattinson suggests that the introduction of the Continuous Linked Settlement (CLS) system has taken us much closer to this position. There are already a cat’s cradle of links between the world’s largest wholesale payment systems, with CLS at its epicentre. Whether payment systems and regulators have fully taken into account the systemic risk implications of this increased interconnection between systems is, however, a moot point. In particular, liquidity risk – and the accompanying need for effective cross-border liquidity management – may have increased inadvertently as a result of this increased interconnection between infrastructures. However wholesale payment systems develop, there are likely to be implications for the demand for central bank money and, it could be argued, for the ability of the central bank to carry out monetary policy. The chapter by Millard and Saporta in this book surveys some of the literature on this. Woodford (2004) notes that as long as central bank money is used to settle payments there will always be some demand for it, even in the absence of central-bank-issued cash. This line of argument is discussed further in the chapter by Schmitz in this volume. Schmitz argues that there are strong economies of scale reasons for having a single unit of account. Given that the current unit of account corresponds to units of central bank money, and that changing to an alternative unit of account would be costly, it is efficient for central bank money to remain as the ultimate settlement asset. As long as this remains the case, the central bank can continue to carry out monetary policy, more or less, as it does now. And if the demand for central bank money were to fall to zero, the central bank could still carry out monetary policy provided it could control the supply of, or demand

10

A.G. Haldane et al.

for, the ultimate settlement asset (whatever it was) via reserve requirements in, and an interest rate corridor on central bank borrowing and lending of the ultimate settlement asset. Freedman makes a similar point in his chapter in this volume. He suggests, though, that how a central bank might act to control the supply of a settlement asset that were not its own money is unlikely to become a practical issue in the foreseeable future. As long as central banks maintain a stable value for their money, it is likely that it will remain the settlement asset of choice.

Conclusion We have not solved the conundrum with which we began this introduction. We hope, however, that this volume is some small step towards beginning to close the gap between the perceptions of practitioners, academics and the general public on the role and importance of payment systems. In particular, we hope this volume helps raise the interest of those who had previously thought payment systems obscure and demystified, at least to a degree, the technical language built up around them. With luck, readers might even feel spurred to tackle some of these fascinating and important issues for themselves.

Notes 1 Lacker (2006) makes a similar point, arguing that the interest rate should be related to the rate at which agents discount the utility they obtain from consumption if incentives – in this case, banks’ payment incentives during the day – are not to be adversely affected. 2 Definitions all taken from Bank for International Settlements (2003). 3 In his chapter in this volume, Rochet also mentions two other factors that have led to an increase in liquidity risk: increases in banking sector concentration, as well as in the complexity and size of financial markets, and the increased use of derivative products, which generate a large demand for liquidity.

References Allen, H., Christodoulou, G. and Millard, S.P. (2006) ‘Financial infrastructure and corporate governance’, Bank of England Working Paper No. 316. Bank for International Settlements (2003) A glossary of terms used in payments and settlement systems. Bech, M. and Garratt, R. (2003) ‘The intraday liquidity management game’, Journal of Economic Theory, 109: 198–219. Cifuentes, R. and Willison, M. (2006) ‘Why payment systems matter: measuring their benefits for the economy’, unpublished thesis, Bank of England. He, P., Huang, L. and Wright, R. (2005) ‘Money and banking in search equilibrium’, International Economic Review, 46: 637–70. Kahn, C.M. and Roberds, W. (2006) ‘An introduction to payments economics’, unpublished thesis, University of Illinois. Lacker, J. (2006) ‘Central Bank credit in the theory of money and payments’, speech


11

given at Federal Reserve Bank of New York Conference on Economics of Payments II, 29 March. Lester, B. (2005) ‘A model of interbank settlement’, unpublished thesis, University of Pennsylvania. McAndrews, J. and Rajan, S. (2000) ‘The timing and funding of Fedwire funds transfers’, Federal Reserve Bank of New York Economic Policy Review. McAndrews, J. and Trundle, J. (2001) ‘New payment system designs: causes and consequences’, Bank of England Financial Stability Review, 11: 127–36. Millard, S.P. (2006) ‘The foundations of money, payments and central banking: a review’, unpublished thesis, Bank of England. Millard, S.P. and Willison, M. (2006) ‘The welfare benefits of stable and efficient payment systems’, Bank of England Working Paper No. 301. Millard, S.P., Speight, G.E. and Willison, M. (2006) ‘Why do central banks observe a distinction between intraday and overnight interest rates?’, unpublished thesis, Bank of England. Norman, B., Shaw, R. and Speight, G.E. (2006) ‘The history of interbank settlement arrangements: exploring central banks’ role in the payment system’, unpublished thesis, Bank of England. Rosenblat, T.S. (1999) ‘What makes the money go round?’, PLD thesis, Massachusetts Institute of Technology. Schanz, J. (2006) ‘Innovation and ownership structure in payment systems’, unpublished thesis, Bank of England. Woodford, M. (2004) Interest and Prices: Foundation of a Theory of Monetary Policy, Princeton, NJ: Princeton University Press.

Part I

Payment systems and public policy

1

Central banks and payment systems Past, present and future Stephen Millard and Victoria Saporta1

Introduction Central banking and payment systems – mechanisms that enable the transfer of monetary value – are inextricably linked. In the past, institutions that developed into modern central banks stood at the top of the inter-bank payments hierarchy, providing the ultimate settlement asset exchanged by commercial banks when settling payments with each other. At present, modern central banks devote a considerable proportion of their resources to operating, overseeing and influencing developments in payment systems. In the future, innovations in payment system technology might permanently change the role of central banks, possibly even leading to their demise. And yet, the economics literature in the field is surprisingly scarce. With some honourable exceptions (including papers by the contributors to this volume), mainstream monetary economics has largely ignored the mechanics of how payments are actually made and banking theory has largely ignored the management of liquidity intraday. Even within central banks, payment systems are often treated as simply ‘the plumbing’ and left to technocrats. The aim of this chapter is to paint a broad-brush picture of the economic links between central banks and payments in the past, the present and the future. The purpose is ambitious and impossible to cover comprehensively in a single chapter – hence ‘broad-brush’. In particular, we start by arguing that the modern roles of central banks can be seen as natural outgrowths of their historical role in the inter-bank payments hierarchy. We then proceed to ask what are the characteristics of payment systems modern monetary authorities should be interested in? And how should this interest be made operational? Should central banks own, operate and/or oversee payment systems? We conclude with some tentative thoughts on how the payments landscape may evolve in the future and what that may mean for the future role of central banks. The chapter is organised as follows. We first provide background on the development of payment systems and central banking, arguing that historically they have been closely linked – the past. We then go on to analyse the role of modern-day central banks in the payment system, in particular in which systems should they be interested and how should they exercise this interest – the

16

S. Millard and V. Saporta

present. Finally, we offer ideas about the future direction of payment systems and of central bank involvement therein – the future.

Payment systems and central banking – the past Natural pyramiding Historically, the evolution of central banking can be traced back to the market’s natural demand for an efficient way to make payments. This natural demand can lead to the development of a hierarchy or pyramid in payments with the liabilities of a proto central bank at its apex, as the ‘settlement asset’ of choice. In other words, where institutions could provide a safe settlement asset that other banks use to settle obligations ultimately between themselves, they often developed the characteristics that in the twentieth century we came to associate with modern central banks. Payment systems form the means by which monetary value is transferred. Agents have a natural demand for a safe and verifiable asset – money – that they can use to transfer value in exchange for goods. This demand is derived from the low probability of the ‘double coincidence of wants’ necessary for trade in a barter economy (Jevons (1875) and, in a modern context, Kiyotaki and Wright (1989, 1993)). Given this asset, agents will eventually wish to find a way of being able to make payments – transfers of this asset – without having to carry it. There are at least two reasons for this. First, as suggested inter alia by He et al. (2005), money is susceptible to theft. Banks developed as places where people could deposit their gold for safekeeping. The banks would then issue their customers with receipts. These receipts represented a form of debt and, eventually, this debt became ‘transferable’ in the sense that it became possible for a merchant who wished to make a purchase to transfer the debt to the seller as payment for his goods. Final settlement occurred when the sellers went back to the bank to call in the debt. Second, as suggested inter alia by Kohn (1999), it was hard to verify the true value of different coins (the predominant form of money at this stage). Banks developed as places where agents could have their money counted and valued by money changers. As it was efficient for this process to only happen once, agents would leave their money – once counted and valued – with the money changers who would issue them with receipts. Payments were made with both payer and payee present at the bank. Where the payee did not hold an account at the payer’s bank, he either opened one or could ask the bank to transfer the money to his own bank. Since banks were close to each other, this was done by the payer’s banker walking over to the payee’s banker with the money. But, in an economy with many banks, it is inefficient for every agent to have an account with each and every bank and the banks themselves might be a long distance from each other. One solution is for each bank in the economy to have an account with all other banks and net obligations bilaterally with them. In a world with many banks this will tend to result in an inefficiently large number of

Central banks and payment systems

17

inter-bank accounts. A more efficient solution is for a hierarchy – or pyramid – of banks to develop, with banks at the bottom of the pyramid having accounts with correspondent banks in its upper tier which in turn have accounts with banks at the apex of the pyramid. Indeed, there is plenty of historical evidence that such pyramiding evolved naturally in a free-banking environment without the need for the state to superimpose and/or guarantee a ‘settlement institution’ at the apex of the pyramid.2 One example is the case of England (and, later, the United Kingdom). The Bank of England was founded in 1694 and was granted a number of privileges by the British Government, in return for its services in raising finance and managing the Government’s accounts.3 Due to these privileges, the Bank has been the largest and best capitalised bank in the United Kingdom for most of its history. Its large capital base and creditworthiness meant that it became the ‘custodian’ of choice – other banks naturally felt that it was the safest institution in which to hold their gold reserves, which they exchanged against Bank of England notes. Consequently, Bank of England notes (and later deposits) became the ultimate settlement asset for making payments, placing the Bank at the top of the payments pyramid in the United Kingdom. But, for most of its history, despite being the Government’s banker, the Bank did not enjoy an explicit government guarantee, nor was there an explicit or implicit acceptance that if the Bank chose to put the capital of its shareholders at risk the Government would step in to cover any resulting loss. For example, in 1890 the Chancellor of the Exchequer refused a request by the Governor of the Bank to guarantee its shareholders against loss if it were to support Barings Bank. A second example of natural pyramiding is the development of the Suffolk Bank system in Boston in the early nineteenth century. The development of this system is discussed in Goodhart (1988), Trivoli (1979) and Calomiris and Kahn (1996). At the time, ceteris paribus, Boston banks could issue fewer notes than their New England country competitors because the probability of a note being presented for payment varied negatively with the difficulty of travelling to the bank that issued it. This put the Boston banks at a competitive disadvantage to country banks and encouraged them to develop secure and systematic ways to redeem the various note issues that were circulating freely around the city. The Suffolk Bank ran the most successful system – it undertook to redeem at par the notes of country banks as long as they maintained sufficiently large deposits, topped up as necessary so as to make redemption at par possible. Moreover, the Suffolk Bank refused entry to its clearing system to banks it deemed not to have the requisite degree of integrity. In effect, it undertook an early form of supervision of banks. A third example relates to the arrangements for inter-bank payments in the United States during the period 1837–1913 (when there was no central bank in the country). Green and Todd (2001) explain that a hierarchy of correspondent bank relationships developed. Each small city had one or more correspondent banks and New York City had a number of banks that facilitated interregional

18


payments; that is, there was essentially a ‘mutualised cooperative’ at the top of the pyramid. Put in the words of Smith (1936): The conspicuous position held by the banks of New York City in this respect – in 1912 six or seven of them held about three-quarters of all banks’ balances – seemed to point to the existence of spontaneous tendencies to the pyramiding and centralisation of reserves and the natural development of a quasi-central banking agency, even if one is not superimposed. (our italics)4 There is, of course, the issue of whether natural pyramiding is socially optimal or whether the government may wish to intervene by creating an institution that sits at the summit. On one view – referred to as the ‘jaundiced view’ in Calomiris and Kahn (1996) – private systems such as the Suffolk Bank system are driven by large banks seeking to limit the supply of money and engage in monopoly pricing. Any gains are at the expense of the smaller banks and the public as a whole. An alternative view – the ‘sanguine view’ – is that such arrangements increase efficiency and reduce risk in the banking system. Calomiris and Kahn (1996) suggest that empirical evidence backs the sanguine view in the case of the Suffolk Bank system (see also Selgin and White (1994) for a similar view).5 But regardless of whether such natural pyramiding is socially optimal, the fact that it seems to occur raises the question of how many banks would naturally take this role at the top of the hierarchy? Does the market, in each currency, tend to one proto central bank or more? The relative standing of different banks and the structure of capital market flows in a country are important factors – as in the case of the Bank of England. Another important factor is the structure of the banking market.6 In an oligopolistic ‘free banking’ market with few banks, it may still be efficient for banks to hold bilateral correspondent accounts with each other, settling in each others’ monies, rather than in an outside settlement asset. According to Green and Todd (2001), in Canada banks did just this, until recently. In consequence, markets with a few large banks dominating the system may tend to develop flatter upper-tier structures. In contrast, in a unitbank system – that is, a system consisting of a large number of small independent units – efficiency considerations will lead the smaller units to seek an arrangement that would decrease the number of inter-bank relationships. In such systems ‘proto central banking agencies’ may develop naturally.7 Features of the settlement institution What are the financial features that such proto central banks need to display to enjoy a comparative advantage in performing the functions of the settlement institution at the apex of the pyramid? First, if a central bank is commercially-oriented – in practice, if it is privately-owned – it needs to find ways of overcoming various conflicts of interest. Banks whose payments are made via the settlement asset of the proto central


19

bank would need assurances that it is not using the information in their accounts to compete unfairly with them. Further, there may be a tension between a bank’s pursuit of profit and its role as a central bank. There is plenty of historical evidence that suggests that when the provider of the ultimate settlement asset is also a commercial bank, conflicts of interest ensue, especially during periods of financial crisis. For example, in 1793, the Bank of England was asked to aid some large country banks – it refused to do so and some important failures occurred that spilled over to the London market. Henry Thornton (1802) writes that ‘a sense of unfairness of the burden cast on the Bank by the large and sudden demands of the banking establishments in the country, probably contributed to an unwillingness to grant them relief’. Goodhart (1988), inter alia, describes a number of other examples, involving commercial rivalries among the Suffolk Bank and the New England country banks, the Bank of England and the London bill brokers and the Banque de France and potential commercial competitors. One way of overcoming the conflict of interest problem is for the central bank to stop competing with the other banks for non-bank business. In effect, this was how the Bank of England overcame the problem, withdrawing from all (new) commercial activity with non-bank entities between around 1880 and 1910 (Goodhart, 2004). Second, the banks whose payments are made using the central bank’s liabilities as settlement asset need to be confident in the credit quality and liquidity of this asset – where liquidity should be taken to mean the acceptability of this asset as a means of payment by others. At least one of three features recur in the development of institutions as central banks and help explain other banks’ willingness to use these assets: (a) the provider’s bank notes and deposits are backed by a commodity with intrinsic value (such as gold); (b) the provider has a very large capital base such that the probability of failing to realise its obligations is very small; and (c) the provider holds an explicit or implicit government guarantee. During different times in its history, the Bank of England had each of the features (a), (b) and (c). The fact that the Bank had the largest capital base of any bank in the United Kingdom well into the nineteenth century – feature (b) – was the key factor in explaining why the Bank’s liabilities became the settlement asset of choice. During this time, the Bank’s relative standing as the banker to the Government might have created the impression that it had an implicit government guarantee – feature (c) – but any such impression was certainly less firmly held than today. The Barings episode in 1890 is a concrete example of the Government refusing to underwrite the capital of the Bank. It was not until 1844 that the Banking Act placed restrictions on the Bank’s ability to print notes that were not backed by gold. It stipulated that the Bank had to hold gold reserves against all the notes it issued in excess of a fiduciary issue of £14 million – essentially forcing the Bank to display feature (a). However, the regulation was suspended during subsequent periods; in particular, during the liquidity crises that occurred in 1847, 1857 and 1866, the government allowed the Bank to issue additional notes not backed by gold. Again, however, there was no indication that the government would underwrite any losses the

20


Bank made as a result of intervention. In 1946, the Bank was nationalised – effectively giving it feature (c).8 But the Bank of England’s history may be unusual. In many cases, more than one bank exhibited features (a) and (b). Either you would expect to see these banks ‘jockeying for position’ as the central bank or you would expect to see flatter upper tier structures as in Canada or the New York Clearing House system with ultimate settlement (where necessary) carried out in gold or government bonds.9 In such cases, eventually central banks tended to be superimposed by the state to provide the ultimate settlement asset – as, for example, in the cases of the Bank of Canada, the Federal Reserve, the Reichsbank and the Swiss National Bank, among others. Often, but not always, this happened in response to banking crises that were perceived to result from the lack of a central bank being able to provide lender-of-last-resort assistance. For example, the Federal Reserve System was set up in 1914 in response to the banking panic of 1907 as a direct result of a perceived need for a government-backed lender of last resort in such circumstances. This is discussed in more detail below. Proto central banks, financial and monetary stability At present, central banks all around the world generally share two closely related core purposes – monetary and financial stability. In this sub-section we ask why central banks evolved as the natural candidates for taking on these two responsibilities and argue that the answer lies in the key role proto central banks played in the payment system. That is, these core purposes can be seen as natural outgrowths of a central bank’s role in payments. In Section 3, we reverse the question and ask, given that modern central banks are currently the public institutions commonly charged with the preservation of monetary and financial stability, what do these core functions imply for their modern interest, and active involvement, in payment systems? Historically, privately-owned settlement institutions that supplied the settlement asset at the top of the payments pyramid had a natural interest in ensuring the ability of their client base – the banking sector as a whole – to meet the public’s demand for liquidity. The reason for this is that if it allowed a solvent commercial bank to fail as a result of a run, it would only aggravate the situation and this could ultimately result in a run on itself. Also, assuming the commercial bank stayed in business, the central bank would make a high return on this lending (given that lender-of-last-resort assistance would typically be given at high rates of interest). Put differently, profit maximisation is consistent with Bagehot’s (1873) rule that a central bank should always lend to liquid but solvent institutions against collateral – that is, be a ‘lender of last resort’. Historical evidence backs this assertion. For example, the Bank of England provided lender-of-last-resort assistance during the financial crises of 1857 and 1866. Equally, the status of a proto central bank at the top of the payments pyramid derived from the fact that it was perceived to be ‘safe’ – that is, an institution with a large capital base, holding high quality assets. So a commerciallyoriented central bank would also need to be concerned about its own soundness.


21

This would give it incentives to be careful about to whom it should provide settlement accounts and to monitor these banks; one can think of this as an early form of banking supervision. In addition, it also had to weigh carefully the advantages of providing lender-of-last-resort assistance to the banking system to avoid a drop in its revenue stream against the risk of lending to an insolvent institution and making a loss that could decrease its capital base and threaten its reputation as the supplier of the ultimate settlement asset. As a result, proto central banks were more likely to let healthy banks go down than risk lending to unhealthy banks by mistake. Hence, in a fractional reserve system, central banks without a government guarantee have incentives to maintain financial stability – that is, grow their balance sheet to avoid crises – but at a sub-optimally low level. This is the typical justification for a role for the public sector in providing financial stability. Indeed, the full title of the Federal Reserve Act of 1913 reads, ‘An act to provide for the establishment of Federal reserve banks, to furnish an elastic currency, to afford means of rediscounting commercial paper, to establish a more effective supervision of banking in the United States, and for other purposes’ (our italics). Privately-owned providers of the ultimate settlement asset also have incentives to maintain the value of their liabilities. In particular, if the proto central bank printed more and more of its notes without a corresponding increase in the demand for them, the notes would fall in value relative to those of other banks. Eventually, the proto central bank would no longer be seen as ‘safe’ and it would lose the revenue it obtained from acting as the settlement institution. This is why these proto central banks emerged as natural candidates for ultimately being charged with maintaining ‘integrity’ and ‘confidence’ in the currency in modern fiat monetary systems.

Payment systems and central banking – the present Payment systems and central banks have evolved in tandem. In addition, the ultimate development of the core functions of modern central banks – monetary and financial stability – has been closely linked to their role in the provision of the ultimate settlement asset in the payment system. But, apart from providing the ultimate settlement asset, central banks have played a number of other roles in the provision of payment services including the ownership and operation of some payment systems, both retail and wholesale, and overseeing systems which they do not own or operate themselves. This raises a current pertinent policy issue for central banks. History aside and given their core functions of monetary and financial stability, to what extent, and why, should central banks play these other roles in payment systems in modern times? In what follows, we first tackle the ‘why’ question, asking specifically in which payment systems a modern central bank should be interested. We then tackle the question of how this interest might best be made operational.

22


In which payment systems should a central bank be interested? Modern central banks are typically charged with the provision of monetary and financial stability. Any attempt therefore to answer the question of why central banks should have an interest in payment systems should start with an analysis of how disruptions in payment systems could affect monetary and financial stability. And this, in turn, should start with a definition of what we mean by monetary and financial stability. To define ‘monetary stability’ it is useful to start from a definition of ‘money’. Money is normally defined by its four functions: unit of account, store of value, medium of exchange and means of deferred payment (settlement). By controlling inflation – that is, helping to ensure stable prices – the central bank enables money to perform the first two of its functions, though its roles as a medium of exchange and means of deferred payment are likely to be severely compromised in a situation of high inflation. But it is possible to think of situations in which, although inflation is low, money is still not able to perform its roles as a medium of exchange and means of deferred payment. Historically, shortages of coin have meant that agents in the economy have been unable to use money as a medium of exchange and have had to resort to barter. Alternatively, and in a modern context, the failure of a payment system might mean that agents are unable to use money held in bank accounts to make payments, leading to a loss of confidence in money more generally. This could happen regardless of the rate of inflation.10 This suggests a broad – monetary stability – objective for a central bank of ensuring that money can perform its functions of unit of account, store of value, medium of exchange and means of deferred payment in all states of the world. Financial instability is normally thought of as a situation in which shocks to the financial system – institutions, markets and financial infrastructure (including payment systems) – have contagious external effects elsewhere within the financial system with consequences for social welfare.11 It is worth noting that the financial stability responsibility of central banks can be derived from the broad monetary stability responsibility as we have defined it above, as problems in financial institutions and markets would clearly disrupt the ability of money to perform its functions of medium of exchange, store of value and means of deferred payment. Having defined the objectives of a central bank, we now consider what this implies for its interest in payment systems. To understand the characteristics of payment systems – means of transferring monetary value – one first needs to understand the economic purposes of money and of banks (given that commercial bank deposits form the bulk of the stock of money in modern economies). Money developed as a way of overcoming the problem of a lack of ‘double coincidence of wants’ in a barter economy, i.e. as a medium of exchange. The most basic form of money is ‘cash’. Where a central bank is the monopoly issuer of banknotes, a key element of its broad monetary stability remit will be ensuring that its banknotes can act as a reliable medium of exchange in all cir-


23

cumstances – in effect, to ensure that cash can always act as the payment medium of last resort. But cash forms only a small part of the total money supply; the bulk of money in the economy today exists in the form of bank deposits. ‘Banks’ originally developed to deal with two specific problems with cash: susceptibility to theft and the difficulty in verifying its true value. Because of these problems, cash was not always able to perform its functions of medium of exchange, unit of account, store of value and means of deferred payment. Banks enabled payments to be made with ‘bank money’ – banknotes, bills of exchange and payments ‘in bank’ – rather than with cash. In essence, banks and bank money enable ‘money’ to perform its functions more effectively. Where economic agents wish to make a payment to someone who has an account at the same bank, they can make the payment ‘in bank’. One can think of the internal accounting systems of banks over which such payments are made as a ‘payment system’ since they are allowing agents to make effective use of ‘bank money’. Where the payer and payee do not hold accounts with the same bank, then the payer will need to transfer a claim on his bank to the payee. Eventually, this claim will be deposited by the payee, leaving his bank with a claim on the payer’s bank. As this process will be happening for many agents banking with different banks, all the banks will be building up claims on each other, which need to be settled at some time. ‘Final settlement’ takes place via the transfer of an asset that the creditor bank is happy to accept. Although some banks will be happy to accept settlement in the money of other banks, there will be some level at which the banks will not accept settlement in each other’s money; settlement between such banks can only then take place through an ultimate settlement asset. This whole process – from the point at which a payer transfers a claim to a payee through to final settlement – defines an inter-bank payment system. Such systems enable money to perform its roles of a store of value, medium of exchange and means of deferred payment. Different payment systems have evolved over time to handle different types of payments. Some distinctions can be made between payments that support particular financial market transactions (e.g. unsecured inter-bank loans, securities purchases), payments that are made electronically or by paper, regular payments that can be preset in a bank’s systems, payments that are made for retail purchases in shops, etc. Disruptions to the systems used for making certain types of payment are likely to matter more to a central bank than to other systems. But in deriving the key set of characteristics for a central bank, we need to assess what matters for the ability of money – and bank money in particular – to perform its functions. This ability can be disrupted if a system that processes large values and/or volumes of payments is disrupted and the participants in the system cannot divert their payments to another system. Further, if disruptions to a single system led to a general loss of confidence in all payment systems, then bank money would again be prevented from effectively fulfilling its functions of a medium of exchange and means of deferred payment.12 Indeed, in this case, it is likely that agents would turn to cash to make their payments. This possibility

24


again emphasises the importance to a note-issuing central bank of ensuring confidence in its banknotes. The above analysis suggests that payment systems can be characterised in terms of three features: •

•

•

Size – other things equal, problems in a system handling a large value of payments are more likely to lead to contagious losses than problems in a system handling only a small value of payments, since the inter-bank exposures within them are likely to be higher relative to banking system capital. Problems in a system handling a large volume of payments are also likely to lead to the disruption of more transactions than problems in a system handling a small volume of payments. Types of payment – payment systems that support particular financial markets within which problems could result in financial instability; or systems that enable the payment of non-discretionary payments (such as wages, salaries and bills) as opposed to discretionary payments (such as retail purchases); or handle payments that are more or less urgent than payments handled by other systems. Availability of substitutes – that is, if a system were to develop problems, would agents still be able to make their payments by switching easily/ costlessly to an alternative system?

We can now attempt to map these characteristics into the broad monetary and financial stability objectives of a central bank. Agents’ demand to make payments is what creates a demand on the part of the banks for the ultimate settlement asset – central bank money (whether held as cash or accounts at the central bank). And it is precisely this demand for its liabilities that enables the central bank to carry out monetary policy. So, in order to carry out monetary policy so as to maintain stable prices, a central bank will need to take an interest in the payment systems in which agents use the ultimate settlement asset. In most developed economies, these will almost certainly include the large value payment systems where unsecured money market transactions are settled and the security settlement systems for government bonds and other central bank-eligible securities.13 In addition, with the broad objective of ensuring that money can perform its functions in all states of the world, a central bank should, in principle, also take an interest in disruptions to systems that could lead to contagious losses among banks and other financial institutions, disruptions to other systems and/or problems elsewhere in the financial system more generally – that is, financial instability – since these problems could, in turn, prevent money from fulfilling its functions. In most developed economies, the set of relevant systems is likely to comprise the large-value payment and clearing and settlement systems that support financial market transactions in money, securities and derivatives. A central bank should also take an interest, in conjunction, where relevant, with the prudential supervisory authority, in disruptions to the internal systems of


25

banks where such disruptions prevented customers of that bank from making payments with contagious effects on the rest of the financial system. A recent example is the disruption of Bank of New York (BoNY) following the 11 September 2001, terrorist attacks.14 Beyond these systems, the central bank will need to consider what would be the minimum necessary involvement to enable bank money to carry out its functions in both normal and crisis states of the world. One answer would be to rely on cash. But in a situation where there were no other means for agents to make payments, banks could potentially run into liquidity problems with systemic consequences. In such situations, the monetary authority would need to engineer a huge increase in the supply of cash, which it may not be able to do with adequate speed (depending on the amounts involved). Moreover, reverting to a ‘cash economy’ would reintroduce the welfare costs that bank money eliminates and will be welfare-reducing in itself. A better answer would be to ensure that, in each state of the world, there was at least one payment system available to a sufficient proportion of the population. There would be different options available to a central bank to make this objective operational. The central bank could provide such a system itself or ensure that it was in place via regulation or provide a back-up system ready to come into operation in times of crisis. Alternatively, it could ensure – via regulation, oversight, encouraging competition, etc. – that there were many substitute systems available for agents to use. How should central banks’ interest in payment systems be made operational? Central banks have an interest in ensuring that the payment systems that are important to the functioning of the monetary and financial system remain open in both normal and crisis states. But this interest does not necessarily imply that a central bank or any other public authority should intervene in the payment system. Public sector intervention can only be justified in the presence of market failures and only then if cost-effective instruments that can mitigate these failures exist. In this section, we first set out the externalities that may justify public sector intervention in payment systems before going on to describe and evaluate a set of stylised models of intervention that could mitigate such externalities. Market failures and payment systems Payment systems can give rise to the following externalities that may justify public sector intervention: •

As discussed in Bank of England (2005), payment systems create systemic risk externalities that private sector owners and operators of payment systems may not internalise, in the absence of intervention. Although there is no universally agreed definition of ‘systemic risk’, it is generally accepted

26

•

•

•

S. Millard and V. Saporta that it includes the following four contagion-related risks, all of which may be mitigated by the public sector influencing the way payment systems are designed and operate: (i) the risk that the failure of one member of a system will lead to other members suffering losses and potentially also failing; (ii) the risk that operational disruptions to one member, or features of the design of the payment system, may lead to ‘liquidity sinks’, whereby a member may receive large pay-ins but for technical reasons cannot, or for other reasons does not, make pay-outs, which in turn exposes other members of the system to liquidity pressures; (iii) the risk that the operational or financial failure of a payment system itself may disrupt financial markets with knock-on consequences for banks and other financial intermediaries; and (iv) the risk that news about the failure of one payment system leads to failures of other payment systems that were not subject to the original shock through ‘loss of confidence’ effects operating through agents’ expectations. Systemic risk externalities fall squarely within the remit of a central bank. That said, to the extent that some of the systemic risks can be mitigated through strengthening the credit and operational risk management of individual members of the payment system, the prudential regulatory authority – in countries where it is separate from the central bank – should also be involved. Moreover, for the case of the internal payment systems of individual banks, where these are judged to be of systemic importance to the payment and settlement system as a whole, the interest would clearly lie with the prudential regulator. As discussed in Bolt and Humphrey (2005), payment systems are characterised by network externalities and increasing returns to scale which in turn imply a strong tendency to monopoly. While relevant to the central bank, one could argue that regulating against this type of market failure would more naturally fall to competition authorities.15 A number of payment systems are organised as member-owned cooperatives and are typically beset by collective action problems.16 Such problems can potentially lead to an ‘inefficient’ set of payment systems being available to consumers. Again, while relevant to the central banks, it is not clear that the central bank is best placed in dealing with such problems; again, the competition authorities may be better suited. Finally, the informational asymmetries between banks and their customers, which lead to ‘consumer protection/conduct of business issues’ within the banking industry, may give rise to a need for regulating the information provided by banks on their provision of payment services.17 Clearly, the government agency charged with regulating the conduct of business of banks has the natural locus here.

Models of intervention No form of public sector intervention to correct market failures will be effective without appropriate instruments for monitoring performance and appropriate


27

powers of enforcement; that is, some lever to ensure compliance with a set of standards or principles. Together, these determine the degree of control over the payment system that the public sector is able to apply in mitigating the social costs of the market failures set out above. While the central bank will be in a position to exercise a degree of control through its role as the provider of the ultimate settlement asset and its influence as banker to the banks, it may be desirable/necessary for the public sector at large to supplement this by taking additional actions. In practice, these will be drawn from choices taken along three dimensions: (i) public sector ownership of the system; (ii) public sector operation of the infrastructure; and (iii) public sector oversight or regulation of the system. Before going any further, it is useful to define each of these terms. By ownership of the system, we refer here to the case in which the public sector, typically the central bank, has an ownership stake in the entity governing the payment system, or a role in its governance. A controlling stake (greater than 50 per cent) affords the pubic sector the ability to design the system in accordance with its own objectives (preferences) and enforce continued compliance. By operation of the infrastructure, we refer to active public sector engagement in the design, implementation and operation of all, or a sub-set, of the elements such as software, hardware, communication networks, data centres and contingency sites that underpin modern-day payment infrastructures. In practice, the central bank may be the authority best-placed to take on this role, as it can leverage the systems it maintains to hold, monitor and control accounts for financial institutions that bank with it. It is worth noting, however, that operation of the payments infrastructure is a separable activity from the provision of the ultimate settlement asset. While acting as the settlement agent gives the central bank direct access to information on the payment flows of system members that settle in central bank money, operation of the infrastructure provides a complementary, yet distinct, instrument for exercising direct control over operational capacity and performance. By oversight of the system we refer to day-to-day regulatory activity that ensures continued compliance with a set of minimum standards and design principles (e.g. the Core principles for systemically important payment systems set by the Bank of International Settlements (2001)). In practice, this activity is invariably carried out by the central bank but, both conceptually and in practice, it need not be. Oversight, with adequate powers/influence, may be a substitute vehicle for enforcement in the absence of public ownership of the system. Each activity can, in practice, be carried out with varying degrees of formality and intensity: ownership can range from no role in governance, through a seat on the Board, through to a controlling stake; operation can range from complete outsourcing with appropriate control/monitoring procedures through to design, maintenance and operation of all key components of the infrastructure; and oversight, if carried out at all, can be carried out with limited and informal powers through to extensive and formal powers of direction and enforcement. For illustrative purposes, however, we can think of a spectrum of possible models of intervention, in which these three activities are somehow combined.

28


Model 2

Model 1

OWNS

Model 4

Model 5

OVERSEES Model 6

Model 3 OPERATES

Figure 1.1 Stylised models of intervention in payment systems.

These models can be shown to lie along the surfaces and in the interior of a three-dimensional cube (as shown in Figure 1.1). An evaluation of the effectiveness of the various alternative models of intervention can then start with an assessment based around the six relevant corner models identified in the diagram: model 1 (‘owner/operator’) lies at one corner of the cube, with the central bank assuming a controlling ownership stake and operating the key components of the infrastructure; model 2 (‘owner’) involves a controlling stake in the ownership of relevant payment systems, but with the infrastructure operated by a private sector provider; model 3 (‘operator’) involves the central bank operating key components of the infrastructure on behalf of a private sector owner; model 4 (‘overseer’) involves pure oversight with powers of enforcement (without excluding the role of central banks in providing the ultimate settlement asset for systems that settle in central bank money); model 5 (‘overseer/operator’) combines oversight and operation of the infrastructure, without ownership; and, finally, in model 6 (‘laissez-faire’), there is no active public sector intervention. The two further corner models, not identified in the diagram, combine both ownership and oversight. While this might be feasible – i.e. central banks may wish to perform an internal audit function on their provision of payment services – it is difficult to imagine that, in such circumstances, oversight offers incremental value in achieving central bank objectives. Hence, we do not consider these in our analysis. Evaluation of corner models of intervention Models in which the central bank assumes ownership (i.e. models tending towards 1 and 2 in Figure 1.1) may be highly effective in achieving the central


29

bank’s objective of ensuring that payment systems that are critical in the operation of the monetary and financial system are free from systemic risk externalities. With complete power of enforcement and access to information for monitoring purposes, these models can give the central bank power to determine the level of resilience and systemic risk mitigation built into payment systems. A counter argument here, however, is that public ownership can undermine the continued engagement of members of the payment system. Without a stake in how the system is run, the banking system may be more inclined to divert flows to competing, and perhaps more risky, systems and payments vehicles. Indeed, it is important to recognise that powers of direction over the characteristics of the system do not translate into powers of direction over its usage: a central bank may be able to force every bank to have an account, but cannot force them to use it. An additional consideration might be that models incorporating ownership, especially those combining this with operation, will entail a cost to the central bank and ultimately the taxpayer. If full-cost recovery could not be achieved, the central bank under these models would, in effect, be subsidising the provision of the payments infrastructure. And although a subsidy might be justified if central bank intervention was directed towards addressing the private sector’s underprovision of a public good in the absence of negative externalities, a priori it appears less appropriate when private sector solutions are plagued by negative externalities. In addition, a central bank might not be indifferent between models with and without operation. First, and most critically, outsourced operation of the infrastructure raises the question of control over access to central bank money settlement. Ensuring balance sheet integrity is an important element of a central bank’s monetary stability objective and systems settling in central bank money, particularly those settling in real time, can potentially expose the central bank to losses arising from system-error or mismanagement. Where settlement is outsourced, the central bank can only control such exposure at arms’ length. Different central banks may have different risk tolerances in this regard. Second, the higher the degree of operational control exercised, the greater the central bank’s access to information and data and the greater its ability to both monitor and enforce operational standards applied to the system. Third, outsourced operation potentially introduces principal-agent frictions, if the private firm’s incentives cannot be aligned perfectly with those of the central bank. Finally, the cost considerations outlined above are likely to be an order of magnitude greater when the central bank is also the operator. Under models tending towards model 3 (‘operator’), the central bank has a high degree of operational control, but no formal powers of enforcement afforded by either ownership or oversight. Here, the central bank’s monitoring capability may be enhanced and, subject to its ability to recover costs (or its willingness to subsidise), the central bank can assume a certain level of control over initiatives to address operational risk. It may also be that regular and close engagement with the users may afford the central bank additional influence and

30


control over other aspects of the owner’s strategy. However, this model in itself affords no legal means to enforce risk-mitigating actions over which it has no direct control – e.g. over the settlement model adopted by the payment system, the rules covering direct credit caps, membership rules, etc. Under an appropriately designed pure-oversight model (one tending towards model 4), on the other hand, the overseer would have powers to force a private system owner to comply with any desired risk-mitigating action, which may be as effective as under an ownership model. However, being a step-removed from the operation of the system, it is unlikely that the overseer would be able to obtain the same quality of information on operational processes and performance. That said, with adequate powers, the overseer would be able to prescribe the requisite level of reporting/information provision, and hence the gap here might not be so large. Also, with any specific actions imposed by the overseer implemented at the members’ own expense, there is no issue around cost recovery: oversight essentially constitutes taxation of the members. Finally, as noted above, a model with mutual ownership might encourage the continued engagement of the members. Model 5 is essentially a hybrid of models 3 and 4, whereby the public sector operates the payment system and also has regulatory powers to impose penalties to enforce performance. With close operational control supported by powers of enforcement, and no issues around disengagement of members, such a model would perhaps be the most effective in achieving the central bank’s objectives. A caveat here is that the combination of operation of the infrastructure and oversight might expose the central bank to conflicts of interest, either apparent or real. In the presence of market failures and cost-effective instruments allowing their mitigation, model 6 would clearly be the least appropriate in achieving public sector objectives. So far we have focussed on evaluating models of intervention in terms of their effectiveness in mitigating systemic risk externalities that fall clearly within the remit of modern central banks. But as we noted earlier, there are several other market failures that may raise distinct but equally valid public sector concerns that may or may not fall within a central bank’s official remit. In seeking to eliminate these market failures, the public sector at large will have an interest in facilitating the development of an infrastructure that promotes welfare through providing the optimal set of payment instruments (direct debits, credits, cards, cash and the like), infrastructural platforms (the IT networks that ensure that payment messages are sent, processed and settled) and settlement assets (combination of bank and central bank money), at the optimal price-quality combinations.18 The models considered in Figure 1.1 might equally be applied in the pursuit of these public policy objectives. Something close to a pure oversight model (model 4) is likely to be most effective in this regard. A regulator, if appropriately empowered, can catalyse infrastructural development beneficial to welfare by resolving coordination problems in the industry under the threat of regulation


31

– this is effectively the line the Office of Fair Trading in the United Kingdom pursued in persuading the industry to build a Faster Payments Service.19 A regulator is also well-placed in taking steps to eliminate excessive rents in a sector characterised by network externalities and scale economies (not unlike in other network industries, such as telecoms or electricity generators). And by addressing market failures through regulation rather than subsidy, oversight would seem to be the appropriate public policy instrument to mitigate the frictions that lead to the innovation and consumer protection market failures outlined above. Of course, if different authorities have different interests and multiple overlapping oversight regimes emerge, conflicts or duplication of effort can arise. In this regard, clarity as to each authority’s responsibility is essential. If this strategy does not work, the public sector may seek to solve the problem by building and operating the service itself, for a period at least.20 This is the approach many central banks have taken historically. But permanent public sector operation of the payments infrastructure is unlikely to be as beneficial to this objective as private sector ownership subject to regulation because the public sector is unlikely to be as innovative as a private sector provider in developing efficient, high quality and cheap IT network solutions. Model 2, in which the central bank owns but does not operate the system, will rank higher than models that also involve operation because, despite owning the system, private sector providers of infrastructure are free to compete for the central bank’s business. Table 1.1 ranks the stylised models of intervention against their effectiveness in mitigating the systemic risk externalities that fall squarely within most central banks’ remits and the other frictions set out on pages 25–6. It is clear from the table that, while models involving ownership may be effective in delivering on central banks’ monetary and financial stability objectives, these models could stifle competition and innovation in the payments sphere and hence potentially compromise other public sector objectives. If combined with outsourced operation, however, there may be a lower efficiency cost to public ownership. Taking all public sector objectives together, an oversight-only model might be preferred, as long as sufficient powers of enforcement could be conferred upon the overseer whose responsibilities were clearly defined. Table 1.1 Ranking models of intervention Model

Degree of effectiveness in mitigating systemic risk externalities

Degree of effectiveness in achieving additional public sector objectives

Model 1: owner/operator Model 2: owner Model 3: operator Model 4: overseer Model 5: overseer/operator Model 6: laissez-faire

High/medium High/medium Medium/low High/medium High Low

Low Medium Medium/low High Medium/low Low

32


Practical considerations Whether such an oversight-only solution will be chosen will depend on a number of practical considerations, including the risk preferences and budget sets of the public authorities involved (which in turn may depend on political economy considerations) and a weighing of the costs of introducing a statutorybased oversight regime relative to the expected welfare benefits. One factor affecting costs would be the prevailing institutional structure for financial system regulation and oversight in the country concerned. In countries where multiple authorities have overlapping interests in systemically important payment systems – e.g. in countries where the prudential regulator is not the same agency as the central bank – the net (of costs) social benefits of introducing formal legislation delineating responsibilities and powers between agencies may turn out to be lower than in an arrangement whereby the central bank and the prudential regulator agree on how they should co-operate in discharging their respective roles, leveraging on their existing powers.21 Moreover, in an increasingly globalised financial and monetary system, payment systems that are of systemic importance to a particular country may not be located in that country at all – e.g. Euroclear, which owns the UK securities settlement system CREST, is incorporated in Belgium and CLS Bank, which settles wholesale foreign exchange transactions between sterling and all major currencies and which is incorporated in the United States. For these systems, an oversight arrangement based on domestic-only statutory powers is unlikely to be enforceable; instead, international co-operative oversight arrangements between ‘host’ and ‘home’ authorities become necessary. There is also the question of whether a uniform approach to intervention is necessary. Indeed, on an examination of models of intervention across various countries, consistency of approach seems to be the exception rather than the rule. A series of tables in the Annex maps a set of international payment systems to the corner models shown in Figure 1.1. It is immediately clear that, of the countries included in the tables, Canada is the only one adopting the same model of intervention for all of the payment systems covered in the tables. Elsewhere, central banks have tended to adopt a mix of models, with models involving ownership and/or operation of the system more prevalent in the case of largevalue payment systems or payment systems embedded in securities settlement systems. Historical (or path) dependence appears to be one factor behind the cross-country variation in models of intervention.22 That said, the different models applied may also reflect a judgement on the extent of each system’s contribution to the market failures the intervention is intended to address, or the balance of central bank interests versus those of other public authorities. For example, a ‘lighter central bank touch’ might be envisaged for retail versus wholesale payment systems, where competition, innovation and consumer protection concerns may weigh more heavily.


33

Payment systems and central banking – the future In this section we offer some tentative ideas about the future direction of payments and central bank involvement in payment systems. We first examine whether cash will endure and we then move on to discuss how wholesale financial transactions might be made in the future. The chapters by Freedman, Leinonen, Pattinson and Schmitz in Part IV of this volume discuss these issues in more detail. We also tentatively explore the implications for the ability of central banks to carry out their core monetary and financial stability functions in the future. How will consumers pay for goods and services in the future? When thinking about the question of how consumers will pay for goods in the future, one question to ask is whether cash will endure or whether it will be replaced by some form of electronic money. Cash transactions, though declining, still form about 75 per cent of personal payments in the United Kingdom, although their value tends to be small (abstracting from illegal black market transactions). In thinking about why cash is so enduring, it is important to note that any replacement would have to offer its user the same level of anonymity, universal acceptability and recognisability; no current alternative has ever done this. Another advantage of cash over, say, credit and debit cards is the fact that final payment takes place simultaneously with the provision of the good or service; the seller is not exposed to settlement risk. Agents like anonymity. Indeed, the lack of anonymity in credit cards, for example, has led to the large and increasing problem of identity theft; in the United Kingdom, for instance, credit card fraud in 2004 totalled $966 million.23 An alternative view – expounded by Buiter (2005) – is that since anonymity is of most use to criminals, there may be a case for the government to do away with legal tender currency issue by the state, while making sure that private note issuance continued to be banned. In that case, all payments would have to be made using a medium – such as credit cards – in which the purchaser of any good could always be identified. Drehmann et al. (2002) point out that such a policy ‘would be appallingly illiberal’ and so unlikely to be contemplated. But the fact that no current alternative to cash is able to match its attributes does not mean that such an alternative will not exist in the future. One might think that eventually, in place of cash, some form of e-money will exist that offers the same complete anonymity, universal acceptability and recognisability as cash but will not be useable by anyone other than the holder of the e-money. This will reduce the incentive of others to steal the e-money and so make it a ‘safer’ asset to hold than cash. But it may well be that cash and e-money can coexist. A key issue here is anonymity. In practice, regulators have forced e-money transactions to be auditable (that is, limited anonymity); if they continue to do so, cash would always have the advantage of anonymity. E-money will always be subject to ‘operational risk’ where cash is not – at least once it has been taken out of the

34


ATM.24 Finally, unless the central bank were to issue the e-money (or, equivalently, the government were to underwrite the issuing companies), then it would always be subject to credit risk. Dowd (1998) and Friedman (1999) suggest that the demand for central bank money – and cash in particular – has fallen dramatically over recent years and that it will, possibly, fall to zero eventually. Given this, they argue that changes in the supply of central bank money – that is, monetary policy – will increasingly have less impact on the wider economy, in the limit having no impact at all. In effect, central bank money is just one of a number of competing monies; the price level itself, at that point, would need to be tied to a commodity or, alternatively, a bundle of financial assets. But Woodford (2004) notes that as long as central bank money is the ultimate settlement asset – that is, there is a need for it in order that banks can make payments to each other – there would always be some demand for it even in the absence of central bank notes and this would mean that central banks could carry out monetary policy exactly as before. Buiter (2005) argues along similar lines, noting that even if there were no demand for central bank money in normal times, there would still be a need for central banks to supply liquidity to the banking sector in times of stress. Hence, central bank deposits are likely to be replaced by overdraft facilities, lines of credit or other contingent claims on central bank money; if the demand for such claims were sufficiently stable, the monetary authority could still set short-term interest rates. How will wholesale payments be made in the future? With respect to wholesale payments, the trends seem to point to two opposing corner outcomes: one integrated payment system perhaps covering the whole world (a mega-version of the Continuous Linked Settlement (CLS) system that settles foreign exchange transactions for the major world currencies) or a large number of competing private payment systems. The benefits of the first corner outcome would be large savings in collateral, IT communications and other costs; the downside would be the ‘single point of failure’ problem associated with a massive concentration of risk in one system and the general inefficiencies usually associated with monopoly providers. In terms of implications for central banks, the net effect on monetary and financial stability would appear to depend upon the level of systemic risk in the system (or systems), together with the degree of control that central banks can exercise over this system(s). Aligning incentives of different national overseers-regulators will become an even more important determinant of which outcome prevails. A future of competitive private systems granting agents the ability to transfer financial assets to pay for goods in real time forms the basis of the vision of the future put forward by King (1999).25 In a world of competitive systems, the systems would need to be linked, but this need not be via a central bank provided there was general agreement on acceptable settlement assets. In such a world, central banks would become pure regulators operating monetary policy by tightly defining the unit of account – like ‘weights and measures’ inspectors.


35

The unit of account could be one of the settlement assets but need not be. Greenfield and Yeager (1983), for example, analyse an economy in which the unit of account is a bundle of standardised commodities (conceivably financial assets) and is not used as a medium of exchange. In addition, accounts held in it cannot be redeemed for the ‘bundle’ but instead are redeemed for something of intrinsic value. In this world, a central bank could define this unit of account and offer a continuous quote of its value against all the financial assets used as media of exchange. But, as pointed out by Selgin and White (1994), the information costs associated with doing this are large; information and transactions costs for agents in the economy can be minimised by everyone quoting prices in the dominant medium of exchange. Financial stability policy in this world would involve ensuring the operational integrity of the payment systems and the appropriate level of systemic risk in the ‘custodian sector’ – as this is what the banking sector would effectively become. Whether it would make sense for the central bank to do this rather than another government body is, of course, an open question. But there is still the issue of what the acceptable settlement asset would be. Kiyotaki and Moore (2004) argue that payments must be made in terms of money rather than other assets because there is a commitment problem with repaying loans with returns from investment. Kocherlakota (2004), reviewing this paper, says that it fails to address why assets that could pay a higher return than money and that have no enforcement problems are not used: e.g. government debt.26 Wallace (1988) makes the same point by noting that the acceptability of central bank money today hinges on the fact that it is ‘legal tender’ and the fact that legal restrictions stop other banks from issuing interest-yielding bearer bonds that could act as ‘money’ while offering a higher return than central bank money. Selgin and White (2005) make the point that central bank money is likely to continue being preferred to other assets for two reasons: (a) because it defines the unit of account, it will not be subject to bid-ask spreads, and (b) payment in it is final given its ‘legal tender’ status. If central bank money remained, then central banks would remain at the top of the payments pyramid. This is the vision presented by Harry Leinonen in his chapter in this volume. He envisages a situation in which individuals can settle payments between each other in real time and in central bank money over a network without a central payments-processing infrastructure. For this to work, the central bank will need to provide liquidity in the form of ‘bytes of encrypted information’ (i.e. central bank e-money!) that it can distribute to banks over this same network. Once these have been distributed, the central bank would not need to keep a tab on them until the end of the day (or any other point at which they may charge interest or remunerate balances). On the other hand, if government bonds became ‘money’, then the top of the payments pyramid would instead be represented by a Central Securities Depository (CSD) that would be responsible for transferring securities from one account to another across its books. Essentially, this is exactly like a payment system run by a proto central bank, as we described in the first section of this chapter. In this case, the CSD would have become the central bank.

36


As the rate of inflation would be determined by government borrowing – which would fix the supply of government bonds – monetary policy would become indistinguishable from fiscal policy. Guarding against systemic risk externalities would involve ensuring that the CSD were operationally robust and reliable. The rationale would be the same as for payment systems today: if the CSD were to go down, agents would have to find an alternative – and expensive – way of making payments such as transferring a valuable commodity or, at the very worst, resorting to barter. If financial assets could be transferred from hand to hand there would be no need for a settlement institution at the ‘apex’ of the pyramid. Again, monetary policy would involve ensuring that these financial assets were what agents said they were – just as goldsmiths and money changers assured the quality of money in the past – and ensuring that the supply of these financial assets did not grow quickly. Likewise, guarding against systemic risk externalities would involve ensuring that the process of transferring financial assets from hand to hand was operationally secure and that the custodian sector (which would presumably intermediate in these assets) was appropriately regulated. The likelihood and desirability of these scenarios is open to question. And even if market forces were to drive the financial system in one direction, governments could still enforce, to some degree, settlement in certain assets via certain institutions through legal restrictions.

Concluding remarks Based on an eclectic choice of raw material, this chapter has painted a broadbrush picture of the economic links between central banks and payments in the past, the present and the future. In particular, we argued that the core functions of monetary and financial stability of modern-day central banks can be traced back to their payments role as providers of the ultimate settlement asset. We then argued that central banks charged with the preservation of monetary and financial stability should have an interest in payment systems that process large values/volumes and/or payment types, disruptions to which can give rise to significant social welfare costs. We then evaluated different models of public intervention, concluding that an oversight model backed with appropriate enforcement powers might provide the best balance between central bank objectives and other broader public policy objectives. Whether such a model is chosen in practice will depend on a number of practical considerations, including the risk preferences and budget sets of public authorities and the weighing of social benefits versus the costs of introducing a statute-based regime; this, in turn, will depend on country-specific institutional arrangements for the regulation and oversight of the financial system at large. Finally, we ended with some tentative thoughts about how retail and wholesale payments may be made in the future and how this might affect central banks’ ability to conduct monetary policy and to guard against systemic risk.


37

Annex: International models of intervention in payment systems Tables 1.A1 to 1.A3 explore existing models for intervention in selected countries’ large-value payments systems, automated clearing houses (ACHs) and embedded payment systems of securities settlement systems. These are mapped to the first five corner models illustrated in Figure 1.1. (There are currently, in the countries considered, no existing examples of a laissez-faire (model 6) approach for these categories of payment system, so this model has been excluded. To re-cap, the models considered are: Model 1: own and operate Model 2: own only Model 3: operate only Model 4: oversee only Model 5: operate and oversee.

Table 1.A1 G10 models for intervention in large-value payment systems Country

Domestic LVPS(s)

Australia Belgium Canada France

HVCS ELLIPS LVTS TBF PNS RTGSplus BI-REL BOJ-NET TOP NICS SCP RIX SIC CHAPS* Fedwire CHIPS TARGET EURO1

Germany Italy Japan Netherlands Norway New Zealand Sweden Switzerland UK US ECB

Model 1

Model 2

✓

Model 5 ✓ ✓

✓

✓ ✓ ✓ ✓

✓

Model 4

✓

✓

✓

Model 3

✓ ✓

✓ ✓

✓ ✓

Note *It should be noted that, although operated and overseen by the Bank of England, CHAPS does not map perfectly to model 5 as oversight is conducted without formal powers of enforcement.

38


Table 1.A2 G10 models for intervention in ACHs Country

Domestic ACH(s)

Australia

BECS CECS CEC ACSS SIT RPS BI-COMP* Zengin Interpay NICS Retail ISL BGC DTA LSV BACS EPN ACH

Belgium Canada France Germany Italy Japan Netherlands Norway New Zealand Sweden Switzerland UK US

Model 1

Model 2

Model 3

Model 4 ✓ ✓ ✓ ✓

✓ ✓

✓

✓

✓ ✓ ✓ ✓

Model 5

✓

✓

✓ ✓ ✓

Note *Operation of one of the two sub-systems settling across BI-COMP is carried out by SIA on behalf of the Bank of Italy

In the case of large-value systems, the majority of central banks own and operate (model 1). The most notable exceptions here are the United Kingdom, Australia, New Zealand and Belgium, where the infrastructure for the largevalue system is operated by the central bank, but the system is privately owned; and Switzerland, where the central bank owns but does not operate the system. There are some other examples of pure-oversight models for large-value systems, but these occur where private and public large-value systems co-exist (CHIPS in the United States; PNS in France; and Euro-1 in the Euro area). The pure oversight model is most prevalent in the case of ACHs. Only in the cases of the United States and Germany does the central bank own and operate an ACH (in the United States, this is in competition with a private sector provider, EPN), while in Italy the central bank owns the system but operates only part of the infrastructure. In Belgium, oversight is combined with a role in operating the infrastructure. Finally, in the case of securities settlement systems, most central banks have adopted either model 1, 4 or 5. As for its large-value system, Switzerland is an outlier here, having adopted model 2 (ownership with outsourced operation). The examples of model 1 tend to be where either a separate system exists for settlement of government securities (United States, Japan and Belgium) or where the cash leg of securities transactions occurs via a central bank owned and operated large-value system. Otherwise, most central banks have adopted an oversight model, with some also assuming some operational involvement, depending on whether an interfaced or integrated settlement model is applied.

US

Netherlands Norway New Zealand Switzerland UK

Fedwire Securities DTC

Austraclear NBB Clearing CIK CDSX VPC Euroclear France Clearstream Banking AG Monte Titoli BOJ-NET JGB Services JASDEC Necigef VPO Austraclear NZ SWX CREST

Australia Belgium

Canada Sweden France Germany Italy Japan

Domestic SSS(s)

Country

✓

✓

✓ ✓

✓

✓

Model 1

✓

Model 2

Model 3

✓

✓

✓

✓

Model 4

Interfaced model Settles through LVPS Settles through LVPS Integrated model Interfaced model Interfaced model Settles through LVPS Settles through LVPS Interfaced model Interfaced model

✓

✓ ✓

Cash transfers onto DTC accounts take place through Fedwire

Settles through LVPS A hybrid-integrated model, sitting between models 4 and 5. DvP occurs with finality in CREST across memo. accounts mirrored by cash movements in the Bank’s RTGS system

Interfaced model

✓

✓ ✓

Notes

Model 5

Table 1.A3 G10 models for intervention in the embedded payment systems of securities settlement systems

40


Notes 1 The views expressed are those of the authors and do not necessarily reflect those of the Bank of England. The chapter has benefited from comments from Helen Allen, Andrew Bailey, Morten Bech, Paul Bedford, Alastair Clark, Charles Goodhart, Andrew Gracie, John Jackson, Nigel Jenkinson, Charles Kahn, Ana Lasaosa, Harry Leinonen, Chris Mann, Mark Manning, Adrian Penalver, David Rule, George Speight, Matthew Willison, Al Wilson and Jing Yang. A number of the arguments presented were refined following discussions with participants at the June 2006 Bank of England workshop on ‘payment economics’. Any errors are entirely our own. 2 Here we follow BIS (2001) and define as a ‘settlement institution’ the institution across whose books settlement takes place, or, put differently, the institution that supplies the asset which members of the system have agreed to accept for settlement of their obligations – the ‘settlement asset’. 3 For example, until 1826, the Bank was the only joint stock bank – other banks’ capital was constrained to the fortunes of a maximum of six partners. And early in the eighteenth century, forging Bank of England notes was punishable by death (Collins, 1988). 4 Cited in Goodhart (1988), page 35. The term ‘quasi-central banking agency’ in the quote includes both private banks – as in the Suffolk Bank example – and mutuallyowned clearing houses – as in the New York Clearing House example. 5 Even if tendencies that lead to the centralisation of reserves do not result in monopoly pricing, there might still be a rationale for government intervention in ‘superimposing’ a central bank. In particular, during financial crises, conflicts of interest might prevent commercially-oriented central banking agencies from behaving in a socially optimal manner. 6 There is a substantial literature that investigates the economic reasons why different firms adopt different internal management ‘hierarchies’ – see, for example, Rajan and Zingales (2001). It is possible that this literature can shed light on the development of different payment hierarchies. 7 The clearing arrangements developed by the Suffolk Bank and by the New York Clearing House in the nineteenth century in the United States fall into this category. Another example is provided by the Schulze-Delitsch savings banks in Germany which in the nineteenth Century sought to clear through Dresdner Bank which in turn developed into a proto central bank (Goodhart, 1988). 8 The Act of 1833, gave other banks the right to redeem their own liabilities with Bank of England notes rather than gold – that is, Bank of England notes became legal tender. The impact of the Act was far from startling, however, because Bank of England notes had been de facto (rather than de jure) legal tender for a long time (Collins, 1988). 9 At times when liquidity was particularly tight (that is, when gold had been leaving the system), the New York Clearing House (CH) issued liabilities backed by commercial bank notes that banks could use to settle their multilateral net obligations. It issued them to member banks against holdings of the notes of other banks; so, when a bank received another bank’s notes as a deposit, it could post these with the CH in exchange for a clearing house loan certificate. On such occasions, the CH would operate as a proto central bank in a fractional reserve system – that is, it would supply a settlement asset (the clearing house loan securities) that was not fully supported by either a commodity with intrinsic value or a government guarantee (Timberlake, 1984). 10 The Irish banking strike of 1966 is a good example where agents were unable to use bank accounts to make payments and were unable to withdraw cash. In this case, money was completely unable to fulfil its role as a medium of exchange at a time when inflation was low. See Murphy, 1978.


41

11 For a broader definition of financial stability see Haldane et al. (2004). 12 For example, an ‘informational externality’ can be based on imprecise information about fundamentals – e.g. all affected systems depend on the same telecoms infrastructure which agents believe with some probability to have been the source of the disruption of the original system – or conceivably be purely ‘sunspot’ driven – that is, agents stop using another payment system based on the belief that others will stop using it and their beliefs turn out to be self-fulfilling) 13 As an aside, it is worth noting that for ‘balance sheet’ control reasons, there is also a clear need for a central bank to be interested in the system for getting ‘cash’ out into the economy (e.g. the Note Circulation Scheme in the United Kingdom) as well as any payment system whose workings lead to the injection/withdrawal of central bank money into/out of the economy (e.g. in the United Kingdom, the large value payment of CHAPS and the security settlement system of CREST). 14 BoNY and JP Morgan Chase – the so-called ‘clearing banks’ – settle across their books over 75 per cent of transfer volume in US government securities, placing them at two critical nodes in the network of inter-bank payment flows (McAndrews and Potter, 2002). Following the attacks on 11 September, BoNY became a liquidity sink as payments were made to it which it was temporarily unable to recycle by making pay-outs. At some point BoNY was reported to be overdue of $100 billion in payments and unable to make payments or lend funds. Failures to deliver US government securities rose from $1.7 billion per day the week of 5 September to $190 billion the week ending 19 September. Absent massive intervention by the Federal Reserve, money and securities markets are likely to have been severely disrupted (Lacker, 2004). 15 It is possible of course that the central bank is formally charged with guarding against both systemic risk and competition-related externalities. This is the case, for example, in Australia. International comparisons, however, reveal that the Reserve Bank of Australia is the exception rather than the rule in this respect. See Bank of England, 2005: Table 1. 16 The classic reference on collective action problem in a mutual is Hart and Moore (1995). 17 This suggestion was put forward by Cruickshank (2000). 18 By ‘quality’ we mean features such as robust anti-fraud security, speedy clearing cycles and ease of use. 19 See www.oft.gov.uk/News/Press+releases/2005/94–05.htm for the press announcement by the OFT. Further information on how this position was reached can be found by following the links on this page. There are numerous other examples. In the late 1970s, for example, the Securities Exchange Commission (SEC) pursued a similar strategy in catalysing the establishment of DTC (the user-owned company that settles equities and corporate bonds in the US) and the European Commission is currently pursuing a similar approach for catalysing the creation of a Single Euro Payments Area (SEPA) in retail payments. 20 This is effectively what the Bank of England did when it built CREST, the central securities depository and large-value securities settlement system in the United Kingdom, in the mid-1990s, following a failed private sector initiative to build an electronic settlement service for equities. And it is, in effect, what the Eurosystem is suggesting in its recent proposal to build a central-bank owned and operated securities settlement system for Euro-denominated securities (TARGET2-Securities). See the press release issued by the European Central Bank on 7 July 2006 (www.ecb.int/ press/pr/activities/paym/html/index.en.html). 21 This is the current arrangement in the United Kingdom. A publicly available memorandum of understanding between Her Majesty’s Treasury, the Bank of England and the Financial Services Authority allocates responsibility for oversight of payment systems that are systemically significant to the United Kingdom to the Bank of

42

22

23 24 25

26

S. Millard and V. Saporta England. That said, the Bank of England appears to be an outlier among central banks, internationally, in not having such an arrangement backed by statutory powers. See Bank of England, 2005: Table 1. In countries with fragmented banking systems, one of the purposes of central banking institutions when they were founded was to solve the coordination problems that prevented the establishment of a single inter-bank payments network (see Norman et al., 2006). The large fixed costs involved in setting up payment systems from scratch explains why, once a reasonably efficient model of ownership and operation is established (regardless of whether it is central bank owned or privately owned), shifts to different models are observed infrequently. This is the path-dependence element of the historical explanation. Kahn and Roberds (2005) analyse this issue in the context of a model of a simple economy with credit arrangements. Of course, cash can be lost or stolen; indeed, the papers by He et al. (2005) and Millard and Willison (2006) suggest this possibility as a reason why payment systems developed in the first place. Capie et al. (2003) interpret King’s (1999) vision as one of electronic barter where goods are exchanged for goods in real time using electronic means. They argue that the transaction and information costs that led to the development of money in the first place cannot ever be removed by technology; hence, there will always be the need for (central bank) money. Of course, if another financial asset became an ‘acceptable’ asset in the sense we go on to describe, such an asset would realise the advantages of money described in their paper; we would be back at the conclusion that central bank money would vanish. Of course, there is nothing to stop the central bank paying interest on its money, at which point the distinction between central bank money and government debt essentially vanishes and fiscal and monetary policy would be, literally, the same thing.

References Bagehot, W. (1873) Lombard Street: A description of the money market, London: Henry S. King. Bank for International Settlements (BIS) (2001) Core principals for systemically important payment systems, Committee on payment and settlement systems. Bank of England (2005) Payment System Oversight Report 2004, January. Bolt, W. and Humphrey, D. (2005) ‘Public good issues in TARGET: natural monopoly, scale economies, network effects and cost allocation’, European Central Bank Working Paper No. 505. Buiter, W.H. (2005) ‘New developments in monetary economics: two ghosts, two eccentricities, a fallacy, a mirage and a mythos’, Economic Journal, Vol. 115, pages C1–31. Calomiris, C.W. and Kahn, C.M. (1996) ‘The efficiency of self-regulated payments systems: learning from the Suffolk system’, Journal of Money, Credit and Banking, Vol. 28, No. 4, pages 766–97. Capie, F.H., Tsomocos, D.P. and Wood, G.E. (2003) ‘E-barter versus fiat money: will central banks survive?’, Bank of England Working Paper No. 197. Collins, M. (1988) Money and banking in the UK: A history, London: Croom Helm. Cruickshank, D. (2000) Competition in UK banking: A report to the Chancellor of the Exchequer, Norwich: Her Majesty’s Stationary Office. Dowd, K. (1998) ‘Monetary policy in the 21st century: an impossible task?’, CATO Journal, Vol. 17, pages 327–31. Drehmann, M., Goodhart, C.A.E. and Krueger, M. (2002) ‘The challenges facing foreign


43

currency usage: will traditional transactions medium be able to resist competition from new technologies?’, Economic Policy, Vol. 17, pages 193–227. Friedman, B.M. (1999) ‘The future of monetary policy: the central bank as an army with only a signal corps?’, International Finance, Vol. 2, No. 3, pages 321–38. Goodhart, C.A.E. (1988) The evolution of central banks, Cambridge, Massachusetts: MIT Press. Goodhart, C.A.E. (2004) ‘Financial supervision from an historical perspective’, Paper given to Bank of Finland Conference on ‘The structure of financial regulation’. Green, E.J. and Todd, R. (2001) ‘Thoughts on the Fed’s role in the payment system’, Federal Reserve of Minneapolis Quarterly Review, Winter, pages 12–27. Greenfield, R.L. and Yeager, L.B. (1993) ‘A laissez-faire approach to monetary stability’, in White, L (ed.), Free banking. Volume 3: Modern theory and policy, Aldershot: Elgar. Haldane, A., Hall, S., Saporta, V. and Tanaka, M. (2004) ‘Financial stability and macroeconomic models’, Bank of England Financial Stability Review, June, pages 80–8. Hart, O. and Moore, J. (1995) ‘The Governance of Exchanges: members’ co-operatives versus outside ownership’, LSE Financial Markets Group Discussion Paper No. 229. He, P., Huang, L. and Wright, R. (2005) ‘Money and banking in search equilibrium’, International Economic Review, Vol. 46, pages 637–70. Jevons, W.S. (1875) Money and the mechanisms of exchange, New York: D. Appleton and Company. Kahn, C.M. and Roberds, W. (2005) ‘Identity, identity theft and credit’, University of Illinois, mimeo. King, M.A. (1999) ‘Challenges for monetary policy: old and new’, Bank of England Quarterly Bulletin, November, pages 428–38. Kiyotaki, N. and Moore, J. (2004) ‘Liquidity and asset prices’, London School of Economics, mimeo. Kiyotaki, N. and Wright, R. (1989) ‘On money as a medium of exchange’, Journal of Political Economy, Vol. 97, No. 4, pages 927–54. Kiyotaki, N. and Wright, R. (1993) ‘A search-theoretic approach to monetary economics’, American Economic Review, Vol. 83, No. 1, pages 63–77. Kocherlakota, N.R. (2004) ‘Optimal monetary policy: what we know and what we don’t know’, University of Minnesota, mimeo. Kohn, M. (1999) ‘Early deposit banking’, Dartmouth University Working Paper No. 99–03. Lacker, J. (2004) ‘Payment system disruptions and the federal reserve, following September 11, 2001’, Journal of Monetary Economics, Vol. 51, pages 935–65. McAndrews, J. and Potter, S. (2002) ‘Liquidity effects of the events of September 11, 2001’, Federal Reserve Bank of New York, Economic Policy Review, Vol. 8, No. 2, pages 59–79. Millard, S.P. and Willison, M. (2006) ‘The welfare benefits of stable and efficient payment systems’, Bank of England Working Paper No. 301. Murphy, A.E. (1978) ‘Money in an economy without banks: the case of Ireland’, Manchester School of Economic and Social Studies, Vol. 46, No. 1, pages 41–50. Norman, B., Shaw, R. and Speight, G. (2006) ‘The history of interbank settlement arrangements: exploring central banks’ role in the payment system’, Bank of England, mimeo. Rajan, R. and Zingales, L. (2001) ‘The firm as a dedicated hierarchy: a theory of the origins and growth of firms’, Quarterly Journal of Economics, Vol. 116, pages 805–51.

44


Selgin, G.A. and White, L.H. (1994) ‘How would the invisible hand handle money?’, Journal of Economic Literature, Vol. 32, No. 4, pages 1718–49. Selgin, G.A. and White, L.H. (2005) ‘The future of fiat money: a Mengerian perspective’, University of Georgia, mimeo. Smith, V. (1936) The rationale for central banking, London: P&S King Ltd. Thornton, H. (1802) An enquiry into the nature and effects of paper credit of Great Britain, London: Hatchard. Timberlake Jr, R.H. (1984) ‘The central banking role of clearinghouse associations’, Journal of Money, Credit and Banking, Vol. 16, No. 1, pages 1–15. Trivoli, G. (1979) The Suffolk Bank: a study of a free-enterprise clearing system, London: Adam Smith Institute. Wallace, N. (1988) ‘A suggestion for oversimplifying the theory of money’, Economic Journal, Vol. 390, pages 25–36. Woodford, M. (2004) Interest and prices: foundation of a theory of monetary policy, Princeton, NJ: Princeton University Press.

2

The role of a central bank in payment systems Edward J. Green

Introduction The central banks of all industrialised countries specialise to some extent in what roles they play in their respective payment systems. Each defers to other entities in its respective economy, both private and public, to assume roles that it elects not to fill. Nevertheless, there is considerable variation across industrialised economies in how broad or focused a role the central bank assumes. Does the body of economic learning about central banking and payment systems have anything to say about what is the preferred point along this spectrum, or about the range of acceptable points? I suggest here that this learning does have an implication. Specifically, the central bank is a specialised organisation that is uniquely able to offer free, short-term credit on illiquid collateral to other financial intermediaries. It provides this service via a set of settlement accounts on its books for those intermediaries. A central bank does well, and arguably does best, by specialising in providing these services while leaving other roles (including clearing of retail and commercial transactions, transmission of payment messages and commercial regulation of payment intermediaries) to different entities that can specialise in those tasks.

Commencing the argument Let me commence this argument by defining some terms, and also by indicating briefly how some of its premises are justified. A payment system is a comprehensive system for settling the obligations of purchasers of goods, services and financial assets, and of their intermediaries. Examples are large-value payment systems that settle obligations between banks, and payment-card networks through which many consumer transactions are settled. Sometimes an individual component of such a comprehensive system is also called a payment system, but that is not relevant here. Rather, the discussion will concern systems in which numerous entities, including providers of such specialised components, interact to provide an overall service. The institutional arrangement for meshing these various contributions functions essentially as a market. As in other markets, comparative advantage determines which

46

E.J. Green

entity should best do what. So the topic of this chapter might be restated as: What is the comparative advantage of a central bank with respect to the other participants in a payment system? Define a central bank to be an institution that • •

•

has both the government and private financial intermediaries (that will generically be called ‘banks’) as account holders; is therefore in a position to influence overall interbank credit market conditions through its credit policies towards account-holding banks and its intermediation on behalf of the government; and has been given lead public policy responsibility for achieving credit market conditions that foster prosperity and economic stability, and price stability in particular.

In view of the position and responsibility just described, the central bank is in a privileged position to • • •

manage a system of accounts for interbank settlement; provide short-term credit to facilitate settlements; and accept illiquid collateral for that credit.

These are the three activities in which a central bank should specialise. The argument for this proposal rests on three premises. The first premise is that it is economically efficient for financial intermediaries to have access to credit at a cost (either a direct interest price or an indirect collateral cost) just high enough to balance the small risk that the central bank assumes by extending it. This assertion is supported by a model of payment system credit due to Freeman (1996a, 1996b). Freeman derived the optimality of such essentially free credit along with a zero inflation-rate policy, and Zhou (2000) has further shown in that framework that free credit is conditionally optimal even when inflation is positive. I provide an informal discussion of Freeman’s model and its logic, and of why I consider the model to be a convincing economic theory, in a separate chapter (Green, 2006) elsewhere in this volume. The second premise is that a central bank is uniquely able to make short-term loans on illiquid collateral. A collateral asset is illiquid if it will possibly take time considerably beyond the maturity date of the loan to sell the asset for its full value, although the full value (including a market rate of appreciation) can ultimately be obtained. A private-sector intermediary that lends subject to a balance-sheet constraint cannot wait a long time to recoup the value of loans in default, so it cannot afford to take illiquid assets for collateral. In contrast, a central bank in a fiat-money regime issues credit by creating outside money. Suppose that the only constraint on such money creation, imposed by the requirement of price stability, is that the money so emitted must be reabsorbed eventually by sale of the collateral, or through repayment of the loan. This constraint does not rule out acceptance of illiquid assets as collateral. This premise

The role of a central bank in payment system

47

can be derived formally in a suitably parameterised overlapping-generations model of money, and is a consequence more generally of formal or informal theories in which inflation is a function of the long-run rate of money growth. It is not a consequence of a ‘fiscal theory’ of the price level, such as the theory of Woodford (2003) discussed in Green (2006), nor of a generic overlappinggenerations model in which the price level is indeterminate (cf. Brock and Scheinkman, 1980.) However, neither of those models (or any other model of which I am aware) implies that a central bank cannot accept illiquid collateral or that doing so would make it more difficult or less likely than otherwise to achieve price stability. The third premise is that not only does a central bank have a comparative advantage relative to most other organisations in supplying credit to banks, but there are likely to be some diseconomies of scope when a central bank attempts to play other roles as well. Because of those diseconomies, at least prima facie, a central bank has a comparative disadvantage at playing such other roles. Richard Todd and I have argued in some detail for this premise in a recent paper (Green and Todd, 2001) concerning the role of the US Federal Reserve System in consumer and commercial payments.

Must a monetary authority be involved with payment systems? The topic of this chapter has been framed as a question about the industrial organisation of the payment system: What should a central bank do, or refrain from doing, in order to enhance the performance of the market for payment services? However, a central bank is the natural candidate to play the macroeconomic roles of issuing base money and regulating the price level. If performing those macroeconomic functions is a central bank’s primary role, then perhaps the question framed above is of secondary importance. The primary question should be, what level or kind of involvement with payment systems must the central bank maintain, in order to function effectively as a monetary authority? Let me briefly examine this question, before returning to issues of industrial organisation. Stipulate, for a moment, that a central bank must issue money in order to be an effective monetary authority. There is a long-standing issue in monetary economics, regarding whether a central bank needs to monopolise the provision of money. To a large extent, arguments for a central bank monopoly have assumed that money is being used to exact a seignorage tax. If issuing money is profitable for the government, the argument goes, then it is profitable for everyone, so there will be a glut of money in the absence of monopoly. Today, though, seignorage is recognised to be an inefficient tax. It is universally thought that, except perhaps during a temporary public-finance emergency, a central bank can better contribute to public welfare by avoiding inflation than by raising seignorage that results in inflation. Issuing money according to a low-inflation policy is not a profitable activity, so a glut of money from private issuance need not be

48

E.J. Green

feared. That is, the traditional argument for making the central bank a monopolist in money issuance is moot, according to current understanding of how monetary policy should be conducted. There are other, more recent arguments regarding the potential benefits and risks of permitting private-sector intermediaries to issue money. Research by Cavalcanti and Wallace (1999) and Wallace (2005) suggests a potential benefit, essentially that private issuers can target more precisely than the central bank the recipients of the money that is created. Research by Smith (1988) suggests a potential cost – exposure to the possibility of sunspot equilibrium. These are representative of economists’ arguments about this topic. It is noteworthy that the arguments on all sides of the topic assume that central bank money and private monies are perfect substitutes. This assumption contrasts with the discussion among central bankers and bank supervisors, regarding whether or not central bank money is an inherently superior settlement asset. (For example, one of the CPSS ‘Core Principles for Systemically Important Payment Systems’ (2001) seems to suppose that central bank money is superior. Some scepticism towards that position is expressed in a comment letter on the draft core principles that was submitted to the CPSS by the Federal Reserve Bank of Chicago (2000). This contrast evidences a more fundamental distinction. The policymakers’ discussion focuses on whether or not some degree of central bank monopoly in money issuance would improve the stability of payment systems, while the economists’ discussion focuses on whether or not monopoly would improve the stability of the monetary system and of the real economy. It is the economist’s discussion that is relevant to my preliminary question here. Lately there has been some discussion of whether or not, in principle, a central bank really does need to issue money, or indeed will be able to issue money in a future environment where advances in payment technology have taken full effect. One view, expressed informally by King (1999), is that money might become difficult to issue, and that such difficulty might significantly impair the effectiveness of monetary policy. An alternative view, taken by Woodford (2003) and others, is that the effectiveness of monetary policy is based on the central bank’s ability to set a nominal interest rate, and that such rate setting can be accomplished with zero net issuance of money in equilibrium. This ‘fiscal theory’ view implies that a central bank can be an effective monetary authority without being involved in the payment system in any way whatsoever. The considerations just discussed have to do with the possibility that some form of involvement with payment systems may intrinsically benefit the conduct of monetary policy or the pursuit of financial stability. One also hears arguments, at least in the United States, that involvement with payment systems can provide some extrinsic benefit to the central bank. One such argument is that information obtained from involvement with payment systems can profitably be factored into monetary policy decision making. There is scant evidence for that supposition, though. The minutes of the FOMC, the Fed’s policy-making committee, have seldom, if ever, mentioned such information. Another argument,


49

usually privately expressed, is that involvement with payment systems garners valuable political support for the central bank. Setting aside the serious question about whether or not it is proper for (or in long-term interest of) the central bank to engage in interest group politics, consider whether or not involvement with payment systems is politically effective in the short run. In the US situation, this claim is superficially plausible because the US banking system includes thousands of small ‘community banks’, scattered throughout every Congressional district. The owners of these banks are a cohesive, national, political lobby. However, they mobilise to support Fed policies that they perceive to give them subsidy or regulatory advantage over their competitors. They are not known for supporting the Fed when it is necessary to raise interest rates, which is the occasion when political support might be helpful to monetary policy. In summary, the argument that involvement with payment systems confers extrinsic, informational or political advantages for monetary policy making is not convincing. In general, the idea that involvement with payment systems confers any extrinsic advantage on monetary policy making has no support, in the context either of the US or other countries. Economists differ among themselves about whether or not some form of involvement in payment systems confers an intrinsic monetary-policy advantage, but the intrinsically beneficial involvement, if any, does not extend beyond the three roles that I identified in the introduction. Even if a central bank becomes a better monetary authority on account of performing these functions, there is not a consensus among economists that it must have a monopoly position. In particular, there is no consensus among economists that private clearinghouses should be prevented or discouraged from playing the same settlement roles as a central bank plays.

The market structure of payments Now I return to the main investigation, regarding the benefit that some form of central bank involvement may confer on payment systems. This discussion draws heavily on Green and Todd (2001), which pertains specifically to the Federal Reserve’s involvement in US consumer and commercial payment systems. Recall that a payment system can be viewed as a market in which various types of entity provide complementary services to payors and payees, as well as intermediate-good services to the payment intermediaries with which those end-users deal directly. Which services should be provided by which form of entities involves questions of market structure and comparative advantage. Regarding market structure, on the production side, many payment technologies have high fixed costs and exhibit declining average costs throughout a very wide quantity range. In addition there are economies of scope, such as the opportunity for several payment systems to share a common informationtransmission infrastructure. On the demand side, there are ‘network externalities’ – the preference of payors and payees to co-ordinate with one another on which payment system they use. These features distinguish markets for payment services from the neoclassical model of a competitive market. However, the

50

E.J. Green

assumptions of the neoclassical model constitute sufficient – not necessary – conditions for competitive equilibrium to exist and to be efficient. Several decades of experience with ‘deregulation’ largely corroborate the theory of contestable markets, according to which fairly nonintrusive regulation supports efficient outcomes in industries having economies of scale and scope on the production side. Winston (1993) estimates that ‘deregulation’ has achieved a 7–9 per cent improvement in contribution to GNP from the US sectors affected by regulatory reform, without significant distributional side-effects. With respect to network externalities, the situation is less clear-cut. One analysis, based on decentralised, myopic decision making by consumers, suggests that this demand side phenomenon may lead to inefficient outcomes. A different analysis, based on a cooperative solution concept for markets with network externalities, predicts efficient outcomes. (Weinberg (1997) discusses these two theories.) The conflict between these two theories was essentially the crux of the contentious Microsoft case in competition law. The preceding paragraph has dealt with competition among payment system operators. Another issue is strategic interaction among participants in payment systems. Bech and Garrett (2003) formalise one theory. They study strategic interaction within a single payment date, and find that ‘gridlock’ can be an inefficient outcome of interaction. Each bank desires to be in a credit position versus the payment system as much of the time as possible, but this is a zero-sum game among the participants. Participants non-cooperatively delay their payments in order to retain their clearing balances for as long as possible, so in equilibrium payments are not made until the end of the settlement day. In Bech and Garrett’s model, there is a public cost attached to this, and thus an economic inefficiency. McAndrews and Rajan (2000) document an intraday pattern of Fedwire payments that they attribute to strategic interaction. They emphasise the very high level of payment activity in the late afternoon, close to the time when the CHIPS payment system settles over Fedwire. They suspect that even payment orders received in the morning are often not executed until this time. This phenomenon plausibly reflects banks’ desire to avoid being in a debit position vis-à-vis Fedwire, a motivation that is easy to understand because the Fed charges interest on intraday credit. Many Fedwire payments are not time sensitive for the transactors, as long as they are settled by the end of the day of submission to the payment system. Thus, while banks seem to be playing the zero-sum game modelled by Bech and Garrett, the feature of their model that early settlement is a public good does not plausibly hold for Fedwire. Of course, there is a public good problem if time sensitive payments are being delayed, but McAndrews and Rajan do not provide evidence of that. If a payor informs its bank that a payment is time sensitive, then the bank’s cost from alienating that customer by holding back the payment would credibly exceed the gain from minimising interest on an intraday debt by doing so. Moreover, the private gain to a bank from delaying a payment is certainly exacerbated by, and may be nothing but an artifact of, the Fed’s pricing of intraday credit. If it is efficient to offer free intraday credit, as Freeman (1996a) and Zhou (2000) argue, then any inefficiency in the timing of


51

Fedwire transactions is arguably a result of the Fed’s daylight credit policy, rather than being a market failure requiring pricing of daylight credit to remedy.

The comparative advantage of a central bank I explained earlier that a payment system is, in effect, a market where various specialised entities collaborate to produce services. These entities notably include commercial banks, correspondent banks, clearinghouses and the firms that provide technology infrastructure, as well as central banks. The division of labour among these entities is determined by comparative advantage, and by economies of scope internal to each of the entities. The general presumption in economics as a whole is that the operation of these forces produces efficient outcomes. The gist of my preceding discussion of market structure is that this presumption applies to payment systems in particular. The comparative advantage of the central bank consists in maintaining deposit accounts for banks and in providing short-term credit to, and effecting transfers of balances among, those accounts as a means of settling interbank obligations. This characterisation reflects both economic history and economic theory. Historically, central banks have been chartered to perform two functions. One is to be an intermediary between the government and its lenders, enabling the government to obtain credit by ensuring that implicit default through inflation will occur only in genuine national emergencies. The other is to serve broad public interest as the trustworthy and neutral apex of a hierarchy of banks that, in turn, provide the nonbank public with accounts used to settle financial business and personal payments by transfer balances. Indeed, there is an economy of scope between these functions that gives the central bank comparative advantage in performing the latter. Since almost all banks need to transfer funds from their customers to the government to pay taxes, the government’s bank is in a natural position to serve as apex of the correspondent-banking hierarchy. This role as apex puts the central bank in a unique and distinguished position in the payments business. Its role with respect to banks is closely analogous to the role the banks play with respect to their nonbank customers – including netting, extension of credit and concomitant monitoring of creditworthiness. Moreover, just as banks are often structured to avoid conflicts of interest with their own nonbank customers, central banks evolved in part to avoid conflicts of interest with banks. Market demand rose for a special-purpose intermediary (that is, one that does not do business with nonbank customers) that could play the roles just mentioned without the incentive conflicts to which a commercial bank serving its peers would be exposed. The most immediate incentive conflict – the temptation to steal customers’ profitable nonbank customers – is ruled out (except in emergency conditions) explicitly by the central bank charter, which prohibits lending to nonbank entities under normal circumstances. Nonprofit status and other features of governance are designed further to control potential conflict of interest that might arise through the central bank’s discharge of its payment system functions.

52

E.J. Green

This historically-oriented description of the role of the central bank in the payment system is consonant with the body of recent economic theory regarding central banks described earlier. Together, history and theory suggest that there are two payment system functions that a central bank is better able than other institutions (except, perhaps, a clearinghouse) to perform for banks. First, the central bank can manage, in the broad public interest, a system of accounts that all banks are eligible to own, and that the banks can use to settle interbank transactions. Second, by extending credit to banks, the central bank can provide the benefits of netting and immediate finality of payments. Its ability to perform these functions, and particularly its position of neutrality and trust among the public and the institutions that it serves, is the unique strength of the central bank in the payment system. From this finding, together with the general principle that the public is best served when each institution in the economy focuses its resources in its area of unique strength. I conclude that central banks should restrict their involvement with payment systems principally to the accountmanagement and lending roles just discussed.

Do economies of scope justify broader involvement? The restrictive conclusion that has just been stated might be relaxed to some extent by taking the possibility of economies of scope into account. An economy of scope is a situation in which an entity is able to perform one task more efficiently than otherwise, because it is also performing a related task. Thus, while the central bank might not have a comparative advantage in performing some task on a stand-alone basis, the relationship of the task to account management and lending activities might give the central bank an advantage over other entities to perform it. The aftermath of the terrorist attack in the United States that took place in 2001 was a notable instance in which officials asserted that an economy of scope had been demonstrated between operation of a payment system by a country’s central bank and the core financial stability objective of that central bank. Among its effects, the attack impaired the interbank collection of cheques for the better part of a week. During that interval, the Federal Reserve agreed to accept legal possession of cheques that had not been transferred physically from its customer banks, and to give those banks immediate credit for the value of the cheques. This credit for cheques in the process of collection was a significant source of central bank credit beyond what was given overtly through the discount window. Numerous Federal Reserve officials characterised this extension of credit, during a crisis in which financial stability was at risk, as something that the Fed would not have been able to accomplish if it had not been the operator of a retail payment system. In fact, the extension of credit exemplified lending on illiquid collateral (in this case, cheques in the process of collection), which is one of the activities in which a central bank possesses a comparative advantage over other institutions. The Fed allows borrowers from the discount window to retain custody of collateral assets under appropriate safeguards, so


53

the fact that customer banks’ cheques could not be brought to Federal Reserve premises did not automatically disqualify them as collateral assets for a discount loan. While the fine print of the Fed’s discount regulations might have disqualified their acceptance, an ad hoc arrangement to accept them during the crisis could have been made very easily and immediately. Moreover, it seems likely that cheques in the process of collection were not contemplated as discount collateral precisely because they already serve routinely as collateral for credit extended in the course of the Fed’s cheque collection business. This oversight would surely be corrected in the course of an exit by the Fed from that business. In conclusion, because the Federal Reserve’s operation of a retail payment system does not enable it to lend on illiquid collateral that could not otherwise be accepted, the perceived economy of scope between operating a payment system and responding to a financial-stability crisis is illusory. More generally, in practice there are almost always pros and cons associated with prima facie economies of scope between central bank functions. For example, historically account management and lending to banks were argued to have had an economy of scope with bank supervision, but many countries have established separate supervisory authorities in recognition that any economy of scope is outweighed by the costs of giving that responsibility to the central bank. The careful econometric research required to substantiate that an economy of scope exists has seldom, if ever, been conducted for payment systems. One claimed economy of scope, between maintaining a system of settlement accounts and operating the ‘backbone’ system for making large-value transfers between those accounts, has sufficient intuitive plausibility to be widely accepted in the absence of such research. Despite this exception, it would be difficult to argue that those central banks that have taken on diffuse missions in their country’s payment systems on the basis of such arguments are actually serving the public well. One role for central banks that is conspicuously missing from the very short list enumerated here is regulation. There are at least three types of regulation to which payment systems might be made subject: prudential (or ‘safety and soundness’) regulation, regulation of competitive conduct, and consumer protection regulation. Central banks seem to have no comparative advantage in regulation of competition or consumer protection. Nevertheless, there is a recent tendency (exemplified in CPSS advice) to take on these responsibilities. This seems to me particularly ill-advised in the case of competition regulation. In terms of both cost structure on the producer side and the presence of network externalities on the demand side, payment systems closely resemble the telecommunications industry, on which national competition authorities expand very significant resources to maintain expertise that bears little or no relation to central banking. Given that another authority possesses such expertise, and that the central bank faces a dilemma of either regulating incompetently or else duplicating public investment in expertise, the wisest course would be to let the competition authority regulate payment systems in that regard. Prudential regulation is the type of regulation that people intuitively feel most

54

E.J. Green

strongly should be in the province of the central bank. There is a view that is widely held, even in countries where a specialised agency regulates and examines individual banks, that the central bank should regulate payment systems. This view is recommended on the grounds that the payment system is closely connected with the ‘systemic contagion’ of bank failures and other financial shocks, and the authority to regulate this transmission channel ought to go along with the central bank’s responsibility to deal with widespread financial troubles when they occur. That rationale is not an assertion that there is an economy of scope in the precise sense of economic theory, however. If the payment system is regulated by a separate agency, and the central bank believes that new regulations or a different regulatory stance would be conducive to financial stability, then the central bank can make its case to the agency and can use its political capital, if necessary, to exert pressure for change. A consideration against having the central bank regulate payment systems directly, on the other hand, is that the arrangement puts one payment system participant in the position of regulating others. There is a potential for conflict of interest. This potential is illustrated well by the preference expressed by the CPSS (2001) for the use of central bank money rather than inside money as a settlement asset. If we consider this issue in light of Freeman’s model of payments, a good settlement asset is one that can be emitted and reabsorbed (in a manner analogous to open market operations) by the entity with which the settlement accounts are held. The model depicts a central bank and a clearinghouse as being equally capable in this regard. Indeed, in the nineteenth century, US clearinghouses did emit and reabsorb inside settlement assets that they used for crisis management. In the context of this theoretical conclusion and of economic history, a regulatory presumption by central banks against creation by a clearinghouse of an inside settlement asset has the outward appearance of abuse of regulatory authority to disadvantage competition with the central bank from clearinghouses. If there is a cogent and creditable reason to discourage use of inside settlement assets, then a separate agency might be in a better position than the central bank to solicit public support for such a policy.

Conclusion A central bank does several things that are of immense value to the payment system, and it is better for the central bank to focus resources on doing those crucial things excellently than to dissipate resources by taking on additional tasks. Such a policy of deliberate focus is a difficult one for any organisation to adopt and maintain. Indeed, many central bankers probably believe that it would be difficult in practice to adopt such a policy. There are too many demands from the banking industry, the legislature, and various other stakeholders to embark on additional tasks, and those demands cannot be ignored. To rebut those sceptical views, I would point out that several OECD central banks actually do approach the high degree of focus recommended here. The closest of all is the Bank of Canada. In the early 1990s, the Bank and the Canadian banking indus-


55

try embarked on an initiative to consider deliberately what kind of payment system would serve Canada well. As this initiative progressed, the Bank of Canada concurrently revised its monetary-policy operating procedures to take full advantage of the improved payment environment. The new regime went into effect in 1999. Currently the Bank of Canada plays the three roles (maintaining settlement accounts, providing short-term credit and accepting illiquid collateral) enumerated at the beginning of this chapter. The Bank neither owns nor operates either Canada’s Large Value Transfer System (LVTS) or other retail components of the Canadian payment system. The Bank does serve as the final guarantor of LVTS obligations, a role that is considered not to create a material problem of moral hazard because the LVTS is well capitalised. The Bank regulates several payment systems that are deemed to be systemically important, but shares regulatory authority with Canada’s Department of Finance. One might argue that Canada is more fortunately situated than some other countries with respect to its ability to adopt such a highly focused stance. However, anyone making this argument should bear the burden of proof for it, and that burden is especially heavy because several other, highly regarded and successful, central banks have also chosen payment system roles that approximate the Bank of Canada’s focused role. In summary, a central bank can best contribute to the payment system by maintaining a system of accounts for interbank settlement, denominated in a money of stable value, and by offering short-term credit on illiquid collateral to the holders of those accounts. These services are of immense value to the payment system, and a central bank has a comparative advantage over most other institutions in providing them. In a developed economy where other institutions are able to provide services complementary to these, the social consequence of broader central bank involvement in the payment system is likely to be a marginal gain, at best.

References Bech, M.L. and Garrett, R. (2003) ‘The intraday liquidity management game’, Journal of Economic Theory, 109, pages 198–209. Cavalcanti, R. and Wallace, N. (1999) ‘Inside and outside money as alternative media of exchange’, Journal of Money, Credit and Banking, 31, pages 443–57. Committee on Payment and Settlement Systems (CPSS) (2001) ‘Core principles for systemically important payment systems’, Bank for International Settlements. Federal Reserve Bank of Chicago. (2000) ‘Comments regarding implementation of the core principles for systemically important payment systems’, www.chicagofed. org/bankwide_public_policy/files/\FinalStaffComCPVI09080.pdf (accessed 26 January 2006). Freeman, S. (1996a) ‘The payments system, liquidity, and rediscounting’, American Economic Review, 86, pages 1126–38. Freeman, S. (1996b) ‘Clearinghouse banks and banknote over-issue’, Journal of Monetary Economics, 38, pages 101–15. Green, E.J. (2006) ‘Some challenges for research in payments’, this volume.

56

E.J. Green

Green, E.J. and Todd, R.M. (2001) ‘Thoughts on the Fed’s role in the payment system’, Federal Reserve Bank of Minneapolis Quarterly Review, pages 12–27. King, M.A. (1999) ‘Challenges for monetary policy: new and old’, Bank of England Quarterly Bulletin, 39, pages 397–415. McAndrews, J. and Rajan, S. (2000) ‘The timing and funding of Fedwire funds transfers’, Federal Reserve Bank of New York Economic Policy Review, pages 17–32. Smith, B. (1988) ‘Legal restrictions, “sunspots” and Peel’s Bank Act: the real bills doctrine versus the quantity theory reconsidered’, Journal of Political Economy, 96, pages 3–19. Wallace, N. (2005) ‘From private banking to central banking: ingredients of a welfare analysis’, International Economic Review, 46, pages 619–31. Weinberg, J. (1997) ‘The organization of private payment networks’, Federal Reserve Bank of Richmond Economic Quarterly, 83, pages 25–44. Winston, C. (1993) ‘Economic deregulation: days of reckoning for microeconomists’, Journal of Economic Literature, 31, pages 1263–89. Woodford, M. (2003) ‘Interest and prices: foundations of a theory of monetary policy’, Princeton, NJ: Princeton University Press. Zhou, R. (2000) ‘Understanding intraday credit in large-value payment systems’, Federal Reserve Bank of Chicago Economic Perspectives, 24, pages 29–44.

3

Some challenges for research in payments Edward J. Green

In this chapter I discuss four directions in payment research that provide particular challenges in both pure and applied economics, chosen from among the many important topics in this active field. They are: 1 2 3 4

Formulating better basic models. Making market-microstructure data publicly available. Providing sound advice about payment systems risk. Understanding the relationship between payments and other business processes.

Challenge 1: to formulate better basic models As a working definition, suppose that payment economics comprises the topics that pertain to both monetary economics and industrial organisation. Monetary economics is the study of economic environments where: • • •

•

limited trading opportunities do not exhaust economy-wide gains to trade; institutions such as monetary and banking regimes can link these opportunities to enlarge the feasible gains; centralised, command-and-control institutions or all-encompassing contingent contracts that would provide prior resolution of all decisions are infeasible; and good outcomes require trust based on self-fulfilling expectations of favourable institutional performance.

Industrial organisation is an approach to economic analysis that recognises that: • •

economic agents are strategic players, not passive price takers; and economic activity involves increasing returns and externalities.

These definitions help define what a good economic model of payments should contain. One should begin by specifying an environment in terms of the agents who populate it, their technological opportunities, their preferences, the information that

58

E.J. Green

they possess, the protocols for communication among them, and so forth. From this specification, it should be clear what are the institutions that would be feasible to operate in this community. Then an equilibrium concept – specifically, one that recognises agents’ strategic incentives – should be set forth, as well as a welfare criterion that ranks the allocations that potential equilibria would implement. The equilibria of various institutional frameworks can then be studied, and recommendations about optimal institutions can be made in a way that is forthright about the combination of analytic and normative assumptions on which they rest. If these are the foundations for models in payment economics, then what facts should be explained according to it? The most basic fact is that virtually all trade utilises one of two institutions that coexist in the economy. One is the transfer tokens of stored value. Historically, these tokens have generally been coins or pieces of paper currency, either publicly or privately issued. Recently, there are also electronic implementations of stored-value transfer such as ‘smart cards’. The other institutional framework for trading is the recording of a pair of offsetting ledger entries in accounts of the parties to the transaction on the books of intermediaries, supported by another pair of offsetting entries in intermediaries’ accounts at a higher level intermediary such as a correspondent bank or central bank if the transactors have accounts at two different intermediaries. There was no economic theory, or model, of payments that could be regarded as even a serious attempt to explain this fact until about 15 years ago. Today there are two such models. One is an overlapping-generations model with settlement frictions, due to Freeman (1996a, 1996b). A second is a model of bilateral trade that formalises the neoclassical idea of ‘lack of double coincidence of wants’, the prototype of which was developed by Kiyotaki and Wright (1989), that Cavalcanti et al. (1999), Cavalcanti and Wallace (1999) and Wallace (2005) have augmented by a representation of financial intermediaries. Each of these models is now examined, to see how they fit the principles outlined above and also to suggest the respects in which they are not entirely successful. The overlapping-generations model with settlement frictions, which might be called the settlement-friction model for short, is a descendant of Sargent and Wallace’s (1982) model of the coexistence of transactions using outside money and those using debt. It adopts the modelling strategy of that paper in positing two types of trader within each generational cohort, one of which must acquire goods from the other at the beginning of life in order to fully exploit potential gains to trade, and also in positing a fixed sequence of trade meetings with limited participation at each date. The various traders must utilise these opportunities to accomplish their transactions. As this model is specified by Freeman (1996a, 1996b), an agent whose trade meeting to acquire first-period consumption from a contemporary comes before he has had a chance to meet someone who wants to purchase his endowment does not yet have money, so the only way to make his purchase is on credit. Fortunately, since both sides of this transaction are contemporaries, the two will

Some challenges for research in payments

59

have a subsequent trading meeting during the second (and final) period of their lives, at which the debt can be settled. All endowments are perishable and are received in the first period of life, so the repayment must be monetary. That is, a nominal asset must be used for settlement. That is not a problem, if the secondperiod meeting comes in time for the creditor to purchase consumption from young members of the next cohort later in the period. In that case, the traders who are creditors in equilibrium value money for the same reason as the traders in Samuelson’s classic model. They are also willing to accept debt issued by the traders who are debtors in equilibrium because it is safe debt that settles in outside money, and is therefore a perfect substitute for the outside money that they would alternatively have got by selling their endowments to members of the preceding cohort. The problem is that the second-period trade meetings between contemporaries are not so nicely timed for everyone. If debt could not be intermediated, then creditors would be unwilling to sell to debtors who could not make timely repayments. Or, if those specific debtors could not be identified in advance, then creditors would be unwilling to sell to debtors at all. However, if there are appropriately timed trade meetings among agents in their second period of life, then creditors who cannot receive timely repayment are able to sell their claims to others who can meet those debts later, and those others recover the purchase price of the claims by receiving payment from the debtors. That is, payment debt is intermediated by a subset of the creditors. What this structure accomplishes is to motivate three features of actual payment systems: 1

2 3

Some trade (between members of adjacent cohorts) has to be conducted using outside money, while other trade (between creditors and debtors in the same cohort) has to be conducted using trade credit. The trade credit has to be nominal – that is, settled in fiat money rather than in a commodity. In general, an intermediary must be active in order for all debt to be settled.

There will be inefficiency if an intermediary is active, but is unable to hold enough outside money to purchase at face value all of the debt offered for sale. The situation when this problem is severe, so that debt claims have to be sold at far below par, is one of a high interest rate prevailing in the intermediation market. There are two solutions that turn out to be equivalent in this model. One is to allow the intermediary to emit outside money but require him to reabsorb it by the end of the period. This is tantamount to making the intermediary a central bank that conducts open market operations. The other is to allow the intermediary to issue debt of his own in return for the debt claims that he acquires from other agents for settlement. The intermediary must settle these debt claims later, of course. This debt issued by the intermediary closely resembles the securities that clearinghouses issue to members in return for their claims on the members’ debtors. In summary, the model implies that:

60

E.J. Green

•

an interest rate on debt created in the payment system that is above the level warranted by the riskiness of that debt is inefficient; a central bank and a clearinghouse operate, although having different asset structures and payment flows among payment intermediaries, support identical real allocations.

•

The first implication of the settlement-friction model mentioned above – the coexistence of money and credit transactions – is just the basic fact that was identified earlier as a challenge for a theory of payments to explain. The second and third implications – that payment-system debt is nominal and requires intermediation – provide further, accurate details about this basic fact that were mentioned in framing the challenge. The fourth implication shows that the settlement-friction model has the sort of welfare implication that one expects from a good theory in economics: it identifies a significant efficiency issue, and it also provides an observable, potentially decidable criterion (the interest-rate level relative to the level of risk) for how well the payment system is working with respect to that issue. The criterion reflects a classical view that the main contribution of monetary policy to economic efficiency is to facilitate the operation of the payment system by regulating conditions in the interbank market for the short-term credit generated in the payment process. This is a cogent view, well aligned with what a central bank is able directly to affect. Certainly, central banks have focused their attention on this matter for weeks or months during periods of financial stress, such as 1987, 1998 and 2001 in the United States. If one were to emphasise the settlementfriction model very heavily as a model of the economic role of central banks, though, the upshot might be a view of optimal monetary policy that is more focused on this role, and correspondingly less focused on attempting to control broad real or nominal aggregates, than most central banks embrace today. The fifth implication of the model is a feature that is a hallmark of good scientific theories – the ability to explain facts beyond the handful for which it was explicitly designed. For, besides the coexistence of cash and trade-credit transactions, it is also a basic fact that central banks and clearinghouses have long coexisted in most of the world’s mature economies, with neither institution seeming to have so pronounced an advantage over the other from participants’ perspective to drive it out of business. Indeed, factual questions about coexistence of, and competition among, institutional forms, and normative questions about whether the formation and survival of efficient institutions is an outcome of competition, are challenging and important. The value of the settlementfriction model in this regard is further established by the research of Fujiki (2003) concerning institutional structures of foreign exchange settlement. The foregoing discussion establishes that the settlement-friction model is, on the whole, a scientifically successful economic theory, the policy implications of which should be taken seriously. Nevertheless it has three limitations in common with all overlapping-generations models. First, monetary equilibrium would not survive if the model were enriched and made more realistic by adding


61

a real asset with a positive rate of return. Second, while the model as typically formulated has a unique equilibrium, uniqueness is an artifact of the restrictive assumption that agents have zero endowments in the second period of their lives. If that assumption is relaxed, then the resulting model will have a continuum of equilibria with deterministic price paths, and will also be replete with ‘sunspot’ equilibria. Because an eminently reasonable generalisation of the model makes equilibrium indeterminate, welfare analysis and policy advice based on the special version of the model are fragile. (Note, however, that fragility takes the form that there is simply nothing to say if the special assumption is relaxed, rather than that a contradictory conclusion must be drawn if the assumption is relaxed.) Third, while the model provides qualitative insight about how the payment process works and why it is structured as it is, it does not seem to provide a helpful framework for the quantitative study of data. For those who think that empirical work should be organised around the estimation of explicit, coherent theoretical models, the unsuitability of this model for that purpose is a significant limitation. To further consider the difficulty of using the model as a framework for empirical research, recall that one of Samuelson’s motivations to formulate the original overlapping-generations model was to study social security systems. Recent advances in computing power have made it possible to formulate versions of the model – adapted in ways such as specifying 30 or 40 periods (interpreted as years) of working life and 20 periods or so of retirement, instead of one period of each – that do provide suitable frameworks for quantitative studies. An analogous strategy of building a larger-dimension version of the model and analysing it computationally probably cannot work for the settlement-friction model. The reason is that the specification of trading opportunities in the model, while judicious and fruitful for explanation, is ad hoc. Consider a bank’s operational problem of making payments, for instance. Settlements at different times are distinct services, and the resources for making those payments – such as staff with specialised competence in the bank’s operations centre – are production inputs with economies of scope in their application. It would be costly to bring in part-time ‘bubble staff’ for only an hour or two each day to handle the spike in demand for Fedwire (the United States Federal Reserve RTGS system) payments that precedes the closing time of CHIPS (clearinghouse owned/operated net settlement system for US dollar payments) that McAndrews and Rajan (2000) have documented, for example. Exogenously or endogenously, banks may be heterogeneous in their respective customers’ levels of intraday demand variability or in their ability to accommodate that variability. These are the sort of issues that make it difficult for banks to coordinate their settlements directly, and that therefore make it socially valuable for Fedwire to grant intraday credit to the banks that use it. To analyse convincingly a question like how the development of automated, ‘straight-through’ payment processing will affect the level of demand for Fedwire credit, one would need to derive the sequencing of the meeting opportunities envisioned in the settlement-friction model under both the manual and the automated

62

E.J. Green

technologies. Providing such a derivation is part of the challenge to formulate a more satisfactory payment model. The second noteworthy model of an environment in which use of outside money coexists in equilibrium with use of the services of payment intermediaries focuses on bilateral trade between pairs of agents who lack a double coincidence of wants, although there would be economy-wide gains to trade if everyone could trade simultaneously. A formal setting with this feature, and in which the agents are explicitly modelled as having a high degree of anonymity that makes multilateral trading agreements infeasible to negotiate or enforce, is the random-matching model introduced by Kiyotaki and Wright (1989). The population is a continuum of infinite-lived traders who maximise expected utility, who have preferences to consume one another’s endowments in a pattern that excludes double coincidence of wants, and who are randomly matched into trading pairs at each date. Kiyotaki and Wright assume that no trader can learn any other’s trading history, so an agreement to participate in a pattern of multilateral trades over time could not be enforced. They (and subsequent researchers who study the use of fiat money) show how coordination on the use of either commodity money or fiat money can partially compensate for traders’ anonymity. Payment intermediation has been introduced into a random-matching model by Cavalcanti et al. (1999), and also Cavalcanti and Wallace (1999) and Wallace (2005). Payment intermediaries are modelled as traders who possess the same, random, meeting of technology as other traders, but whose trading histories are publicly known. This publicity makes it possible for intermediaries to subject themselves to self-enforcing agreements to repay their own debts and also the debts of other intermediaries. In fact, the probability that an intermediary will have its own notes returned to it is zero. What is important is that each intermediary commits itself to provide valuable goods in return for the notes of others, and to limit its net emission of notes – that is, the excess of its cumulative note issuance over its cumulative acceptance of other intermediaries’ notes. Claims on intermediaries – broadly resembling private banknotes – thus become acceptable to other traders. In fact, they become a circulating medium of exchange that is accepted by one non-intermediate trader from another, as well as when offered directly by an intermediary. Such inside money complements the outside money in the economy so that a higher level of welfare is reached. Let us compare this sort of random-matching model of payments with the settlement-friction model. Both models posit constraints on traders’ ability to deal directly with one another at mutually convenient times, and both succeed in explaining the coexistence of transactions made directly with outside money and those made via financial intermediaries that issue inside-money debt claims. The settlement-friction model has a richer set of implications, but the additional implications are fruits of a style of modelling that allows assumptions to be made ad hoc to produce an equilibrium with preconceived characteristics. In contrast, the random-matching model is highly stylised, but parsimonious. Traders’ anonymity and the lack of double coincidence of wants (or the more


63

general idea that immediate gains to trade in small coalitions do not exhaust the gains that economy-wide trade would afford) are modelled in a conceptually straightforward way. This is also a model in which the acceptability of a medium of payment is significantly a matter of social convention, so that a security that everyone agrees to treat as money can be useful as such, despite being intrinsically less appropriate (for example, lower in rate of growth) than another asset.

Challenge 2: to make market-microstructure data publicly available A second challenge for payments research is to gather, and make available to the research community a body of statistical data from which models of payments could be assessed and estimated. From the preceding discussion, it should be clear that data about what finance researchers call ‘market microstructure’ is particularly important. Consider large value payments among banks and other financial intermediaries, for example. What is the pattern of these payments (including value, volume and concentration of payors and payees) through the course of a typical day? How is that pattern affected by various special circumstances, ranging from ‘triple witching days’ in markets for cash-settled derivatives to episodes of malfunction of critical payment-technology infrastructure? Having this sort of information is basic to understanding how large value payment systems work and to proposing and evaluating policy towards them. Obviously this is a shared challenge for researchers, the payments industry, and especially central banks. There is the usual problem of assembling and maintaining very large data sets: finding someone to do it and to finance it. There are also issues such as data confidentiality that are somewhat specific, although by no means unique, to payments data. Owners of other data that present such problems have shown determination and creativity in making it available to researchers. For example, the US Census Bureau operates a number of facilities around the country where researchers can submit their computer programs to be run by Bureau staff. Since researchers are almost exclusively interested in the overall statistical measures that are computed from the data in this way, rather than in the names of individuals whose privacy the delegated-computation arrangement safeguards. This way of providing access is very successful. Another strategy that might be considered for the types of data that I have mentioned would be to create synthetic data sets, using a computer program for random number generation, that would have the same statistical moments or other relevant statistical features as the actual data, but that would not contain actual, individual transactions. If the statistical verisimilitude of the synthetic data set were credibly vouched for, then journal editors and other pivotal members of the scientific community would be disposed to accept its use as a proxy for the actual data. However, researchers’ access to payment data might be provided, the returns to providing it would be large. Today, lack of data access is a factor that

64

E.J. Green

discourages empirical researchers from working on payment-system issues. And that fact tends to discourage economic theorists, in turn, from investing the extra effort to turn insightful, schematic theories into estimable and testable ones. The costs of data not being publicly accessible is a silent one of researchers, who look for good problems to address and simply find problems in other areas, and perhaps not even being aware that they would have chosen payment topics if they had had the opportunity. Nevertheless, the cost to payment economics is substantial. A broad-based, scientific community provides rapid and relatively straightforward growth of knowledge, and correction of mistaken belief. Without a good infrastructure of data conveniently available to independent researchers, analogous to the published national income statistics that researchers on other issues of interest to central banks take for granted, the scale of research in payment economics is limited and the benefits of more intensive research are missed.

Challenge 3: to provide sound advice about payment risk A third challenge is to understand better the role of payment-system credit, and correspondingly to be able to provide better founded advice about the risk management issues that this credit raises. Here is the issue: there is broad consensus (although it is not quite unanimous) among policy makers that an RTGS system is the gold standard for a systemically important large value payment system. There is also consensus, supported by some recent contributions in paymenteconomics theory (such as Freeman and Hernandez-Verme, 2004), that an RTGS system that does not provide credit to its participants can be at least as risky as a net settlement system. Essentially, what is done in virtually all RTGS systems operated by central banks is exactly what the settlement-friction model recommends. The central bank effects RTGS by becoming the central counterparty to all participants, providing intraday bridge loans for the difference between the payments that they would have to pay in an RTGS system without credit and the smaller payments that they would have to make in a net settlement system. As a result, at the peak of its exposure during the day, the central bank holds a huge portfolio of short-term loans that are exceptionally safe on the whole, but never absolutely safe. A first question is, how risky is this loan portfolio? To answer this question from a public perspective is subtle. Simply to say, for example, that the central bank has never lost a penny on its short-term credit to the payment system. But that may not be relevant if the RTGS design creates incentives for participants to become more exposed to losses than they would otherwise be, and if there are other creditors of those participants who would be junior to the central bank in an insolvency. If so, then would that situation be an externality that the RTGS operator is unwisely imposing and that those other creditors cannot control? That is the sort of assumption that Lagunoff and Schreft (2001) make in their modelling of systemic risk, for example. A polar assumption, analogous to a strong-form efficient market hypothesis in asset-pricing research, would be that


65

junior counterparties have reasonably current and accurate knowledge of how an RTGS participant uses intraday credit from the operator, and that those counterparties price their own credit extensions to the participant. The participant thus has the right incentive to economise on RTGS credit, since the operational cost of doing so will result in savings from more favourable pricing of other credit that it receives. How would researchers distinguish between these hypotheses? A threshold question is, are RTGS participants heterogeneous in their levels and patterns of use of intraday credit? With the sort of market-microdata for payments that were described above, this question would be answerable. If the answer is affirmative, as seems likely, then it could be asked whether participants’ strategies of RTGS credit use are correlated with, for example, the terms that they receive in the overnight money market (see Furfine, 2000). Even if it is determined that RTGS credit does not impose an uncontrollable externality on participants’ counterparties, there is the second question of how an RTGS operator should manage the credit risk that it takes on directly. One, decision-theoretic, aspect of this issue is how the very small risk of an extremely large loss to the operator ought to be quantified. Once it is quantified, one approach to managing it is to price it, as is done by the US Federal reserve. An alternate approach, taken by numerous other central banks, is to require collateral. Participants tend to view collateral as being costly, since the assets eligible to be used as collateral have lower rates of return than other investments. Thinking about collateral in terms of the Diamond–Dybvig theorem (along lines suggested by Wallace (1996)) suggests that there is also a social cost if RTGS collateral requirements force the banking system to hold a portfolio more heavily weighted towards low-return assets than would otherwise be necessary. An estimate of the magnitude of this cost would be useful. Moreover, if one thinks of central banks – the RTGS operators – as having much less need for liquidity than other market participants because they can create it themselves, then there is a welfare-economic question of whether they should accept collateral that is less liquid than other secured lenders would accept. Finally, if so, there are financial-economics questions of how such collateral should be valued, and there are other questions – partly in the domain of law and economics – about how it should be managed. A third aspect of the challenge regarding the welfare economics of RTGS credit is the issue of whether or not such credit, when offered cheaply or for free, is a subsidy. A traditional view is that payment credit is indeed a subsidy if it is priced below the intertemporal rate of substitution or the intertemporal marginal rate of return of agents in the economy. The settlement-friction model suggests a contrasting view – that payment credit is a unique economic institution that, despite a superficial resemblance to investment credit, has a completely different rationale. In fact, Zhou (2000) has analysed a version of the model implying that (abstracting from risk) payment credit should be priced at zero, even when there is a positive rate of inflation. In the settlement-friction model in its usual form and as she uses it, however, there is no arbitrage opportunity for the diversion of

66

E.J. Green

intraday credit. In the actual economy, there is a possibility that an RTGS participant with very low transaction costs might be able to profit from doing something like using RTGS credit every morning to purchase government securities, and then selling the securities at the end of the day to fund the RTGS payment from the proceeds. If there were a tendency for government securities to be priced higher at the end of the day than at the beginning, then this daylighttrading strategy might be profitable on average over time for a participant that is large enough to take huge positions. Whether or not there actually does exist any such arbitrage possibility is worthy of study.

Challenge 4: to understand the relationship between payments and other business processes A fourth challenge is substantially different than the others, which have all been towards the monetary-economics end of the payment-economics spectrum. This last challenge is better to integrate models of payments and related business processes. Much emphasis has been placed in payment economics on understanding the benefits of net settlement and on the transaction costs of payment float. Yet, for moderate-size commercial payments as well as for household payments, the costs and benefits of these for a single transaction must be fairly minor. In contrast, US bankers and corporate treasurers tend to quote numbers in a range of something like $5–$20 as the all-in cost of making a payment when related costs, such as invoicing and reconcilement, are included. Besides these variable costs, there are substantial fixed costs of payment-related information technology – particularly of comprehensive enterprise-management software that structures how payments and other business processes are related. There are other sorts of complementarity as well, such as retailers’ use for marketing purposes of information generated by customer payments. Models that treat payment explicitly as a component of a more comprehensive transaction, and that incorporate these various considerations, will likely be essential for understanding what drives the choices that transactors and their intermediaries make among payment options.

References Cavalcanti, Ricardo and Neil Wallace (1999) ‘Inside and outside money as alternative media of exchange’, Journal of Money, Credit, and Banking, 31, pages 443–457. Cavalcanti, Ricardo, Andres Erosa and Ted Temzilides (1999) ‘Private money and reserve management in a random-matching model’, Journal of Political Economy, 107, pages 929–945. Freeman, Scott (1996a) ‘The payments system, liquidity, and rediscounting’, American Economic Review, 86, pages 1126–1138. Freeman, Scott (1996b) ‘Clearinghouse banks and banknote over-issue’, Journal of Monetary Economics, 38, pages 101–115. Freeman, Scott and Paula Hernandez-Verme (2004) ‘Default and fragility in the payments system’, unpublished thesis Texas A&M University.


67

Fujiki, Hiroshi (2003) ‘A model of the Federal Reserve Act under the international gold standard system’, Journal of Monetary Economics, 50, pages 1333–1350. Furfine, Craig (2000) ‘Interbank payments and the daily federal funds rate’, Journal of Monetary Economics, 46, pages 535–553. Kiyotaki, N. and R. Wright (1989) ‘On money on a medium of exchange’, Journal of Political Economy, 97, pages 927–954. Lagunoff, Roger and Stacey L. Schreft (2001) ‘A model of financial fragility’, Journal of Economic Theory, 99, pages 220–264. McAndrews, James and Samira Rajan (2000) ‘The timing and funding of Fedwire funds transfers’, Federal Reserve Bank of New York Economic Policy Review, pages 17–32. Sargent, Thomas and Neil Wallace (1982) ‘The real-bills doctrine versus the quantity theory: a reconsideration’, Journal of Political Economy, 90, pages 1212–1236. Wallace, Neil (1996) ‘Narrow banking meets the Diamond–Dybvig model’, Federal Reserve Bank of Minneapolis Quarterly Review, 20, pages 3–13. Wallace, Neil (2005) ‘From private banking to central banking: ingredients of a welfare analysis’, International Economic Review, 46, pages 619–631. Zhou, Ruilin (2000) ‘Understanding intraday credit in large-value payment systems,’ Federal Reserve Bank of Chicago Economic Perspectives, 24, pages 29–44.

4

Payment economics and the role of central banks Jeffrey Lacker

The role of payment economics A distinct and coherent field of payment economics appears to be emerging, and it deserves some attention, especially among central bank policymakers (Lacker and Weinberg, 2003). In this chapter, I will say a few words about payment economics, and then discuss the role of the central bank in the payment system and implications of that role for several current issues. At the core of payment economics are systems of exchange financed by private and/or public liabilities and the institutions that facilitate the clearing and settlement of these instruments. In other words, payment economics can be defined as the study of the mechanics of exchange. It is based on the core insight of monetary economics that the instruments that people use to pay for goods and services serve to communicate reliably (that is, in an incentive-compatible way) about the buyer’s past actions (Townsend, 1989; Kocherlakota, 1998). Payment economics extends banking theory to encompass the role of banks as private issuers of payment instruments, and reflects the observation that virtually all institutions usually thought of as banks are significantly involved in payments. Indeed, the defining feature of banks appears to be their issuance of payment instrument liabilities, as opposed to their role as balance sheet intermediaries between savers and borrowers. Banks, from this perspective, are specialized institutions for facilitating the transmission and recording of relevant payment information. The industrial organization of the banking system therefore affects the characteristics of the monetary system. Payment economics thus lies at the intersection of monetary and banking economics with industrial organization. The fact that payment instruments and specialized institutions are at the core of the economics of payment arrangements has important methodological implications. It means that the choice of payment instruments, and the structure of the institutional arrangements that support them, should be viewed as endogenous. This defines an approach known as mechanism design – the cornerstone of modern monetary theory. Under this approach, payment instruments are seen as messages that embody contingent contracts, and one can model the information and risk allocation characteristics of alternative payment arrangements in a way that takes into account the limitations imposed by real world payment technologies

Payment economics and the role of central banks

69

– for example, the costliness and falsifiability of communication, verification, and authentication. For example, the mid-1980s presumption that paper cheques were socially inefficient because of their higher processing costs ignored other, apparently consumer-relevant, characteristics of cheques. The central role of communication in payments instruments and institutions has implications for the organization of payments activities. Communication technologies invariably are characterized by such features as economies of scale, common costs, and joint production. These features often take the form of network effects, in which much of the benefits and costs of network activities are shared among multiple participants. Private organizations that deal effectively with such characteristics can be described as clubs, in which terms of membership are just as important as the pricing and terms of service provision in inducing efficient participation. There is a tradition in industrial organization of questioning the extent to which competition ensures efficient performance in markets with these characteristics. This hinges on the extent to which markets are contestable, as has been emphasized by Ed Green and Dick Todd in their essay for the Minneapolis Fed’s 2000 Annual Report (Green and Todd, 2001).

The role of central banks in payments I would like to sketch out a tentative view that seems consistent with the emerging lessons of payment economics. It is not the only possible view one could take, but it strikes me as compelling. Until a better one comes along, I view it as a logical benchmark model. This view is built on two core ideas. First, central banks have more or less nationalized the clearinghouses at the ‘apex’ of the payment system. One can debate whether this was efficiency enhancing, as Goodhart (1988) argues, or whether it arose instead to re-allocate the costs and benefits of clearing and settlement activities. Clearinghouse activities appear to have aspects of club goods, as I noted earlier, and for club goods there is often a range of allocations consistent with efficiency – that is, with Pareto optimality. Central bank intervention sometimes alters the distribution of net benefits among payment system participants. For example, the Fed’s entry into cheque clearing seems to have been less about efficiency improvements than it was about shifting the costs of clearing cheques drawn on country banks. In any event, legal restrictions nowadays more or less compel many banks to settle at least some transactions through the transfer of central bank account balances. The second core idea is that many, if not most, of the private sector institutions that are the major players in the payment systems benefit from a substantial public sector safety net. In many cases explicit deposit insurance provides the most visible government support. But in addition, significant support is provided in conjunction with central bank payment operations. Central banks generally supply credit, both intraday and overnight, to key payment system participants. (The Swiss, until recently, were notable exceptions.) Moreover, there is a widespread perception among private payment system participants that central bank

70

J. Lacker

credit will be made available, perhaps even overnight, to facilitate the resolution of operational problems or other settlement disruptions. As Marvin Goodfriend and I have emphasized in a joint paper, this constitutes a backstop line of credit provided by the central bank (Goodfriend and Lacker, 1999). Indeed, operational protocols and the routine provision of daylight credit in some cases leave the central bank with no other choice but to lend. For example, in the case of the disruption at the Bank of New York in November 1985, the extension of overnight credit was a fait accompli (Lacker, 2004).

Moral hazard Taking the terms on which central banks clear, settle, and lend as given, the usual presumption is that competitive pressures will drive private sector institutions towards second-best efficiency. Underpriced access to central bank credit will of course distort private sector choices. Absent countermeasures, banks will take excessive risks and central bank credit will be overused, a distortion often referred to as moral hazard. It is in the nature of lines of credit, however, that they are underpriced at the point in time at which they are utilized. Credit lines provide guaranteed access to funds at a prespecified rate that does not vary with the borrower’s ex post creditworthiness. Thus borrowers essentially obtain insurance against adverse shocks to their creditworthiness. Private line of credit lenders are generally compensated for this insurance provision through up-front fees. Other features of typical credit lines act to constrain moral hazard. Lenders limit the extent of their liability through loan covenants that let them deny credit if certain financial conditions are not satisfied. In addition, lenders generally monitor borrower financial conditions on a regular basis, and often reserve the right to audit borrowers. The potential for moral hazard due to a public sector safety net, and in particular the provision of central bank credit in connection with payment operations, is to my mind the central rationale for central bank oversight of payment system participants. Such oversight should be aimed at measuring and efficiently constraining private risk taking that could affect the extension of central bank credit or the provision of public sector support. Much central bank payment system supervisory activity obviously fits this description well. Having said that, it is my sense that central banks have not come close to offsetting fully the safety net’s moral hazard distortion, although I would be hard pressed to document that claim, except to note the extent to which access to central bank settlement seems to be highly prized by financial institutions. This description of central bank payment activities implies a minimal service provision role – basically just offering clearing accounts that are used to settle interbank obligations. And this role is a byproduct of having de facto monopolized interbank settlement. In this, I find Green and Todd (2001) persuasive when they argue that the rationale for more extensive central bank service provision depends on the extent to which there are economies of scope between additional activities and the basic clearing account function. A focus on payment

Payment economics and the role of central banks

71

systems as communications mechanisms suggests the importance for this question of the relative effectiveness of alternative configurations of communications architectures, and potential economies in verifying messages and safeguarding information. My sense, however, is that there are far less by way of economies of scope than would be needed to justify, on economic efficiency grounds, the current scale of Federal Reserve service provision, particularly in clearing ‘retail’ payments such as cheques and automated clearing houses (ACHs). In fact, I have argued elsewhere that the evidence suggests that the Fed’s role in clearing retail payments rests on altering the allocation of clearing costs that would result from purely private provision. The imminent transition away from paper cheque clearing makes the Green and Todd question a live issue right now in the United States. Notice that I have made no use of the notion of ‘market failure’. My own working hypothesis is that market failures are largely absent from the payment system. After all, participants in any given payment arrangement are all linked by voluntary contractual relationships. Thus I find it hard to see how an externality, in the classic sense, could possibly arise. (The only genuine payment system externality I know of occurred when the Federal Reserve incinerated worn paper currency, thus polluting the air.) Note that the lack of an observed market does not mean market failure. For example, large banks do not clear cheques for rural banks in the United States. Surely this is due to the terms on which the Fed provides the same service. After all, there was an active market before the Fed did it. But as I argued earlier, we do not need a market failure to motivate central bank supervision of private payment system activities. To me, central banks’ policy interest is amply motivated by the presence of a substantial public sector safety net to payment system participants, and the central bank’s role in providing and setting the terms and conditions of important elements of that safety net.

Questions about the role of central banks in payments The perspective I have just outlined implies that the terms and operational conditions on which central banks extend daylight and overnight credit are of central importance. Years ago, when many aspects of current arrangements were put in place, operational considerations made it costly to implement systems that did not automatically extend daylight central bank credit in one form or another. New technologies may have significantly altered this cost–benefit trade-off, and in my opinion a re-examination of daylight credit policies is in order. For example, many banks monitor and control the extension of daylight credit to many of their corporate customers, and indeed, supervisors expect them to. It would be ironic for central bank risk management to lag behind private sector practices in this regard. A focus on central bank credit also makes clear that paying interest on reserves is more important than is commonly appreciated. The prohibition of interest on central bank deposits, as in the United States, greatly enhances the demand for daylight credit, in the sense that larger overnight balances act as

72

J. Lacker

substitutes for daylight overdrafts. As a result, limitations on central bank credit extension would be less costly if reserves earned interest. More broadly, it seems plausible that a huge fraction of settlement activity originates in transactions whose main purpose is to allow entities to evade interest prohibitions, and thus to some extent are socially wasteful. The relationship between central bank credit and the broader public safety net has implications that are sometimes overlooked. For example, the collateralization of central bank credit extension may reduce risks to the central bank, but it can increase risk to the deposit insurance fund. Therefore, the central bank ought to consider more than just its own balance sheet risk in making lending decisions. This is especially important because, as the lender of last resort, the central bank can often force an institution’s closure by refusing credit. Notwithstanding several seemingly strong policy pronouncements, I do not believe that our understanding of the economics of intraday credit is at this point sufficient to provide quantitative guidance on the optimal pricing of daylight credit, even apart from moral hazard considerations. In that light, volumes such as those devoted to the advance of payment economics are among the most noble uses of central bank resources.

References Goodfriend, M. and Lacker, J.M. (1999) ‘Limited Commitment and Central Bank Lending’, Federal Reserve Bank of Richmond Economic Quarterly, 85, 4, pages 1–27. Goodhart, C.A.E. (1988) The Evolution of Central Banks, Cambridge: The MIT Press. Green, E.J. and Todd, R.M. (2001) ‘Thoughts on the Fed’s Role in the Payments System’, Federal Reserve Bank of Minneapolis Quarterly Review, 25, pages 12–27. Kocherlakota, N.R. (1998) ‘Money is Memory’, Journal of Economic Theory, 81, pages 232–251. Lacker, J.M. (2004) ‘Payment System Disruptions and the Federal Reserve After September 11, 2001’, Journal of Monetary Economics, 5, 1, pages 935–965. Lacker, J.M. and Weinberg, J.A. (2003) ‘Payment Economics: Studying the Mechanics of Exchange’, Journal of Monetary Economics, 50, pages 381–387. Townsend, R.M. (1989) ‘Currency and Credit in a Private Information Economy’, Journal of Political Economy, 97, pages 1323–1344.

Part II

New approaches to modelling payments

5

New models of old (?) payment questions Ricardo Cavalcanti and Neil Wallace1

Introduction In both the United States and the United Kingdom, a monopoly on ‘currency’ issue grew out of a system in which there were many issuers of banknotes. In the United Kingdom, that monopoly was created in 1844, and was accompanied by a 100 per cent specie marginal reserve requirement against banknote issue. The 1844 law, Peel’s Act, was a victory for the currency school, whose members advocated some version of hard money, or what much later came to be called monetarism. The 1844 law was opposed by members of the banking school: those who advocated some versions of laissez-faire in intermediation. Among the questions alluded to in the debates were: Was the private note-issuing system accomplishing anything? If it was, then would it be desirable to have the Bank of England manage its monopoly so as to emulate what the private note system was accomplishing? In this chapter, we revisit those questions and do so for at least three reasons. First, one test of progress in monetary theory is its ability to provide new insights about old questions that have never been satisfactorily resolved. Second, those old questions have modern analogues: should central banks operate lending facilities and, if so, how? Should stored value, and other modern analogues of private note-issue, be regulated and, if so, how? Third, the modelling ideas that throw light on those questions have implications for seemingly unrelated questions: for example, how best to model cashless economies. Why do we assert that the nineteenth-century debates were never satisfactorily resolved? At the beginning of the twentieth century, the dominant monetary theory consisted of the classical dichotomy. While that theory could accommodate private credit instruments that to some extent substitute for outside or base money, either by treating such substitutes as part of the stock of a broader concept of money or by treating them as increasing the velocity of outside or base money, neither treatment could say anything about the welfare consequences of different monetary systems or, for that matter, the welfare consequences of money. At the beginning of the twenty-first century, the dominant monetary theory consists of descendents of the classical dichotomy: models with real balances in utility or production functions or models with cash-in-advance constraints. These descendents were designed to overcome the blatant inconsistencies of the classical dichotomy:

76

R. Cavalcanti and N. Wallace

the kind of inconsistency that Patinkin (1951) pointed out. They were not designed to and cannot address the questions raised in the nineteenth-century debates any better than could the classical dichotomy. In this chapter, we set out some ideas about how such questions might be approached. Our goal is to convince readers that the ideas are fruitful: both for the nineteenth-century questions about good monetary systems and for other questions concerning monetary systems. However, one warning is in order; we have essentially no results about the implications of the modelling ideas we set out.

Some general ideas One challenge to any model of money is: why is trade being modelled using money when trade could conceivably be accomplished with some version of borrowing and lending between people? There is, by now, a well-known answer. Individuals cannot commit to future actions and to some extent their histories are not known. This answer goes back at least to Ostroy (1973). (See also Townsend (1989) and Kocherlakota (1998).) Neither part is controversial. The inability to commit, although inconsistent with the Arrow–Debreu model, is a standard and plausible assumption of game theory. The problem of partially unknown histories is, in modern game theory, labelled imperfect monitoring. It means that previous actions of some people are not common knowledge. It is the assumption in moral-hazard models and is implicit in the idea that money is used in trade among strangers and the related idea that money is evidence of past actions that are otherwise imperfectly known. We like the answer, and, therefore, build a model that rests on it. Throughout we maintain the assumption that people cannot commit to future actions. As regards monitoring, we assume that some people are not monitored at all and others are perfectly monitored. The unmonitored people are the demanders of tangible media of exchange; the monitored people are the potential issuers of private money and, in most respects, are the focus of our discussion. The kind of private money we analyse is best thought of as payable-to-thebearer bills of exchange that have only the issuer’s name on them. The private money has this form, a form which bypasses banks as we ordinarily describe them, because this form more easily gets us to a model in which the welfare consequences of different systems can be analysed. (Something like this is done in the Diamond–Dybvig model (Diamond and Dybvig, 1983) of banking in which what is described as a banking system is best thought of as a mechanism in a model consisting only of consumers, who are the owners of a consolidated banking-business sector.) Throughout, we work against the background of a model in which each person, including any issuer of private money, is individually a small part of the total economy and in which for purposes of production and consumption people meet in pairs. In our model, people do not choose to meet in pairs and a pair need not be viewed as a natural production unit, as it is in models of marriage and seems to be in many search models of labour. (In our setting, larger

New models of old (?) payment questions

77

production-consumption meetings, if they could occur, would enhance welfare.) In our model, one pairwise meeting per date for the purposes of production and consumption is free and any other kind of meeting for that purpose is infinitely costly. The pairwise meeting structure helps us in several respects; it is consistent with absence-of-double-coincidence difficulties, with imperfect monitoring (each person may know only what the person has seen in the meetings in which the person has been a participant), and with non-trivial float (although float will not play a significant role here). As in any model of private money, potential over-issue has to be prevented. In our model, it is prevented by threatened punishment. The punishment we use here, although rather mild, always includes the loss of the ability to issue valuable money. To make that feasible, we assume throughout that the private money issued by one person is potentially distinguishable from that issued by anyone else. This is a strong recognizability assumption. We suspect that weakening it, by permitting some sort of counterfeiting, would matter a lot.

The model The model is almost identical to that in Wallace (2005), which, in turn, builds closely on our previous work (Cavalcanti and Wallace, 1999a, 1999b). In particular, the sense in which there is imperfect monitoring is carried over from the specification in our earlier papers. A background specialization environment We use the familiar specialization setting of Shi (1995) and Trejos and Wright (1995). Time is discrete. There is a unit measure of each of K 3 specialization types of infinitely lived people and there are K distinct, produced, and perishable goods at each date. A specialization-type k person, k {1,2,...K}, produces only good k and consumes only good k + 1 (modulo K). Each person maximizes expected discounted utility with discount factor (0,1). For a specializationtype k person, utility in a period is u(qk+1) – qk, where qk+1 ℜ+ is consumption of good k + 1 and qk ℜ+ is production of good k. The function u : ℜ+ → ℜ is strictly concave, strictly increasing, differentiable, and satisfies u(0) = 0 and u'(∞) = 0. In addition, u'(0) is sufficiently large. A word is in order about the assumption that the number of people is uncountable. So far as we can see, this assumption plays only one role. It implies that a person’s action in a two-person meeting does not influence his or her future trading opportunities except by way of what happens to the person: not by way of what happens to the person’s trading partner. That should hold approximately for a sufficiently large finite number of people. In other words, we suspect that the outcomes we describe resemble those of the comparable model with a sufficiently large finite number of people, provided there is discounting that is held fixed as the number of people is allowed to get large.

78


Imperfect monitoring and the sequence of actions We make one other distinction among people. We assume that the set of each specialization type is partitioned in an exogenous way into two parts. Throughout, the fraction n, is not monitored at all. The history of each such person, except as may be revealed by the person’s money holdings, is private to the person. The rest, the fraction m = 1 – n (m for monitored) are perfectly monitored. That is, the history of each monitored person is common knowledge. It is as if each monitored person wears a computer chip that transmits actions of the person to everyone else. In this model, m represents the economy’s monitoring capacity. As part of not being monitored, each unmonitored person can hide money. We use the following sequence of actions in discrete time. At the start of each date, each person has a state consisting of the person’s type, history, and money holding. A person’s type, specialization type, and whether monitored or not, is assumed to be common knowledge and is permanent. Money holding, a scalar, is defined to be the sum of outside money plus private money acquired from others. As this suggests, we only consider allocations in which all valuable monies, all private monies and outside money, are perfect substitutes. (Richer allocations that distinguish among valuable monies, both subsets of private money issuers and between private money and outside money, could be considered.2) Then, there are pairwise meetings at random, during which there is production and consumption which gives payoffs according to the preferences and technologies described above. After those meetings conclude, monitored people simultaneously meet the planner and, to be consistent, are all together. However, by assumption, there is no production or consumption that goes on after the pairwise meetings. At best, there are transfers of money. (The planner can be thought of as a benevolent central bank running a discount window and having unlimited access to outside money, while trades among the monitored people after the pairwise meetings can be thought of as being something like a federal funds market.3) Weakly implementable allocations and welfare We will be doing a limited kind of mechanism design analysis. We start by defining a set of allocations. Then, we describe a simple coordination game in which people choose individually either to cooperate or defect. If everyone cooperates and unmonitored people, who can hide money, choose to truthfully self-select, then the allocation is weakly implementable; otherwise not. Our goal is to describe the best weakly implementable allocation, where best is defined below. An allocation describes what happens in pairwise meetings and what happens when monitored people meet the planner, all conditioned on the date and on the states of the people in the pairwise meeting and the state of each monitored person when meeting the planner. Then, given an initial condition in the form of


79

a distribution over money holdings and histories, such a description of what happens in meetings at each date is sufficient to describe the evolution of the economy. Given a suggested allocation, the following game is played. Consider a pairwise meeting. The allocation includes a suggested trade in the meeting. Both parties simultaneously choose between cooperate and defect. If both cooperate, then the suggested trade is carried out. If either says defect, then they leave the meeting without trading. If an unmonitored person defects, then there are no further consequences. The person goes on to the next date with what the person has. If a monitored person defects, then there are further consequences to be described momentarily. In the meeting with the planner, each monitored person again chooses between cooperate and defect. As regards the consequences of defection for a monitored person, we assume that the person can at any time join the ranks of the unmonitored people and suffer no additional punishment except that the person’s private money is no longer accepted. In describing the consequences of defections, we are explicitly ruling out punishment of a large segment of the economy in response to individual defections. For example, we are ruling out permanent autarky for the entire economy as a response to an individual defection. Notice that our defection scheme permits free exit from the set of monitored people. However, we do not permit free entry into that set. Definition 1: An allocation is weakly implementable if there is a sub-game perfect Nash equilibrium in which each person cooperates and each unmonitored person also self-selects the trade intended for people with the person’s actual holding. Two comments are in order about this definition. First, it only requires that there is some equilibrium that implements the allocation. Second, it permits only individual defections, not group defections. In particular, it does not permit cooperative defection by the pair in a meeting. There are several obvious consequences of the definition. It is weakly implementable to have any recognizable money be worthless and for the usual reason: if a person thinks that others will not accept an intrinsically useless object in the future, then the person will not accept it now. Thus, it is weakly implementable to have all private monies be worthless, to have outside money be worthless (it is important that we are assuming outside money to be uniform), and to have all money be worthless. In particular, allocations in which there is no valuable private money are special cases of more general allocations that include valuable private money. (According to our model, the Peel’s Act monetary system could arise without a law.) Therefore, in order not to dwell on the completely obvious, our focus will be on describing as carefully as we can what is sacrificed by not having valuable private money. The simplest welfare criterion for the model is an ex ante representativeagent criterion: one that treats people as identical before the assignment of types and states. In particular, according to this criterion, the probability of being in the monitored set is m and the probability of being in the unmonitored set is n.

80


Finally, because there is no capital in this model, the initial condition, the distribution of money holdings and histories, can be treated as something that the planner chooses along with the allocation. The ex ante welfare criterion can be expressed as the expected discounted value of the gains from trade over all single-coincidence pairwise meetings: gains from trade in the sense of the magnitude of g(q) u(q) – q. Obviously, maximum ex ante welfare is achieved by production and consumption equal to q* = arg maxq[u(q) – q] in every single-coincidence meeting. As we will see, the constraints on monitoring and punishments rule out that allocation. However, the maximum nicely summarizes the economic problem represented by the model; good arrangements will tend to weaken the tie between what happens in single-coincidence meetings and the individual histories of the participants in the meeting.

The role of inside (private) money: an example with {0,1} money holdings We present a simple example that shows that private money can actually play a role. Although we do this in the context of individual money holdings in the set {0,1}, the forces at work are general. We describe stationary and symmetric allocations that are weakly implementable with valuable private money, but that are not in its absence. Consider an allocation in which the same output level, some y (0,q*], is produced in all single-coincidence meetings except in two circumstances: when an unmonitored producer has money or when an unmonitored consumer does not have money. Suppose that nothing is produced in those single-coincidence meetings. (The exception for unmonitored producers is implied by the bound on money holdings and their participation constraint; that on unmonitored consumers is an arbitrary part of the allocation and will be discussed below.) Moreover, suppose that unmonitored consumers surrender money when they consume y and that unmonitored producers receive money when they produce y. In single-coincidence meetings between monitored and unmonitored people, the monitored consumer provides (newly issued private) money to the unmonitored producer and the monitored producer collects (outstanding private) money from the unmonitored consumer, which is then turned over to the planner. In meetings between monitored people, no money changes hands. Suppose further that half the unmonitored people start without money and half with money and that all monitored people start without money. The above trades imply that that distribution persists (is a steady state). In order to express the participation constraints, it is a helpful short-hand to compute the discounted values implied by this allocation. Let v n(z) be the discounted value for an unmonitored person at the beginning of a date with money holdings z {0,1}. These values satisfy K(1 – )v n(0) = (–y + )

(1)


81

and K(1 – )v n(1) = (u(y) – )

(2)

where = v n(1) – v n(0) and = m + n/2. (These linear equations have a unique solution that implies = u(y) + y, where = 2 + K(1 – )/.) And let v m be the discounted value for a monitored person at the beginning of a date without money. It satisfies K(1 – )v m = (u(y) – y)

(3)

We do not need to express the discounted value for a monitored person of starting a period with money, money issued by another monitored person, because (i) there are no such people in equilibrium, and (ii) a defection does not give rise to such a person. For incentive feasibility, there are three relevant constraints. One is the participation constraint for an unmonitored producer: v n(1) – y v n(0)

(4)

The others are two constraints for monitored people: v m – y v n(0) and v m v n(1)

(5)

The first is the participation constraint for a monitored producer (the pay-off for a monitored producer who defects is that of an unmonitored person without money because the defector’s printing press becomes worthless); and the second says that a monitored person is willing to surrender to the planner the money received in a trade. Because v m = v n(0) + v n(1) (see equations (1) to (3)), participation constraint (equation (4)) implies participation constraints (equation (5)). Next, we describe necessary conditions for duplicating the above consumption and production pattern without private money. In order to duplicate the pattern, each monitored person must begin a period with outside money. Otherwise, when a monitored person is a consumer in a meeting with an unmonitored producer without money, the producer cannot be induced to produce y. In the simpler set-up of our earlier paper (Cavalcanti and Wallace, 1999b), there was nothing like a discount window or a federal funds market, and the stock of money was constant. Hence, it was simply impossible to have the spending described in the allocation: the monitored people who spent money in the previous period would not have money at the start of the next date. Now, that argument does not apply because the planner could give money to those monitored people who spent money and could collect money from those who have acquired money. If that is done and the trades are as described by the allocation, then vm as given by equation (3) again describes the discounted value for any monitored person. In addition, the v n(z) are unaffected. However, the

82


constraints are now different. In place of the constraints on v m in equation (5), there is just one relevant constraint: v m – y v n(1)

(6)

The constraint says that there will be no defection when a monitored person with money is called on to produce y. The new constraint is tighter than the two it replaces and is not implied by equation (4). In fact, it is easy to describe magnitudes of y and the other parameters for which equation (4) holds, but for which equation (6) does not. For example, if y = > 0, as is implied if y is the outcome of a take-it-or-leave-it offer by an unmonitored consumer to an unmonitored producer, then v n(0) = 0 and v n(0) = v m. Therefore, equation (4) holds, but equation (6) does not. Hence, implementabilty can fail without private money. The greater temptation to defect when a transfer of outside money replaces private money issue does not seem to depend on the special assumption about money holdings. The result does, however, depend on two features of the model. One is the assumption that outside money is uniform. If each unit of outside money were unique, then a defection could render worthless the particular unit held in the same way as the person’s printing press is rendered worthless. And it depends on the uncertainty about spending. If future spending were known when the monitored person meets the planner, then without private money the planner’s transfer could be made just sufficient to support that spending. Although the above comparison is suggestive, it is not decisive even about this simple setting with the special {0,1} money holdings. The example does not establish that private money is necessary for an optimum, even among stationary allocations. Even with y = q*, the allocation described above does not maximize welfare. In that allocation, a monitored producer does not produce for an unmonitored consumer who has no money. But some production in such meetings – even if offset by lower production in other meetings in order to satisfy participation constraints – would almost certainly increase welfare because u is strictly concave. Given {0,1} money holdings, an upper bound on welfare is given by y = q* in all single-coincidence meetings except those in which the unmonitored producer has money and y = 0 in those meetings. However, it is immediate that any allocation with the same positive output in all meetings except those in which the unmonitored producer has money is not implementable. Given such an allocation, in a meeting with a monitored producer, an unmonitored consumer with money will envy the trade of an unmonitored consumer without money unless the former is not asked to turn over money. But, if not, then money never flows from the set of unmonitored people to the monitored, which, in turn, implies that money cannot flow the other way. But that contradicts the presumed spending of monitored consumers in meetings with unmonitored producers. This immediately tells us that the optimum will have some binding truth-telling or participa-


83

tion constraints. That, in turn, makes it challenging to describe optima even in the highly special case of money holdings in the set {0,1}.

The planner or a market as a potential substitute for private money The above discussion points to the potential gain from private money. Suppose, however, that we are stuck with only outside money. Is there a presumption that there is a role for an active planner? For this question, the simple case of money holdings in the set {0,1} is misleading. So let’s think about general money holdings. It should be evident that dispersion of money holdings is not a good thing in this model. In general, if the consumer has small money holdings, then it will be impossible to get an unmonitored producer to produce much for such a consumer. And, as we have seen in the example above, it will be difficult to get even monitored producers with large money holdings to produce much. Hence, it would seem desirable for the planner to transfer money from those monitored people with large holdings and to transfer money to those with small holdings. Of course, those who are asked to give up money have to be willing to do so because they have the option to defect.4 A scheme of such transfers is an insurance arrangement. One of the things sacrificed by a monitored person who defects is the right to continue in it. Another is participation as a monitored consumer in meetings with monitored producers; in such meetings, an optimal arrangement will tend to have output be less dependent on the consumer’s money holdings if the consumer is monitored than if the consumer is not monitored. It, too, is a kind of insurance. Obviously, the binding constraints for transfer schemes arise when taking money from monitored people. One way to avoid those constraints is to inflate. An extreme is to give only non-negative transfers to monitored people and to make them a decreasing function of the wealth of monitored people. That will shift purchasing power towards the monitored people with little money. Of course, that will also produce a falling value of money, which, itself, tends to have undesirable effects because it tightens participation constraints. It should be emphasized, by the way, as in our earlier papers, that inflation and deflation are not the only ways to produce non-zero returns on money in this model. Even in the simple case of {0,1} money holdings, there is no reason why output in meetings should not depend on the monitoring status of the participants. In particular, a positive average return on money for unmonitored people can be achieved by having a monitored consumer get less in a meeting with an unmonitored producer than does an unmonitored consumer in any single-coincidence meeting. Is a market among monitored people a perfect substitute for activity by a planner? This question seems particularly relevant in our model because there are no aggregate shocks in the model. In the model, the market would be one in which people are insured against

84


the kind of pairwise meetings they experience. Moreover, the market would have to be subject to participation constraints because individuals can defect. One possible specification is a special case of the formulation in Kehoe and Levine (1993), a competitive formulation in which each person faces a budget set of the usual sort and the person’s own participation constraints: constraints that are common knowledge. And, obviously, the market would be subject to a feasibility constraint on total money holdings, a constraint that the planner does not have to satisfy. (By the way, the scheme of transfers of outside money described above in the case of {0,1} money holdings could be accomplished by a market with the following trades: each monitored person who ends up after pairwise trade with two units of money willingly surrenders one unit in the market (the second unit would violate the bound) and each monitored person who ends up with zero acquires one unit.) One way to think of a market is as a constraint on what the planner can accomplish (Hammond, 1987). This view of a market seems to be the same as imposing the stronger requirement on allocations that a group does not want to defect to anything that is feasible for the group. A surmise is such, that under a more stringent notion of implementability, the only advantage of a planner over a market is the planner’s freedom to change the total amount of outside money.

Generalizations of the model Given that we have done little but pose questions of the simple model that we have set out, it seems gratuitous for us to suggest generalizations of the model. However, showing that the model lends itself in a straightforward way to various generalizations is part of its attractiveness. Imperfect monitoring is, of course, consistent with having people experience private-information shocks to preferences. One extreme version of such shocks was described in Cavalcanti and Wallace (1999a). There, we assumed that people receive at each date a private-information realization that determines whether or not they can produce at that date. The presence of such a shock has essentially no consequences for how we describe the unmonitored people because they cannot gain by misrepresenting their realization. For monitored people, in contrast, such shocks introduce into the model the kind of truth-telling constraints in Green (1987). One of the consequences is to make the planner’s dealings with monitored people dependent on individual histories. The model above has the simplest timing consistent with uncertainty about spending opportunities. Obviously, there are many alternatives that would retain that feature. And nothing in the model is inconsistent with aggregate shocks or with something like a deterministic seasonal. The imperfect monitoring we have assumed is very special. A troubling aspect of imperfect monitoring is that there are innumerable ways of specifying it. A lag in updating each person’s history is adopted in Kocherlakota and Wallace (1998). Such a lag is applied to the monitored people of the model above in Mills (2001). And, although they do not attempt a mechanism-design


85

analysis, implicit in Cavalcanti et al. (1999) is the assumption that the planner’s only information about issuers of inside money comes from the money that shows up in a clearing house run by the planner. In some respects, the crucial assumptions we have made are about recognizability. We have assumed that outside money is uniform, but that private monies can be distinguished according to the issuer. Missing from the model is the notion that uniformity of money is desirable.

Concluding remarks This volume is about the future of payments and the challenges that that future poses for central banks. We have focused on seemingly old questions: is private money useful? In the absence of private money, is there a role for a central bank discount window over and above what a federal funds market could accomplish? Our model hints at affirmative answers to both questions. Moreover, the model seems relevant for some new questions. Is management of central-bank, intra-day credit a new question or is it a version of the question about a role for a discount window as we have posed it? That depends in part on whether it is sensible to think of intra-day credit as being extended to perfectly monitored agents who have a demand for it because of their dealings with strangers. And what sort of model of a cashless economy should we focus on? Presumably, the relevant cashless economy should be a limit of a cash economy as cash becomes less important. Because we like the ideas we described at the outset that explain why cash rather than IOUs are used, we are inclined to use such a model as our model of a cash economy. But what sort of limit should we take? In such a model, we can get a cashless economy in one of two ways: we can let the ability of individuals to commit to future actions get perfect or we can let monitoring get perfect. To us, the choice is clear. We should let monitoring get perfect; after all, that is what improved information technology makes possible. This has an immediate implication: the limiting cashless economy is not an Arrow–Debreu economy. We have suggested some ideas about how to deal with a fundamental issue in monetary theory: the margin between money and credit. And we think that those ideas are fruitful both for old questions about monetary systems and for new ones related to the future of payment systems.

Notes 1 We are indebted to Stacey Schreft of the Federal Bank of Kansas City and to John Moore of the University of Edinburgh for helpful comments on an earlier draft. 2 Distinctions among the money issued by subsets of monitored people are discussed in Wallace (2003). 3 In a sense, excluding the unmonitored people from meeting the planner and others after pairwise meetings is without loss of generality. Because unmonitored people can hide money, the planner can at best give non-negative transfers to them that are weakly

86


increasing in their money holdings. And even that can be regarded as problematic. How does the planner prevent the same unmonitored person from showing up many times at a date for a transfer? 4 There is a literature on matching models that avoids heterogeneity of money holdings. One route is the so-called large family model (Shi, 1997). Another is the device introduced by Lagos and Wright (2005): quasi-linear preferences in a good that is traded in a centralized market. These models have two limitations. First, the assumptions that do away with the heterogeneity are special. Second, the possible role of policy in dealing with heterogeneity is lost.

References Cavalcanti, R. and Wallace, N. (1999a) ‘A model of private banknote issue’, Review of Economic Dynamics, 2: 104–36. Cavalcanti, R. and Wallace, N. (1999b) ‘Inside and outside money as alternative media of exchange’, Journal of Money Credit and Banking, 31(2): 443–57. Cavalcanti, R., Erosa, A., and Temzelides, T. (1999) ‘Private money and reserve management in a random matching model’, Journal of Political Economy, 107: 929–45. Diamond, D. and Dybvig, P. (1983) ‘Banks runs, deposit insurance, and liquidity’, Journal of Political Economy, 91: 401–19. Green, E.J. (1987) ‘Lending and the smoothing of uninsurable income’, in E.C. Prescott and N. Wallace (eds) Contractual Arrangements for Intertemporal Trade, Minneapolis, MN: University of Minnesota Press. Hammond, P. (1987) ‘Markets as constraints: multilateral incentive compatibility in continuum economies’, Review of Economic Studies, 54: 399–412. Kehoe, T. and Levine, D. (1993) ‘Debt-constrained asset markets’, Review of Economic Studies, 60: 865–88. Kocherlakota, N. (1998) ‘Money is memory’, Journal of Economic Theory, 81: 232–51. Kocherlakota, N. and Wallace, N. (1998) ‘Optimal allocations with incomplete recordkeeping and no commitment’, Journal of Economic Theory, 81: 272–89. Lagos, R. and Wright, R. (2005) ‘A unified framework for monetary theory and policy analysis’, Journal of Political Economy, 113: 463–84. Mills, D.C. (2001) Outside and Inside Money: A Mechanism Design Approach, PhD dissertation, Pennsylvania State University. Ostroy, J. (1973) ‘The informational efficiency of monetary exchange’, American Economic Review, 63: 597–610. Patinkin, D. (1951) ‘The invalidity of classical monetary theory’, Econometrica, 19: 134–51. Shi, S. (1995) ‘Money and prices: a model of search and bargaining’, Journal of Economic Theory, 67: 467–98. Shi, S. (1997) ‘A divisible search model of money’, Econometrica, 65: 75–102. Trejos, A. and Wright, R. (1995) ‘Search, bargaining, money and prices’, Journal of Political Economy, 103: 118–41. Townsend, R. (1989) ‘Currency and credit in a private information economy’, Journal of Political Economy, 97: 1323–44. Wallace, N. (2003) ‘Commentary’ in D. Altig and B. Smith (eds) Evolution and Procedures in Central Banking, Cambridge, England: Cambridge University Press. Wallace, N. (2005) ‘From private banking to central banking: ingredients of a welfare analysis’, International Economic Review, 46: 619–36.

6

Optimal settlement rules for payment systems Benjamin Lester, Stephen Millard and Matthew Willison1

Introduction Payments are transfers of value between agents. For all payments that are not made in cash, finalization of payment occurs separately to the exchange of goods and will involve a payment system: a specification for when and how the actual funds are delivered consisting of a settlement asset, credit arrangements, infrastructure and rules. Indeed, Zhou (2000) defines a payment system as a ‘contractual and operational arrangement that banks and other financial institutions use to transfer funds to each other’. Such systems support a vast amount of economic activity. For example, on an average day in 2005 CHAPS Sterling, the United Kingdom’s largevalue payment system, processed about 120,000 transactions with a total value of around £210 billion, about 20 per cent of the United Kingdom’s annual gross domestic product.2 Given this, problems in a payment system could affect the functioning of the financial system and in turn the wider economy. As part of their role in ensuring the stability of their financial systems, central banks ‘oversee’ a number of payment systems with the goal of assessing and, if necessary, reducing the amount of risk that they bring to the financial system.3 Historically, interbank payments have been settled via end-of-day deferred net settlement (DNS) systems. As the volume and value of interbank payments passing through such systems increased rapidly in the 1980s and 1990s, central banks became increasingly concerned about the risk that stemmed from such systems. In particular, where payments are credited to customer accounts before being finally settled, credit exposures can build up and a failure of one participant in the system can then lead to the failure of other participants in the system. Fry et al. (1999) report that, at the same time as these exposures were becoming larger, advances in IT meant that it became increasingly technologically feasible to settle payments gross and in real time. Since doing this eliminates credit risk from a payment system, central banks increasingly favoured real-time gross settlement (RTGS) as the settlement rule within their countries’ large-value payment systems. In particular, in 1995 Switzerland and the United States were the only major countries relying on RTGS systems for their large-value payments. The Bank of Japan, which had offered both DNS and RTGS systems, switched to only offering RTGS in the late 1990s; in the United Kingdom,

88

B. Lester et al.

CHAPS switched to settling payments on an RTGS basis in 1996; and Euro area central-bank-administered, wholesale, systems have operated as RTGS systems since 1997.4 But, RTGS systems can be more costly than DNS systems. In addition, to the higher IT costs involved in setting up and running such systems, RTGS systems are ‘liquidity hungry’ relative to DNS systems. That is, participant banks require more liquidity to settle their payments in an RTGS system than in a DNS system. In turn, this liquidity is costly as when banks do not have it to hand, they will need to borrow. To a degree, central banks can mitigate this cost by providing intraday liquidity at low cost – typically free, so long as it is collateralized – but, even then, this liquidity will still carry an opportunity cost. So, it is not at all clear that moving from a DNS to an RTGS payment system is necessarily welfare improving. Indeed, in a comparison of the costs of secured net settlement on CHIPS – at the time a DNS system – to those of an otherwise equivalent RTGS system, Schoenmaker (1995) concludes that ‘the estimated extra cost of RTGS exceeds the estimated reduction in settlement risk’.5 Furthermore, George Selgin, in his chapter in this volume, argues that the ‘credit risk’ between banks in DNS systems used as a justification for imposing RTGS does not exist. He argues that this is because a bank is only exposed to the risk of another bank failing to meet a net obligation in a DNS system if it credits customer accounts before settlement occurs, something it does not actually have to do. In addition, such customer credits can typically be reversed in the event of settlement not taking place. Therefore, Selgin argues, all agents involved face the right incentives to manage these risks and there is no market failure in payment systems. As a result, he suggests that the imposition of RTGS, where the market had settled on a DNS system, must be welfare reducing. The purpose of this chapter is to construct a model within which we can begin to explore the trade-off between cost and risk in payment systems. The model shows that under certain assumptions, DNS and RTGS can both be a payment system’s settlement rule in equilibrium. The presence of multiple equilibria opens up the possibility that private agents fail to coordinate on the optimal equilibrium settlement rule. However, we stop short of claiming that there can be a case for government intervention to help coordinate private agents to choose the optimal rule. Rather, the purpose of the chapter is to highlight some of the features of the economy that play an important part in determining a payment system’s settlement rule. Relaxing some of the restrictive assumptions we make remains avenues for future research. We consider a banking economy in the spirit of He et al. (2005), ignoring theft as a motive for banking and instead focussing on the case of interestbearing deposits. Moreover, we introduce into the He et al. model two possible payment systems, and endogenize the choice of payment system for both buyers and sellers. Incorporating this creates a framework within which we can analyse the given payment system as an equilibrium outcome of the economy. In particular, in our model the end recipients of payments placed through a DNS system are exposed directly to the possibility of default by the banks of the

Optimal settlement rules for payment systems

89

payees. We make three key assumptions in the model. The first is that prices are exogenously fixed. One good always transfers for one unit of money. The implication of this is that settlement rules can have different effects on buyers and sellers. The second assumption is that banks are unable to charge buyers for the cost of RTGS. Third, sellers cannot offer incentives to buyers to switch settlement rules. This assumption partly derives from the first – sellers cannot use prices to induce buyers to switch settlement rules – but goes further by prohibiting sellers from using other strategies to get buyers to switch; e.g. sellers cannot base the quality of the good on the settlement rule. We start by considering an economy in which a DNS equilibrium exists and first show that if the costs of using an RTGS system are too high, then there will not be an equilibrium in which agents use an RTGS system. We think of this economy as representative of a time when IT had not developed to the extent to allow RTGS to occur at an economical price. We also show that if these costs become low enough an RTGS equilibrium will exist, in addition to the DNS equilibrium. We think of this economy as representative of developed economies in the early 1990s. Finally, we show that if the costs of RTGS are low enough relative to the costs of DNS, under the assumptions we make, the equilibrium in which all agents use the RTGS system generates a higher value of social welfare than that in which all agents use a DNS system.6 Since our model cannot handle dynamics and, in particular, the endogenous decision to move from one equilibrium to another, we cannot handle the question of which equilibrium – RTGS or DNS – will be selected by agents in the economy. We would argue that historically, DNS was the only system that could be used in equilibrium and that we have now moved to a situation in which either a DNS or an RTGS system could be used in equilibrium. An analysis of how private agents coordinate on one of the equilibria and whether they can overcome potential coordination failures is left to subsequent research. This research would help us better understand the role of public authorities in the move towards RTGS that occurred in the early to late 1990s in many countries. The chapter is structured as follows. We first outline our model before discussing equilibria within it. We then compare welfare over regions of the parameter space within which both DNS and RTGS equilibria exist showing that, under the assumptions we make, we can find a critical value for the costs of the RTGS system below which the RTGS equilibrium welfare dominates the DNS equilibrium. Finally, we conclude with some suggestions for future work.

The model We begin by describing a simple random matching model of money that we use as a platform for our analysis. In the economy there is a unit continuum of infinitely-lived agents. A proportion M [0,1] of agents are each endowed with one indivisible unit of fiat money. Agents produce and consume indivisible goods. In each period, agents are randomly and anonymously matched with one another. In any pair-wise meeting, a double coincidence of wants occurs with

90

B. Lester et al.

zero probability. Single coincidence meetings, where one agent wants the other’s good but not vice versa, occur with probability, x. Given the absence of any double coincidence of wants, the only feasible trades involve the exchange of one unit of money for one good. In any single coincidence meeting, the buyer receives utility, u while the seller incurs cost of production, c. Banks We build on this simple random matching model of money by allowing agents either to hold money in the form of cash or to deposit it in a bank. Banks are modelled in the same way as in He et al. (2005) and, in particular, are perfectly competitive. An agent deposits his money at a bank with probability, and pays a fee, , which is derived from the setup of the banking sector. Banks face a fixed cost, a, for managing each account. Banks also make loans, L, to agents without money but must retain a fraction of deposits (D) as reserves. We denote the measure of agents holding their money as cash as M0 and the measure of agents holding either cash or having a bank account as M1 (total money supply). So L + M = M1 and D + M0 = M1. Banks charge an upfront fee, , for loans. Settlement Whether an agent chooses to hold his money in the form of cash or a bank deposit it has implications for when money is transferred between buyers and sellers. To capture this we assume that each period is divided up into two subperiods. Trade between agents takes place during the first sub-period, which we refer to as the morning. If a buyer uses cash, money is transferred between the buyer and the seller in the morning since cash changes hands at the point of sale. If a buyer makes a payment to a seller from his bank account, the money has to be transferred via an inter-bank payment system.7 There is a single inter-bank payment system in the economy. When inter-bank payments are received depends on the rule in place governing how payments are settled. Payments are settled in the morning if they are made through a real-time gross settlement (RTGS) payment system and are settled in the afternoon if they are made through a deferred net settlement (DNS) payment system. The timing of the settlement process would be irrelevant if there is perfect commitment among banks, as there is in He et al. (2005). Agents would simply choose the cheaper of the two settlement rules as money is transferred from the buyer’s bank to the seller’s bank before the next trading sub-period with certainty. However, the timing of the settlement process is crucial if there is a possibility of bank insolvency between when trading occurs and when the transfers of money are completed. Sellers may fail to receive funds altogether or receive them only at a cost if a bank becomes insolvent before the settlement of payment occurs. We introduce the possibility of bank default into the model by assuming that there is an exogenous risk that each bank could become insolvent and consequently default on any outstanding payment obligations it may have. Banks


91

can become insolvent between the morning and the afternoon in every period. We assume that there is full deposit insurance implying that depositors face no costs as a result of bank default. In the following morning, their money is transferred to accounts at new banks that replace the insolvent institutions. Any default situation is resolved and sellers receive their money but only at a cost. The expected per payment cost incurred by a seller is . This cost represents a deadweight loss to the economy and can be thought of as being the real resource cost of winding up insolvent institutions and sorting out the residual claims on it (i.e. ‘bankruptcy’ costs). Sellers are not exposed to default risk when they receive an RTGS payment because they receive a payment before buyers’ banks can become insolvent. But reducing the lag between trade and settlement comes at a cost. First, there exist bureaucratic costs to settling every transaction more promptly. Moreover, a higher frequency of settlements requires banks to hold larger amounts of idle reserves, thereby decreasing the revenue earned per unit of money deposited.8 Finally, some banks will also be forced to borrow from the monetary authority or enter the interbank market in order to obtain sufficient reserves to complete settlement of all of their customers’ transactions. These additional costs will result in customers facing higher fees in a perfectly competitive banking sector. The cost of making RTGS payments is captured by a per-period cost that is levied on all bank accounts in proportion to the amount of RTGS payments made. In a single coincidence meeting, the buyer proposes whether he wishes to pay in cash or in the form of a payment from his bank account. The seller then chooses between accepting a payment through these means and not trading. Trade does not take place if they cannot agree. Trade always occurs when a buyer offers to pay in cash because a seller incurs no cost and is exposed to no risk when receiving cash. If a buyer offers to make a payment from his bank account and the interbank payment system is RTGS, the seller accepts with probability one since there is no cost or risk to him from receiving a payment this way. If the interbank payment system is DNS, a seller accepts a payment with probability . Let = 1(0) if the interbank payment system is DNS (RTGS). Therefore, trade takes place in a single coincidence meeting, when the buyer wishes to make a payment from his bank account, with probability P( ,) = (1 – ) + . It also follows that the cost to a bank of providing payment services is K( ) = (1 – ). This implies that the per-period fee that each (perfectly competitive) bank charges a customer for providing deposit services is ( ) = a – (1 – )rp + (1 – ) where r is the discount rate. Bellman equations We can now derive the Bellman equations for the value of being a buyer holding cash, V1m, the value of being a buyer who has a bank deposit, V1d, and the value of being a seller, V0. These are shown in equations (1)–(3).

92

B. Lester et al. rV0 = (M0x + (M1 – M0)xP( ,))(V1 – V0 – c) – (M1 – M0)x

(1)

rV1m = (1 – M1)x(u + V0 – V1) + V1 – V1m

(2)

rV1d = (1 – M1)xP( ,)(u + V0 – V1) + V1 – V1d – (1 + r)( )

(3)

where V1 = max {V1m,V1d} is the value of being a buyer prior to the decision of whether or not to deposit his money in the bank. The first component of equation (1) is the gains from trade enjoyed by a seller multiplied by the probability of meeting someone to trade with in any given period. The second component is the expected cost of failing to receive a DNS payment times the probability of accepting a DNS payment. The first part of equation (2) is the gains from trade enjoyed by a buyer times the probability of trading in a given period when he offers to make payments in cash. The first component of equation (3) is the same except that it represents the gains from trade times the probability of trading when a buyer deposits his money in a bank and makes payments from his account. The second component of equation (3) is the fee paid by a buyer for depositing his money in the bank.

Equilibrium analysis In this section we examine pure-strategy equilibria in our model in which all money is deposited in banks; i.e. V1 = V1d, M0 = 0 and M1 = M/. Therefore, a buyer’s Bellman equation is: rV1 = (1 – M1)xP( ,)(u + V0 – V1) – (1 + r)( )

(4)

We derive conditions for an equilibrium in which trade occurs with only a DNS payment system and one in which trade occurs with only an RTGS payment system. We then compare the conditions under which each type of equilibrium exists. Such equilibria will exist if the following conditions hold: Individual rationality: the value of being a seller or a buyer is at least as good as leaving the market and living in autarky: V0 0 and V1 0. Incentive compatibility: a seller has an incentive to produce his good in exchange for one unit of money and a buyer has an incentive to trade his unit of money for one unit of the good: V1 – V0 c + and u V1 – V0. Banking constraint: agents choose to deposit money in a bank: ( ) 0. The probability that trade occurs equals one in any single coincidence meeting when we restrict attention to pure-strategy equilibria. Thus, the only difference between holding money as cash and depositing it in the bank is the banking fee. Buyers have incentives to deposit money in banks as long as they receive a return from doing so; that is, the banking fee is negative. Hence, ( ) 0 is both necessary and sufficient for agents to choose to deposit money. The buyer’s individual rationality constraint holds if the seller’s individual


93

rationality constraint and the buyer’s incentive compatibility constraints are both satisfied. Thus, we only need to derive the conditions under which V0 0, V1 – V0 c + , u V1 – V0 and ( ) 0 hold. Now, it is not feasible for all agents without money to borrow from a bank because then there would be no sellers. So, in equilibrium, the loan market clears at a rate at which sellers are indifferent between remaining a seller and borrowing. As stated in He et al. (2005), they will be indifferent when the cost of borrowing, , is equal to the difference between the value of being a buyer and the value of being a seller, V1 – V0. Using this result, equations (1) and (4) then imply the following expressions for the values of being a buyer and of being a seller, respectively: M1xP( ,)[(1 – M1)xP( ,)u – (1 + r)(a + (1 – ))] rV0 = r((1 + r) – r) + xP( ,) M1x[(1 – M1)xP( ,) + r((1 + r) – r)][P( ,)c + ] – r((1 + r) – r) + xP( ,)

(5)

[r + M1xP( ,)][(1 – M1)xP( ,)u – (1 + r)(a + (1 – ))] rV1 = r((1 + r) – r) + xP( ,) M1x[(1 – M1)xP( ,) – r(1 + r)(1 – )][P( ,)c + ] – r((1 + r) – r) + xP( ,)

(6)

We use these equations to derive ranges of parameter values for which there exists an equilibrium in which there is trade when there is only a DNS payment system ( = = 1) and when there is only an RTGS payment system ( = 0). The ranges of values of x and a for which there exist such equilibria are illustrated in Figures 6.1–6.4. In each figure, the dashed line represents the seller’s individual rationality and incentive compatibility conditions. These conditions hold for values of x that are sufficiently high, relative to the value of a, that (x,a) lies to the right of the dashed line. This is because the value of being a seller is increasing in the probability of meeting an agent with whom to trade (x) but decreasing in the cost of operating a banking account (a). The dashed line is upward sloping since when a rises, sellers must be compensated by an increase in x for their individual rationality and incentive compatibility conditions to still hold. The solid line represents the banking constraint. This condition is met for values of x that are sufficiently high relative to the value of a because the banking fee is only negative if the cost of operating an account (a) is low enough compared with the loan market rate (). The loan market rate is increasing in the gain from moving from being a seller to a buyer, which is obviously increasing in the probability of trading (x). It follows that the solid line is upward sloping.9 The grey, highlighted, areas show possible equilibria.

94

B. Lester et al.

Figure 6.1 shows the DNS equilibrium. RTGS equilibria are shown in Figures 6.2–6.4. The figures show that the range of parameter values for which there exists an RTGS equilibrium grows smaller the more costly is the RTGS (i.e. the higher is ). When is high enough no RTGS equilibrium will exist (see Figure 6.4). Historically, was very high and hence, trade was only possible if payments settled on a deferred net basis. It is likely that the value of has fallen over time 0.03 a

0.02

0.01

0

0.1

x

0 0.2

Figure 6.1 DNS equilibrium.

0.03 a

0.02

0.01

0

0.1

Figure 6.2 RTGS equilibrium ( = 0).

x

0 0.2


95

0.03 a

0.02

0.01

0

0.1

x

0 0.2

Figure 6.3 RTGS equilibrium ( = 0.005).

0.03 a

0.02

0.01

0

0.1

x

0 0.2

Figure 6.4 RTGS equilibrium. Note: Figures 6.1 to 6.4 are drawn for M = 0.5, = 0.7, u = 5, c = 2, = 0.003, r = 0.02.

(say, with improvements in information technology). This means that an equilibrium in which payments settle on an RTGS basis is now possible, in addition to one in which they settle on a DNS basis (Figures 6.2 and 6.3). Further reductions in would expand the range of parameter values for which there exist both RTGS and DNS equilibria.

96

B. Lester et al.

Welfare In this section we will analyse the level of social welfare under the two settlement rules. We identify the (positive) values of and under which one of the settlement rules has superior welfare properties under the particular assumptions we have made about fixed prices and the difference between buyers and sellers in terms of how they incur the costs of DNS. Social welfare, W, is defined as the average of the value of being a buyer and the value of being a seller: W = M1V1 + (1 – M1)V0

(7)

The social planner chooses whether payments settle on a DNS or RTGS basis to maximize W subject to agents’ individual rationality constraints, incentive compatibility constraints and banking constraint. Using equations (5) and (6) and setting both and to unity, we can show a DNS equilibrium will exist if [1,2], where 1 ensures that the banking constraint holds and = 2 ensures that the seller’s individual rationality and incentive compatibility conditions hold. The values of 1 and 2 are (r + x)a – rx(1 – )[(1 – M1)u + M1c] 1 = rM1x(1 – ) and (1 – M1)xu – (1 + r)a 2 = – c. (1 – M1)x + r((1 + r) – r) Similarly, using equations (5) and (6) and setting to zero, we can show that an RTGS equilibrium will exist if min {1,2}, where 1 ensures that the banking constraint holds and 2 ensures that the seller’s individual rationality and incentive compatibility conditions hold. The values of 1 and 2 are rx(1 – )[(1 – M1)u + M1c] 1 = – a r+x and (1 – M1)xu – [(1 – M1)x + r((1 + r) – r]c 2 = –a. 1+r It is straightforward to show that (1 – M1)x 1 ()c. r + (1 – M1)x


97

RTGS

Autarchy

DNS and RTGS

DNS

2

Min {1, 2}

Figure 6.5 Existence of DNS and RTGS equilibria.

When c is relatively low, the relevant constraint on is the banking constraint; the cost of settlement must be sufficiently small so that banks can continue to pay interest on deposits. Alternatively, when c is relatively high, the relevant constraint are the seller’s incentive constraints because must be sufficiently small that a seller finds it profitable to produce and trade. Before comparing welfare levels under either settlement rule, note that we can ignore the constraint 1 because of the fact that 1 0 implies 1 0. In other words, if the banking constraint can be satisfied under RTGS for a positive value of it will definitely be satisfied (i.e. not bind) under DNS. The banking constraint is satisfied under RTGS as long as the gains from trade enjoyed by a seller are sufficiently high relative to the cost of RTGS. Under DNS, there is no cost from using this settlement rule and there is also an additional benefit to the seller of not facing the cost in the following period. It follows that if the banking constraint can be satisfied under RTGS the banking constraint will not bind under DNS. Figure 6.5 depicts the different equilibria that can occur for different values of and . Both the DNS and the RTGS equilibria are possible when the costs of default and the costs of making payments on an RTGS basis are low enough that 2 and min {1,2}. When exceeds 2, but min {1,2}, there exists an RTGS equilibrium but no DNS equilibrium. There is a DNS equilibrium but no RTGS equilibrium when 2 and > min {1,2}. Finally, if both costs are sufficiently high, no DNS or RTGS equilibrium will exist. When either a DNS equilibrium or an RTGS equilibrium could exist, welfare would not be unambiguously higher in one case than the other. We can use equations (5), (6) and (7) to calculate welfare under each of the two settlement rules, again setting both and to unity in the DNS case and setting to zero in the RTGS case. Welfare is the same under either settlement rule if

98

B. Lester et al. L

2 A B

Min {1, 2}

Figure 6.6 Welfare.

(1 + r)(r + x) = . x[r((1 + r) – r) + (1 – M1)x – rM1] When = 0 (that is, RTGS is costless) welfare is the same under both settlement rules only if DNS is also free of cost ( = 0). Thus, the locus values of and for which welfare is the same under either settlement rule passes through the origin and divides the area in which both equilibria exist into two.10 Welfare is higher under RTGS above the locus (region A) and is higher under DNS below the locus (region B). The locus is depicted in Figure 6.6. In summary, when trade could occur under either settlement rule in equilibrium, it may be possible to rank the equilibria in terms of welfare. But without modelling how agents select settlement rules, the analysis does not indicate whether they could coordinate on the equilibrium providing the highest welfare.

Conclusions and future work In this chapter we have constructed a model for examining the trade-off between cost and risk in DNS and RTGS payment systems. The model showed that when the costs of settling payments on an RTGS basis are high, only a DNS equilibrium can exist. The opposite is true when the costs of default are high: only an RTGS equilibrium can exist. Either settlement rule may hold in equilibrium for intermediate values of costs. The model as it stands makes several restrictive assumptions and so is not general enough to answer the question of whether agents can coordinate on a welfare-superior settlement rule and any role for public policy in preventing coordination failures. Nonetheless the model offers a starting point for developing a general model with which these questions can be tackled.


99

Notes 1 The views expressed in this chapter are those of the authors and not necessarily those of the Bank of England. 2 Bank of England (2006). 3 Bank for International Settlements (2005). For more on the Bank of England’s roles and responsibilities in the area of payment systems, see Bank of England (2005). 4 The chapter by Morten Bech in this volume discusses the diffusion of RTGS systems across the world. 5 The Clearing House Interbank Payment System (CHIPS) is a private payment system operated by the New York Clearing House. At the time of Schoenmaker’s work (1995) it was a DNS system; as of 2001 settlement happens almost continuously. 6 While in this chapter we assume the existence of a cost–risk trade-off between RTGS and DNS, this is derived from first principles in Lester (2005). 7 We assume that although a seller may not already have a bank account when a buyer offers to make a payment from his bank account, he will open an account if he agrees to receive money in this way. The seller pays a fee for holding a bank account in the following period when he is a buyer. 8 Taken literally, this would suggest that would be higher if the payment system in operation were RTGS than if it were DNS. But, to keep the model simple while making our point, we lump this effect in and leave exogenous. 9 For the parameter values we use, the buyer’s incentive compatibility condition holds for all values of x and a. 10 The locus is upward sloping because the value of being a buyer (V1) is decreasing in the production cost (c) and the cost of default ().

References Bank for International Settlements (2005) New developments in large-value payment systems, Committee on Payment and Settlement Systems Publication No. 67. Bank of England (2005) Payment systems oversight report 2004, London: Bank of England. Bank of England (2006) Payment systems oversight report 2005, London: Bank of England. Fry, M.J., Kilato, I., Roger, S., Senderowicz, K., Sheppard, D., Solis, F. and Trundle, J. (1999) Payment systems in global perspective, London: Routledge. He, P., Huang, L. and Wright, R. (2005) ‘Money and banking in search equilibrium’, International Economic Review, 46: 637–70. Lester, B. (2005) ‘A model of interbank settlement’, unpublished thesis, University of Pennsylvania. Schoenmaker, D. (1995) ‘A comparison of alternative interbank settlement systems’, London School of Economics Financial Markets Group Special Paper, No. 204. Zhou, R (2000), ‘Understanding intraday credit in large-value payment systems’, Federal Reserve Bank of Chicago Economic Perspectives, 24(3): 29–44.

7

The microstructure of money James McAndrews1

Introduction In his chapter in this volume, Ed Green suggested that ‘payment economics comprises the topics that pertain to both monetary economics and industrial organization’: loosely paraphrasing, the study of the industrial organization of money. In this chapter, I approach the study of payment systems as the study of the market microstructure of money. In doing so, I will also use the lens of market microstructure to address issues relating to trends in payment systems. The study of the microstructure of financial markets has focused on the institutional arrangements for the exchange of financial instruments. This study has examined the efficiency of different market structures. Various measures of efficiency including the size of the bid–ask spread and the speed of execution have been explored. In addition, the conditions under which one market structure or another is more appropriate, given the underlying economic environment, has been a question of interest. Payment systems are analogous to financial markets. In this chapter I will explore this analogy in depth. I will examine first the analogies between alternative arrangements for the exchange of financial contracts and for money. The primary focus is on payment systems in which only money (or deposit balances) is exchanged. A key difference between payment systems and financial exchanges is that the price is held fixed in payment systems (at least at first glance). Nonetheless, the organization of payment systems has been examined in theory by many authors who argue that the behavior of participants differs in payment systems of differing designs. Given these analogies we can apply the microstructure literature of financial markets to that of payment systems. This application yields various insights into the conditions in which one payment system design is more appropriate than another based on the volatilities of payment sizes, arrival rates, and likelihood of offsetting other payments. A second line of application of microstructure literature focuses on the priced aspects of payment systems. This line of thought suggests that large-value payment systems provide the settlement system for the overnight money market. As a result, the microstructure of the payment system and the money market are

The microstructure of money

101

linked. I explore this application by presenting empirical estimates of an inventory model of dealer behavior in the fed funds market to a sample of large US banks. The null hypothesis of no inventory effects, as would be the case in a frictionless market, would suggest that the current balance of the bank would not affect the price quotes of the bank. Such a hypothesis would be consistent with a risk neutral dealer that had no limits to the balance it could hold in the central bank. I find some evidence of inventory effects in the US federal funds market. The examination of the empirical results of the inventory model leads to a discussion of how the design of the payment system affects the overnight money market. A review of the features of payment systems suggest that daylight credit policies, the design of the payment system as an RTGS, a DNS, or a queueaugmented system, the importance of links to ancillary payment systems, discount window policies, and reserve accounting rules all affect both the precautionary demand for overnight balances and the tightness, or elasticities of demand, of the money market. This discussion suggests new methods of examining payment system policy and its effects in the money market. The chapter concludes with a discussion of the role of payment systems in the money market. In addition, I discuss recent trends in payment systems, how the discussion of microstructure can assist us in understanding those trends and what they portend for payment system developments.

Designs for payment systems and security markets The design of large-value payment systems has been closely examined in Bank for International Settlements (1989, 1990, 1997, 2005). The alternative designs for large-value payment systems can be roughly characterized as follows. A netting system, often referred to as a designated-time net settlement system (DNS), cumulates payment orders until a specified time. At the designated time a settlement agent calculates the multilateral net amount that each participant owes or is owed. The net debtors are required to send funds in the amount of their net debts, and upon receipt of these funds, the net creditors are paid out the amounts owed to them.2 A real-time gross settlement system (RTGS), in contrast, settles each payment order against balances on account of the settlement institution. Each payment order is settled, if possible, on arrival and in the full amount of the payment order.3 Recently, a number of systems that are not easily classified as either DNS or RTGS have been put into operation. These systems, such as the ‘new CHIPS’ system in the United States and ‘RTGS plus’ in Germany, are not easily characterized. One aspect of these systems, perhaps easiest to understand in the RTGS plus system, is that participants can specify a priority for each payment order. An express payment is meant to be settled as in an RTGS system: immediately against balances. A limit payment is intended to be settled in a liquidity savings mode. The participants can establish the limits and they can be limits on the total of payments to be settled in that mode, and the bilateral or the multilateral net amount settled in that mode. The purpose of the limits is

102

J. McAndrews to reserve liquidity exclusively for the execution of urgent payments (Express payments); to prevent one-sided dissipation of liquidity in comparison with individual participants; to synchronise the payment flow with other participants; and to promote its early submission.4

There are numerous possible variations on the possible designs of payment systems of this type, that is, payment systems that allow prioritizing payments by criteria other than the time of submission, and that allow payments to be settled out of a queue against incoming balances. This type of payment system has been called a ‘hybrid system’ but in this chapter I will refer to it as a ‘limit payment order’ system.5 Each of the designs for payments system has an analogue in the designs of security markets. A RTGS system is analogous to a continuous auction market, in which trades can be made throughout the market day. In continuous auction markets, trades arrive throughout the day, and specialists or dealers match buy and sell orders, concluding transactions quickly. In a continuous auction for a security, the price moves based on market conditions, while in a payment system only the quantity of deposit account balances moves.6 Payment systems typically settle trades for securities, debts, or commercial goods or services that are delivered outside of the payment system. For example, the title to a building or security is transferred in an office, and the banks representing the buyer and seller transfer the funds in an RTGS system. A DNS system is analogous to a call market, such as occurs at the opening of the New York Stock Exchange, in which orders are accumulated and trade occurs at a price that clears the market. In a DNS, payment orders are accumulated and netting of trades occurs. In call markets, liquidity is concentrated and various authors have posited reasons why call markets might be efficient in certain environments. Similarly, a DNS system concentrates funding liquidity, as offsetting payments reduce the amount of account balances that need to be transferred between parties. Finally, limit-order books and crossing markets accumulate trade orders on either side of the market at particular prices, and settle them when a market order or a new limit order enters the queue and ‘crosses’ or satisfies the limit price of the queued order. The analogy to payment systems is found in limit payment order systems in which payment orders are queued pending the satisfaction of the limit condition established by the participant. For example, the limit condition might be the arrival of funds into the participant’s account, such as an offsetting payment order from a bilateral counterparty. In this analogy, an express payment is like a market order, crossing the limit payment order, and allowing it to settle without pre-establishing account balances, as would be necessary in a pure RTGS system. Large-value payments are usually intended to settle debts of fixed nominal value. As such, much of the market microstructure literature of securities markets, which studies the efficiency of price-setting in markets of various designs, would not apply directly to payment systems. However, several aspects of the market microstructure literature do apply to payment systems.


103

The participants in payment systems are typically banks. In most payment systems, banks either have to post collateral or pay fees if they borrow during the day (from the central bank) to acquire more account balances. By the end of the day, the bank has to repay their intraday borrowing, borrow from the discount window (usually at a rate that exceeds the overnight rate in the market), or incur an overnight overdraft on which the bank pays a penalty rate. As a result, banks treat the balances that they transfer in payment systems as scarce. A number of market microstructure concerns are applicable to the analysis of payment systems. Before delving into the areas which have been investigated in the literature, I will contrast some of the fundamental assumptions in the market microstructure literature. Generally, in securities markets, traders are often modeled as being risk averse, or being subject to convex costs of trading; the basic risk facing participants in securities markets is market risk of inventory: the risk that the price of the security will change during the planning period; and securities markets are subject to relatively high bid–ask spreads. In contrast, in the money market, banks are usually modeled as risk-neutral, but subject to a convex cost function related to penalties to missing their reserve requirement (although even these might be only slightly convex on most days because of a high degree of substitutability of reserves across the days of a maintenance period); the basic risk they are subject to is liquidity risk, rather than market risk; and money markets typically display low bid–ask spreads. The market microstructure literature contrasts the outcomes of continuous and call markets, suggesting that call markets can be useful when liquidity is particularly scarce. Admanti and Pfleiderer (1988) and Pagano (1989) show that clustering of trades occurs even in continuous auctions. Because multiple equilibria can occur, however, call markets can assist in selecting a particular clustering equilibrium, minimizing the costs of trade. In the case of payment systems, this would translate into minimizing the costs of balances needed to settle the day’s payments. McAndrews and Rajan (2000) conduct an empirical investigation of clustering of payments in the US Fedwire RTGS system. Other studies of continuous versus call auctions focus on adverse selection effects of market design. Vayanos (1999) models a situation in which market participants with private information about their own risk sharing demands limit trades in a continuous auction to limit the market impact. This motive, and other similar motives, might result in payment system participants limiting the amount they wish to pay at one time in an RTGS, instead preferring to wait on the arrival of expected payments from their counterparties. Such motives are discussed in Bank for International Settlements (2005) and in Kahn et al. (2003), and may result in situations in which netting is preferable to RTGS. Freixas and Parigi (1998) consider interbank contagion effects as depositors seek to avoid bank failure in a gross or net payment system. Kahn and Roberds (1998) consider a rich model that reveals that the choice of payment system design creates alternative adverse selection and moral hazard incentives for banks. Another issue discussed by market microstructure theorists is the value of transparency in securities markets. Transparency has been shown theoretically to

104

J. McAndrews

reduce the problems of adverse selection in securities markets, as in Pagano and Roell (1996), although empirical results suggest mixed effects. The effect of greater transparency in payment systems has been examined in the paper by Willison (2005). A number of other issues, including the strategic supply of liquidity (Angelini 1998; Bech and Garratt 2003), intermarket competition and liquidity fragmentation are examined in the market microstructure literature, all of which can be applied to the study of payment systems. How does a bank manage to pay back its intraday borrowings from the central bank if it experiences a net outflow from its account as a result of payments its customers request to be made, or that it makes for itself? The bank can enter the overnight market and borrow funds from another bank. We now turn to a discussion of the market microstructure of the overnight market.

Payment systems and the money market Large-value payment systems provide the settlement infrastructure for the overnight money market in most countries. In the United States, federal funds market sales are settled by the seller of funds making a funds transfer on Fedwire, the Federal Reserve System’s large-value, RTGS payment system. The money market assists banks as they make payments. As Skeie (2004) points out in a model of banking and money, as banks make net outflows from their payment activity, they experience a demand to borrow on the overnight market to avoid overnight negative positions that would otherwise result. The recognition that payments activities of banks and their overnight borrowing are so closely linked suggests that the microstructure of the payment system influences activity on the money market. Consider first an RTGS payment system. In such a system, information about a bank’s balance evolves over the day as payments flow into its account and customers request that payments are made on their behalf. In this case, a bank may not have uncertainty about its balance resolved until very near the close of the payment system. The reserve accounting rules that govern the reserve requirements in the United States allow banks to meet their requirements by averaging reserve levels over a two-week reserve maintenance period.7 The ability to average shortages on one day with excess amounts on another day tends to increase the elasticity of a bank’s demand for balances late in the day (on an average day). This lessens the bank’s need to be extremely precise in its reserve management and maintain its reserves in an extremely tight band. However, because reserve balances do not earn interest, banks have an incentive to avoid holding excess reserves, and as shortfalls in reserve holdings incur penalties, banks wish to avoid shortfalls in reserves as well. In addition, banks’ balances should not fall below zero on any day. These requirements suggest that banks take care that their balances come close to the level that they are targeting. As they make and receive payments in the late afternoon, they manage their account balances to achieve their desired


105

end-of-day balance. These considerations suggest that inventories might play a role in the federal funds market. Modern microstructure theory has focused on informational differences in traders, but I will focus on inventory effects in the federal funds market as an example of the microstructure approach to payment systems and the money market. The federal funds market has been well described in the papers by Demilralp et al. (2004), Furfine (1999, 2003), Goodfriend and Whelpley (1993), Griffiths and Winters (1995), Hamilton (1996, 1997), Lee (2003), and Stigum (1990). These papers have established many regularities of the federal funds market. Inventory models One of the most basic models of market microstructure suggests that there are inventory effects in dealer markets. Garman (1976) examines a risk-neutral dealer market. In his model the stochastic flow of buy and sell orders is pricedependent. The dealer has an obligation to maintain continuous trading, and because the orders arrive stochastically, the dealer is motivated to carry an inventory. Garman goes on to examine an inventory-independent pricing policy. Amihud and Mendelson (1980) extend Garman’s result to derive the optimal inventory-dependent pricing policy by the monopoly dealer when it has constraints on its short and long position. They show that prices, both bid and ask, are monotonically decreasing in inventory. This result, and similar ones for risk-averse dealers (as in Stoll (1978) and Ho and Stoll (1981, 1983)) have led to various empirical tests of the theory. Hasbrouck and Sofianos (1993) and Madhavan and Smidt (1993) show mean reversion in specialist inventories; Lyons (1995) applies the inventory model to the foreign exchange market; Manaster and Mann (1996) find that market makers in the Chicago Mercantile Exchange with long positions tend to sell. Biais et al. (2004) review many other studies as well. The result that bid and ask prices are monotonically decreasing in inventories is often implemented by showing that the midpoint of the bid–ask spread of dealers is decreasing in the inventory of dealers. In what follows I apply this to the federal funds market. Sales in the federal funds market (again, excellent descriptions of the market are available in Demilralp et al. (2004) and in Furfine (1999)) are delivered on Fedwire. Federal funds trades can either be brokered or direct. Furfine (1999) examines the patterns of participant, timing, and concentration in the federal funds market. He finds that the largest five banks by asset size in 1998 accounted for 24 percent of industry assets, but an even greater share of both federal fund purchases and sales. Those banks accounted for 38 percent of federal funds sold and purchased 34 percent of federal funds bought (by value). Similarly the top ten banks in asset size sold 47 percent of the federal funds transactions, and bought 48 percent of the federal funds by value in the period. Taking advantage of this concentration in federal funds markets we collect a sample of likely federal funds transactions from the Fedwire transactions journal.8 Figure 7.1 displays the intraday pattern of identified federal funds

106

J. McAndrews

1,800 1,600

14 Volume Value

12

US$ (billions)

10 1,200 1,000

8

800

6

600 4

Volume (thousands)

1,400

400 2

200 0 12:30 a.m.

2:30 a.m.

4:30 a.m.

6:30 a.m.

8:30 a.m.

10:30 a.m.

12:30 p.m.

2:30 p.m.

4:30 p.m.

0 6:30 p.m.

Time

Figure 7.1 Intraday pattern of activity: total value and volume of federal funds traded (2 October 2001–31 September 2004) (source: Federal Reserve Bank of New York: author’s calculations). Note Values shown are the average values in a calendar minute calculated over the days of the sample period.

activity averaged over the sample period, from October 2001 through September 2004. The levels of the lines in the chart are per-minute averages for each minute of the Fedwire operating day, averaged across the sample period.9 There is a great deal of clustering in federal funds trading activity, as there is in payments activity generally in Fedwire. Our test of inventory effects focuses on banks that are active on both sides of the market for Fedwire during the period of heaviest trade in federal funds. Recognizing that large banks are active on both sides of the market, we treat the banks as dealers. There may well be economies of scope between making and receiving large amounts of payments and buying and selling federal funds. It may be that these economies of liquidity generation can explain the concentration of banks actively buying and selling federal funds. We treat our sample of banks as dealers, and test whether their balances have an effect on the transaction prices of their purchases and sales. While the perfect market hypothesis suggests that dealers face no inventory constraints, a finding of inventory effects on prices does not imply that the market for federal funds is inefficient. As Amihud and Mendelson (1980) point out, there can be inventory effects in an efficient market, in that no one can profit from knowledge of the market-maker’s inventory position and its pricing policy.


107

Data and tests The data consists of the federal funds transactions of several US commercial banks that are regularly active on both sides of the market. In addition, I supplement this with the bank’s balances throughout the day (calculated from the transaction journal) and the bank’s transactions deposits from the Call Reports. The measure of the balance we use is the bank’s actual balance divided by that quarter’s level of transactions deposits from the Call Reports; this provides a useful way to scale the level of a bank’s balance in relation to a rough gauge of its reserve demands. I will call this variable the bank’s transactions balance. Using this data, I first calculate the value-weighted spread of interest rates on a bank’s sales of federal funds and its purchases of federal funds over the period between 4.00 p.m. and 4.15 p.m. (as well as between 4.30 p.m. and 4.45 p.m.). Then I calculate the midpoint of the value-weighted average federal funds bidask spread over the period and subtract the target federal funds rate. I’ll call this variable the bank’s midpoint minus target rate. The test we construct is to test whether the midpoint of the excess of the bidask spread over the target rate of dealers is decreasing in the inventory of dealers. I first regress the midpoint minus target rate between 4.00 and 4.15 on the bank’s transactions balance (its inventory of balances) at 4.00 p.m. I choose 4.00 p.m. because the market is so active at that time. The estimation tests whether a bank’s balance at 4.00 p.m. influences its federal funds activity over the following 15 minutes. This might be considered a rigorous test, as banks may not pay close attention to the balance at any one moment, and they may not adjust their federal funds activity in response to a particular balance. Instead, banks may have a better estimate of the average balance they wish to achieve over some time period late in the day. If any inventory effects are found, it might be useful to test these less restrictive approaches to identifying inventory effects. The null hypothesis is that there are no effects on the prices at which the bank transacts. The interpretation of this hypothesis is that a finding of no inventory effects is consistent with an efficient market: banks can substitute expected funds for current balances and make loans even with low current balances late in the day. This suggests that banks have strong expectations for their ability to borrow later in the day. As shown in Figure 7.2, which plots the 5th and 95th percentile of the excess of the average federal funds rate (calculated as the average federal funds rate in the particular minute across the days of the sample period) over the target federal funds rate, it is clear that the very sparse trading in the early morning hours leads to greater variability in observed trades during those periods. The regression also includes a number of calendar and event effects. In particular each regression we report has dummies for the month, for days that precede or follow a holiday, for Good Friday (a day of low activity), for the days of the reserve maintenance period, for the first and last banking days of the month, and for the end of the quarter. Finally, we include dummies for days of

108

J. McAndrews 4.00 5th percentile 95th percentile 3.00

1.00

0.00

Average midpoint target (%)

2.00

1.00

12:30 a.m.

2:30 a.m.

4:30 a.m.

6:30 a.m.

8:30 a.m.

10:30 a.m.

12:30 p.m.

2:30 p.m.

4:30 p.m.

2.00 6:30 p.m.

Time

Figure 7.2 Fifth and 95th percentile of the average federal funds rate minus the target rate (2 October 2001–30 September 2004) (source: Federal Reserve Bank of New York: author’s calculations). Note Values shown are the 5th and 95th percentile of the distribution of the average federal funds rate minus the target rate for each minute calculated over the days of the sample period.

federal funds target rate increases and, separately, one for decreases in the target rate.10 I perform two other tests. First, I conduct the same test described above for the 4.30 to 4.45 time period. As the 4.00 time was chosen somewhat arbitrarily among those times in which federal funds trading is active, I wanted to test whether any findings were robust to changes in the time considered. Second, I also regress the deviations in the midpoint minus target rate on the bank’s deviations in its transactions balance, both at 4.00 and 4.30 p.m. Results The results of these tests are shown in Table 7.1 (all coefficients and standard errors in Table 7.1 have been multiplied by 10,000 to reduce the number of zeros in the Table). First, we find weak results in favor of inventory effects, with the coefficient on the transactions balance variables always negative and significant at the 10 percent level for the 4.00 p.m. estimations. The estimated coefficient is negative but smaller in absolute value, and statistically insignificant, for the


109

Table 7.1 Regression results Dependent variable

Midpoint – target rate (4:00 – 4:15)

Transaction –0.456 balances (0.242)* Deviation from mean transaction balances January February March April May June July August September October November December Pre-holiday Post-holiday Good Friday 1st Thursday 1st Friday 1st Monday 1st Tuesday 1st Wednesday

0.613 (0.73) 0.407 (0.731) –0.116 (0.729) 0.844 (0.739) 1.232 (0.739)* 0 (0) 0.137 (0.749) 0.598 (0.78) 0.787 (0.775) 0.784 (0.729) 1.87 (0.759)** –0.817 (0.741) –3.57 (0.79)*** 4.939 (0.822)*** 0.134 (2.967) 0.435 (0.667) –4.364 (0.68)*** 0.466 (0.685) –3.87 (0.667)*** –2.617 (0.677)***

Deviation Midpoint – from midpoint – target rate target rate (4:30 – 4:45) (4:00 – 4:15)

Deviation from midpoint – target rate (4:30 – 4:45)

–0.075 (0.146) –0.426 (0.248)* 0.633 (0.731) 0.427 (0.731) –0.092 (0.729) 0.846 (0.739) 1.234 (0.739)* 0 (0) 0.13 (0.749) 0.587 (0.78) 0.777 (0.775) 0.779 (0.729) 1.874 (0.759)** –0.817 (0.741) –3.559 (0.79)*** 4.939 (0.822)*** 0.081 (2.967) 0.437 (0.667) –4.366 (0.68)*** 0.473 (0.685) –3.869 (0.667)*** –2.613 (0.677)***

–0.068 (0.146) –1.179 (0.509)** –0.38 (0.518) –1.463 (0.507)*** –0.396 (0.513) –0.826 (0.508) 0.398 (0.521) 0.435 (0.513) –0.286 (0.521) 0.046 (0.524) –0.754 (0.504) 0 (0) –1.031 (0.513)** –3.712 (0.589)*** 3.794 (0.579)*** 5.672 (2.639)** –0.295 (0.481) –4.361 (0.471)*** –1.021 (0.472)** –3.578 (0.469)*** –3.683 (0.476)***

–1.179 (0.509)** –0.379 (0.518) –1.461 (0.507)*** –0.394 (0.512) –0.825 (0.508) 0.4 (0.521) 0.435 (0.513) –0.286 (0.521) 0.047 (0.524) –0.755 (0.504) 0 (0) –1.031 (0.513)** –3.713 (0.589)*** 3.793 (0.579)*** 5.671 (2.639)** –0.295 (0.481) –4.361 (0.471)*** –1.021 (0.472)** –3.577 (0.469)*** –3.682 (0.476)*** continued

110

J. McAndrews

Table 7.1 continued Dependent variable

2nd Thursday

Midpoint – target rate (4:00 – 4:15)

–1.556 (0.691)** 2nd Friday –2.22 (0.669)*** 2nd Monday 0 (0) 2nd Tuesday –2.153 (0.686)*** 2nd Wednesday –0.468 (0.675) First of month 3.987 (0.806)*** Last of month 9.207 (0.818)*** End-of-quarter –3.398 (1.617)** Fed funds target –12.078 increase (2.978)*** Fed funds target 11.512 decrease (1.75)*** Constant 1.525 (0.73)** Observations 1640 Adjusted R2 0.21

Deviation Midpoint – from midpoint – target rate target rate (4:30 – 4:45) (4:00 – 4:15) –1.555 (0.691)** –2.217 (0.67)*** 0 (0) –2.157 (0.686)*** –0.466 (0.675) 3.986 (0.806)*** 9.214 (0.819)*** –3.401 (1.617)** –12.101 (2.979)*** 11.533 (1.75)*** –145.181 (0.722)*** 1640 0.21

–1.381 (0.468)*** –1.934 (0.488)*** 0 (0) –2.927 (0.479)*** –2.101 (0.475)*** 6.458 (0.526)*** 8.382 (0.581)*** –0.276 (1.082) –2.746 (1.879) 16.586 (1.868)*** 3.108 (0.507)*** 2648 0.23

Deviation from midpoint – target rate (4:30 – 4:45) –1.381 (0.468)*** –1.933 (0.488)*** 0 (0) –2.927 (0.479)*** –2.101 (0.475)*** 6.459 (0.526)*** 8.384 (0.581)*** –0.275 (1.082) –2.746 (1.879) 16.588 (1.868)*** –139.205 (0.506)*** 2648 0.23

Notes Standard errors in parentheses: * significant at 10%; ** significant at 5%; *** significant at 1%. All coefficients and standard errors reported at e + 4 level.

4.30 p.m. regressions. The estimated coefficients are not particularly significant in an economic sense for the 4.00 p.m. estimate. Column 1 suggests that a doubling of the average transactions balance (in absolute value) would lower the midpoint minus target rate by about 5–18 percent (using estimated coefficients from Column 1 or Column 2, respectively). But the average midpoint minus target rate is itself very small, with the constant in the regression (from Column 1) being 0.015 basis points. These results then suggest that the federal funds market appears quite efficient on average at the 4.00 and 4.30 p.m. times. The result on the calendar and event variables also point to some interesting microstructure stylized facts. As found in Griffiths and Winters (1995), Hamilton (1996), Lee (2003) and Demilralp et al. (2004), most of the reserve maintenance period days are significant. On both Fridays, and the last Tuesday of the maintenance period, the banks in the sample post lower midpoints over target


111

rates. As the other authors have suggested, especially for the last Tuesday, that may reflect a desire to avoid being ‘locked-in’ with excess reserves at the end of the maintenance period. None of the monthly effects are consistently significant. Good Friday is insignificant. The pre-holiday and post-holiday dates, and the first and end-of-the-month days are significant. The pre-holiday is correlated with lower rates, but the other days, all days of typically high payment volume are correlated with higher rates. The end-of-quarter days are significant at the 5 percent level, and although they are typically high payment days, they are associated with marginally lower rates. Most notable, perhaps, are the estimated effects of increases and decreases in the federal funds target rate. On days on which the target rate increased the midpoint spread over the target rate was approximately three basis points lower for an increase in the target rate of 25 basis points. Similarly a decrease in the target rate was correlated with a similar size increase in the spread over the target rate. These variables are the largest in terms of economic significance in the estimation. It appears that the market, after controlling for all of the calendar effects and balances, is slightly softer after adjusting to a new, higher, target rate. These results suggest that the market for federal funds displays relatively slight calendar and inventory effects, in terms of economic significance. Of somewhat more economic significance are the effects that changes in the target rates have on behavior in the market, with transaction rate spread midpoints actually moving downward slightly, relative to the target rate, on days of rate increases. These slight effects suggest that on average the market is quite efficient in that it would be hard to profit from these spreads. These tests are weak in the sense that they measure current balances only during a short time period during the day. Other possible tests would be to measure the change in a bank’s balance (excluding all of its fed funds transactions) over the course of the day, or over some lengthy period during the day. However, the test reported here gives a view of the efficiency of the market at a particular time late in the trading day; this might be a good measure of the view that traders have of their ability to obtain funds in the market later in the day.

Discussion: market microstructure of payments and current trends Market microstructure gives us a lens through which we can view payment systems and developments in payment systems. Consider the federal funds market. This market is decentralized and operates through brokers and directly as banks use telephones to contact counterparties. It settles and convenes in the context of an RTGS system. A large number of participants can enter the market if they wish.11 The reserve accounting procedures appear to have slight, but significant, effects on participants’ behavior. Now imagine how the market would differ if trades were settled only in a net settlement system (DNS). One could imagine many different outcomes, depending on how early participants were informed of their net obligations. Suppose

112

J. McAndrews

that participants were informed of their net obligations in the payment system only shortly before the designated time of settlement. Furthermore, suppose that positions of all parties were transparent to all participants. Then, depending on the distribution of the positions, those on one side of the market might face a credit squeeze if the other side of the market is quite concentrated. Alternatively, the presence of a net settlement payment system might encourage the market participants to develop a more centralized call market for the money market. These remarks are intended to focus attention on the money market effects of payment system design. Whether a payment system is a DNS, an RTGS, or a limit payment order system has implications for the money market that utilizes the payment system as its settlement system. Continuous auction money markets are naturally associated with RTGS systems, although it is not necessary that the settlement must be continuous. In fact, it is likely that the causation runs the other way in that RTGS systems, with binding constraints on intraday overdrafts, may require continuous trading in money markets to assist banks in maintaining the level of their balances in a desirable range. In an RTGS then, it is not surprising that many banks both buy and sell money market loans, as they utilize the overnight money market to assist in their intraday money management. This realization in turn suggests that, as technology improves, net settlement of the next day’s money market loans might be possible and desirable, to reduce the outstanding dues to and dues from in the banking system. Looking at current trends in the payment system, the development of limit payment order systems is certainly a notable trend, as discussed in Bank for International Settlements (2005). The improvements in the technology for communication and computation that make these technologies feasible and economical for payment system participants continue. It is likely therefore that these systems will continue to develop. How will the money market be affected? As in other forms of trading, one can imagine the development of electronic limit order books for overnight money markets.12 It is possible to imagine that the liquidity suppliers post interest rates, perhaps through brokers, and banks that need to make payments then ‘cross’ or ‘hit’ the offer of funds. Such a market development could assist banks in planning their liquidity needs at different times, and may lead to a smoother method for providing liquidity for payments. The development in other securities markets, including stock markets, of electronic limit order book systems has led to a voluminous debate over the merits of floor-based trading systems and electronic limit order books and between the benefits of transparency of the ‘book’ of limit orders, and lack of transparency. As summarized by Biais et al. (2004) most theoretical analyses of transparency suggest that it reduces adverse selection in markets. However, empirical analyses suggest some mixed results. Biais et al. (2004) suggest this may be because transparency may make it more difficult for large traders to supply liquidity. The analysis of Willison (2005) suggests that transparency in limit payment order systems is not as crucial as the common knowledge of the limit order priority system itself.


113

Payment systems have also become somewhat more tied together in recent years, both within a currency area, as the development of the Target system in Europe shows, and across currencies, as the development of CLS Bank International shows. These links have assisted banks in managing their positions in different currencies and in different countries. It is likely that, as banks continue to pursue international opportunities, payment system participants – banks and central banks – will continue to work to improve and develop these links, similarly to the way that securities markets in both Europe and the United States have done so.

Summary The microstructure approach to the study of payment systems views payment systems as analogous to security settlement systems. Payment system designs have analogues in the designs of securities settlement systems. Similarities in the economics of payment systems and security settlement systems suggest that further study of payment systems, borrowing from the financial market microstructure literature, would be fruitful. The tests for inventory, calendar, and event effects in the US federal funds market suggested that there are only slight effects of these variables on the midpoint of the spreads of federal funds loans above the target rate. Furthermore the spread above the target rate was very small, indicating that the market is quite efficient on average, and that the open market desk is quite accurate in its operations. Notable developments in payment systems include the recent adoption of ‘limit payment order’ systems, and the linking of payment systems through Target and CLS Bank International. These developments are similar to developments in securities markets, and are likely to be pursued further. The study of market microstructure in financial markets has been greatly affected by the increasing presence of high-quality, high-frequency data. Recent work by central banks in simulating the performance of payment systems has revealed that central banks have been maintaining data on payment system activity. This is a first step in organizing data that can be quite useful in studying the market microstructure of money.

Notes 1 I wish to thank Adam Ashcraft, Bruno Biais, and Xavier Freixas, and the participants at the Future of Payments Conference at the Bank of England for helpful comments. I thank Kurt Johnson for excellent research assistance. The views expressed in this chapter are those of the author and do not necessarily reflect the views of the Federal Reserve Bank of New York or the Federal Reserve System. 2 This rough characterization does not fully specify a DNS as it does not describe the rules of the DNS in the event of a default. See Bank for International Settlements (1989, 1990) for more complete descriptions of DNS systems. 3 If there are not sufficient balances on account, a payment order may be queued, pending the arrival of additional balances, or rejected. See Bank for International Settlements (1997) for a full discussion of various alternatives and implementations of RTGS systems.

114

J. McAndrews

4 From the RTGS plus website of the Deutsche Bundesbank (surveyed April 20, 2005): www.rtgsplus.de/en/leistungsumfang/limitsteuerung/inhalt_e.htm. 5 See McAndrews and Trundle (2001) and Bank for International Settlements (2005) for more detailed descriptions of the range of alternatives for these payment systems. 6 Again, I confine myself here to ‘pure’ payment systems, and ignore delivery-versuspayment systems in which both a security and account balances are exchanged simultaneously. 7 See Regulation D, 12 CFR 204, Board of Governors of the Federal Reserve System. 8 The method for identifying federal funds purchases from the Fedwire transactions journal was pioneered by Furfine (1999), and applied by Demilralp et al. (2004), as well. The method I use is very similar to Demilralp et al. (2004) ‘N-to-N’ method. We search for payments from a bank to its counterparty on day t that is in round values of $100,000, and a matching return payment on day t + 1 that is slightly larger than the first payments, and within a reasonable range given federal funds rates reported by brokers for that day. Specific details are available from the author. 9 On May 22, 2004, the Fedwire operating day lengthened from its previous hours of 12.30 a.m. to 6.30 p.m. to 9.30 p.m. the previous day to 6.30 p.m. We do not include the hours of 9.30 p.m. to 12.30 a.m. in this chart. 10 We note that open market operations are usually conducted in the late morning, well before the 4.00 p.m. time of our tests. 11 Demilralp et al. (2004) have a good discussion of who is eligible to participate in federal funds trading. 12 See www.e-mid.it/index.php for a description of the screen-based money market platform.

References Admanti, A. and Pfleiderer, P. (1988) ‘A theory of intraday patterns: volume and price variability’, Review of Financial Studies, 1: 3–40. Amihud, Y. and Mendelson, H. (1980) ‘Dealership market: market-making with inventory’, Journal of Financial Economics, 8: 32–53. Angelini, P. (1998) ‘An analysis of competitive externalities in gross settlement systems’, Journal of Banking and Finance, 22: 1–18. Bank for International Settlements (1989) Report on Netting Schemes (Angell Report), Committee on Payment and Settlement Systems Publication No. 2. Bank for International Settlements (1990) Report of the Committee on Interbank Netting Schemes of the Central Banks of the Group of Ten Countries (Lamfalussy Report), Committee on Payment and Settlement Systems Publication No. 4. Bank for International Settlements (1997) Real-time Gross Settlement Systems, Committee on Payment and Settlement Systems Publication No. 22. Bank for International Settlements (2005) New Developments in Large-value Payment Systems, Committee on Payment and Settlement Systems Publication No. 67. Bech, M. and Garratt, R. (2003) ‘The intraday liquidity management game’, Journal of Economic Theory, 109: 198–219. Biais, B., Glosten, L., and Spatt, C. (2004) ‘Market microstructure: a survey of microfoundations, empirical results, and policy implications’, IDEI Working Paper. Demiralp, S., Preslopsky. B., and Whitesell, W. (2004) ‘Overnight interbank loan markets’, Board of Governors of the Federal Reserve System Finance and Economics Discussion Series. Freixas, X. and Parigi, B. (1998) ‘Contagion and efficiency in gross and net interbank payment systems’, Journal of Financial Intermediation, 10: 3–31.


115

Furfine, C. (1999) ‘The microstructure of the federal funds market’, Financial Markets, Institutions and Instruments, 8(5): 24–44. Furfine, C. (2003) ‘Interbank exposures: quantifying the risk of contagion’, Journal of Money Credit and Banking, 35: 111–28. Garman, M. (1976) ‘Market microstructure’, Journal of Financial Economics, 3: 257–75. Goodfriend, M. and Whelpley, W. (1993) ‘Federal funds’, in T.Q. Cook and R.K. Laroche (eds) Instruments of the Money Market, Richmond, VA: Federal Reserve Bank of Richmond. Griffiths, M. and Winters, D. (1995) ‘Day-of-the-week effects in federal funds rates: further empirical findings’, Journal of Banking and Finance, 19: 1265–84. Hamilton, J. (1996) ‘The daily market for federal funds’, Journal of Political Economy, 104: 26–56. Hamilton, J. (1997) ‘Measuring the liquidity effect’, American Economic Review, 87: 80–97. Hasbrouck, J. and Sofianos, G. (1993) ‘The trades of market makers: an empirical analysis of NYSE specialists’, Journal of Finance, 48: 1565–93. Ho, T. and Stoll, H. (1981) ‘Optimal dealer pricing under transactions and return uncertainty’, Journal of Financial Economics, 9: 47–73. Ho, T. and Stoll, H. (1983) ‘The dynamics of dealer markets under competition’, Journal of Finance, 38: 1053–74. Kahn, C., McAndrews, J., and Roberds, W. (2003) ‘Settlement risk under gross and net settlement’, Journal of Money Credit and Banking, 35(4): 591–608. Kahn, C. and Roberds, W. (1998) ‘Payments system settlement and bank incentives’, Review of Financial Studies, 11: 845–70. Lee, Y.S. (2003) ‘The federal funds market and the overnight eurodollar market’, Journal of Banking and Finance, 27: 749–71. Lyons, R. (1995) ‘Tests of microstructural hypotheses in the foreign exchange market’, Journal of Financial Economics, 39: 321–51. Madhavan, A. and Smidt, S. (1993) ‘An analysis of changes in specialist inventories and quotations’, Journal of Financial Economics, 30: 99–134. Manaster, S. and Mann, S. (1996) ‘Life in the pits: competitive market making and inventory control’, Review of Financial Studies, 9: 953–75. McAndrews, J. and Rajan, S. (2000) ‘The timing and funding of Fedwire funds transfers’, Federal Reserve Bank of New York Economic Policy Review. McAndrews, J. and Trundle, J. (2001) ‘New payment system design: causes and consequences’, Bank of England Financial Stability Review. Pagano, M. (1989). ‘Trading volume and asset liquidity’, Quarterly Journal of Economics, 104: 255–76. Pagano, M. and Roell, A. (1996) ‘Transparency and liquidity: a comparison of auction and dealer markets with informed trading’, Journal of Finance 51: 579–611. Skeie, D. (2004) ‘Money and modern bank runs’, unpublished thesis, Federal Reserve Bank of New York. Stigum, M. (1990) The Money Market, Homewood, IL: Dow Jones-Irwin. Stoll, H. (1978) ‘The supply of dealer services in securities markets’, Journal of Finance, 33: 1133–51. Vayanos, D. (1999) ‘Strategic trading and welfare in a dynamic market’, Review of Economic Studies, 66(2): 219–54. Willison, M. (2005) ‘Real-time gross settlement and hybrid payment systems: a comparison’, Bank of England Working Paper No. 252.

Part III

Current payment policy issues

8

Wholesale payments Questioning the market-failure hypothesis George Selgin1

The kind of situation which economists are prone to consider as requiring corrective governmental action is, in fact, often the result of governmental action. Coase (1988: 133)

Introduction When a bank, acting on its own behalf or that of a client, needs to transfer a large sum, say $10 million, to another bank, the transfer will almost certainly be made through a ‘wholesale’ payments system. Wholesale payments systems receive payments messages from sending banks, relay them to receiving banks, and oversee the final settlement of interbank accounts. Two basic kinds of wholesale payments systems exist. A Deferred Net Settlement (DNS) system gathers payment orders throughout the business day (or some other preset period), calculates end-of-day multilateral net obligations, and then arranges for the transfer of reserves (‘good funds’) from net senders to net recipients of funds. A Real-Time Gross Settlement (RTGS) system executes payment orders as they arrive, at once transferring reserve credits representing the gross value of individual payments. The strictest Real-Time Gross Settlement (RTGS-GF) systems require their members to possess clearing balances or ‘good funds’ sufficient to cover all payments, while less strict versions (RTGSDO) allow their participants to rely on intraday credits or ‘daylight overdrafts’ to cover payments in excess of their available balances, on the understanding that the credits must be repaid at day’s end. An RTGS-DO arrangement combines the continuous settlement feature of RTGS-GF with at least some of the reserve efficiencies found in DNS systems. However, unless intraday credits are provided free of charge, a sequence of wholesale payments administered by either sort of RTGS system will give rise to a greater demand for bank reserves than an identical sequence of payments using DNS. This is one reason why DNS has historically been the preferred means for handling both retail (small value) and wholesale transfers. The 1990s, however, witnessed a remarkable change in wholesale payments arrangements, with traditional DNS arrangements giving way to RTGS systems,

120

G. Selgin

and especially to RTGS-DO systems. Government monetary authorities, including the Fed, have actively promoted the change both by recommending or mandating changes in private wholesale payment systems and by competing against private networks with their own RTGS systems. Government promotion of RTGS has typically been defended on the grounds that traditional net settlement systems are beset by serious externalities that could have catastrophic consequences. Here, I critically assess market failure arguments for reforming or abolishing traditional DNS systems, showing that these arguments have been based, not on empirical evidence or on careful consideration of actual DNS procedures, but on a spurious analogy drawn between DNS systems and Fedwire, the Federal Reserve’s own RTGS-DO system. While there are good reasons for thinking that Fedwire does indeed harbor a serious externality problem, or at least that it did so prior to reforms initiated in 1994, the frequently made claim that unregulated DNS systems suffered from an analogous but independent externality problem overlooks crucial differences between Fedwire and traditional DNS arrangements, including CHIPS prior to the 1990s, and especially the different ways in which each generates and assigns intraday credit risk. My position is not that traditional DNS arrangements were trouble-free. Nor do I deny that such arrangements may have permitted excessive risk taking. However, I argue that, to the extent that risks were excessive, the cause was not market failure but implicit guarantees extended by regulatory authorities themselves, which tended to corrupt otherwise sound market-based arrangements. My modest aim is to show how the literature on wholesale payments has tended to confuse regulatory failure with market failure, thereby diverting attention from potential first-best solutions to alternatives that may not even qualify as second-best.

The trouble with Fedwire At the end of the 1990s, the United States was the only industrialized nation that relied heavily upon both DNS and RTGS arrangements to handle wholesale payments. The Clearing House Interbank Payment System (CHIPS) was a private DNS system operated by the New York Clearing House.2 Fedwire was (and remains) a RTGS-DO system operated by the Federal Reserve. Each system handled about $1.5 trillion in wholesale transfers annually. The coexistence of these two arrangements, despite superior reserve-holding economies generally realized through net settlement, was largely due to the Fed’s willingness to supply sending banks with low-cost daylight overdrafts. Prior to 1994, the Fed supplied intraday credit routinely and free of charge, effectively matching the reserve economies of net settlement. While the Fed allowed sending banks to overdraw their accounts to execute payments, it also provided for immediate settlement by crediting the accounts of receiving banks by the full amount of gross payments. The Fed guaranteed the ‘finality’ of these payments, meaning that it could not have recourse to credited

Wholesale payments

121

accounts if a sending bank with an overdrawn account failed to settle up with it by the end of the business day. The risks associated with Fed extensions of intraday credit, including the risk of settlement failure, were therefore borne approximately by the Federal Reserve System (which might find itself holding Fedwire-generated claims on a failed institution) and ultimately by the general public. Thus, until 1994 at least, Fedwire suffered from a potentially serious flaw, in that it deprived both receiving banks and their customers (the ultimate recipients of payments) of any strong incentive to monitor sending banks or to limit their acceptance of payments orders sent to them. It also relieved senders of any reason to fear that, in selecting unsound banks as their agents, they would remain liable for promised payments in the event that their banks failed to settle with the Fed. The result was a serious moral hazard problem, with private agents initiating transactions that exposed third parties to credit risk. One way out of this moral hazard was to have the Fed charge a fee on its intraday credits sufficient to cover the risk of non-payment. Although the Fed acknowledged the need for some such remedy during the 1980s, when the volume of Fedwire transactions was increasing rapidly, only in April 1994 did it begin charging a marginal daylight overdraft fee of 24 basis points (annual rate), which was increased to 36 basis points a year later.3 This reform led to a significant reduction in measured Fedwire overdrafts (Richards, 1995). Nonetheless, some experts maintain that the Fed’s current intraday lending rates, which are levied only on overdrafts in excess of 10 percent of a bank’s risk-based capital, and which are waived if they sum to less than $25 in two weeks, may still involve some under-pricing of settlement risk.4 In fact, the Federal Reserve Board had originally intended to raise the Fed’s overdraft fee above 36 basis points, but changed its mind in part because it feared that such a move might prompt a substantial shift of wholesale payments volume to private payments networks, including CHIPS (Richards, 1995: 1068).

DNS: guilt by association Fed officials hesitated to implement any Fedwire reforms that might have sponsored a substantial shift in wholesale payments activity from Fedwire to CHIPS. Although bureaucratic turf-preserving and budget-maximizing behavior might account for this hesitation, it appears to have been due at least in part to officials’ belief that CHIPS’ traditional operating procedures were no less in need of reform than Fedwire’s had been. Their reasoning, which is shared by many monetary authorities across the globe as well as by some economists, goes as follows. In a traditional, ‘unsecured’ DNS system, like CHIPS before 1990, payment orders are, essentially, binding IOUs that come due at settlement time.5 Therefore, if at any time of the day system participant A has received payment orders from participant B exceeding by $X the value of such orders B has received from A, A may be said to have granted B $X of ‘intraday credit.’ (Some authorities even claim that sending banks in a traditional DNS system

122

G. Selgin

routinely rely on ‘daylight overdrafts’ on their clearinghouse accounts, implying that intraday credit is granted, not by receiving banks, but by the clearinghouse itself.) DNS-system intraday credit is, moreover, provided free of any nominal or explicit charge: just like Fedwire’s pre-1994 intraday credits. DNS-system participants have no other (non-price) means for controlling the volume of intraday credits they grant one another (or, as some would have it, obtain from the clearinghouse), except by rejecting payments orders altogether. A free-rider problem makes it less than worthwhile for receiving banks to go to the trouble of finding out whether a sending bank might be unworthy of an intraday loan. DNS systems therefore suffer from the same moral hazard problem present in Fedwire. The tendency of regulators and economists to treat traditional or ‘unsecured’ DNS systems as close cousins of Fedwire, with its reliance upon free or underpriced intraday credits or ‘daylight overdrafts,’ is evident throughout the literature on wholesale payments: •

•

•

•

•

•

Board of Governors of the Federal Reserve System (1988: 7): ‘CHIPS participants [that] have initiated transfers with a total dollar value greater than that of the transfers they have received . . . are essentially receiving intraday credit from the participants that have received transfers with a total value higher than that of transfers they have sent.’ Rochet and Tirole (1996: 840): ‘[I]n net payment systems, intraday overdrafts do not appear explicitly and are therefore necessarily free’; ‘Since its inception, CHIPS has operated with explicit [sic] intraday overdrafts.’ Folkerts-Landau (1997: 5): ‘Netting arrangements . . . expose the participants to credit risks as they extend large volumes of payments-related intraday credit to each other.’ Kahn and Roberds (1999: 30): Large-value payment systems ‘have traditionally enjoyed access to significant amounts of virtually free intraday credit. . . . In net settlement systems, such credit is granted when a bank accumulates a large net debit position vis-à-vis other banks in the system.’ Roberds (1999: 2): ‘The principal disadvantage of a net settlement system is that the central counterparty . . . ends up bearing most of the credit risk and liquidity risk associated with the settlement of payment obligations.’ Mengle et al. (1987: 7): ‘On private net settlement networks such as CHIPS, the [intraday credit] supply curve faced by system participants diverges from the supply curve reflecting risks to society the same as occurs on Fedwire.’

Using such arguments regulators have claimed, not only that unsecured DNS arrangements involve externalities comparable to those found in Fedwire, but also that DNS-system externalities pose a different and perhaps greater hazard. This last conclusion stems from beliefs concerning the different consequences of a bank’s failure to settle in the two kinds of systems. Under Fedwire, as we have seen, if a bank fails to pay-off its intraday debts to the Fed, settlement proceeds

Wholesale payments

123

regardless, with the Fed footing the bill. (Meanwhile the defaulting bank may be granted a discount-window loan, which unlike other central bank loans may have to be fully collateralized.) In a traditional DNS system, in contrast, contractual agreements would generally call for a payments ‘unwind.’ An unwind commences with cancellation of all of the previous settlement period’s payment orders to and from the failed institution. Remaining multilateral net positions are then re-calculated, and settlement is once again attempted. Should one or more other banks then find themselves unable to settle owing to a deterioration of their net positions, payments messages to and from those banks are also cancelled. The process continues in this fashion until surviving banks are able to settle. In principle, the failure of a single DNS-system participant could cause many other participants to default. Just how great was the risk of a DNS-system participant failing unexpectedly, and how extensive would the fallout from such a failure have been in practice? The empirical record supplies only negative evidence: there had never been an actual settlement failure on CHIPS or any other important DNS wholesalepayments network.6 Economists and policy-makers therefore had to rely on simulations to assess the likely consequences of a DNS-system unwind. According to Humphrey’s (1986) influential simulation, the failure of a major CHIPS participant, given rules in place at the time, might have caused dozens of large banks to fail, triggering a system-wide crisis. Fear of such a catastrophe led regulators to treat traditional DNS systems as being especially in need of regulation, or (an increasingly popular option) of replacement by some form of RTGS. The Bank of Japan, which for many years offered both DNS and RTGS services to Japanese banks, switched to offering RTGS only in the late 1990s, despite Japanese financial firms’ apparent preference for the DNS alternative: just prior to the change, only about 3 percent of Japan’s wholesale payments, measured in value terms, were handled by the Bank of Japan’s RTGS system. The United Kingdom converted CHAPS – its counterpart to CHIPS – into an RTGS system in 1996. Central bank administered wholesale payments systems throughout the rest of the EU have, in response to BIS recommendations, operated on a real-time gross basis since 1997. The Reserve Bank of New Zealand switched to RTGS in March 1998, and in 2000 the Canadian Payments Association established, at the Bank of Canada’s urging, a hybrid Large Value Payment System that relies on the Bank of Canada as a source of collateralized intraday loans and as an ultimate guarantor of end-of-day settlement. CHIPS itself, finally, was converted into a hybrid system offering near continuous and irrevocable settlement for executed payments in January 2001.7 Traditional DNS systems have thus been largely abolished, despite having long been generally favored by parties engaged in wholesale payments.8

‘Mirage’ externalities The wholesale payments system reforms of the 1990s are supposed to have addressed excessive risk taking, and excessive risk of payment unwinds in

124

G. Selgin

particular, stemming from externalities inherent in unregulated DNS systems. However, a close look at traditional DNS-system contractual arrangements suggests that the externalities that are supposed to have given rise to an inefficient risk–return trade-off were not genuine market failures at all, but market failure ‘mirages’ that tend to appear when DNS systems are viewed through Fedwirecoloured glasses. Putting those glasses aside allows us to get an undistorted picture of traditional, voluntary contractual arrangements of commercial banks and private clearinghouses, where by ‘voluntary’, I mean free from pressure, edicts, or (subsidized) competition from government agencies. As noted earlier, the overwhelming historical tendency has been for financial firms to establish clearinghouses that rely on unsecured DNS to handle both small- and largevalue payments. In the absence of government involvement, clearinghouses have generally been bankers’ associations or ‘clubs’ (Dowd, 1994). Among their organizational features, the following are especially relevant: •

•

•

•

Because the clearinghouse is not a bank, it receives payment orders and calculates net multilateral positions, but (unlike a central bank) does not maintain account balances for its members. Because clearinghouse members do not maintain accounts with the clearinghouse, ‘daylight overdrafts’ in the strict meaning of the term play no part in traditional DNS systems, or play only an indirect part (as when net settlement transfers are administered through a RTGS-DO system). Nor do private clearinghouses normally provide intraday credit in any form to their members. Such clearinghouses therefore do not usually assume any payments-related risk.9 Intraday DNS payments have traditionally been provisional payments only. Private agents contract for ‘check’ rather than ‘receiver’ finality in both retail (check) and wholesale (wire) payments. Check finality means that a bank receiving a payment order is under no obligation to release funds to the payee until end-of-day settlement is complete. Moreover, should the bank grant the payee access to funds prior to settlement, the funds in question generally remain revocable in the event of a settlement failure. In legal terms, the ‘acceptance’ of payment orders by receiving banks, which renders them liable to payees for the amount of the orders (while in turn obliging sending banks to pay them) is conditional upon settlement.10 Because acceptance of payments orders in a traditional, unsecured DNS system is conditional upon settlement, a settlement failure in such a system renders orders sent by a failed participant null and void. The effect is as if the orders in question had never been sent. This is the essence of the ‘unwinding’ process. It is therefore misleading to speak of receiving banks in DNS systems providing sending banks with ‘intraday credits’: in a traditional DNS system, payment orders are not legally binding IOUs; they are properly regarded as non-binding pledges only.11

Wholesale payments

125

Taken together these points imply that the only ‘intraday credits’ actually extended in traditional DNS systems are credits granted by receiving banks to their own customers when they grant beneficiaries immediate access to funds. No basis exists, therefore, for claiming that receiving banks in DNS systems are – in the absence of government interference with private contracts – unable to effectively regulate their exposure to intraday credit risk, or that such risk is an external or third-party consequence of transactions undertaken, not by receiving banks, but by sending banks and their clients. Finally, because the intraday credits to beneficiaries can generally be revoked in the event of a settlement failure (the one circumstance in which credits might conceivably need to be repaid) risk is borne not primarily by receiving banks but by the beneficiaries themselves and also by payment originators who remain obliged to the beneficiaries even though their banks’ failure may have deprived them of access to their accounts.12 This bearing of risk by payment senders and beneficiaries, as opposed to some externality, may account for the extension of intraday credits free of charge in unsecured DNS systems. To insist that unsecured DNS systems involved underpriced intraday credit ‘the same as occurs [or occurred] on Fedwire,’ while simultaneously finding fault with such systems because they expose payment senders and beneficiaries to the possibility of a payments unwind, is self-contradictory.13 The use of terminology taken from Fedwire (or other RTGS-DO systems) to describe DNS arrangements has thus caused economists and regulators to overlook features of deferred settlement aimed at containing and controlling payment-related risks in a cost-effective way. By making the acceptance of payment orders conditional only, pending successful settlement in good funds, receiving banks in private DNS systems avoid having to monitor sending banks – a burden that would be great enough in a small DNS system such as CHIPS, and impossible in a network with as many participants as Fedwire. Receiving banks can instead concentrate on keeping informed of the credit worthiness of their own customers – the beneficiaries of wire transfers – in order to decide whether to release funds to them prior to settlement. Bankers are, presumably, capable of assessing the credit-worthiness of their own account holders. In practice, receiving banks in traditional DNS systems often grant immediate credit to customers. This tendency reflected customers’ apparent creditworthiness as well as receiving banks’ low estimate of the probability of a settlement failure. The low perceived risk of a settlement failure was no doubt informed by the extreme rarity of actual settlement failures in traditional DNS systems, which was partly a consequence of careful up-front selection (by means of capital and liquidity standards and the like) and monitoring of clearinghouse members. The presumption that a settlement failure was highly unlikely may also have been informed at least in part by the belief that regulators would intervene rather than allow such a failure: a possibility considered below. Payment beneficiaries nevertheless retained an incentive to limit their use of revocable funds, and to insist that funds be sent to them through banks that they considered sound. Finally, payment originators had reason to select sending banks carefully,

126

G. Selgin

knowing that a sending bank’s failure prior to settlement would leave beneficiaries’ claims against them intact, while depriving them of access to most, if not all, of their undelivered bank balances. As Mengle (1990: 158) observed: In wholesale wire transfers, the sender is most likely a corporation, possibly a bank. Given the size of the transfer, it is plausible that senders are of sufficient sophistication to monitor the soundness of the banks with which they do business. Failure of a sending bank is something against which a prudent sender can protect itself.14 That unsecured DNS systems did not involve any inherent externality problem does not necessarily mean that such systems were efficient. Economists’ standard practice is, nonetheless, one of assuming that, unless special reasons can be found for thinking otherwise, ‘parties in privity will contract for the most efficient allocation of risk’ (Scott, 1990: 182). There was nothing to stop receiving banks in private DNS systems from voluntarily agreeing to accept payment orders unconditionally, making irrevocable payments to beneficiaries in anticipation of settlement. But then these banks, unlike receiving banks in Fedwire, would have had reason to adjust their fees to reflect any perceived risk of a settlement failure. Faced with the choice of having to pay receiving banks a fee sufficient to compensate them for bearing such risk, and bearing the risk themselves, payment senders and beneficiaries apparently preferred the latter option, and did so presumably, because they believed themselves capable of controlling or absorbing such risks for less than what receiving banks would charge them for performing those same services. Indeed, the sophistication of large payment originators and beneficiaries makes check finality appear to have been even better suited to satisfying the ‘least-cost avoider’ principle for optimal risk assignment with respect to large-value wire transfers than it has been with respect to check payments.

The systemic risk problem Although individual participants in a traditional DNS system might have been fully in control of credit risks associated with their own intraday lending activities, they may also have been exposed to losses resulting from other participants’ imprudence, which could have exposed them to the diffuse adverse effects of a payments unwind. Did the existence of such ‘systemic’ risk itself justify restricting activity in DNS systems or replacing DNS with RTGS? Strictly speaking, it did not. Although it is true that a bank’s multilateral net position may be adversely affected by failure of another bank even if it had no bilateral transactions at all with the failed bank, such interdependence does not imply a market failure. As Scott (1990: 185) observes: a bank that joins a voluntary payments network enters a contractual arrangement with the network operator and, indirectly, with all other network

Wholesale payments

127

participants. Network members have an incentive to establish rules designed to limit systemic risk generated within the network without having to be encouraged or forced to do so by regulatory authorities.15 Systemic risk is, moreover, not a problem unique to DNS systems. All banks, regardless of the payment systems they rely upon, depend to some extent on payments from other banks to finance their own payments. Participants in a DNS system expose themselves to systemic risk when, upon receipt of a payment order, they credit their customers’ accounts in anticipation of settlement. In an RTGS-GF system, payment orders are settled immediately, or are rejected, so that the orders cannot be said to be the basis for payment expectations that may ultimately be disappointed. But this does not mean that RTGS-GF system participants cannot act upon expectations that are later falsified regarding the flow of good funds through the banking system. For example, a bank might promise to make a payment to a second bank at 16:00 h, anticipating the 15:00 h arrival of a payment promised to it by a third bank. But the 15:00 h payment may never be sent. The same thing can of course happen in an RTGS-DO system. The claim that participants in RTGS systems ‘cannot respond to payments that have not been received’ (Van den Bergh and Veale, 1994: 103) is therefore invalid. It follows that, with regard to systemic risk, the difference between RTGS and DNS is a difference of degree rather than a difference in kind; and this difference may be warranted in view of the much higher liquidity or collateral costs (or more frequent payment-order rejections and delays) that real-time gross settlement entails.16 Comparing the costs of secured net settlement on CHIPS to those of RTGS, Schoenmaker (1995: 26) concludes that ‘the estimated extra cost of RTGS exceeds the estimated reduction in settlement and settlement risk.’17 What about the potentially catastrophic consequences of a DNS unwind? Are there not sufficient grounds for preferring RTGS to DNS even if the former arrangement is much more costly and does not completely eliminate systemic risk? One reply to this question asks, ‘What catastrophic consequences?’ To repeat: there has never been a settlement failure and consequent unwind on CHIPS or any other major DNS wholesale-payments system; and the chances of a DNS-system participant failing suddenly enough to precipitate a settlement crisis (as can happen if other participants have no inkling of troubles at a participating bank even on the very day on which it fails) can only be judged remote.18 Of course, were an unwind to have truly catastrophic consequences, even a tiny risk might be considered unacceptable, and especially so in light of the possibility that market participants might underestimate or choose to ignore risks that they are scarcely able to contemplate. Research in cognitive psychology and behavioral economics suggests that, contrary to the predictions of orthodox rational choice theory, individual inferences concerning extreme events can exhibit systematic biases. However, the same research offers no consistent prediction concerning the direction of the bias, and therefore supplies little guidance for the formulation of public policy (Gerson, 2001). Private markets do

128

G. Selgin

supply catastrophe insurance, after all. Moreover, the assumption that an unwind is likely to prove catastrophic is itself open to serious doubts. That assumption has been based largely on the results of Humphrey’s (1986) simulation, which itself assumed that CHIPS participants could have had no recourse at all to beneficiaries’ account balances following cancellation of failed banks’ payment messages. This is tantamount to assuming that every beneficiary of an intraday DNS payment draws its account balance down to zero prior to settlement time, which is unrealistic as well as question-begging. Intraday payment beneficiaries in DNS systems function like lenders of last resort, except that, instead of augmenting their banks’ liquidity by supplying them with last-minute reserves, they allow the banks to make last minute, negative adjustments to the beneficiaries’ deposit credits. If beneficiaries’ end-of-day account balances were always at least equal to their provisional intraday credits (the opposite of Humphrey’s assumption), a payments unwind could never cause a bank to fail even in a banking system bereft of capital. Reality falls between the two extremes, with end-of-day beneficiary balances offering some protection against settlement failure.19 Provisional payment clauses would never have arisen in the first place had they not been capable of re-directing risk, if only to a limited extent. Nor would they still exist, despite regulations aimed at making unwinds less probable than ever, if banks could not bring themselves to contemplate the possibility of a settlement failure. In general, the more frequent the clearing-and-settlement sessions in a DNS system, the less likely it is that any participant will fail before its payment orders are rejected by the system. RTGS can thus be viewed as a limiting case in which settlement occurs with each and every payment. The predominance of daily clearings in past DNS systems suggests that such clearings typically sufficed to reduce the odds of an unwind to tolerably low levels. This interior solution to the ‘optimal frequency of settlement’ problem – a solution informed by decades of experience – stands in stark contrast with the ad hoc corner solution of realtime gross settlement favored by regulatory authorities. Besides exaggerating the likelihood and adverse consequences of a DNS unwind, regulators have also tended to overlook certain advantages that an unwind rule offers relative to an alternative rule guaranteeing intraday finality of payments. Consider, for example, how a large (uninsured) deposit holder might respond under each rule to a rumor that his bank is going to fail later that day. Under guaranteed finality, the depositor can rescue most of his balance by wiring funds to another bank, thereby shifting default risk from himself to the payment system. Under provisionality, in contrast, the same depositor would have no choice but to run on his bank for cash, knowing that the failure of his bank before the end of the day will result in the cancellation of any payment messages sent by it. Although a run to currency may seem more disorderly than a wire transfer ‘run’ to other banks, it does not necessarily expose third parties to default risk.20 In this respect at least, an unsecured DNS arrangement is more incentive-compatible than a secured one, and also more incentive-compatible than RTGS systems with intraday credit, including Fedwire.21

Wholesale payments

129

Government, not market, failure Although no convincing grounds exist for the claim that unregulated DNS systems suffer from inherent market-based imperfections, good reasons exist for holding regulatory authorities themselves responsible for undermining the safety of wholesale payments arrangements that might otherwise be expected to achieve an efficient and acceptable risk–return trade-off. The most important source of ‘government failure’ in past DNS systems consisted of central-bank-provided finality guarantees.22 Central bankers’ empirically unsupported views concerning the likelihood of systemic failures, and bureaucratic or political considerations as well, have inclined them to bail out banks – and big banks especially – that might otherwise have been unable to meet their net settlement obligations. In the past such guarantees, if they existed at all, tended to be implicit only. But some recent DNS-system reforms, based upon the misguided policy desideratum of zero unwind risk, have made them explicit. Whether implicit or explicit, government-based finality guarantees undermine private incentives to monitor and control payments-related risks in DNS arrangements, including receiving banks’ incentive to limit beneficiaries’ access to unsettled payments and beneficiaries’ incentive to resist employing advanced funds prior to settlement. The extent to which the routine practice, in CHIPS and other DNS systems, of immediately releasing intraday funds to payment beneficiaries was encouraged by the presence of central bank guarantees (or by the presence of alternative central-bank administered payments arrangements in which intraday finality was provided for less than its true social cost) remains a crucial but as yet unexamined empirical question.23 If excessive risk taking in DNS systems has been due, not to market failure, but to the presence of central-bank guarantees, then DNS systems might be made ‘perfectly’ safe by ending the guarantees. In principle, this requires nothing beyond enforcement of the generally approved but frequently broken ‘classical’ rule limiting last-resort assistance to illiquid but solvent banks.24 The 1991 FDICIA reform limited Fed lending to undercapitalized banks to no more than 60 days within any 120-day period, and sanctions stricter than those included in FDICIA could further discourage Fed lending to insolvent institutions (Kaufman, 1999: 5–6). Alternatively, the Fed and other central banks could be altogether prohibited from providing extended credit to banks. As Kaufman (1999: 6–9) and several other economists have observed, well-organized modern markets for both government securities (or commercial paper) and bank reserves make direct central bank lending to troubled banks unnecessary in most industrialized economies. Central banks need only provide adequate supplies of base money, which they can do by means of open-market operations, leaving to the private market the task of reallocating reserves among solvent banks, perhaps through pre-established lines of credit. In traditional DNS arrangements, solvent banks that found themselves short of funds for settlement routinely relied on interbank loans to tide them over until they were able to replenish their reserves

130

G. Selgin

by liquidating non-reserve assets. CHIPS participants, for example, had up to 1 hour after clearing to acquire needed settlement funds.25 Most monetary authorities are of course unwilling to give up all or part of their power to extend ‘last resort’ loans, a power they regard as essential for containing systemic risk. But where deposits are largely insured, systemic risk has come increasingly to be identified with the risk of a wholesale payments crisis. Monetary authorities have tended, in other words, to argue in a circle, appealing to systemic risk problems originating in their own implicit guarantees as reason for providing those guarantees, while disguising the circularity of their arguments by misrepresenting DNS problems as market failures. In fact, the persistence and prominence of provisionality clauses in bank deposit contracts suggests that participants in DNS systems are, after all, not entirely convinced that their central banks will intervene to prevent a payments unwind.26 Although any positive probability of central-bank support does presumably reduce private parties’ incentives to monitor and control settlement risk, recent reforms, including modified DNS risk-sharing arrangements aimed at ‘securing’ those arrangements while converting provisional intraday payment messages into ‘final’ payments and (in the US) measures aimed at favoring Fedwire over CHIPS, only worsen the implied moral hazard, by replacing uncertain ex post finality guarantees with certain ex ante ones.27 Another source of government failure in DNS systems is the failure of courts to enforce private contracts. In the United States both the Federal Reserve and the courts can override private payment contracts. Prior to 1990 the rights and liabilities of parties to wholesale payment transactions were unclear, particularly with regard to the consequences of a settlement failure, owing to the lack of any case or statutory law addressing this problem (Mengle et al., 1987: 4). The very rarity of unwinds had the ironic effect of generating uncertainty regarding the enforceability of standard unwinding rules and provisionality clauses. Section 4A of the Uniform Commercial Code – the law that governed wire transfers in most of the United States throughout the 1990s – clarifies matters, but did so in part by declaring bank contracts providing for provisionality of wholesale payments to be generally unenforceable! The Code allowed for two exceptions, one to accommodate CHIPS, the other to accommodate the Automated Clearing House (ACH): a government-operated DNS system used mainly for smaller payments. The CHIPS exception, however, allowed its members to revoke payments in the event of a settlement failure only so long as they also took part in a special Lamfalussy-type loss-sharing arrangement that was not a traditional DNS feature.

‘Secured’ net settlement: chipping away at CHIPS Attempts to ‘secure’ DNS systems through regulations aimed at reducing their exposure to systemic risk have been a third source of government failure in wholesale payments. The special risk-sharing provisions first adopted by CHIPS are one of three means now employed on a worldwide basis to achieve such

Wholesale payments

131

security, the other two being ‘caps’ on DNS participants’ net bilateral and multilateral intraday clearinghouse ‘debits.’ Were such special arrangements and restrictions a response to genuine market failures, serving to ‘internalize the cost of third party risk’ (Folkerts-Landau, 1997: 6), they might contribute to the overall efficiency of payments. But when adopted in response to market-failure ‘mirages’ the efforts can undermine efficiency instead of enhancing it. Consider, for example, the effects of bilateral and multilateral debit caps. If it were really true that receiving banks in DNS systems, by failing to reject payment orders, became creditors to banks sending those orders and thereby passively exposed themselves to credit risk, then such restrictions might be the only reliable means (apart from switching to RTGS) for keeping intraday credit exposures within efficient bounds. But we have seen that receiving banks in DNS systems actually expose themselves to intraday credit risk only when they grant payment beneficiaries pre-settlement access to anticipated funds, and then only to the extent that they may be unable to recover funds advanced to those beneficiaries following an actual settlement failure. It follows that bilateral and multilateral caps may serve no purpose other than to alter and restrict artificially the flow of wholesale payments, presumably in a manner that makes such payments less efficient. Nor does the fact that many DNS systems, including CHIPS, who have adopted CAPS voluntarily, necessarily contradict this conclusion. Many ‘voluntary’ restrictions have been adopted only in response to encouragement by central bankers, whose advice may be coupled with implicit threats, including (for example) the threat to deny private DNS systems access to central bank final settlement facilities. Much the same may be said regarding special risk-sharing arrangements that CHIPS and other DNS systems were encouraged to adopt as a substitute for the standard practice of making all payments conditional upon settlement. Under the Lamfalussy provisions for secured DNS, if any DNS participant defaults, surviving participants are required to share responsibility for its net settlement obligations. Ironically, such special risk-sharing provisions can create precisely the sort of externality problem that was originally supposed to justify regulation of DNS systems, but which may not have been present in such systems before they became objects of regulators’ scrutiny (Rochet and Tirole, 1996). Market-failure mirages have thus served as the basis for lifting wholesale payments out of the unwind frying pan and into the moral-hazard fire.28 If attempts to ‘secure’ DNS systems may only have served to render them less efficient (if not more risky) than before, reforms that have abolished DNS systems altogether in favor of RTGS may involve even larger welfare losses. As noted previously, unless supplemented by underpriced intraday credits (which introduce a moral-hazard problem), RTGS imposes much higher liquidity or collateral costs on banks, forcing them to maintain higher average reserve balances, or to incur more liquidity risk, than would be the case under deferred net settlement. Wherever RTGS reforms have been motivated, not by private agents’ voluntary quest for efficiency but by regulators’ desire to correct non-existent market failures, reason exists for suspecting that the reforms have not been efficient.

132

G. Selgin

Conclusion: private contracts versus government policy The widespread view that unregulated private wholesale payment arrangements, and deferred net settlement in particular, promote excessive risk taking has been based upon regulators’ misunderstanding of (1) the manner in which private DNS arrangements generate and assign intraday credit risk and (2) the role that government guarantees have played in undermining otherwise sound privatemarket wholesale payments arrangements. Unlike Fedwire, CHIPS and other traditional DNS arrangements do not rely upon underpriced intraday credits, or rely upon them only to the extent that they must participate in an RTGS-DO system to effect final settlements. The only credits that are directly connected to intraday DNS payments are credits granted by receiving banks to their own customers; and these credits shift settlement risks to third parties only in so far as they are backed by extra-market finality guarantees. Recent reforms doing away with DNS systems, or subjecting them to restrictions or loss-sharing provisions that may run counter to what participants would have contracted for, have been inefficient if not counterproductive substitutes for reforms aimed at eliminating uncalled for finality guarantees extended by regulatory authorities. Although it may not be possible to reverse the wave of central bank sponsored reforms aimed at abolishing or at least ‘securing’ DNS systems, it is not too late for economists to strive for greater clarity concerning what such reforms have and have not accomplished. The reforms have succeeded in reducing or eliminating unwind risk in wholesale payments. In some cases, the reforms may have countered inefficiencies stemming from the presence of central bank guarantees or from the failure of courts to enforce private contracts. What the reforms cannot be shown to have achieved is the correction of any genuine market failure.

Notes 1 The author thanks Bill Bergman, Sandra Haasis, David Humphrey, George Kaufman, David Mustard, Harold Nitsch, Will Roberds, Art Snow, Larry Wall, Ron Warren, Lawrence H. White, several anonymous referees, and participants in the University of Georgia Economics Department workshop for their helpful comments and suggestions. 2 In January 2001, CHIPS switched from deferred net settlement to near-continuous settlement. 3 Since 1986 the Fed has also imposed limits, known as ‘net debit caps,’ on the maximum overdraft individual banks could obtain from it without being subject to special administrative actions. These caps appear, however, to have had only a very limited effect on the overall value of overdrafts (Hancock and Wilcox, 1996). 4 According to Zhou (1999), allowing for deductibles Fedwire’s ‘imputed average annual rate [was] only around 11 basis points’ after 1995. Zhou’s study is also valuable for its excellent critique of models (e.g. Freeman, 1999) that claim to demonstrate the optimality of a zero intraday lending rate. The essence of Freeman’s argument is that the assumption of aggregate default risk by central banks transforms that risk into less problematic inflation risk. Because Freeman’s model assumes an exogenous default rate, it cannot allow for the moral hazard problem induced by

Wholesale payments

5

6

7

8 9

133

having the costs of intraday borrowing borne largely by third parties. A model aimed at addressing the question of the optimal intraday borrowing rate should at least be capable of representing the problem of moral hazard! By a traditional or ‘unsecured’ DNS system I mean a system in which there are no caps or limits imposed on the volume of payment flows and where the failure of any participant to settle results in the cancellation of payment messages to and from that participant. The oft-cited Bank of New York computer glitch of December 1985 affected the government securities market only and did not result in any CHIPS unwind. The Fed automatically debited the Bank of New York’s reserve account to pay for securities it received on behalf of that bank’s clients, but because the glitch kept the Fed from knowing where to deliver the securities, the Bank of New York ended up with a $32 billion overdraft, which lasted for 90 minutes. CHIPS’s conversion to continuous settlement coincided with the establishment of a new private-sector network for foreign-exchange settlement: the Continuous Linked Settlement (CLS) Bank, aimed at avoiding Herstatt risk. Herstatt was a small German bank that failed in 1974, after having received irrevocable mark payments, but before the dollar leg of its transactions had been settled. Note that this was not a settlement failure in the sense used throughout the paper: Herstatt was closed at 08:30 h New York time, with no outstanding CHIPS payment orders. In 1995, wholesale payments in most nations were processed by DNS systems, and Switzerland was the only major country that relied upon real-time gross settlement for all of its large-value payments. Referring specifically to the New York Clearing House Association Cannon (1910: 209) observes that: The association is in no way responsible for the balances, except in so far as they are actually paid into the hands of the manager, and then its responsibility is strictly limited to the faithful distribution by him among the creditor banks of the amounts which he has received.

CHIPS today likewise disclaims responsibility for any obligations incurred by its members. 10 Even today most US banks provide their customers with ‘deposit account agreement and disclosure’ statements that include language like the following (from the Seaway National Bank of Chicago): All non-cash items (for example, checks) deposited to your Account are posted subject to our receipt of final payment by the payor bank. If final payment is not received . . . you authorize us to charge any of your Accounts, without prior notice and at any time, for the amount of the returned item, our returned item fee, any interest paid on the item, and any other fee we pay or incur . . .. With respect to wire transfers . . . you agree to enter into and comply with our wire transfer agreement. . . . Credit given by us to you with respect to [a] wholesale (wire) funds transfer entry is provisional until we receive final settlement for such entry through a Federal Reserve Bank. If we do not receive final settlement, you are hereby notified and agree that we are entitled to a refund of the amount credited to your account in connection with such entry, and the party (the originator of the entry) making payment to you via such entry shall not be deemed to have paid you the amount of the entry. 11 The practice of ‘unwinding’ transactions of failed clearinghouse participants, although rarely resorted to, goes back to the early days of clearinghouses. According to Cannon (1910: 277–278), members of the Chicago Clearing House (founded in 1865)

134

G. Selgin who have been so unfortunate as to fail have seldom attempted to clear on the day of their failure. . . . When, however, a failing bank has attempted to make its regular clearings and has not been able to pay the balance against it . . . all the exchanges presented by and against it have been returned and a new settlement made, the same as if it had not participated in the exchanges of the day. In this way, those who had presented checks against the failing member were enabled to return them to their customers, thereby intrenching themselves against loss.

Cannon goes on to observe that, although [n]o formal provision was made for such action at the onset, . . . as time passed on this course was found to be the best means by which to avoid serious complications. Accordingly, the rule has been embodied in the constitution. It is similar in its provisions to that existing in the constitutions of nearly all the clearinghouse associations of the country. 12 I abstract here from any possible influence of government guarantees, which are discussed below. 13 Some writers (e.g. Dale, 1998: 231–232) recognize that receiving banks might avoid intraday credit exposure in traditional DNS systems simply by refraining from releasing funds prior to settlement, but insist nevertheless that, insofar as funds are released during the day, resulting credit exposures ‘are the result, not of credit judgments made by the participant banks, but of customer transactions over which the participants have little or no control.’ The argument begs the question: what is it that compels the banks to release funds at once? 14 Mengle goes on to observe that this argument no longer holds if a sending bank participates in a DNS network indirectly, via a correspondent, because it is not reasonable to expect senders to ‘know’ their banks’ correspondents. This is true enough; however, the problem in question reflects, not any inherent shortcoming of DNS or check finality, but yet another regrettable consequence of historical restrictions on nationwide and international branch banking. In places (like Canada) where DNS arrangements have developed along with unrestricted branch banking, senders generally have the option of doing business directly with clearing banks. 15 In 1990, for example, CHIPS, anticipating the Bank of International Settlements’ Lamfalussy Report (which recommended that participants in DNS systems be required to collectively post enough collateral to cover default by any single participant, thereby further limiting the potential for a payments unwind), adopted rules and fees especially designed to limit its members’ exposure to systemic risk. It is not clear, however, that CHIPS considered these changes worthwhile except as a means for pre-empting more stringent government regulation. 16 Shen (1997: 51–53) offers an excellent discussion of systemic (‘liquidity’) risk in RTGS systems. He notes (Shen, 1997: 53) that, in the Swiss Interbank Clearing System (in which banks do not have access to central-bank supplied intraday credit), on an average day c.1997 ‘at least 45% of payments experience[d] some delay in their execution due to the lack of liquidity.’ See also Kahn et al. (2003). The reality of frequent payment order rejections in some actual RTGS systems is to be contrasted with the (so far) purely hypothetical possibility of a DNS-system unwind. The Fed, of course, avoids the problem of liquidity risk in Fedwire only by exposing itself to credit risk. 17 An anonymous referee, while agreeing that ‘the choice between RTGS versus DNS is a matter of relative costs,’ observes that ‘which system is more cost-effective is by no means a settled issue,’ and observes that I fail to supply ‘any new evidence’ of the cost-effectiveness of traditional DNS. That is true enough. However, the referee appears to forget that standard practice, informed by the welfare theorems of neoclas-

Wholesale payments

18

19

20 21 22

23

24

25

26 27

28

135

sical economics, places the burden of proof not on those who deny the presence of a market failure, but on those who insist that such a failure is present. The likelihood that either conventional accounting ratio analysis or CAMEL (capital, assets, management, earnings, and liquidity) ratings will fail to identify banks that are about to fail diminishes as the frequency of monitoring increases. Thus, of more than 1600 banks that failed between 1980 and 1994, only 16 percent had high (1 or 2) CAMEL ratings based on examinations made within a year of the banks’ failure, whereas 36 percent had ratings of 1 or 2 two years prior to failing. More frequent financial monitoring and public disclosure of CAMEL ratings and such could presumably go far in further reducing the (already low) probability of a DNS settlement failure. Angelini et al. (1996) performed a Humphrey-style simulation for Italy’s DNS system, which differed from CHIPS mainly in having many more (288) direct participants. They found that only 4 percent of these participants had the potential to trigger a systemic ‘crisis,’ and that in no case would the ‘crisis’ have caused more than seven other participants to default. The probability of no chain of defaults occurring following a single participant’s failure was 96 percent. This was again assuming, following Humphrey, zero recovery of provisional (intraday) credit to account holders. This is not to deny that a run to currency would, ceteris paribus, reduce the money multiplier, and would therefore necessitate some expansion of the monetary base to maintain the money stock. For a more general discussion of the incentive advantages of net settlement see Roberds (1999) and Kahn et al. (2003). In personal correspondence Roberds acknowledges the inherent risk-control advantages of net settlement, observing that ‘The problem with more modern net settlement systems is that the incentive to undertake [traditional] risk-mitigating activities has necessarily been diminished by the presence of central banks.’ That CHIPS participant practices were influenced by Fed guarantees of one sort or another is strongly suggested by the fact, reported to me by Humphrey, that those participants persistently refused to follow regulators’ suggestion that they purchase ‘settlement insurance’ on the grounds that ‘the risks [meaning, presumably, the premiums] were too great.’ During the 1980s, for example, approximately 90 percent of 418 banks that received extended emergency credit from the Fed failed subsequently, and most of them were known or at least suspected by regulators to have been insolvent when assistance was granted to them (Kaufman, 1999: 4). From 1857 to 1907, when interbank lending markets were poorly developed, members of CHIPS’ predecessor, the New York Clearing House Association, operated an effective coinsurance scheme by agreeing to accept fully collateralized clearinghouse ‘loan certificates’ in lieu of gold for settlements during financial emergencies (Timberlake, 1984). Were banks and their customers truly certain that last-resort loans would be employed to prevent a payments unwind, they would be indifferent between receiver finality and check finality rules (Scott, 1990: 186). Dale (1998: 229) points out that the very involvement of central banks ‘in the settlement process can lead to market expectations of official support in the event of any threatened disruption’ and that for this reason settlement risk exposure is best contained by means of settlement arrangements in which central bank involvement ‘is least evident.’ In CHIPS this particular externality problem was partly avoided by linking each participant’s share of losses to the bilateral credit line the participant has set for a defaulting bank. A reduced moral hazard is thus achieved at the expense of more rigidly constrained payments flows.

136

G. Selgin

References Angelini, P., Maresca, G., and Russo, D. (1996) ‘Systemic risk in the netting system,’ Journal of Banking and Finance, 20(5): 853–868. Board of Governors of the Federal Reserve System (1988) Controlling risk in the payments system, Washington, DC: Board of Governors of the Federal Reserve System. Cannon, J.G. (1910) Clearing houses, Washington, DC: Government Printing Office. Coase, R.H. (1988) The firm, the market and the law, Chicago, IL: University of Chicago Press. Dale, R. (1998) ‘Risk management and public policy in payment, clearing and settlement systems,’ International Finance, 1(2): 229–259. Dowd, K. (1994) ‘Competitive banking, bankers’ clubs, and bank regulation,’ Journal of Money Credit, and Banking, 26(2): 289–308. Folkerts-Landau, D. (1997) ‘Wholesale payments and financial discipline, efficiency, and liquidity,’ International Monetary Fund Working Paper No. 154. Freeman, S. (1999) ‘Rediscounting under aggregate risk,’ Journal of Monetary Economics, 43: 197–216. Gerson, J.E. (2001) ‘Theoretical approaches to catastrophic risk,’ unpublished thesis, University of Chicago. Hancock, D. and Wilcox, J.A. (1996) ‘Intraday management of bank reserves: the effects of caps and fees on daylight overdrafts,’ Journal of Money, Credit and Banking, 28: 870–908. Humphrey, D.B. (1986) ‘Payments finality and the risk of settlement failure,’ in A. Saunders and L.J. White (eds) Technology and the regulation of financial markets, Lexington, KY: Lexington Books. Kahn, C.M. and Roberds, W. (1999) ‘The design of wholesale payments networks: the importance of incentives,’ Federal Reserve Bank of Atlanta Economic Review, 84(3): 30–39. Kahn, C.M., McAndrews, J. and Roberds, W. (2003) ‘Settlement risks under gross and net settlement,’ Journal of Money Credit and Banking, 35: 591–608. Kaufman, G. (1999) ‘Do lender of last resort operations require bank regulation?,’ unpublished thesis, Loyola University Chicago. Mengle, D.L. (1990) ‘Legal and regulatory reform in electronic payments: an evaluation of payment finality rules,’ in D.B. Humphrey (ed.) The US payment system: efficiency, risk and the role of the Federal Reserve, Boston, MA: Kluwer. Mengle, D.L., Humphrey, D.B., and Summers, B.J. (1987) ‘Intraday credit: risk, value, and pricing,’ Federal Reserve Bank of Richmond Economic Review, 73(1): 3–14. Richards, H.W. (1995) ‘Daylight overdraft fees and the Federal Reserve’s payment system risk policy,’ Federal Reserve Bulletin, 81(12): 1065–1077. Roberds, W. (1999) ‘The incentive effects of settlement systems: a comparison of gross settlement, net settlement, and gross settlement with queuing,’ Bank of Japan Institute for Monetary and Economic Studies Discussion Paper No. 99-E-25. Rochet, J.-C. and Tirole, J.T. (1996) ‘Controlling risk in payment systems,’ Journal of Money Credit and Banking, 28: 832–862. Schoenmaker, D. (1995) ‘A comparison of alternative interbank settlement systems,’ London School of Economics Financial Markets Group Discussion Paper No. 204. Scott, H.S. (1990) ‘Commentary,’ in D.B. Humphrey (ed.) The US payment system: efficiency, risk and the role of the Federal Reserve, Boston, MA: Kluwer.

Wholesale payments

137

Shen, P. (1997) ‘Settlement risk in large-value payment systems,’ Federal Reserve Bank of Kansas City Economic Review 82(2): 45–62. Timberlake, R.H. (1984) ‘The central banking role of clearinghouse associations,’ Journal of Money Credit and Banking, 16: 1–15. Van den Bergh, P. and Veale, J.M. (1994) ‘Payment system risk and risk management,’ in B.J. Summers (ed.) The payment system: design, management, and supervision, Washington, DC: International Monetary Fund. Zhou, R. (1999) ‘Understanding intraday credit in large-value payment systems,’ unpublished thesis, Federal Reserve Bank of Chicago.

9

Central bank intraday collateral policy and implications for tiering in RTGS payment systems John P. Jackson and Mark J. Manning1

Introduction In this chapter we present a model of a Real-Time Gross Settlement (RTGS) payment system with tiered membership where settlement is facilitated by intraday credit extensions from the central bank. RTGS systems process and settle payment instructions individually in real time, ensuring intraday finality. Furthermore, central banks typically provide the settlement accounts across which payments are processed; hence, settlement is typically effected in central bank money, thereby eliminating counterparty risks between members once settlement has taken place. The model allows us to examine the key factors that influence both an agent’s decision over whether to participate directly in an RTGS payment system, and a central bank’s decision as to whether to require collateralization of intraday credit extensions to payment system participants. The design of all payment arrangements must reflect a trade-off between cost and risk. As noted in BIS (2005), ‘if a system was so costly or burdensome that no one used it, the system would have no effect on risks no matter how extensive its risk controls’. This applies as much to RTGS systems as to any other system design, for while addressing the counterparty credit risks associated with Deferred Net Settlement (DNS) systems, RTGS of payments can be a significant, and costly, drain on a bank’s liquidity (Kahn and Roberds, 2001). To alleviate this burden, central banks also typically offer intraday credit to payment system participants. In the absence of such credit, users would have to pre-fund their settlement accounts at the central bank, thereby incurring a substantial opportunity cost of holding liquidity.2 Chakravorti (2000), Kahn and Roberds (2001) and Bech and Garratt (2003) as well as the chapter by Bech et al. in this volume all highlight the behavioural implications of costly liquidity needs. They note that system participants might then seek to reduce these costs by delaying the submission of payments, with potentially adverse consequences for liquidity recycling in the system, operational risk, and, to the extent that obligations are not settled as expected, ultimately social welfare. Central bank provision of intraday credit to payment system participants entails a potential credit exposure. Several risk-mitigating measures might be taken, but these impose costs on payment system participants, and potentially

Central bank intraday collateral policy

139

also on society. Many central banks, including the Bank of England, require full collateralization of intraday credit exposures; this greatly reduces credit risk, but imposes an opportunity cost of posting collateral. By contrast, the US Federal Reserve does not require collateralization, but instead charges an interest rate on all intraday overdrafts and imposes credit limits on agents’ usage of intraday credit. Costly intraday credit can have similar behavioural consequences to prefunding requirements: prompting banks to economize on their usage of liquidity in the system by delaying payments. The trade-off faced by central banks, between assuming greater credit exposure and imposing costs on participants, has been subject to considerable scrutiny in recent years. Furfine and Stehm (1998), for instance, highlight the deadweight welfare losses associated with costly collateral requirements. They conclude that, from a social welfare perspective, a policy of free liquidity provision would be preferred to full collateralization unless the opportunity cost of collateral tended to zero. However, more recent work by Mills (2005) suggests that models of this type may not have adequately accounted for the credit risks faced by the central bank under zero collateralization. A related strand of literature focuses on how central banks, if they do require that intraday credit be collateralized, can reduce the opportunity costs incurred by system participants in posting such collateral. Manning and Willison (2005) show that allowing cross-border usage of collateral enables agents to economize on their total collateral holdings, while the chapter by Green in this volume suggests that central banks could accept less liquid (and hence lower cost) collateral than might other secured lenders. Alternatively, Willison (2005) considers recourse to more liquidity-efficient payment system designs, so-called hybrid systems, to reduce the amount of intraday credit needed to settle a given set of payments.3 Where significant costs of obtaining intraday credit remain, agents might choose not to participate directly in an RTGS system at all. The chapter by Rochet in this volume argues that to the extent that an agent chooses to by-pass an RTGS system, by entering into bilateral agreements with other agents or by shifting flows to a competing DNS system, systemic risk may be increased. Such alternative arrangements typically include recourse to a correspondent bank, who processes payments on behalf of indirect system participants. This phenomenon, known as ‘tiering’, is a commonly observed feature in many RTGS systems internationally. CHAPS Sterling, with only 13 out of around 350 commercial deposit-taking banks operating in the United Kingdom participating directly, is particularly highly tiered (Harrison et al., 2005). In this chapter we explore the implications tiering can have on the welfare costs associated with risk mitigation in payment arrangements. Our starting point is an insight from Kahn and Roberds (2006), who identify delegated monitoring as an alternative enforcement device to collateralization. They show that, in the presence of private information about the reliability of agents, delegated monitoring can economize on the need for agents to post collateral to guarantee repayment of intraday credit. This monitoring is achieved

140

J.P. Jackson and M.J. Manning

through a tiered structure, whereby a direct participant of a payment system absorbs the risk associated with credit extensions to its customer banks, hence maintaining good incentives to monitor. When default states occur only a fraction of the time, a full collateralization policy can achieve only a second-best outcome; any collateral posted to the central bank in non-default states is a deadweight loss to society, arising as a result of the central bank’s imperfect information about settlement banks’ credit quality and their wish to minimize their own credit losses (and hence costs to the taxpayer). Under such a scenario, if monitoring is sufficiently accurate, and monitoring costs sufficiently low, delegated monitoring can achieve a smaller deviation from the first-best. In this chapter we highlight two additional channels by which tiering in payment systems might lead to a reduction in the deadweight social costs associated with full collateralization of intraday credit extensions by the central bank. First, we consider internalization of payments. This refers to a situation where payments made between customers of the same correspondent bank are settled internally across the correspondent’s books, without being processed through the payment system. Internalization allows payments to be made without recourse to intraday credit from the central bank, thereby avoiding any costs associated with collateral posting requirements. Therefore, to the extent that tiering facilitates the internalization of payments, it can reduce the costs imposed by a central bank’s full collateralization policy. Furthermore, to the extent that payment flows from a correspondent’s customers to other first-tier participants are likely to be spread out through the day, there may be a diversification, or ‘collateral-pooling’ benefit. That is, unless customers’ payment flows are perfectly correlated, the total pool of collateral required to generate intraday credit on behalf of several customer banks, will be smaller than that required were each customer’s collateral needs served from segregated pools of collateral. Where payments made by second-tier participants are not internalized, but rather are effected by the correspondent over central bank settlement accounts, any collateral-posting requirements at the central bank would apply. But here too, tiering may reduce the deadweight social costs of collateral if agents with high opportunity costs of posting collateral are able to take advantage of lower collateral-posting costs enjoyed by their correspondent. The opportunity cost of posting collateral may be proxied by the reverse-repo spread (i.e. the spread between secured and unsecured borrowing costs). This may vary across agents, according to differences in credit-worthiness which affect the unsecured cost of borrowing. Also, market imperfections and regulatory policy may influence agents’ relative opportunity cost of collateral. For example, in the United Kingdom, banks subject to the Stock Liquidity Requirement (SLR), a prudential liquidity regime, are able to meet intraday collateral requirements using assets that they have to hold, in any case, to meet their prudential requirement at the end of the day; hence they have a very low opportunity cost of posting collateral intraday.4


141

The foregoing discussion highlights the potential benefits of tiering. However, tiering can also introduce additional risks to the system. First, a correspondent bank might not have sufficient incentive to monitor because it does not fully internalize the potential systemic consequences of a customer’s default which triggers liquidity or solvency problems of its own. Even if sufficient incentives were to exist, monitoring of second-tier agents by a correspondent might not be accurate, causing the correspondent to incur credit and liquidity exposures that might in turn lead to wider contagion. Harrison et al. (2005) analyse the credit risk implications of the highly tiered structure of the UK large value payment system, concluding that this channel might not impose significant risks on the system as a whole, except in extreme circumstances. However, other risks exist. For instance, internalized payments might be subject to greater legal risks as they are not likely to be covered by provisions providing protection against bankruptcy law (such as the Settlement Finality Directive in the European Union). Tiering increases the risk that operational or financial problems at a settlement bank lead to disruption of payments in a large part of the system. Another important concern is that, in response to liquidity problems among second-tier banks, settlement banks might decide to cut or restrict intraday credit, further exacerbating these liquidity problems. In this chapter, we apply a simple model of an RTGS payment system, in which agents rely on the central bank to provide intraday credit to facilitate settlement of a single payment obligation. We allow the paying agent to choose whether to be a direct participant of the payment system, or to settle its obligation via an existing direct participant. With complete information as to the determinants of the agent’s choice, the central bank chooses whether or not intraday credit should be fully collateralized.5 Using this framework, we are able to show that, when the central bank requires full collateralization, it may be optimal for an agent to become an indirect participant, so as to take advantage of cost-efficiency benefits arising from monitoring as a substitute for collateralization. These benefits are increased to the extent that payments can be internalized and that agents can take advantage of their correspondent’s lower collateral posting costs.6 Furthermore, in the absence of spillover risks from tiering, private and social costs are aligned when the central bank opts for full collateralization. Therefore, any private costefficiency benefit derived translates directly into a social welfare improvement. We do show, however, that a wedge between social and private costs is likely to exist under zero collateralization and that, although welfare might be maximized if the central bank requested zero collateral and the agent chose to access the system indirectly, this outcome is not achievable: it will always be in the agent’s interest to access directly under zero collateralization. Unless the probability of default is very low, this will rarely be optimal for the central bank. With imperfect monitoring and the potential for systemic spillovers from tiering, a wedge will also emerge between social and private costs when the central bank opts for full collateralization, with this wedge increasing in the degree of monitoring imperfection, the value of payments to be settled and the

142


degree of spillover per unit of exposure. In this case, policy intervention might be desirable to address the risks introduced. The chapter is organized, as follows. We first outline our analytical framework. We then apply this framework to analyse agents’ decisions under alternative scenarios for the quality of monitoring and the existence of tiering externalities before offering some conclusions.

Analytical framework In this section, we present a simple model of payment arrangements to explore two key decisions: a bank’s decision as to whether to access an RTGS payment system directly, or via a correspondent banking arrangement and a central bank’s decision as to whether to collateralize intraday credit extensions. Our analysis draws on the framework presented in Kahn and Roberds (2006), but applies this in a much-simplified, stylized and reduced-form fashion.7 We first provide an overview of the model set-up, going on to describe in greater detail the actions taken and costs incurred under each alternative arrangement. The model set-up and timeline for actions The essence of the model is a game of complete information, with actions taken sequentially by two players: the central bank; and a commercial bank, C. There are two further agents in the game, banks A and B, both of whom are direct settlement members of the payment system, with A also a potential provider of correspondent banking services to C. Bank B never provides payment services. All agents are assumed to be risk-neutral. Neither bank A nor bank B take any direct actions in the game, although we do establish the terms on which A provides correspondent banking services if called upon to do so, ensuring that it would be rational for A to offer such services. Bank C makes a single payment, of value unity.8 No other payments are made. Time consists of a single day, divided into four periods. In period 0 the central bank sets its collateral policy with respect to bank C, choosing actions from the set {F,Z}, where F = full collateralization; and Z = zero collateralization, so as to minimize expected social costs.9 In period 1, observing (with certainty) the central bank’s policy choice, C minimizes its expected costs with respect to its decision as to whether to fulfil a single payment obligation directly in the RTGS payment system, or via correspondent banking services provided by direct payment system participant, A. Its set of potential actions is then {D,I}, where D = direct participation; and I = indirect participation. We assume that, if indifferent, C will always choose to participate directly. The state of the world is characterized by {e,}. Parameter e [0,1] describes the possible orientation of payment flows in the system; with probability (1 – e), C is obliged to make a payment to A, whereas with probability e, C has an obligation to B. As will be discussed below, the orientation of payment flows has


143

implications for the degree of internalization possible when C’s payments are settled indirectly via A. The orientation of payment flows is realized by a draw from nature in period 2 and revealed immediately to the agents. Parameter [0,1] is the probability that C suffers an exogenous default shock which prevents it from repaying intraday credit extended by either A or the central bank.10 The incidence of a default shock is also realized by a draw from nature in period 2. The outcome is not revealed to agents until period 3, although a signal as to whether or not a shock has occurred can be obtained by monitoring in period 2. Settlement of C’s payment obligation occurs in the payment system in period 2. It is assumed that C has no endowment of the settlement asset at the start of the period and hence always requires an intraday credit extension, either from the central bank or from A, before settlement can be effected. Equally, we assume that A requires intraday credit from the central bank before it can make a payment on C’s behalf. All intraday credit extended in period 2 is to be repaid by the end of period 3, by which time C expects to have received a sufficient quantity of the settlement asset (from a maturing investment). Parameter may be interpreted as the probability that this investment fails and returns nothing. If settling directly in the system under full collateralization, C also posts collateral in period 2 to support its request for intraday liquidity. C then settles its obligation with finality in central bank money.11 If bank A is settling on behalf of C, A first obtains intraday credit from the central bank. Any cost to A of posting collateral to the central bank is passed on directly to C.12 This collateral requirement in respect of C’s payments under an indirect arrangement can, however, be reduced by internalization. When C has a payment obligation to A, this can be settled directly on A’s books. Only if C has to make a payment to B will settlement occur in central bank money. This is shown in Figure 9.1 below. Whether settling C’s obligation in central bank money, or internalizing across its own books, A settles C’s payment obligation in advance of the receipt of funds from C; that is, A extends intraday credit to C. A can potentially economize on collateral sought from C in respect of such a credit extension by carrying out monitoring.13 More specifically, A monitors to obtain a signal as to whether C has suffered a default shock in period 2. If monitoring reveals that a A

B

C C has an obligation to A: Probability (1e) Payment in central bank money

A

B

C C has an obligation to B: Probability e Payment in commercial bank money

Figure 9.1 Payment flows when C accesses the system via bank A.

144


Period 0: The central bank sets its collateral policy with respect to intraday credit extensions to C.

Period 2: Nature determines the recipient of C ’s payment obligation and C ’s default shock is realised (but not revealed to agents). If settling for C, A monitors to obtain a signal as to whether C has suffered a default shock. Intraday credit is granted and C posts collateral if required to do so. Settlement occurs.

Time

Period 1: C decides whether to fulfil its payment obligation directly in the payment system; or indirectly via A.

Period 3: C ’s default shock is revealed to agents. Intraday credit repaid if no shock has occurred.

Figure 9.2 A time-line for actions.

default shock has occurred, collateral will be sought from C in respect of intraday credit granted; otherwise A will not require payments to be collateralized. Any monitoring costs incurred by A are passed on to C. If no default shock arises in period 3, all intraday credit is repaid. Otherwise, C defaults on the repayment of its intraday credit. Unless sufficient collateral has been posted, this will impose default costs upon agents in the system. These costs will be described below. The timeline of the model is shown in Figure 9.2. Agents’ actions and costs Characterization of bank C’s costs Bank C’s costs are characterized by: E[CC] = f (M,C,A), where M is A’s cost of monitoring C, which, to the extent incurred, will be passed on to C; C is C’s private opportunity cost of posting collateral either to A or the central bank; and A is A’s private opportunity cost of posting collateral to the central bank, which again will be passed onto C.14 We consider each in turn. As noted above, by incurring a monitoring cost, M, A can obtain a signal as to whether C has suffered a default shock and hence will be unable to repay an intraday credit extension. It is initially assumed that the signal obtained by monitoring is perfectly correlated with the shock; we later relax this assumption and allow for Type I and Type II errors. Private monitoring costs might be expected to be relatively low to the extent that the normal-course interconnection between financial institutions ensures a steady flow of information between them. Indeed, this implies that larger banks with diversified activities and a wide network of clients in non-payments-related businesses, will have access to more private information, and hence be better monitors. Furthermore, there are likely to be economies of scale in monitoring activity.


145

Turning to collateral costs, it is worth noting that where the costs faced by a correspondent bank are lower than those faced by its customer banks, indirect participation may allow agents with high collateral costs to take advantage of lower costs enjoyed by other agents. Recall that in the model, any costs incurred by A in respect of settlements effected on behalf of C will be passed on in full; but the lower these costs are relative to C’s own direct collateral costs, the more efficient indirect participation will be relative to direct membership. Should C suffer a shock in period 3 and consequently fail to repay its intraday loan, additional private costs of default might be incurred. For simplicity such costs are normalized to zero. To further simplify the exposition, we also normalize to zero any fixed costs (technological and fees) associated with linking directly to the payment system or indirectly through A. When settling directly, C faces a collateral cost of C if the central bank requires full collateralization. Bank C’s costs under indirect participation depend on the relative costs of monitoring and collateralization, and the potential for internalization. We assume that A prices its correspondent services competitively, subject to a full cost–recovery constraint, and offers two alternative correspondent banking services: one involving monitoring; and another involving full collateralization. Bank C chooses the service that offers the lowest expected cost. We assume that C has no private information about its likelihood of experiencing a default shock, and that this is common knowledge. As the full collateralization service offered by A can never be cheaper than settling directly at the central bank (C faces cost, C, in both cases), and under the assumption that, where the costs of direct (D) and indirect participation (I) are equal, C will choose D, it is clear that C will never choose A’s full collateralization service. Hence, indirect participation will always be associated with monitoring. Where A does monitor C, and the signal obtained is perfectly correlated with the incidence of default, it will only be optimal for A to request collateral in default states (at a cost of C to C), but not otherwise. Bank C will, however, still have to compensate bank A for any additional collateral it may be required to post to the central bank, implying an extra cost above the direct cost of posting collateral to A of e(1 – )A. This assumes that of the time A will simply use collateral posted by C to cover any requirement at the central bank. The expected cost to C is then: E[CC] = M + C + e(1 – )A

under (F,I)

E[CC] = M + C

under (Z,I)

Central bank’s costs We assume that the central bank makes decisions in order to minimize expected social costs. These costs are characterized by E[CCB] = f (M,C,A,S), where the

146


first three cost parameters are private costs, as above, and S is the social cost of default by C when intraday credit is not fully collateralized. Monitoring and collateral costs are included here as these are deadweight costs to society arising from informational asymmetries and limited enforcement. A social cost S 1 arises when C cannot repay an uncollateralized intraday loan at the central bank and hence the central bank is forced to cover the resulting loss via taxation of unmodelled agents. The extent to which S exceeds unity reflects any distortion imposed by the tax. If, on the other hand, default occurs and collateral has been pledged, the defaulting agent’s creditor (either the central bank, or bank A under a tiered arrangement) may attach the defaulter’s collateral up to the amount of the pledge. For simplicity, we assume no market risk to the value of collateral posted.15

Agents’ decisions In this section, we apply the framework described above to establish equilibrium outcomes for central bank collateral policy and the degree of direct participation in payment systems. We consider three alternative cases: (i) perfect monitoring; (ii) imperfect monitoring; and (iii) imperfect monitoring with tiering spillovers. Case 1: perfect monitoring and agent-specific collateral costs In this case, we assume that monitoring reveals default states with certainty, and allow for banks A and C to face different opportunity costs of posting collateral. We solve the model by backwards induction, applying sub-game perfection as an equilibrium concept. Accordingly, we begin with C’s decision. Consistent with the earlier discussion, C faces the expected private costs detailed below: (Z,D): 0 (F,D): C (Z,I): min[M + C,C] (F,I): min[M + C + e(1 – )A,C] where, to establish expected costs, states of the world are weighted by the probabilities e and . Comparison of C’s expected costs immediately reveals that it will always be optimal for C to access the payment system directly if the central bank adopts a policy of zero collateralization (since min[M + C,C] > 0). In the event that the central bank chooses F, we see that C constitutes an upper bound for costs under indirect participation. Hence, in this case, C will certainly opt for indirect participation if the inequality in (1) holds.

Central bank intraday collateral policy M C > M + C + e(1 – )A ⇒ C > + eA (1 – )

147 (1)

If this inequality does not hold, expected costs under (F,I) will equal C, leaving bank C indifferent. Again, we assume that, if indifferent, C will settle directly.16 It is clear from the inequality in (1) that C’s choice under full collateralization will depend on several parameters. In particular, the inequality in (1) is more likely to hold, and hence C is more likely to participate indirectly, the higher is C and the lower are , M, e and A. Inequality (1) illustrates that, with A < C, C can take advantage of A’s lower collateral costs by choosing to be an indirect member. With A < C = , this potential efficiency is no longer available and the inequality in (1) reduces to > M/(1 – )(1 – e), which is less likely to hold. In period 0, the central bank chooses its actions, anticipating the choices C will make in response in period 1. The central bank’s expected (social) costs are given by: (Z,D): S (F,D): C (Z,I): min[M + C,C] (F,I): min[M + C + e(1 – )A,C] It is important to note that social and private costs are equivalent in all cases with the exception of when uncollateralized exposure is retained (and hence there is some probability that social default costs are suffered): i.e. the case with (Z,D). No externality exists under (Z,I) as A fully absorbs the shock of any default by C. This assumption will be relaxed later. Given that C will choose to participate directly if the central bank chooses Z, the central bank will compare expected social costs under (Z,D) with those under C’s optimal response to a policy of full collateralization. If the inequality in (1) holds, the relevant comparison is with expected costs under (F,I); if the inequality in (1) does not hold, the relevant comparison is with expected costs under (F,D). Depending on whether the inequality in (1) holds, the central bank will choose zero collateralization if either S < C or S < M + C + e(1 – )A (whichever is relevant, given parameter values); and full collateralization otherwise.17 Intuitively, then, ceteris paribus, zero collateralization and direct participation by C is more likely, the lower the probability that C defaults and the social costs associated with default; the higher the opportunity cost of posting collateral for either (or both) A and C; the lower the cost of monitoring; and the lower the probability of internalization.

148


A graphical illustration It is instructive to illustrate agents’ choices graphically, so as to draw out their important determinants and identify potential sources of divergence of public and private interests. Figures 9.3 and 9.4 trace private and social costs with varying M, for given values of A, C, S, and e. Figure 9.3 presents a case with a high degree of internalization (a 90 per cent probability that payments will be internalized). It is clear that, with zero collateralization, C will choose direct participation; and, with full collateralization, C will prefer indirect participation for all values of M up to the threshold X shown in the figure. Thereafter, direct participation will be chosen. Given these responses, the central bank will choose full collateralization: it is clear that the social cost of the combination (Z,D) is higher than that associated with C’s optimal choices under full collateralization, for all M. However, this is not the socially optimal outcome. It is clear from the figure that (Z,I) would maximize social welfare for all values of M shown. However, this first-best outcome is unachievable because C makes its choice after the central bank, and (Z,D) offers lower private costs for all M. Hence, under such a scenario, there is a wide interval of monitoring costs within which it seems that policy intervention might be justified to steer the market towards the socially optimal outcome of indirect participation when the central bank chooses not to require that credit extensions be collateralized. However, it should be recognized that this initial scenario has perfect monitoring and no tiering risks/spillovers, which, as we will show, will leave (Z,I) socially preferred for a narrower range of monitoring costs, if preferred at all. Also, we have assumed that A faces no risk of an exogenous default shock and hence the central bank’s credit extension to A is, in this scenario, riskless. 0.0020 (Z,D) Social

0.0018 0.0016

0.0012

X

0.0010

(F,I ) Social Private

0.0008 0.0006 (Z,I ) Social Private

Social/private cost

0.0014

(F,D) Social Private

0.0004

(Z,D) Private

0.0002 0

0

0.0003

0.0006

0.0009

0.0012

0.0015

Monitoring cost (M)

Figure 9.3 High degree of internalisation (C = A = 0.0015; S = 1.1; = 0.0015; e = 0.1).


149

0.0020

X

0.0018


0.0016

0.0012

(F,I ) Social Private (Z,D) Social

Y

0.0010 0.0008 0.0006

(Z,I ) Social Private

Social/private cost

0.0014 W

0.0004

(Z,D) Private

0.0002 0

0

0.0003

0.0006

0.0009

0.0012

0.0015

Monitoring cost (M)

Figure 9.4 Low default probability (C = A = 0.0015; S = 1.1; = 0.00075; e = 0.5).

Figure 9.4 presents a scenario with a low default probability. Here, the lowest expected private cost for C is again associated with the combination (Z,D). Under full collateralization, C would prefer indirect participation for values of monitoring cost up to threshold value X; above these values direct participation would be optimal. Given C’s responses, and the low default probability in this scenario, the central bank would choose full collateralization for values of monitoring cost up to threshold value W. Beyond this point, the central bank would favour zero collateralization. Interestingly, social and private preferences are aligned beyond Y in this case, reflecting C’s low default probability and hence the relatively low social costs associated with outcome (Z,D). Below Y, however, the first-best outcome (Z,I) is again unachievable, although the caveats noted above remain relevant in this regard. The profile of payment system participation in the UK The UK experience is consistent with the broad predictions of the model as stated above. In the United Kingdom the Bank of England requires full collateralization of intraday credit extensions.18 A sub-set of banks (the UK-owned banks) face very low opportunity costs of collateral due to the fact that assets held to meet prudential regulatory requirements can be used to back intraday liquidity needs. Furthermore, correspondent banking is highly concentrated, and becoming more so, with just three banks providing the bulk of these services. Thus, a high degree of internalization takes place.19 These correspondent banks are all UK-owned, and hence all benefit from, and offer, a low opportunity cost of collateral. Foreign-owned banks therefore have a strong incentive to participate indirectly, taking advantage of a high C – A and a

150


low e (i.e. a high degree of internalization). The trend towards concentration in correspondent banking is thus largely self-fulfilling, particularly to the extent that economies of scale exist in monitoring. And, with the three large correspondent banks in the UK all major banks with diversified businesses, it is likely that they also have better access to their customer banks’ private information through other business lines than would smaller banks, and hence can offer a ‘cheaper’ correspondent service. Thus, despite the UK’s role as an international financial centre and a high foreign presence in sterling markets, just 15 per cent of daily value flowing through the large-value payment system is represented by the three foreign-owned direct participants.20 Case 2: imperfect monitoring and agent-specific collateral costs To assume perfect monitoring, as in case 1 is, perhaps, a bit strong. In this subsection, we relax this assumption to allow for Type I and Type II errors in monitoring. That is, with some probability, , bank A fails to ask for collateral and a default occurs (a Type I error); and with some probability, , bank A mistakenly identifies a state as a default state, and hence requests collateral unnecessarily (a Type II error). As bank A fails to obtain collateral in some default states, some uncollateralized exposure will be retained in the system under indirect participation. More specifically, when A monitors under full collateralization and indirect participation, C’s expected private costs are augmented by ( – )(C – eA) + . The first term captures the direct collateral cost incurred by C, adjusted for the correspondingly smaller pass-through of A’s collateral costs vis-à-vis the central bank. The second term captures costs imposed upon A, , in the event that C defaults and A is uncollateralized. Under the assumption that A is aware that it will retain uncollateralized exposure of the time, these costs would be passed on to bank C. But, of course, with ( – )(C – eA) + strictly positive when C > A, it is less likely that A’s correspondent service with monitoring will entail a lower cost to C than direct participation with full collateralization; that is, it is less likely that the inequality in (2) below will hold. C > M + ( + – )C + e(1 – ( + – ))A +

(2)

Imperfect monitoring therefore has significant implications for both private and social costs under each of the policy/participation states involving indirect participation. As it has the effect of increasing costs, indirect participation is less likely than in case 1. More formally, bank C’s expected costs under the four possible strategy pairs become: (Z,D): 0 (F,D): C


151

(Z,I): min[M + ( + – )C + ,C] (F,I): min[M + ( + – )C + e(1 – ( + – ))A + ,C] It remains the case that, under zero collateralization, C will choose direct participation. With the imperfection in monitoring, however, it becomes less likely that indirect participation with monitoring will be a low-cost outcome and hence more likely that C’s costs under (F,I) will be equivalent to those under (F,D). Hence, given that, if indifferent, C will participate directly, (F,D) is more likely to be favoured. The central bank faces the following expected social costs: (Z,D): S (F,D): C (Z,I): min[M + ( + – )C + ,C] (F,I): min[M + ( + – )C + e(1 – ( + – ))A + ,C] As before, it is known that C will choose to participate directly if Z is chosen and with (F,D) now more likely to be favoured by bank C in the event that the central bank chooses strategy F, the most relevant comparison may well be between expected social costs under (Z,D) and (F,D). Also, given that the costs associated with indirect participation and monitoring are higher in this scenario, the (unachievable) (Z,I) outcome is likely to be the socially preferred outcome for a much smaller interval of values for M (if, indeed, socially preferred at all). The foregoing has an interesting and important implication. In particular, given that A is more likely to resort to costly full collateralization when monitoring is imperfect and there is a risk of retaining uncollateralized exposure, the potential efficiency benefits associated with delegated monitoring are lost. Hence, it would appear that welfare could be improved if the banks accessing the system directly and providing correspondent banking services were ‘better monitors’: i.e. they had better skills or better information in this regard, which ensured that both and (and particularly the latter) were low. To the extent that monitoring quality is improved when banks are large and diversified, and, hence, have better access to private information, such banks should be encouraged to participate directly and to provide correspondent banking services. Figure 9.5 illustrates that imperfect monitoring creates a smaller range of monitoring costs for which C will choose to be an indirect participant (the threshold value of M moves from Y to X) and shows that the effect of better monitoring would be to narrow the horizontal distance between the perfect and imperfect monitoring thresholds. Case 3, however, offers a qualification to this conclusion where tiering imposes a spillover.

152

J.P. Jackson and M.J. Manning 0.0020 (Z,D) Social

0.0018 0.0016

0.0012

X

0.0010

(F,I ) Social Private

0.0008 0.0006 (Z,I ) Social Private

Social/private cost

0.0014


0.0004

(Z,D) Private

0.0002 0

0

0.0003

0.0006

0.0009

0.0012

0.0015

Monitoring cost (M)

Figure 9.5 The impact of imperfect monitoring (C = A = 0.0015; S = 1.1; = 0.0015; e = 0.5; = 0.00075; = 0.0005).

Case 3: imperfect monitoring and spillovers under indirect participation In the model, as presented in cases 1 and 2, indirect participation introduces no spillover risk to the payment system, or to the wider economy. Indirect participation is an equilibrium choice for C only when the central bank adopts a strategy of full collateralization; thus, even if C were to default with A uncollateralized, there would be no disruption to payments in the system and no spillovers. We do, however, allow for the possibility, under (F,I), that C’s default could impose costs on A, with an expected value of . While A seeks compensation from C for these expected losses (by passing through the expected cost), an actual loss of one unit would be suffered in those states of the world in which default occurred. This could cause liquidity, or even solvency, problems at A which might spill over into the system more widely. While we do not explicitly model the channels by which losses might be transmitted under such a scenario, we can attempt to capture possible spillovers by introducing an additional ‘tiering risk’ term, T, to expected social costs under indirect participation. By definition, T is an externality, and hence will not be internalized by C or A. It might be interpreted as reflecting the potential that A fails to fully internalize the risk that a failure of C will trigger A to suffer liquidity or solvency problems that could have knock-on effects on the financial system. In addition, it might capture the potential that A fails to internalize the risk that operational problems to itself would disrupt C’s payments. Expected social costs under (Z,I) and (F,I) thus become:


153

(Z,I): If C > M + ( + – )C + , then M + ( + – )C + T (F,I): If the inequality in (2) holds: M + ( + – )C + e(1 – ( + – ))A + T otherwise: C Allowing for this tiering externality, a wedge is introduced between social and private costs under full collateralization and indirect participation (illustrated in Figure 9.6). Specifically, there exists a range of monitoring costs, between X and Y in the diagram, in which (F,I) is preferred to (F,D) for C, but (F,D) is preferred to (F,I) for the central bank. That is: ∑ + T > C > ∑ +

(3)

where ∑ = M + ( + – )C + e(1 – ( + – ))A. The range of monitoring costs for which the inequality in (3) holds is increasing in T. However, we know from the inequality in (2) that, for C, (F,I) is only likely to be preferred to (F,D) for low values of . For larger values, the costs associated with indirect participation and monitoring would exceed those associated with full collateralization. Hence, it would never be optimal for indirect participation to be chosen by C, limiting the potential incidence of tiering externalities, and hence the potential impact of T. Nevertheless, the existence of this wedge indicates that there might be a role for additional policy intervention to mitigate tiering externalities. 0.0020 (F,I ) Social

0.0019 Y


0.0018

0.0016 0.0015 0.0014 0.0013 (F,I ) Private

Social/private cost

0.0017

X

0.0012 0.0011 0.0010

0

0.0001

0.0002

0.0003

0.0004

0.0005

Monitoring cost (M)

Figure 9.6 The impact of tiering risk (C = A = 0.0015; S = 1.1; = 0.0015; e = 0.5; = 0.00075; = 0.0005; T = 1.4).

154


Indeed, until now we have worked with a value of C’s payment obligation of unity. Generalizing this, by allowing C’s payment to take the value, P, we can capture the policy implications associated with payment system participants of different ‘sizes’. In particular, we find that the interval of monitoring costs over which social and private incentives under full collateralization are misaligned is increasing in the value of payments. That is, for a given error in monitoring, , and a given degree of spillover, T, higher payment values lead to higher potential uncollateralized exposures and hence potentially greater social losses. More formally, bank C’s and the central bank’s expected costs under full collateralization are given below. In these expressions, we assume that monitoring costs are invariant with respect to P.21 But we also recognize that total collateral costs and spillover costs will, ceteris paribus, be increasing in step with the value of payments. Hence, as P increases, the expected cost schedules shift upwards. Bank C: (F,D): PC (F,I): min{M + P[( + – )C + e(1 – ( + – ))A + ],PC} Central Bank: (F,D): PC (F,I): If M + P[( + – )C + e(1 – ( + – ))A + ] < PC, then M + P[( + – )C + e(1 – ( + – ))A + T]; otherwise, PC Figure 9.7, drawn in an analogous fashion to Figure 9.6, illustrates the implication of increasing payment values, or increasing size of payment system participants. Note the difference in the scales of the two figures, reflecting the fact that, with a ‘high P’, expected costs are significantly higher. The important observation from Figure 9.7 is that the interval X–Y covers a much wider range of monitoring costs and hence tiering spillovers may be a much more significant policy concern when ‘large’ payment system participants settle indirectly. Policy alternatives As a result, policymakers may wish to consider policy options that either reduce the size of the wedge, or encourage large payment system participants to settle directly. A number of policy options might be considered in this regard. First, banking supervisors might consider more stringent ex ante capital


155

0.0090 (F,I ) Social

Y


0.0085

X

0.0080

0.0070 0.0065

(F,I ) Private

Social/private cost

0.0075

0.0060 0.0055 0.0050 0

0.0003

0.0006

0.0009

0.0012

0.0015

Monitoring cost (M)

Figure 9.7 The impact of tiering risk with high payment value (C = A = 0.0015; S = 1.1; = 0.0015; e = 0.5; = 0.00075; = 0.0005; T = 1.4; P = 5).

and/or liquidity regulation to encourage correspondent banks to internalize the externalities associated with intraday credit extensions to indirect payment system participants. Second, steps might be taken to ensure the quality of monitoring carried out by correspondent banks, perhaps via improved accounting and disclosure standards for financial institutions. Third, direct participation might be encouraged by efforts to reduce either the quantum of liquidity required to effect payments in the system; or the opportunity cost of collateral faced by prospective direct members. The liquidity burden might be addressed by introducing a more liquidity-efficient payment system design; e.g. introducing queuing or netting algorithms, or allowing certain payments to be settled on a deferred net basis. And collateral costs might be lowered by broadening the eligible collateral list to include less liquid assets; or allowing greater fungibility of collateral across borders or across systems.22 Were such measures to be successful in encouraging direct participation of large and diversified banks that enjoyed superior access to private information and were such banks to then offer correspondent banking services, this might also ultimately achieve an improvement in monitoring quality.

Conclusions The model developed in this chapter can be used to examine the key factors influencing both an agent’s decision over whether to participate directly in a payment system and a central bank’s decision as to whether to require collateralization of intraday credit extensions to payment system participants.

156


Consistent with the existing literature in this area, we show that a central bank will be more likely to require full collateralization of intraday credit when participants have high default probabilities, or face low collateral costs. However, a contribution of this chapter is to show that, for full collateralization to be a rational policy choice for the central bank, it is only necessary that a subset of agents have low collateral and monitoring costs; other agents can take advantage of these low costs if they become indirect participants. We also find that internalization is likely to make indirect participation more attractive under full collateralization. This suggests that economies of scale exist in correspondent banking and implies that significant concentration is likely to be observed in the provision of correspondent banking services (particularly where a subset of agents face particularly low collateral posting costs). And to the extent that economies of scale also exist in monitoring, one would expect banks to gravitate towards the larger, cheaper service-providers. Collateral pooling benefits, which we do not model here, are also likely to support such concentration. The model’s predictions are, prima facie, consistent with UK experience, with full collateralization, a high degree of indirect participation, and significant concentration in correspondent banking all key features of the UK landscape. Although the model does not capture legal and operational risks, and hence the full implications of internalization, our model can go some way towards offering some policy guidance as to whether such a profile of participation is desirable. Given that the agent’s and central bank’s decisions are taken sequentially, we show that, under certain circumstances, the first-best outcome might not be achievable. In particular, we find that zero collateralization and indirect participation might be optimal for a range of monitoring costs, but that, if the central bank chooses zero collateralization, the agent will always find it privately optimal to access directly. We do show, however, that zero collateralization and indirect participation is likely to be socially optimal for a smaller range of monitoring costs as monitoring becomes less perfect. When we allow for both imperfect monitoring and tiering spillovers, a wedge also emerges between private and social choices under full collateralization. In particular, we show that there will exist a range of monitoring costs in which the bank will prefer to participate indirectly under full collateralization, while the central bank would prefer direct participation. While we do not model the precise channels by which such spillovers arise, we can draw some broad policy conclusions in this regard. In particular, we show that the key determinants of the size of this wedge will be: the error in monitoring; the magnitude of any spillover, reflecting both the size of exposures arising through correspondent banking and the spillover per unit of exposure; and the cost of collateral if participating directly. This implies that policy to reduce the size of the wedge should be directed towards: (i) ensuring the capacity of correspondent banks to absorb either capital losses or liquidity shocks arising from the failure of a customer bank, perhaps via enhanced ex ante solvency and/or prudential regulation; (ii) encouraging correspondent banks


157

to improve monitoring quality; and (iii) facilitating a low opportunity cost of posting collateral for all prospective payment system members. Our analysis could be improved by a more complete and sophisticated treatment of the interaction between payment system participants and the provision of payments services. For example, with just one agent making choices and a single payment made, we cannot address the implications of factors such as collateral pooling, or the intraday liquidity management game. And, with the direct members A and B playing only a passive role in this model, we cannot assess the risk they bring to the system, and hence cannot perform a complete welfare analysis. Finally, a more detailed analysis of tiering spillovers would be useful, in order to refine our policy conclusion that such spillovers should be addressed in regulatory design.

Notes 1 The views expressed in this chapter are those of the authors, and not necessarily those of the Bank of England. We would like to thank the following for helpful comments during the preparation of this work: Victoria Saporta, Steve Millard, Matthew Willison, Ana Lasaosa, Jochen Schanz and Will Roberds. 2 For example, in September 2005, participants of the CHAPS and CREST systems used an average of £65 billion of intraday liquidity at the Bank of England to facilitate settlement. This includes more than £50 billion generated by self-collateralizing repos in CREST. 3 See also BIS (2005) for a discussion of alternative hybrid system designs. 4 In some securities settlement systems, where the cash leg is settled gross (equally relevant to the issues considered in this chapter), the cost of generating cash liquidity is significantly reduced via the implementation of self- or auto-collateralization techniques. CREST and Euroclear France, for instance, apply such procedures, allowing the immediate pledge/repo of (eligible) securities to the central bank to generate liquidity to fund their own purchase. 5 Other possible central bank policies, such as restricting access or imposing quantity limits, are not considered in this chapter. 6 Our model cannot capture the potential effects of collateral pooling because we only consider the decision of a single agent. 7 Specifically we use the framework in arrangements 4 and 5 of Kahn and Roberds (2006), which deal with payments settling across central bank accounts. 8 We later generalize the value of the payment. 9 It is assumed that Bank C is eligible to participate directly in the system, but that the central bank’s preferences over direct versus indirect participation by Bank C, from the perspective of social cost, will be reflected in the collateral policy chosen. 10 We assume that only Bank C faces the possibility of an exogenous default shock in this model. A more complete framework might allow Bank A to suffer such a shock also. We note in our discussion the potential implications for our results of this simplification. 11 With only a single payment here, we rule out the possibility of strategic behaviour among settlement banks. In particular, we abstract from the possibility that banks delay outgoing payments until incoming payments have arrived so as to economize on collateral costs. This behaviour is well documented in the literature, e.g. Bech and Garratt (2003). 12 We assume that the central bank’s collateral policy is applied to all direct members. However as Banks A and B take no decisions and make no payments on their own

158

13 14

15 16 17 18 19 20

21 22


behalf in this model, it is not necessary that the central bank consider the implications of its collateral policy on these agents’ behaviour. Consistent with empirical observation, we assume that the central bank either does not have the capacity, or finds it excessively costly, to monitor. In practice, these costs are often not passed on explicitly. Given that there is a clear economic rationale for full pass-through, it is likely that these costs are fully reflected in the price for a bundle of services provided by correspondent banks, which includes monitoring costs. Furthermore, explicit charging for intraday liquidity may become increasingly common in future, as payments become more time-critical intraday. Alternatively, it may be that the providers of these services simply face very low opportunity costs to posting collateral. It is worth noting that the linearity of all components of social cost implies that a central bank policy of partial collateralization will never be a dominant strategy. This might be justified if technological costs and the profit component of costs under indirect participation, which are normalized to zero here, exceeded the fixed costs of joining and accessing the payment system directly. It is worth noting that, under full collateralization, the central bank’s and Bank C’s expected costs are the same. Such an intraday credit policy is adopted by most G10 central banks, although some notable exceptions do exist, in particular the Federal Reserve, which charges an explicit fee on intraday overdrafts. It is thought that upwards of 20 per cent of sterling large-value payments are internalized across the accounts of correspondent banks. It is worth noting that in many comparable economies, large value payment arrangements are far less highly tiered. For example in the United States the Fedwire system has over 7,000 member banks, and the Japanese large-value payment system BOJ-NET has over 300 members, compared with 15 members of the UK CHAPS system and 14 members of the Canadian LVTS system. Several factors might help to explain the structural differences that can be observed between countries. In some jurisdictions authorities have made greater efforts to encourage wider membership of large value payment systems, either through imposing a specific regulatory requirement, applying moral suasion, or subsidizing the cost of such payment arrangements. In addition the impact of prudential liquidity requirements in encouraging concentration is not relevant in a number of jurisdictions. Finally, some countries have historically always had highly concentrated banking systems (this is true of the United Kingdom) while in others highly fragmented banking arrangements are observed. This assumption seems reasonable, although one could argue that a bank’s monitoring intensity might increase when the size of its potential exposures was greater. A more extreme policy option might be to simply compel certain payment system participants to settle directly. However, even if these banks were required to open settlement accounts with the central bank, it is not clear that they could be compelled to actually use them.

References Bank for International Settlements (BIS) (2005) New developments in large-value payment systems, Committee on Payment and Settlement Systems Publication No. 67. Bech, M. and Garratt, R. (2003) ‘The intraday liquidity management game’, Journal of Economic Theory, 109(2): 198–219. Chakravorti, S. (2000) ‘Analysis of systemic risk in multilateral net settlement systems’, Journal of International Financial Markets, Institutions and Money, 10: 9–30.


159

Furfine, C.H. and Stehm, J. (1998) ‘Analyzing alternative intraday credit policies in realtime gross settlement systems’, Journal of Money Credit and Banking, 30 (4): 832–48. Harrison, S., Lasaosa, A. and Tudela, M. (2005) ‘Tiering in UK payment systems: credit risk implications’, Bank of England Financial Stability Review. Kahn, C.M. and Roberds, W. (2001) ‘Real-time gross settlement and the costs of immediacy’, Journal of Monetary Economics, 47: 299–319. Kahn, C.M. and Roberds, W. (2006) ‘Payments settlement: tiering in private and public systems’, unpublished thesis, University of Illinois. Manning, M.J. and Willison, M.D. (2006), ‘Modelling the cross-border use of collateral in payment systems’, Bank of England Working Paper No. 286. Mills, D.C. (2005), ‘Alternative central bank credit policies for liquidity provision in a model of payments’, Board of Governors of the Federal Reserve System Finance and Economics Discussion Series No. 2005–55. Willison, M. (2005), ‘Real-Time Gross Settlement and hybrid payment systems: a comparison’, Bank of England Working Paper No. 252.

10 Central banks’ interest calculating conventions Deviating from the intraday/overnight status quo George Speight, Matthew Willison, Morten Bech and Jing Yang1

Introduction Central banks lend central bank balances and accept deposits on terms designed to ensure that the overnight market interest rate is close to its ‘policy’ rate, which it sets to meet its monetary policy objectives. This rate anchors market interest rates for longer maturities. However, central banks also typically lend central bank balances and accept deposits at low or zero interest rates intraday. This anchors the intraday market interest rate at or close to zero, in so far as money is traded intraday at all. Dale and Rossi (1996) show that the central bank can set a low or zero intraday interest rate and ensure the overnight rate is close to its (higher) policy rate because all intraday lending is repaid by the end of the day; there is no spillover overnight. The existence of a distinction between intraday and overnight interest rates has its origins in the move from settling large-value payments on a deferred net settlement basis to settling them on a gross basis in real time. (A description of the diffusion of RTGS across the world’s large-value payment systems can be found in the chapter by Bech in this volume.) When banks exchanged central bank money at the end of each day there was simply no need for banks or the central bank to lend at maturities of less than one day because money was not being exchanged on a more frequent basis than once per day.2 But in a RealTime Gross Settlement (RTGS) payment system, and in equivalent ‘hybrid’ payment systems, it is possible to lend central bank money for periods strictly within a day. Indeed, the prospect of a genuine market in intraday money is one which generates great interest among market practitioners and policy makers alike. So far, central banks’ practice of providing intraday credit without charge or at a very low rate (and similarly not remunerating positive intraday balances) has been sufficient to hold back any such development.3 Broadly speaking, central banks have not charged interest on the basis of balances during the day because of a belief that if they did, banks would have a strong incentive to delay payments until late in the day. It would be individually

Banks’ interest calculating conventions

161

rational for banks to wait until late in the day in the hope that incoming payments would provide them with the liquidity they needed to make their payments, rather than having to borrow from the central bank. Of course, if all banks followed the same strategy none would actually gain from delaying payments. Payment delays are not desirable because they can imply the following risks and inefficiencies in the payment system: 1 2

3

the failure to receive payments in a timely manner may represent a liquidity risk to recipients; delay could also reduce the number of times liquidity is recycled in the payment system during the day, which may actually increase the amount of intraday credit required for payments to settle; delay of payments until late into the day implies greater exposure of the payment system to operational risk since if an operational problem hits later during the day there is less scope to overcome it in time for payments to settle.

This chapter questions this line of reasoning. It does this by analysing the effects of the central bank imposing its interest constraint more frequently. (We choose more rather than less frequently because it moves us towards the real-time paradigm, but the results are broadly generalizable to any frequency.) We find that banks would indeed delay payments, to the extent that they had no reason to send them more promptly. But it may be that in many cases, customers would opt to send payments in the morning rather than the afternoon, because money would now have value on a more frequent basis. To assume that banks would delay all payments until the end of the day is to take for granted that customers would continue in all cases to contract to make payments on an ‘end of day’ basis. However, this is not necessarily the case. Either way, payments would be compressed against deadlines to a greater extent than at present. But if customers opted to make some payments early, the payments would at least be spread across the day and would not be concentrated at the end of the day. Also, the change would likely affect the amount of intraday credit which banks extend to their customers. There may be customers who would be content for their banks systematically to delay their payments until later in the day. For these customers, settlement banks would extend less credit (at least on a duration-weighted basis) than if their payments were spread more evenly across the day: they would only incur intraday overdrafts towards the end of the day. And if the frequency with which settlement banks actively monitor customers’ accounts is anchored by how often they calculate interest, then a shift to calculating interest twice a day may mean that settlement banks’ charges to their customers become more closely related to the amount of credit the customers actually use. So if this bank-imposed ‘monitoring constraint’ contributes to credit being extended beyond the socially optimal level, the change in interest

162

G. Speight et al. Interest rate

Next possible policy change Policy rate

End of End half day of day

Time

Figure 10.1 Change in shape of yield curve.

calculating convention may generate financial stability benefits by easing this constraint. These effects could produce system-wide risk-reducing financial stability benefits if the private costs which settlement banks face in extending intraday credit are lower than the social costs. The chapter works through these effects in a simple scenario: where the central bank imposes its policy rate at the end of every half day rather than at the end of every day (Figure 10.1). Within each half day it provides credit at no charge and does not remunerate positive balances. The central bank now constrains the market rate of interest to trade at close to its policy rate between half-day periods.4 Our analysis in this chapter is very much in the spirit of a thought experiment, to begin exploring the issues. There are several important factors that we do not analyse. One is whether the possible financial stability benefits we identify could be achieved via other means. For example, as shown in Willison (2005), concentrating payments activity in smaller periods of the day in order to reduce the duration of exposures between settlement banks and customers could also be achieved by introducing liquidity saving features such as payment offsetting. Any wedge between private and social costs of intraday credit could be reduced by imposing capital requirements that bind more frequently than at the end of the day. It is also possible that the private sector may move to monitoring and charging of intraday credit on a more timely basis on its own accord as information technology costs decline. Another factor we do not take into account is the costs of shifting from one way of calculating interest to another (e.g. costs of upgrading IT systems). Of course, these costs would be mainly up front and would need to be contrasted with any recurring longer-term benefits. The chapter proceeds as follows. We first review the existing literature before summarizing the monetary stability implications of an increase in the frequency at which the central bank calculates interest. We then set out our model and work through the implications for financial stability of the change in how interest is calculated. Finally, we consider some further issues before offering some conclusions.


163

Literature review Angelini (1998), Bech and Garratt (2003) and Kobayakawa (1997) each develop game-theoretic models of an RTGS payment system which emphasize how banks’ decisions about when to make payments during the day depend on other banks’ decisions and that banks’ individual incentives can lead to an inefficient outcome when intraday credit is costly. They show that a bank may choose to delay payments when other banks, from which it receives payments, do not delay because the gain from reducing the amount of intraday credit it must obtain exceeds the cost of delaying. Since all banks have this incentive to delay, all banks delay their payments in equilibrium. But the equilibrium outcome is inefficient since intraday credit needs are unchanged and all banks incur delay costs. Banks would each be better off if none of them delayed. The central bank can reduce the likelihood of delay arising in equilibrium by reducing the cost of intraday credit; e.g. by setting a zero intraday interest rate. Other papers also explore the adverse effects of payment delays and the relationship these effects have with the cost of intraday credit using models where the timing of payments is treated as exogenous but where it is assumed that payment times are asynchronous. Freeman (1996, 2002) considers an economy where in any period agents need liquidity to make payments and before other agents from who they are due payments can make theirs. Agents can go to the market to obtain liquidity but if there is insufficient market liquidity they will be forced to delay some payments. Freeman (2002) shows that the liquidity constraints in the payment system can be alleviated if a central bank extends credit. Agents that need liquidity early borrow from the central bank and repay the central bank once they have received payments. Further, liquidity constraints in the payment system are completely eliminated when the central bank lends at a zero interest rate. Kahn and Roberds (2001), Martin (2004) and Zhou (2000) develop similar models and demonstrate that the central bank should lend intraday at a zero interest rate. In summary, most of the existing literature concludes that a central bank should always extend intraday credit at a zero interest rate to avoid liquidity constraints and delays occurring in RTGS payment systems.

Implications of our thought experiment for the implementation of monetary policy This section discusses the implications for implementation of monetary policy of the central bank calculating interest on settlement banks’ accounts at mid-day as well as at the end of the day. Broadly speaking, this would have relatively little effect on the central bank’s ability to implement monetary policy. Woodford (2003) shows that when central bank money is the ultimate settlement asset, central banks can control the price of their money at a particular maturity by standing willing to accept deposits at one rate and lend at another, thereby bounding the money market rate of interest for similar duration loans on

164

G. Speight et al.

the lower and upper sides, respectively. Central banks typically target the overnight interest rate. Dale and Rossi (1996) show that the implementation of monetary policy is not compromised by the central bank setting a lower intraday interest rate than the central bank’s overnight policy rate, provided that intraday credit has to be repaid in full by the end-of-day. By imposing this ‘quantity constraint’, the central bank ensures that overnight borrowers have to pay the overnight cost of central bank money. In our thought experiment, there is no reason why the central bank’s ability to implement monetary policy should be reduced. Banks face the same conditions for their holdings of central bank money at mid-day as at the end of the day under the current regime. It follows that banks will treat central bank money in the same way across the mid-day point as they do currently across the end-ofday point and the central bank would exercise the same control over market rates of interest as it does at present, but now starting from the maturity of half a day.5 The shortest maturity point on the yield curve would shift from one day to half a day. It is difficult to believe that the change would have any first-order macroeconomic implications, e.g. for the speed or effectiveness of the transmission of monetary policy. The inflation control mechanism would essentially be unaltered. However, to enforce a new regime with a shorter interest-calculating period, the central bank may need to implement operational changes. It would need to calculate interest on accounts and supply the necessary funds to meet the reserve maintenance requirement more regularly. It may also need to conduct open market operations (OMOs) more frequently, though this would depend on whether it had a reserve-averaging framework and, if so, on its length given that a central bank needs to observe the general principle that there should be at least one OMO round per reserve maintenance period.

Implications of our thought experiment for system-wide risk in the financial system The implications for system-wide risk or financial stability depend on the pattern of payments. The model sets out in a more formal way how the pattern of payments would be determined. Payments fall into two kinds: •

•

payments on behalf of customers where the customer does not express any preference for which time of day he wants his payment made; it seems reasonable to assume that end of day would remain the ‘default’ for these customers; payments which the bank has a particular reason for sending in a specific half of the day including customer payments, where the customer has requested the payment be sent in a particular half of the day, and some payments on the bank’s own account (e.g. for funding, trading or risk management purposes).


165

The risk implications are different for the different kind of payments. Overall, the key implications are for operational risk and credit risk in the payment system. There will be an increase in operational risk, regardless of the mix of the two kinds of payments. If a bank submits a payment to the payment system close to the deadline by which it should settle, and at that time there is an operational problem either with the bank or with the system itself, there is less scope to overcome the problem and allow the payment to be settled by the deadline. So as a general principle, as the time between a bank submitting a payment and the deadline for that payment to settle falls, operational risk increases. Payments of the first kind are systematically delayed until the second half of the day, reducing the average amount of time available for them to settle. Payments of the second type are submitted in the half of the day when they are expected to settle. Customers who had been expecting to receive a payment by a certain time but do not and who are not insured against this risk, clearly face a cost: either they will fail to meet their own commitments and face default or delay penalties, or they will have to resort to (possibly expensive) funding from alternative sources, or they may postpone consumption or investment. Sending customers whose payments fail may have to pay compensation, which may or may not cover the costs faced by receiving customers. Importantly, there may be knockon consequences for other agents, if the customer expecting to receive a payment needed it to fulfil an obligation to another customer, which is in turn delayed or fails. The social cost may be greater than the net private costs. A change in the way that the central bank calculates interest could also affect the amount of intraday credit which settlement banks provide to their customers and its price. Where banks systematically delay payments until late in the day, such payments will typically require less credit in order to settle. Holding other things constant, the duration of any extensions of credit occurring within the day will be shorter compared to when some payments are made earlier in the day. This will have implications for system-wide credit risk if the private cost to settlement banks of extending credit within the day does not fully reflect the social costs. There could be a wedge between private and social costs if banks fail to internalize the impact of a customer default on the wider financial system. If this externality exists, a fall in the duration of intraday credit may benefit financial stability. There could be further benefits for financial stability if the change in the central bank’s interest-charging convention affects how intensely banks monitor exposures to their customers. When the central bank only charges and remunerates at the end of the day, banks may not actively monitor customers’ actual balances through the day. Indeed, this has typically been the case: evidence presented in Harrison et al. (2005) suggests that, broadly speaking, banks have only tended to watch for limit breaches on a day to day basis. So any charge which they make for intraday overdrafts – including a charge to cover the credit

166

G. Speight et al.

risk they bear – may be only loosely linked to the amount of risk actually faced within the day, thereby contributing to any wedge between private and social costs. A central bank, by calculating interest at mid-day, and thus inducing settlement banks into doing the same, can ease this bank-imposed monitoring constraint and as a result reduce the private–social cost wedge. The remainder of this section develops these arguments more thoroughly, using a simple formal model to highlight the key issues. The model represents a single day in an RTGS payment system. There are two banks (bank 1 and bank 2) that are members of the RTGS system (they are ‘settlement’ banks), each with one customer (customer 1 and customer 2). Each customer has one payment with unit value to make to the other customer during the day. This is common knowledge. For simplicity we assume that settlement banks and customers both begin the day with zero balances. This implies that to make a payment the first settlement bank must run an overdraft with the central bank. Likewise, the first customer to make a payment must run an overdraft with their settlement bank. Time within the day is divided into two periods: morning and afternoon. When customers trade they write contracts with one another. A contract specifies the value of the payment (unity), and the period or periods within the day in which the payment should be made. The central bank requires extensions of intraday credit to be fully collateralized. We assume that a settlement bank faces a cost (C) when it posts collateral with the central bank. Given that time is modelled as discrete, we assume that a settlement bank has to obtain intraday credit from the central bank if it makes its payment in the period before it receives a payment and if it makes it in the same period as it receives a payment.6 Other costs include a delay cost (D), which a settlement bank incurs when it delays making its payment until the afternoon. This delay cost captures the greater risk that payments fail due to operational problems if they are delayed until the afternoon. A settlement bank also incurs a cost in extending credit to its customer from the morning to the afternoon (E) which it may not be able to control through other means and which is not fully reflected in the charges faced by customers. The final cost that settlement banks face is the interest incurred or received (R) on their accounts at the central bank at mid-day. Zero intraday interest rates The case where interest is calculated on account balances at the end of each day only – i.e. the current situation – serves as the benchmark case. Customers make their payment requests at the beginning of the day. The rationale for this is that they do not incur or receive interest according to their balance positions intraday, and that they wish to minimize delay costs. Each settlement bank then faces a choice between making its payment in the morning or the afternoon. Since each bank’s costs depend on the other bank’s decision as to when to make its payment, we model the situation as a simultaneous-move game. The banks’ costs are shown in the strategic-form game below.


167

Table 10.1 Pay-offs when central bank charges and remunerates at end of day only Bank 2 Bank 1

Morning Afternoon

Morning C,C D,C + E

Afternoon C + E,D C + D,C + D

Table 10.2 Equilibrium strategies when the central bank charges and remunerates at end of day only Bank 1’s decision

Bank 2’s decision

Providing that

Morning Afternoon Morning or afternoon Morning or afternoon

Morning Afternoon Afternoon or morning Morning or afternoon

C and E < D C and E > D E

The Future of Payment Systems (Routledge International Studies in Money and Banking)

Understanding Pensions (Routledge International Studies in Money and Banking)

Money and Payments in Theory and Practice (Routledge International Studies in Money & Banking)

Central Banking in Eastern Europe (Routledge International Studies in Money and Banking)

Money, Uncertainty and Time (Routledge International Studies in Money and Banking)

Money, Credit and Price Stability (Routledge International Studies in Money and Banking)

Institutional Change in the Payments System and Monetary Policy (Routledge International Studies in Money and Banking)

The Structure of Financial Regulation (Routledge International Studies in Money and Banking)

The Role of Central Bank Money in Payment Systems

Monetary and Banking History: Essays in Honour of Forrest Capie (Routledge International Studies in Money and Banking)

Dollarization: Lessons from Europe for the America (Routledge International Studies in Money and Banking, 22)

Monetary Stability in Europe (Routledge International Studies in Money and Banking, 16)

Tax Systems and Tax Reforms in South and East Asia (Routledge International Studies in Money and Banking)

Organisational Change and Retail Finance: An Ethnographic Perspective (Routledge International Studies in Money and Banking)

Bank Deregulation and Monetary Order (Routledge International Studies in Money and Banking, 2)

The Future of Banking

Financial Geography: A Banker's View (Routledge International Studies in Money and Banking)

Monetary Macroeconomics: A New Approach (Routledge International Studies in Money and Banking, 15)

Finance: A Characteristics Approach (Routledge International Studies in Money and Banking, Number 7)

Banking on Innovation: Modernisation of Payment Systems (Contributions to Economics)

The Future of Money

The Routledge International Handbook of Globalization Studies (Routledge International Handbooks)

Monetary Policy and Unemployment: The US, Euro-area and Japan (Routledge International Studies in Money and Banking)

Monetary Policy in Central Europe (Routledge International Stuidies in Money and Banking)

Handbook of International Banking

Banking on Basel: The Future of International Financial Regulation

Handbook of International Banking

Information Systems: Critical Perspectives (Routledge Studies in Ortganization and Systems)

Islam and the Future of Money

Islam and the Future of Money

The Future of Payment Systems (Routledge International Studies in Money and Banking)

Understanding Pensions (Routledge International Studies in Money and Banking)

Money and Payments in Theory and Practice (Routledge International Studies in Money & Banking)

Central Banking in Eastern Europe (Routledge International Studies in Money and Banking)

Money, Uncertainty and Time (Routledge International Studies in Money and Banking)

Money, Credit and Price Stability (Routledge International Studies in Money and Banking)

Institutional Change in the Payments System and Monetary Policy (Routledge International Studies in Money and Banking)

The Structure of Financial Regulation (Routledge International Studies in Money and Banking)

The Role of Central Bank Money in Payment Systems

Monetary and Banking History: Essays in Honour of Forrest Capie (Routledge International Studies in Money and Banking)

Dollarization: Lessons from Europe for the America (Routledge International Studies in Money and Banking, 22)

Monetary Stability in Europe (Routledge International Studies in Money and Banking, 16)

Tax Systems and Tax Reforms in South and East Asia (Routledge International Studies in Money and Banking)

Organisational Change and Retail Finance: An Ethnographic Perspective (Routledge International Studies in Money and Banking)

Bank Deregulation and Monetary Order (Routledge International Studies in Money and Banking, 2)

The Future of Banking

Financial Geography: A Banker's View (Routledge International Studies in Money and Banking)

Monetary Macroeconomics: A New Approach (Routledge International Studies in Money and Banking, 15)

Finance: A Characteristics Approach (Routledge International Studies in Money and Banking, Number 7)

Banking on Innovation: Modernisation of Payment Systems (Contributions to Economics)

The Future of Money

The Routledge International Handbook of Globalization Studies (Routledge International Handbooks)

Monetary Policy and Unemployment: The US, Euro-area and Japan (Routledge International Studies in Money and Banking)

Monetary Policy in Central Europe (Routledge International Stuidies in Money and Banking)

Handbook of International Banking

Banking on Basel: The Future of International Financial Regulation

Handbook of International Banking

Information Systems: Critical Perspectives (Routledge Studies in Ortganization and Systems)

Islam and the Future of Money

Islam and the Future of Money

Recommend Documents