Market Infrastructures - Risksubscriptions.risk.net/wp-content/uploads/2016/03/JFMI.pdf · Incisive...

Incisive Media, Haymarket House, 28-29 Haymarket, London SW1Y 4RX

PEFC Certified

This book has been produced entirely from sustainable papers that are accredited as PEFC compliant.

www.pefc.org

Volume 4 Number 2December 2015

The Jo

urn

al of Fin

ancial M

arket Infrastru

ctures

Volume 4 N

umber 2 D

ecember 2015

■ Identification of over and under provision of liquidity in real-time payment systemsEdward Denbee, Rodney J. Garratt and Peter Zimmerman

■ Reaction functions of the participants in Colombia’s large-value payment systemConstanza Martínez and Freddy Cepeda

■ Interoperability between central counterpartiesJürg Mägerle and Thomas Nellen

FORUM■ Central counterparties in crisis: the Hong

Kong Futures Exchange in the crash of 1987Robert T. Cox

The Journal of

Financial MarketInfrastructures

JFMI_4_2_1215.indd 1 01/12/2015 14:38

Tria

l Cop

y For all subscription queries, please call:

UK/Europe: +44 (0) 207 316 9300

USA: +1 646 736 1850 ROW: +852 3411 4828

in numbers

140,000

Users

Page views

19,400+ on Regulation

6,900+ on Commodities

19,600+ on Risk Management

6,500+ on Asset Management

58,000+ articles stretching back 20 years

200+

New articles & technical papers

370,000

21,000+ on Derivatives £

Visit the world’s leading source of exclusive in-depth news & analysis on risk management, derivatives and complex fi nance now.

(each month)

(each month)

See what you’re missing

(each month)

RNET16-AD156x234-numbers.indd 1 21/03/2016 09:44

The Journal of Financial Market InfrastructuresEDITORIAL BOARD

Editor-in-ChiefRon Berndsen De Nederlandsche Bank and Tilburg University

Associate EditorsSujit Chakravorti The Clearing House Mark Manning Reserve Bank of AustraliaMassimo Cirasino World Bank Alistair Milne Loughborough UniversityRodney Garratt University of California Masayuki Mizuno Bank of JapanTerry Goh Monetary Authority of Singapore Daniela Russo European Central BankGerard Hartsink CLS Bank International Edwin Schooling Latter FinancialRichard Heckinger Federal Reserve Bank Conduct Authority (UK)of Chicago Manmohan Singh Washington, DC

Ronald Heijmans DNB Jeff Stehm Promontory Financial GroupLex Hoogduin University of Amsterdam Lawrence Sweet Federal Reserve Bank ofCharles Kahn University of Illinois New YorkThorsten Koeppl Queen’s University John Trundle Euroclear UK & IrelandEsmond Lee Hong Kong Monetary Authority Leo Van Hove Free University of BrusselsGottfried Leibbrandt SWIFT Wolf Wagner Erasmus UniversityCarlos León Central Bank of Colombia Stuart E. Weiner Santa Fe Group andKlaus Löber Bank for International Dartmouth CollegeSettlements Froukelien Wendt Washington, DC

SUBSCRIPTIONSThe Journal of Financial Market Infrastructures (Print ISSN 2049-5404 j Online ISSN 2049-5412)is published quarterly by Incisive Risk Information Limited, Haymarket House, 28–29 Haymarket,London SW1Y 4RX, UK. Subscriptions are available on an annual basis, and the rates are set outin the table below.

UK Europe USRisk.net Journals £1945 €2795 $3095Print only £735 €1035 $1215Risk.net Premium £2750 €3995 $4400

Academic discounts are available. Please enquire by using one of the contact methods below.

All prices include postage.All subscription orders, single/back issues orders, and changes of addressshould be sent to:

UK & Europe Office: Incisive Media (c/o CDS Global), Tower House, Sovereign Park,Market Harborough, Leicestershire, LE16 9EF, UK. Tel: 0870 787 6822 (UK),+44 (0)1858 438 421 (ROW); fax: +44 (0)1858 434958

US & Canada Office: Incisive Media, 55 Broad Street, 22nd Floor, New York, NY 10004, USA.Tel: +1 646 736 1888; fax: +1 646 390 6612

Asia & Pacific Office: Incisive Media, 20th Floor, Admiralty Centre, Tower 2,18 Harcourt Road, Admiralty, Hong Kong. Tel: +852 3411 4888; fax: +852 3411 4811

Website: www.risk.net/journal E-mail: [email protected]

To subscribe to a Risk Journal visit Risk.net/subscribe or email [email protected]

http://subscriptions.risk.net/subscribe/

mailto:corpsubs%40risk.net?subject=

The Journal of Financial Market InfrastructuresGENERAL SUBMISSION GUIDELINES

Manuscripts and research papers submitted for consideration must be original workthat is not simultaneously under review for publication in another journal or otherpublication outlets. All articles submitted for consideration should follow strict aca-demic standards in both theoretical content and empirical results. Articles should beof interest to a broad audience of sophisticated practitioners and academics.

Submitted papers should follow Webster’s New Collegiate Dictionary for spelling,and The Chicago Manual of Style for punctuation and other points of style, apart froma few minor exceptions that can be found at www.risk.net/journal. Papers should besubmitted electronically via email to: [email protected]. Please clearlyindicate which journal you are submitting to.

You must submit two versions of your paper; a single LATEX version and a PDFfile. LATEX files need to have an explicitly coded bibliography included. All files mustbe clearly named and saved by author name and date of submission. All figures andtables must be included in the main PDF document and also submitted as separateeditable files and be clearly numbered.

All papers should include a title page as a separate document, and the full names,affiliations and email addresses of all authors should be included. A concise andfactual abstract of between 150 and 200 words is required and it should be includedin the main document. Five or six keywords should be included after the abstract.Submitted papers must also include an Acknowledgements section and a Declarationof Interest section. Authors should declare any funding for the article or conflicts ofinterest. Citations in the text must be written as (John 1999; Paul 2003; Peter and Paul2000) or (John et al 1993; Peter 2000).

The number of figures and tables included in a paper should be kept to a minimum.Figures and tables must be included in the main PDF document and also submittedas separate individual editable files. Figures will appear in color online, but willbe printed in black and white. Footnotes should be used sparingly. If footnotes arerequired then these should be included at the end of the page and should be no morethan two sentences. Appendixes will be published online as supplementary material.

Before submitting a paper, authors should consult the full author guidelines at:

http://www.risk.net/static/risk-journals-submission-guidelines

Queries may also be sent to:The Journal of Financial Market Infrastructures, Incisive Media,Haymarket House, 28–29 Haymarket, London SW1Y 4RX, UKTel: +44 (0)20 7004 7531; Fax: +44 (0)20 7484 9758E-mail: [email protected]




The Journal of

Financial MarketInfrastructures

The journalToday, in the light of the financial crisis, it has become part of the political agendato strengthen payment, clearing and settlement systems, as well as repositories fordata on the trades they process. In 2012 a new set of internationally agreed CPSS-IOSCO Principles carved out financial market infrastructures (FMIs) as a distinctarea in financial policy. The Journal of Financial Market Infrastructures is the firstjournal to focus on this exciting and dynamic sector, and aims to bring together acommunity of contributors from the constituent sectors to analyze FMIs to further thedevelopment of this emerging field. The journal provides a balanced representation ofacademic and practitioner-focused papers that are dedicated to analyzing operationaland regulatory effectiveness and efficiency of payment, clearing, settlement and traderepository systems; and the risks they manage, transmit and create.

The Journal of Financial Market Infrastructures considers submissions in theform of technical papers and policy-oriented papers (forum discussions) on topicsincluding, but not limited to, the following:

� systemically important payment systems,� securities settlement systems,� central counterparties,� central securities depositories,� trade repositories,� settlement risk and other FMI-related risks including interdependencies,� infrastructure-related systemic risk,� network analysis of an FMI,� critical service providers and non-bank payment service providers,� correspondent banking,� FMI liquidity and collateral management,� exchanges and multilateral trading platforms,� oversight and supervision of FMIs, and� FMI-related standardization and legislation.




The Journal of Financial Market Infrastructures Volume 4/Number 2

CONTENTS

Letter from the Editor-in-Chief vii

RESEARCH PAPERSIdentification of over and under provision of liquidity in real-timepayment systems 1Edward Denbee, Rodney J. Garratt and Peter Zimmerman

Reaction functions of the participants in Colombia’s large-valuepayment system 21Constanza Martínez and Freddy Cepeda

Interoperability between central counterparties 49Jürg Mägerle and Thomas Nellen

FINANCIAL MARKET INFRASTRUCTURES FORUMCentral counterparties in crisis: the Hong Kong Futures Exchange inthe crash of 1987 73Robert T. Cox

Editor-in-Chief: Ron Berndsen Subscription Sales Manager: Aaraa JavedPublisher: Nick Carver Global Head of Sales: Michael LloydJournals Manager: Dawn Hunter Information and Delegate Sales Director: Michelle GodwinEditorial Assistant: Carolyn Moclair Composition and copyediting: T&T Productions LtdMarketing Executive: Giulia Modeo Printed in UK by Printondemand-Worldwide

©Copyright Incisive Risk Information (IP) Limited, 2015. All rights reserved. No parts of this publicationmay be reproduced, stored in or introduced into any retrieval system, or transmitted, in any form or by anymeans, electronic, mechanical, photocopying, recording or otherwise without the prior written permission of thecopyright owners.

Marketing Manager: Rainy Gill




LETTER FROM THE EDITOR-IN-CHIEF

Ron BerndsenDe Nederlandsche Bank and Tilburg University

In this issue of The Journal of Financial Market Infrastructures we have two paperson large-value payment systems and two papers on central counterparties.

In their paper “Identification of over and under provision of liquidity in real-timepayment systems”, Edward Denbee, Rodney J. Garratt and Peter Zimmerman studythe issue of liquidity provision in the context of payment systems where participatingbanks have flexibility on the timing of their outgoing payments during the settlementday. Using the observed difference between a bank’s share of liquidity provision andits share of liquidity usage, it is found that smaller banks provide more liquidity tothe system than larger banks. The interesting underlying question is whether bankssteer their behavior during the settlement day. Based on data from the Bank of Eng-land’s CHAPS system, the authors perform a simulation by randomly reordering thepayments that have occurred in reality on that settlement day. They find evidence thatthere are structural or behavioral reasons for participating banks to consistently underprovide or, as the case may be, over provide liquidity.

The second paper, “Reaction functions of the participants in Colombia’s large-value payment system” by Constanza Martínez and Freddy Cepeda, investigates howparticipants react to major incidents in a payment system, contributing to the literatureon payment reaction functions. They study four cases in the period 2007–12 of theColombian central bank’s system Cuentas de Depósito (CUD) relating to a relativelylong power outage (4.5 hours), a multiday technical failure of the largest participantand two brokerage firms’ defaults. The results show, for example, that the responseof the participants depends significantly on the type of incident (eg, technical versusa financial root cause) and the type of financial institution involved.

Moving to central counterparties (CCPs), the third paper explores the conceptof interoperability between CCPs. The authors, Jürg Mägerle and Thomas Nellen,investigate interoperability from the perspective of the multilateral netting propertyof central clearing. They compare the setup of three interoperable cash-equity CCPsto the benchmark case of a single CCP. This three-way arrangement is an existingconfiguration in Europe and can be compared to the situation of full consolidation,ie, when all interoperable CCPs would merge into a single CCP. This paper looksat alternative setups to reduce the amount of overcollateralization while respectingrequirements for containing systemic risk.

The last paper in this issue takes us back to October 19, 1987: Black Monday,a regime change comparable to the Lehman Brothers crisis. I still vividly remem-ber Black Monday because on the next day I had to defend my MA thesis at Tilburg

www.risk.net/journal




University, dealing with … regime changes. In this Forum contribution, entitled “Cen-tral counterparties in crisis: the Hong Kong Futures Exchange in the crash of 1987”,Robert Cox shows the reader the lessons that can be learned from one of the fewoccasions in history when a CCP got into severe difficulty. One striking differencefrom the modern-day CCP is that, back then, the clearing house was responsible fortrade feed processing and initial margin setting but it did not act as the principal forthe guarantee of settlement. The guarantee function was attributed to a separate legalentity, which complicated procedures and obscured responsibilities.

This last paper provides an appropriate link to the next issue of The Journal ofFinancial Market Infrastructures, preparing the reader for our special issue on CCPs,to appear in March 2016.

I hope you enjoy reading this issue of The Journal of Financial Market Infrastruc-tures.

Journal of Financial Market Infrastructures 4(2)




Journal of Financial Market Infrastructures 4(2), 1–20

Research Paper

Identification of over and under provision ofliquidity in real-time payment systems

Edward Denbee,1 Rodney J. Garratt2 and Peter Zimmerman3

1Bank of England, Threadneedle St, London EC2R 8AH, UK;email: [email protected] of California, Santa Barbara, CA 93106, USA; email: [email protected]ïd Business School, University of Oxford, Oxford OX1 1HP, UK;email: [email protected]

(Received August 20, 2015; revised October 28, 2015; accepted November 9, 2015)

ABSTRACT

It can be difficult to tell to what extent the amount of liquidity that an individualbank provides is intentional and to what extent it results from external factors thatare beyond its control. In this paper, we describe two methods for evaluating liquidityprovision in real-time gross settlement payment systems. We also utilize a recombi-nant approach to detect instances where observed patterns of liquidity provision areunlikely to have occurred in the absence of some behavioral or structural factors, suchas differences in banks’business models. We apply our techniques to crisis-period datafrom the Clearing House Automated Payment System (CHAPS), the UK large-valuepayment system. We find that smaller banks provide more liquidity to the system thanlarger banks relative to their payment flows. Moreover, we observe an increase in thedegree of inequality of liquidity provision relative to usage across banks followingthe collapse of Lehman Brothers. Our results suggest that the instances of over andunder provision of liquidity that appear in our data are more frequent than would beexpected from random payment flows.

Keywords: inequality; liquidity; payment systems.

Corresponding author: R. J. Garratt Print ISSN 2049-5404 j Online ISSN 2049-5412Copyright © 2015 Incisive Risk Information (IP) Limited

1




2 E. Denbee et al

1 INTRODUCTION

A payment system consists of the procedures and associated computer networks usedby its participants to transfer money. Sometimes called the plumbing of the finan-cial system, smoothly functioning payments systems are essential to the operationof financial markets. Large-value wholesale payment systems, such as the ClearingHouse Automated Payment System (CHAPS) in the United Kingdom, are generallyconsidered to be systemically important because of the value and nature of the finan-cial transactions that they facilitate. On a typical business day, transactions with a totalvalue of around £269 billion flow through CHAPS, roughly equivalent to one-sixthof the United Kingdom’s annual gross domestic product (see Bank of England 2015).

In a real-time gross settlement (RTGS) system, payments settle immediately andwith finality in central bank money, providing that the payer has sufficient liquidityto fund the outgoing payment. But the aggregate amount of liquidity needed to fundpayment obligations is often much less than the gross value of payment flows. Duringthe course of the day, each payment system participant typically makes and receivesthousands of payments. Thus, outgoing payments are funded not only from liquid-ity made available from payment system participants’ own resources, but also fromliquidity obtained from incoming payments.1

If banks were required to process payment requests as soon as they received them,they would have little discretion over the liquidity they provide to the rest of thepayment system. But this is not the case: with the exception of some time-criticalpayments, or some payments systems designed to process customer payments in realtime, banks do not usually have to process payment requests as soon as they receivethem. Rather, banks may choose to delay processing payments in order to conserveliquidity and make use of incoming funds.2 If too many banks withhold liquidity, thepayment system can fall into gridlock.3 Consequently, central banks have an interestin monitoring banks’ liquidity provision in order to ensure the continued smoothfunctioning of the payments system.

In this paper, we measure liquidity provision in two ways. First, we look at themaximum net debit position that settlement banks reach in their settlement accounts

1 These are generally either in the form of central bank reserves, or liquidity generated via arepurchase agreement (repo) of eligible assets at the central bank.2 The incentive to conserve liquidity arises because funds that banks deposit in their settlementaccounts to facilitate payments have an opportunity cost to the bank in terms of foregone investmentopportunities, or to mitigate against the possibility of liquidity shocks later in the day.3 The Committee on Payments and Market Infrastructures (1993) defines a gridlock as a situationin which “the failure of some transfer instructions to be executed … prevents a substantial numberof other instructions from other participants from being executed”. See Soramäki and Bech (2001)for a discussion of the cost of delay in payment systems.

Journal of Financial Market Infrastructures www.risk.net/journal




Identification of over and under provision of liquidity 3

over the course of each day, during a historic period.4 The sum of these net debitpositions across all banks is the total amount of liquidity that was actually used tomake the day’s payments. Therefore, each bank’s own net debit position, divided bythe sum of the net debit positions of all banks, gives the share of liquidity providedby each bank. Whenever the value of a bank’s payments into the system exceeds thatof those it has received, the difference has to be made up either from central bankreserves, or from eligible collateral that a settlement bank pledges intraday in orderto obtain liquidity from the central bank.5 Therefore, we can assume that a net debitposition imposes an opportunity cost of using central bank reserves or of pledgingeligible collateral, so our first measure reflects the nominal monetary cost of liquidityprovision.

In addition to this monetary cost, making payments earlier can result in a greaterexposure to the liquidity risk of counterparties. For example, if the paying bank relieson its own recycled liquidity to fund future payments, then it faces the risk that itscounterparty fails to recycle the liquidity back into the payment system in a timelyfashion. This may happen, for example, if the counterparty has an operational problemor enters bankruptcy.6 A settlement bank’s average net debit position throughout theday is a better proxy for this liquidity risk than the maximum net debit position, sowe use this as our second measure of liquidity provision.

Some banks have a higher value of payment activity than others and, hence, mayreasonably be expected to provide more liquidity, in absolute terms. Therefore, wetake usage of payments system liquidity (ie, gross payment outflows) into accountin our two measures of the cost and risk of liquidity provision. We compute ourmeasures for CHAPS settlement banks using data from January 2008 to May 2010,and we present aggregated results for groups of banks in two size categories.Althoughlarger banks do provide the bulk of the liquidity in absolute terms, we find that thesmaller banks almost invariably provide a larger share of liquidity to the system thantheir share of payments. This is true under both of our measures.

The ratio of the liquidity provided by a bank to the gross value of payments it makesis the average liquidity cost of the bank’s payments. As the above analysis suggests,

4 The term “settlement bank” refers to a bank that has a settlement account with the central bankfor the purposes of making intraday payments through the payments system. Other banks that makepayments through CHAPS do so via one or more of these settlement banks.5 Banks can also fund payments using liquidity received in interconnected payments systems. Forinstance, banks that are settlement banks in both CHAPS and the Certificateless Registry for Elec-tronic Share Transfer (CREST), which is the UK securities settlement system, could use a netreceiver position in one system to fund payments in the other. We do not consider this in our analy-sis, partly because in the United Kingdom these other interconnected payments systems do notallow settlement banks the same degree of control over intraday timing as CHAPS does.6 Manning et al (2009) examine issues around strategic payments system behavior.

www.risk.net/journal Journal of Financial Market Infrastructures




4 E. Denbee et al

different banks have different average liquidity costs. We measure inequality in theliquidity cost of payments across CHAPS banks by computing Gini coefficients.We examine these coefficients over time. The series shows a significant increase inthe period surrounding the collapse of Lehman Brothers. This finding is importantbecause it tells us that CHAPS participants became more dependent on the liquidityprovision of others during the crisis.

Unobserved factors may explain some of these differences in liquidity provision.7

Average payment sizes are likely to be important; a bank that sends and receives asmall number of very large payments would be expected to use more liquidity than abank with a large number of smaller payments. This is because the latter bank wouldbe better able to offset payments and receipts and, thus, less likely to assume a largenet sender position.

Clearing on behalf of clients may have an impact too. There is evidence that thiscan create efficiencies in liquidity usage (Jackson and Manning 2007). Client clearingtends to form a greater proportion of payments for larger banks than it does for smallerbanks. Further, it is possible that settlement banks, when making payments on behalfof clients, may on occasion intentionally delay in order to reduce credit exposures tothese clients.8

These factors could explain the lower levels of liquidity provision, relative to pay-ment values, that we observe for larger banks. In any case, some heterogeneity betweenindividual banks’ liquidity provision and usage is inevitable, and it does not necessar-ily imply unfairness. Since the arrival of payment requests from customers is typicallyoutside of the control of the banks, there will be net liquidity providers and users onany given day, even if all banks process payment requests immediately. This meansthe patterns of liquidity provision that we observe could just be an artefact of the waypayment requests happened to arrive. We would like to know when observed differ-ences in liquidity provision are so marked that they are very unlikely to have beensolely a result of external factors. We provide a method for identifying when overand under provision of liquidity is unlikely to have occurred by accident. The ideais to ask, given all the different permutations for how payments might have arrived,what a very unlikely level of liquidity provision would be. We answer this questionby reshuffling each day’s payment schedule many times to generate distributions forour liquidity provision measures. We then check how often actual values of these

7 Becher et al (2008) describe CHAPS and the factors that affect the timing of payments in moredetail.8 According to Valukas (2010, Section III.A.5), Lehman Brothers’ settlement banks tried to reducetheir unsecured intraday exposures to the institution once its financial condition began to deteriorate.






measures lie in the tails of these distributions. We find that the instances where banksare in these tails occur far more frequently than we would expect to see in the absenceof behavioral or structural factors.

The analysis presented in this paper can be applied to any payment system. It maybe particularly useful to assess liquidity provision in instances where settlement bankshave a substantial degree of freedom in choosing when to make payments during theday. To our knowledge, this is the case in most systems used for making wholesaleunsecured payments. Care should be taken when comparing results across systemsor across time periods that may involve significant structural changes to the paymentsystem. In the particular case of CHAPS, it should be noted that the system hasundergone several structural changes since the end of our data period in May 2010that may have led to changes in the patterns of liquidity provision.9

2 DATA

We use data on payments activity for all CHAPS settlement banks from January 2,2008 to May 28, 2010. This data is obtained from the payments database maintainedby the Bank of England in its role as operator of the RTGS system.

The CHAPS settlement banks during this period wereABNAmro, Bank of England,Bank of Scotland, Barclays, Citibank, Clydesdale, Co-operative Bank, CLS Bank,Danske Bank, Deutsche Bank, Lloyds, HSBC, NatWest, The Royal Bank of Scotland(RBS), Santander, Standard Chartered and UBS. Bank of England and CLS Bank areexcluded from our analysis. Membership is not constant throughout this period: ABNAmro left on September 19, 2008 and Danske Bank joined on April 20, 2009.

We aggregate any figures that are reported separately for NatWest and RBS, sincethese banks belong to the same group. A merger with Lloyds meant that the Bank ofScotland reserves account was withdrawn on February 5, 2009, so those two settlementbanks have effectively operated from a single pool of liquidity since this date.

3 MEASURES OF LIQUIDITY PROVISION

3.1 Measuring the cost of liquidity provision

One way to evaluate liquidity provision is to look at the share of total liquidity a bankprovides to the system and to relate this to its share of total payments. Suppose thereare n banks, which are indexed by i D 1; : : : ; n. Let xs

i .t/ be the amount sent by banki up to time t on day s, and let ys

i .t/ be the amount received. t lies in the interval

9 Davey and Gray (2014) analyze changes to banks’ liquidity requirements in CHAPS as a result ofthe introduction of a liquidity-saving mechanism in April 2013; Finan et al (2013) describe recentwidening in CHAPS settlement bank membership.





6 E. Denbee et al

FIGURE 1 Frequencies of observations of cost measure of liquidity provision for (a) largerand (b) smaller banks over the period January 2, 2008–May 28, 2010.

0

5

10

15

20

25

30

35

40

%

0

5

10

15

20

25

30

35

40

%

–0.40 –0.20 0 0.20 0.40 –0.40 –0.20 0 0.20 0.40

(a) (b)

Œ0; T �, where t D 0 denotes the start of the day and t D T denotes the end. Then, thenet debit position at time t on day s is

N si .t/ D xs

i .t/ � ysi .t/: (3.1)

The net debit position identifies the liquidity provided by bank i to the rest of thesystem by time t on day s. The liquidity burden of bank i on day s is determined bythe largest net debit position,

Lsi D max

t2Œ0;T �N s

i .t/: (3.2)

The largest net debit position incurred by a bank on a given day is the total amountof its own cash and collateral that it actually used to fund its own payments. It isthe minimum amount of liquidity that the bank could have held to meet its paymentobligations on that day in order to make its payments at the time that they occurred.Note that Ls

i > 0, since xsi .0/ D ys

i .0/ D 0 for all i; s.Let P s

i D xsi .T / denote the total value of payments sent by bank i on day s. Our

cost-based measure of bank i ’s liquidity provision on day s is

csi D Ls

iPnj D1 Ls

j

� P siPn

j D1 P sj

: (3.3)

This measures the observed difference between bank i ’s share of liquidity provisionin the system and its share of liquidity usage.

If the difference is less than 0, the bank provides less liquidity than might beexpected, given its share of payment activity. If it is greater than 0, the bank providesmore than might be expected, given its share of payment activity.






TABLE 1 Moments of distributions of the cost-based measure of liquidity provision forlarger and smaller banks over the period January 2, 2008–May 28, 2010.

Larger banks Smaller banks

Mean �0.039 0.029Variance 0.010 0.002Skewness �0.417 1.924Excess kurtosis 0.093 4.101

Figure 1 on the facing page comprises two histograms showing our measure ofthe cost of liquidity provision over the period January 2, 2008–May 28, 2010. Wepartition the banks into two groups based on their average daily values sent over theperiod: the larger banks in part (a) and the smaller banks in part (b). Larger banksaccount for over 90% of total payment values sent through CHAPS over this period.The makeup of each group is recalibrated on February 5, 2009.10 We emphasize that“larger” and “smaller” here refer only to average daily values through CHAPS overthe period and are not necessarily correlated with other measures of size, such asbalance-sheet size or payments in non-sterling currencies.

The bin size on the x-axis of Figure 1 on the facing page is 0:01, with the globalpeaks occurring in the interval containing zero Œ�0:005; 0:005�. Figure 1 shows thatthe larger banks tend to have the widest range of values. For smaller banks, themeasure over this period takes an average value of 0:03, while for larger banks itis �0:04. This translates into an average provision for each small bank of £600–700million a day in excess of what might be expected from their share of payment activity.Difference in means tests reveal significance. Table 1 shows that there are differencestoo in the higher moments of the two distributions: the distribution of scores for thesmaller banks has a lower variance, a higher (positive) skewness and a higher kurtosiscompared with the distribution for the larger banks.

These differences may not be due to strategic behavior. It may be that, structurally,larger banks pay later than smaller ones, due to the nature of their business. Theactivity of many of the banks in the smaller group is dominated by retail business,while the payment activity of many of the larger banks may be driven by wholesalebusiness (for example, lending and borrowing in the money markets). Many of thelarger banks also make payments on behalf of other institutions. These factors may

10 Before February 5, 2009, seven banks are classed as larger, and six are classed as smaller.Following this date, there are six in each of the two groups.





8 E. Denbee et al

FIGURE 2 Frequencies of observations of risk measure of liquidity provision for (a) largerand (b) smaller banks over the period January 2, 2008–May 28, 2010.

0

5

10

15

20

25

30

35

40

%

0

5

10

15

20

25

30

35

40

%

–0.40 –0.20 0 0.20 0.40 –0.40 –0.20 0 0.20 0.40

(a) (b)

lead to structural differences in banks’ payment schedules. These results may alsosuggest that there are liquidity efficiency benefits to being a larger bank.11

3.2 Measuring the risk of liquidity provision

Measuring liquidity provision based on largest net debit position has its drawbacks.While this measure addresses the direct cost to a bank of providing liquidity, it does notfully reflect exposure to the liquidity risk of its counterparties in the payment system.Banks rely on each other to recycle payment system liquidity. When a bank assumesa debit position, it has made outward payments before receipt of incoming payments,so it is exposed to risk if a counterparty delays or fails to send its own payments. Toreduce this risk, the bank may prefer to delay its own payments.12 However, if allbanks delay their payments, then the system would fall into gridlock; so, banks musthave some willingness to assume net debit positions vis-à-vis their counterparties,at least for some period of time. A measure of the risk of liquidity provision shouldtherefore consider how willing the bank is to hold net debit positions for periods oftime during the payment day, as well as the size of those positions.

11 Galbiati and Giansante (2010) model liquidity usage as a symmetric random walk. In their model,it can be shown that expected liquidity usage varies asymptotically with the square root of paymentvolumes, but linearly with payment values. This suggests liquidity usage is concave with the sizeof bank.12 Benos et al (2014) show that banks delay payments to counterparties with higher credit risk.






TABLE 2 Moments of distributions of the risk-based measures for larger and smallerbanks over the period January 2, 2008–May 28, 2010.

Larger banks Smaller banks

Mean �0.045 0.033Variance 0.019 0.004Skewness 0.111 2.398Excess kurtosis �0.148 6.571

A bank takes on liquidity risk when it is a net sender, that is, its net debit positionN s

i .t/ > 0. The average risk taken for bank i on day s is

�si D 1

T

Z T

0

maxŒN si .t/; 0� dt: (3.4)

Using this metric for liquidity risk, we can construct a measure comparing the shareof risk borne by each bank to its share of payments, which is given by

� si D �s

iPnj D1 �s

j

� P siPn

j D1 P sj

: (3.5)

If this is less than 0, then the bank takes on less liquidity risk than might be expected,given its share of payment activity. If it is greater than 0, then the bank takes on morerisk than might be expected, given its share of payment activity.

Figure 2 on the facing page comprises two histograms showing observed values ofthe measure � s

i over the sample period from January 2, 2008 to May 28, 2010. Thebanks are grouped in the same way as for the measure of cost (see footnote 10). Therelationship between size and risk is similar to before. The larger banks bear less riskthan their shares of payment activity would suggest. The larger banks have a meanvalue of �0:05, compared with 0:03 for the smaller banks. Again, difference in meanstests reveal significance.

Table 2 shows the higher moments of the two distributions. Again, the distributionfor smaller banks is more positively skewed and has greater kurtosis.

Ball et al (2011) explain that many banks have internal schedulers that allow themto place bilateral limits against individual counterparties. By exercising these limits,a bank can directly control its positions against each counterparty and, thus, manageits liquidity provision, as captured by the measure of cost. A bank that uses theselimits dynamically to control the duration of its bilateral liquidity exposures wouldbe able to manage its liquidity risk, as captured by our measure. The results of the





10 E. Denbee et al

FIGURE 3 The relationship between measures of cost and risk of liquidity provision,January 2, 2008–May 28, 2010.

–0.3 –0.2 –0.1 0 0.1 0.2 0.3 0.4 0.5 0.6–0.4–0.4

–0.3

–0.2

–0.1

0

0.1

0.2

0.3

0.4

Measure of risk of liquidity provision

Mea

sure

of c

ost o

f liq

uidi

ty p

rovi

sion

next section suggest that CHAPS settlement banks effectively manage the durationas well as the magnitude of these exposures.13

3.3 Comparing the measures of cost and risk of liquidity provision

Intuitively, the two measures of liquidity provision are likely to be correlated. Inmost cases, a bank that builds up a large net debit position will do so graduallyover the course of the day, in which case it may have a high score on both metrics.This correlation is shown in Figure 3, which plots the pairs .� s

i ; csi / for each of the

settlement banks on each day over a period extending from January 2, 2008 to May28, 2010. Banks that provide more liquidity than their share of payments in bothdimensions appear in the top right corner of the figure, whereas banks that provideless in both dimensions appear in the bottom left. This correlation is positive andsignificant, with an adjusted R2 value of 0.871.

13 This is confirmed by others too. For example, Jurgilas and Žikeš (2014) find evidence that settle-ment banks in CHAPS assign a positive and significant interest rate to the intraday timing of theirpayments.






4 PAYMENT SYSTEM INEQUALITY

In this section, we measure payment system inequality in terms of the liquidity costto the payee of making their daily payments. There are various approaches we couldtake to measure this inequality. We believe the Gini coefficient, a common measure ofincome inequality, has desirable properties that make it most suitable for this applica-tion.14 The Gini coefficient has an intuitive interpretation and can be calculated for asample of entities with different-sized populations. This is essential to our application,where the number of payments differs both across banks and from day to day.

Entropy measures, such as Theil’s T measure, are less desirable for our purposes.15

The main advantage of Theil’s T is that it is decomposable, so that inequality within agroup and between groups can be estimated. For instance, economists have comparedincome inequality within racial groups in a given country (see Bellù and Liberati2006). We see no immediate use of the decomposition property in our application.Moreover, Theil’s T cannot be computed if members of a population sometimes havea 0 value.Again, this is unlikely when considering income, but it is feasible that a bankmay provide zero liquidity and, hence, have a cost per payment equal to 0. Finally,Theil’s T is not ideal for comparing populations with different sizes. If the numberand size of groups differs, then the limit of the index will differ.

Following Ray (1998), the Gini coefficient for a population with a finite numberof groups, each with a finite number of members, on day s is represented by

Gs D 1

2M 2�

� nXj D1

nXkD1

mj mkj`sj � `s

kj�

; (4.1)

where n is the number of banks, mj is the number of payments made by bank j , M isthe total number of payments made by all banks, � is the average liquidity cost of allpayments and `s

j D Lsj =P s

j is the average liquidity cost of payments made by bankj . Rather than looking at differences between individual values and the mean, thismeasure sums the pairwise differences over all. This is convenient for our purposes,since we can weight each settlement bank by the number of payments sent. Gs liesin the interval Œ0; 1�. It takes a value of 0, denoting perfect equality, if all banks make

14 See Gini (1912), Atkinson (1975) and Kakwani (1977) for the development of this methodol-ogy. Barro (2000) and Demirgüç-Kunt and Levine (2009) use inequality to investigate economicphenomena.15 The connection to entropy is very simple. Maximum entropy occurs with a uniform distribution,implying perfect equality. Minimum entropy occurs in a world of perfect certainty; that is, whereone person has all the wealth.





12 E. Denbee et al

FIGURE 4 Gini coefficient Gs for CHAPS over the period January 2, 2008–May 28, 2010.

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Jan2008

Apr2008

Jul2008

Oct2008

Jan2009

Apr2009

Jul2009

Oct2009

Jan2010

Apr2010

Lehman End of break point test window

Gray line represents daily data. Black line represents twenty-day moving average.

TABLE 3 Quandt–Andrews break test for the Gini coefficient series.

Date maximized F -statistic

Likelihood ratio September 5, 2008 16.48��

Wald ratio September 5, 2008 6828.93��

The period tested is January 2, 2008–February 4, 2009. ��Significance at the 1% level.

payments using the same amount of liquidity per unit of payment. Conversely, Gs

takes a value of 1 if one bank provides all the liquidity with which the other banksmake their payments. Figure 4 plots the twenty-day moving average value of Gs usingdata from January 2, 2008–May 28, 2010.

Figure 4 shows that this measure is highly volatile, changing significantly from dayto day. Moveover, the volatility of the series increases throughout 2008, suggestingthe distribution of liquidity usage became less consistent. We test for the presence of astructural break in the series with a Quandt–Andrews test for an unknown breakpoint.We use the individual daily data from January 2, 2008 to February 4, 2009, since wemight expect the merger of two settlement accounts on February 5, 2009 to causea structural change to liquidity usage. The test identifies September 5, 2008 as theday when the test statistics were maximized (significant at the 1% level). Table 3shows the results. A Chow test for September 15, 2008 rejects at the 1% level the






null hypothesis that there is no breakpoint on this date. We conclude that there wasa significant increase in payment system inequality across CHAPS members aroundthe time of the Lehman Brothers default on September 15, 2008.16

The Gini coefficient can also be computed using the measure of the risk of liquidityprovision. Since the two measures are highly correlated, the results are very similarand are not included here.

Benos et al (2014) discuss the changes in CHAPS settlement banks’ behavior thatoccurred following the Lehman default. Increased concerns about bank-specific andsystem-wide risks led to slower payment processing, with evidence of targeted delay.This may have caused some banks to use less liquidity, as they slowed down theirpayments, while others may have been forced to use more liquidity, as they could notrely on recycling incoming payments.

Central bank actions can also have an effect on the aggregate measure. For exam-ple, if one day the Bank of England makes earlier payments in CHAPS, then othersettlement banks might be able to use less of their own liquidity, as they can recyclecentral bank liquidity instead. That would result in a decrease in liquidity provisionmeasures. But, if these early payments had taken place in another linked system andthe liquidity transferred into CHAPS by the settlement banks, then the opposite effectcould occur.17 The actual effect on the Gini coefficient measure will be determinedby the degree of heterogeneity in the impact of central bank actions on differentsettlement banks.

5 A RECOMBINANT APPROACH TO IDENTIFYING OVER ANDUNDER PROVISION OF LIQUIDITY

Our measures of liquidity provision capture differences in the amount and durationof liquidity provision by settlement banks. However, payment requests that come tosettlement banks from customers are likely to be determined by factors exogenous to

16 Other studies have observed effects on participant behavior in payment systems in responseto extraordinary events. McAndrews and Potter (2002) showed that, in the immediate aftermathof the events of September 11, 2001, the propensity of banks to send out payments through theFedwire Funds transfer service declined. Likewise, Bech and Garratt (2012) documented dynamicadjustments in payment processing behavior by Fedwire participants following the collapse ofLehman Brothers. Neither of these studies considered relative liquidity provision across banks.17 For example, a settlement bank recycling central bank liquidity in CREST may appear to be usingrelatively little of its own liquidity in that system, but if the liquidity is transferred to CHAPS andused to make payments, then the settlement bank’s apparent liquidity provision in CHAPS wouldincrease.





14 E. Denbee et al

the banks, so they can be thought of as random from their perspective. Our liquidityprovision measures depend not only on the payments made and received during aday, but also on the order in which these payments take place. This means that whenwe look at a payment file for a given day and compute our measures, the liquidityprovision we observe might just be accidental in the following sense: had the samepayments arrived in a different order, then, assuming all banks process paymentsimmediately, the measures for individual banks would change.

We would like to be able to distinguish between cases in which over or underprovision is likely to be an artefact of random variations in the timing of payments, andcases in which they are not. To sort out the random component, we take the observedpayments files over a sample period and re-order each day’s payments many timesto produce thousands of simulated days in which all the same payments are made,but in different orders. We compute our measures of liquidity provision for each ofthese simulated days and construct a distribution of each measure over each day ofthe sample.

5.1 Simulation

Since the schedule of payments on each day varies, we need to consider a periodof several days. We use 102 days of data (January 4, 2010 to May 28, 2010). Eachday’s simulation involves randomly reordering around 125 000 transactions, so it istoo computationally expensive to consider a longer period. Moreover, if we takea longer period, there is a greater risk of structural changes in banks’ liquidityusage (for example, when a settlement bank takes on a large new customer). It isnot feasible to consider all possible permutations for every day. Instead, we simu-late each day 200 times. We find that 200 is sufficient to produce stable empiricalresults.

We treat payments from or to the Bank of England and CLS Bank as exogenous,since these settlement banks do not have incentives to engage in the strategic provisionof liquidity. These payments therefore retain their order in the simulations. For exam-ple, if such a payment was the tenth payment of the day in reality, then it remains thetenth payment in our recombinant simulations of that day. As discussed previously,there may be other time-critical payments, but there is no means of identifying suchpayments from the data.

At the time of undertaking this analysis, CHAPS was a simple RTGS system withno liquidity-saving mechanism. Payments were processed in the order that they weresubmitted to the system, subject to sufficient liquidity being available to the payer set-tlement bank. Therefore, our simulations use the actual payments that were submittedto the system, without any netting or other manipulations.






TABLE 4 Cost of liquidity provision: comparison of actual values to threshold valuesproduced from simulations over the period January 4–May 28, 2010.

Threshold‚ …„ ƒ5% 95%

Frequency of threshold value 27% 39%breach, across all banksProportion of banks that never 0.17 0.17breach the thresholdProportion of banks that breach the 0.25 0.50threshold on over half the occasions

TABLE 5 Risk of liquidity provision: comparison of actual values to threshold valuesproduced from simulations over the period January 4–May 28, 2010.

Threshold‚ …„ ƒ5% 95%

Frequency of threshold value 18% 56%breach, across all banksProportion of banks that never 0.42 0.17breach the thresholdProportion of banks that breach the 0.17 0.67threshold on over half the occasions

5.2 Results

Table 4 provides statistics on the number of occasions when csi , our measure of the

cost of liquidity provision of bank i on day s, falls below the 5th percentile or risesabove the 95th percentile. Table 5 does the same for the measure of risk of liquidityprovision. In the absence of behavioral or structural factors that influence the timingof payments, we would expect each of these threshold values to be breached on around5% of occasions. This is not the case: they are breached much more frequently. Thereappears to be some heterogeneity across these twelve banks: some never breach thesethresholds, while others breach them on more than half of occasions. This providesvery strong evidence that there are additional structural or behavioral reasons thatmay cause a bank to provide a share of liquidity to payment systems that differs fromits share of payment activity.

Figure 5 on the next page shows the frequency with which individual banks arebelow the 5% threshold (x-axis) and above the 95% threshold (y-axis) for the two





16 E. Denbee et al

FIGURE 5 Frequencies of extreme values of liquidity provision measures for twelvesettlement banks in CHAPS over the period January 2, 2008–May 28, 2010.

00

0.2

0.2

0.4

0.4

0.6

0.6

0.8

0.8

1.0

1.0 0 0.2 0.4 0.6 0.8 1.00

0.2

0.4

0.6

0.8

1.0(a) (b)

% o

f tim

es a

bove

95%

% of times below 5% % of times below 5%%

of t

imes

abo

ve 9

5%

(a) Cost. (b) Risk.

measures of liquidity provision. If liquidity provision were entirely random, then eachof these thresholds should be breached around 5% of the time, and we would expectall of the banks to be near the (5%,5%) point. Instead, they are widely scattered, andthere is a clear negative correlation.

5.3 Alternative explanations for simulation results

There is no way to disentangle behavioral and structural causes of over and under pro-vision of liquidity. As discussed above, banks can be under providers if they activelyseek to reduce costs of liquidity provision or reduce risk by delaying payments. In whatfollows, we summarize structural reasons that would lead to over or under provisionabsent any intentional efforts to manage liquidity usage.

Differences in banks’ business models could be a structural factor that affects theirliquidity provision. For example, overnight loans involve an advance late in the daywith a repayment the next morning. Therefore, banks that tend to be net lenders mayappear to be providing relatively less liquidity (as they pay late, they can do so fromrecycled liquidity), while those that are net borrowers may appear to be providingrelatively more (as principal repayments are made early, they may need to be fundedfrom reserves).

More generally, some banks may have payments that are time-critical intraday(they need to be made at some point before the end of the day). For example, pay-insto CLS must be made in the morning UK time. This means that the bank may appear






to be generous with its liquidity, when in fact the early payment timing is driven byfactors outside of the control of the bank’s payment system operators.18

A bank that sends and receives a small number of very large payments would beexpected to use more liquidity than a bank with a large number of smaller payments.This is because the latter bank would be better able to net payments and receipts and,thus, less likely to assume a large net sender position.19

Another issue is “tiering”, that is, the use of the payment system via an account at asettlement bank rather than via direct membership. One of the reasons for doing so isliquidity pooling: Jackson and Manning (2007) argue that, unless customers’paymentflows are highly correlated with those of the settlement bank, the total pool of liquidityneeded will be smaller than that required if each customer were to become a settlementbank. This may mean that settlement banks with a large number of clients may needto use less liquidity than those of an equivalent size that do not have a large numberof clients.20 Banks with clients may behave differently, for example, to manage theircredit risk exposure to their customers.21

For banks with a lot of international customers, their location can have an importanteffect. For example, consider the case of a CHAPS settlement bank with a lot of clientsbased in the United States. These clients might only send same-day instructions in theUK afternoon, due to time zone differences. Such a bank may appear to be hoardingliquidity, when in fact it may be making its payments as soon as the instructions arrive.Clients based in Asia may have the opposite effect.

6 CONCLUDING REMARKS

Our liquidity provision measures allow us to see whether banks provide amounts ofliquidity that are comparable to their share of payment flows, and whether this changesover time. We provide evidence that small banks tend to provide more liquidity thanlarger banks in relative terms (though not necessarily in absolute terms). We also findevidence that the patterns of liquidity provision by some participants are unlikely tobe explained solely by random variations in the timing of payment requests.

18 According to Ball et al (2011), such time-critical payments comprise around 4% of values in thesystem.19 This pooling effect is described in Galbiati and Giansante (2010).20 We cannot observe the internalization effect of tiering. Flows between a settlement bank and itscustomers do not need to be submitted to CHAPS at all: they can simply be cleared across thesettlement bank’s own accounts, without any central bank liquidity being required.21 The Bank of England has identified tiering in UK payment systems, particularly CHAPS, as asource of risk to financial stability, as it leads to intraday credit and liquidity exposures betweenmembers and indirect participants (see Bank of England 2015).





18 E. Denbee et al

We cannot distinguish whether observed differences in liquidity provision resultfrom banks intentionally conserving liquidity (see Ball et al 2011; Norman 2010), orwhether they result from structural factors. This distinction would be important if thegoal was to make a normative assessment of a particular bank’s behavior. It is possiblethat a regulator, operator or other participants in the system might want to do this,and perhaps take action in response. CHAPS has throughput guidelines designedto encourage banks to complete a certain percentage of their day’s payments by aparticular point in time during the day, and banks that fail to do this over an extendedperiod of time have to explain themselves to a panel of their peers. Their argumentscould relate to structural factors, such as those discussed in Section 5.3. An individualinvestigation of a bank and its business model could be useful for shedding light onthe extent to which it may be intentionally providing an amount of liquidity differentto that which might be expected, given its payment activity.

The Gini coefficient we compute provides a measure of how the average liquiditycost of payments is distributed across the system. High values of the Gini coefficientindicate that some banks are much larger providers of liquidity relative to paymentsthan others, and that others are recycling others’ liquidity rather than providing theirown. The implication is that disruptions in liquidity provision due to failures or opera-tional outages could have significant effects, because some banks are dependent on theliquidity provision of others. We found that the Gini coefficient in CHAPS increasedaround the time of the Lehman Brothers failure, elevating this concern.

An interesting by-product of our simulations is that we obtain a measurement ofsystem-wide liquidity usage that abstracts from intraday payment arrival timing andprocessing decisions. The total liquidity actually used by these twelve banks averaged£18.7 billion each day over the sample period, while in the simulations the systemrequired £17.9 billion on an average day. These results suggest that, to the extent thatsettlement banks in CHAPS submit payments strategically, this does not result in asubstantially less efficient system, in aggregate, compared with our model in whichpayments are simply made randomly.22

DECLARATION OF INTEREST

This paper was completed while Rodney J. Garratt was working at the Federal ReserveBank of New York. The views expressed in this paper are those of the authors, andnot necessarily those of the Bank of England, the Federal Reserve Bank of New Yorkor the Federal Reserve System.

22 This is not to say that such strategic behavior would never result in a much less efficient system;only that it does not occur in practice during our sample period.






ACKNOWLEDGEMENTS

The authors wish to thank Evangelos Benos, Anton Braun, James Chapman, RogerClews, Toby Davies, Marco Galbiati, Matti Hellqvist, Matthew Hunt, John Jackson,Will Roberds, Warren Weber, two anonymous referees, and seminar participants atthe Bank of England and at the Bank of Finland 8th Simulator Seminar and Workshopand the Federal Reserve Bank of Atlanta for useful comments and suggestions. Wealso thank Lewis Webber for invaluable help with simulations. Any errors are ours.An earlier version of this paper appeared as Bank of England Working Paper 513.

REFERENCES

Atkinson, A. B. (1975). The Economics of Inequality. Oxford University Press.Ball, A., Denbee, E., Manning, M., and Wetherilt, A. (2011). Intraday liquidity: risk and

regulation. Financial Stability Paper 11, Bank of England.Bank of England (2015). The Bank of England’s supervision of financial market infrastruc-

tures. Annual Report.Barro, R. J. (2000). Inequality and growth in a panel of countries. Journal of Economic

Growth 32, 5–32.Bech, M. L., and Garratt, R. J. (2012). Illiquidity in the interbank payment system following

wide-scale disruptions. Journal of Money, Credit and Banking 44(5), 903–929Becher, C., Galbiati, M., and Tudela, M. (2008). The timing and funding of CHAPS sterling

payments. Federal Reserve Bank of New York Economic Policy Review 14(2), 113–133.Bellù, L.G., and Liberati, P. (2006). Inequality Analysis:The Gini Index.Food and Agriculture

Organization, United Nations.Benos, E., Garratt, R.J., and Zimmerman, P. (2014).The role of counterparty risk in CHAPS

following the collapse of Lehman Brothers. International Journal of Central Banking10(4), 143–171.

Committee on Payments and Market Infrastructures (1993).Payment Systems in the Groupof Ten Countries. Bank for International Settlements.

Davey, N., and Gray, D. (2014). How has the liquidity saving mechanism reduced banks’intraday liquidity costs in CHAPS? Bank of England Quarterly Bulletin 54(2), 180–189.

Demirgüç-Kunt, A., and Levine, R. (2009). Finance and inequality: theory and evidence.Working Paper 4330, World Bank Policy Research.

Finan, K., Lasaosa, A., and Sunderland, J. (2013). Tiering in CHAPS. Bank of EnglandQuarterly Bulletin 53(4), 371–378.

Galbiati, M., and Giansante, S. (2010). Emergence of networks in large value paymentsystems (LVPSs). Working Paper 01/2010, Department of Economic Policy, Financeand Development, University of Siena.

Gini, C. (1912). Variabilità e mutabilità. Reprinted in Memorie di Metodologia Statistica(1955), Pizetti, E., and Salvemini, T. (eds). Libreria Eredi Virgilio Veschi, Rome.

Jackson, J. P., and Manning, M. J. (2007). Central bank intraday collateral policy andimplications for tiering in RTGS payment systems. In The Future of Payment Systems.Routledge.





20 E. Denbee et al

Jurgilas, M., and Žikeš, F. (2014). Implicit intraday interest rate in the UK unsecuredovernight money market. Journal of Financial Intermediation 23(2), 232–254.

Kakwani, N. C. (1977). Applications of Lorenz curves in economic analysis. Econometrica45(3), 719–727.

Manning, M., Nier, E., and Schanz, J. (2009). The Economics of Large-Value Paymentsand Settlement. Oxford University Press.

McAndrews, J. J., and Potter, S. M. (2002). Liquidity effects of the events of September 11,2001. Federal Reserve Bank of New York Economic Policy Review 8(2), 59–79.

Norman, B. (2010). Liquidity saving in real-time gross settlement systems: an overview.Financial Stability Paper 7, Bank of England.

Ray, D. (1998). Development Economics. Princeton University Press.Soramäki, K., and Bech, M. L. (2001). Gridlock resolution in interbank payment systems.

Working Paper 9, Bank of Finland.Valukas, A. R. (2010). Lehman Brothers Holdings Inc Chapter 11 proceedings examiner’s

report. Report, US Bankruptcy Court Southern District of New York, Jenner and BlockLLP.






Research Paper

Reaction functions of the participants inColombia’s large-value payment system

Constanza Martínez and Freddy Cepeda

Financial Infrastructure Oversight Department, Banco de la República, Carrera #14-78,Bogotá, Colombia; emails: [email protected], [email protected]

(Received April 14, 2015; revised August 24, 2015; accepted September 14, 2015)

ABSTRACT

Large-value payment flows can be disrupted by several types of failures, such asoperational incidents, problems experienced by the administrator of the paymentssettlement system, outages in communications networks and the inability of a partic-ipant to submit payments due to insufficient liquidity. During any of these incidents,the participants in the system can decide to stop, delay or continue sending paymentorders, depending on the elements that originated the disruption as well as on thealternative liquidity sources available to each entity. By means of Tobit models withrandom effects, we evaluate the payments activity of Colombian financial institutions.Our results suggest that their reactions vary in accordance with the type of incident,along with the type of financial institution and its role in the market.

Keywords: payment system; operational incidents; payment reaction function; coordination ofpayments; payment system resilience.

1 INTRODUCTION

Large-value payment systems (LVPSs) settle liquidity transfers (the cash leg) oftransactions between institutions that participate in the financial markets. The settled

Corresponding author: C. Martínez Print ISSN 2049-5404 j Online ISSN 2049-5412Copyright © 2015 Incisive Risk Information (IP) Limited

21




22 C. Martínez and F. Cepeda

payments in this system include all trading in financial assets, monetary policy oper-ations, central bank liquidity provision, direct funds transfers between financial insti-tutions, and operations between the central bank and the Ministry of Finance. InColombia, as in many other countries, the LVPS is operated by the central bank.

From 1998 until the beginning of 2006, the Colombian LVPS (Cuentas de Depósito,CUD) functioned as a pure real-time gross settlement system (RTGS), and, accord-ingly, the payments were liquidated in the same order that they were entered. As inany pure settlement system, the payments were processed as long as the funds inthe sender’s account were sufficient to allow their finalization. Otherwise, they wererejected. With the aim of alleviating the high liquidity requirements of the system, onJanuary 10, 2006, the Colombian central bank incorporated a queuing structure andtwo liquidity optimization mechanisms (netting cycles and retrial) into the CUD.

A netting cycle is an automated mechanism that checks an entity’s account bal-ance and calculates the offset value of incoming transfers and outgoing payments.If the funds in an entity’s account are sufficient, this value is settled. Otherwise, thesystem continues to make multiple attempts at removal payment orders until it cansettle the maximum value of payment. Five netting cycles are executed daily, at 11:50,14:20, 15:30, 16:15 and 17:45. However, under special circumstances the central bankcan activate more netting cycles, especially when the system exhibits considerablyhigh liquidity needs. Hence, the unresolved transactions that came from the centralsecurities depositary (Depósito Central de Valores, DCV) or from the CUD could beprocessed through forwarding payments or netting settlements.1 The retrial mecha-nism consists of an automatic checking of participants’ accounts balances, in order tosettle delayed payment orders. The DCV begins the retrial mechanism at 14:30 andrestarts every thirty minutes, at the end of each previous execution.2 In this settle-ment system, enhanced with this queuing structure, a financial entity that is expectingincoming transfers could send orders of payments without having enough money inits account, as the payments that it is about to receive will permit the achievement ofthe payments orders sent on that same day.

As noted by McAndrews and Potter (2002), the entities that participate in an LVPSmay have access to several sources of liquidity, such as their own deposits at the centralbank, money market loans, central bank liquidity and the expected incoming transfersfrom other financial institutions. The latter source determines the payments reaction

1 Netting cycles are not continuously activated during a day, given that every time this mechanismis activated it freezes the balance in each entity’s account with the aim of calculating the net valueof payments, considering both legs of transaction (securities and cash) (Bernal et al 2012).2 A detailed description of how the netting cycles and the retrial mechanism operate can be foundin Banco de la República (2010).





Reaction functions of the participants in Colombia’s large-value payment system 23

function, which, as McAndrews and Potter defined it, describes the interdependencebetween the participants of the payment system in terms of liquidity. A paymentreaction function is measured as a linear relationship between the payments sent andreceived by a financial institution.

For the Colombian central bank this subject is of particular importance given that,as LVPS owner and system operator, it may consider it optimal to induce more coop-erative behavior from participants that delay the orders of payments (exhausting liq-uidity), especially when the system is suffering from failures. The literature on theeconomics of payments has recognized the relevance of this subject, pointing out thatthe systemic impact of disruptions could be reduced if the participants were morewilling to cooperate with the execution of payments they owe to participants withhigh liquidity needs (Ledrut 2007).

In this paper we present several payments reaction functions of the financial institu-tions that participate in the CUD, estimated by means of random-effects Tobit models.In particular, we compare the estimated payments patterns on the days on which thepayments system has suffered from an incident both with those of a predeterminedbenchmark period (which contains data from prior to each incident) and with thoseof three days following the incident. Due to the fact that the CUD is a non-tieredpayment system, we include the entities with the greatest participation in the execu-tion of payments, of which there are three types: banks, brokerage firms and mutualfunds.3 Within each type, the entities included represent 80% of the average monthlypayments sent. The entities included comprise eight banks, eight brokerage firms andten mutual funds.4 To the best of our knowledge, this is the first attempt to empiricallyestimate payments reaction functions of the financial institutions that participate inthe CUD.

The results obtained suggest that there is coordination in the payments sent byfinancial market participants, but the coordination of the payments’ timing is ratherlow, as can be inferred from the size of the estimated parameters. Under normal

3 In a non-tiered LVPS all participants have direct access to the system, and hence it is the systemthat settles all payment orders.4 Other entities having access to the CUD include pension fund managers, commercial financialcorporations (leasing), financial corporations, financial cooperatives, insurance companies and spe-cial official institutions (Fogafin, Bancoldex). In December 2012, a total of 158 entities had directaccess to CUD. The average daily value of settled payments in 2012 was Ps38 132 billion, whichcorresponded to 8196 transactions per day. The average value of settled payments per day corre-sponded to 5.7% of the annual gross domestic product of 2012. For the same year, the system’sfee before 17:00 was Ps2360 per transaction, and after that time this ad valorem fee was Ps2.5per million Colombian pesos of value transferred, with a minimum value for this time window ofPs3410 per transaction.






circumstances a higher coordination of payments (or strategic complementarity) isvery likely, as long as it allows financial institutions to reduce their demand forliquidity, making payments less costly. However, after a failure, the coordination ofpayments weakens, potentially deteriorating the liquidity position of some financialinstitutions (eg, brokerage firms) that may also encounter difficulties in accessing thecentral bank’s liquidity, due, for example, to the lack of sufficient collateral.

This paper is organized as follows. Section 2 summarizes the literature on reactionsfunctions for the large-value payment system. Section 3 explains the methodologyused to estimate the payment reaction functions in specific periods and incidents.Section 4 contains the main estimation results, and Section 5 concludes.

2 LITERATURE REVIEW

As noted by McAndrews and Potter (2002), an entity that participates in an LVPS canfund its payments using its own deposits at the central bank, loans from the centralbank, money market loans and incoming transfers from other participants in the sys-tem. Given that some of the entities might not have enough liquidity in their accountsat the central bank and that the loans are available at a cost, the payments that theysend are mainly funded by the transfers that they receive from their counterparties.5

The received payments are attractive to the system participants as a source of liquid-ity because they are cost-free. But the excessive dependence on this liquidity sourcemay encourage the adoption of unwanted strategies (generating problems as the freerider) that could affect the flow of payments in the system. Thus, participants face thedilemma of incurring liquidity costs or delaying their payments (Bernal et al 2012).

The financial market participants’decisions about making early payments or delay-ing them critically depends on the relative costs of liquidity (the opportunity costof transferring securities as collateral, central-bank liquidity operations or money-market repo transactions) and the cost of delaying payments, as was shown by Bechand Garrat (2003). Hence, for a given structure of costs, participants could face theprisoner’s dilemma by choosing to delay their payments orders, although they wouldhad benefited even more from an early submission.

But, beyond the decision of making early payments or delaying them, what reallymatters is the coordination of the payments (synchronization) (McAndrews and Rajan2000), since this can become an essential source of liquidity, given by the directrelationship that exists in the frequent transactions between the same counterpartiesand the certainty of the incoming payments. As a result, there will be a higher degreeof certainty in relation to the payments an entity is expecting from the counterparties

5 In accordance with Becher et al (2008), incoming payments represented around 25% of the totalpayments sent through the UK sterling CHAPS system in October 2006.






with which it usually deals than in those coming from unusual counterparties. Thus,the coordination of payments could be increased as long as the patterns of payments,given by the same entities and with the same timing, are repeated day-to-day. Inthis setting, an unanticipated disruption could magnify the uncertainty with respectto the payments that entities are expecting. The strategic complementarity (ie, thecoordinated harmonious interaction between the decisions by all system participants)in the sending of payments can make incoming payments one of the main sources ofliquidity, smoothing the payments flows and hence contributing to financial stability(Bernal et al 2012).

In the economics of large-value payments the reaction function denotes the conceptof the relationship that exists between the payments sent and received by an entity,in an attempt to measure the coordination of payments between participants. Forthis reason, it is important to central banks, especially during times of LVPS failure(see McAndrews and Potter 2002; Ledrut 2007; Mills and Nesmith 2008; Perlin andSchanz 2010; Merrouche and Schanz 2010).

The operational incidents of the payments system are usually related to failuresthat affect the system operator, failures in the communications networks and failurescaused by the inability of one (or more) settlement participant to submit paymentorders to the system. Among these incidents, those that emerge from the systemoperator could be generated by failures in information technology, human error bythe staff operating the system or failures produced by external events such as naturaldisasters, power failures and terrorist attacks (Bedford et al 2005).

An empirical estimation of payments reaction functions of the US LVPS (Fed-wire) participants after the terrorist attacks on September 11, 2001 was developedby McAndrews and Potter (2002), who suggested that these attacks generated opera-tional disruptions in the communication networks and hence also caused a drop in thepayments sent through the system. But this situation was rapidly contained by meansof the actions taken by the Federal Reserve System, which were specially designedto inject liquidity into the system and restore the payments coordination equilibrium.In this setting, payments coordination has been recognized as an essential tool tomitigate the impact of disruptions in the payments system (Bech and Garrat 2012).

The temporary or permanent insufficiency of funds in the sender’s account couldalso have systemic effects on the LVPS. Ledrut (2007), Mills and Nesmith (2008),Merrouche and Schanz (2010) and Perlin and Schanz (2010) examine how the entitiesreact in response to an operational failure experienced by one of their counterpartiesthat also makes use of the payments system. According to them, a simulated shockto the biggest entity of the system will make the remaining participants stop sendingpayments to that entity, in an attempt to save liquidity. In line with these results, Benoset al (2012) found that after the collapse of Lehman Brothers the banks of the United






Kingdom (participants of CHAPS) delayed payments to their counterparties due toconcerns about both bank-specific and system-wide risks.

3 PAYMENT REACTION FUNCTIONS

A reaction function relates the payments that a participant of the LVPS sends to thepayments that it has received from its counterparties, as the following linear expressionsuggests (McAndrews and Potter 2002):

P At D ˛ C ˇRA

t C "t : (3.1)

Consistent with this function, participant A sends and receives payments at time t ,denoted by P A

t and RAt , respectively. The parameter ˛ represents the autonomous

willingness of this participant to send payments, regardless of the payments received.The slope of the reaction function, ˇ, represents the marginal propensity of partic-ipant A to send payments in response to those received from other participants inthe system. In terms of liquidity, this parameter could be considered as a signal ofhow cooperative an entity is. In such a way, a positive and significant ˇ is consideredto be a signal of the existence of payments coordination among entities, since thiscorresponds to a participant that sends payments as soon as new funds transfers arereceived into its account. Similarly, disregarding whether or not the funds recentlycollected (from its counterparties) by an entity allow it to complete a payment order,a negative ˇ will indicate that this entity delays its payments orders, while ˇ D 0

corresponds to an entity that does not send payments.

3.1 The estimation methodology

The chosen estimation methodology closely follows the structure of analysis proposedby McAndrews and Potter (2002), which consists of defining the dependent variableas the total amount of payments sent by an entity per minute. This variable is setas a function of the intercept term ˛ and the total value of payments received fromthe other entities via CUD in the previous fifteen minutes (ˇ).6 Other explanatoryvariables are the opening balance, and the cumulative receipts minus the cumulativepayments sent up to sixteen minutes prior to each minute.

In our data, the dependent variable exhibits a large number of observations thatare equal to zero, which reflects the cases in which an entity did not send payments

6 The fifteen-minute time interval, proposed by McAndrews and Potter (2002), has also beenaccepted in other empirical studies as a standard unit for estimating payments reaction functions(Afonso and Shin 2010). Thus, when the same time interval is considered, comparisons among dif-ferent LVPS can be easily established. Besides, it is quite common in the local market that an entityin trouble asks the system’s operator for fifteen minutes in which to restore its normal paymentsflow.






through the system. This type of variable has been termed a “corner solution response”in the literature, given that it relies on a continuous distribution over positive values,but there also exists a nonzero probability of that variable taking a value of zero(Yermack 1995; Wooldridge 2010). Linear models such as the standard fixed-effectpanel data model were not considered in this study, because these models ignoredthe fact that the dependent variable (measured by the total payments sent) can takea positive value or be equal to zero and therefore they could produce inconsistentparameter estimates. Nonlinear panel data models, such as the Tobit model withfixed effects, were also discarded given that in small and fixed samples the varianceestimator could be affected by the fixed effects, which could weaken the inferencesdrawn from the estimated parameters (Greene 2004). An alternative nonlinear paneldata model, the Tobit model with random effects, adequately captures the statisticalproperties of the dependent variable, and for that reason it is preferred for estimatingthe reaction functions.

In this context, the random-effects Tobit model is given by

P A�t D ˛ C XA0

t ˇ C "t ; t D 1; 2; : : : ; T; (3.2)"t

XA0t

� Normal.0; �2/: (3.3)

The value of the payments sent (P A�t ) takes the form of a corner solution response:

P At D

(P A�

t if P A�t > 0;

0 otherwise:

The information matrix XAt contains the set of regressors mentioned above and

two dummy variables. The first dummy reflects the facility that CUD provides tosettle transactions coming from DCV or money transfers that were completed usingnetting cycles. This dummy takes the value 1 when the CUD’s liquidity-saving mech-anisms are activated and 0 otherwise. The other dummy represents the automaticpayments in the system, that is, the settlement of the net value resulting from clearingat the CEDEC (local check clearing system) and ACH-CENIT (the automatic clearinghouse), as these payments do not represent the financial institution’s willingness tosend payments as the reaction function describes.7

Equation (3.3) assumes normality in the error term "t , and implies that the strictexogeneity of XA

t does not hold, given that the exogenous variables .XAt / could

7 Other “automatic payments”, such as those executed through the retrial mechanism, were notincluded in the models because it was not possible to separately identify this type of payment in theCUD. Likewise, the debits coming from DECEVAL (the central securities depository for corporateand government, nonsovereign securities) were not included, given that these payments cannot beregarded as automatic payments given that each entity can manage the balances of the accountassociated with the CUD and hence decide when a payment order will be executed.






be affected by the possible non-contemporaneous feedback that could emerge fromlagged values of the dependent variable .P A

t�S /.

3.2 Entities selected and incidents tested

In order to select the entities for our estimations, we examine data registered duringApril 2011, given that this is very close to the average yearly payments executedthrough the CUD. The payments sent by banks, brokerage firms and mutual funds inthat month represented 84.6% of the total payments that were settled in the system.For each type of entity, individual entities were chosen to account for at least 80%of the payment flows. This corresponded to eight banks (80.1%), eight brokeragefirms (80.6%) and ten mutual funds (80.5%). The payments sent by these twenty-sixentities represented 67.9% of the total payments sent through the system. In order tocomprehend how these entities reacted to different types of incidents, we estimated thereaction functions in response to events that would have affected the normal patternof payments activity.

Four incidents are considered. The first was originated by a power outage, whilethe other three were caused by specific entities’ problems. Within this last groupof incidents, the two that occurred most recently are related to liquidity problemssuffered by an entity. However, they do not correspond to the type of failure causedby the inability of one settlement participant to submit payments, in the sense ofBedford et al (2005), given that in the CUD all participants have direct access tothe system and hence it is not possible to classify them as settlement and customerfinancial institutions.

These incidents, which are public knowledge, are briefly described next.

3.2.1 The blackout (April 26, 2007)

In Colombia, the power outage that occurred on April 26, 2007 is the only case ofoperational failure that has affected the CUD so far. The disruption in the electricityflows, caused by a power overcharge, started in Bogotá at 09:58 and lasted until 14:30.This outage interrupted the supply of electricity to the entire country.

3.2.2 Bancolombia’s operational failure (February 26, 2010)

The technical failure of this bank, which is the country’s largest by asset size, gen-erated several difficulties in its transactional channels, eg, stopping its clients fromwithdrawing or depositing money, and delivering incorrect deposit account balanceinformation. This incident was originated by a failure of the bank’s technologicalplatform, which started on Friday, February 26, 2010 and lasted around three days.At the beginning of this incident, the highest delay by value in Bancolombia’s settled






payments via CUD (10.8%) was registered at 15:00; this delay corresponded to 5.5%of settled payments in the system.

3.2.3 The failure of Proyectar Valores (June 23, 2011)

The first of two disruptions related to financial resources management was caused bythe brokerage firm Proyectar Valores, which suspended its payment orders at the endof May 2011. On May 27 the Financial Superintendency of Colombia (FSC) adopteda preventive measure consisting of a special supervision measure to this brokeragefirm (FSC Resolution number 0826). On June 22 the FSC decided to take over thefirm (FSC Resolution number 1000) and announced the suspension of its activities inthe market. On October 4 the FSC ordered its compulsory administrative liquidation(FSC Resolution number 1714).

3.2.4 The failure of Interbolsa (November 2, 2012)

The second case related to financial resources management was that of Interbolsa,which was then the largest brokerage firm operating in the Colombian stock and moneymarkets. The failure to pay an intraday credit of Ps20 billion (US$11 million) to a localbank obligated an intervention through the FSC (FSC Resolution number 1795) onNovember 2, 2012. Two working days later, on November 7, 2012, the FSC orderedits compulsory liquidation (FSC Resolution number 1812).

4 THE MAIN ESTIMATION RESULTS

4.1 Payment reaction functions

For the empirical estimation of the reaction functions, we used minute-by-minutedata of the payments registered from 07:00 until 20:00, given that this is the periodin which the CUD operates.8 The random-effects Tobit models were estimated usingthe method of maximum likelihood.9 The main estimation results are summarized inTable 1 on the next page, Table 3 on page 34, Table 5 on page 36 and Table 7 onpage 39. Each table includes the parameters estimated for the benchmark period, forthe day of the incident and for the following three days. The outcomes referred to as“all” correspond to the three types of entities included: banks, brokerage firms andmutual funds.

8 Before 07:00, CUD settles the fees charged for its services and the financial transactions’ taxes.After 20:00 the transfers are related to the constitution or retrocession of the central bank’sremunerated deposits, and interest payments and capital amortizations of sovereign bonds (TES).9 We obtained robust standard errors estimated by bootstrapping, to account for the possible problemof serial correlation in data that may emerge from this static model.






TABLE 1 The blackout. [Table continues on next page.]

(a) Benchmark (April 1–25)

Brokerage MutualAll Banks firms funds

Reaction function slope 0.025 0.023 0.008 0.101(3.11)�� (3.85)�� (0.76) (10.08)��

Autonomous willingness 1.9E+09 2.2E+09 1.5E+09 2.5E+09to send payments (8.90)�� (6.77)�� (4.85)�� (4.12)��

Number of observations 39 444 19 875 15 627 3 942

Number of participants 25 8 7 10

(b) The blackout (April 26)


Reaction function slope 0.024 0.025 0.000 0.120(1.47) (1.19) (0.02) (4.05)��

Autonomous willingness 1.9E+09 2.8E+09 1.7E+09 1.3E+09to send payments (4.31)�� (1.59) (1.37) (2.28)��

Number of observations 2 486 1 275 950 261


(c) Benchmark (April 27)



Autonomous willingness 1.8E+09 3.2E+09 1.8E+09 2.0E+09to send payments (5.59)�� (1.71)� (2.54)�� (4.44)��



The benchmarks

As long as the patterns of payments change with time, and provided that the incidentsare separated between them for more than a year, we can examine them separatelyby defining particular benchmark periods per incident. Each benchmark includesinformation on the payments settled in the days prior to the occurrence of the incident,within the month in which the incident took place (when it was possible). In this section






TABLE 1 Continued.

(d) April 30


Reaction function slope 0.028 0.019 0.042 0.065(1.66)� (0.84) (0.99) (0.40)

Autonomous willingness 2.0E+09 2.4E+09 2.8E+09 2.2E+09to send payments (6.25)�� (1.88)� (2.62)�� (1.78)�



(e) May 2


Reaction function slope 0.022 0.019 0.021 0.080(4.09)�� (1.72)� (1.45) (0.70)

Autonomous willingness 1.9E+09 2.5E+09 2.8E+09 2.5E+09to send payments (8.43)�� (3.38)�� (0.76) (3.43)��

Number of observations 2 779 1 429 1 066 284


Source:authors’calculations (t -statistic values are given in parentheses).�, �� and �� denote statistical significanceat the 10%, 5% and 1% levels, respectively.

we compare the estimated parameters for the day of the incident with those obtainedfor the benchmark period and the next three days.

The incidents

Operational failures can have several causes, which may have different consequenceson the flow of payments. In this section we examine the effects that some incidents hadon the settlement system, starting with a power outage (the blackout). The estimatedparameters correspond to the average marginal effects. The remaining results as wellas some additional statistics (total value of payments, opening balance, central bankliquidity and the number of send out payments) can be consulted in the tables presentedin Annex A online.10

10 Even excluding the dummy of clearing at the CEDEC and ACH-CENIT, the estimated slopesremain practically unaltered, due to its low level of participation in the total payments settled in theCUD. In 2012, for example, the net value resulting from this clearing barely represented 1.2% ofthe average yearly payments settled in the system.






TABLE 2 Average liquidity savings achieved through liquidity savings algorithms overDCV’s transactions, April 2007.

Mutual Brokerage TotalAverage Banks funds firms system

Gross value 1 274 787 64 126 884 504 2 554 520Net value 132 212 22 706 129 997 366 661Liquidity saving (%) 89.6 64.6 85.3 85.6

Statistics correspond to the type of entities analyzed in this paper. All values are given in millions of Colombianpesos. Source: authors’ calculations with data from CUD.

The benchmark period designated to evaluate the power outage includes data ofthe payments registered from April 1 to April 25, 2007. As can be seen in Table 1on page 30, the marginal propensity to submit payments (reaction function slope) ofbrokerage firms before, after and during this incident were nonsignificant, suggestingthat this variable that measures their sending of payments in response to the liquidityreceived from its counterparties was null. The same occurred with the estimatedintercept term. These results, considered along with those of the dummy of liquidity-saving algorithms (netting cycles), are almost always positive and significant (as canbe seen in Annex A online): evidence that this group of entities usually has low levelsof liquidity, which makes them more dependent on the activation of this mechanismto fulfill their payment obligations.

During April 2007, the average liquidity saving reached through the liquidity-saving mechanisms (LSMs) for compliance of transactions originated in the centralsecurities depositary (DCV) was greater than 85% for the entire system. Table 2 showsthe average liquidity saving per entity type, obtained as a benefit of the LSM duringthe month in which the power outage occurred.

The brokerage firms’ liquidity savings were 85.3%, slightly lower than the per-centage registered for banks (89.6%). For mutual funds and banks the results indicatethat although the activation of the LSM partly explains the payments that these enti-ties made, their autonomous willingness to send payments, which depends on theirdeposit account balances at the central bank, is also an important determinant.

Both the size and scope of the reaction functions of mutual funds and banks arevery different. For the group of banks this parameter turned out to be nonsignificanton the day of the disruption, while it was significant for mutual funds. Nevertheless,the ability of banks to smooth the impact of the power outage on the payments systemwas low, given that the estimated willingness to send payments (“reaction functionslope”) for the entire system (“all”) was usually similar to that exhibited by banks.A plausible explanation for this outcome is that banks are the type of entities thatcontribute the most to the payments in the system.






Once the electricity flows were restored on the afternoon of April 26, mutual fundswere the only group that continued to exhibit significant payments reaction slopes;while banks decided to stop sending payments on the day of the incident but returnedto their earlier payments strategy three days later.

Similar results were obtained from the examination of Bancolombia’s technicalfailure (Table 3 on the next page).11 The mutual funds were again the only entity typeto be most likely to send payments during the incident. Nevertheless, their marginalpropensity parameter fell from 0.085 in the benchmark period to 0.057 on the day ofthis failure.

The system as a whole (“all”) reacted in a very similar manner to banks and broker-age firms, which, as the estimating results indicate, were reluctant to send paymentsduring this failure. These outcomes coincide with their participation in the monthlyaverage gross value settled during the LSM activation (banks: 59.4%; brokerage firms:21.1%); but also with the liquidity savings that they attained in the month in whichthe failure occurred (86.0% and 90.7%, respectively; see Table 4 on page 35).

Brokerage firms are the market participants that benefit the most from the LSM;this could be explained by the low level of deposits that they have in their centralbank’s accounts and the sizable volume of transactions (conducted as part of theirbusiness). The benefits in terms of liquidity savings that brokerage firms and banksachieved under LSM (to DCV’s transactions) have become an incentive for delayingthe compliance of payments until the activation of the netting cycles.

Some similarities and differences can be identified when comparing these twoincidents. The central bank did not adopt contingency arrangements in any of thesecases, and hence the settlement system went back to its normal pattern of paymentswithout any intervention. These failures were both short-lived, but only the poweroutage generated a temporary payments disruption, since the system was not con-stantly available to settle payments.12 The electricity flow was restored in less thanfive hours after this incident, allowing the disbursement of delayed payment orderson that day. Accordingly, the payments coordination was fully recovered three dayslater, although the CUD showed some signs of recovery a day before.

The technical problems of Bancolombia started on a Friday and lasted two moredays (all weekend). But the system’s payments coordination was not restored until

11 The operational failure of this bank lasted three days, but this incident was evaluated using onlyinformation regarding the payments registered on February 26, 2010, which was a Friday, giventhat the markets are closed during weekends.12 During this incident quite a few payments were settled using the contingency arrangements offeredby the central bank. However, the number of payments that were settled was considerably smallerthan the number registered in a normal day.






TABLE 3 Bancolombia’s operational failure. [Table continues on next page.]

(a) Benchmark (April 1–25)


Reaction function slope 0.009 0.007 0.008 0.085(1.84)� (1.44) (1.02) (5.63)��




(b) Bancolombia’s operational failure (February 26)


Reaction function slope 0.008 0.002 �0.005 0.057(1.33) (0.14) (�0.26) (8.18)��




(c) March 1


Reaction function slope 0.000 �0.007 0.011 0.070(�0.02) (�0.62) (1.08) (2.66)��




three days afterward. Perhaps the longer duration of this incident could have madethe other market participants less confident in the bank’s ability to solve its technicalproblems. Hence, it is reasonable to think that the negative consequences on the stabil-ity of the payments pattern would have been exacerbated by the market’s perceptionof systemic importance regarding Bancolombia.






TABLE 3 Continued.

(d) March 2


Reaction function slope 0.035 0.028 0.023 0.020(2.17)�� (1.52) (1.04) (1.12)

Autonomous willingness 1.8E+09 3.9E+09 2.0E+09 2.7E+09to send payments (3.71)�� (2.65)�� (1.15) (4.36)��



(e) March 3


Reaction function slope 0.002 �0.003 0.002 0.031(0.32) (�0.33) (0.09) (1.34)





TABLE 4 Average liquidity savings achieved through liquidity savings algorithms overDCV’s transactions, February 2010.


Gross value 3 790 582 137 470 1 349 388 6 386 608Net value 529 104 49 478 125 638 862 546Liquidity saving (%) 86.0 64.0 90.7 86.5


The consequences that the inability of a financial institution to submit paymentscould have on the payments system are examined with the failures of ProyectarValoresand Interbolsa. The duration of these incidents was slightly longer than that of theincidents presented above, given that, as a result of the FSC investigations, bothbrokerage firms were liquidated.






TABLE 5 The failure of Proyectar Valores. [Table continues on next page.]

(a) Benchmark (June 1–22)






(b) Proyectar failure (June 23)


Reaction function slope 0.027 0.022 0.019 0.092(1.58) (0.89) (1.77)� (5.23)��




(c) June 24



Autonomous willingness 2.0E+09 4.0E+09 1.7E+09 2.5E+09to send payments (4.58)�� (3.62)�� (1.77)� (6.27)��



For the benchmark period selected to test the failure of Proyectar Valores(June 1–22, 2011), the estimation results indicate that the payments settled in thesystem (group “all”) were explained by the liquidity that all types of entities provided(Table 5). However, this outcome exclusively corresponds to the strategy of paymentsselected by banks and mutual funds, given that brokerage firms, as usual, exhibited anull payments reaction function slope.






TABLE 5 Continued.

(d) June 28


Reaction function slope 0.010 0.006 0.034 0.118(1.30) (0.75) (1.77)� (5.20)��




(e) June 29







Banks stopped their payment instructions on the day that the FSC intervened withthis brokerage firm (June 23), and continued to withhold payment instructions fortwo more days, while mutual funds continued to send payments on June 23 and thefollowing three days. The brokerage firms reacted in a very different way, since theyincreased their marginal willingness to send payments during the incident, contraryto their payment strategies one day and three days later. All these results indicate thateach entity (group of entities) could adjust its own strategies of payments (timingin the sending of payments) according to its level of liquidity, alternative sources offunding, information and expectations.

In relation to DCV’s transactions, the data registered in the month that this incidentoccurred indicates not only that banks and brokerage firms were once more the largestparticipants (Table 6 on the next page), but also that they were the entities that attainedthe greatest liquidity savings as a result of the netting cycles (brokerage firms 91.7%;banks 84.3%).






TABLE 6 Average liquidity savings achieved through liquidity savings algorithms overDCV’s transactions, June 2011.




The reaction function slopes estimated for the benchmark period selected to test thefailure of Interbolsa (October 1 to November 1, 2012) are significant for banks andmutual funds. These results, in other words, suggest that the payments flows comingfrom these entities were self-funded by means of a sound matching of the paymentssent with the payments received (Table 7 on the facing page).

Mutual funds estimations coincide with those obtained for the entire system (“all”),as the parameters of interest (reaction function slope) for both types of entities aresignificant on the day of the incident (November 2) and on the following three days.This seems to indicate that the Interbolsa failure stimulated the cooperation of theother brokerage firms in the sending of payments, as can be seen from the increase intheir marginal willingness to send out payments.

Despite this failure, the payments activity was rapidly resumed by the immediateintervention of the FSC, under whose authority all payments were totally settled. Thecentral bank increased access to liquidity facilities, widening the range of collateralto nonsovereign securities (eg, corporate) and increasing the size of overnight repoauctions by 39.7% with respect to the limit offered on the previous day. But theliquidity used by financial entities was considerably lower in percentage terms (71.7%)than that used on November 1 (98.1%).13 On the day of the failure, the Colombiancentral bank scheduled six additional netting cycles. In total, eleven netting cycleswere activated (at 11:28, 11:50, 12:30, 13:00, 13:30, 14:00, 14:20, 15:30, 16:15,16:55 and 17:45), but the total value of settled payments barely reached one-fifth ofthe average daily value of payments settled a month before this incident.

The amounts of registered payments in the CUD suggest that the increasing percep-tion of risk that arose from the entities’ knowledge of the Interbolsa situation reduced

13 On November 2, 2012 the daily cap was set at Ps8.1 billion, of which Ps5.8 billion were used byfinancial entities.The Colombian central bank’s External Resolution of November 7, 2012 included credit securitiesrated by rating agencies, which broadened the eligible securities that central bank could (permanentlyor temporarily) acquire in order to regulate the liquidity of the economy.






TABLE 7 The failure of Interbolsa. [Table continues on next page.]

(a) Benchmark (June 1–22)






(b) Interbolsa failure (November 2)


Reaction function slope 0.049 0.048 0.036 0.161(1.96)�� (1.36) (1.35) (7.94)��




(c) November 6


Reaction function slope 0.079 0.078 0.073 0.172(4.24)�� (1.82)� (6.11)�� (4.04)��

Autonomous willingness 2.7E+09 4.5E+09 2.6E+09 1.9E+09to send payments (5.43)�� (2.50)�� (4.86)�� (1.84)�

Number of observations 2 144 1 216 668 260Number of participants 26 8 8 10

trading in the stock and money markets. There was a 25.9% reduction in the numberof settled payments coming from all financial entities on that day (2495 transactions)compared with the previous day (November 1), and 36.2% compared with the dailyaverage of the preceding month (3911 transactions).

In relation to the average daily value liquidated a month before (in October), thesettled payments in the debt government securities market decreased on November 2






TABLE 7 Continued.

(d) November 7


Reaction function slope 0.060 0.059 0.064 0.127(3.69)�� (1.51) (4.15)�� (4.08)��

Autonomous willingness 2.6E+09 4.0E+09 1.8E+09 2.2E+09to send payments (5.04)�� (1.91)� (2.13)�� (3.79)��



(e) November 8


Reaction function slope 0.034 0.033 0.043 0.054(6.21)�� (2.25)�� (9.60)�� (2.64)��





TABLE 8 Average liquidity savings achieved through liquidity savings algorithms overDCV’s transactions, October 2012.




by more than Ps6.6 billion (82.3%), while for the sell/buybacks that value was reducedto Ps1.3 billion (54.1%). Banks and brokerage firms were cautious and preferred toreduce their trading activity on that day. That reduction in the volumes traded easedthe fulfilment of the payments owed between entities without the necessity to exhaustthe repo limit offered by the central bank. The levels of liquidity savings reached in






the month before the incident were 95.2% for brokerage firms and 87.6% for banks(Table 8 on the facing page). Nevertheless, the failure of Interbolsa produced differenteffects in the CUD due to its considerable relative size in the market.

For the days after the incidents of Proyectar Valores and Interbolsa the estimatedreaction parameters slopes are positive, suggesting the existence of coordination inthe payments’ timing, especially in banks and mutual funds. These results coincidewith those obtained by McAndrews and Potter (2002), in that as long as the pay-ments coordination among the participants of the LVPS increases, the number ofpayments sent by each entity to its counterparties will also increase. Nevertheless, forthese specific cases, the degree of payments coordination is rather low, given that theparameters for banks do not exceed a value of 0.08, while those for mutual funds arestill located at low levels (all lower than 0.18), albeit higher than banks.

Two similarities between these incidents can be identified. First, the FSC, notingthat these financial institutions were unable to make payments, took them over andannounced its liquidation process. Second, the central bank also intervened, opting toextend the LVPS’s closing time as a contingency arrangement. In the case of ProyectarValores the CUD was closed at 21:00 (one hour later than the scheduled time on anormal day), while in the case of Interbolsa the system operated until 23:00.

On the day that Interbolsa failed to pay its obligation with a local bank, the centralbank offered additional liquidity to the system to prevent liquidity shortages that couldhave made other entities fail to meet their payment obligations. However, not a singleentity used these liquidity facilities. For the Proyectar Valores incident, the liquidityprovision of the central bank remained unchanged, perhaps due to the relatively smallsize that this entity represented in the stock and money markets.

Disregarding the nature of the four tested incidents, the system’s payments coordi-nation was affected to the extent that all of them altered the payments flows. Never-theless, the blackout was the only one that caused a disruption to the payments flow.This incident and Bancolombia’s technical outage lasted a short time, and hence thesettlement system rapidly returned to its normal pattern of payments (the paymentscoordination was restored between the second and third day after each incident). Theother incidents (caused by the inability of a participant to submit payments) had adifferent ending, given that in both cases the financial entity that suffered from liq-uidity problems was closed and liquidated. Consequently, the reestablishment of thecoordination of payments was accompanied by the loss of one of the participants inthe settlement system.

All in all, the results obtained suggest that there is coordination in the sending ofpayments among the CUD participants. Yet this coordination is rather low, as can beinferred from the small size (low value) of the estimated payment reaction functionslopes. Under normal circumstances an increase in the coordination of payments (thereaction function parameter) will be desirable given that this will allow an entity to






make less costly payments; at the same time this will reduce its demand for liquidity.The coordination could be increased further if a large number of system participantssend payments during the netting cycles. But, as noted by Bernal et al (2012), as longas the entities with less access to central bank liquidity and lower opening balances(such as brokerage firms) become more dependent on the payments received fromtheir counterparties, the negative impact of temporary disruptions in the system couldbe exacerbated, given that these latter entities can suddenly stop the payments ordersdue to their risk concerns.14

Even when some policy options, such as the implementation of liquidity-savingalgorithms (netting cycles) and the incentives for early submission of payments, havebeen adopted in the CUD, other alternatives that could allow mitigation of the riskof payments delays remain pending. The most prominent of these strategies consistsin the implementation and enforcement of binding throughput rules, successfullyadopted in other LVPSs such as CHAPS in the United Kingdom.15. This strategy hasrepresented risk-reduction benefits (Buckle and Campbell 2003) and has facilitatedliquidity recycling in that system (Ball et al 2011).

4.2 Analyzing the failure of Interbolsa

The failure of Interbolsa, being the most recent of the incidents examined, is againconsidered here. With the aim of a deeper understanding of the way in which thesettled payments evolved over time after this failure, we describe Interbolsa’s day-to-day payments activity, using graphical analysis of the estimated reaction functionsslopes over time, holding all other variables constant. To this end, the set of graphsin Figure 1 includes the parameter estimates of the marginal propensity to send outpayments for the benchmark period (represented by the dashed line), and the resultsobtained for each day until the date in which the reaction function slope returned tothe benchmark (November 15).

As can be seen in Figure 1, the slopes of the payment reaction functions for thethree groups of entities (“all”; part (a)) closely follow the behavior exhibited bybanks (part (b)), as their trends and values evolve in a very similar manner. In otherwords, the dynamics of the entire LVPS are mostly explained by banks’ paymentinstructions. This result coincides with those obtained from the Chernoff faces (ascan be seen in Annex B online), from which we identified similarities in the dynamics

14 Bernal et al (2012) measured the contribution of liquidity sources in the Colombian LVPS inMay 2010 and found that brokerage firms and mutual funds (trust companies) relied heavily on therecirculation of account-balance funds in 85% and 75% of their payments.15 Binding throughput rules consist of specific policies that force entities to send payments early, soas to complete a predetermined percentage of payments at a specific hour of a day






FIGURE 1 Estimated slope of reaction function per type of entity.

0.160.130.100.070.040.01

–0.02

0.160.130.100.070.040.01

–0.02

0.160.130.100.070.040.01

–0.02

0.160.130.100.070.040.01

–0.02

Ben

chm

ark

Nov

2

Nov

6

Nov

7

Nov

8

Nov

9

Nov

13

Nov

14

Nov

15

Ben

chm

ark

Nov

2

Nov

6

Nov

7

Nov

8

Nov

9

Nov

13

Nov

14

Nov

15

Ben

chm

ark

Nov

2

Nov

6

Nov

7

Nov

8

Nov

9

Nov

13

Nov

14

Nov

15

Ben

chm

ark

Nov

2

Nov

6

Nov

7

Nov

8

Nov

9

Nov

13

Nov

14

Nov

15

(a) (b)

(c) (d)

(a) All. (b) Banks. (c) Brokerage firms. (d) Mutual funds. Dashed line denotes the benchmark. Source: authors’calculations.

of their reaction parameters (slopes of the reaction function represented by the mouthcurvature), as well as in their access to central bank liquidity and hub centrality.16

Two days after the failure of Interbolsa, banks were reluctant to submit paymentsorders to the system, which seems to have worsened (a day later, on November 8) thepayments coordination among all types of financial institutions.

Brokerage firms reacted in a totally different way than usual in terms of payments.Their reaction function slope overpassed its corresponding benchmark parameter(0.004) on the day of the failure (passing to 0.036 on November 2) and continued that

16 This graphical method, proposed by Chernoff (1973), allows the inclusion of multivariate data ina single graphic that enables development of a sensitivity analysis per period.Based on the Hyperlink Induced Topic Search (HITS) information retrieval algorithm (Kleinberg1999), hub centrality is a measure of the global importance of a participant as a distributor or senderwithin a network or system. This measure determines the importance of a participant as a weightedaverage of the importance of the participants it distributes or sends to. As in León and Pérez (2012),in this paper hub centrality metric is related to payments.






way until November 14, reaching a peak (0.134) four days later (on November 9).From these results it is reasonable to think that these entities adopted a cooperativebehavior, but the change in their payments activities is explained by reasons otherthan just a sense of cooperation within the group. Among these reasons we found

(i) a temporary concentration of unsettled payments (from this and other brokeragefirms) that occurred during the takeover by the FSC and the beginning of itsliquidation process: time in which trading in the stock market slowed down;and

(ii) the early termination of obligations that brokerage firms (and other financialinstitutions) were subject to under the Colombian Securities Act of 2005.17

Thus, when payments activities were resumed, several pending payments coming frombrokerage firms were sent through the system, increasing their marginal propensity tosend out payments (reaction function slope, on November 9) further than this groupof entities usually exhibit under normal circumstances.

Unlike banks and brokerage firms, mutual funds exhibited the highest marginalpropensity to submit payments on all these days, but they also showed a parameterthat remained positive since the beginning of the failure. However, the ability of thesefinancial institutions to mitigate the liquidity problems that emerged in those dayswas low, as can be seen when comparing this group with the entire system (“all”).

The settlement system remained under stress until November 15 (seven days afterthe failure) when these financial institutions finally converged to the benchmark pointestimated for each one of them (see Figure 1 on the preceding page and Annex Bonline). As expected, this recovery emerged only until the banks’ reaction functionparameters went back to normality. In contrast to banks and mutual funds, the bro-kerage firms reaction function slope became negative, but not statistically significant,which could indicate that these entities returned to their noncooperative paymentsstrategy.

Our results indicate that banks are the most important institutions in terms ofliquidity, but their marginal propensity to send out payments during this incident wasmoderate. Mutual funds continued cooperating (sending payments), but their capacityto smooth the downward liquidity cycles suffered by the entire system was nil. Thepayments patterns of the brokerage firms, which are represented by a low marginalpropensity to submit payments along with high liquidity needs, could explain whytheir contribution to restore the payments coordination after a disruption is negligible.

17 Securities and Exchange Law 964 (2005).






5 CONCLUSIONS

In the payments literature the concept of reaction functions had been used to measurehow the strategy of sending payments could be affected by disruptions. The estimationof this concept has been done by means of a linear function that relates the paymentssent and received by an entity through the LVPS.According to the pioneering researchof McAndrews and Potter (2001), a positive estimation of the reaction function isindicative of the existence of payments coordination among the entities, which is adesirable condition especially during periods of disruptions.

The Colombian LVPS (CUD) has suffered from a few failures, from among whichwe evaluated the most recent: the power outage, a failure in Bancolombia’s operationalplatform and two other cases caused by the inability of an entity to submit payments(Proyectar Valores and Interbolsa). All of them temporarily affected the system’spayments coordination. However, the resiliency, understood as the time required by thesystem to recover its normal payments pattern, was longer for the incidents generatedby the liquidity problems of an entity, given that in both cases the entity was liquidated.

In response to the disruption of the payments activity of Interbolsa, for exam-ple, the remaining brokerage firms stopped sending payments as the market sloweddown. After the FSC intervened and the central bank offered some liquidity facil-ities, these entities reestablished payments flows even further than what was pre-viously registered. The estimated results allowed us to recognize that the reac-tion of the remaining financial market participants to this incident was representedby a null change in the pattern of payments sent by mutual funds, which con-tinued to exhibit the highest degree of cooperation, and a sudden stop in banks’payments. The heterogeneous responses exhibited by each entity type to this fail-ure could be considered to be a signal of a breakdown in payments coordina-tion. From the benchmark period selected to test this incident, we noted that thereestablishment in the sending of payments (resilience) took around seven workingdays.

Some possible extensions on this topic include the development of a methodologythat allows specific system participants’ behavior to be characterized in regard to theavailability of their liquidity sources and their payment strategies, such as free riders.

The methodology used in this paper could be helpful in overseeing the functioningof the payment reactions between entities.


The views expressed in this paper are the sole responsibility of the authors and do notreflect those of Banco de la República or its board of directors.






ACKNOWLEDGEMENTS

The authors thank Pamela Cardozo, Clara Machado, Carlos León, Fabio Ortega andJhonatan Pérez for their helpful comments and suggestions. The comments receivedfrom the Banco de la República Monetary and International Investment Division’sstaff are also acknowledged and appreciated. The authors also thank all commentsfrom the participants to the Thirteenth Payment and Settlement System SimulationSeminar and Workshop, held August 27 and 28, 2015, in Helsinki, Finland.

REFERENCES

Afonso, G., and Shin, H. (2010). Precautionary demand and liquidity in payment systems.Staff Report 352, Federal Reserve Bank of New York.

Ball, A., Denbee, E., Manning, M., and Wetherilt, A. (2011). Intraday liquidity: risk andregulation. Financial Stability Paper 11, Bank of England.

Banco de la República (2010).Reporte de Sistemas de Pago, June.Banco de la República,Bogotá, Colombia.

Bech, M., and Garratt, R. (2003). The intraday liquidity management game. Journal ofEconomic Theory 109(2), 198–219.

Bech, M., and Garratt, R. (2012). Illiquidity in the interbank payment system followingwide-scale disruptions. Journal of Money, Credit and Banking 44(5), 903–929.

Becher, C., Galbiati, M., and Tudela, M. (2008). The timing and funding of CHAPS sterlingpayments. Federal Reserve Bank of New York Economic Policy Review, September,113–133.

Bedford, P., Millard, S., and Yang, J. (2005). Analysing the impact of operational incidentsin large-value payments systems: a simulation approach. In Liquidity, Risks and Speedin Payment and Settlement Systems: A Simulation Approach, Bank of Finland Studies,Volume E31, Chapter 9, pp. 249–276.

Benos, E., Garratt, R., and Zimmerman, P. (2012). Bank behaviour and risks in CHAPSfollowing the collapse of Lehman Brothers. Working Paper 451, June, Bank of England.

Bernal, J., Cepeda, F., and Ortega, F. (2012). Estimating the contribution of liquiditysources in the Colombian large-value real-time gross settlement payment: a preliminaryapproach. Journal of Payments Strategy and Systems 6(2), 159–182.

Buckle, S., and Campbell, E. (2003). Settlement bank behaviour and throughput rules inan RTGS payment system with collateralised intraday credit. Working Paper 209, Bankof England.

Chernoff, H. (1973). The use of faces to represent points in K-dimensional spacegraphically. Journal of the American Statistical Association 68(342), 361–368.

Greene, W. (2004). Fixed effects and bias due to the incidental parameters problem in theTobit model. Econometric Reviews 23(2), 125–147.

Kleinberg, J. (1999). Authoritative sources in a hyperlinked environment. Journal of theACM 46(5), 604–632.

Ledrut, E. (2007). How can banks control their exposure to a failing participant? In Sim-ulation Studies of Liquidity Needs, Risks and Efficiency in Payment Networks, Bank ofFinland Studies, Volume E39, Chapter 8, pp. 227–252.






León, C., and Pérez, J. (2014). Authority centrality and hub centrality as metrics of systemicimportance of financial market infrastructures.Journal of Financial Market Infrastructures2(3), 67–87.

McAndrews, J., and Potter, S. (2002). Liquidity effects of the events of September 11, 2001.Federal Reserve Bank of New York Economic Policy Review, November, 59–75.

McAndrews, J., and Rajan, S. (2000). The timing and funding of Fedwire funds transfers.Federal Reserve Bank of New York Economic Policy Review, July, 17–32.

Merrouche, O., and Schanz, J. (2010). Banks’ intraday liquidity management during oper-ational outages: theory and evidence from the UK payment system. Journal of Bankingand Finance 34, 314–323.

Mills, D., and Nesmith, T. (2008). Risk and concentration in payment and securitiessettlement systems. Journal of Monetary Economics 55(3), 542–553.

Perlin, M., and Schanz, J. (2010). System-wide liquidity risk in the United Kingdom’s large-value payment system: an empirical analysis. Working Paper 427, Bank of England.

Raciborski, R. (2011). Graphical representation of multivariate data using Chernoff faces.Stata Journal 9(3), 374–387.

Wooldridge, J. (2010). Econometric Analysis of Cross Section and Panel Data, 2nd edn.MIT Press, Cambridge, MA.

Yermack, D. (1995). Do corporations award CEO stock options effectively? Journal ofFinancial Economics 39, 237–269.





We also offer daily, weekly and live news updates

Visit Risk.net/alerts now and select your preferences

Asset Management

Commodities

Derivatives

Regulation

Risk Management

CHOOSE YOUR PREFERENCES

Get the information you needstraight to your inbox

RNET16-AD156x234-NEWSLETTER.indd 1 21/03/2016 09:27


Research Paper

Interoperability between centralcounterparties

Jürg Mägerle and Thomas Nellen

Swiss National Bank, Bundesplatz 1, PO Box, CH-3003 Berne, Switzerland;emails: [email protected], [email protected]

(Received June 22, 2015; revised October 7, 2015; accepted November 3, 2015)

ABSTRACT

This paper analyzes the risk management and regulation of financial exposures result-ing from links between central counterparties (CCPs). Interoperability is shown toenable a fragmented clearing system to reduce exposures between CCPs and theirparticipants to the optimal benchmark of a single CCP. This comes at the cost ofnewly established exposures between CCPs. Due to concerns about the systemic riskimplications of a formerly applied cross-CCP risk management model of Europeancash equity CCPs, responsible regulators have issued guidelines to eliminate systemicrisk. These guidelines are shown to come at the cost of collateral requirements exceed-ing the optimal level. Two approaches are investigated that help to reduce collateralrequirements while containing systemic risk.

Keywords: interoperability between central counterparties; financial network; systemic risk;multilateral netting; netting efficiency.

1 INTRODUCTION

This paper analyzes the risk management and regulation of financial exposures result-ing from links between central counterparties (CCPs). Such links are referred to as

Corresponding author: T. Nellen Print ISSN 2049-5404 j Online ISSN 2049-5412Copyright © 2015 Incisive Risk Information (IP) Limited

49




50 J. Mägerle and T. Nellen

interoperability arrangements. These arrangements enable participants to clear posi-tions in any linked CCP without needing to maintain multiple CCP memberships.On the one hand, this allows traders to multilaterally net trade positions from diversesources as if dealing with a single CCP. Insofar, interoperability allows us to reduce afragmented clearing system’s overall financial exposure and its cost of collateraliza-tion to the optimal level achievable with a single CCP. On the other hand, interoper-ability creates financial exposures among CCPs that could give rise to systemic riskif not appropriately collateralized. Thus, there is a trade-off between the efficiencygains associated with multilateral netting and the cost of collateral to secure exposuresamong CCPs.

Building on a simple model that allows us to study this trade-off, we first analyzea formerly applied cross-CCP risk management model, which draws on an exist-ing interoperability framework operating in Europe.1 Interoperating CCPs applied arehypothecation model to cover their cross-CCP margin, ie, the CCPs pledged collat-eral collected as regular margin contributions from their members at interoperatingCCPs.2 This model raised concerns among responsible European authorities that itmight be prone to insufficient collateralization.3 More specifically, regulators wereconcerned that CCPs might not be able to finance cross-CCP margins by means oftheir members’ margin contributions at all times. Indeed, if other CCPs are left insuf-ficiently collateralized when a CCP defaults, a domino effect may result. We deriveconditions for sufficient cross-CCP collateralization that validate regulators’concerns.Insufficient collateralization may be realized even in bilateral interoperability frame-works if CCPs’ margin models are too divergent. Worse, multilateral interoperabilityarrangements are generally prone to insufficient collateralization.

In order to eliminate systemic risk, regulators have released guidelines for cross-CCP risk management.4 In particular, cross-CCP exposures must be covered by

1 Interoperability has been applied in Europe for more than ten years. Around 65% of cash equitytrading in Europe has switched to clearing through the analyzed interoperability framework, whichconsists of EuroCCP, LCH.Clearnet Ltd and SIX x-clear Ltd (see Chan 2012). However, the scopeof interoperability is not limited to cash equity trading: a link between LCH.Clearnet SA and CG&Gclears bonds, and a link between LCH.Clearnet Ltd and SIX x-clear Ltd Norwegian branch clearsexchange-traded derivatives.2 As of today, CCPs exchange cross-CCP margin as if they were normal members of each other. Formore information about cross-CCP risk management, see Mägerle and Nellen (2011).3 See “LCH.Clearnet in last-minute hitch”, Financial Times, Friday, October 30, 2009, and “Clearers’linkages may introduce new risks”, Financial Times, Saturday, February 15, 2010.4 See European Commission (2009) and Mägerle and Nellen (2011) for discussions of these guide-lines. These guidelines are also reflected in the “Principles for financial market infrastructures”(PFMI) by the Committee on Payment and Market Infrastructures (CPMI) and the InternationalOrganization of Securities Commissions (IOSCO) (see CPMI/IOSCO 2012).





Interoperability between central counterparties 51

additional collateral that is funded via financial resources other than and on top oftheir members’ ordinary margin contributions. While this framework allows us toeliminate the risk of insufficient collateralization, this comes at the cost of overallmargin requirements that exceed the optimal level of a clearing system with a singleCCP. Worse, collateral requirements could exceed those of a bilateral clearing system,providing traders with an incentive to bypass CCPs.

We investigate two alternative cross-CCP risk management approaches that allow usto reduce or avoid overcollateralization without rendering CCPs prone to insufficientcollateralization. We argue that if a CCP pledges its members’ margins to anotherCCP, the characteristic of these margin contributions is transformed from a defaulter-pays into a survivors-pay instrument.5 As a consequence, the rehypothecated marginsof a defaulting CCP’s members may now be used to cover the losses of the survivingCCPs, as they will not be redeemed to the defaulting CCP’s surviving members. Thus,rehypothecation gives rise to a transformation that creates additional collateral, whichis otherwise unavailable to cover losses. The “scalable margin model” allows CCPsto reduce overcollateralization by asking members for additional collateral only ifinsufficient collateralization is realized. The establishment of a “meta-CCP”, ie, aCCP for CCPs, eliminates the risk of insufficient collateralization without requiringany additional collateral.

While recent research by Koeppl et al (2012), Biais et al (2013), Acharya andBisin (2014) and Monnet and Nellen (2014) analyzes the economics of clearingmore generally, we focus on one particular service of a CCP, namely multilateralnetting, in order to analyze cross-CCP risk management and its regulation. Insightsare derived by means of a simple model that captures the basic features of Duffie andZhu (2011) and Duffie et al (2015). While these papers illustrate the efficiency gains ofcentral clearing via a single CCP compared with a fragmented clearing system withmultiple but isolated CCPs, we show that interoperability may help to resolve theinefficiencies related to fragmentation. Closest to our paper are Anderson et al (2013)and Cox et al (2013), who analyze the efficiency gains from linking CCPs. We gobeyond this, focusing on the lessons that can be learned from the experiences gainedusing the existing interoperability frameworks of cash equity markets in Europe forcross-CCP risk management and its regulation. Recently, Cœuré (2015) emphasizedthe importance of a well-regulated framework for interoperability in terms of itscontribution to the efficiency and stability of the clearing system.

5 While “members’ margin” is a defaulter-pays instrument (the defaulting member’s margins areused to cover the losses to the CCP, while surviving members’ margins are redeemed), the defaultfund is a survivors-pay instrument (surviving members’ collateral may be used to cover the lossesof a defaulting member and, thus, may not be redeemed).






This paper is organized as follows. Section 2 presents the basic model. Section 3derives conditions for sufficient collateralization in a bilateral framework. Multilat-eral interoperability is analyzed in Section 4. Section 5 evaluates alternative riskmanagement models. Section 6 concludes and suggests areas for future research.

2 MODEL

We consider a market for one security that is traded among an arbitrary number oftraders, i D 1; : : : ; k. Denote by ti;j a sale by trader i to trader j , ie, ti;j denotes theshort position of trader i toward trader j , or the long position of trader j toward traderi .6 We further assume that ti;i � 0, saying that traders do not trade with themselves.Trader i ’s gross short position toward all other traders is defined as the sum of his shortpositions:

Pkj D1 ti;j . As a consequence, the clearing system’s total gross position isPk

iD1

Pkj D1 ti;j .

Let the net position of trader i toward trader j be ti;j � tj;i D xi;j . This denotestrader i ’s short-minus-long positions. The net positions of a pair of traders always off-set each other as xi;j C xj;i D 0. This also holds true for the entire clearing system’snet position:

kXiD1

kXj D1

.ti;j � tj;i / DkX

iD1

kXj D1

xi;j D 0: (2.1)

We define the net open position of trader i toward trader j to be the absolute value ofthe net position, ie, jti;j � tj;i j D jxi;j j. The net open positions of a pair of traders areidentical, ie, both traders have the same exposure toward each other: jxi;j j D jxj;i j.7We determine the clearing system’s net open position to be the sum of all traders’net open positions,

PkiD1

Pkj D1 jxi;j j > 0. In contrast to the clearing system’s net

position, the clearing system’s net open position does not offset. Therefore, in theevent of a defaulting trader, the net open position rather than the net position is a validindicator of the exposures in the clearing system.

2.1 Bilateral clearing

Assume that, in a bilaterally cleared market, a regulator requires traders to backnet open positions with their own capital. For this purpose, the regulatory authority

6 Because of linearity, the model applies to any arbitrary number of trades in one security and toseveral securities.7 Risk management models for cash equity CCPs often assume that the replacement cost risk ofshort and long positions is identical. Thus, short and long positions offset each other. If we consideranything other than cash equity products, the existence of a hedge instrument may be necessary toachieve linearity.






FIGURE 1 Example of a clearing system with bilateral clearing.

Trade flow

Trader

23

2

1

2

2

determines a capital coefficient of 0 < � < 1. The regulatory capital charge for traderi is �

Pkj ¤i jxi;j j. The clearing system’s own capital requirement results in

�

kXiD1

kXj D1

jxi;j j > 0: (2.2)

Figure 1 illustrates a stylized example of a bilateral clearing system. We will usethis example’s basic setup throughout the paper. A trade is indicated by an arrow.Each trade involves 100 securities at a price of US$1 with a capital charge of � D1%. The rectangles depict traders and contain the sum of their own capital charges,defined as the aggregated net open position of each trader times the capital coefficient.The rectangle from which an arrow originates denotes a trader that encounters ashort position, whereas the rectangle the arrow points to denotes a trader with thecorresponding long position. Without a central clearing facility, netting only occursbilaterally. Summing up the numbers in the rectangles results in the clearing system’stotal capital charge of US$12.






2.2 Single CCP

A single CCP requires its traders to provide margins on their net open positions. Wedefine the margin requirement coefficient to be 0 < � < 1.8 Because the CCP con-centrates counterparty risk, it typically applies multilateral netting. We denote traderi ’s multilateral net open position toward the CCP by j

Pkj D1 xi;j j. Thus, aggregated

margins collected by the CCP are equal to

�

kXiD1

ˇ̌ˇ̌

kXj D1

xi;j

ˇ̌ˇ̌: (2.3)

Lemma 2.1 The clearing system’s bilateral net open position is greater than themultilateral net open position:

kXiD1

kXj D1

jxi;j j >kX

iD1

ˇ̌ˇ̌

kXj D1

xi;j

ˇ̌ˇ̌: (2.4)

Proof This can easily be seen when applying the triangle inequality. �

Multilateral netting reduces the clearing system’s overall net open position, totalexposures and margin requirements. Inequality (2.4) states that the clearing system’snet open position under bilateral clearing is greater than the multilateral net openposition of a centrally cleared system.

Certainly, the margin requirements of a CCP are lower if � > �, which is consideredto be the case under Basel III.9 If � D �, the margin requirements of a centrallycleared system are no greater than the capital charges of a bilaterally cleared market.We take this to be the reference case. Depending on the reduction of exposures dueto multilateral clearing, margin requirements by a CCP might be lower even if � <

�. This crucially depends on the trade-flow pattern. To draw conclusions that aregenerally valid, we abstain from imposing any restrictions on the trade-flow pattern.

Figure 2 on the facing page illustrates the effect of a CCP for the same clearingsystem as represented in Figure 1 on the preceding page. The multilateral net open

8 Standard models applied by CCPs, such as the historical value-at-risk (VaR) model, result in amargin requirement of a certain percentage of the aggregate net open position of a trader. Thisjustifies the linearity of the model.9 The Basel III capital framework imposes higher capital and margin requirements for bilateral thanCCP-cleared trades. As initially agreed in 2009, the Group of Twenty’s (G-20’s) reform programcomprised an element to ensure that noncentrally cleared derivatives contracts should be subjectto higher capital requirements (see www.g20.utoronto.ca/2009/2009communique0925.html). Sim-ilarly, in 2011, the G-20 agreed to add margin requirements on noncentrally cleared derivatives tothe reform program (see www.g20civil.com/documents/Cannes_Declaration_4_November_2011.pdf). In our model, this is reflected as � > �.






FIGURE 2 Example of a clearing system with a single CCP.

Trade flow

Trader

CCP

03

2

1

0

2

position is calculated as the absolute value of the sum of all arrows pointing to atrader’s rectangle, minus the sum of all arrows pointing away. The number in a trader’srectangle now indicates the margins charged by the CCP. In our example, the clearingsystem switches from a capital charge of US$12 to a margin contribution of US$8(given � D � D 1%).

2.3 Multiple CCPs and interoperability

Interoperability enables traders to consolidate their participation in a single CCP andapply multilateral netting to a larger set of trades and traders across all interoperat-ing CCPs. Thus, interoperability allows us to reduce exposures and related marginpayments by means of multilateral netting, as if there were a single CCP.10

To illustrate this, let us assume that the clearing system consists of two CCPs clear-ing two different securities, t1 and t2. CCP1 clears t1 and CCP2 clears t2. All traders

10 The same logic essentially holds true for a trader’s default fund contributions.






participate in both CCPs. CCPs apply identical margin requirement coefficients0 < � < 1.

Let us first investigate a situation without interoperability. CCPs’ margin require-ments sum up to

�

kXiD1

ˇ̌ˇ̌

kXj D1

x1i;j

ˇ̌ˇ̌ C �

kXiD1

ˇ̌ˇ̌

kXj D1

x2i;j

ˇ̌ˇ̌: (2.5)

If traders can net their open positions multilaterally through a single CCP that appliesthe same margin coefficient �, the following overall margin requirement results:

�

kXiD1

ˇ̌ˇ̌

kXj D1

.x1i;j C x2

i;j /

ˇ̌ˇ̌: (2.6)

Applying the triangle inequality, it is easy to see that the margin requirements oftwo distinct CCPs are greater than the margin requirement of a single CCP. Althoughwe assume complete netting between the two securities, the linearity of the modelensures that multilateral netting through a single CCP increases netting efficiency,even with a lower level of netting. A merger will not increase netting efficiency if andonly if netting between different securities is excluded. Linearity also ensures thatthe results apply for any arbitrary number of markets and CCPs. Similar to Duffieand Zhu (2011); Duffie et al (2015), we conclude that a single CCP is optimal, as itreduces margin requirements and counterparty risk to the greatest possible extent.

Interoperability enables multilateral netting across traders of linked CCPs. We lettraders 1; : : : ; l be members of CCP1 and traders l C 1; : : : ; k be members of CCP2,where k D l C m. Again, margin requirements are taken to be identical. Then, CCP1collects margins equal to

�

lXiD1

ˇ̌ˇ̌

kXj D1

.x1i;j C x2

i;j /

ˇ̌ˇ̌; (2.7)

while CCP2 collects margins equal to

�

kXiDlC1

ˇ̌ˇ̌

kXj D1

.x1i;j C x2

i;j /

ˇ̌ˇ̌: (2.8)

Lemma 2.2 The multilateral net open position of a segmented clearing system isgreater than that of an interoperability framework, which is identical to that of asingle CCP.

Proof Neglecting identical margin coefficients, the triangle inequality reveals that(2.5) is greater than (2.6) (proving the first claim) and the sum of (2.7) and (2.8) equals






(2.6) (proving the second claim). Due to the linearity of the model, the propositionholds true for any arbitrary number of CCPs and securities. �

Thus, interoperability reduces collateral needs and exposures in the clearing system,as it increases the potential for multilateral netting.

3 BILATERAL INTEROPERABILITY

We start by investigating whether the margins collected by CCPs suffice to covercross-CCP exposures in a bilateral interoperability framework. We then analyze theincentive effects of the new regulatory guidelines.

Consider a clearing system consisting of one security and two CCPs with the samepartition of traders as before. The margin requirement coefficient for CCP1 is definedas 0 < � < 1 (and for CCP2 as 0 < � < 1). Suppose that CCPs apply identical riskmanagement models, � D �.

Interoperating CCPs enable members to multilaterally net their trades across allcounterparties in the interoperability framework. Thus, CCP1 requires margins fromits traders equal to

�

lXiD1

ˇ̌ˇ̌

kXj D1

xi;j

ˇ̌ˇ̌: (3.1)

The cross-CCP net position between CCPs is determined in a similar way as the netposition of a CCP’s member toward its CCP. For instance, the cross-CCP net positionof CCP1 toward CCP2 is simply the sum of net positions, xi;j , between traders ofCCP1, i D 1; : : : ; l , and traders of CCP2, j D l C 1; : : : ; k. Thus, the cross-CCP netposition of CCP1 is determined by

lXiD1

kXj DlC1

xi;j : (3.2)

We obtain the cross-CCP margin requirements by multiplying the cross-CCP net openposition by the margin requirement coefficient. The cross-CCP net open position isdefined as the absolute value of (3.2). Because the short and long positions are givenequal weight, CCPs’ cross-CCP net open positions are identical:

ˇ̌ˇ̌

lXiD1

kXj DlC1

xi;j

ˇ̌ˇ̌ D

ˇ̌ˇ̌

kXiDlC1

lXj D1

xi;j

ˇ̌ˇ̌: (3.3)

Therefore, CCP1 is exposed to a cross-CCP margin requirement by CCP2 equal to

�

ˇ̌ˇ̌

lXiD1

kXj DlC1

xi;j

ˇ̌ˇ̌: (3.4)






If margins collected by either CCP are greater than the margins that interoperatingCCPs require from each other, we say that the CCPs are sufficiently collateralized.We define an interoperability framework to be insufficiently collateralized or ratherundercollateralized if the margin collected by any CCP does not suffice to fulfill thecross-CCP margin requirements of an interoperating CCP. For instance, the conditionfor CCP1 to be sufficiently collateralized is

�

lXiD1

ˇ̌ˇ̌

kXj D1

xi;j

ˇ̌ˇ̌ > �

ˇ̌ˇ̌

lXiD1

kXj DlC1

xi;j

ˇ̌ˇ̌: (3.5)

Skipping identical margin coefficients, the right-hand side of (3.5) can be expandedaccording to (2.1): ˇ̌

ˇ̌lX

iD1

kXj DlC1

xi;j

ˇ̌ˇ̌ D

ˇ̌ˇ̌

lXiD1

kXj D1

xi;j

ˇ̌ˇ̌: (3.6)

Equation (3.6) enables us to apply the triangle inequality to (3.5) and demonstrate thatsufficient collateralization of CCP1 is ensured. The same can be derived for CCP2.

Proposition 3.1 In a bilateral interoperability framework, the multilateral netopen position of a CCP is greater than the cross-CCP net open position.

Proof See derivation above. �

Thus, in a bilateral interoperability framework, traders cannot increase cross-CCPexposures without an equivalent and simultaneous expansion of exposures towardtheir CCP.

This result implies that sufficient collateralization is ensured if CCPs apply identicalmargin requirements, � D �.

Corollary 3.2 In a bilateral interoperability framework, both CCPs are suf-ficiently collateralized, independent of the underlying trade-flow pattern, if CCPsapply identical margin requirement coefficients, � D �.

The equality of the margin requirement coefficients is a sufficient but not necessarycondition to avoid insufficient collateralization. CPMI/IOSCO (2012) just requiresthat margins, at the minimum, are able to cover future expected losses under normalmarket conditions. Thus, depending on the specification chosen by CCPs, the marginrequirements of any two CCPs can easily differ, ie, � ¤ �. However, if � ¤ �,whether or not a CCP is sufficiently collateralized depends on the difference betweenmargin coefficients chosen by CCPs and the trade-flow patterns CCPs face. The largerthe difference between coefficients, the smaller the cross-CCP exposure has to be inrelation to the margins collected by CCPs. Otherwise, insufficient collateralizationmay occur, as discussed below.






FIGURE 3 Example of a bilateral interoperability framework.

Trade flow

CCP

22

Trader

03

2

1

0

2

Figure 3 illustrates that the cross-CCP margin requirements, as a result of bilateralnetting between the two CCPs, are equal to US$2 (denoted by the italic numbers).Within the ellipses, which denote CCPs, are rectangles, which represent the tradersthat clear through these CCPs. Traders from the right-hand CCP sell 300 securities totraders from the left-hand CCP, and traders from the left-hand CCP sell 100 securitiesto traders from the right-hand CCP. Both CCPs are sufficiently collateralized, sincefor each CCP the sum of margins collected (the numbers in the trader’s rectangles) isgreater than the cross-CCP margin requirement (the italic numbers in the ellipses). Theoverall margin requirement of US$8 is identical to that of a single CCP. As predictedby the model, the cross-CCP margin requirements of both CCPs are identical for� D �.

The importance of the trade-flow patterns can be highlighted by assuming that onlyone trader participates in CCP1. If so, the margins collected by CCP1 are exactlyequal to the cross-CCP margin requirement of CCP2, because all margin require-ments of CCP1 necessarily result from trades across CCPs. Thus, CCP1 is constantly






undercollateralized if � < �. However, Figure 3 on the preceding page illustratesthat insufficient collateralization can also occur in less extreme circumstances. In thegiven example, margins collected by the right-hand CCP are exactly sufficient to meetthe cross-CCP margin requirement. Therefore, the right-hand CCP would be under-collateralized if its margin requirement coefficient was less than that of the left-handCCP.

Let us reconsider the above example to illustrate the cost implications of additionalcollateral, as required by the new guidelines. US$4 of additional collateral is requiredto cover cross-CCP exposures (the sum of the italic numbers). Thus, margin require-ments for the clearing system sum up to US$12 (the sum of the normal and italicnumbers). This is identical to the capital charges asked in bilateral clearing, as illus-trated in Figure 1 on page 53. For the chosen example, traders are indifferent towardcentral clearing and bilateral clearing. However, it can be the case that interoperabilityunder the new guidelines requires more collateral than bilateral clearing.

More formally, under the guidelines, traders prefer interoperability to bilateralclearing if the benefits of multilateral netting outweigh the cost of additional collateral.For instance, bilateral interoperability is preferred to bilateral clearing if the followingcondition holds:

�

ˇ̌ˇ̌

lXiD1

kXj DlC1

xi;j

ˇ̌ˇ̌ C �

ˇ̌ˇ̌

kXiDlC1

lXj D1

xi;j

ˇ̌ˇ̌

6 �

kXiD1

kXj D1

ˇ̌ˇ̌xi;j

ˇ̌ˇ̌ �

��

lXiD1

ˇ̌ˇ̌

kXj D1

xi;j

ˇ̌ˇ̌ C �

kXiDlC1

ˇ̌ˇ̌

kXj D1

xi;j

ˇ̌ˇ̌�

: (3.7)

This is generally true, as (3.7) can easily be extended to any multilateral interoper-ability framework. The left-hand side of (3.7) denotes the additional margins requiredto fulfill regulators’ guidelines (the sum of the cross-CCP margin requirements). Theright-hand side denotes the benefits from multilateral netting (the difference betweencapital requirements under bilateral clearing and the sum of both CCPs’ marginrequirements under bilateral interoperability). Thus, in terms of collateral savings,traders only prefer bilateral interoperability to bilateral clearing if the benefits frommultilateral netting are large enough to compensate for the additional margins requiredby regulators.11

11 For a comprehensive cost–benefit analysis, savings on default fund contributions would have tobe taken into account.






FIGURE 4 Example of a multilateral interoperability framework.

Trade flow

CCP

Trader

0 3

11

0 3

03

2

1

0

2

4 MULTILATERAL INTEROPERABILITY

Unlike in a bilateral framework, traders in a multilateral framework may accumulateoffsetting net open positions toward their own CCPs, while their CCP’s cross-CCP netopen position does not offset. For instance, CCP1’s traders can offset a long positiontoward CCP2 by a short position toward CCP3. As a result, CCP1 is undercollateral-ized, because CCP1’s traders are flat toward their CCP, while CCP1 has an exposuretoward CCP2 and CCP3. As a consequence, interoperability with three or more CCPsis generally prone to insufficient collateralization, even if the CCPs apply identicalmargin coefficients.

Returning to our example and expanding interoperability from a bilateral to amultilateral framework with three CCPs results in cross-CCP margin requirementsthat violate the condition for sufficient collateralization, as illustrated in Figure 4.Again, the net open position of a trader can be calculated as the absolute value of thenumber of arrows pointing toward a rectangle minus the number of arrows pointingaway. The resulting margin requirements for traders are displayed as numbers inthe respective rectangles. The cross-CCP margin requirement based on the net open






position between each pair of CCPs is calculated on the basis of bilateral netting(indicated as italic numbers in the ellipses closest to the respective interoperatingCCP’s ellipse). Summing the italic numbers of a CCP results in the total cross-CCPmargin requirement that a CCP must be able to cover in order to be sufficientlycollateralized.

Therefore, a CCP is sufficiently collateralized if the sum of traders’ margin con-tributions (ie, the sum of the numbers within an ellipse’s rectangles) is at least equalto the sum of cross-CCP margin requirements (the italic numbers within an ellipse).In Figure 4 on the preceding page, all but the right-hand CCP are sufficiently col-lateralized. The trader at the bottom of the right-hand CCP ellipse illustrates thatin a multilateral framework a trader may accumulate offsetting net open positionstoward its own CCP, while its CCP’s cross-CCP net open position does not offset. Inour example, this trader’s margin contribution is 0, while his trade pattern imposes apositive cross-CCP margin requirement of US$2 on their CCP.

Thus, in a multilateral framework, where CCPs rehypothecate margins and applyidentical margin requirement coefficients, trade-flow patterns may result in undercol-lateralized CCPs if traders’ cross-CCP exposures toward a CCP are offset by cross-CCP exposures toward another CCP. Depending on the trade-flow pattern, one, twoor all involved CCPs may end up undercollateralized. This result can be generalizedto any multilateral framework with more than three CCPs.

Proposition 4.1 Multilateral interoperability involving more than two CCPs isprone to insufficient collateralization, even if the CCPs apply identical margincoefficients.

To see what factors foster the occurrence of insufficient collateralization, let usconsider the following thought experiment. We steadily increase the number of inter-operable CCPs in the clearing system, while all else remains equal. At the limit, k

traders are distributed over k CCPs (with each serving a single trader). This impliesthat all trades are necessarily cross-CCP trades. As CCPs clear bilaterally among eachother, they face the same cross-CCP margin requirements as traders under bilateralclearing. More formally, if the number of linked CCPs converges to the number oftraders, the sum of cross-CCP exposures converges to the sum of the bilateral net posi-tions of a bilateral clearing system. Thus, under a given trade pattern, the maximumof aggregated cross-CCP margin requirements is reached when each trader is servedby a separate CCP. While the sum of cross-CCP margins increases with the numberof CCPs, the total margins the CCPs collect from their members remain constant, asthey are based on multilateral netting across all traders in the interoperability frame-work. Thus, the likelihood that insufficient collateralization occurs increases with thenumber of CCPs.






Considering the example on multilateral interoperability in Figure 4 on page 61,regulators’guidelines result in US$8 of additional margins to secure cross-CCP expo-sures (the sum of the italic numbers). Thus, the total margin requirements of the entireclearing system sum up to US$16 (the sum of the normal and italic numbers). In com-parison with multilateral interoperability with three CCPs, traders save US$4 undera bilateral clearing system and US$8 under a single CCP. Thus, for the given tradeflow, netting benefits do not compensate for the additional margins required, as sug-gested in Proposition 4.1 and illustrated in (3.7). In this particular case, even bilateralclearing is preferred to multilateral interoperability.

For a single CCP, the left-hand side of (3.7) equals 0 by definition. Under the newguidelines, and assuming multilateral netting is beneficial (ie, the right-hand side of(3.7) is greater than 0), total margin requirements are minimized with a single CCP.Relative to the optimal level with a single CCP, the new guidelines result in excessivecollateralization that can even go beyond the level of a bilateral clearing system.

5 OTHER REHYPOTHECATION MODELS

Interoperating CCPs find themselves in a paradoxical situation. Even though the sumof margins collected by CCPs equals the margins collected by a single CCP, marginsmay not suffice to collateralize cross-CCP exposures. It is evident that insufficientcollateralization cannot be circumvented by simply raising margin requirement coef-ficients. Even though more margins would be collected from traders, proportionatelymore collateral has to be delivered to the other CCPs. Therefore, the probability ofinsufficient collateralization remains the same.

Regulators’ guidelines to avoid insufficient collateralization are straightforward.CCPs are expected to collect additional, specifically dedicated margins to cover cross-CCP exposures. Further, CCPs are not allowed to use existing margins to cover cross-CCP exposures. As a consequence, more collateral is needed to comply with themargin requirements, and the clearing system ends up overcollateralized in compari-son with a clearing system with a single CCP. Thus, it may be worthwhile to consideralternative regulatory regimes.

We argue that members’ rehypothecated margins represent additional collateralbecause of their transformation from a defaulter-pays instrument to a survivors-payinstrument. As a consequence, we ignore the guideline of collecting additional col-lateral on top of existing collateral by allowing the rehypothecation of members’margins. Thus, we evaluate whether less collateral-intensive cross-CCP risk manage-ment models can be designed so that they comply with the remaining guidelines, ie,fully collateralized with prefunded collateral. Under this premise, the following twosubsections consider a scalable margin model and a meta-CCP approach.






5.1 Scalable margin model

To reduce excessive collateralization in a multilateral interoperability framework,CCPs could require additional margins if and only if the sum of regular marginsis insufficient to cover cross-CCP exposures. For this purpose, we derive a marginallocation rule that ensures sufficient collateralization for any arbitrary number ofinteroperating CCPs by closing the collateral gap between the margins collected andcross-CCP margin requirements.

For the sake of simplicity, we analyze the above-defined multilateral interoperabil-ity case with h D 3 CCPs and their respective margin coefficients �, � and �. Let usdefine

lXiD1

Ci D �

lXiD1

ˇ̌ˇ̌

kXj D1

xi;j

ˇ̌ˇ̌

as the sum of collected margins from CCP1’s members i D 1; : : : ; l . Further, X1h

represents CCP1’s cross-CCP margin requirement by CCP h, and X1 DP

h¤1 X1h

is the sum of cross-CCP margin calls by CCP1’s interoperating CCP2 and CCP3.Thus, CCP1 must meet the following condition to avoid insufficient collateralization:Pl

iD1 Ci > X1. If this inequality does not hold, CCP1 calls for additional marginsfrom its members by applying the following allocation rule:

Ai D�

X1 �lX

iD1

Ci

�CiPl

iD1 Ci

C Ci : (5.1)

Ai denotes the adjusted margin requirement that trader i has to meet in orderfor CCP1 to avoid insufficient collateralization. The first term of the right-hand siderepresents the gap between the cross-CCP margins required and the margins collectedfrom members. This gap is multiplied by an allocation rule, which is simply trader i ’sshare in the sum of collected margins. This could either be a real-time measure or anaverage share over a predetermined period of the past. Because some members mightregularly be flat, a CCP can alternatively relate the allocation rule to the traders’ sharein the gross position.12 The last term is the regular margin requirement that a traderhas to meet. It is easy to see that summing the adjusted margin requirements over alltraders of CCP1 results in the sum of cross-CCP margin calls for CCP1:

lXiD1

Ai D X1 D �

ˇ̌ˇ̌

lXiD1

mXj DlC1

xi;j

ˇ̌ˇ̌ C �

ˇ̌ˇ̌

lXiDl

kXj DlCmC1

xi;j

ˇ̌ˇ̌: (5.2)

To avoid insufficient collateralization, we determine the effective margin requirementfor trader i as follows:

Mi D maxfCi ; Aig: (5.3)

12 For instance, algorithmic traders usually end the trading day flat.






In comparison with regulators’ guidelines, collateral requirements can be reduced,since additional collateral is collected only to close the gap between margins collectedand cross-CCP margin requirements. As can easily be seen from the example inFigure 4 on page 61, the insufficiently collateralized CCP on the right-hand sidewould have to collect US$2 of additional margins. Therefore, the clearing systemrequires US$6 less than the new risk management model imposed by regulators’ newguidelines and US$2 less than bilateral clearing. Thus, the scalable margin model ismore costly than a single CCP approach, but it is able to restore incentives to clearthrough interoperating CCPs rather than via a bilateral clearing framework. This isgenerally true and can be formalized by comparing the clearing system’s aggregatedmargin requirement,

PkiD1 Mi , with the capital requirements of a bilateral clearing

system.

Proposition 5.1 The scalable margin model’s aggregated margin requirementsare lower than

(i) the aggregated capital requirements of a bilateral clearing system if � > �; �,

(ii) the aggregated margin requirement of an interoperability framework complyingwith regulators’ guidelines.

Proof (i) Consider the most favorable case of a capital coefficient equal to themargin coefficients of any number of interoperating CCPs. In case of bilateral inter-operability, apply Lemma 2.1. In case of multilateral interoperability and sufficientcollateralization, apply Lemma 2.2. In case of multilateral interoperability and insuf-ficient collateralization, consider (5.4). On the left-hand side, CCP1’s cross-CCPmargin requirements are depicted. The right-hand side denotes the aggregate capi-tal requirements of CCP1’s traders under bilateral clearing. The triangle inequalityconfirms that inequality (5.4) holds true:

�

ˇ̌ˇ̌

lXiD1

mXj DlC1

xi;j

ˇ̌ˇ̌ C �

ˇ̌ˇ̌

lXiDl

kXj DlCmC1

xi;j

ˇ̌ˇ̌ 6 �

lXiD1

kXj D1

jxi;j j: (5.4)

The same conditions can easily be derived for the other two CCPs.

(ii) This is straightforward, as existing members’ margins can be used to covercross-CCP margins. �

Ultimately, the entire clearing system is better off when moving from bilateralclearing, or, rather, an interoperability framework that complies with regulators’ newguidelines, to an interoperability framework that applies a scalable margin model.13

13 Depending on the trade-flow pattern that determines the multilateral netting effect, the inequalitycan hold true for lower values of �.






The advantage of this model is that it is readily applicable without the need forany harmonization of risk management models. However, there might be room forimprovement, as the scalable margin model is less efficient than a single CCP.

Proposition 5.2 If interoperating CCPs apply margin coefficients greater thanthat of a single CCP, a single CCP’s margin requirement is less than that of aninteroperability framework that applies the scalable margin model.

Proof Under bilateral interoperability, Lemma 2.1 proves equality. Under multi-lateral interoperability, Lemma 2.2 proves equality in the case of sufficient collat-eralization. Given insufficient collateralization, the CCPs have to call for additionalmargins, which establishes the proposed inequality. �

5.2 Meta-CCP

Insufficient collateralization in a multilateral framework occurs because cross-CCPexposures are not linked to the exposures between the CCP and its members, as inbilateral interoperability. A way to mitigate the risk of insufficient collateralizationis to reestablish this linkage. For this purpose, we introduce the concept of a meta-CCP, ie, a CCP for CCPs. The intuition is simple: a meta-CCP reduces multilateralrelationships between CCPs into bilateral relationships between CCPs and the meta-CCP (essentially, this is what a CCP does for its members).

Without loss of generality, we focus our attention on the multilateral interoperabilityframework with three CCPs. We derive the results for CCP1, assuming all CCPsand the meta-CCP apply identical margin requirement coefficients (omitted). Thus,exposures directly translate into margins:

C1 DlX

iD1

ˇ̌ˇ̌

kXj D1

xi;j

ˇ̌ˇ̌: (5.5)

The meta-CCP does not have to account for bilateral exposures between CCPsbecause it can multilaterally net cross-CCP exposures. The multilateral net openposition of a CCP toward the meta-CCP is defined as the absolute value of the sumof total multilateral net open positions that traders of a CCP have toward traders ofall interoperating CCPs. For instance, CCP1 has the following exposure toward themeta-CCP:

M1 Dˇ̌ˇ̌

lXiD1

kXj DlC1

xi;j

ˇ̌ˇ̌: (5.6)

Multilateral netting by a meta-CCP transforms cross-CCP margin requirements intothe requirement of a merged CCP2 and CCP3, as denoted in (5.6).






Proposition 5.3 If �meta CCP 6 min.�CCP1; : : : ; �CCPN /, then interoperatingCCPs are sufficiently collateralized to meet the meta-CCP’s margin requirements.

Proof Without loss of generality, we assume all margin coefficients are equal. Toinvestigate whether CCPs are sufficiently collateralized under a meta-CCP frame-work, we derive the condition for sufficient collateralization by combining (5.5) and(5.6) to obtain

C1 DlX

iD1

ˇ̌ˇ̌

kXj D1

xi;j

ˇ̌ˇ̌ >

ˇ̌ˇ̌

lXiD1

kXj DlC1

xi;j

ˇ̌ˇ̌ D M1: (5.7)

By applying (2.1), the right-hand side of (5.7) can be extended as follows:

ˇ̌ˇ̌

lXiD1

kXj DlC1

xi;j

ˇ̌ˇ̌ D

ˇ̌ˇ̌

lXiD1

kXj D1

xi;j

ˇ̌ˇ̌:

The triangle inequality confirms that sufficient collateralization is ensured. It isstraightforward to show this for all other CCPs. Linearity ensures that this holdstrue for any nonnegative integer N of interoperating CCPs. �

The establishment of a meta-CCP enables CCPs to avoid insufficient collateraliza-tion in a multilateral framework without requiring any additional margins. A meta-CCP reintroduces the bilateral relationship that links cross-CCP margins to members’margins, because it allows for multilateral netting between CCPs. In other words, themeta-CCP’s margin requirement toward a CCP is less than the margins collected bythe respective CCP (ie, multilateral net open positions between the meta-CCP and aCCP are less than the regular margins collected by the CCP).

Proposition 5.4 An interoperability framework that clears through a meta-CCPrequires the same amount of margins as a clearing system with a single CCP, if thesingle CCP’s and interoperating CCPs’ margin requirement coefficients are greaterthan that of the meta-CCP.

Proof Because Proposition 5.3 applies, no additional margins are required to avoidinsufficient collateralization. Whether a clearing system is served by a single CCP oran interoperability framework, regular margin requirements remain the same for eachtrader. �

Figure 5 on the next page illustrates a clearing system with a multilateral inter-operability framework secured by a meta-CCP (represented by the triangle). CCPsnow cover their cross-CCP exposures via the meta-CCP. This allows them to net theircross-CCP positions on a multilateral basis. The bold numbers in the triangle denote






FIGURE 5 Example of a multilateral interoperability framework with a meta-CCP.

0 3

11

1 2

3

Trade flow

CCP

MetaCCP

Trader

0 3

03

2

1

0

2

the margin requirements of the meta-CCP toward the CCPs. Sufficient collateraliza-tion is ensured if the sum of margins a CCP requires from its members (the sum ofnumbers in the rectangles of a CCP) is greater than the bold number closest to therespective ellipse. As predicted by the model, all CCPs collect sufficient margins fromtheir members to cover the meta-CCP’s margin requirement. In particular, while theright-hand CCP still collects US$2 from its members, it also has to deliver exactlyUS$2 as cross-CCP margin to the meta-CCP (instead of US$4 without a meta-CCP,ie, the sum of the italic numbers).

6 CONCLUSIONS

This paper draws attention to the key roles that interoperability and its regulationmay play in the establishment of a safe and efficient clearing system. Interoperabilitymay allow a fragmented clearing system to increase the scope of multilateral netting.However, linking the clearing system comes at the cost of additional risks, as financialexposures are created between CCPs. This creates a trade-off between multilateralnetting benefits and additional costly collateral.






We study formerly and currently applied cross-CCP risk management models toshed light on this trade-off. Regulators’ systemic risk concerns are found to be valid.Insufficient collateralization may be realized if members’ margins are the only finan-cial resources available to cover cross-CCP exposures. We further demonstrate thatregulators’guidelines to cover cross-CCP exposures by means of additional collateralon top of existing financial resources eliminate the risk of insufficient collateraliza-tion. However, this comes at the cost of an overcollateralized clearing system. Weinvestigate two alternative cross-CCP risk management approaches that reduce oravoid excessive collateralization without rendering CCPs prone to insufficient col-lateralization. Whereas the scalable margin model reduces overcollateralization, theestablishment of a meta-CCP, ie, a CCP for CCPs, eliminates insufficient collateraliza-tion without requiring any additional financial resources. Thus, regulators’ guidelinesincrease the cost of collateralization without necessarily increasing the safety of theclearing system.

While interoperability’s outreach is currently limited to Europe and exchange-traded financial instruments, it is of a wider and global relevance (Cœuré 2015). Asa reaction to the financial crisis, CCP clearing has gained momentum, driven by bothmarket demand and the G-20 commitment to mandatory clearing of over-the-counter(OTC) derivatives. While Sidanius and Zikes (2012) provide a compilation of costestimates for the mandatory clearing of OTC derivatives, Duffie et al (2015) pointout that the chosen clearing structure influences collateral demand. Increasing thescale and scope of multilateral netting is key to reducing exposures and collateralrequirements. However, Duffie et al (2010) assume that the market structure of theOTC derivatives clearing system will remain fragmented. A similar conclusion isdrawn by a report of the Committee on the Global Financial System (2011). Oneoption for reducing the inefficiencies associated with fragmentation is interoperability.

Against this background, further research on cross-CCP risk management and reg-ulation may help to improve existing interoperability guidelines. While we focus onthe reduction of cross-CCP margins using members’ existing margins, other routesmay be promising too. In particular, future research may also consider measures toreduce cross-CCP exposures. CPMI/IOSCO (2012) has substantially raised the barin terms of CCP resilience. In addition, CCPs are required to establish recovery plansin order to continue clearing in times of stress (CPMI/IOSCO 2014). Further, underthe umbrella of the Financial Stability Board, authorities are meant to work out res-olution regimes that allow for the continuation of clearing when recovery plans failand a CCP defaults (Financial Stability Board 2014). How do these measures affectexposures among interoperable CCPs, and how should these measures be structuredto reduce cross-CCP exposures? The answers to these questions would provide a basisfor increasing the safety and efficiency of the clearing system.







The views expressed in this paper are those of the authors and do not necessarilyrepresent those of the Swiss National Bank.

ACKNOWLEDGEMENTS

We are indebted to Ernst Baltensperger, Marco Cecchini, Darrell Duffie, PhilippHaene, Béatrice Kraus, Antoine Martin, Cyril Monnet, Klaus Neusser, RobertOleschak, Andy Sturm, and participants of the Brownbag Seminar at the Univer-sity of Berne, the GFRI/FINRISK/SFI Conference on Financial Networks, the Bankfor International Settlements Research Seminar, the Economics of Payments V Con-ference (hosted by the Federal Reserve Board and the Bank of Canada) and theResearch Seminar at the Bank of Canada, as well as an anonymous referee, for helpfulcomments.

REFERENCES

Acharya, V., and Bisin, A. (2014). Counterparty risk externality: centralized versus over-the-counter markets. Journal of Economic Theory 149, 153–182.

Anderson, S., Dion, J. P., and Perez-Saiz, H. (2013). To link or not to link? Netting andexposures between central counterparties. Journal of Financial Market Infrastructure1(4), 3–29.

Biais, B., Heider, F., and Hoerova, M. (2013). Clearing, counterparty risk and aggregaterisk. IMF Economic Review 60(2), 193–222.

Chan, D. (2012). CCP clearing: from a CCP perspective. Presentation, Nordic Back OfficeConference, May 30, 2012.

Cœuré, B. (2015). The international regulatory agenda on CCP links. Speech, EuropeanSystemic Risk Board Workshop on CCP Interoperability Arrangements, November 2.URL: www.ecb.europa.eu/press/key/date/2015/html/sp151102.en.html.

Committee on Payment and Market Infrastructure and Technical Committee of the Inter-national Organization of Securities Commissions (CPMI/IOSCO) (2012). Principles forfinancial market infrastructures. Report, Bank for International Settlements.

Committee on Payment and Market Infrastructure and Technical Committee of the Inter-national Organization of Securities Commissions (CPMI/IOSCO) (2014). Recovery offinancial market infrastructures. Final Report, Bank for International Settlements.

Committee on the Global Financial System (2011). Macro-financial implications of alter-native access configurations to central counterparties in OTC derivatives markets.Report 46, November, Bank for International Settlements.

Cox, N., Garvin, N., and Kelly, G. (2013). Central counterparty links and clearing systemexposures. Discussion Paper RDP2013-12, Reserve Bank of Australia.

Duffie, D., and Zhu, H. (2011). Does a central clearing counterparty reduce counterpartyrisk? Review of Asset Pricing Studies 1(1), 74–95.






Duffie, D., Li, A., and Lubke, T. (2010). Policy perspectives on OTC derivatives marketinfrastructure. Staff Report 424, Federal Reserve Bank of New York.

Duffie, D., Scheicher, M., and Vuillemey, G. (2015). Central clearing and collateral demand.Journal of Financial Economics 116, 237–256.

European Commission, Internal Market and Services DG, Financial Services Policyand Financial Markets, Financial Markets Infrastructure (2009). The code of conducton clearing and settlement: three years of experience. Commission Staff WorkingDocument.

Financial Stability Board (2014). Key attributes of effective resolution regimes for financialinstitutions. Report, Financial Stability Board. URL: www.fsb.org.

Koeppl, T., Monnet, C., and Temzelides, T. (2012). Optimal clearing arrangements forfinancial trades. Journal of Financial Economics 103, 189–203.

Mägerle, J., and Nellen, T. (2011). Interoperability between central counterparties.WorkingPaper 2011-12, Swiss National Bank.

Monnet, C., and Nellen, T. (2014).The collateral costs of clearing.Working Paper 2014-04,Swiss National Bank.

Sidanius, C., and Zikes, F. (2012). OTC derivatives reform and collateral demand impact.Financial Stability Paper 18, Bank of England.






Forum Paper

Central counterparties in crisis: theHong Kong Futures Exchange inthe crash of 1987

Robert T. Cox

Federal Reserve Bank of Chicago, 230 South LaSalle Street, Chicago, IL 60604, USA;email: [email protected]

(Received August 24, 2015; revised October 23, 2015; accepted November 3, 2015)

ABSTRACT

The Hong Kong Futures Exchange faced a major clearing house crisis during theglobal equity market crash of 1987. During a four-day market closure initiated by thestock exchange, clearing member performance failures were of sufficient magnitudeto overwhelm the solvency of the guarantor of the clearing house’s trades and causea dramatic loss in confidence. The crisis triggered a resolution effort that was ledby the Hong Kong government. This is the only known example of the successfulresolution of a central counterparty (CCP). This paper examines the development ofthe interlinked equities and futures markets in Hong Kong, the market infrastructuralweaknesses that contributed to the crisis, the challenge that the crisis presented tothe government’s prevailing market dogma, the influence of special interests and theexecution of the resolution itself. The findings of the Davison Report, commissionedafter the crisis, and the actions taken to implement it are reviewed. This episode givesus insight into the contemporary problem of CCP resolutions and the choices andchallenges that CCP stress present to resolution authorities.

Keywords: Hong Kong Futures Exchange (HKFE); 1987 crash; default management; centralcounterparty (CCP) resolution; International Commodities Clearing House Ltd (ICCH).

Print ISSN 2049-5404 j Online ISSN 2049-5412Copyright © 2015 Incisive Risk Information (IP) Limited

73




74 R. T. Cox

1 INTRODUCTION

To complete a thing, a hundred years is not sufficient; to destroy it, a day is morethan enough.

Old Chinese proverb

The global equity crash of 1987 hit Hong Kong hard. On October 19 (Black Mon-day), the benchmark Hang Seng Index (HSI) fell by over 10%. The stock exchangesuspended trading for the rest of the week, and trading in HSI futures also came toa halt. When the market reopened on the following Monday, HSI futures fell 44%in one day. These market movements threatened the solvency of many equity marketparticipants. Their liquidity was challenged too, partly by the febrile environment andpartly by delays in settling securities. Many clearing members failed to perform atthe futures exchange, and as a result the central counterparty (CCP) was threatened.The Hong Kong government was forced to intervene, arranging and in part financinga rescue of the clearing house. This is the only known instance of a resolution actionhaving been applied to a CCP during a crisis. It is, moreover, a successful example:the clearing house was rehabilitated once sufficient resources were deployed.

As the Bank of England’s deputy governor Paul Tucker remarked in 2011,

There was a risk that a modestly “recapitalised” clearing house would go broke againif positions had to be marked down immediately [following further market falls].In the event, the Hong Kong government and the clearing banks underpinned theclearing house. This episode warrants more study than it has received. Had it beenLondon, Chicago or New York, it would have entered the folklore of policy memory.

The resolution was neither defined nor well understood from the outset. It wentagainst the grain of the government’s guiding market dogma. Its execution was clumsyand left many market participants in the cold. Some parties were well served; others,less so. It attracted political fallout. The trading participants of the stock market, ratherthan those of the futures market, ultimately paid for it. The government that had donetoo little before the crisis in turn did too much after it.

For all of these reasons, the study of this crisis, and its redress, offers an instructivetale about CCP risk management as well as insight into the choices that must be madewhen the financial resources of a CCP have been exhausted.

2 BACKGROUND

We begin by introducing some of the entities and individuals who played importantroles in the crisis.

2.1 Stock exchange of Hong Kong

Until the 1970s, Hong Kong had a fairly immature capital market. Most financingwas done through the colony’s banks. The Hong Kong Stock Exchange grew out





CCPs in crisis 75

of an association of stockbrokers that was established in 1891, with the exchangeacquiring its current name in 1914. By the 1960s, the Hong Kong Stock Exchangewas a relatively small market run in a clubby manner, typical of commonwealth stockmarkets of the era.

Prominent Hong Kong businessman Ronald Li Fook-shiu started a new market, theFar East Stock Exchange, in 1969. This proved to be a boon for the listing of smallercompanies, and its turnover soon eclipsed that of the Hong Kong Stock Exchange.Membership of the new market was cheap (HK$500 000), and its small board lotsappealed to the retail investors in the Chinese community. Its success led to the launchof two additional new exchanges: the Kam Ngan Stock Exchange in 1971 and theKowloon Stock Exchange in 1972 (Securities Review Committee 1988, Appendixes,p. 372).

The Stock Exchange Control Ordinance was enacted in early 1973 to halt the frag-mentation of the exchange landscape. A Securities Ordinance and the establishmentof the Securities Commission and the Commissioner for Securities followed a yearlater, establishing a regulatory framework.

Attempts began to unify the four markets in 1976, culminating in the opening ofthe Unified Exchange (Stock Exchange of Hong Kong) in 1986. At this stage, HongKong had a capital market that could compete with the local bank finance.

2.2 Ronald Li Fook-shiu

Ronald Li was a scion of the Li family, one of Hong Kong’s most prominent families.His father was the founder of the Bank of East Asia; his nephew David Li was thechairman of that bank and a member of the government’s Legislative Council at thetime of the crisis; his brother Simon was a noted jurist and acting Chief Justice of theSupreme Court of Hong Kong; and his second cousin Andrew Li would serve in theCourt of Final Appeal as Chief Justice from 1997 to 2010.

Elected chairman of the Stock Exchange of Hong Kong upon its creation in 1986,Ronald Li was considered the godfather of the stock market. He controlled the pow-erful Listing Committee and his push for a listing of Club Volvo, a 70 000-square-foothostess lounge in Kowloon (of which he was a part owner), prompted concern thatinternational institutional investors “may be disturbed by the head of the Hong KongStock Exchange in effect promoting a girlie bar” (Kristof 1987).

2.3 Hong Kong Futures Exchange

The government enacted the Commodities Trading Ordinance in 1976, enabling theestablishment of the Hong Kong Commodities Exchange (HKCE). This began tradingin 1977 with cotton and sugar; soybeans were added in 1979, and gold in 1980. Trading





76 R. T. Cox

volumes were light, and a review completed in 1983 found that the exchange had notbeen a success.

The HKCE was reauthorized for three years with the conditions that the member-ship, board and executive staff were to be widened, and that new contracts with largerlocal appeal and international characteristics were to be developed. A banking con-sortium had concurrently proposed the establishment of a financial futures exchange.The government did not want two exchanges, so both ideas were taken on: in 1985,the HKCE was renamed the Hong Kong Futures Exchange (HKFE), and a proposedstock index futures contract was approved as part of the exchange’s license renewal(Securities Review Committee 1988, Appendixes, p. 401). Dr. Kim Cham Yau-sum,an economist, was named chairman, and Ronald Li was named vice chairman.

Trading in HSI futures commenced in May 1986.

2.4 International Commodities Clearing House Hong Kong Ltd

The International Commodities Clearing House Ltd (ICCH) of London was a leadingglobal CCP, acting as a clearing house for numerous derivatives exchanges in London,including the London International Financial Exchange (LIFFE) as well as the SydneyFutures Exchange and the New Zealand Futures and Options Exchange in Auckland.1

The International Commodities Clearing House Hong Kong Ltd (ICCH (HK)) wasits local operating entity. In a departure from the typical modern clearing housedesign, ICCH (HK) was responsible for trade processing and some risk management,including initial margin setting, but it did not guarantee trades as principal.

2.5 Futures Guarantee Corporation

Instead, a separate entity, the Futures Guarantee Corporation (FGC), was the guar-antor of HKFE trades. This body was owned primarily by a consortium of local andinternational banks that had been traditionally involved with commodity trade financeand thus had been involved from the start of the HKCE. The FGC’s owners were ICCH(20%), the Hongkong and Shanghai Bank (20%), Standard Chartered via CharteredCapital (15%), Chase Manhattan (15%), Wing On (Hongkong and Shanghai owned;10%), Barclays (10%) and Credit Lyonnais (10%).

FGC’s paid-up capital was HK$15 million, and its net worth at the time of thecrisis, after including retained earnings, was HK$22.5 million.

FGC had neither staff nor premises, its management was contracted to ICCH (HK)and it shared a common chairman. There was no clearing or default fund involving

1 ICCH was owned by a consortium of UK clearing banks: Barclays (20%), Lloyds (20%), NatWest(20%), Midland (20%), Royal Bank of Scotland (10%) and Standard Chartered (10%).





CCPs in crisis 77

FIGURE 1 HKFE trade flow.

ICCH (HK)HKFE FGC

MemberA

Client X(buy)

Client Y(sell)

MemberB

Price discoveryTrade registration

margin calls Trade guarantee

members or outside parties, which, again, is in contrast to modern clearing housedesign.

2.6 Robert Ng Chee Siong

Robert Ng was and remains the chairman of Hong Kong property developer SinoGroup. He is the son of Singaporean billionaire property magnate Ng Teng Fong.

2.7 Piers Jacobs

Piers Jacobs was the Financial Secretary of the Hong Kong government until1991. He was the torchbearer of his predecessors’ economic policy of “positivenon-interventionism”.

2.8 David Nendick

David Nendick was the Secretary for Monetary Affairs and, as a result, Hong Kong’schief banking regulator.

2.9 The market regulators

Hong Kong’s market regulatory framework was lightweight in 1987. There was aSecurities Commission and a Commodities Trading Commission, but they shared acommon chairman. The positions of all of the members of the commission and the





78 R. T. Cox

BOX 1 The dogma of positive non-interventionism.

In 1980, Sir Philip Haddon-Cave said:

I have frequently described the government’s economic policy stance asbeing one of positive non-interventionism. Some have claimed that this isjust a fancy term for laissez-faire. Others have equated it to a “do-nothing”approach. This is simply not so: positive non-interventionism involves takingthe view that it is normally futile and damaging to the growth rate of an econ-omy, particularly an open economy, for the government to attempt to planthe allocation of resources available to the private sector and to frustrate theoperation of market forces.... I do qualify the term “non-interventionism” withthe adjective “positive”.... What it means is this: that the government, whenfaced with an interventionist proposal, does not simply respond that such aproposal must, by definition, be incorrect … the government weighs up care-fully the arguments for and against an act of intervention … [and] comes to apositive decision as to where the balance of advantages lies, [even though] itis true that, more often than not, we come to the conclusion that the balanceof advantage lies in not intervening.

Tsang (2006)

Sir Philip was Financial Secretary from 1971 to 1981.

chairman were part time. An executive commissioner, who had the small office ofthe Securities Commissioner, which was staffed by civil servants, supported boththe commissions. Understaffed and underfunded, the market regulators were largelyineffective, despite their powers. Repeated pleas for additional resources were deniedby the positive non-interventionist government (Securities Review Committee 1988,Davison Report (DR) 9.23, 1.7).

3 RUN-UP TO THE CRISIS

The timing of the commencement of HSI futures trading could not have been moreauspicious. The Hong Kong market was set to boom.

3.1 The running of the bull

Hong Kong benefitted from the certainty gained from pegging its currency to theUS dollar and the Sino-British Joint Declaration on the future of Hong Kong. Rapiddevelopment in the coastal zones of eastern China under the economic liberalizationsof Deng Xiaoping, a global secular bull market in equities, and the rapidly increasingwealth of Hong Kong all combined to make the Hong Kong equity market especiallyattractive in the mid-1980s.





CCPs in crisis 79

Volume in the HSI futures contract was a major beneficiary of this trend, risingfrom 31 000 in May 1986 to over 600 000 contracts being traded during the month ofSeptember 1987 (Securities Review Committee 1988, Appendixes, p. 407). Indeed,the volume figures eclipsed every equity index future globally except Standard &Poor’s 500 (S&P 500) futures, although the well-trumpeted comparison was some-what deceiving: at HK$50 times the HSI, the contract was about an eighth of the sizeof its S&P rival (Securities Review Committee 1988, Appendixes, p. 407).

The small contract size of the HSI future was, however, a boon to retail investors.Those looking to play the market could do so in small amounts and cheaply by usingfutures. Many brokers would lend margin money, and the client base extended acrossmuch of society to include everyone from taxi drivers to domestic helpers (SecuritiesReview Committee 1988, Appendixes, p. 411). Rumors of a possible hostile takeoverof Hongkong Land in the fall of 1987 added more froth to an already overheatedmarket.

3.2 Strain on the plumbing

The high volume of the mid-1980s bull run took its toll on the stock market’s infra-structure. Settlement for share trading was done by the physical exchange of checksfor bearer negotiable share certificates, as there was no central securities depositoryin that era. Seller members presented physical certificates at the back offices of thebuyer members on t C 1, based on the information contained on the trade tickets. ByJune 1987, the backlog of unsettled transactions was acute, and some firms had “backoffice” staff working around the clock in a vain effort to keep up. Many firms began“honoring” each other, accepting promises to deliver later, an uncollateralized expo-sure, in the absence of deliverable scrip, as the chain of “fails” lengthened (personalcommunication). An already serious problem was exacerbated in July when the Com-missioner of Inland Revenue advised the stock exchange members that stock lendingand borrowing lay within the scope of the Stamp Duty Ordinance. Enforcement of a0.3% duty on the loan and return of lent/borrowed shares for both parties to a loanwas tightened (Securities Review Committee 1988, DR, p. 112).

3.3 Concentrated risk

The high volumes traded at the futures exchange made it look very successful. How-ever, a major problem was building in the positions being held and the inability ofanyone to manage the risk.

Net open interest, or the number of contracts that remained outstanding, had risen tonearly 36 000. This represented approximately HK$7 billion worth of stock againstan overall market capitalization of roughly HK$600 billion. While the size of the





80 R. T. Cox

BOX 2 Who’s minding the store? (Securities Review Committee 1988, DR, pp. 180, 183.)

It is hard to see who was monitoring the position-taking of HKFE members. The HKFEwas conducting some surveillance on its members by monitoring their financial returns.This would spot a member’s balance sheet but not market risk. ICCH (HK) was settinginitial margins for the clearing members based on volatility and saw the clearing members’positions in aggregate. FGC saw very little and instead relied on ICCH (HK) to manageaffairs. There was no effort by the three to coordinate on risk management.

HKFE:

� membership requirements,

� surveillance based on financial returns,

� sets minimum client margins,

� monitors floor trading.

ICCH (HK):

� trade matching,

� calculates and settles daily pays and collects,

� sets initial margin levels for members,

� no auditing or inspection powers.

FGC:

� admission of expulsion of members,

� approves settlement banks,

� bears the clearing risk,

� no auditing or inspection powers.

BOX 3 Links.

The futures and stock exchanges had many links, starting with the HSI futures contractand including many common participants. The popularity of margin trading was high inthe futures market, but this created a need to be able to realize cash quickly to meetmargin calls in case of adverse price movements. The settlement backlog in the stockmarket undermined the ability of investors to raise cash promptly by selling shares formargin purposes. This connection underscores the need for robust market infrastructure,especially for settlement and straight-through processing. It also highlights that when twomarkets are linked, a crisis situation will demand coordinated action by the two separateexchanges and their regulator(s). This is especially true if, as was the case with the HongKong stock and futures exchanges, the links include substantial cross-market arbitragepositions.





CCPs in crisis 81

open interest may or may not have been overly large, the danger lurked in how it wascomprised.

� On the long side of the open interest, retail speculators were poorly capitalized,and some had not even posted margins, as they were being financed by theirbrokers. These speculators were aligned with Robert Ng, who used a pair oftwo-dollar Panamanian nominee companies to amass a long position, withassociates, that constituted over 50% of the market. Accordingly, just threemembers held more that 50% of the long position (Securities Review Committee1988, Appendixes, p. 411).

� The short side of the open interest was dominated by arbitrageurs, largely fromoverseas. The buying pressure of the enormous long positions had forced anextraordinary premium into the pricing structure of the futures contract: futurestraded well above their normal pricing (ie, cash price plus cost of interest lessthe dividend rate). This excess futures premium had attracted the arbitrageurswho had sold expensive futures against purchases of stock, sometimes earn-ing an annualized return as high as 30% (Securities Review Committee 1988,Appendixes, p. 408). They held the vast majority of their positions at just fourmembers, who collectively had nearly 80% of the short position.

Thus, the market was highly concentrated. The failure of any of these memberswould present a major crisis for the clearing house and guarantee corporation inparticular, and the futures market in general.

3.4 Dependence on margin

Given the de minimis resources in the FGC, protection of the integrity of the futuresmarket rested solely on the margin system.2 A client margin of HK$15 000 per contractrepresented roughly 8% of the full contract value, not an unreasonable level by modernstandards. The system had been undermined, however. Reports later surfaced of clientsplacing half margin with their brokers, giving post-dated checks and being allowedto pyramid their position size based on winnings (Mulcahy 1987b, p. 69). This wasfacilitated by ICCH (HK)’s policy of margining the members’client positions on a netbasis, whereby members who had offsetting client positions could use the collectedmargins to fund uncollected margins of other clients.

The margin system collapsed like a house of cards when the storm blew in.

2 In a modern-day CCP, an Achilles’heel in volatility-based margins is recognized: an extreme priceevent that represents volatility in excess of the observations used in the risk modeling. To providefor this weakness, modern CCPs have financial safeguards, such as a members’ default fund, toensure that the CCP has ample resources should an extreme but plausible event occur.





82 R. T. Cox

4 CRISIS AND CRISIS MANAGEMENT

Any serious shock to confidence can cause sales by those speculators who havealways hoped to get out before the final collapse, but after all possible gains fromrising prices have been reaped. Their pessimism will infect those simpler souls whohad thought the market might go up forever but who now will change their mindsand sell. Soon there will be margin calls, and still others will be forced to sell. So thebubble breaks.

Galbraith (1954)

4.1 The crash

Having peaked on October 1, the HSI followed overseas markets lower. The dropof 10.9% on Monday, October 19 was the first clear sign of trouble. This followedsmaller falls the previous week.

The price declines of October 19 drove ICCH (HK) to make intraday calls on longposition holders in the futures market at midday and again at 15:00 for a total ofHK$192 million. All was paid except HK$4 million, which was received on October20 (Securities Review Committee 1988, Appendixes, p. 350).

There was more to come, though. During the Hong Kong evening, the Dow JonesIndustrial Average fell by 22.6%.

4.2 Market closed

Following the drop in New York, the Hong Kong Financial Secretary Piers Jacobswas awakened at 04:00 on October 20 by a phone call from Ronald Li. Li explainedthat, due to a backlog of 250 000 unsettled trades (equivalent to a week’s trading),the Stock Exchange of Hong Kong would close for the rest of the week (SecuritiesReview Committee 1988, Appendixes, p. 350). The move, and Jacobs’s acquiescenceto it, would ultimately prove controversial, prompting the Hongkong and ShanghaiBank chairman William “Willie” Purves to remark, “I don’t know what my responsewould have been at that hour, but I hope it would be to ring back after breakfast”(Kirkland and Kraar 1987).

Despite objections from the Commissioner for Securities over the length of theclosure, the stock exchange’s committee voted to suspend trading for four days underits emergency powers. In the public announcement, it said that its decision “wasbased on the possibility of panic selling, confusion and disorder in the market, theliquidity of members, the possibility of bank runs and the uncertainty caused bythe settlement backlog” (Securities Review Committee 1988, Appendixes, p. 350).Sydney Morning Herald correspondent Eric Ellis was forcibly removed from thepress conference after questioning Ronald Li’s authority to close the exchange andLi’s personal position in the market in front of television cameras and global newsagencies. Ellis would subsequently receive death threats and be denounced in Wen





CCPs in crisis 83

Wei Pao, a Communist Party newspaper (Asia Sentinel 2007). The HKFE followedthe lead of its vice chairman Ronald Li and closed for the balance of the week.

Under questioning the following day before the Legislative Council, Piers Jacobsrepeatedly described the closure as “sensible”. He went on to explain that

had the markets opened for trading on Tuesday the market would undoubtedly havebeen disorderly. It would not have worked in the interests of investors. It would nothave worked in the interests of the community.

He stated that “indeed if, as I hope, markets stabilize shortly we may well be seenas having avoided some of the major problems experienced elsewhere” (Hong KongLegislative Council 1987b, pp. 138–139). Jacobs also rebuffed a call for an investi-gation into the trading positions of the stock exchange’s committee members as “anunwarranted intervention and interference” (Hong Kong Legislative Council 1987b,p. 139).

4.3 Trouble at the till

Although it had closed for trading on Tuesday, October 20, the futures market stillneeded to settle accounts from the previous day’s market decline. This did not go well.Margins called from the membership of HK$382 million were not met in full, withHK$108 million (28%) left unpaid at day’s end.3 Chairman Kim Cham notified theSecretary for Monetary Affairs David Nendick that members were facing difficultieshonoring their obligations as many clients were “walking away from their commit-ments” (Securities Review Committee 1988, Appendixes, p. 350). Kim Cham alsoconveyed “serious doubts” about the ability of the FGC, as currently capitalized, tofulfill its role as the guarantor of contracts in this environment (Securities ReviewCommittee 1988, Appendixes, p. 350). This was consistent with the initial reaction ofsome of the FGC’s principals: ICCH, the Hongkong and Shanghai Bank, and Stan-dard Chartered were reported to support the notion of tearing up or “ringing out” thecontracts at the October 19 closing levels, as they wished to limit their exposure tothe capital of the FGC (Mulcahy 1987a, p. 61).

The situation was seen as critical by much of the futures exchange membershiptoo: it was generally expected that the HSI futures contract would be closed out at anartificial level (personal communication).

4.4 Two approaches

The futures exchange’s membership was divided in its approach to the problem ofthe inadequacy of the FGC’s resources. Both sides appealed to the government on

3 ICCH (HK)’s intraday margin calls on the previous day had not fully kept pace with the market’sdecline.





84 R. T. Cox

BOX 4 CCP insolvency.

In modern CCP risk management, there is a waterfall of resources available to be used toallow the CCP to make variation margin payments due from the positions of a defaultedmember. A typical waterfall would be the initial margin deposits of the defaulting member,its contribution to the CCP’s default fund, CCP capital committed to the waterfall, the non-defaulting clearing members’ contributions to the default fund and, in some cases, assess-ments made on nondefaulting clearing members. According to the “Principles for financialmarket infrastructures” (Committee on Payments and Settlement Systems/InternationalOrganization of Securities Commissions (CPSS–IOSCO) 2014), the pre-funded elementsshould cover the risk of the failure of the clearing member that would potentially cause thegreatest credit exposure in extreme but plausible circumstances.For systemically importantCCPs, they should cover the risk of failure of the two such largest clearing members.

Should the waterfall resources prove insufficient, the CCP may employ recovery tools.Those listed in the “Recovery of financial market infrastructures” (CPMI–IOSCO 2014)recovery toolbox include variation margin gains haircutting and cash calls from nonde-faulting clearing members.

A CCP becomes insolvent if the resources available to it (after employing the relevantrecovery tools and exhausting any other potential options) are insufficient to allow it tomake variation payments on behalf of its defaulted clearing members.

In the case of the HKFE in 1987, there was no waterfall, as there was no mutualizedguarantee by the members. Instead, the guarantee came from the FGC. With its capitalresources of HK$22.5 million, the FGC was unable to meet its obligation to pay HK$108million on behalf of the defaulted members from the close of Monday, October 19.

At that time, the FGC was insolvent.

Wednesday, October 21. The division between them was based on a simple distinction:some were long and some were short. The longs wanted a close out, while the shortswanted a recapitalization (and thus a lower risk of losing their positions and paymentof their winnings).

David Nendick met with the FGC’s chairman and representatives of major brokersin the morning. Their argument was that overseas arbitrageurs and hedgers held themajority of the short open interest, and these parties would be forced to sell theirlong shares positions if their futures positions were taken from them. The concernarticulated by the group was that such forced selling would aggravate an already fragilemarket, causing “a massive downward pressure … major economic disruptions andserious damage to Hong Kong’s reputation as a financial center” (Securities ReviewCommittee 1988, Appendixes, p. 351).

The group’s recommendation was for a bailout of the FGC, comprising HK$1 bil-lion from the government’s Exchange Fund, HK$500 million from FGC’s sharehold-ers and HK$500 million from the major brokers. This was believed to be sufficient to





CCPs in crisis 85

enable the FGC to make payments in the place of the defaulting longs in the event ofa further fall of 1000 points in the HSI futures.

Futures exchange members Wardley and James Capel, both part of the Hongkongand Shanghai group, were holding over 50% of the short open interest (Scott 1989,p. 261; Mulcahy 1987a, p. 61). As it began to understand its market risk in the eventof a ring out of the contract, the Hongkong and Shanghai group “quickly lent supportto the notion of a rescue package” (Mulcahy 1987a, p. 61).

Nendick rejected the proposal, voicing his view that the FGC should be recap-italized by its shareholders and that the short futures position holders should vol-untarily agree to refrain from dumping shares (Securities Review Committee 1988,Appendixes, p. 351; Mulcahy 1987c, p. 63).

Later in the day, Kim Cham appealed for a ring out of the contract at the closingprice of October 19 “in order to prevent a collapse of the exchange”. There was noagreement (Securities Review Committee 1988, Appendixes, p. 351).

4.5 Intervention

The government’s dilemma cut to the heart of its dogma of positive non-interven-tionism. Not intervening would mean allowing a ring out. This would destroy not onlythe futures market but in all probability Hong Kong’s reputation as an internationalfinancial center. A bailout of the FGC would mean intervention and interference,actions that were to be avoided under the dogma and would doubtless attract politicalbacklash. With Hong Kong’s governor away on an overseas trade mission, the burdenfell squarely on Piers Jacobs.

There was an out in the positive non-interventionist dogma, however: interventionistproposals were to be evaluated as to “where the balance of advantage lies”.

For reasons that were never entirely clear, Piers Jacobs reversed his views about abailout on Thursday, October 22, and engaged the London merchant bank Hambrosto put together a plan. Hambros’ vice chairman Charles Perrin, a workout specialist,led a team that included London International Financial Futures Exchange chairmanBrian Williamson. The plan bore a very close resemblance to that rejected by DavidNendick on October 21. The FGC was to be given a loan of HK$2 billion, morethan enough to cover a 1000-point drop in the index. This was to be funded byHK$1 billion from the government’s Exchange Fund, HK$500 million from the FGCshareholders in proportion to their ownership and HK$500 million from a number offutures exchange members, including Wardley, James Capel, Vickers da Costa andSun Hung Kai. The net short position of those four firms was 28 400 contracts; forHongkong and Shanghai group-owned Wardley and James Capel alone, the figurewas 22 200 (Scott 1989, p. 261).





86 R. T. Cox

BOX 5 Ringing out.

Modern CCP default management recognizes the importance of reestablishing a “matchedbook” at the CCP, in essence liquidating the positions of the defaulting member(s) so thatthe CCP has matching longs and shorts. The tools for accomplishing this are typically thefollowing:

� liquidating the defaulter’s position in the market;

� holding an auction of the defaulter’s positions (perhaps after executing immunizinghedge trades, and preferably within one to two days); and

� in extremis, forced allocation of positions to members; or

� contract tear-up, either in part or whole.

Forced allocation to members and a partial or complete contract tear-up carry the risk oftransmitting contagion. Forcing the positions onto members may affect a limited number ofplayers, but if the action topples any of the affected members, a cascade may be triggeredwith unforeseeable results.Tear-up could transmit contagion to a wider universe of marketparticipants, including customers with even greater uncertainty as to the adverse rippleeffects.

Liquidation or auction within one to two days was impossible with a four-day market closure.

By proposing a ring out of the contract at the last settlement price from October 19, KimCham was proposing what we would today term a full tear-up.This would have terminatedall positions and left the FGC with uncollected amounts owed by members of about HK$108million. Given the initial posturing of the FGC’s owners that their liability was limited to theirpaid in capital and retained earnings, there would have been a shortfall in the variationpayments due to the shorts.

After a ring out, the holders of long futures positions would have had no exposure to a furthermarket drop, while the futures shorts would have had no protection from a further declinein prices. This would have been catastrophic for the long stock–short futures arbitrageurs.The obvious contagion effect would have been the forced liquidation of the arbitrageurs’long stock positions.

Kim Cham’s hope of averting a collapse of the futures market might have been achievedin the very short run, but the crisis would have intensified at the stock exchange.

Other members who were included in the loan did not have meaningful short posi-tions but were “persuaded” to join by senior executives of the Hongkong and Shanghaigroup with “veiled threats about retribution if support was not forthcoming” (Mulc-ahy 1987c, p. 63). Indeed, during deliberations, the only member of the government’sExecutive Council who reportedly understood what was going on was Hongkong andShanghai Bank’s chairman Willie Purves (Lau 1987, p. 69).

The loan was to be repaid, at market interest rates, by a special levy on stocktransactions of 0.03%, of HK$30 per trade on futures transactions and collection





CCPs in crisis 87

BOX 6 The Exchange Fund.

Following the abandonment of silver as a monetary standard in 1934, the Hong Konggovernment enacted the Currency Ordinance of 1935. This established a fund (called byits function the Exchange Fund) to purchase silver from the public and sell it in Londonfor pound sterling. The assets of the fund backed the currency note issuance of the localnote issuing banks. The purchases later shifted to US dollars as sterling became morevolatile beginning in 1967 and ahead of the implementation of the Hong Kong dollar’s pegto the US dollar (Kasa 1999). By 1987, the Exchange Fund, which also served a role as thegovernment’s fiscal reserve depository, had accumulated earnings of over HK$50 billionand held fiscal reserves of over HK$30 billion (Jao 1998, p. 57).

By tradition, the note-issuing banks of Hong Kong were expected to act as the lenders of lastresort, essentially fulfilling a function that elsewhere would have fallen to a central bank.TheExchange Fund was used in the 1980s, however, in the instance of some bank failures,including the government’s takeovers of the Hang Lung Bank (1983) and the OverseasTrust Bank (1985) (Jao 1998, p. 60). The justification was that the Financial Secretary wasgranted discretion to use the fund for such purposes as he thought fit to stabilize the valueof the Hong Kong dollar. Protecting the financial system to avoid capital flight was seenas within this purview; indeed, the statutory function of the fund was extended in 1992 toexplicitly provide for the fund so that it could assume the secondary and subsidiary roleof maintaining the stability and the integrity of the monetary and financial systems, witha view to maintaining Hong Kong as an international financial center (Legislative CouncilSecretariat 2010, 3.2).

The potential use of the Exchange Fund for the assistance of the Hongkong and ShanghaiBank had also been contemplated prior to the crash of 1987. No less than (then) FinancialSecretary Philip Haddon-Cave, during the bank’s unsuccessful bid to acquire the RoyalBank of Scotland in 1981, assured the Bank of England’s deputy governor Kit McMahonthat, should anything go wrong with the Hongkong and Shanghai Bank, the well-resourcedExchange Fund would act as the lender of last resort, and thus the Bank of England wouldnot be called upon (Roberts and Kynaston 2015, p. 77).

Of course, no one contemplated the use of the Exchange Fund to rescue the FuturesGuarantee Corporation prior to the crash.

action against the defaulters. The Exchange Fund was to be repaid ahead of theprivate sector contributors. The arbitrageurs undertook not to sell any of their sharesunless they lifted their corresponding futures hedge.

The government also imposed leadership changes at the exchanges. Kim Chamand Ronald Li were removed from their posts at the futures exchange and replacedby Wilfred Newton, the chairman of the Mass Transit Railway Corporation, andPhilip Thorpe, the Assistant Securities Commissioner. Ronald Li also stepped downas the chairman of the stock exchange and the government had Robert Fell, a formerSecurities Commissioner, appointed as chief executive.





88 R. T. Cox

BOX 7 A toxic chain of default.

A chain of failures-to-pay can sometimes arise in clearing systems. The problem beginswhen some of the customers will not pay. Some of the system’s members will not pay inturn because they cannot collect from their customers and do not want to lose their capital.There is no bulwark if, as was the case in Hong Kong, the guarantor of the market is notrobust enough to be credibly seen to guarantee all trades. This can lead to parties whocan fail to pay opportunistically, defaulting to limit their exposures.

A resolution authority needs to be mindful of the corrosive effect of the loss of confidencein a clearing system in situations like this.

Preventive measures can be put in place to mitigate these risks. These include sufficientresources at the guarantor, clearly understood ex ante rules about member obligations,and the firm expectation that they will be fulfilled. Members need to have an indisputableunderstanding that they will be sharing in losses, both by guaranteeing their customers’performance as well as that of the CCP. In a run such as this, though, the resolutionauthority must stand very tall to break the toxic chain and reestablish market confidence.

4.6 Market reaction

Panic selling ruled the day when the markets finally reopened on Monday, October 26.Investors, trapped by the four-day closure, stampeded for the exits. The HSI closeddown 1120 points (�33%) at the stock exchange, while at the futures exchange thespot contract crashed 1544 points (�44%). The futures premium was gone, withfutures trading at a 266-point discount to the index by the end of the day. Marginswere increased to HK$25 000 per contract at the beginning of the day and doubled toHK$50 000 at day’s end (Reuters 1987b).

Of the HK$2 billion loan to the FGC, about HK$200 million remained as FGC paidthe shorts, standing in for the defaulted longs. Piers Jacobs was informed that in theabsence of additional resources the FGC would have to stop guaranteeing contracts.The government and Hambros worked overnight and arranged a second bailout fund,this time with HK$1 billion from the Exchange Fund, and HK$1 billion split evenlybetween the Hongkong and Shanghai Bank, Standard Chartered and the Bank ofChina. This second facility was never drawn upon (Securities Review Committee1988, Appendixes, p. 353).

The presence of the new facility, a waiver of some of the provisions in the takeovercode, a reduction in the prime rate by the banks and outright buying by the banks tosupport the market stopped the rout on Tuesday, October 27.

As was feared, many clients, including Robert Ng, failed to pay; forty-five default-ing members were then suspended. Some members were broke, while others sim-ply refused to honor their obligations. “They’re boycotting … we’re one of them”,remarked a broker at a Japanese member firm. “They did it on purpose … they don’t





CCPs in crisis 89

BOX 8 Reestablishing a matched book.

The sale of 9000 of the defaulted contracts by the FGC to magnates Li Ka-shing, ChengYu-tung and Lee Shau-kee provides an interesting tool if considered as an action followinga failed auction. Faced with the alternative of using a contagion transmitting tool in orderto reestablish a matched book, a CCP may well decide that it can do better by “cutting adeal” off-market with parties outside of the CCP’s membership who are willing to take theposition.

While it is understandable that members and market participants want ex ante certainty ofthe steps that can be taken by a CCP in a default management, the price of such certaintyis to limit flexibility. The CCP, or its resolution authority, in extremis may sometimes needto take actions that were hitherto unforeseen.

want to dish out their own capital … until investors pay them.”Among the suspensionswas Chine Kit, a firm in which Ronald Li had an interest (Reuters 1987a).

Many retail clients refused to pay too, as they felt that they were not given a chanceto get out of their positions due to the four-day market closure. Others felt that theirlosses should be limited to the margin that they posted. A member delegation metwith the newly installed chairman Wilfred Newton on Thursday, October 29 to pleadin vain for a settling of the long positions at a “reasonable price” (Reuters 1987c).

Some of the futures exchange’s defaulting members, including Chine Kit, eventu-ally paid, and the FGC liquidated the positions of the remaining thirty-four defaultedmembers between October 27 and November 2. Part of the liquidation was an off-market sale of 9000 contracts to the magnates Li Ka-shing, chairman of HutchisonWhampoa; ChengYu-tung, chairman of New World Development; and Lee Shau-kee,chairman of Henderson Land (Marchand and Bowring 1987, p. 80).4

5 THE AFTERMATH AND THE DAVISON REPORT

The HKFE crisis had moved from no bailout to one to two bailouts in a short period.This was bound to attract comment and to highlight the need to learn the lessons ofthe crisis. In this section, we discuss this aftermath.

5.1 Political reaction

Adverse reaction to the way the crisis was managed was not confined to the retailclients and the members but spread into the press and the Legislative Council.

When pressed in a council meeting on October 28, Piers Jacobs reiterated hisjustification for using the Exchange Fund to bail out the FGC. Gone was the hope

4 The sale was allegedly at around 1950, about a 200-point discount to the spot index value, or adiscount of around 9–10%.





90 R. T. Cox

that Hong Kong was to be “seen as having avoided some of the major problemsexperienced elsewhere”. Instead, he offered that he had

looked at the circumstances that faced us over the last week, certainly the beginningof the period, and I decided that immediate action was needed. I perceived a possi-ble threat to the exchange value of our currency.... I think the disorder that I havedescribed would have spilt out into other sectors of the community and may wellhave spilt out on to the streets.

Hong Kong Legislative Council (1987a, p. 152)

Council member David Li offered:

The events of the past two weeks have shown the government at its ineffective worst.At a time when the collapse of the local securities markets indicated a need for majorsurgery, the government was seen to be running round sticking on Band-Aids.

Of the bailout, he observed:

This loan of public money is to ensure that speculators are paid their profits on futurescontracts. It is to be repaid not by the speculators themselves but by a levy paid bythe investing public. This surely is adding insult to injury … the widely held viewis that the government would have done well to have let the futures markets and itsaccompanying speculators find their own solution, and the government itself shouldhave concentrated its attention and our funds on handling the problems of the StockExchange.

Of the justification for using the Exchange Fund, he noted:

The Financial Secretary has so far failed to advance any cogent economic or otherreasons as to how the collapse of the Futures Exchange would have affected the valueof the Hong Kong dollar.

Hong Kong Legislative Council (1987c, pp. 249–50)

Finally, council member Lydia Dunn waxed existential, asking:

Do we need a stock index futures market?… In what way does a futures marketcontribute to Hong Kong’s economy? We managed for a long time without one andit seems to have become little more than a licensed casino.

Hong Kong Legislative Council (1987c, p. 196)

5.2 Further actions

Piers Jacobs announced the formation of a six-member Securities Review Committeeon November 11. Led by Ian Hay Davison, a former chief executive of Lloyd’s ofLondon and managing partner of Arthur Andersen, the committee was tasked with acomprehensive review of the exchanges and its regulators.

Ronald Li was arrested by the Independent Commission Against Corruption inearly January 1988 and charged under the Prevention of Bribery Ordinance over an





CCPs in crisis 91

allotment of shares. He was convicted, the conviction was upheld on appeal in 1991,and he served over two years in prison.

Robert Ng escaped prosecution (despite the Commercial Crime Bureau’s findingsof prima facie evidence of a breach of the Commodities Ordinance) when the gov-ernment accepted the argument from David Nendick that taking him on would posea risk to the stability of the overall market (Stillwell 2007, p. 116). In a controversialmove, he settled the HK$1.1 billion owed to the FGC for HK$750 million to be paidover eight years. His original debt was 61% of the HK$1.8 billion owed to the FGC.

In the meantime, volume and open interest at the HKFE plummeted, beginning afive-year drought of any meaningful trading activity.

5.3 The Davison Report

The Davison Report was released on May 27, 1988 (see Securities Review Committee1988). It is a remarkable document not only for the breadth of its coverage of issuesbut also for its candor. It provided an extensive examination of the issues affectingthe Stock Exchange of Hong Kong, the HKFE and the regulatory environment.

Despite the numerous systemic flaws the commission was obliged to examine,it remained true to its main objective: “to examine critically the relevant structuresand systems to build a blueprint for the future” (Securities Review Committee 1988,DR 2.26). This was done in the context of Hong Kong’s role as an international finan-cial center, with the stated intention of restoring confidence in Hong Kong’s securitiesmarkets. In addition to meeting with local market participants and institutions, thecommittee’s views were informed by extensive interviews with the exchanges and reg-ulatory authorities in the United Kingdom, United States, France, Japan, Singaporeand Australia.

5.4 Poor risk management

The report noted poor risk management and lax credit controls in both the stock andfutures markets and “at every level of the system from individual brokers throughto the clearing house” (Securities Review Committee 1988, DR 6.5). The main rec-ommendations for the futures exchange were for new arrangements to be in placeto

(a) ensure that the clearing members are robust,

(b) require clearing members to undertake to support the risk of the clearing house,

(c) give the exchange de facto control of the clearing house’s risk management,

(d) put in place a new regulatory authority to ensure that adequate safeguards arein place (Securities Review Committee 1988, DR 6.58).





92 R. T. Cox

The report was especially critical of the futures market’s risk management structure.As the exchange was not directly at risk from member or client failures, it was foundnot to have an “immediate incentive to play a day-to-day role in ensuring the financialintegrity and stability of the market” (Securities Review Committee 1988, DR 7.4).The FGC did not have the resources to do the job either. Indeed, FGC was solelydependent on its relationship with ICCH (HK), and it had no direct relationship tothe HKFE.

5.5 Where is the guarantee?

The report pointed out that the futures guarantee arrangements were flawed, with thereport citing four points.

(a) The guarantee was not direct, there being no direct legal relationship betweenthe FGC and the clearing members.

(b) The mechanics were vague, with uncertainty over whether the FGC was thecounterparty to each trade or just guaranteed contracts. The following questionwas raised: if the latter were true, would the FGC be guaranteeing the members’clients?

(c) The timing was unclear as to when the guarantee became effective.

(d) The form was seen as inappropriate, with the members having no responsibilityto support the market in the event of a collapse, and the FGC, as a limitedliability company, having no legal obligation to meet losses beyond its resources(Securities Review Committee 1988, DR 7.21).

The report also noted that when the FGC finally contemplated boosting its resourcesduring the summer of 1987, nothing was done. “In any case, discussions centered onincreasing the capital in case of a default, rather than strengthening the arrangementsto reduce the prospect of a default” (Securities Review Committee 1988, DR 7.28).

5.6 Suggested changes

The report called for changes to at least

(a) include a clarification of the terms and mechanics of the guarantee so that thesewere “unequivocal and easily understood”,

(b) require ICCH (HK) to act “as a risk manager first and a processor second”,

(c) require HKFE and ICCH (HK) to work “closely together to monitor markets”(Securities Review Committee 1988, DR 7.30).





CCPs in crisis 93

The report concluded that the existing risk management structure needed a compre-hensive overhaul. Central to its recommendations were the integration of the clearingand guarantee functions in the clearing house, and the requirement that clearing mem-bers underwrite (at least in part) the risk of the clearing house. It examined the issueof maintaining a separate clearing house or having it housed within the Exchange:independence versus integration. It settled on the latter option, a separate companywithin the exchange with stipulations about the rank of the clearing house head inthe exchange organization, the representation on its board of both the exchange CEOand representatives of those backing the risk, and an enjoinder that “there should bea strong independent element” (Securities Review Committee 1988, DR 7.38).

Further recommendations included a suggestion that the clearing house seek defaultresources beyond a member’s clearing fund in the form of external cover from insur-ance or bank guarantees, and assessment powers over member’s capital (SecuritiesReview Committee 1988, DR 7.41–51).

There was sharp criticism of the regulatory framework as well, with the reportrecommending the creation of a statutory regulator separate from the government,with professional rather than civil service staff, funded by the market and chargedwith ensuring market integrity and investor protection. This led to the creation of theSecurities and Futures Commission in 1989.

5.7 Epilogue

The HKFE remained under effective control of the government until 1992. The volumeeroded to less than a thousand contracts a month, and a substantial overhaul of therisk management system of the exchange was completed, largely implementing therecommendations of the Davison Report. In the interim, the backing of the FGC wasscaled back, commensurate with the sharply lower exchange open interest.

Hong Kong Futures Exchange Clearing Corporation (HKCC) was created in 1989to reform the clearing house. It was integrated into the exchange management structurebut had a separate board of directors for an independent element. Clearing membercommitments and bank guarantees provided the main financial backing of its clearingfund, known as the Reserve Fund. HKCC novated contracts, and it provided a clearand explicit guarantee of positions to clearing members. ICCH (HK) was relegated toa processing role until the clearing corporation was staffed at the beginning of 1992.

Financial safeguards included gross margining for client accounts, concentrationmargins for large member positions, capital-based position limits for clearing mem-bers and full assessment rights on members’ capital. Clearing member entities wererequired to be incorporated in Hong Kong. Management was delegated the power toraise volatility based margin requirements as needed, but only a vote at a regularly





94 R. T. Cox

scheduled monthly board meeting could lower margins. A provision for increasingthe HKCC’s Reserve Fund in line with market growth came later.

The collection rate from the 1987 defaulters allowed for the levy to be abolished in1993. The bailout loan of the FGC was fully repaid in 1995. A HK$71 million surpluswas transferred to the clearing corporation’s Reserve Fund in late 1997. As there wasvery little futures volume before 1993, the weight of the levy fell most heavily onthe customers of the stock exchange. This engendered bitter resentment of the futuresexchange in many stock exchange members that endured beyond the merger of thetwo exchanges to form Hong Kong Exchanges and Clearing (HKEX) in 2000.

The government relinquished direct control of the futures exchange in 1992 withthe election of Leong Ka-Chai, a member, as chairman. Trading volumes began torecover. International confidence returned after a senior American Stock Exchangeexecutive, Ivers Riley, became the chief executive in 1994.

The new risk management system was put to a major test when the exchange’slargest clearing member, Barings, failed in 1995. The crisis was resolved in threedays without employing the guarantee resources.

The reforms had worked.

6 LESSONS LEARNED

The 1987 futures crisis in Hong Kong should be viewed as a CCP resolution thatsucceeded. Here are some lessons that present themselves from these events.

6.1 Aim of the intervention

A resolution authority that is brought in to resolve a failed CCP must consider thedesired outcome of its action. There are essentially two alternatives.

(1) A short-term set of measures to assure continuity of service long enough tofacilitate an orderly wind down of positions, or to buy enough time to facilitatethe sale of the CCP to another entity.

(2) Putting the CCP on a sufficiently sound footing so that it can continue in businesswith the market’s confidence for the foreseeable future. This can be seen as arehabilitation of the CCP.

The 1987 crisis in Hong Kong was an example of the latter.

6.2 Resources

The resolution authority must decide what financial resources are required and howthey should be sourced. As the Hong Kong example indicates, there may be pressure





CCPs in crisis 95

to abandon previous commitments to avoid the use of taxpayer funds. In any case,consideration should be given to involving private institutional funds as well as anypublic resources available, with the resolution authority leading the effort.5

A major factor in sizing the rescue funding is whether the CCP has been able toreestablish a matched book.Without a matched book, the CCP’s ultimate losses cannotbe fixed, only estimated, and those estimates will only be as good as the estimates ofclose-out prices that the CCP can get as it continues to liquidate or auction defaultedpositions. Even if the contagion transmitting tools of forced allocation or tear-up areused, the ultimate magnitude of the CCP’s losses will still be a function of the invoiceprice selected when the tools are employed.

To restore confidence, which will surely have vanished in a CCP crisis that provokesa resolution action, the resolution authority must employ enough funding to cover theCCP’s losses and provide a suitable buffer to support ongoing operation. It also needsto have quick access to supplemental funding, should the process of reestablishinga matched book prove more costly than previously estimated.6 More simply, newarrangements must be decisive and sufficiently robust to restore confidence.

6.3 Fallout

The political fallout in the HKFE crisis is also instructive. It suggests that any action,or perhaps even inaction, by a resolution authority will inevitably generate an adversepolitical reaction in at least some quarters. This underscores the importance of makinga proper inquiry to establish the causes, recommend solutions and plot a responsible,confidence-instilling road for the future. As can be seen in the Davison Report, themore professional and less political this type of endeavor, the better the likely result.

6.4 Review

The experience of ICCH suggests that, after a failure, it makes sense to take a goodlook in the mirror. The failure of the futures market in Kuala Lumpur in 1984 wasnot on ICCH’s watch as it was no longer the clearing house operator, but the systemin place was of their design. The market collapse was precipitated by an overly largeclient position default. Failure to spot this design flaw and take remedial action led toits replication in the large client position failure of Robert Ng in Hong Kong. Again,ICCH did not identify and address its design flaw. This left the door open for yetanother overly large client position that defaulted in New Zealand in 1989.

5 In the case of Hong Kong in 1987, while some of the FGC shareholders may have been thebeneficiaries of the rescue, they were also required to help fund that rescue.6 Piers Jacobs was fortunate in being able to cobble together a supplemental package overnight. Itwould be preferable, especially in terms of market confidence, if a resolution authority could bespared such nocturnal drama.





96 R. T. Cox

6.5 Responsibilities

Those who manage the risk must have a meaningful stake in it. Failure to do so willcreate a false sense of comfort. The situation of ICCH (HK) managing the risk withno direct consequences of a failure, and the Nelsonian “blind eye” given to the risk bythe FGC who bore it, was not conducive to robust risk management. Indeed, it was amajor structural flaw that enabled the crisis to occur. The solution was an integratedclearing house with a separate, member-populated board of directors. Those bearingthe risk were thus put in a position to manage it, at least from a policy level and bycontrol of the setting of margin levels.

6.6 Involving stakeholders

If the goal of a CCP resolution is rehabilitation, success cannot be taken as a given.The Hong Kong futures episode was marked by a prolonged, multi-year slump intrading activity. Major reforms were made during this period, and the rank and filemembership was excluded from the process. Instead, reforms were drawn up bythe government and announced to the membership without consultation. Tradingactivity did not return to viable levels until control of the exchange was returned to themembership. The lesson here is that it is vitally important to include the membershipin the reform process, even if only by consultation.

6.7 Rehabilitation is often preferable to wind down

Looking at the aftermath of the crisis and the resolution action, it would seem thatrehabilitation is preferable to wind down. Had there been no action on behalf ofthe government, the Hang Seng contract would have been rung out at an off-markethigh price, and the futures exchange would have collapsed from the ensuing lack ofconfidence, with unforeseeable consequences. Instead, the resolution and the marketand regulatory reforms introduced by the Davison Report were crucial in restoringconfidence and laying the foundation for a viable financial market infrastructure inHong Kong.

7 CONCLUSION

The story of the great crash of 1987 in Hong Kong is a tale of the triumph of specialinterests, especially those of the Hongkong and Shanghai group and Robert Ng. Incontrast, many retail investors were trapped in a four-day market closure, and, in atragic irony, those retail investors who were fortunate enough to be short but unluckyenough to have accounts with members who failed lost both their positions and theirmoney.





CCPs in crisis 97

The story is also one of a government that was justifiably criticized for “doing toolittle before and too much after”. It underscores the need for the sound prudentialregulation of financial market utilities as well as regulators who truly understand therisks.

Finally, it is a story of the resolution of a CCP in crisis. Though the motives forthe resolution remain partly obscure, the outcome was the restoration of confidence,a blueprint for a successful future as an international financial center and an eventualrehabilitation of a failed CCP.


The views herein are the author’s own and must not be construed in any way asrepresenting policy of the Federal Reserve Bank of Chicago or any part of the FederalReserve System.

The author was a member of the board of directors of the Hong Kong FuturesExchange Clearing Corporation from 1997–2000 and a member of Hong KongExchanges & Clearing (HKEX)’s Clearing Consultative Panel from 2000–2002.

ACKNOWLEDGEMENTS

The author thanks David Murphy, Senior Policy Advisor, Bank of England; RicoLeung, Senior Director, Securities and Futures Commission; and David Marshall,Senior Vice President, Associate Director of Research and Director of FinancialMarkets, Federal Reserve Bank of Chicago, for their comments and suggestions.

REFERENCES

Asia Sentinel (2007). Remembrance of things past. Asia Sentinel, October 23.Committee on Payments and Settlement Systems/International Organization of Securi-

ties Commissions (CPSS–IOSCO) (2014). Principles for financial market infrastructures.Report, Bank for International Settlements.

Committee on Payments and Settlement Systems/International Organization of Securi-ties Commissions (CPSS–IOSCO) (2014). Recovery of financial market infrastructures.Report, Bank for International Settlements.

Galbraith, J. K. (1954). The Great Crash, 1929. Penguin.Hong Kong Legislative Council (1987a). Official report of proceedings (Hansard), October

28, 147–182. URL: www.legco.gov.hk/yr87-88/english/lc_sitg/hansard/h871028.pdfHong Kong Legislative Council (1987b). Suspension of trading activities on the Stock

Exchange of Hong Kong, October 21. 123-145. URL: www.legco.gov.hk/yr87-88/english/lc_sitg/hansard/h871021.pdf

Hong Kong Legislative Council (1987c). Hansard, November 4, 183–276. URL: www.legco.gov.hk/yr87-88/english/lc_sitg/hansard/h871104.pdf





98 R. T. Cox

Jao, Y. C. (1998). Money and finance in Hong Kong: retrospect and prospect. NationalUniversity of Singapore, East Asian Institute, World Scientific.

Kasa, K. (1999). Why attack a currency board? Economic Letter, November 26, FederalReserve Bank of San Francisco.

Kirkland, R. I., Jr., and Kraar, L. (1987). Global traders head for home. Fortune, December7.

Kristof, N. (1987). Hong Kong’s hottest hot spot. New York Times, October 11.Lau, E. (1987). Selling shoddy goods. Far Eastern Economic Review, November 5.Legislative Council Secretariat (2010). Management of the Exchange Fund and fiscal

reserves. Fact Sheet FS16/09-10, Research and Library Services Division, LegislativeCouncil.

Marchand, C., and Bowring, P. (1987). Handle with care. Far Eastern Economic Review,November 12.

Mulcahy, J. (1987a). No confidence trick. Far Eastern Economic Review, November 5.Mulcahy, J. (1987b). When surveillance is pushed to the sidelines. Far Eastern Economic

Review, November 5.Mulcahy, J. (1987c). Coming to the aid of one party. Far Eastern Economic Review,

November 5.Reuters (1987a). Brokers leave H.K. futures mart over margin calls. October 27.Reuters (1987b). H.K. futures exchange raises margin, curbs selling. October 26.Reuters (1987c). H.K. futures trader says 80 firms could fail. October 29.Roberts, R., and Kynaston, D. (2015).The Lion Wakes: A Modern History of HSBC. Profile

Books.Scott, I. (1989). Political Change and the Crisis of Legitimacy in Hong Kong. University of

Hawaii Press.Securities Review Committee (1988).Operation and regulation of the Hong Kong securities

industry. (Davison) Report, May 27, Securities Review Committee.Stillwell, J. (2007). Asian Godfathers. Atlantic Monthly Press.Tsang, D. (2006). Big market, small government. Press Release, September 18, Chief

Executive, The Government of the Hong Kong Special Administrative Region.Tucker, P. (2011) Clearing houses as system risk managers. Speech, DTCC-CSFI Post

Trade Fellowship Launch, June 1, London. Bank of England.





Market Infrastructures - Risksubscriptions.risk.net/wp-content/uploads/2016/03/JFMI.pdf · Incisive...

Documents

Transcript of Market Infrastructures - Risksubscriptions.risk.net/wp-content/uploads/2016/03/JFMI.pdf · Incisive...