The CLS Bank: A Solution to the Risks of

International Payments Settlement?

Charles M. Kahn and William Roberds
Working Paper 2000-15a
November 2000

Working Paper Series

The CLS Bank: A Solution to the Risks of

International Payments Settlement?
Charles M. Kahn and William Roberds
Federal Reserve Bank of Atlanta
Working Paper 2000-15a
November 2000

Abstract: Foreign exchange transactions are subject to a unique type of settlement risk. This risk ultimately
stems from the difficulty of coordinating separate settlements in two different currencies. Settlement of foreign
exchange transactions through the proposed CLS (Continuous Linked Settlement) Bank has been discussed
as a potential solution to this problem. This paper describes the CLS proposal and analyzes the incentives it
places on banks engaged in foreign exchange transactions. The analysis shows that while settlement through
the CLS Bank may represent an improvement over current arrangements, some important problems associated
with foreign exchange settlements will remain.
JEL classification: G20, F31
Key words: foreign exchange, settlement risk

This paper was prepared for the Carnegie-Rochester Series on Public Policy, April 2000. Research for this paper was begun
while Roberds was visiting the Institute for Monetary and Economic Studies at the Bank of Japan. The authors thank
Yoshiharu Oritani and Shuhei Aoki of the Bank of Japan and Darryll Hendricks and Jamie McAndrews of the Federal
Reserve Bank of New York for explanations of the CLS settlement system. They also acknowledge the useful comments of
Jeff Lacker, Larry Neal, Ed Stevens, the editor, and an anonymous referee, as well as participants in seminars and
discussions at the Bank of Norway, the London School of Economics, Nuffield College Oxford, the Bank of England, the
Bank for International Settlements, and the European Central Bank. The views expressed here are the authors and not
necessarily those of the Federal Reserve Bank of Atlanta or the Federal Reserve System. Any remaining errors are the
authors responsibility.
Please address questions regarding content to Charles M. Kahn, Department of Finance, University of Illinois, Urbana,
Illinois, or William Roberds, Research Department, Federal Reserve Bank of Atlanta, 104 Marietta Street, N.W., Atlanta,
Georgia 30303, 404-521-8970, 404-521-8956 (fax), [email protected].
The full text of Federal Reserve Bank of Atlanta working papers, including revised versions, is available on the Atlanta Feds
Web site at http://www.frbatlanta.org/publica/work_papers/index.html. To receive notification about new papers, please
use the on-line publications order form, or contact the Public Affairs Department, Federal Reserve Bank of Atlanta, 104
Marietta Street, N.W., Atlanta, Georgia 30303-2713, 404-521-8020.

The CLS Bank: A Solution to the Risks of International Payments Settlement?

When foreign exchange transactions fail to settle, commercial banks can suffer
major losses. While settlement failures are rare, they lead to significant disruption when
they occur. Thus improvements to the system are also of interest to central banks,
concerned with the size of their potential liabilities as settlement guarantors under the
current arrangements (see Bank for International Settlements 1993, 1996, 1998).
The arrangements for settling foreign exchange transactions may undergo a major
transformation in the coming year. In 1997, a group of major commercial banks
established CLS Services, Limited, to develop a new centralized system (see Bank for
International Settlements 1998 and CLS Services Limited 1998). Under the system,
foreign exchange trades in major currencies would be settled through the CLS
(Continuous Linked Settlement) Bank, a special-purpose bank chartered in the U.S.
Not only will this arrangement involve a private entity in a role heretofore taken
on by central banks, it will also introduce a mechanism for settlement which is different
in several important ways from current techniques.1 Given the significance and the scale
of the potential changes, it is important to understand how the arrangement will alter the
risks of participation in foreign exchange markets.
In this paper, we describe the major provisions of the CLS arrangement and
develop a series of simple models of bank incentives and settlement risk that can be used
to analyze the CLS proposal. Unlike previous descriptions of the legal and technological

It should be noted that even after CLS becomes operative, some foreign exchange transactions may

continue to be settled through traditional channels, although the common expectation is that within a few
years, most transactions will be settled through CLS.

aspects of the proposed arrangement, this paper focuses on the incentives the
arrangement provides for the participants.
We find that, while settlement through the CLS Bank represents an improvement
over current arrangements, it leaves several problems associated with foreign exchange
settlement untouched. The system also leads to some complex selection effects: On the
one hand, some of the advantages of the system are likely to encourage its rapid adoption,
since they will prove most attractive to the strongest financial institutions. On the other
hand, in comparison with current arrangements, the system discourages monitoring of
counterparty quality. The analysis also points to new issues which central banks will
have to address in dealing with the new system.

1. Background
The market for foreign exchange is the worlds most active financial market,
turning over an estimated $1.5 trillion daily. By way of comparison, the daily trading
volume in the U.S. Treasury market averages about $227 billion, and the New York
Stock Exchange turns over about $29 billion per day. The FX market has also grown
quite rapidly in recent years, its dollar-equivalent nominal volume expanding roughly 11
percent annually since 1989.2

FX trading volume is the daily average volume reported by Bank for International Settlements (1999)

based on a global survey conducted in April, 1998. This figure is adjusted for double counting. Today most
foreign exchange trades are not spot trades but forex swaps. A forex swap combines a spot trade with a
later spot trade in the opposite direction of the original trade. The risks associated with the settlement of
swaps are similar to those associated with the settlement of spot trades.

The market for foreign exchange is also exceptional in other respects. Compared
to most securities markets, it is loosely organized, with the bulk of trading occurring in
the decentralized interdealer market.3 Unlike most developed financial markets, it has
no centralized mechanism for settling trades. This is at least partly due to the nature of
what is being traded: A typical spot foreign exchange trade, for example, consists of a
commitment from two parties to deliver good funds to each other by the end of the
second business day following the trade. Thus, a typical spot FX transaction is actually an
exchange of short-maturity forward commitments. Settlement of each side or leg of a
trade is governed by the respective laws and institutions of the two nations whose
currencies are involved. In practice this means that separate settlements are required in
each currency.4
Standard FX trading agreements call for banks to irreversibly commit to
settlement on the day following the trade (Bank for International Settlements 1996,
1998), but such commitments cannot be honored in all states of the world. Suppose for
example, that on day t Bank A sells yen to Bank B in return for dollars, and that on day
t+1, both A and B commit to deliver yen and dollars, respectively. If, however, Bank A
fails before it delivers yen to B, then As commitment to settle might not be honored. Nor
do existing arrangements always allow Bank B to back out of its commitment to deliver
dollars to A, even in cases where B knows of As default before it has sent any dollars to
A.

On the organization of the foreign exchange markets, see Lyons (forthcoming).

In the tradition of the international finance literature, the term currency refers to bank deposits

denominated in a given countrys currency.

An obvious question that arises is why settle FX trades in this fashion? For
example, why not settle trades immediately, or at least on the same day as the trade?
Clearly, if the funds are needed immediately then it useful to have them sooner rather
than later. The shallow answer to such questions is that the present legal and operational
infrastructure would not allow for such changes in the settlement process. Over the longer
run, the necessary infrastructure could be put in place, but doing so would entail certain
costs (see for example, Nol 1994). Accelerating settlement would require streamlining
trade confirmation procedures and more precise management of settlement funds. Also, if
certain FX settlements are seen as too big to fail and therefore implicitly guaranteed by
central banks, accelerating settlement would effectively shorten the maturity of banks
default option and thereby lessen the value of these guarantees.
As a practical matter, however, the deadline by which funds are needed can likely
be forecasted sufficiently far in advance that existing techniques for cash management
can be used to minimize the associated risks. Thus the delay may not be a major
consideration. The CLS system is focused instead on the benefits that arise from a
reduction in counterparty risk, i.e., the risk that one of the parties to a trade may not
settle. Counterparty risk will be a small consideration in any individual transaction, but
may become significant when aggregated over a system as a whole.5
Situations in which settlement does not occur because a bank cannot honor its
obligations fall into two categories: The first category is operational failure resulting from
5

That is, the likelihood of any individual counterparty being unable to settle is usually quite small, but as

the number of trades increases, settlement failures become more likely. The presence of a large number of
interconnected trades also increases the probability that a settlement failure by one trader could result in a
more costly sequence of settlement failures.

say, a natural disaster or a computer crash. The second category is bank failure,
including (more typically) closure by a regulator. From the viewpoint of the defaulting
banks counterparties, the problems associated with failures are exacerbated by the
inability to cut off settlement of one leg of the transaction when the other leg does not
settle.
The possibility of bank failure in the course of settlement is not simply a
theoretical issue. Failures occur frequently enough for them to be a concern both to
regulators and to participants. An important case of settlement problems stemming from
bank failure occurred in June 1974, when bank regulators closed a German bank,
Bankhaus Herstatt, at the end of the German business day (mid-morning in New York). A
number of Herstatts counterparties in foreign exchange transactions had settled their
Deutschmark obligations to Herstatt, in anticipation of offsetting dollar settlements. The
dollar settlements could not be made once Herstatt had been declared insolvent, resulting
in a loss of principal for Herstatts counterparties.6
Today, it is common to refer to the risk that foreign exchange transactions will not
be settled as Herstatt risk, although this name has sometimes given the misleading
impression that settlement failures can only be caused by a banks inability to settle. In
fact, settlement failures can result when either or both counterparties are able, but simply
unwilling, to settle their leg of the transaction. For example, the 1990 collapse of Drexel
Burnham Lambert resulted in a severe liquidity crisis for one of its London subsidiaries,
6

See the Allsopp report (Bank for International Settlements, 1996) for additional details of the Herstatt

episode. In cases of bank failure, unsettled obligations may or may not be offsetting depending on the laws
of the country where the failure occurs; see Bank for International Settlements (1992). As a general
principle, however, it is better to be a debtor rather than a creditor of a bankrupt.

Drexel Burnham Lambert Trading (DBLT). DBLT had been active in foreign exchange
markets and came under severe liquidity pressure as problems became apparent with its
parent group. Fearing default by DBLT, its counterparties became less willing to honor
their obligations to DBLT. At the same time, DBLT was reluctant to pay amounts due to
its counterparties, for fear that the counterparties would not settle the other leg of the
foreign exchange transaction, but instead offset any funds received from DBLT against
outstanding debts that had been incurred by other DBL companies. Central banks feared
that withholding of these settlements could have resulted in a gridlock scenario, in
which DBLT and its counterparties (and potentially its counterparties counterparties)
might have all withheld settlements, despite the fundamental solvency of everyone
involved.7
In the absence of a real-time or coordinated system for settling trades, settlement
in foreign exchange markets has been supported by banks belief in their mutual
creditworthiness. In some cases, as in the Herstatt episode, settlement failures have
occurred and losses have resulted. In other instances, central banks have intervened to
prevent settlement failures, out of concern for the systemic or knock-on effects of a
settlement failure. In the case of the DBLT, for example, a facility was set up whereby
the Bank of England, in coordination with other central banks, guaranteed that neither
DBLT nor its counterparties would take the money and run. The current system for
settling foreign exchange transactions thus poses difficulties both for market participants
and for central banks.

Again see Bank for International Settlements (1996) for more information on the DBLT episode, as well

as for other examples of settlement failures or near-failures.

2. Policy Responses and the CLS Proposal

Regulatory interest in the area of FX settlement was spurred by the 1991 failure of
the Bank of Credit and Commerce International (BCCI). BCCIs failure resulted in losses
of principal for at least two of its FX counterparties (Bank for International Settlements
1996) and prompted interventions by several central banks (George 1994). The Nol
Report (Bank for International Settlements 1993; see also the discussion in Nol 1994)
provided the first extensive discussion of policy tradeoffs in the area of FX settlement
risk. It suggested that central banks needed to find some way of coordinating settlements
across currencies through the actions of a common agent, but did not provide a detailed
plan for how such an arrangement would work. At the same time, two private-sector
groups, Multinet (Glass 1994) and Echo (Duncan 1994), advanced their own initiatives
for controlling FX settlement risk. Both schemes involved both within-currency netting
of settlement obligations and coordination of settlements across both legs of a given FX
transaction.
The successor to the Nol Report, the Allsopp Report (Bank for International
Settlements 1996), strongly endorsed the development of additional private-sector
initiatives for the coordination of settlements across currencies. The Allsopp Report was
widely interpreted as a regulatory warning that FX market participants should assume a
more proactive role in managing settlement risk (see, e.g., Davidson 1996, Sesit 1996, or
Marjanovic 2000). In response, a group of commercial banks began development of a
more comprehensive system for coordinating settlements through CLS. In December
1997, Multinet and Echo were formally merged with CLS Services.

Under the CLS proposal, foreign exchange trades would be settled through a
private intermediary known as the CLS Bank. Settlement through the CLS Bank will
differ from todays settlement practices along a number of dimensions.8 Below, we
analyze and compare settlement risks in the present, decentralized system and in the
centralized system being developed by CLS. We do this by presenting some of the issues
involved in the context of stylized examples. Subsequent sections focus on the incentive
effects of CLS in both bilateral and multilateral situations.

Examples and Details of the CLS Procedures

We will illustrate the key features of CLS through a series of examples. The first
example focuses on the basics of the CLS systems payment-versus-payment feature,
which is the centerpiece of the new arrangement. The second shows the main aspects of
the systems overdraft facilities, and in particular how the system mitigates the risks
associated with the use of overdrafts. The third example focuses on the way that the
system economizes on the use of central bank funds in settlement through quasinetting.9
Example 1. On day t, Bank A needs $1 million dollars and is willing to sell yen to
get it. Bank B needs 100 million and is willing to sell dollars. Suppose that the exchange
8

Description of the CLS settlement procedure can be found in CLS Services Limited (1998) and Roscoe

(1998). Some information is also available at the CLS website: www.cls-services.com. Since the scheme is
not yet in operation, some details presented in this paper will undoubtedly change during implementation.
9

In the examples below, all banks are assumed to be able to directly settle their own accounts. It should be

noted that under the proposed CLS system, only the largest and most creditworthy institutions will have
direct access to settlement via the CLS Bank.

rate is $1.00 = 100. What is the normal procedure by which A and B trade dollars for
yen? Today the standard arrangement would be for a trader for Bank A to contact B with
a request to buy $1 million dollars. Say Bs trader agrees to the trade at a price of
$1.00=100. In effect the trade is an exchange of promises to deliver good funds in the
appropriate currency, two business days hence.10 The first pair of accounts in Table 1
shows the situation after the agreement is made on date t. At some point during day t+2,
Bank A is obligated to send 100 million to B over FEYCS (the Japanese large-value
payment system used to settle the yen legs of foreign exchange transactions). Likewise,
Bank B is obligated to send $1 million to A over CHIPS (the corresponding large-value
system for settling dollar legs) at some point during t+2.11 As illustrated in Table 1, we
suppose that the yen transaction happens to be executed first. In most cases, both funds
transfers occur and the trade is settled. However, there is a coordination problem, under
current arrangements. For instance, if Bank B is closed down by regulators before its
funds are sent to Bank A, then Bank A suffers a capital loss. On the other hand suppose it
is Bank A that is shut down early on date t+2. Then Bank B is likely to suffer a capital
loss even if the shut down occurs before any settlement takes place, since it is difficult or
impossible for either bank to cancel its leg of transaction, should it learn of the failure of
its counterparty.
<Insert Tables 1&2 about here>

10

To ensure that no mistakes have been made there is a confirmation process the next day.

11

CHIPS is a private system operated by the New York Clearing House Association. Some dollar

settlements are also made through Fedwire, the large-value payment system operated by the Federal
Reserve.

Now consider settlement of the same transaction under the system proposed by
CLS. Suppose Banks A and B are both members of CLS; for illustrative purposes, assume
that no other transactions take place on day t. Once the two banks have verified the
transaction, each notifies the CLS of their intent to settle. The deadline for such
notification is midnight the day before the settlement (i.e., the end of day t+1). On the
morning of day t+2, each CLS member is required to make payments on its short
positions. For the moment we will assume that the payments required are equal to the full
value of the short position; we will consider more complicated cases later. Table 2
illustrates the process. The initial position on date t+1 is the same as in the previous case.
Each bank begins by making a payment of its short position to CLS. These payments are
made on the central bank RTGS systems. Once CLS has both currencies available to it,
settlement is effected on the books of CLS, by moving funds between accounts. After
that, the currencies can be sent out to the banks.

As example 1 shows, the key feature of the new arrangement is that there is never
a point at which one leg is settled and the other is not. In this sense, CLS is a paymentversus-payment (PvP) system. Under CLS, final settlement of each side of a transaction
is simultaneous and mutually conditional. Under the current arrangement there is an
instant where one bank (Bank A in Table 1b) has paid out funds to its counterparty but
not received funds in return. At this point, Bank A is a creditor of Bank B and therefore
vulnerable to a failure by Bank B. By contrast, at no point in the process in Table 2 is
either bank a net creditor of the other. Under CLS, if Bank B fails at any instant before
settlement, the transaction does not go through, and the funds paid in by Bank A are

10

returned to Bank A. If Bank B fails at any instant after settlement, it is of no concern to

Bank A. After settlement, Bank A has no remaining financial links with the failed
institution, all such links having been transformed into links with the CLS Bank.
Of course at that vulnerable point Bank A is instead a creditor of CLS until the
time when CLS sends the payments to Bank A. For this situation to be an improvement
over current arrangements, it must be the case that the CLS bank is a better counterparty
than any individual bank. But this is indeed the case. In this simple example, the CLS
Bank is never the creditor of any bank and therefore is invulnerable to failures of other
banks. Note the importance to the CLS arrangement of the finality of payments on RTGS
systems. Since the CLS Banks assets are simply good funds, not due froms, they
dont disappear in the event of the disappearance of the bank paying them in.
The extreme safety of the arrangement described in example 1 stems from the fact
that full pay-in by both sides occurred before settlement. In some circumstances, the CLS
proposal will allow settlement to occur before all net funds have been paid in.12 Example
2 examines the case where settlement takes place with only a small initial pay-in.
<Insert Table 3 about here>
Example 2. As before, but suppose that initially Bank B pays in $1 million or 10%
of its due-to position in its short currency, and Bank A pays in a corresponding amount:
10 million (changes in bank balance sheets are indicated in Table 3). As before,
12

Payment, on the other hand cannot be completed until all participants pay-in all net amounts due in each

currency. (Again, care must be taken to distinguish between settlement and payment. In this context, from
the point of view of a participating bank, settlement means substitution of the CLS Banks due from
for the initial counterpartys due from. Payment means substituting central bank funds for the CLS
Banks due from.)

11

settlement occurs by transferring the required balances between the sub-accounts of the
two banks on the books of CLS: 100 million from Bank As yen sub-account to Bank
Bs yen sub-account and $1 million from Bank Bs dollar sub-account to bank As dollar
sub-account. Now, however, these transactions leave overdrafts in a sub-account for each
of the banks. Once the pay-ins are completed, the situation is the same as in the earlier
description, and pay-out can proceed safely. But until pay-in is completed, the system is
vulnerable again to a failure by either of the banks. For example, if Bank B fails before
completing its pay-in, CLS Bank will owe Bank A $1 million but will only have
$100,000 in good funds.

Although CLS permits member banks to have overdrafts during the day, it places
several sorts of limits on the overdrafts. A transaction will not be settled if it causes either
bank to exceed its overdraft limits; instead both legs of the transaction will be held in a
queue to await inflows of funds to the banks account.
Moreover, in order to handle the possibility of a failure by a bank with an
overdraft, the CLS arrangement includes a back-up provision. The CLS Bank will hold
lines of credit with a set of liquidity providers, who agree to provide the funds that CLS
needs to cover settled transactions. There are two features of the CLS arrangement that
should make it possible for the CLS Bank to obtain these lines of credit at small cost:
First, there is an upper limit on the amount of liquidity that ever needs to be provided,
namely the limit on the overdrafts permitted to the various banks. A transaction will not
be settled if it causes either bank to exceed its overdraft limits; instead both legs of the
transaction will be held in a queue to await inflows of funds to the banks account.

12

Second, although a bank is permitted an overdraft in individual currency sub-accounts, its

net position across all currencies in its account must be positive. Again, CLS Bank is
never in the position of being an overall creditor to any bank. Thus the event of the
failure of Bank B does not adversely affect the value of the CLS Bank.
These arguments demonstrate why the liquidity providers should regard the CLS
Bank as carrying no credit risk. But why should the CLS Bank be satisfied with the
reliability of its liquidity providers? Indeed, who are these liquidity providers? They turn
out to be the owners of the CLS Bankthat is, none other than the member banks
themselves. Might the protection offered by them be illusory? There are two counter
arguments: first it is the group of liquidity providers as a whole that provides protection
to the CLS bank against failure of any individual member. Second, the limits on
overdrafts under CLS, while explicitly protecting liquidity providers, may also be there to
convince any potential additional liquidity providersincluding in extreme situations,
central banksof the ultimate safety of any liquidity infusion.13 To the extent that the
participating banks regard the liquidity facilities of CLS as sound, and to the extent that
the relative value placed on currencies by the participating bank are identical to market
values, participant behavior will be unaffected by early settlement.

13

This discussion should also make it clear that the safety of the CLS Bank from the point of view of

liquidity providers depends on the maintenance of several underlying regulatory principles by the relevant
American regulatory bodies (recall that the CLS Bank will have a U.S. charter). First, all payments to or
from CLS Bank over RTGS systems of the various central banks must be regarded as final. Second, subaccounts at CLS of a failed bank must be treated as a single netted account. Finally, CLS can not be treated
as a source of strength to any of the banks owning it.

13

The principle that a banks net position over all currencies must be positive
becomes even more important once we take into account currency fluctuations. In the
absence of exchange rate fluctuations, settlement could begin before the pay-in of any
funds, without violating the principle that a banks net position at CLS must not be
negative. With exchange rate fluctuations, the out-of-the-money party (at least) must
make some pay-in before settlement can begin.14

Example 2, continued. Suppose that on day t+1 the value of the yen falls, so that
by the close of trading on day t+1, 100 is now worth only $.90, so that Bank Bs
position vis--vis Bank A is out of the money. CLS then calculates the day t+2
settlement schedule as follows. Bank B is short dollars so has a minimum initial pay-in
obligation of $1 million (.90$1 million) = $100,000 or 10% of its due-to position in its
short currency.15

So far we have dealt with a single payment. In fact pairs of participants in CLS
will make large numbers of exchanges during the day, repeatedly swapping currencies
back and forth in offsetting or near-offsetting trades.16 In such circumstances netting
arrangements are used to reduce the amount of currency needed to settle the trades. The
14

In other words, the requirement of a net positive balance plays much the same role as margin

requirements under marking-to-market in an organized futures market. See e.g., Baer, France, and Moser,
(1995), or Moser (2000).
15

In practice, an additional amount would have to be paid in to provide a cushion against additional

exchange rate volatility during day t+2.

16

Such patterns are common in foreign exchange trading. See e.g., Lyons (1997, 1998).

14

CLS system will also economize on the use of currency through a quasi-netting
arrangement that avoids some of the undesirable consequences of standard netting.

Example 3. Suppose that Bank A buys $2 million from Bank B for 200 million
during the first trade of day t, and then buys 100 million from Bank B for $1 million
during the second trade of the same day. As before, assume that the dollar rises to 100 =
$.90 by the close of trading on day t+1, so that each banks initial pay-in requirement
would be the same as in example 2. That is, at the beginning of day t+2, Bank B is net
short $1 million and long 100 million, so once again B would need to pay in $100,000.
When settlements now occur depends on how large an overdraft the two banks are
permitted. For simplicity we will assume that the overdraft each bank is permitted is
sufficient to handle each trade. Even so, settlement still cannot take place until the payins are sufficient to ensure that the trade leaves each bank with a net positive balance.
Since bank B is out of the money $200,000 on the first trade, that trade will not settle
until bank B puts a further $100,000 in its account. Once it does, the first trade will settle.
Although A is out of the money on the second trade, the settlement of the first trade
leaves the net position of A sufficiently positive to enable the second trade to be settled as
well.

Although the trades are settled, the CLS Bank still lacks the funds to make a
payout. These must await the pay-in of funds by each of the banks. As those funds
appear, payouts are made subject to two restrictions: 1) the CLS Bank can never

15

overdraw its account with any RTGS system, and 2) all settlement banks accounts with
CLS must remain net positive.
Note the ways that this arrangement is and is not like a netting arrangement. Like
netting arrangements, CLS economizes on the use of central bank funds. In the example
we have described, the payment process can be completed with each bank paying in only
its net position in the short currency. If we had assumed the banks faced more stringent
caps on their overdraft positions, the necessary pay-in could have exceeded this net
amount, but typically by smaller amounts as the number of transactions engaged in by
each bank becomes large.
Nonetheless, in its handling of settlements the CLS arrangement is not a netting
arrangement. Each trade is settled or not on its own. It would be possible for one trade
between the two banks to settle and (either because of subsequent bank failure, or
because of operational problems with a pay-in) for another trade not to settle. Payouts on
the settled trade become the responsibility of CLS Bank. On the unsettled trades, each
bank is returned its initial pay-ins.

3. Payment vs. Payment and Bilateral Incentives

The remainder of this paper uses a series of simple models to explore the
incentives on settlement banks induced by the proposed CLS arrangement as compared
with existing or potential alternatives. This section of the paper focuses on the basic
effects of a payment-versus-payment system. Appendix A describes the model underlying
this section.

16

Since CLS is billed as (and is) a method of reducing certain kinds of settlement
risks, there is a danger that it will be interpreted in terms of insurance. In fact, if
insurance were the goal, it would easy to concoct arrangements more effective than CLS.
It has become a commonplace as a criticism of certain payments arrangements that they
do not in fact reduce risks, but simply shift them from one party to another. The CLS
arrangement does do its share of risk shifting, but it also has interesting effects on
participants behavior. Indeed, CLS is in most respects the opposite of insurance: it
reduces the risks by penalizing the undesired behaviors.
We will examine the effect of the CLS system on several types of behavior. The
first, and most basic form of behavior we will examine is compliance: a banks
willingness to carry out the pay-ins of the amounts they have promised. In any payment
system with a prayer of survival, compliance must be the norm. The decision to cancel a
trade only occurs in extreme circumstances (financial distress, operational failures), since
the consequences are severe, including possible loss of future ability to trade. But these
are the sorts of circumstances that matter for the design of a system. The payments
system arrangements determine how extreme the circumstances must become before the
bank fails to comply. A social planner will likewise prefer that in some exceptional
circumstances the trade (or possibly one leg of the trade) not go through,17 although there
may be a difference between private and public calculations of when cancellation is
desirable.

17

As a response to Herstatt risk, Central Bankers are not generally keen on one strike and youre out

arrangements, according to Clementi (1999) implying that, at least in some extreme circumstances, nonpayment ought to be tolerated.

17

All of the models we consider have the following structure: Initially two banks
agree to make an exchange of currencies. Subsequently either party can cancel (default
on) the trade (what this means exactly will vary with the arrangement) but it is not
possible to modify the terms in any way.18 This seems a reasonable modeling of the
situation. There are fixed costs to each act of establishing a trade and to each act of
canceling a trade. Situations inducing firms to cancel their payments are rare and nonreplicable and central authorities have limited time to respond. Thus, it is natural to
assume that the mechanisms allow only a single message (cancel/dont cancel) from each
agent and consequently only four (2x2) possible responses to their behavior.
In this framework the fundamental difference between the current arrangement
and a payment-versus payment (PvP) system like CLS is this: under a PvP system, if a
bank does not pay in, it cannot receive the payments from the counterparty. Under the
current system, receipt of counterparty payments is independent of the pay-in decision,
because it depends only on the other partys decision. In other words, the CLS system
requires lower penalties to enforce the same degree of compliance. In the model of
Appendix A, compliance under CLS is achieved with zero penalty, while the current
system requires positive penalties. In more general situations, holding other penalties
constant,19 the CLS arrangement increases the cost of default and decreases the likelihood
of default.

18

Thus in terms of CLS default corresponds to a decision to cancel the trade after it becomes irrevocable

(midnight of the day after making the arrangement).

19

This includes any cost to reputation from a default decision. Implicitly we are holding these non-

pecuniary costs constant when comparing regimes. However, this assumption may not be valid: one of the

18

Now there will certainly be extreme circumstances in which a social planner

would prefer that trades not be carried out given some new information arriving at the
intermediate period. But in general we might expect that the central planners goal is to
make parties comply more readily. If system administrators would like to increase
compliance, but penalties are already as severe as possible, then the CLS arrangements
offer a means to improve the situation, by increasing compliance without further
increasing penalties. Similarly, if penalties are socially costly to administer, then the CLS
arrangement allows the same outcomes to be enforced at lower cost.
Moreover, the model in the appendix shows that for a range of penalties, the
current system is subject to coordination failure to which the CLS system is immune. In
other words, there can be multiple equilibria to the current arrangement: expectations that
counterparties will fail to honor their commitments can be self-fulfilling. By making
payments conditional on counterparty behavior, a PvP system reduces this problem.
On the other hand, the CLS arrangement may not be superior to the current
arrangements in terms of risk sharing or optimal allocation of currencies. In the case of a
unilateral default,20 under the current system the counterpartys payment goes through;
under the CLS system it is stopped. Those considerations might lead a central planner to
favor an arrangement that continues to send payments to banks in financial distress or
under operational problems, even if those banks are unable to send payments in return.

effects of CLS may be to make any failure of pay-in more public and thus more costly to reputation (we are
grateful to seminar participants at Norges Bank for pointing this out).
20

It should be noted that the systems also differ in the case of a mutually agreed upon rescission of a

transaction: CLS permits this (CLS Services, 1998, p. 13) , but there is no way to allow for it without
penalty in the current uncoordinated systems.

19

In the model, the relevant concern in ranking the two systems is the social costs of
penalties under current arrangements, versus the ex post cost of misallocations under
CLS.
Individual banks could conceivably prefer the risk sharing arrangements under the
current system: True, the CLS system affords insurance against counterparty risk in sense
that, if the counterparty defaults, the arrangement will not go through. But in the long run
a bank could be either the counterparty of a distressed firm or the distressed firm itself.
The symmetry of the situation could make it possible that the current system offers better
insurance.
In fact, however, it is unlikely that banks will prefer the implicit insurance
embodied in current arrangements. First, limited liability means that a firm that is already
in the extreme situation of contemplating defaulting on payments is likely to behave from
that point on in a risk loving, rather than a risk averse fashion. A firm is unlikely to desire
what is in effect small amounts of ex ante income insurance over such extreme states.
Second, even if this insurance were desired, any effects of such a system providing
insurance would likely be swamped by the adverse selection characteristics of the
arrangements. Once in place, a CLS like system is likely to cream skim the best risk
categories, as financial institutions regarding themselves as relatively unlikely ever to
choose unilateral default migrate to the system that penalizes them least in the case of a
counterparty default.21
21

Finally, to the extent that the participants in the system themselves bear the enforcement costs then the

reduction in these costs for similar performance levels will encourage agents to switch to the new system.
On the other hand, to the extent that enforcement costs in public arrangements are publicly subsidized,
there will be resistance to migration.

20

In summary, CLS represents an improvement over the status quo because it

provides a mechanism for coordinating settlement. Coordination through PvP reduces the
need for penalties: it increases the penalty to the defaulting party and reduces the penalty
to the party defaulted against.
Although the effects on banks incentives to default in a crisis are the
fundamental feature of the new system, it is also of considerable practical importance to
understand the incentives CLS places on a bank to avoid such crises in the first place.
The CLS arrangements will impose direct restrictions and requirements on its member
banks for managing various sorts of risks, including operations risks. But it is also
worthwhile to determine the extent to which the system itself encourages or discourages
risk taking by participants.
A simple extension of the model shows that, to the extent that a PvP system puts
additional penalties on a bank which does not make its pay-in, the CLS system
encourages banks to reduce risks of operational problems or financial distress. On the
other hand, since a PvP system reduces the costs of dealing with unreliable
counterparties, it encourages banks to take on trades with other banks who would have
been excluded from trading under current arrangements. In a two-trader world this is of
little concern, since the bank can gauge the costs and benefits of dealing with less reliable
counterparties. In a multiparty world, negative effects from the relaxation of standards
may be concentrated on third parties. Appendix B outlines an extension of the basic
model to illustrate this possibility.
While the design of the CLS arrangement may give its participants an incentive to
relax their standards for monitoring counterparty quality, there will still be an interest in

21

maintaining standards for the system as a whole. In the CLS structure, the responsibility
for scrutinizing the quality of counterparties falls on the CLS bank itself, which sets
standards for membership. Current standards are rather stringent, e.g., a member must
have Tier 1 capital in excess of $1 billion and a BBB+ or better rating of its long-term
debt. The question of membership standards is a vexing one for regulators: on the one
hand they too have an interest in maintaining the quality of counterparties so as to
minimize the chances of settlement disruption. On the other hand the CLS bank may be
interested in setting standards higher than those which an efficiency analysis will pose,
since restriction of membership will (assuming CLS is successful) be a way of enhancing
the profits of the members.

4. CLS and Multilateral Incentives

In this section we describe several additional multilateral examples illustrating
other incentive effects of the CLS arrangement. We begin with the simplest case, where
burdens of payment failure are shared between the counterparty and other uninvolved
traders. The more complex case, where a failure between one pair of traders triggers a
second failure, is dealt with only in a preliminary fashion.

Burden Sharing
Our examination of the sharing of burdens of counterparty failure is designed to
show the differences between CLS and two possible alternatives: a true netting system
and a pure gross settlement system with PvP. By true netting we mean a system

22

incorporating netting by novationthat is, one where original obligations are legally
replaced by net obligations.22
So far we have emphasized how the CLS system resembles a netting system
through its economizing on use of currency. It is also important to understand how it
differs from a true netting system. There are at least three aspects to this difference. First,
in the event of default on one leg of a transaction, the other leg ceases to be an obligation
under CLS. In a true netting arrangement, the link between performance in specific pairs
of obligations is lost in the new replacement arrangements. Second, CLS avoids the
regulatory costs (collateralization requirements and position limits) associated with true
netting under the Lamfalussy standards.23 Finally, CLS differs from the netting
arrangements found in many organized futures exchanges, where obligations originally
described in terms of delivery of a variety of specific items can all be reduced to and
settled for dollar equivalents. This last difference is closely related to the lack of a true
spot market for foreign exchange, so we will confine the present discussion to withincurrency netting arrangements. The issue of cross-currency netting is briefly addressed in
the concluding section of the paper.
Suppose Bank A contemplates trading with a counterparty B with a known
probability of failure. Under a pure PvP system, if the trade fails, the principal is
returned. Under a true netting arrangement, all trades are aggregated, and any failure is
shared among the surviving parties, through paid-in collateral. Although some
arrangements put an additional penalty on the counterparty of the failed institution, in

22

Under current arrangements netting by novation is only possible within-currency.

23

The Lamfalussy standards are described in Bank for International Settlements (1990).

23

general netting arrangements will penalize the counterparty less than a pure PvP system
would.
The CLS system would be in effect a pure RTGS system with PvP except for the
presence of the overdraft privileges. Because of these privileges, it is possible for an
institution to fail after its transaction has settled but before its payment has arrived. In this
case the CLS Bank, through its liquidity providers and ultimately its members, bears the
costs of the liquidity problem, thereby reducing the cost borne by the counterparty
relative to the pure system.
But how do the costs to the rest of the participants under CLS compare with those
costs under a pure netting system? That depends on the constraints put on activity in a
netting system, but unless the constraints were infeasibly tight, costs to others will be
smaller under CLS. First of all, the CLS arrangement limits the liquidity cost to the
system as a whole through its limits on overdrafts. A netting arrangement could maintain
an equivalent limitation only by imposing a tight limit on positions and a strict collateral
requirementin each case, probably tighter than current netting systems achieve.
To summarize, a pure PvP system encourages trading with risky counterparties
more than the current system does. However, since it concentrates the cost of a
settlement failure on the individual counterparties, a pure PvP system gives less
encouragement than does a true netting system. The CLS system lies between PvP and
true netting: the overdraft privileges speed up settlement and reduce the need for
currency, but they deflect part of the burden of counterparty default to the system as a
whole.

24

Ex Post Inefficiencies
As we have seen in the two-bank case, the CLS arrangement is more likely to lead
to an ex post inferior allocation of currencies, since it stops both legs of a trade. Our next
example shows that in multilateral settings this phenomenon is exacerbated, leading to a
larger role for central banks in solving ex post coordination problems under CLS.
The traders decide whether to trade in period 0 and whether to default in period 1
under CLS rules. For simplicity we deal with the case of no penalties for default. The
central bank does not trade with any bank initially, but may intervene (supplying liquidity
ex post) in period 2 if it so wishes. The goal of the central bank is to maximize ex post
social welfare, defined as the sum of traders utilities. In the example we treat dollars as
numeraire, with the same value to all holders. Other currencies ex post will can have
different possible values to the bank trading them. With a probability near one they will
have either a low marginal value L<1 or a high marginal value H>1 (in each case,
relative to dollars). These values are known ex ante, and determine the pattern of planned
trade. However with a small probability, these marginal values will be discovered at date
1 to be different, resulting in changes in the desire to trade. Specifically, we consider the
consequences of individual firms facing an idiosyncratic shock to its valuations. One
natural interpretation of such a shock is financial distress for the individual firm.
There are three traders A, B, & C. Each buys a foreign currency with U.S. dollars
and sells another foreign currency for dollars, so each trader has a zero net position in
dollars (not an unreasonable scenario since 80-90% of FX trades have a dollar leg). All
traders have a MU of dollar consumption equal to unity. Suppose all trades are one for
one.

25

Trader A buys euros from B and sells pounds to C; trader B buys Swiss francs
from C and sells euros to A; trader C buys pounds from A and sells francs to B. Initially
each trader believes that he will have marginal utility H for the bought currency and L for
the sold currency. Suppose, however, that after the fact it turns out with small probability
that any firm finds that its valuations are reversed. These probabilities are independent for
each of the firms. For example if it is firm C that suffers from this shock, C has MU of H
for francs and L for pounds and defaults on his dollar settlement to A and franc settlement
to B. Suppose also that A is liquidity-constrained in dollars, so that without the dollar
settlement from C, A cannot settle the trade with B.
Note that after the shock to C the original sequence of trades still has positive
social value equal to H-L = 2(H-L)-(L-H). Thus there is an incentive for the central bank
to step in and settle Cs trades, by supplying francs and selling pounds. Even if the central
bank places values on these currencies that are the same as Cs valuations (the worst
possible case) and therefore suffers a loss of H-L on the settlement, the total benefit
increases.
The basic problem here is that C cannot renegotiate with A and B, once he realizes
that he has no incentive to complete the original trade. Since the overall gains to trade are
positive, presumably A and B would be willing to give up some of their gains from trade
in order to compensate C for his loss. Since there is no practical way of doing this, the
central bank finds it optimal to step in and take the loss after the fact. Now if the reversal
occurs instead for A or B, the result is the same. Since simultaneous reversals by two
banks are second order small, this completes the analysis.

26

Note that this phenomenon requires at least three traders. Consider the analogous
situation with two traders: A and B. A buys euros from B and sells pounds to B (if desired,
each leg of this trade can be accompanied by a flow of dollars in the reverse direction).
Suppose that As valuations now reverse, so that it is unwilling to complete its trades. The
two banks now agree on the valuation of all currencies. If the central bank intervenes to
complete the trade with B, and its valuations are identical as well, no welfare gain
arises.24
In contrast, netting arrangements have an advantage in such circumstances (at
least from the point of view of a central bank), since they spread the trades themselves
and the costs of failed trades on a pro-rata basis, implying a diminished need for central
bank intervention to complete desirable trades.

Exchange rate fluctuations

In this section we consider the effects on CLS of exchange rate fluctuations. We
use the same example as the previous section, only changing the details of the small
probability shock. Suppose the shock takes the following form: the value of a currency
(say, pounds) to all traders falls in a proportionate fashion. Specifically, suppose the
value of pounds to C falls from H to L, (and falls from L still lower for A). Note that it
still remains socially desirable to have all trades consummated, since currencies then end
up in the hands of traders who value them most. However, while C still wishes to
24

Of course this specific description depends on the fact that the currencies can only take on two

valuations. But in a more general formulation with a multiplicity of possible valuations by the traders, the
difference between the two and three bank cases continues to hold if we assume the natural (conservative)
valuation of currencies by the central bank.

27

consummate the franc/dollar trade with B, he wishes to cancel the pound/dollar trade with
A.
The first case to consider is the one where bank C still prefers consummating both
trades to consummating neither.25 The outcome then depends on the extent to which the
two trades are tied.
Under current arrangements, each currency would be transacted on a different
facility. The dollar legs of the transaction, once entered into CHIPS, would be subject to
netting. However, the two other currency transmissions, going on separate systems,
would be subject to the coordination failures described in the earlier sections.
Complete integration would tie all transactions to each other so that they stand or
fall as a group. Doing so increases the likelihood of compliance by a participant, since in
the money transactions support the traders willingness to take the out of-the-money
transactions. CLS achieves this by making it impossible to selectively repudiate
transactions after midnight on day 1.
But there are additional layers to the CLS arrangement which are designed to
increase compliance. The pay-in requirements on the short positions are adjusted to
reflect the current exchange rate, and to ensure that as the day goes on, the value of the
participants net short position is limited. If the participant values all currencies at their
market values, then the CLS arrangement guarantees that the account is in the money
so that he has no desire to default. The pay-in amounts thus provide additional protection
of the system against adverse information that the participants could acquire during the
course of the settlement day. The arrangement therefore represents a trade-off between

25

The alternative assumption reduces the situation to that described in the previous section.

28

the benefits of economizing on pay-ins through the quasi-netting behavior of recycling

reserves, and increasing pay-ins to ensure compliance.

5. Conclusion
A characteristic feature of modern payment systems is their capacity to
economize on the use of outside money. Through the use of netting and similar
techniques, banks and other financial intermediaries can, to a large degree, allow inside
money to substitute for outside money in market transactions.26 The introduction of CLS
will open a new avenue to such substitution, by allowing for simultaneous, coordinated
settlements in inside funds across different currencies.
Previous descriptions of CLS have tended to focus on the mechanics of the
arrangements for payment and settlement. In this paper we have concentrated on a more
limited set of aspects in order to focus on participants incentives under CLS. The
payment versus payment arrangement eliminates the risk of loss of principal. The CLS
system is not, nor has it been advertised as, a system designed to eliminate the liquidity
risk associated with settlement failure. Nonetheless we have shown how the arrangement
reduces these risks by encouraging compliance by participants. We have also examined
how specific provisions of the arrangement encourage its use relative to existing
uncoordinated arrangements.

26

Green (1999, 2000) has suggested that many innovations in payments systems can best be thought of as

an increase in such substitution, as measured by an aggregate netting ratio (the ratio of gross to net
transactions within an economy).

29

The idea that I will give you this valuable object if and only if you give me that
valuable object in return is the essence of what we mean by a spot trade. It is this
conditionality, not mere simultaneity, which is the essential feature of CLS. And it is the
lack of arrangements for such spot trades in foreign exchange markets which has made
them special.
A promise of dollars or pounds from any private party is a slightly different good
from dollars or pounds themselves. A currency is good as long as the central bank
survives. A promise of a currency from a private party is the same as the currency only in
states where the private party honors its promise. Thus in the rarefied world of foreign
exchange trading, where even the remote probability of a bank default is relevant to
pricing and trade, it becomes important to find a way to exchange pounds for dollars, not
just for the mere promise of currencies. But this poses a problem. For the only place that
large denominations of a currency exist is on the books of that central bank. Thus
coordination of funds transfers would require a higher degree of communication and
cross-border cooperation between central banks than present arrangements provide.
We conjecture that if such a spot market existed (imagine for a moment a
medieval money changer transported to a stall in present-day London, with billions of
dollar, yen and euro notes piled high on his bench), then the additional innovations
corresponding to CLS would be relatively minor, and could mirror instead the standard
repertory of arrangements in organized futures exchanges, with the posting of margin
collateral on netted positions. A similar conjecture might apply if much of todays foreign
exchange trades were to be replaced with trades in contracts for difference. A contract
for difference requires one party to pay the other a payment equal to the home-currency

30

value of the market gain or loss that would have resulted from a foreign exchange
transaction (see Bank for International Settlements 1998).
Clearly incentives are fundamental to settlement risk, and central to a regulators
evaluation of a settlement system. And the incentives associated with the decision of
whether or not to pay up are clearly the most fundamental incentives that there are,
which, together with their simplicity, justifies focusing upon them in this initial analysis
of CLS. However, other incentives must also be considered in examining settlement risk:
we have briefly considered the inducements that a system gives for risk taking
behaviorin particular the extent to which they encourage or discourage increasing the
magnitude of foreign exchange activity for a bank, and the degree to which they
encourage or discourage monitoring of counterparty quality.
We have also considered in a preliminary fashion, the effects of the CLS
arrangement on incentives in a multiparty context. We found that the CLS system lay
between pure PvP and netting arrangements in its encouragement of trading with risky
counterparties. Compared to netting arrangements it leaves a greater role for ex post
intervention by Central Banks to complete uncompleted trades. And the ability of the
arrangement to hold completion of some trades by a bank hostage to the completion of
other trades also serves to increase compliance. Although these are significant
implications of the new arrangement, our analysis has clearly only scratched the surface
in investigating CLS in a multiparty context.

31

Appendix A: A Basic Two-Bank Model

At stage 0, each bank can agree to enter a proposal to make a trade. If both banks
agree, then the game begins, and there are two stages to the banks activities. At stage 1,
each bank can choose to make an investment27 at a cost k. If the investment is made, then
one indivisible unit of initial asset arrives with probability (1 p). If the investment is not
made, then there is zero probability of the assets arrival. The agreement to trade
mandates that each bank deliver its initial asset to the other bank, in return for its more
preferred asseti.e., the initial asset of the other bank.
The stage 2 decision for a bank is whether to carry out the agreement by
delivering the initial asset, or to renege on the agreement by refusing to deliver. (If the
asset did not in fact arrive, then delivery is impossible).
Depending on these actions, and on the rules for handling clearing, a bank can end
up holding its initial asset, its preferred asset, both, or neither. The following equations
define the parameters associated with each holding, H(x,y), where x represents the
holding of the initial asset, and y represents the holding of the preferred asset:

H(0,0) = 0

(1)

H(1,0) = c

(2)

H(0,1) = b

(3)

H(1,1) = a

(4)

where
27

For some applications it is more natural to consider these investments as reliance investments taken

before the agreement is made. For the issues in this appendix, however, no substantive difference arises,
and technicalities are avoided by reversing the order.

32

a > b > c>ab

(5)

A banks total payoff is the sum of H, investment costs, if any, and penalties assessed by
the clearing mechanism, if any.
The current arrangements for clearing and settlement can be described as
follows: each party makes a unilateral decision to deliver. A fine is imposed on a party
failing to deliver. A PvP system can be described as follows: each party makes a
unilateral decision to deliver. If one party chooses not to deliver, then neither asset is
delivered. For completeness, we include the possibility of a fine assessed for choosing
not to deliver in a PvP system as well; let fc and fp represent the fines under current
arrangements and the PvP system, respectively. Then Table 4 describes payments as a
function of various consequences in the subgame after investment has already been made.
<Insert Table 4 here.>
We will assume that the social costs of a penalty are equal to the private costs, so
that the social value of the arrangement is the sum of the payoffs to the two banks.28 Thus
it is ex post always socially beneficial for a bank to deliver when (and of course, only
when29) it receives the initial asset. Also we assume
(1-p) b > k

(6)

so that investment is socially valuable if the asset is then transferred.

28

I.e., there is complete dissipation of penalties paid by a defaulter.

29

Non-arrival is simply short hand for a circumstance where delivery becomes prohibitively costly, both

privately and socially.

33

Once the two banks have entered the agreement, there are three pure strategies in
the full game: 1) Invest and Deliver, 2) Invest and Renege, and 3) Dont Invest. The
payoffs are therefore those given in Table 5.
<Insert Table 5 here>
Elementary calculations demonstrate:
Lemma 1: If
fc > (1 p)(a b) + p c

(7)

fc > k /(1-p)

(8)

and

Then there is a Nash equilibrium under current arrangements in which each party invests
and delivers. Moreover, given these assumptions, the Nash equilibrium is the unique pure
strategy equilibrium if and only if
fc > c

(9)

Lemma 2: Under the PvP system, it is a Nash equilibrium for each party to invest and
deliver for any non-negative fp . Moreover, if fp > 0 the equilibrium is unique.

Thus the PvP system can be implemented without penalties, while current
arrangements require positive penalties. Moreover provided
c > k / (1-p)

(10)

there is in general a range of penalties over which there will be multiple equilibria under
current arrangements.
If we assume instead that

34

c < k / (1-p)

(11)

then equilibrium is unique under both current arrangements and under PvP. Under this
assumption it is not socially valuable to invest unless the asset will be transferred to the
other party. Thus the social value of the current arrangement is
2(1-p) b 2 p fc

(12)

while the social value of a PvP system (assuming zero penalty) is

2(1-p)2 b +2 p(1-p) c

(13)

Since the minimum feasible penalty under the current system is

Min {(1 p) (a b) + p c, k /(1 - p)},

(14)

the question of which system dominates depends on parameters. As an illustration, let p

approach 0. Then if b-c > k, the current system dominates. If the inequality is reversed a
PvP system dominates. Roughly speaking, the question boils down to the issue of
whether the social costs of the penalties necessary under the current system to maintain
incentives are greater than or less than the social costs of the ex post misallocation of
assets under PvP.

Appendix B: A Three-Bank Extension (outline)

Label the three banks A, B, and C. Banks A and C are as described in the previous
appendix with the following exceptions: Bank A is perfectly reliable that is, bank As
initial asset arrives with probability 1. We will continue to use 1-p for the probability that
bank Cs initial asset will arrive. However, p is now stochastic, with probability
distribution F(p). The realization of p is observed by C, but not by A. For concreteness,
we will call As initial asset dollars and Cs initial asset yen.

35

Clearly in a two-bank world trade between A and C will frequently break down: A
will not be willing to trade if his expectation of the value of p is too high. To facilitate
comparison, we will assume that the fine under the current arrangements is set so that the
same types of C are willing to trade in the bilateral case under the current arrangements
as under the PvP arrangements.
Now suppose there is an intermediary bank which observes the probability p. This
bank makes its profits by facilitating trade between the other two, but it is not wholly
reliable, either. Specifically, we assume that with probability 1-q bank B receives an
initial allocation of a unit of dollars and a unit of yen, and with probability q it
receives neither. These probabilities are independent of all probabilities for bank C. The
payoffs to bank B based on its final holdings are:
For holding 1 unit of yen and 1 unit of dollars: v
For holding one unit of either: w
For holding two units of either: 2w
For holding nothing: 0.
We will ignore the incentives for B to renege (while they could be included, they
are not the point of this section and would only increase notational clutter). Bank B will
receive a payoff in addition to payoffs for final holdings, from the profits h it collects on
each consummated trade.30

30

It would be feasible, but not particularly instructive, to keep this section consistent with the previous

section by increasing the number of assets in the model and folding h into the payoff for particular asset
holdings by B.

36

The trades that will be worth considering under any arrangement are
arrangements for bank B to buy dollars from A and yen from C. Bank B will be able to
carry out its side of these arrangements if it receives its initial holdings or if its
counterparties do, so that it can use the proceeds of each to satisfy the others
requirements. We assume that
v > 2w + h

(15)

so that bank B prefers to enter trades only under the anticipation that each side of the
trade will settle successfully.
The payoff to bank B is the same under the current system as it is under PvP, as
long as bank C pays up. However, under current arrangements it is not possible for a
bank to monitor whether trades have gone through before commencing its own trades.
Thus exchanges will go through unilaterally as soon as the funds are available. For this
reason, bank B faces greater chance of losses under current arrangements than under PvP.
Specifically, compare the outcomes when bank B receives its initial assets but bank C
does not. Under current arrangements bank B sends assets to bank C but receives nothing
in return, while the trade with bank A is consummated. Thus the payoff to B under the
circumstances is
w + h.

(16)

Under PvP, bank B does not send assets to bank C but continues to consummate the trade
with bank A. Thus the payoff to B under the circumstances is
2w + h.

(17)

Thus the outcome is worse for bank B under current arrangements than under PvP.

37

(If both B and C fail to receive initial assets, then under both current arrangements
and under PvP, bank B will end up with no assets and with no trade consummated; thus
the only difference between the two will be in terms of the size of penalties assessed of B
under the two systems. Under CLS no specific penalties would be attached to these
circumstances. In any event, as long as the penalties are greater under the current system
than under CLS, then bank B is also worse off in this case under PvP than under current
arrangements and the rest of the discussion in this section is unaffected.)
In summary, bank B will choose to make arrangements with both bank A and
bank C or with neither of them. Which it will do will depend on the quality of bank C.
Under a PvP system, bank B will be willing to make arrangements when bank C has a
higher realization of p than under the current arrangements. Let Epc and Epp be the
expectation of p in the distribution F conditional on bank Bs accepting a trade under,
respectively the current arrangements and a PvP system. Thus,
Epc < Epp

(18)

We now turn to a consideration of the values of the two arrangements from the
point of view of bank A. If the trade is successful, As payoff is b-k. If the trade fails, As
payoff is c-k under PvP and k under the current arrangements. Under either regime, the
trade will fail when both B and C have no assets, so the probability of failure is
(Ep) q

(19)

and A will participate if

(b-k)[1-(Ep) q] - (Ep) q m > 0

(20)

Ep = Epc or Epp

(21)

where

38

and
m = k or k - c

(22)

according to whether the market operates under the current arrangements or PvP
respectively. Under PvP, the principal risk is smallerthat is, the amount lost when a
counterparty defaults is smaller. But the liquidity risks are greaterthat is, counterparties
engage in trades with lower quality third partiesmaking shortfalls more likely.
The effect of the regime switch on As payoff is ambiguous: Depending on
parameters, bank A will be willing to participate in both regimes, in either regime or in
neither.

39

References
Baer, H.L., V.G. France, and J.T. Moser, 1995. Determination of collateral deposits by
bilateral parties and clearinghouses, in: Proceedings of the 31st annual conference
on bank structure and competition (Federal Reserve Bank of Chicago, Chicago).
Bank for International Settlements, 1990, Report of the committee on interbank netting
schemes of the central banks of the group of ten countries (Bank for International
Settlements, Basel).
Bank for International Settlements, 1992, The insolvency liquidation of a multinational
bank (Bank for International Settlements, Basel).
Bank for International Settlements, 1993, Central bank payment and settlement services
with respect to cross-border and multi-currency transactions (Bank for
International Settlements, Basel).
Bank for International Settlements, 1996, Settlement risk in foreign exchange
transactions (Bank for International Settlements, Basel).
Bank for International Settlements, 1997, Clearing arrangements for exchange-traded
derivatives (Bank for International Settlements, Basel).
Bank for International Settlements, 1998, Reducing foreign exchange settlement risk: A
progress report (Bank for International Settlements, Basel).
Bank for International Settlements, 1999, Central bank survey of foreign exchange and
derivatives market activity 1998. (Bank for International Settlements, Basel).
Clementi, D., 1999, Keynote address at S.W.I.F.T. U.K. regional conference, 25 May.
CLS Services Limited, 1998. An introduction to continuous linked settlement. Mimeo,
October.
Davidson, C., 1996. Settlement saga rolls on. Risk 9 (November), 49-52.
Duncan, G.M., 1994, Clearing house arrangements in the foreign exchange markets, in:
Symposium proceedings: international symposium on banking and payment
services, (Board of Governors of the Federal Reserve System, Washington, D.C.)
168-174.
George, E., 1994, International banking, payment systems, and financial crises, in:
Symposium proceedings: international symposium on banking and payment
services, (Board of Governors of the Federal Reserve System, Washington, D.C.)
73-79.

40

Glass, G.R., 1994, Multinets FX netting solution, in: Symposium proceedings:

international symposium on banking and payment services, (Board of Governors
of the Federal Reserve System, Washington, D.C.) 152-167.
Green, E.J., 1999, We need to think straight about electronic payments, Journal of
Money, Credit, and Banking 31, 668-670.
Green, E.J, 2000, Central banking in an information age, mimeo, February.
Lyons, R.K., 1997, A simultaneous trade model of the foreign exchange hot potato,
Journal of International Economics 42, 275-298.
Lyons, R.K., 1998, Profits and position control: A week of FX dealing, Journal of
International Money and Finance 17, 97-115.
Lyons, R.K., forthcoming, The microstructure approach to exchange rates (M.I.T. Press,
Cambridge, Mass.).
Marjanovic, S., 2000, CLS forex settlement system seen gaining industry nod, American
Banker, March 10.
Moser, J.T., 2000, Origins of the modern exchange clearinghouse: A history of early
clearing and settlement methods at futures exchanges, in L.G. Telser, ed.: Classic
futures: lessons from the past for the electronic age (Risk Publications, London)
277-320.
Nol, T.E., 1994, Report on central bank payment and settlement services with respect to
cross-border and multicurrency transactions, in: Symposium proceedings:
international symposium on banking and payment services, (Board of Governors
of the Federal Reserve System, Washington, D.C.) 123-132.
Roscoe III, D.L., 1998. Continuous linked settlement, in: Proceedings of the 34th annual
conference on bank structure and competition, (Federal Reserve Bank of Chicago,
Chicago) 32-37.
Sesit, M.R., 1996, Central banks issue warning on trading, Wall Street Journal, March 27.

41

Table 1. Settlement of Spot FX Trade: Current System

Bank A Purchases $1M from Bank B for 100M
1a. After trade, before settlement:
Bank A
Assets
+ $1M due from Bank B

Liabilities
+ 100M due to Bank B

Bank B
Assets
+ 100M due from Bank A

Liabilities
+ $1M due to Bank A

1b. After settlement of one leg

Bank A
Assets
-100M in CB funds
+ $1M due from Bank B
Bank B
Assets
+100M in CB funds

Liabilities

Liabilities
+ $1M due to Bank A

1c. After settlement of second leg

Bank A
Assets
+ $1M CB Funds
-100M CB Funds

Bank B
Assets
- $1M CB Funds
+100M CB Funds

Liabilities

Liabilities

42

Table 2. Settlement of Spot FX Trade: CLS With Full Pay-In

Bank A Purchases $1M from Bank B for 100M
2a. After trade, before settlement:
Bank A
Assets
+ $1M due from Bank B

Liabilities
+ 100M due to Bank B

Bank B
Assets
+ 100M due from Bank A

Liabilities
+ $1M due to Bank A

CLS Bank
Assets

Liabilities
0

2b. After pay-in

Bank A
Assets
+ $1M due from Bank B
+ 100M due from CLS Bank
- 100M CB Funds
Bank B
Assets
+ 100M due from Bank A
+ $1M due from CLS Bank
- $1M CB Funds
CLS Bank
Assets
+ 100M CB Funds
+ $1M CB Funds

Liabilities
+ 100M due to Bank B

Liabilities
+ $1M due to Bank A

Liabilities (Accounts)
Currency Sub Accts.

$
Bank A 100M
Bank B
$1M

43

2c. After settlement

Bank A
Assets
+ $1M due from CLS Bank
- 100M CB Funds
Bank B
Assets
+ 100M due from CLS Bank
- $1M CB Funds
CLS Bank
Assets
+ 100M CB Funds
+ $1 M CB Funds

Liabilities

Liabilities

Liabilities (Accounts)
Currency Sub Accts.

$
Bank A
$1 M
Bank B 100M

2d. After pay-out

Bank A
Assets
+ $1M CB Funds
- 100M CB Funds
Bank B
Assets
+ 100M CB Funds
- $1M CB Funds
CLS Bank
Assets

Liabilities

Liabilities

Liabilities
0

44

Table 3. Settlement of Spot FX Trade: CLS With Delayed Pay-In

Bank A Purchases $1M from Bank B for 100M
3a. After trade, before settlement:
Bank A
Assets
+ $1M due from Bank B

Liabilities
+ 100M due to Bank B

Bank B
Assets
+ 100M due from Bank A

Liabilities
+ $1M due to Bank A

CLS Bank
Assets

Liabilities
0

3b. After initial pay-in

Bank A
Assets
+ $1M due from Bank B
+ 10M due from CLS Bank
- 10M CB Funds
Bank B
Assets
+ 100M due from Bank A
+ $0.1M due from CLS Bank
- $0.1M CB Funds
CLS Bank
Assets
+ 10M CB Funds
+ $0.1 M CB Funds

Liabilities
+ 100M due to Bank B

Liabilities
+ $1M due to Bank A

Liabilities (Accounts)
Currency Sub Accts.

$
Bank A 10M
Bank B
$0.1 M

45

3c. After settlement

Bank A
Assets
+ $1M due from CLS Bank
- 10 M CB Funds

Liabilities
+ 90 M Overdraft at CLS

Bank B
Assets
+ 100M due from CLS Bank
- $0.1M CB Funds

Liabilities
+ $0.9 M Overdraft at CLS

CLS Bank
Assets
+ 10 M CB Funds
+ $0.10 M CB Funds

Liabilities (Accounts)
Currency Sub Accts.

$
Bank A - 90M
$1M
Bank B 100M - $0.9 M

3d. After final pay in

Bank A
Assets
+ $1M due from CLS Bank
- 100M CB Funds
Bank B
Assets
+ 100M due from CLS Bank
- $1M CB Funds
CLS Bank
Assets
+ 100M CB Funds
+ $1 M CB Funds

Liabilities

Liabilities

Liabilities (Accounts)
Currency Sub Accts.

$
Bank A
$1 M
Bank B 100M

3e. After pay-out (same as Table 2)

46

Table 4: Payoffs in Settlement Subgame

4a: Subgame under current arrangements
And My Counter Party
Delivers

Reneges or Holds
No Asset
0

Deliver

Renege

a fc

c fc

Hold No Asset

b fc

fc

If I

4b: Subgame under PvP system

And My Counterparty
Delivers

Reneges or Holds
No asset
c

Deliver

Renege

c fp

c fp

fp

fp

If I

Hold No Asset

47

Table 5: Payoffs in the Full Game

5a: Payoffs under current arrangements
And My Counterparty

If I

Invest and
Deliver
Invest and
Renege
Dont Invest

Invests and
Delivers
(1-p) b p fc k
(1-p)2 a +(1-p) p c
+ (1-p) p b fc k
(1-p) b fc

Invests and
Reneges
p fc k
(1-p) c fc
k
fc

Doesnt
Invest
p fc k
(1-p) c fc
k
fc

5b: Payoffs under a PvP system

And My Counterparty

If I

Invest and
Deliver
Invest and
Renege
Dont Invest

Invests and
Delivers
(1-p)2 b + (p p2) c
p fp k
(1-p) c fp k
fp

48

Invests and
Reneges
(1-p) c k

Doesnt
Invest
(1-p) c k

(1-p) c fp
k
fp

(1-p) c fp
k
fp

Figure 1: Multilateral Settlement

A
GBP

EUR
USD

C
CHF
Ex Post Marginal Utilities
Example 1

Trader

USD

GBP

CHF

EUR

L<1

H>1

CB

Example 2
Trader

USD

GBP

CHF

EUR

L<L

49