BIS Working Papers

No 309
Toward a global risk map
by Stephen G Cecchetti, Ingo Fender and Patrick McGuire

Monetary and Economic Department

May 2010

JEL classification: F3, G01, G15, G21, Y1

Keywords: risk map, international banking, financial crises, yen

carry trade, funding risk
BIS Working Papers are written by members of the Monetary and Economic Department of
the Bank for International Settlements, and from time to time by other economists, and are
published by the Bank. The papers are on subjects of topical interest and are technical in
character. The views expressed in them are those of their authors and not necessarily the
views of the BIS.

Copies of publications are available from:

Bank for International Settlements
Communications
CH-4002 Basel, Switzerland

E-mail: [email protected]
Fax: +41 61 280 9100 and +41 61 280 8100
This publication is available on the BIS website (www.bis.org).

© Bank for International Settlements 2010. All rights reserved. Brief excerpts may be
reproduced or translated provided the source is stated.

ISSN 1020-0959 (print)

ISBN 1682-7678 (online)
 Toward a global risk map

Stephen G Cecchetti, Ingo Fender and Patrick McGuire 1

Revised Draft
May 2010

Abstract
Global risk maps are unified databases that provide risk exposure data to supervisors and
the broader financial market community worldwide. We think of them as giant matrices that
track the bilateral (firm-level) exposures of banks, non-bank financial institutions and other
relevant market participants. While useful in principle, these giant matrices are unlikely to
materialise outside the narrow and targeted efforts currently being pursued in the supervisory
domain. This reflects the well known trade-offs between the macro and micro dimensions of
data collection and dissemination. It is possible, however, to adapt existing statistical
reporting frameworks in ways that would facilitate an analysis of exposures and build-ups of
risk over time at the aggregate (sectoral) level. To do so would move us significantly in the
direction of constructing the ideal global risk map. It would also help us sidestep the complex
legal challenges surrounding the sharing or dissemination of firm-level data, and it would
support a two-step approach to systemic risk monitoring. That is, the alarms sounded by the
aggregate data would yield the critical pieces of information to inform targeted analysis of
more detailed data at the firm- or market-level.

JEL classification: F3, G01, G15, G21, Y1

Keywords: risk map, international banking, financial crises, yen carry trade, funding risk

1
Cecchetti is Economic Adviser and Head of the Monetary and Economic Department, Bank for International
Settlements (BIS), Research Associate of the National Bureau of Economic Research, and Research Fellow
at the Centre for Economic Policy Research; Fender is Special Adviser and McGuire is Senior Economist at
the BIS. This is a revised version of a paper prepared for the Fifth European Central Bank Conference on
Statistics “Central bank statistics: What did the financial crisis change?”. The authors would like to thank
Claudio Borio, Dietrich Domanski, Philippe Mesny, Philip Turner and Paul van den Bergh for comments and
conversations.

Toward a global risk map iii

 Contents

Abstract.................................................................................................................................... iii
1. Introduction......................................................................................................................1
2. Systemic risk: lessons from the crisis..............................................................................2
3. Elements of a strategy: lessons from the BIS banking statistics .....................................6
3.1 Monitoring banks’ funding vulnerabilities ...............................................................7
3.2 Measuring the currency carry trade–the Japanese yen .........................................9
3.3 Implications for statistical strategy: the BIS and beyond......................................12
4. Implications for global risk maps and similar tools ........................................................14
4.1 Supervisory information exchange.......................................................................14
4.2 Public information dissemination..........................................................................15
5. Conclusions...................................................................................................................16
Appendix.................................................................................................................................18
References .............................................................................................................................21

Toward a global risk map v

1. Introduction
Data are the eyes and ears we use to see and hear what is happening in the financial and
economic world. Anecdotes, introspection, personal experience and modelling can help us
figure out where to look and organise our thoughts. In the end, though, it is the data that tell
us what is going on. Without data, we are deaf, blind and, in both senses, dumb. The need
for data is especially acute in macroeconomic analysis for policy, where a bird’s eye
perspective is indispensable. Aggregate data on prices, output, employment, credit, money
and the like provide monetary and fiscal policymakers with the information that allows them
to monitor and meet their stabilisation objectives.
As the financial crisis deepened in the 18 months from August 2007, it quickly became
apparent that we were partially blind because of significant gaps in our statistics. In some
cases, we weren’t collecting the right data, and, in others, we weren’t using effectively what
we had. While it remains unclear whether improved analysis could have reduced the intensity
of the crisis, better and more timely information would almost certainly have simplified the
task of managing and containing the crisis as it evolved. Filling these gaps in the data is thus
an important challenge on our way towards improving our monitoring of systemic risk and
hence the regulation and supervision of financial institutions.
It is worth emphasising at the outset that, because of the global nature of the financial
system, systemic risk is a global problem. This means that our measurement and monitoring
of risk must have a global dimension, unless we wish to give up on the benefits of
globalisation and return to a financial system where institutions and markets are restricted to
operating inside national boundaries. As a result, and as will be apparent from our arguments
below, a clear need exists for increased sharing of international information and for
enhancement of the centralised analysis that is already pursued in a variety of organisations,
including the Bank for International Settlements (BIS).
Over the past year or so, the problem of data gaps has attracted an increasing amount of
attention. Numerous exercises are under way to identify and fill the gaps revealed by the
crisis. 2 This is an enormous challenge, and what can get lost are the big picture implications.
That is, implications going beyond any specific gaps in the data for a particular market or set
of institutions. Are there datasets or statistical concepts that hold lessons for how data should
be collected in the future? In this paper, it is our intention to step back and look for the
broader strategic lessons of the crisis that we can then use to guide us in developing specific
data requirements for identifying and measuring the accumulation of risks in the financial
system as a whole.
In this spirit, we proceed as follows. In section 2, we describe what we see as the primary
lessons for statistics that emerge from the crisis. High on the list is the need to see both the
entire balance sheets of financial institutions and the interconnections and common
exposures arising from them. In section 3, we present some recent work that builds on these
lessons. While our primary focus is on the BIS’s well known and widely used international
banking statistics, we also touch on the lessons the crisis holds for datasets beyond those
collected by the BIS. With this analysis in hand, in section 4, we turn to the problem of
building a global risk map, as well as the challenges this poses for the international exchange
of data and the need for a broader statistical reporting framework to support it. The final
section concludes.

2
See, for example, FSB and IMF (2009), which includes a list of 20 detailed recommendations for improving
data collection and analysis.

Toward a global risk map 1

2. Systemic risk: lessons from the crisis
Events during the fourth quarter of 2008 underscored the importance of understanding the
scale of risk exposures within and between large globally active financial institutions. We now
know that in the run-up to the crisis, much of the explosion in borrowing was underpinned by
cross-currency funding arrangements and large-scale maturity mismatches. While the trigger
for the crisis was common exposure to a particular class of assets, the US subprime
mortgages simply acted as the match that lit the fuse on the combustible mix of long-term
illiquid dollar-denominated assets financed by a combination of short-term dollar liabilities
and foreign currency swaps. When European banks in particular started to experience
losses, concerns about counterparty risk disrupted funding relationships and turned what
should have been a small brush fire into a global conflagration. 3
Looking back at this experience, we can say that the entirety of financial institutions’ balance
sheets – their assets and liabilities, their off-balance sheet positions, including lines of credit
and derivatives positions, and their counterparty arrangements – all played a role. The
challenge now is to find a way to monitor these factors. Doing so means both changing the
way we use existing data and changing the data that we collect.
To understand the problem, we find it useful to start with a set of core questions that
surfaced during the crisis. Every one of these is difficult (or impossible) to answer using the
data we currently have available. Here is a list of three big questions that help organise our
thinking:
Question 1: How can a central bank monitor the global use of its currency?
 If there were an international lender of last resort, how much of each currency would
it need to stockpile in advance?
 In the absence of such a lender, how large are the funding requests that I would
face in a crisis, and where would they come from?
 How large are speculative carry trade positions in my currency? Are these trades
contributing to volatility in my exchange rate?
Question 2: What information do policy makers need (in addition to home country
supervisory data) to monitor systemic financial stresses?
 How reliant on cross-border financing are my corporate borrowers? How reliable are
the lenders?
 How similar are the international portfolios of financial institutions headquartered in
my country to those of equivalent entities in other countries? Are there clustered
exposures or crowded trades?
Question 3: What data could help financial market participants improve market discipline?
 We now know that, before the crisis, European banks had long (and long-term) US
dollar positions they had funded using short-term FX swaps. Would euro-dollar swap
spreads have been 20 basis points wider in 2006 if the global asymmetry in the size
of these positions had been widely known and publicised?
 Would pricing for interbank funding be more efficient if participants could see the
build-up of system-wide funding stresses? If, for example, institutions had known the
extent of reliance on short-term wholesale sources of funds, would prices have been
different? How about concentrated asset exposures?

3
See BIS (2008), chapter VII, and BIS (2009), chapter II, for a detailed description of these events.

2 Toward a global risk map

 Box 1: Systemic risk—definition and measurement

Definition
Systemic risk in the financial system is analogous to pollution. It is an externality that an individual
institution, through its actions, imposes on others. As commonly understood, this externality takes
two forms. The first is the joint failure of institutions at a particular point in time resulting from their
common exposures to shocks from outside the financial system or from interlinkages among
intermediaries. The second is what has come to be known as procyclicality. This is the term used to
describe the phenomenon that, over time, the dynamics of the financial system and of the real
economy reinforce each other, increasing the amplitude of booms and busts and undermining
stability in both the financial sector and the real economy. Each has different policy implications and
involves different challenges in terms of monitoring and measurement (see, eg, Caruana (2010)).
Common exposures and interlinkages create the risk of joint failure. Assessing their
importance means focusing on both how risk is distributed and how the system responds to
either an institution-specific shock or to a common shock that damages everyone. In the first
case, we need to assess the risk of contagion through credit or funding exposures on the one
hand, and the possibility of asset fire sales on the other. In the second case, systemic effects
would arise as a direct consequence of similarities in the structure of institutions’ balance
sheets and funding patterns.
In the context of systemic risk, procyclicality is about the progressive build-up of financial
fragility exacerbating booms and increasing the risk of catastrophic collapse. As costly
experience has taught us, the financial sector can endogenously generate systemic risk in ways
that are often difficult to capture. New financial products with unseen risks can be introduced.
Margins and haircuts, increasingly lax during booms and progressively more stringent in busts,
will exacerbate price fluctuations in markets. And institutions have a natural tendency to
become less prudent during cyclical upturns and more prudent during downturns. Add to this
the fact that during periods of steady, high real growth, financial market volatilities tend to be
low and risk premia compressed. Taking all of this together, the implication is that traditional
measures of aggregate risk tend to look lowest precisely when risk is at its highest.

Measurement
Over the past several years, research has progressed along four broad tracks: distributional models
and stress testing, which are designed primarily to address the risk arising from common exposures
and interlinkages; leading indicators, focusing on countercyclicality; and vulnerabilities analysis that
combines everything. We briefly describe each of these in turn.
Distributional models. Using a variety of methods based on assumptions about individual
firms’ probabilities of default or failure, and the correlation of default events, researchers first
measure the extent of systemic risk in the system, and then allocate it across financial
firms. These methods capture systemic risk arising from both common exposures and
interconnectedness. Crucial inputs into the current versions of this analysis are some
combination of equity prices and credit spreads. These are used for the dual purpose of
estimating the likelihood of firm-level failure (where balance sheet data is another crucial input)
and the bilateral correlations that deliver the systemic risk estimate. A key advantage of the
distributional models is that they generate explicit loss estimates from widely available data. In
addition, since they are based on price data, these methods embed both the extent of
institutions’ leverage, which is very difficult to capture directly, and its distribution across the
system. Unfortunately, the advantage gained from using price data is balanced by the
disadvantage that prices used as measures of risk are at their least reliable when the risks are
highest. In other words, price-based, distributional models of systemic risk are going to be at
their worst when we need them most. This leads us to ask: what could help? The answer is
bilateral exposure data.
Stress testing. The goal of stress testing models is to measure how a financial system
will respond to negative shocks and to trace the effects of common exposures and
interlinkages.  While the analysis can be done at the sectoral level, completing the job requires
granular exposure data at the individual firm level. This, in turn, requires access to detailed

Toward a global risk map 3

 supervisory information. Researchers have done such work in some jurisdictions and for some
institutions. Depending on the methodology, it is then possible to distinguish defaults directly
caused by external events from those triggered by defaults of other banks. Interestingly, the
results available thus far, which use national data, suggest that second-round effects
associated with counterparty risk are of second-order importance.  But, as we think of the
need to evaluate systemic risk at the global level, we are naturally led to ask whether the same
would hold true in a cross-border context. Is the probability of joint bank failure for large
international institutions driven primarily by common exposures? Answering this question
obviously requires that we have data to perform stress tests on the global system level. At this
point, any such exercise would be restricted to using sectoral data or would founder due to the
lack of sufficient firm-level information across national jurisdictions.
Leading indicators. Regarding time series measures, a growing body of work focuses on
building leading indicators designed to capture the build-up of systemic risk over time.  Based
on broad, macro variables such as credit-to-GDP ratios, property prices, equity prices and the
like, the objective is to identify and measure misalignments that can be used to provide
warnings of financial sector distress. Balanced against the clear advantage of their relative
simplicity is the fact that early warning indicators constructed using historical episodes may
have difficulty in tracking the build-up of risks as the financial and economic environment
evolves. Nevertheless, given the potential benefits, it is surely worth looking for ways to
improve what has already been done. And here, we suspect that systematic treatment of cross-
border exposures could help, as could credit and asset price data with greater international
comparability.
Vulnerabilities analysis. Finally, we have vulnerabilities analysis. This is less a model-
based approach to data analysis than it is a mindset or a lens through which to look at the
world. We think of it as a collection of qualitative and quantitative techniques (including, but not
restricted to, the ones mentioned above) that are used to gauge all types of systemic
risk. The idea is to track developments for a wide variety of markets and institutions, using
information on both prices and quantities in an effort to spot pressure points in the financial
system as early as possible. Because price data are so plentiful, and quantity data are scarce,
the indicators employed often use pricing models to infer market assessments of risks and
vulnerabilities. Examples include the use of contingent claims models, such as the one in
Merton (1974), to generate market-implied default probabilities for individual institutions and
efforts to extract information on financial risks, such as implied volatilities, from option prices.
______________________________________


This raises the crucial issue of choosing the appropriate delineation of the system on which to focus. See BIS,
FSB and IMF (2009) for more detail on the challenges involved in this context.  Examples include the
portfolio credit risk models that are being adapted for systemic risk analysis. See, for example, Huang et al
(2009).  See Borio and Drehmann (2009b) for a review of the literature.  Examples include Blåvarg and
Nimander (2002) and Graf et al (2005).  See Elsinger et al (2006), based an analysis of comprehensive data
for Austria.  See, for example, Borio and Drehmann (2009a). Davis and Karim (2008) provide a survey.

Examples include the financial stability reports now published by most central banks and the IMF’s Global
Financial Stability Reports (GFSRs).

What data do we need to construct a global risk map that can answer these questions? To
what extent do existing statistics suffice? Where are the gaps that need to be filled for us to
get the answers we need? While we postpone a more detailed discussion of these questions
until later, a cursory review of the models and approaches commonly used for systemic risk
analysis (see Box 1) suggests that five principles are key to finding the answers:
 Quantities. Simple aggregate statistics go quite some way towards conveying a
broad sense of the build-up of risks. But to move beyond leading indicators to more
sophisticated measures of systemic risk, more and better quantity data are
essential. We see an immediate need for information on the extent of financial
institutions’ exposures with respect to their peers and their participation in various
markets. As should be obvious, this means collecting data in a manner that
preserves counterparty information. Importantly, to the extent that trading is

4 Toward a global risk map

 channelled through central counterparties and organised exchanges, they would be
a natural source for such data.
 Financial intermediaries. Ultimately, systemic risk comes from the maturity
transformation that arises in the process of allocating savings to their most efficient
investment uses. This is the function that intermediaries perform for the economy.
Thus, we need to monitor the activities and risk profiles of all the major international
financial institutions, regardless of their legal form of organisation, and their most
4
important counterparties. At the time of writing, timely, comprehensive and reliable
firm-level information on, say, the world’s top 50 banks is not publicly available.
While commercial data providers, such as Bankscope, offer information on a large
number of balance sheet items (often at a very high price), these data are available
only with significant lags and with insufficient detail on such essential items as
capital composition, currency and maturity breakdowns of assets and liabilities, and
off-balance sheet risk exposures.
 Consistency. Across the globe, it is fair to say that mountains of information are
available. The problem is that compilation methods across datasets and data
providers differ in ways that often make it difficult, or even impossible, to combine
information. For example, the compilation of a complete risk profile for a large
internationally active institution would almost certainly require combining supervisory
information from a variety of jurisdictions. Doing this means having consistent
reporting standards. Broadly speaking, we need better and more detailed
disclosures across institutions. Existing aggregate datasets (eg the BIS international
banking statistics, balance of payments and the flow of funds statistics discussed
below) also need to be (further) harmonised so that they provide information in a
consistent and comparable fashion.
 Maturities and currencies. If we are to adequately capture maturity transformation
and (on-balance sheet) leverage, as well as the maturity mismatch embedded in
cross-currency positions, then we need information on the maturity and currency
composition of both assets and liabilities. To understand just one risk we need to
monitor, note that when long-term foreign currency assets are financed by short-
term domestic currency liabilities, which are then swapped into the target currency
via foreign exchange (FX) swap markets, this creates roll-over risks for the liabilities
and possibly the swap position. (This risk relates closely to questions 1 and 3 above;
we will return to this shortly). Such embedded mismatch can only be monitored with
data on combined maturity and currency breakdowns. 5
 Consolidated data. Many datasets suffer from the fact that they are entirely
residency-based. This may be less important for GDP – we really do want to know
where things are produced – but it is a problem for financial information. Take the
case of a bank (or even a non-financial corporate) operating globally, borrowing and
lending in many countries. If all we know is the location where a loan is made, then
we cannot consistently relate the bank’s exposures to the capital base ultimately
supporting it (or the headquarter location where key funding and risk-taking
decisions are made). And, not only do we need consolidated data, but we need
these data with geographical detail. That is, supervisors in Germany need to be able
to see both the subsidiaries of UK banks operating in Germany, which they already

4
This links closely with ongoing work on identifying systemically important institutions; see eg BIS, FSB and
IMF (2009).
5
Note that information on the instrument type or counterparty type of assets and liabilities (eg money market
funds, central banks, other banks, non-banks etc) can be used to infer information about maturities and may
be easier to collect than actual maturity detail.

Toward a global risk map 5

 know, and the exposure and funding-related activities of German subsidiaries in the
United Kingdom, which they may not know (at least not in sufficient detail). This is
not only an obvious argument in favour of supervisory exchange of information; it
also shows that systemic risk assessment cannot proceed adequately without
6
consolidated exposure and funding data (see question 2 above).
To summarise, taking the lessons of the crisis, we believe that relatively simple aggregate
statistics, used properly, can help us to gauge the build-up of systemic risks nationally and
globally. But, to improve our understanding, to fully exploit the potential of the various
approaches used to conduct systemic risk analysis, we need more. We need consistent and
comparable data across institutions. We need information on the level of common
exposures. And we need data on interlinkages.
But how detailed and complete does information on quantities need to be? There are well
known trade-offs between the macro and micro dimensions of data compilation and
dissemination. While more granular data enables better analysis, compiling and sharing
information becomes progressively more difficult as the number of institutions covered and
the detail requested increase. Costs as well as legal restrictions stand in the way. This
suggests that it would be wise to start with a focus on detailed data for a small number of key
institutions (eg the 50 largest). In parallel, broader sets of statistics need to be improved to
provide more and better aggregate (ie sectoral) data. The ultimate goal is a consistent set of
aggregate statistics that will allow us to identify pressure points at the sectoral level, and to
then follow up on these signals with a more targeted analysis of detailed (supervisory or ad
hoc collections of) data at the firm or market level. This is where the BIS’s experience with its
international banking statistics can play an important role.

3. Elements of a strategy: lessons from the BIS banking statistics

The BIS maintains an established reporting framework covering the international activities of
some 7,000 banking entities from about 40 countries. As such, these international banking
statistics (the BIS banking statistics) cover the core of the global financial system. The
relatively long time series that they provide can be used for historical comparisons and
statistical analysis. Furthermore, it would be possible, at least in principle, to increase the
perimeter of reporting to include key non-bank financial institutions. The BIS also maintains
information on derivatives and debt that can be used to complement and enhance any
analysis utilising the BIS banking statistics.
Critically, the BIS banking statistics largely conform to the five principles identified above. Not
only do they cover the core financial system but they are also available on a consolidated
(headquarter-based) as well as a locational (residency-based) basis. 7 Furthermore, since

6
More precisely, uncovering the geographical exposure and funding patterns of financial institutions requires
joint availability of consistent consolidated data summed worldwide over the home office, all branches and
subsidiaries, and consistent locational data, in which activity is reported separately based on the country
where it is taking place.
7
The BIS banking statistics consist of four complementary datasets: (i) residency data: locational banking
statistics by residency; (ii) nationality data: locational banking statistics by nationality; (iii) consolidated (IB)
data: consolidated banking statistics, immediate borrower basis; (iv) consolidated (UR) data: consolidated
banking statistics, ultimate risk basis (ie adjusted for risk-transfers). The residency and nationality data follow
balance of payments reporting concepts (tracking the international asset and liability positions of banks’ offices
located in a particular country with respect to counterparties in other countries, and positions with respect to
residents of the host country in foreign currencies). In contrast, the consolidated statistics track the worldwide
asset exposures, broken down by residency of counterparty, of banking systems, or the set of internationally
active banks headquartered in a particular country (eg Swiss banks, UK banks etc).

6 Toward a global risk map

they provide some detail on currencies and information that allows maturities to be inferred,
we can examine the mapping of geographical exposure and funding patterns for
consolidated national banking systems. These, in turn, can be used to track the evolution of
common exposures and interlinkages. They also enable an analysis of positioning (ie net
short and net long positions) on both sides of any given market covered by the data.
Activities in other markets can, to some extent, be inferred by making a number of
assumptions.
Before turning to the implications for statistical reporting at the BIS and beyond, we illustrate
how current data can be used to assess two sorts of systemic risks: those emanating from
cross-currency funding and maturity mismatches (our questions 2 and 3 above), and those
arising from speculative carry trades (question 1).

3.1 Monitoring banks’ funding vulnerabilities

A prominent feature of the recent financial crisis was the dislocation in funding markets.
Arising from the liability structure of banks’ balance sheets, the problem developed when a
large number of institutions found themselves in need of US dollars that they had incorrectly
assumed they could either borrow directly or obtain through the foreign exchange swap
market. Indeed, it was the resulting scramble that led to the need for central bank swap
facilities. As already mentioned, the measurement of funding vulnerabilities requires
knowledge of commercial banks’ globally consolidated balance sheets, rather than the
balance sheets constructed along national borders. Consolidation matters because
(i) stresses build up across the balance sheet (as mismatches between the maturity,
currency and counterparty of assets and liabilities) and (ii) the balance sheets of the relevant
decision-making economic units (ie, internationally active banks) span national borders.
It is important to keep in mind that international funding stresses can only be identified by
tracking the patterns in cross-currency funding – that is, the degree to which banks’ assets
are denominated in a currency different from the one in which they issue liabilities. Picture,
say, a German bank with a large euro deposit base. This bank would like to invest in long-
maturity US dollar-denominated mortgage-backed securities (MBS). To hedge the exchange
risk that naturally arises from this position, the bank uses some combination of FX swaps
and futures. But the swaps and futures are, by their very nature, short-term contracts. The
bank will need to roll them over regularly during the period for which it holds the US dollar
MBS. What this means is that, regardless of the maturity structure of the euro-denominated
liabilities (they could be the same as the MBS) the mere fact that the bank engages in cross-
currency financing creates a funding risk. Put another way, these positions have a maturity
mismatch embedded in them. And, if a large number of institutions are doing the same thing
– taking euro or yen deposits, swapping them into US dollars and buying US dollar assets –
then they will pose a systemic risk. 8
To see the build-up of this risk, we need data on the FX swap market. Some information is
already available in the BIS semi-annual OTC derivatives statistics. But these data only
include gross figures, so they do not allow us to see the position’s direction. That is, from
gross FX swap positions we cannot tell whether the reporting entity provided or received a
particular currency in the FX swap market. 9 And without knowing the direction, we cannot tell

8
This is the funding equivalent of a crowded trade; see McGuire and von Peter (2009) for details.
9
The BIS semi-annual OTC derivatives statistics contain information on forwards, swaps and options of foreign
exchange, interest rate, equity and commodity derivatives. The BIS also releases statistics on concentration
measures for foreign exchange, interest rate and equity-linked derivatives, and data on credit default swaps
(CDS) including notional amounts outstanding and gross market values for single- and multi-name
instruments.

Toward a global risk map 7

whether the derivatives positions mitigate or exacerbate the existing on-balance sheet
mismatches.
Yet, by making what is an admittedly bold assumption, we can get a picture of what positions
in the FX swap market might look like at the level of national banking systems. The BIS
banking statistics allow us to see on-balance sheet net (assets minus liabilities) positions by
currency for each national banking system. If we assume that banks have very small open
FX positions, then any on-balance sheet long and short position in a particular currency
provides an estimate of the banks’ net FX swaps and futures positions in that currency. The
left-hand and centre panels of Graph 1 aggregate these on-balance sheet positions by
currency separately for two groups of banks: those with either more or less on-balance sheet
US dollar assets than US dollar liabilities. We label these long-USD banks and short-USD
banks. If banks hedge their foreign exchange risk, then these figures imply that long-USD
banks (shown in the left panel) demand an estimated aggregate of $1.2 trillion (net) in US
dollars from the FX swap market. In return, these banks provide an equal amount of euros,
pounds, Swiss francs and other currencies in exchange. On the other side, the short-USD
banks (shown in the centre panel) are net providers of roughly $800 million to the FX swap
market. The difference of $400 billion is accounted for by non-bank participants that these
data do now allow us to capture.

Graph 1
Long- and short- USD banks’ positions, by currency
In trillions of US dollars
1
USD long banks USD short banks3 Net, by counterparty sector
1.5 1.2 1.8
USD Monetary
4
EUR 1.0 0.8 authorities 1.2
JPY
0.5 0.4 5 0.6
Other banks
6
Non-banks
0.0 0.0 0.0

–0.5 –0.4 –0.6

2
GBP
2
CHF –1.0 –0.8 7
–1.2
Other Cross currency
–1.5 –1.2 –1.8
00 01 02 03 04 05 06 07 08 09 00 01 02 03 04 05 06 07 08 09 00 01 02 03 04 05 06 07 08 09
1
Includes Canadian, Dutch, German, Japanese, Swiss and UK banks. 2 Positions booked by offices located in Switzerland (for
CHF) and in the United Kingdom (for GBP). CHF and GBP positions reported by offices located elsewhere are included in “Other”.
3
Includes Australian, Belgian, Danish, Spanish, Finnish, French, Italian, Luxembourgian, Norwegian, Portuguese, Swedish, Hong
Kong, Greek, Turkish and Taiwanese banks. 4 Cross-border positions in all currencies and local positions in foreign currencies
vis-à-vis official monetary authorities. Excluding liabilities to Japanese monetary authorities placed in banks located in Japan.
5
Estimated net interbank lending to other (unaffiliated) banks. 6 The net position vis-à-vis non-banks is estimated as the sum of
net international positions vis-à-vis non-banks and net local US positions (vis-à-vis all sectors). By construction, net claims on non-
banks is the sum of net positions vis-à-vis other banks, vis-à-vis monetary authorities and cross-currency funding, which is the lower
bound estimate of the US dollar funding gap. The dashed green line is the estimate after adding back writedowns of assets.
7
Implied cross-currency funding (ie FX swaps) which equates gross US dollar assets and liabilities.

Sources: BIS consolidated statistics (immediate borrower and ultimate risk basis); BIS locational statistics by nationality.

Measured in any particular currency, total FX swap contracts must balance: total dollars
supplied equal dollars demanded. But a whole range of ways exists in which supply can
equal demand. And which one actually transpires can or might have implications for market
pricing. For example, Graph 1 seems to suggest a growing asymmetry – diverging large net
positions in the swap market across banking systems – as the crisis was building. That is,
dollar borrowing via FX swaps became increasingly important for the long-USD banks. We
now know what happens when the net providers of these dollars (the short-USD banks and

8 Toward a global risk map

non-banks) become skittish about lending to the net demanders when credit risk concerns
arise. Are large price movements in a particular currency more likely when the market is
asymmetric in this way? Perhaps. But we can hardly expect market prices to reflect such
market risks a priori when the participants are blind to the overall structure of the market.
The funding risks faced by banks involve more than just rollover risk in the swap market, also
including the rollover risk inherent in borrowing directly in foreign currencies. To get some
sense of the size of this risk, we look at the long-US dollar banks’ investment and funding
positions in US dollars. The shaded area in the right-hand panel of Graph 1 indicates the size
of cross-currency financing into dollar positions (ie FX swaps) by these banks – this is simply
the inverted quantity plotted as the red line in the left-hand side panel. In addition to this
funding, long-US dollar banks also borrowed dollars from the interbank market (the blue line
in the right-hand panel) and from central banks (the red line), and then used these funds to
finance their net investment in non-bank entities (the green line). While direct information on
maturity is not available, we can safely assume that interbank borrowing and FX swaps are
generally short term. Thus, Graph 1 suggests a significant degree of maturity transformation
across the balance sheets of these long-USD banks. The estimated funding gap – the gap
between the long-term US dollar holdings and the short-term US dollar financing implied by
these figures – reached well over $1 trillion for these banks at its mid-2007 peak. 10
Combining these numbers with the experience during the crisis leads us to conclude that the
combination of currency and maturity mismatches creates systemic risks that need to be
monitored at a global level. And that will require the provision of information on bank and
non-bank institutions’ globally consolidated cross-currency financing positions – something
that is only available through international cooperative mechanisms such as the BIS banking
statistics.

3.2 Measuring the currency carry trade–the Japanese yen

How big is the carry trade? The answer to this apparently simple question has eluded
analysts for more than a decade. In part, this is because the term carry trade is used to refer
to several types of financial activity: portfolio diversification by Mrs Watanabe in Japan, low
interest rate foreign-currency mortgages in Eastern Europe, and high-frequency speculative
positions held by hedge funds and other leveraged investors. Measuring the size of the latter,
the speculative positions, has proven extraordinarily difficult largely because there are
different ways to implement the same speculative trade. But, the potential for a disruptive
unwinding of such positions – which is independent of the exact form a speculator may use
for placing the bet – has been a long-standing policy concern.
As we will explain, we believe that the only way to gauge the magnitude of speculative carry
trade positions funded in a particular currency is through the interactions in the FX swap
market. As in the previous section, we have to infer these positions from mirror images of
observable on-balance sheet positions. Using the example of the yen-funded carry trade, we
sketch how this can be done.
To avoid potential confusion, it is worth starting with a definition. We define a carry trade as
the combination of borrowing at a low interest rate (a short position that we can think of as
issuing a bond or taking out a bank loan) and lending the proceeds out at a high interest rate
(a long position that is the purchase of a bond or the provision of a loan). For clarity, let’s
assume that the speculator is selling a bond and using the proceeds to purchase a bond.
Importantly, borrowing and lending can occur at different maturities and in different

10
This is a lower-bound estimate which implicitly assumes that US dollar liabilities to non-banks are long term.
The assumption that these liabilities to non-banks are short term gives an upper-bound estimate of $6.5
trillion.

Toward a global risk map 9

currencies. If they do, we can decompose the carry trade into a yield curve position (which
seeks to profit from any term premium, where typically the borrowing is shorter term than the
lending) and an open exchange rate position (which seeks to profit from the failure of
uncovered interest rate parity, ie the existence of a forward premium where the forward
exchange rate does not accurately reflect the future expected spot rate). 11
In the textbook version of the trade, the investor borrows in a low-interest rate currency, such
as the Japanese yen. He then sells the yen in the spot market for US dollars, and purchases
a US dollar bond with a yield higher than the rate paid on the yen-denominated loan.
Provided that the dollar does not depreciate relative to the yen, the investor will earn at least
the (maturity-matched) cross-currency interest rate differential – that is, the difference
between the US dollar bond rate and the Japanese yen borrowing rate. If there is a maturity
mismatch, then the carry trade investor is exposed to both interest rate risk and exchange
rate risk. As a result, these trades are notoriously vulnerable to interest rate, yield curve, and
exchange rate movements. And, when any of these starts to move against the position, there
can be a rush for the exits as investors try to quickly rid themselves of both maturity and
currency mismatches, thereby creating sudden jumps in interest and exchange rates.
How can we figure out how much of this is going on and whether it poses a systemic risk?
First, even though textbooks discuss it this way, very little speculative carry trade activity is
done exactly as we have just described. This implies that a simple look at cross-border
lending in various currencies won’t work. That is, you will not find evidence for carry-trade
positions directly on the balance sheets of financial institutions. 12 Instead, a speculator will
use derivatives – futures, forwards and swaps – to put on currency carry trade positions. A
typical transaction might look like this. A dollar-based hedge fund borrows dollars (step 1)
and swaps them for yen (step 2), where the swap rate is determined by the standard covered
interest rate parity condition. 13 So far, the investor owes dollars at the dollar interest rate, and
has yen. At step 3, the hedge fund sells the yen for dollars in the spot market (which puts
downward pressure on the yen/dollar exchange rate), and takes the dollar proceeds and
invests them (step 4). At maturity, the position is unwound. The dollar investment matures
(undoing step 4), the investor takes the proceeds and buys yen (undoing step 3), which are
then used to make good on the FX swap (step 2). Finally, the hedge fund repays the original
loan (step 1). While the position is in place, the hedge fund shows a dollar liability and a
dollar asset (probably of different maturities). In other words, the speculation that the yen will
depreciate against the dollar (or, at least, not appreciate) and the income from the interest
rate differential embedded in the FX swap will be invisible if all you see is balance sheet
data.
Importantly, though, for the hedge fund to be able to put on the position in the first place,
someone has to be willing to provide yen for dollars today in the FX swap market (step 2).
And which institutions in the world have both the yen to provide and a need for dollars (and
other currencies)? Japanese banks and, perhaps, insurers come to mind. Indeed, estimates
suggest that Japanese banks have been supplying upwards of $1 trillion worth of yen to the
swap market over the last few years – this is the red line in the left-hand panel of Graph 2. 14

11
On the forward premium, see, for example, Flood and Rose (2002).
12
Support for this assumption is provided by Galati et al (2007), who scoured the BIS locational banking
statistics for yen lending/borrowing related to the carry trade, but did not find much evidence.

That is, (i  i*)  ( f  s ) , where i is the US dollar interest rate, i is the yen interest rate, s is the dollar/yen
13

s (1  i*)
spot rate, and f is the dollar/yen forward rate at the same maturity as the interest rates.
14
See McGuire and von Peter (2009) and section 3.1 above.

10 Toward a global risk map

 Graph 2
Uncovering the speculative bid in yen carry trades

Yen supply and demand in the swap market FX futures positions2

160 1,500 500 16
1
Non-JP banks’ net JPY positions (rhs)
1
148 Debt securities (rhs) 1,200 250 8
1
JP banks’ net JPY positions (rhs)
136 JPYFXRate 900 0 0

124 600 –250 –8

112 300 –500 2

–16
Yen futures position (rhs)
Speculative yen
100 0 –750 1, 3 –24
carry trade (lhs)

88 –300 –1,000 –32

00 01 02 03 04 05 06 07 08 09 00 01 02 03 04 05 06 07 08 09
1
In billions of US dollars. 2 Net non-commercial long positions on the Chicago Mercantile Exchange; derived using average exchange
rates of March 2010; in billions of US dollars. 3 Calculated as the (negative of) the gap between Japanese banks’ net JPY positions
(red line) and the sum of the shaded areas in the left-hand panel.

Sources: CFTC; BIS calculations.

The natural counterparty to Japanese banks in the FX swap market is a foreign investor
seeking to invest in yen assets. This could be a European or a US bank or a pension fund
interested in investing in Japanese equities or bonds. Unlike the hedge fund above, these so-
called real money investors want to buy and hold yen assets, so they do not sell the yen they
get from the swap in the spot market (no steps 3 and 4). That means that their swap market
transactions have no direct exchange rate implications.
How can existing data be used to figure out what is going on? Is there a way to estimate the
size of the speculative yen carry trade from FX swaps? The short answer is yes,
approximately. The trick is to infer the red line – an estimate of the size of the carry trade – in
the left panel of Graph 2 from the information we have. We can do this by subtracting the
amount of yen demand by real money investors in the FX swap market from the total yen
volume supplied to the market by Japanese banks. To do this, we need to construct
estimates of yen demand from real money investors. For (non-Japanese) banks, we again
use the BIS banking statistics to gauge their on-balance sheet net position in yen-
denominated assets. This is shown as the darkly shaded area in the left-hand panel of
Graph 2, and you can see that it is rather small. If non-Japanese banks fully hedge exchange
rate risk, this implies that they are not demanding much yen in the FX swap market, at least
on a net basis.
What about the yen position of the non-bank real money investors? Unfortunately, the
breakdowns available in the aggregate data covering non-banks (from sources such as flow
of funds and balance of payments accounts) do not provide any equivalent measure of yen
asset demand. So, for lack of better data, we use a measure from the BIS debt securities
statistics: the outstanding stock of yen-denominated international bonds. If we assume that
(a) none of these bonds are purchased by banks (ie no double-counting), and that (b) none
of these are purchased by yen-based investors in Japan, then the outstanding stock of yen-
denominated international bonds equals the total international demand for yen assets by
non-Japanese non-bank investors. If these investors hedge their exchange rate risk, then
this stock should approximate their demand for yen funding from the FX swap market. The
result is the lightly shaded area in the left-hand panel of Graph 2.
Putting everything together, we have a measure, albeit rough, of the speculative yen carry
trade. What does it look like? Until 2004–05, yen volumes supplied by Japanese banks in the
FX swap market roughly matched demand for yen assets from real-money investors, as

Toward a global risk map 11

proxied by the sum of non-Japanese banks' net on-balance sheet yen positions and total
outstanding international yen-denominated bonds. After 2005, however, the gap, that is, the
difference between the stacked shaded areas and the red line, widens considerably.
Unsurprisingly, this corresponds to a period of yen depreciation (the green line) – the carry
trade that we have outlined puts downward pressure on the yen – and, at least until the start
of the recent crisis, is consistent with other commonly used indicators of carry trade activity.
For example, analysts often rely on information on yen positioning by non-commercial
entities (assumed to be hedge funds and other non-bank financial institutions) in the Chicago
currency futures markets. While these data capture only a slice of the global currency futures
activity (and miss all OTC positions), they are often used to get a sense of the direction of the
speculative bid for yen. And, as shown in the right-hand panel of Graph 2, these net short
non-commercial yen positions grew substantially after 2004, just as our gap measure widens.
What can we conclude from all this? Is it a coincidence? Or, is the gap that we infer a reliable
measure of speculative carry trade activity? We don’t know for sure (given, for example, the
gap that opens up between the two measures in the right-hand panel for the most recent
observations), but the bigger point is that carry trades are largely off-balance sheet. Without
more consolidated information on derivatives positions, there is no way to track them directly.
To better understand the FX swap market (and the relative importance of participants
therein), what we need is net position data by currency and nationality. And, as in the
example in the previous subsection, here again the necessary information can only come
from datasets that cover all financial institutions and investors, and that feature both a
locational (residency) and consolidated (headquarters) dimension.

3.3 Implications for statistical strategy: the BIS and beyond

Based on the two examples discussed above, it should be clear that the BIS banking
statistics already have most of the properties that we would want from a statistical reporting
framework geared at financial stability analysis. From this, we draw three sets of implications
for the BIS and beyond.
Improving the BIS banking statistics: The crisis highlights the value of consistent banking
system data for the analysis of systemic risk. For the BIS, this implies the need to focus on
statistical improvements that increase the analytical value of the BIS banking statistics for
central banks, the broader financial community and academic researchers. The BIS’s
experience is that relatively small changes to existing reporting frameworks can generate
significant increases in analytical value.
Specifically, we see a need to enhance the ability of the BIS banking statistics to cover the
maturity and currency dimensions of banks’ (international) activities, and to provide
aggregate measures of on-balance sheet leverage. In addition, we hope to be able to
increase the granularity of existing data reporting, refining sectoral breakdowns and the like,
to allow improved integration of supplementary statistics on key financial institutions, and to
selectively add supplementary information on global markets. For the latter, specifically the
integration of information on off-balance sheet activities, central counterparties and trade
warehouses can be a useful additional data source (see Box 2 for a short description of the
BIS’s proposals in these areas; for concreteness, the Appendix sketches out a hypothetical
reporting template, based largely on the existing structure of the BIS banking statistics, and
proposes some broad market and counterparty breakdowns for consideration).

12 Toward a global risk map

 Box 2: Specific improvements to the BIS international banking statistics

Improvements to the international banking statistics are possible along the following dimensions:
 Currency and maturity detail. Our analysis of funding gaps reveals scope for
significant improvements based on relatively small reporting changes: specifically, more
refined and consistent (i) currency breakdowns, (ii) counterparty breakdowns and (iii)
maturity breakdowns across different parts of the BIS banking statistics.
 De-masking and disaggregating. A significant amount of data is actually collected by
central banks but not on-reported to the BIS (either because of aggregation or because
data is masked in the submissions). Compilation is thus possible at low cost, potentially
allowing for significant enhancements of the analytical value of the BIS banking
statistics.
 Leverage and derivatives. Additional enhancements are possible by explicitly covering
on-balance sheet leverage (eg by including information on capital and total assets in the
reporting), as already done by some reporters, and by trying to better capture
derivatives-related exposures (as done, to some extent, in making ultimate risk
adjustments).
 Splicing. Other existing datasets can be used to further enhance the analytical value of
the BIS banking statistics (eg by relating exposures to measures of total assets or equity
for individual banks, which can then be aggregated up in ways consistent with the BIS
banking statistics). The same could be done with other BIS sets of statistics, to better
integrate them with the BIS banking statistics and with information available from other
sources. A key goal would be to get a sense of asset/exposure holdings, for which data
from trade warehouses and central counterparties can be useful (eg ongoing work on
improving the BIS’s over-the-counter (OTC) derivatives statistics by combining the data
on credit default swaps (CDS) with information provided by the Depository Trust and
Clearing Corporation (DTCC). Similar work would be possible for the BIS securities
statistics).
 Supplementary data. Further, supplementary statistics could be compiled selectively to
enhance the existing focus on international banks and global markets. This could involve
the preparation of global aggregates for key indicators already available on a national
basis. Examples include ongoing work by the Committee on the Global Financial
System (CGFS) on aggregate survey measures of credit conditions. Once a framework
for compilation is in place, these efforts could be extended to cover margin
requirements/haircuts and similar metrics that can be used to gauge changes in lending
standards and to aid the monitoring of build-ups in risk.

_______________________________________


See section 3.1 above and McGuire and von Peter (2009).

Core system data: As already mentioned, standardised information on the top 50 banks and
financial institutions is not publicly available in a detailed and timely fashion. Different data
providers provide different levels of detail, but these cannot be easily spliced together. Fixing
this problem means redoubling efforts to improve standardisation and enhance disclosure.
The goal is to provide better public information so that authorities can improve their
monitoring and markets can improve their discipline.
Statistical design more broadly: Finally, if we are to take the crisis-related lessons coming
out of the BIS banking statistics seriously, we need to design statistical frameworks so that
the same sort of analysis of leverage, maturity mismatch and exposure/funding patterns that
is possible for banks can also be done for other sectors. At the risk of repetition, the key here
is consolidation. Specifically, existing aggregate data need to be collected to allow joint
identification of the nationality of the reporting entities’ headquarters and the location of any
particular branch or subsidiary. If all our aggregate statistics follow this principle, we can then

Toward a global risk map 13

complement the locational point of view with a consolidated view, which is necessary for the
measurement of balance sheet stresses.
Fortunately, much of the necessary data are already being collected in existing reporting
frameworks for the creation of the flow-of-funds and balance-of-payments statistics. All that
may be needed are – admittedly non-trivial – changes to existing collection protocols,
combined with adjustments to computation methods (see the Appendix for details). The
payoff would be a complete set of consolidated and locational statistics at the sectoral level
(for banks, other financial intermediaries, non-financial corporates and households) that
could then be used to monitor systemic risk globally and nationally.

4. Implications for global risk maps and similar tools

Global risk maps are the holy grail of systemic risk monitoring. Defined as unified databases
that provide data on risk exposures of financial institutions and markets, they would allow
supervisors and market participants to monitor the evolution of risks in banks, banking
systems and the broader financial market community. 15 Not only would these risk maps
provide data inputs for financial stability analysis, they would also serve as the starting point
and testing ground for the development of new systemic risk measurement and mitigation
tools.
In practice, the creation of a global risk map would require giant matrices of bilateral
exposure data to be assembled for a large set of financial institutions and other systemically
important entities, possibly grouped by different risk categories. The usefulness, in principle,
of such an exercise is self-evident. Done right, it would allow us to read directly from the
matrix, at any point in time, the risk exposures among financial institutions and with respect
to particular risk factors. This would make it possible, at least in principle, to calculate a
whole range of risk and concentration measures both cross-sectionally and over time.
That is the theory. What about the reality? As we noted before, in approaching this sort of
task, we face trade-offs between the macro and micro dimensions of data compilation and
dissemination. The costs of collection and analysis will rise as the scope and scale of the
enterprise increase. This means looking for a balance between the degree of disaggregation,
the universe of firms included, and the amount of detail covered. This, in turn, relates to the
boundaries of supervisory exchange of data and broader statistical frameworks aimed at
public dissemination. Armed with our experience with the BIS statistics, we now turn to a
brief discussion of each of these.

4.1 Supervisory information exchange

We have mentioned this several times because we believe it to be central to the whole
process of systemic risk assessment. The experience of the last few years clearly revealed
that supervisors did not have all the information they needed to track the build-up of risk
going into the crisis. And, perhaps more importantly, they did not have the information they
needed to better contain the crisis once it started. In a world with global banks and global
financial markets, we have no choice but to set up robust frameworks for supervisory
information exchange.
Generally, given the detail needed, such frameworks are likely to focus on a core set of
internationally active institutions. That said, the goal has to be to capture these firms’ risk

15
See Issing Committee (2009).

14 Toward a global risk map

profiles, not only individually but also with regard to concentrations of risk arising from
common exposures among these institutions and from interlinkages between them and their
main counterparties.
What are the main requirements of such frameworks? First, as emphasised above,
comparability and consistency are key. This means standardising reporting across
jurisdictions, including standardisation of any risk measures to be used and the methods for
their construction. 16 Second, reporting will have to be flexible enough to allow regulators to
react in a timely manner to market innovation, structural trends and moves by institutions to
evade what they might see as injurious actions taken in response to the data they are
obliged to report. Third, any legal constraints will have to be sorted out to allow international
exchange of data. Essentially, what is required is a joint reporting template to be used by
supervisors across jurisdictions for core internationally active institutions, based on common
definitions for instruments, counterparties and risk measures. 17
What types of reporting would this entail? Broadly, we can distinguish three types of risk on
which detailed information would be needed: market risk, funding risk and credit risk (plus
insurance risk, for institutions involved in non-life insurance activities). 18 The capture of
exposures across all three risk categories requires detailed information by asset class and/or
market as well as the names of the most important individual counterparties.
Derivatives exposures, which are insufficiently or not at all captured by existing sets of
information and standard disclosures, are a natural starting point. To do the job, supervisors
would need information on gross mark to market exposures (after counterparty netting) with
respect to key counterparties, any collateral held, and net exposures (after netting and
collateral). As market conditions can change rapidly, some measure of potential or worst-
case exposure would also be crucial (based on an agreed methodology).
Next come repurchase agreements, securities lending, and interbank markets, and the list
goes on.
Exploratory efforts to collect counterparty-level credit data for supervisory use are under way
in the context of the so-called Senior Supervisors Group. 19 The idea is to identify the most
important counterparties, to monitor their exposures and to analyse the interconnectedness
among reporters. The resulting risk map, updated over time, will help identify changes in
peer relationships, track the magnitude of these linkages and monitor any concentrations.

4.2 Public information dissemination

Public availability of a global risk map would allow the use of detailed exposure data in the
calculation of system-wide measures of risk outside the strictly supervisory domain. This
would enable investors, academics and authorities beyond supervisors to conduct their own
analyses, improving the monitoring of systemic risks and enhancing market discipline. As a
general rule, to the extent possible, both supervisors and the general public should thus be
supplied with the same information. Where needed, aggregation can be used to address
confidentiality constraints, and reduced reporting frequencies can limit the costs of publicly
disseminated datasets.

16
This may require harmonisation of the regulation and supervision of the core institutions themselves.
17
See Tarullo (2010) for details.
18
See, for example, BCBS, CGFS, IAIS and IOSCO (2001).
19
See also FSB and IMF (2009), especially recommendation #8.

Toward a global risk map 15

The possibility of such reporting is currently being considered by the Financial Stability
Board. 20 And the BIS is contributing to this effort. Yet public dissemination involves
significant challenges. This is where the lessons from the crisis as extracted from the BIS
banking statistics offer a viable alternative. For example, if consolidated (the nationality or
headquarter dimension) as well as locational (the residency dimension) data were jointly
available for the flow of funds and balance of payments statistics (including the IMF’s Co-
ordinated Portfolio Investment Survey), we could then see the entire balance sheet of
households, nonfinancial corporates and financial institutions. This would allow for a rather
detailed analysis of common exposures and interlinkages at the sectoral level, in both
domestic and international terms.
Monitoring at the aggregate, sectoral level is but one step removed from what could be done
with complete global risk maps. Importantly, it could help sidestep the complex legal
challenges surrounding the sharing or dissemination of firm-level data, while supporting a
two-step approach to systemic risk monitoring. That is, the alarms that sound in the
aggregate data will yield the critical pieces of information – nationality of entity, location of
entity and risk type (including funding and exposure risk) – to inform targeted analysis of
more detailed (supervisory or other) data at the firm or market level. 21

5. Conclusions
The recent financial crisis revealed important gaps in our ability to analyse financial
institutions and markets at the system-wide level. Simple aggregate statistics can go some
way in gauging the build-up of risks in a broad sense. But to understand, measure and
mitigate systemic risk we need more. From the crisis experience, we take away five
principles that we believe should guide future statistical collection efforts:
1. Quantities: As a complement to the price data we already have, we need to know
quantities. We need on- and off-balance sheet information to evaluate common
exposures, interlinkages and countercyclicality.
2. Intermediaries: As maturity transformers and funding liquidity providers, financial
intermediaries are the ultimate sources of systemic risk. We need data that allows
us to see what they are doing and how they are doing it.
3. Consistency: We need the ability to put together disparate datasets from different
sources around the world. That means consistency of reporting frameworks.
4. Maturities and currency: Without information on maturity (perhaps inferred from
counterparty information) and currency composition, we will not be able to see the
risks that come from mismatches.
5. Joint residency and consolidated reporting: Officials in one country need to
know what the subsidiaries of their banks are doing in another. Likewise, the
regulators in a particular country will want to see stresses building up on the
consolidated balance sheets of banks operating foreign offices in their jurisdiction.
Using these five principles, we can act to improve data in a way that will allow us to move
significantly in the direction of producing a global risk map – the giant matrices of exposures
and funding relationships that would, in principle, allow a measurement of virtually any
vulnerability in the financial system. One example already in place is the use of the BIS

20
See FSB and IMF (2009), particularly recommendation #9.
21
See Eichner et al (2010) for a detailed description of a very similar approach.

16 Toward a global risk map

banking statistics to produce what we might call a “global risk map in the small”. While
restricted to banks and then only to their international activities, this set of statistics already
allows us to analyse (some) interconnections and common exposures at the sectoral level,
over time. And relatively small adjustments to the existing reporting would further improve the
capabilities of this particular dataset – improvements that the BIS intends to pursue actively
over the coming months.
How about the set of three core questions we raised at the outset of this paper?
 Question 1: How can a central bank monitor the global use of its currency?
Add up consolidated investment and funding positions in your currency across
sectors and countries – these are the FX swap market and carry trade examples we
described.
 Question 2: What information do policy makers need (in addition to home country
supervisory data) to monitor systemic financial stresses?
They need data on the geographical patterns of funding and investment
relationships, including aggregate sources of funding for the domestic corporate
sector, and asset and liability information, including currency and maturity
mismatches, and information on all of the entities and sectors providing funding to
financial institutions.
 Question 3: What data could help financial market participants improve market
discipline?
While we cannot be sure, our sense is that more information on funding sources and
common exposures for consolidated entities, including all bank and nonbank
financial firms, as well as nonfinancial corporates and household worldwide would
help.
All of this leads us to draw an important inference for statistical reporting of aggregate data
more generally. In particular, tremendous value would come from adding a consolidated (ie
nationality) dimension to commonly constructed locational (residency-based) data. This
would make it possible to adapt existing statistical reporting frameworks (such as the flow of
funds and balance of payments statistics) in order to facilitate systemic risk analysis in ways
that are only one step removed from what could be done with complete global risk maps.
This, in turn, would help to sidestep the complex legal challenges surrounding the sharing of
firm-level data. And it would improve our ability to use aggregate data for the identification of
vulnerabilities (including information on the nationality and location of any entity and on the
risk type) for a deeper drill-down on the basis of more granular data.

Toward a global risk map 17

Appendix
What breakdowns of aggregate statistics do we need? Drawing on our experience with the
BIS banking statistics, we outline some tentative answers to that question in this appendix.
In any data collection exercise, additional detail incurs greater costs. Not only that, but there
are many paths to the same result. Our goal here is not to present what is the right or best
template, but rather to provide a point of departure for discussions on what material
improvements we can realistically hope to make. As a first step, we adopt a broad
perspective, describing the maximum of what would be desirable.
Broadly speaking, aggregate data should allow us to assess, at the sectoral and systemic
levels, credit risk, country risk and market risk on the asset side; and currency risk and
maturity transformation across the balance sheet. Getting a sense of these risks requires
measurement of liquidity risk on the asset side, and rollover risk on the liability side. This, in
turn, means having information on markets, counterparty, instrument, maturity and currency.
The question that faces us is what combination(s) of information in our aggregate statistics
would allow us to see where these concentrations of risks are building? Table A.1 presents a
somewhat stylised balance sheet with four breakdowns (indicated in the column headings),
or classifications, for assets and liabilities. Our basic argument is that combinations of
quantity data combined with some comparative static analysis can go a long way in allowing
us to uncover where important risks are located. At the very least, such data should help to
focus targeted discussions with regulatory authorities in various countries that have (or are
able to obtain) the underlying firm-level information. In this context, several points are worth
highlighting:
 Reporting unit: The ideal underlying reporting unit must be a household or the
office/branch/subsidiary of a bank, non-bank financial institution or corporate in a
particular location (country). These are the reporter types. Ideally, the number of
different reporting sectors is the same as the number of counterparty sectors, listed
under counterparty types in Table A.1, column 2.
 Aggregation: The balance sheet positions should be aggregated by location
(country), reporter nationality and reporter type. For example, Table A.1 could be
viewed as the (aggregated) positions of, say, the 10 offices of Belgian-
headquartered non-bank corporates located in Italy. When all countries and all
sectors report, the underlying data blocks can then be further added up to produce
(i) total positions for all entities located in Italy (country-level totals on a residency
basis), (ii) Belgian-headquartered non-bank corporates’ worldwide positions
(consolidated global totals for a particular national sector), and (iii) Belgian-
headquartered entities’ worldwide consolidated positions (country-level consolidated
exposure totals).
 Breakdowns: The breakdowns that we need to gauge things such as maturity-
transformation risk, counterparty risk, and currency risk are illustrated by drilling
down to a particular counterparty type. That is, for a reporting unit located in country
X, the asset side of the balance sheet shows claims on corporate borrowers (Table
A.1, column 2) located in country A (column 3), broken down by currency (column
5), remaining maturity (column 4), and instrument type (column 1). The liability side
shows total liabilities to non-bank financials (column 10), with the same breakdowns.
 Off-balance sheet positions: FX swaps, interest rate swaps and credit default
swaps are all shown at the bottom of the table – they are off-balance sheet. Along
with other derivatives, these add an additional layer of exposures (both on the asset
and liabilities side of the balance sheet) that can magnify, neutralise or reverse the
reported on-balance sheet positions. Ideally, we would require separate reporting of
all items in Table A.1 before and after off-balance sheet activities. In other words,

18 Toward a global risk map

 we would want to add a layer of ultimate risk reporting similar to what is already
available for credit risks in the BIS banking statistics, but covering all risk categories.
For example, for the maturity column we would have exposures after transformation
through interest rate derivatives, for currency it would indicate all open foreign
exchange exposures after FX swaps, and for credit risk, reporting would be by
22
country location and counterparty type on an ultimate risk basis.

Table A.1 A stylised reporting template for aggregate (sectoral) statistics

Assets Liabilities
1 2 3 4 5 6 7 8 9 10

Instrument Counterparty Counterparty Remaining Remaining Counterparty Counterparty Instrument

Currency Currency
type type location maturity maturity location type type

Bank Country A

Country B Country A

Country C Short term Short term

Country D USD Country B Non-bank fin Debt security

Country E Country C

Loan Country A
Long term
Debt security Country B Household
Corporate
Equity Country A
Short term
Other USD Country B Corporate

Household Country A
Country A
Non-bank fin Corporate
Long term
Bank Household

Public sector Deposit

Euro
Central bank Non-bank fin Debt security
Long term
Interoffice Country B Other

Country B Bank

Country C Public sector

Loan Corporate Central bank

Country A
Non bank fin Short term Interoffice

Country B Country A

Debt security Corporate Euro Non-bank fin Equity

Country A Book
Interoffice Country B Public sector
equity
Long term
Country B Corporate

Country C Country C

Exchange rate swaps Interest rate swaps Credit default swaps

22
We realise that this is conceptually challenging and amounts to a very significant reporting burden, which
implies that existing datasets (such as the BIS OTC derivatives statistics) might be enhanced to provide at
least some of the detail mentioned above.

Toward a global risk map 19

 Filling in the balance sheet: Reporting entities will have difficulty filling in all of the
cells in Table A1. In particular, if the reporting institution issues a bond that is traded
on the secondary markets, it is unlikely to know the counterparty that holds the bond
at any point in time. Thus, columns 8 and 9 on counterparty location and type might
be empty. Note that debt securities liabilities are different from, say, banks’ deposit
liabilities, where both the location and the type of counterparty are typically known
by the reporting bank.
 In a world where all entities in all countries are reporters, what’s missing in one
place, can, in principle, be recovered somewhere else. This would require one
additional breakdown (not shown in Table A.1) on the asset side of the balance
sheet: the nationality of the counterparty. For example, the owner of a long-term
bond issued by the London subsidiary of a US-headquartered auto company may
not be known to the UK issuer, but would be picked up if the Saudi pension fund that
owns it reports it as a claim on a US-headquartered corporate located in the United
23
Kingdom. That is, if the asset holder reports both the location and nationality of the
counterparty.
How would this help identify the important risks and vulnerabilities? With a perfect
consolidated view of each sector’s worldwide consolidated exposure, it would be possible to
read off many of the risks we have discussed. For example, in the case described in the text,
we would be able to see the institutions covered by Table A.1 that have a long on-balance
sheet US dollar position financed by short-term dollar funding and long-term euro funding.
We would know if the on-balance sheet exposures were large and growing. And, we would
be able to see the maturity profile of the position, including that of the FX swaps. On the
asset side, we would be able to identify concentrations of particular types of assets and risks,
and tell if they are large relative to capital. And, on the liabilities side, we would be able to
see the scale, timing and frequency of required rollovers.
Finally, note that the breakdowns in Table A.1 are roughly in line with those already available
in the BIS banking statistics, but with several important exceptions. First, comprehensive
information on the maturity buckets is missing in the BIS data. Second, the currently
available breakdown by counterparty type is far less comprehensive (only bank/non-bank; in
some data, the public sector is split out). Third, information on book equity is not explicitly
broken out on the liabilities side. And fourth, there is only limited information on banks’ off-
balance sheet positions.

23
The IMF’s CPIS statistics provide information on individual economy (ie country location) year-end holdings of
portfolio investment securities (equity securities and debt securities), cross-classified by the country of issuer
of the securities.

20 Toward a global risk map

References
BCBS, CGFS, IAIS and IOSCO (2001), Final report of the Multidisciplinary Working Group
on Enhanced Disclosure, Basel, February.
BIS (2008), 78th Annual Report, Basel.
BIS (2009), 79th Annual Report, Basel.
BIS, FSB and IMF (2009), Guidance to assess the systemic importance of financial
institutions, markets and instruments: initial considerations, Basel, October.
Blåvarg, M and P Nimander (2002), “Interbank exposures and systemic risk”, Sveriges
Riksbank Economic Review, no 2, pp 19–45.
Borio, C and M Drehmann (2009a), “Assessing the risk of banking crises – revisited”, BIS
Quarterly Review, March.
——— (2009b), “Towards an operational framework for financial stability: ‘fuzzy’
measurement and its consequences”, BIS Working Papers, no 294, June.
Caruana, J (2010), Systemic risk: how to deal with it?, Basel, February.
Davis, E and D Karim (2008), “Comparing early warning systems for banking crises”, Journal
of Financial Stability, vol 4, June, pp 89–120.
Elsinger, H, A Lehar and M Summer (2006), “Risk assessment for banking systems”,
Management Science, vol 52(9), September, pp 1301–41.
Flood, R and A Rose (2002), “Uncovered interest rate parity in crisis”, IMF Staff Papers, vol
49, pp 252-66.
FSB and IMF (2009), The financial crisis and information gaps, Basel/Washington DC,
October.
Galati, G, A Heath and P McGuire (2007), “Evidence of carry trade activity”, BIS Quarterly
Review, September.
Graf, JP, S Guerrero and F Lopez-Gallo (2005), Interbank exposures and contagion: An
empirical analysis for the Mexican banking sector, Bank of Mexico, May, mimeo.
Huang, X, H Zhou and H Zhu (2009), “A framework for assessing the systemic risk of major
financial institutions”, BIS Working Papers, no 281, April.
Issing Committee (2009), New Financial Order – Recommendations by the Issing
Committee, February.
Eichner, M, D Kohn and M Palumbo (2010), Financial statistics for the United States and the
crisis: what did they get right, what did they miss, and how should they change?, paper
prepared for the European Central Bank Conference “Central bank statistics: What did the
financial crisis change”, Frankfurt.
McGuire, P and G von Peter (2009), “The US dollar shortage in global banking and the
international policy response”, BIS Working Papers, no 291, October.
Merton, R (1974), “On the pricing of corporate debt”, Journal of Finance, vol 29(2), pp 449–
70.
Tarullo, DK (2010), Equipping financial regulators with the tools necessary to monitor
systemic risk, testimony before the Subcommittee on Security and International Trade and
Finance, US Senate Committee on Banking, Housing, and Urban Affairs, February.

Toward a global risk map 21