Bis q3-14

BIS Quarterly Review
September 2014

International banking
and financial market
developments

Monetary and Economic Department

Editorial Committee:
Claudio Borio Dietrich Domanski Hyun Song Shin Philip Turner Christian Upper
General queries concerning this commentary should be addressed to Christian Upper
(tel +41 61 280 8416, e-mail: [email protected]), queries concerning specific parts to the authors,
whose details appear at the head of each section, and queries concerning the statistics to
Philip Wooldridge (tel +41 61 280 8006, e-mail: [email protected]).

This publication is available on the BIS website (www.bis.org/publ/qtrpdf/r_qt1409.htm).

Bank for International Settlements 2014. All rights reserved. Brief excerpts may be
reproduced or translated provided the source is stated.

ISSN 1683-0121 (print)
ISSN 1683-013X (online)

BIS Quarterly Review, September 2014 iii
September 2014
International banking and financial market developments
Volatility stirs, markets unshaken ................................................................................................... 1
Financial markets reflect shifting macro risks ................................................................... 1
Volatility stirs, but reverts to lows ......................................................................................... 8
Box: Volatility concepts and the risk premium ............................................................. 10
Highlights of the BIS international statistics ............................................................................ 13
Recent developments in the international banking market .................................... 13
Box: recent changes in global credit intermediation and potential risks ........... 17
Special Features
Asset managers in emerging market economies .................................................................. 19
Ken Miyajima and Ilhyock Shim
Global asset managers in EME asset markets ................................................................ 20
Correlation of investment in EME assets by asset managers .................................. 22
Box: Emerging market benchmarks .................................................................................. 28
Investor behaviour during recent bouts of market turbulence .............................. 31
Conclusion ................................................................................................................................... 33
Risks related to EME corporate balance sheets: the role of leverage and
currency mismatch .............................................................................................................................. 35
Michael Chui, Ingo Fender and Vladyslav Sushko
Recent patterns in corporate non-financial sector borrowing ................................ 36
Potential risks to the corporate sector ............................................................................. 38
Box: Currency derivatives and corporate losses: this time is different? ............. 43
Implications for local banks and the financial system ................................................ 44
Conclusions ................................................................................................................................. 46
Cross-border bank lending during the taper tantrum: the role of emerging
market fundamentals ........................................................................................................................ 49
Stefan Avdjiev and Eld Takts
Data ................................................................................................................................................ 50
Analysis ......................................................................................................................................... 55

iv BIS Quarterly Review, September 2014
Conclusion .................................................................................................................................... 59
Residential property price statistics across the globe .......................................................... 61
Michela Scatigna, Robert Szemere and Kostas Tsatsaronis
Diversity in the characteristics of house price statistics ............................................. 62
The BIS property price statistics ........................................................................................... 66
Box: Diversity of residential property price statistics: the German case ............. 68
House prices as input to economic analysis ................................................................... 70
Conclusions .................................................................................................................................. 75

Statistical Annex ........................................................................................................................... A1
Special features in the BIS Quarterly Review .......................................................... B1
List of recent BIS publications ............................................................................................ B2

Notations used in this Review
billion thousand million
e estimated
lhs, rhs left-hand scale, right-hand scale
$ US dollar unless specified otherwise
not available
. not applicable
nil or negligible

Differences in totals are due to rounding.

The term country as used in this publication also covers territorial entities
that are not states as understood by international law and practice but for
which data are separately and independently maintained.

BIS Quarterly Review, Month year 1

Volatility stirs, markets unshaken
1

Following a prolonged period of unusual tranquillity, volatility in financial markets
ticked upwards in early August. Risk appetite took a dent, as escalating geopolitical
tensions added to renewed concerns about the recovery. Equity prices fell,
especially in Europe, high-yield credit spreads widened significantly, and yields of
safe haven assets such as short-maturity German bunds fell into negative territory.
But markets quickly rode out the turbulence. By early September, they had already
recovered their losses, as worries over geopolitical tensions gave way to investors
anticipation of further monetary stimulus in the euro area.
After the spell of volatility in early August, the search for yield a dominant
theme in financial markets since mid-2012 returned in full force. Volatility fell back
to exceptional lows across virtually all asset classes, and risk premia remained
compressed. By fostering risk-taking and the search for yield, accommodative
monetary policies thus continued to support elevated asset price valuations and
exceptionally subdued volatility.
Financial markets reflect shifting macro risks
In recent months, shifting risks with respect to the economic recovery in advanced
economies proved to be a main factor affecting asset prices. From early July
onwards, emerging signs of economic weakness in euro area core countries
weighed on markets, and sentiment took another hit as banking sector worries re-
emerged in Portugal (Graph 1, left-hand panel). From late July onwards, heightened
geopolitical tensions added to these headwinds. In particular, the European Unions
announcement of new economic sanctions against Russia on 29 July sparked a sell-
off in financial markets. In this spell of market turbulence, global equities retreated,
led by a fall on the main European bourses of 59% on average. During the market
turbulence, US equities held up despite the geopolitical jitters, supported by

1
This article was prepared by the BIS Monetary and Economic Department. Questions about the
article can be addressed to Christiaan Pattipeilohy ([email protected]) and Andreas
Schrimpf ([email protected]). Questions about data and graphs should be addressed to
Alan Villegas ([email protected]).

2 BIS Quarterly Review, Month year
positive data surprises (Graph 1, centre panel), eg on labour market developments
and a strong earnings season by US corporations. Similarly, emerging market
economy (EME) equities proved by and large resilient against the market ructions.
But high-yield credit markets did experience a hiccup; credit spreads moved up,
especially in the lowest-rated market segment (see below).
Increased geopolitical stress had surprisingly little effect on energy markets. In
the spot market, oil prices actually fell by around 11% between end-June and early
September (Graph 1, right-hand panel). Market expectations for oil demand were
revised down, largely on disappointing growth in the euro area and Japan. Incoming
data from China were mixed, with that countrys manufacturing PMI registering an
18-month high in July, but falling back in August. All in all, demand factors
seemingly offset concerns over potential short-run supply disruptions.
The spell of market volatility proved to be short-lived and financial markets
resumed their rally soon afterwards. By early September, global equity markets had
recouped their losses and credit risk spreads once again consolidated at close to
historical lows. While geopolitical worries kept weighing on financial market
developments, these were ultimately superseded by the anticipation of further
monetary policy accommodation in the euro area, providing support for asset
prices.
Asset prices reflect shifting macro risks Graph 1
Equity market developments
1
Macroeconomic surprises
2
Commodity prices
3

2 June 2014 = 100 Diffusion index

200509 = 100

The dashed vertical lines in the left-hand panel indicate selected news and announcements in 2014: the suspension of trading in shares of
Banco Esprito Santo (10 July); the downing of flight MH17 in eastern Ukraine (17 July); the EU announcement of additional sanctions
against Russia (29 July); Russias announcement of import bans (7 August); and the next business day after the speech by ECB President
Mario Draghi at the annual central bank symposium in Jackson Hole (25 August).
1
Stock market indices, in local currency terms.
2
Citigroup Economic Surprise indices; defined as weighted historical standard deviations
of macro data surprises (actual releases versus Bloomberg survey median). A positive reading suggests that economic releases have on
balance been beating consensus. The indices are calculated daily in a rolling three-month window. The weights of economic indicators are
derived from relative high-frequency spot FX impacts of one standard deviation data surprises.
3
For oil, West Texas Intermediate Cushing
crude oil spot price, in US dollars/barrel; for copper, grade A copper spot price, in US dollars/metric tonne. CRB = Commodity Research
Bureau.
4
MSCI Emerging Markets equity index.
5
CRB BLS spot index.
6
CRB BLS spot foodstuffs sub-index.
Sources: Bloomberg; Datastream.
90
95
100
105
June 14 July 14 August 14
S&P 500
EURO STOXX 50
FTSE 100
EMEs
4
100
50
0
50
Q1 13 Q3 13 Q1 14 Q3 14
United States
Euro area
Japan
EMEs
75
100
125
150
2010 2011 2012 2013 2014
CRB index
5
CRB food
6
Oil
Copper

Diverging economic outlook feeds expectation of asynchronous exit
Differences in the strength of the recovery among advanced economies fed
expectations of a divergence in monetary policies. Despite additional ECB measures
in early June, persistent disinflationary pressures manifested themselves increasingly
in a number of gauges of inflation expectations. By the end of August, following
remarks by ECB President Mario Draghi at the Jackson Hole conference, market
participants started to look for further monetary stimulus, shifting forward rates
down (Graph 2, left-hand panel). Additional ECB measures were announced in early
September, including interest rate cuts and purchase programmes for asset-backed
securities and covered bonds. In response, two-year bond yields moved into
negative territory for a number of euro area sovereigns, including besides
Germany Austria, Belgium, France, Ireland and the Netherlands. These
developments contrast with expectations that the US Federal Reserve will embark
on a gradual scaling-back of monetary policy accommodation. In line with earlier
communications, investors expected the Fed to end its asset purchases by October
2014. Moreover, investors looked for the US policy rate to start rising by mid-2015
and to increase to 125 basis points over the following one-year period. Forward
rates pointed to similar expectations for the United Kingdoms policy rate.
Expectations about the path of monetary policy were also reflected in longer-
maturity bond prices. Central bank communication that interest rates would be
permanently lower in the post-crisis environment has shaped yield curve
movements. For example, the instantaneous forward rate 10 years ahead a gauge
for long-term expectations of future short-term rates dropped well below its
10-year average in both the United States and the euro area (Graph 2, centre panel).
This measure contains a term premium in addition to expectations of future short
term interest rates. Disentangling the two is difficult, but some commonly used
models suggest that the term premium has fallen significantly in 2014, reversing
Bond markets anticipate divergence in the monetary policy outlook Graph 2
Forward curves Instantaneous forward rates, 10 years
ahead
1

Long-term yields
2

Per cent Per cent

Per cent

1
Derived from the Libor/swap curve, month-end data.
2
Ten-year government bond yields.
3
Thirty-day federal funds rate
futures.
4
Three-month Euribor futures.
5
Simple average for the period shown.
Sources: Bloomberg; Datastream; national data; BIS calculations.
0.00
0.25
0.50
0.75
1.00
1.25
Q2 14 Q4 14 Q2 15 Q4 15 Q2 16
2 January 2014
2 June 2014
4 September 2014
United States:
3
Euro area:
4
1.6
2.4
3.2
4.0
4.8
5.6
6.4
04 06 08 10 12 14
United States
Euro area
Long-term average:
5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
Q3 13 Q1 14 Q3 14
United States
Germany
Japan
United Kingdom
France

part of the normalisation in the second half of last year. Anticipation of further
monetary stimulus helped to nudge long-term euro area benchmark yields to
exceptional lows. From mid-August onwards, German 10-year bond yields dropped
below 100 basis points for the first time ever (Graph 2, right-hand panel). French,
Italian and Spanish bond yields moved in lockstep, keeping intra-euro area
sovereign spreads largely unchanged. Surprisingly, in spite of expectations of a first
policy rate hike in 2015, yields on UK and US long-term government bonds fell as
well. However, the decline was considerably less than in the euro area. As a
consequence, US and UK bond spreads widened markedly relative to core euro area
benchmark bonds. In early September, the yield on US 10-year bonds stood at 2.4%,
respectively 110 and 145 basis points over French and German bond yields, with UK
bonds just marginally higher at 2.45%.
Diverging monetary policies were an increasingly important factor in currency
markets too. For a range of advanced economies, interest rate differentials vis--vis
those of the United States have shrunk considerably or even turned negative since
mid-2013 (Graph 3, left-hand panel), dulling the US dollars appeal as a funding
currency. At the same time, low yields on euro-denominated assets for the
foreseeable future led investors to revise down their expectations regarding the
euro exchange rate. In futures markets, CFTC positioning data confirmed that non-
commercial traders (eg speculative investors such as hedge funds) strongly
increased the size of their (net) short positions in the euro against the US dollar
starting in early May and, once again, from mid-July onwards (Graph 3, right-hand
panel). In spot markets, the euro dropped 6.5% against the US dollar between May
and early September, helped on its way by the additional monetary stimulus
announced by the ECB.
as do currency markets Graph 3
Rate differentials (advanced economies) against US
1
Positioning in USD/EUR futures
Per cent

000 contracts USD/EUR

1
Based on interest rate swaps. Weighted average, based on bilateral trade in 200810 (weights in parentheses), across Australia (0.02),
Canada (0.29), the euro area (0.38), Japan (0.19), Norway (0.01), Sweden (0.02), Switzerland (0.03) and the United Kingdom
(0.07).
2
Commitments of Traders Financial Futures report, US Commodity Futures Trading Commission (CFTC); a negative number
indicates aggregate net short positions in euro FX futures taken by non-commercial traders.
3
A decline indicates a depreciation of the
euro.
Sources: Bloomberg; BIS calculations.
1.8
1.2
0.6
0.0
0.6
2009 2010 2011 2012 2013 2014
Term (years): 1
3
5
10
225
150
75
0
75
1.1
1.2
1.3
1.4
1.5
2009 2010 2011 2012 2013 2014
Net positions in EUR FX, by non-commercial traders (lhs)
2
USD/EUR exchange rate (rhs)
3

Emerging market economies prove resilient to market jitters
Many emerging market economies benefited from the benign financial conditions
during most of the review period, which provided these economies with some relief
from the market pressures they had faced in mid-2013 and early 2014.
2
Portfolio
flows into EMEs recovered across all major regions early in the second quarter,
partially reversing the outflows seen in the months before (Graph 4, left-hand
panel).
The recovery of many EME asset prices suggests that the search for yield
despite the jitters in late July and early August remained in full force during the
review period. In this environment, 10-year local currency bond yields fell to around
5.8% in late August, some 50 basis points below their level in February 2014. Yields
on US dollar-denominated EME sovereign bonds fell by as much as 80 basis points,
to 4.8%. At the same time, many EME currencies stabilised (Graph 4, centre panel).

2
EME corporates and sovereigns have issued record amounts of local and foreign currency-
denominated debt securities in recent years (see B Grui, M Hattori and H S Shin, Recent changes
in global credit intermediation and potential risks, BIS Quarterly Review, September 2014,
pp 1718), which makes these borrowers potentially vulnerable to deteriorating global funding
conditions and sharp exchange rate movements. See M Chui, I Fender and V Sushko, Risks related
to EME corporate balance sheets: the role of leverage and currency mismatch, BIS Quarterly
Review, September 2014, pp 3547).
Emerging market economies recover from market pressures Graph 4
Net portfolio inflows
1
Exchange rates
2
Carry-to-risk
3

USD bn 2 January 2013 = 100

Percentage points Ratio

The dashed vertical lines in the centre panel indicate news and announcements by the Federal Reserve on 22 May and 19 June 2013 related
to the prospect of tapering of its asset purchases; the large depreciation of the Argentine peso on 23 January 2014; and the EU
announcement of additional sanctions against Russia on 29 July 2014.
1
Net portfolio flows (adjusted for exchange rate changes) to dedicated funds for individual countries and to funds for which country or
regional decomposition is available. Monthly sums of weekly data across the countries listed.
2
US dollars per unit of local currency. A
decline indicates a depreciation of the local currency.
3
Carry-to-risk is a gauge of the ex ante attractiveness of currency carry trades, and
is defined as the interest rate differential (derived from the forward discount) divided by implied FX volatility. Simple average across
currency pairs of Brazil, Chile, India, Israel, Korea, Mexico, the Philippines, Poland, South Africa, Thailand and Turkey; based on monthly
averages of daily data.
4
China, Chinese Taipei, Hong Kong SAR, India, Indonesia, Korea, Malaysia, the Philippines, Singapore and
Thailand.
5
Argentina, Brazil, Chile, Colombia, Mexico, Peru and Venezuela.
6
The Czech Republic, Hungary, Poland, Russia and
Turkey.
7
Israel, Saudi Arabia, South Africa and the United Arab Emirates.
8
Implied volatility of one-month FX options.
Sources: Datastream; EPFR; BIS calculations.
45
30
15
0
15
30
Q1 13 Q3 13 Q1 14 Q3 14
Emerging Asia
4
Latin America
5
Emerging Europe
6
Other EMEs
7
50
60
70
80
90
100
Q1 13 Q3 13 Q1 14 Q3 14
Brazil
China
India
Russia
Argentina
Turkey
South
Africa
5
7
9
11
13
15
0.1
0.2
0.3
0.4
0.5
0.6
2010 2011 2012 2013 2014
FX volatility (lhs)
8
Carry-to-risk (rhs)

Low foreign exchange volatility was an important factor, as it increased the
attractiveness of carry trades targeting EME currencies (Graph 4, right-hand panel).
In this environment, several central banks in EMEs, including those of Chile,
Hungary, Korea, Peru and Turkey, cut policy rates amid weaker activity and lower
inflation. By contrast, the central banks of Colombia, Malaysia, Russia and South
Africa raised policy rates in the face of macroeconomic or financial stability risks.
The uptick in market volatility in early August had relatively little, and even then
only a temporary, impact on most EMEs. There was no broad-based retreat by fund
investors, in contrast with the sell-off following the mid-2013 tapering
announcement.
3
Instead, portfolio flows to Latin America and emerging Europe
stagnated, whereas flows into emerging Asia kept up their momentum throughout
August. Some repricing of risk took place in EME sovereign and corporate bond
markets, but yields remained well below the levels seen earlier in 2014, hovering at
close to historical lows. In that context, EME bonds were among the top-performing
asset classes in global financial markets in 2014, with US dollar-denominated EME
debt registering a total return of almost 10% up to late July, maintaining most of its
gain during the early August volatility spell.
However, market participants did not entirely ignore country-specific risks. For
example, bond yields in Argentina increased significantly when the country
defaulted for technical reasons on its (restructured) debt, even though trading
volumes were very low. Similarly, the Russian rouble depreciated by 9% from end-
June to early September as investors worried about the escalation of the Ukraine
crisis (Graph 4, centre panel). In general, emerging Europe currencies and equities
underperformed those of other EME regions, reflecting concerns about the possible
effects of geopolitical factors on growth.
High-yield bond markets experience a hiccup
Corporate credit markets experienced a short-lived sell-off during the market jitters
from late July to early August, especially in the advanced economies sub-
investment grade segment (Graph 5, top panels). On the back of low volatility and
investors persistent search for yield, high-yield bonds posted spectacular gains
from mid-2012. At the same time, investors had absorbed an increasing amount of
debt issued by lower-rated corporate borrowers (Graph 5, bottom left-hand panel).
Around 35% of the debt issued by European and US corporates in the second
quarter of 2014 was rated below investment grade (BB+ or lower). Such a large
share of high-yield debt was by no means unprecedented for US corporates; for
instance, it was even higher during the run-up to the global financial crisis in 2004
07. But this proportion represented a record high for corporates in the euro area,
where market-based finance has traditionally played a subordinate role to bank
financing.
With junk bond spreads touching record lows in June (Graph 5, top panels),
investors became increasingly wary of high valuations. Thus, the asset class became
more vulnerable to sudden shifts in sentiment. High-yield spreads had already
ticked up by around 40 basis points between mid-June and mid-July. And, when risk

3
Bank lending to EMEs also slowed significantly in mid-2013, particularly for borrowers in countries
with high current account deficits or a large share of US dollar-denominated bank liabilities (see
S Avdjiev and E Takts, Cross-border bank lending during the taper tantrum: the role of emerging
market fundamentals, BIS Quarterly Review, September 2014, pp 4960).

appetite waned in late July and early August, the sell-off in junk bond markets
accelerated. Selling pressure in secondary markets probably originated from retail
investors redeeming a record amount of almost $20 billion from mutual funds
dedicated to the asset class between early July and early August. When expressed
as a share of net asset values, this outflow exceeds for instance the outflows from
the asset class during the mid-2013 turmoil (Graph 5, bottom right-hand panel).
From 29 July to 8 August, the bond spreads of US and European corporates in the
lowest-rated high-yield segment (CCC+ and lower) spiked by around 70 basis
points and 130 basis points, respectively. Junk bonds with better ratings (BB+ to B)
were less affected, rising only a respective 35 basis points and 48 basis points for US
and European borrowers (Graph 5, top left-hand panel). The brief spike in high-yield
spreads soon abated when institutional investors took the surge in spreads as a
buying opportunity, as indicated by market commentary. By mid-August, markets
had stabilised and spreads started to shrink again. The search for yield and
accommodative global funding conditions were still very much in place at the end
of the review period.
A hiccup in high-yield credit markets Graph 5
Credit spreads: BB+ to B
1

Credit spreads: CCC+ or lower
1

Basis points Basis points Basis points Basis points

Global corporate bond issuance
2
Flows into high-yield funds
3

USD bn

Per cent

1
High-yield option-adjusted spreads on an index of local currency bonds issued by financial and non-financial corporates.
2
Gross
issuance of corporate bonds by private non-financial corporations.
3
Weekly flows into funds that invest in US and western European
high-yield corporate bonds, as a percentage of net total assets.
Sources: Bank of America Merrill Lynch; Dealogic; EPFR; BIS calculations.
0
300
600
900
2011 2012 2013 2014
United States
270
300
330
360
Jun Jul Aug
Euro area
0
650
1,300
1,950
2011 2012 2013 2014
United States
540
650
760
870
Jun Jul Aug
Euro area
0
100
200
300
400
500
03 04 05 06 07 08 09 10 11 12 13 14
Investment grade High-yield
3
2
1
0
1
2
2010 2011 2012 2013 2014

Volatility stirs, but reverts to lows
The short-lived turbulence in late July followed a prolonged period of calm in
financial markets, during which volatility was extraordinarily subdued in all major
asset classes (Graph 6, left-hand and centre panels). By early July 2014, the implied
(forward-looking) volatility of bonds, equities, exchange rates and commodity prices
(green dots in Graph 6, centre panel) had fallen well below historical averages
(yellow dots), and in several cases even below pre-crisis levels (red dots).
The geopolitical tensions in late July triggered a temporary rise in volatility. The
VIX nudged up to 17% in early August, about 7 percentage points higher than a
month before. However, this short-lived pickup in volatility pales in comparison with
the levels seen in the 1987 crash, the 2001 dotcom bubble burst, the 200709
financial crisis or the deepening of the European sovereign debt crisis in 2011.
4
And
by late August, the VIX had already dropped back to 12% as investor risk appetite
recovered and equity markets (in particular in the United States) resumed their rally.
Volatility in other asset classes also fell back to the low levels seen in early July
(black crosses in Graph 6, centre panel).
The current low level of volatility can be partly attributed to reduced
macroeconomic uncertainty. Volatility is generally lower during business cycle
expansions than in recessions, when uncertainty about macroeconomic and firm-
specific fundamentals tends to be higher.
5
In fact, macroeconomic uncertainty has
dropped significantly since the euro area sovereign debt crisis abated in mid-2012.
At the same time, market participants growth expectations have become much less
dispersed, not only in advanced economies but also in major EMEs (Graph 6, right-
hand panel). Greater macroeconomic stability gives rise to fewer surprises
(eg regarding earnings or creditworthiness) and thus less need for portfolio
rebalancing and trading. This, in turn, reduces the volatility of actual asset price
movements.
The exceptionally accommodative monetary policy of recent years is also likely
to have played a key role in driving volatility to such exceptional lows. Policy has
had a direct effect, by compressing volatility in fixed income markets. For example,
the reduction of interest rates to the effective lower bound in all major currency
areas has pinched down the amplitude of interest rate movements at the short end
of the yield curve. More transparent central bank communication, forward guidance
and asset purchases have also removed uncertainty about interest rate changes for
medium- and longer-term maturities.
By fostering the search for yield and influencing risk appetite in the market,
accommodative policies have also had an indirect effect on volatility.
6
An
environment of low yields on high-quality benchmark bonds coupled with

4
Financial volatility (see the box for a discussion of various volatility concepts) is typically persistent,
but also tends to revert to its mean over longer horizons. Bursts of volatility linked to specific
events generally last up to several months, but also herald extended periods of relative calm. This
volatility clustering effect becomes particularly apparent when taking a long-term perspective.
5
In this regard, the current phase is akin to the two prior prolonged low-volatility episodes of 1993
2000 and 200407, which coincided with post-recession recoveries.
6
The response of financial intermediaries to accommodative monetary conditions taking greater
risk is known as the risk-taking channel of monetary policy. See BIS, 84th Annual Report,
Chapter II, June 2014, for a discussion of this monetary transmission channel.

investor confidence in the continuation of favourable market conditions is set to
foster risk-taking behaviour. This then tends to be reflected in lower hedging costs
via options, as well as a general narrowing of risk premia. In fact, the decline in
volatility across asset classes since mid-2012 has gone hand in hand with rising
asset valuations and collateral values more generally. As the capital constraints
faced by financial intermediaries are alleviated, these institutions have an incentive
to take on more risk, sending asset prices higher. This potentially creates additional
feedback effects, since return volatility tends to be dampened when valuations rise
(see box). As market participants further revise down their perceptions of (market)
risk, they may be inclined to take larger positions in risky assets, boosting prices and
pushing volatility even lower.
There are also signs that investor confidence in the continuation of low volatility
and ample funding at low rates has encouraged market participants to take
increasingly speculative positions on volatility in derivatives markets. The popularity
of such leverage-like investment strategies can be gauged from open interest in
exchange-traded volatility derivatives (see Graph A, right-hand panel, and
discussion in box). CFTC positioning data indicate further that speculative (non-
commercial) traders have significant overall net short positions in VIX futures, a sign
of their continued willingness to sell insurance to other investors against rising
volatility, despite a fairly narrow volatility risk premium.

Low volatility everywhere Graph 6
International equity markets
1
Volatility across asset classes
2
Macroeconomic uncertainty
3

Percentage points

Percentage points

Percentage points

1
Implied volatility of S&P 500, EURO STOXX 50, FTSE 100 and Nikkei 225 equity indices; weighted average based on market capitalisation
of the respective indices.
2
Data start in 2001.
3
Cross-sectional standard deviation of Consensus Economics respondents forecasts of
full-year real GDP growth, computed on a monthly basis. The series are weighted averages of current and next-year forecasts, with weights
shifting as the year progresses to proxy a 12-month-ahead forecast. See P Gerlach, P Hrdahl and R Moessner, Inflation expectations and
the great recession, BIS Quarterly Review, March 2011, pp 3951) for details.
4
Simple average for the period shown.
5
Weighted
average, based on 2005 GDP and PPP exchange rates, of the implied volatility of at-the-money options on long-term bond futures of
Germany, Japan, the United Kingdom and the United States.
6
JPMorgan VXY Global index, a turnover-weighted index of implied volatility
of three-month at-the-money options on 23 USD currency pairs.
7
Simple average of the implied volatility of at-the-money options on
commodity futures contracts on oil, gold and copper.
8
Germany, Japan, the United Kingdom and the United States.
9
Brazil, China,
Korea, Mexico and Turkey.
10
Weighted average of the economies listed, based on 2005 GDP and PPP exchange rates.
Sources: Bloomberg; Consensus Economics; BIS calculations.
0
20
40
60
02 04 06 08 10 12 14
Volatility
Long-term average
4
3 July 2014
0
10
20
30
Long-term Equity
1
Exchange Commodity
bond rate
6
futures
7
futures
5
Historical average
and 10th90th
percentile range
End-June 2007
3 July 2014
4 September 2014
0.0
0.3
0.6
0.9
93 96 99 02 05 08 11 14
Advanced economies
8, 10
Emerging markets
9, 10
Long-term
average:
4

Volatility concepts and the risk premium
Marco Lombardi and Andreas Schrimpf
Financial volatility is a measure of the variability of asset prices (or asset returns) over time. As it is a multifaceted
concept, several different volatility measures are used in practice. These fall into two broad categories: statistical
volatility (ie the volatility of the actual return distribution) and implied volatility (ie the volatility of the returns
implied in option prices).
Statistical measures of volatility are based on observed asset returns over a given time interval. This can be
done in various ways. A simple, model-free approach is to compute the standard deviation of the actual returns on a
given asset over a particular time window, so-called realised (or historical) volatility. Model-based approaches
have also been proposed: ARCH (autoregressive conditional heteroscedasticity) models, for example, assume that
the variance of returns fluctuates over time according to a specific time series model.
Implied volatility, by contrast, is derived using option prices. It thus embeds information about market
participants expectations of future movements in the price of the underlying asset as well as their appetite for
holding that risk. The best known example is the volatility index (VIX), a model-free measure of implied volatility
on the S&P 500. The VIX is constructed using option premia from a wide array of calls and puts, with a maturity of
30 days and a broad range of strike prices.
By comparing measures of implied and statistical volatility, researchers and practitioners can infer the volatility
risk premium. This premium can be thought of as the compensation demanded by investors for bearing risk related
to sharp changes in market volatility. To isolate this premium, researchers often compare implied volatility
(eg measured by the VIX) with a projection of realised volatility over the same horizon. For instance, Bekaert et
al (2013; see Graph A, footnote 3) propose a simple approach to estimate expected realised volatility over a one-
month horizon, and suggest that the difference between implied and projected realised volatility can be interpreted

Volatility risk premia and derivatives trading Graph A
Volatility risk premia
1
Positioning in VIX futures
2

Percentage points

000 contracts 000 contracts

1
Monthly averages of daily data.
2
Commitments of Traders Chicago Board Options Exchange report, US Commodity Futures Trading
Commission.
3
Estimate obtained as the difference between implied volatility (ie the volatility of the risk-neutral distribution of stock
returns computed from option prices) and projected realised volatility (ie a projection of the volatility of the statistical return distribution).
The difference between the two risk measures can be attributed to investors risk aversion; see G Bekaert, M Hoerova and M Lo Duca, Risk,
uncertainty and monetary policy, Journal of Monetary Economics, vol 60, 2013, pp 77188.
4
VIX, Chicago Board Options Exchange S&P
500 implied volatility index; standard deviation, in percentage points per annum.
5
Forward-looking estimate of realised volatility
obtained from a predictive regression of one-month-ahead realised volatility on lagged realised volatility and implied volatility. Realised
volatility is computed from five-minute-interval returns on the S&P 500 Index; standard deviation, in percentage points per annum. See T
Anderson, F Diebold, T Bollerslev and P Labys, Modeling and forecasting realized volatility, Econometrica, vol 71, March 2003,
pp 579625.
6
A negative number indicates aggregate net short positions in VIX futures taken by non-commercial traders.
Sources: Bloomberg; Oxford-Man Institute, http://realized.oxford-man.ox.ac.uk; BIS calculations.

0
15
30
45
60
00 02 04 06 08 10 12 14
Risk premium
3
Implied volatility
4
Projected
realised volatility
5
0
100
200
300
400
120
90
60
30
0
2010 2011 2012 2013 2014
Total open interest in VIX futures (lhs)
Net positions in VIX futures, non-commercial (rhs)
6

as a proxy for investors attitude towards risk. The red and blue lines in the left-hand panel of Graph A depict,
respectively, the implied and projected realised volatility, and the green shaded area corresponds to the gauge of
time-varying risk aversion. When volatility spikes in stress episodes, investors attitude towards risk usually
follows, as investors are less willing to hold positions in risky assets or to provide insurance against sharp asset price
changes. More interestingly, estimates of the volatility risk premium have dropped quite substantially since mid-
2012, and now stand at close to pre-crisis levels.
A well known empirical regularity is that volatility tends to be negatively correlated with current and past asset
returns. In other words, volatility tends to be much higher when asset prices drop than when markets rally. The
traditional interpretation of this asymmetric relationship is the so-called leverage effect. According to this
explanation, a fall in equity prices would generally imply a rise in firms leverage, and in turn raise the riskiness of a
given stock. An alternative explanation relates the negative correlation to changes in attitudes towards risk: since
low volatility is associated with increased willingness to take on risk, a low-volatility environment is likely to be
accompanied by rising asset valuations.
Recent economic theory emphasises the endogenous nature of volatility. A prolonged period of low volatility
could paradoxically lead to a build-up in risk. One key mechanism relates to the effect of changes in volatility on
measures such as value-at risk (VaR) or Sharpe ratios, extensively used to inform risk-taking and risk management
by financial intermediaries. For a given VaR threshold, lower volatility increases the fraction of the portfolio that a
financial institution can hold in risky assets. Similarly, for a given portfolio composition and amount of capital, lower
volatility can also encourage the build-up of leverage to finance a larger asset portfolio. A second potential source
for a build-up in risk in a low-volatility environment is crowded investment behaviour by asset managers and similar
types of non-bank players. Expectations of the continuation of benign financial conditions and low volatility may
encourage these market participants to build up large positions in riskier asset classes, which further compresses risk
premia.
There are also signs of intensified speculative activity on volatility. As shown in Graph A, the size of net short
positions on VIX futures held by non-commercial traders (eg hedge funds) ie bets that volatility is going to stay
low has been increasing sharply on the back of low volatility since mid-2012. The graph also suggests that traders
were quick to scale back such short positions during the taper tantrum episode in mid-2013, during the market
turbulence of early 2014, and most recently during the market hiccup of late July to early August 2014.
If one has access to high-frequency financial data (eg five-minute returns), realised volatility has been shown to be a highly accurate
estimator of the diffusion component of the stochastic process driving the evolution of the price of the asset (Anderson et al (2003), op
cit). The pricing of options is based on the principle of no arbitrage. To make this operational, the evolution of the asset on which the
option is written has to be cast in a risk-neutral framework, in which the transition probabilities governing the evolution of the price of the
underlying asset are adjusted for investors attitude towards risk. In this sense, these risk-neutral probabilities differ from the physical
probabilities governing the evolution of observed returns on the underlying asset. See R Engle, Autoregressive conditional
heteroscedasticity with estimates of the variance of United Kingdom inflation, Econometrica, vol 50, 1982, pp 9871007. Recent work
also stresses the possibility that implied volatility may partly reflect the risk-bearing capacity of dealers acting as intermediaries in option
markets (eg N Grleanu, L Pedersen and A Poteshman, Demand-based option pricing, Review of Financial Studies, vol 22, 2009, pp 4259
99). Some market participants use the term risk appetite to characterise investors attitudes towards risk. The term risk aversion is
more technical, and is related by some to innate agents preferences. See F Black, Studies of stock price volatility changes,
Proceedings of the 1976 Meetings of the American Statistical Association, Business and Economic Statistics Section, 1976, pp 17781. This
interpretation of the observed negative correlation between volatility and asset returns was first put forth by R Pindyck, Risk, inflation and
the stock market, American Economic Review, vol 74, 1984, pp 33551. See eg H S Shin, Risk and liquidity, Oxford University Press,
2010; and T Adrian and N Boyarchenko, Intermediary leverage cycles and financial stability, Federal Reserve Bank of New York, Staff
Reports, no 576, August 2012. This has been labelled, inter alia, the financial instability paradox or also volatility paradox; see
C Borio and M Drehmann, Towards an operational framework for financial stability: fuzzy measurement and its consequences, BIS
Working Papers, no 284, June 2009; and M Brunnermeier and Y Sannikov, A macroeconomic model with a financial sector, American
Economic Review, vol 104, no 2, 2014. See K Miyajima and I Shim, Asset managers in emerging market economies, BIS Quarterly
Review, September 2014, pp 1934.

BIS Quarterly Review, September 2014 13

Highlights of the BIS international statistics
1

The BIS, in cooperation with central banks and monetary authorities worldwide, compiles and
disseminates several data sets on activity in international financial markets. This chapter
summarises the latest data for the international banking market, available up to end-March
2014, and analyses those for the international debt securities market, available up to end-June
2014.
Between end-December 2013 and end-March 2014, the cross-border claims of BIS
reporting banks rose by $580 billion, the first substantial quarterly increase since
late 2011. The expansion was broadly spread across countries and sectors. Claims
on both advanced and emerging market economies grew considerably. At
individual country level, claims vis--vis borrowers in China increased the most,
taking the outstanding stock of cross-border claims on the country above $1 trillion
at end-March 2014 (including inter-office transactions by Chinese and other banks).
Claims on the rest of Asia, Latin America and Africa and the Middle East also
increased, albeit at a more modest pace. By contrast, claims on emerging Europe
fell for a fourth consecutive quarter.
The box on securities statistics highlights the recent change in the pattern of
global credit intermediation from bank credit to bond finance and outlines potential
risks. It focuses on emerging market corporate borrowers, whose issuance of
international debt securities has surged in recent years, especially on a nationality
basis that includes issuance by offshore subsidiaries. The box considers the likely
implications of the lengthening of bond maturities for rollover risk. The need to roll
over these debts in any future capital market disruptions has the potential to
impose significant costs on the economy through slower growth and tighter
financial conditions.
Recent developments in the international banking market
The first quarter of 2014 saw the first substantial expansion in international banking
activity since late 2011. The cross-border claims of BIS reporting banks rose by
$580 billion between end-December 2013 and end-March 2014. While not enough

1
This article was prepared by Stefan Avdjiev ([email protected]). Statistical support was
provided by Sebastian Goerlich.

14 BIS Quarterly Review, September 2014
to offset the preceding quarterly declines, the rise caused the annual rate of
contraction in cross-border claims to slow from 4.0% as of end-2013 to 2.0% as of
end-March 2014.
2

The upturn in overall activity in the first quarter of 2014 was boosted by the
first quarterly increase in cross-border interbank claims since late 2011. Interbank
claims, which in the locational banking statistics
3
capture positions with related
offices as well as unrelated banks, rose by $298 billion. As a result, the annual rate
of contraction of cross-border interbank activity slowed from 5.3% at end-
December 2013 to 2.9% at end-March 2014. As usual, the expansion in interbank
claims consisted almost entirely of loans (Graph 1, left-hand panel). Cross-border
claims on other banks and related offices in the euro area, including intra-euro area
activity, increased by $104 billion in the first quarter of 2014, ending a run of seven
consecutive declines.
Matching the rise in cross-border interbank lending to the euro area, euro-
denominated cross-border claims also grew during the first quarter of 2014. The
$163 billion quarterly increase was the first since early 2012. It reduced the annual
rate of contraction from 10% at end-2013 to 7.7% at end-March 2014.
Overall cross-border claims on non-banks mainly non-bank financial
institutions, governments and non-financial corporations also grew between end-
December 2013 and end-March 2014. The $282 billion rise was the largest since
late 2010. While loans accounted for the bulk of the increase in the first quarter of

2
Annual percentage changes are calculated as the sum of exchange rate- and break-adjusted
changes over the preceding four quarters divided by the amount outstanding one year earlier.
3
The locational banking statistics are structured according to the location of banking offices and
capture the activity of all internationally active banking offices in the reporting country regardless
of the nationality of the parent bank. Banks record their positions on an unconsolidated basis,
including those vis--vis their own offices in other countries.
Cross-border claims of BIS reporting banks
By type of instrument and sector of counterparty Graph 1
Annual growth rate
1
Share of outstanding claims
Per cent

Per cent

1
Calculated as the sum of exchange rate- and break-adjusted changes over the preceding four quarters divided by the amount
outstanding one year earlier.
2
Other claims include equities, derivatives and other financial claims not classified as loans, deposits or debt
securities.
Source: BIS locational banking statistics by residence (Table 1).
20
10
0
10
20
2007 2008 2009 2010 2011 2012 2013 2014
Loans to banks
Loans to non-banks
0
15
30
45
60
96 98 00 02 04 06 08 10 12 14
Holdings of debt securities and other claims on banks
2
Holdings of debt securities and other claims on non-banks
2

2014 ($196 billion), BIS reporting banks continued to increase their holdings of non-
banks securities as well ($86 billion). However, the annual growth rate of their
securities purchases remained lower than the growth rates seen in 2012 and 2013
(Graph 1, left-hand panel, purple line).
Cross-border lending to non-banks in the United States was especially strong,
expanding by $73 billion in the first quarter of 2014. The BIS consolidated banking
statistics
4
on an immediate borrower basis suggest that the majority of this rise
reflected increased lending to the US non-bank private sector, whose share of all
international claims on the United States rose by a full percentage point (from
52.0% to 53.0%) between end-December 2013 and end-March 2014. By contrast,
the share of claims on the US public sector declined by 0.7 percentage points
(from 23.5% to 22.8%) during the same period.
Notwithstanding the latest quarterly upturn in cross-border interbank loans,
their share of overall cross-border claims has gradually declined, from roughly two
thirds at end-1995 to 46% at end-March 2014 (Graph 1, right-hand panel, red line).
By contrast, the share of cross-border loans to non-banks, which remained stable at
around 20% between end-1995 and end-2005, has grown over the past few years,
reaching 24% at the end of March 2014 (blue line). In the meantime, the share of
banks cross-border holdings of securities issued by non-banks more than doubled
between end-1995 and end-2005 (from 8% to 19%) before retreating to 17% as of
end-March 2014 (purple line). Finally, the share of banks cross-border holdings of
securities issued by other banks increased from 3% at the end of 1995 to 13% at the
end of March 2014 (yellow line).
Credit to emerging market economies
Cross-border lending to emerging market economies continued to grow in the first
quarter of 2014. The $166 billion increase for the period brought the annual growth
of claims on emerging markets to 10% at end-March 2014.
5

Just as in the preceding several quarters, the main driver of the expansion
during the first quarter of 2014 was lending to China, which rose by $133 billion.
This took the annual growth rate of claims on China to 49%. At end-March 2014, the
outstanding stock of cross-border claims on China stood at just over $1 trillion, with
interbank claims accounting for almost three quarters ($726 billion) of that amount.
The consolidated banking statistics indicate that a significant share of the reported
cross-border claims on China over $400 billion is booked by banks
headquartered outside the BIS reporting area through their offices in BIS reporting
countries: for example, through Chinese banks located in Hong Kong SAR.
6, 7
By

4
The consolidated banking statistics are structured according to the nationality of reporting banks
and are reported on a worldwide consolidated basis, ie excluding positions between affiliates of the
same banking group. Banks consolidate their inter-office positions and report only their claims on
unrelated borrowers.
5
For an analysis of the main drivers in the deceleration in cross-border bank lending to emerging
market economies (EMEs) during the 2013 taper tantrum using data from the Stage 1 enhanced BIS
international banking statistics, see S Avdjiev and E Takts, Cross-border bank lending during the
taper tantrum: the role of emerging market fundamentals, BIS Quarterly Review, September 2014,
pp 4960.
6
A list of countries in the BIS reporting area is available on the BIS website:
www.bis.org/statistics/rep_countries.htm. China does not report international banking statistics to
the BIS. However, the locational statistics and consolidated statistics on an immediate borrower

comparison, the consolidated foreign claims of banks headquartered inside the BIS
reporting area on Chinese residents totalled $796 billion on an ultimate risk basis at
end-March 2014.
8

Claims on the rest of emerging Asia also grew (by $21 billion) during the first
quarter of 2014, but at a much more moderate pace than those vis--vis China
(7.8% on an annual basis).
Just as in the preceding quarter, claims on Latin America and the Caribbean
grew at a very modest pace during the quarter. Claims on the region rose, but only
by $8.9 billion. Internationally active banks increased their claims on Brazil by
$7.6 billion. By contrast, cross-border lending to residents of Mexico contracted by
$3.7 billion.
Cross-border lending to emerging Europe fell for the fourth consecutive
quarter. The $15 billion contraction, which was larger than any of the three that
preceded it, caused claims on the region to fall by 1.9% on an annual basis. At
individual country level, claims on Turkey and Poland shrank the most (by
$5.3 billion and $4.8 billion, respectively). By contrast, claims on Hungary rose by
$1.3 billion in the first quarter of 2014.
As the geopolitical uncertainty surrounding Russia and Ukraine increased in the
first quarter of 2014, internationally active banks reported declines in (the US dollar
value of) their consolidated foreign claims on both countries. The consolidated
banking statistics on an ultimate risk basis reveal that the outstanding stock of
foreign claims on Russia declined from $225 billion at end-December 2013 to
$209 billion at end-March 2014, while those on Ukraine dropped from $25 billion to
$22 billion.
9
However, these reported reductions in foreign claims were amplified
by the sharp depreciation in the value of these countries currencies against the US
dollar during the first quarter of 2014, which reduced the US dollar value of claims
booked in local currencies through local affiliates.
10
The locational banking statistics
indicate that, on an exchange rate-adjusted basis, cross-border claims on residents
of Russia remained virtually unchanged in the first quarter of 2014 ($0.3 billion),
while those on Ukraine declined by $1.5 billion (15% on an annual basis).

basis capture the positions of Chinese and other banks headquartered outside the BIS reporting
area to the extent that these positions are booked through offices in a BIS reporting country.
7
See P McGuire and A Van Rixtel, Shifting credit patterns in emerging Asia, BIS Quarterly Review,
December 2012, pp 1718.
8
Foreign claims comprise cross-border claims and local claims, where local claims refer to credit
extended by foreign banks affiliates located in the same country as the borrower. BIS reporting
banks local claims on China denominated in renminbi totalled $235 billion at end-March 2014.
9
For the outstanding stocks of foreign claims on Russia and Ukraine at end-2013, broken down by
the nationality of the lending bank, see Graph 2 in Bank for International Settlements, Highlights of
the BIS international statistics, BIS Quarterly Review, June 2014, pp 111.
10
The depreciation of a currency against the US dollar results in a decline in the reported US dollar
value of the outstanding stock of claims denominated in that currency.

Recent changes in global credit intermediation and potential risks
Branimir Grui, Masazumi Hattori and Hyun Song Shin
Global credit intermediation has seen a marked shift from the banking sector to the debt securities market,
reflecting the combination of subdued cross-border banking activity in 200912 and portfolio investors search for
yield in a low-yield environment. The shift is particularly notable for emerging market corporate borrowers, where
the amounts outstanding by nationality (based on location of headquarters) exceed the traditional residence-based
amounts (Graph A, left-hand panel), indicating substantial offshore issuance not captured in conventional external
debt statistics. See Chui et al (2014) for a more detailed description of borrower characteristics and exposures.
The average maturity of international debt securities newly issued by EME corporate borrowers has been
increasing. The centre and right-hand panels of Graph A depict the amount and average maturity of gross issuance
of international debt securities by EME non-bank corporates. The size of the bubble represents the dollar amount
while the height of the bubble represents average maturity weighted by issuance amount. Graph A shows increased
issuance amounts in recent years (larger bubbles) as well as longer maturities, especially for the residence-based
series. In terms of the stock of outstanding debt securities by residence, the average remaining maturity excluding
money market instruments now exceeds eight years.
Longer maturities mitigate rollover risk for borrowers, but this is achieved at the expense of greater sensitivity
of bond prices to yield changes greater duration risk for the lender. Although many bond investors such as
institutional investors do not operate with much leverage, if any, they may be susceptible to leverage-like behaviour
if they are subject to risk limits, are sensitive to relative performance metrics, or engage in dynamic hedging to
offset losses from option selling and other return-enhancing practices. To the extent that investor reactions amplify
market disruptions, longer maturities may introduce new vulnerabilities with the potential to affect the availability
and cost of finance. These issues are examined in more detail in Miyajima and Shim (2014).
The projected redemption schedules for EME non-bank corporates based on international debt securities
issued to date indicate that redemptions will peak in 201718. For international debt securities, redemptions in US
dollars take the lions share (Graph B, left-hand and centre panels).
This represents a potential vulnerability. Many emerging market borrowers, such as commodity exporters, will
have US dollar cash flow to offset coupon payments, as detailed in Chui et al (2014). Nevertheless, currency

Annual gross issuance and maturity (weighted average)
Emerging market non-bank corporations
1
Graph A
Amount outstanding
2
Nationality basis
3
Residence basis
3

USD bn Years

Years

1
Non-financial corporations and non-bank financial corporations from Brazil, Bulgaria, Chile, China, Colombia, the Czech Republic, Estonia,
Hong Kong SAR, Hungary, Iceland, India, Indonesia, Korea, Latvia, Lithuania, Malaysia, Mexico, Peru, the Philippines, Poland, Romania,
Russia, Singapore, Slovenia, South Africa, Thailand, Turkey and Venezuela.
2
End-June 2014.
3
Nationality basis refers to firms with
headquarters in the selected countries. Residence basis refers to firms resident in the selected countries. The size of bubbles reflects relative
volume of annual gross issuance of long-term securities. Gross issuance for 2013: $265 billion (nationality basis) and $152 billion (residence
basis). For 2014, data up to June.
Sources: Bloomberg; Dealogic; Euroclear; Thomson Reuters; Xtrakter; BIS.

0
250
500
750
1,000
Nationality Residence
4
6
8
10
12
00 02 04 06 08 10 12 14
4
6
8
10
12
00 02 04 06 08 10 12 14


Projected redemptions
1

Emerging non-bank corporations, in billions of US dollars Graph B
Nationality basis Residence basis Projected redemptions of securities
(nationality basis)
2

1
Projected redemptions on international debt securities issued to date of non-bank corporations in the economies listed in Graph A.
Nationality basis measures projected redemptions of corporates with headquarters in the selected countries. Residence basis measures
redemptions of corporates resident in the selected countries. Domestic currency refers to domestic currency in nationality
terms.
2
Projected redemptions of long-term securities issued on any market (source: Dealogic) and securities that met the BIS criteria
(international debt securities).
Sources: Dealogic; Euroclear; Thomson Reuters; Xtrakter; BIS.

hedging of principal redemptions may be incomplete even for such borrowers. To the extent that market disruptions
are made worse by duration risk, lengthening maturities may have the perverse effect of exacerbating rollover
problems if the US dollar debt securities market shuts down in a distress episode. It should be remembered that the
dollar market for emerging market issuers all but closed between September and October in 2008 and in January
2009, and again for several weeks in mid-2013.
International debt securities constitute around 20% of total debt securities in terms of amounts
outstanding.,

Graph B (right-hand panel) shows the projected redemptions of international debt securities
relative to redemptions of all debt securities, including domestic issuance. Domestic debt securities are
denominated mainly in the domestic currency, eliminating the immediate risk from currency mismatch. However, as
international investors have ventured further into the market for domestic debt securities, even this market may not
be immune to disruptions arising from exits by global investors.
While solvency is less of a problem with non-leveraged investors, the broader macroeconomic impact of capital
market disruptions may impose significant costs on the economy through slower growth and tighter financial
conditions. For these reasons, recent trends in the debt securities market and the changing pattern of credit
intermediation merit attention from policymakers.
M Chui, I Fender and V Sushko, Risks related to EME corporate balance sheets: the role of leverage and currency mismatch, BIS Quarterly
Review, September 2014, pp 3547. The BIS classifies a debt security as international (IDS) if any one of the following characteristics is
different from the country of residence of the issuer: country where the security is registered, law governing the issue, or market where the issue is
listed. All other securities are classified as domestic (DDS). See B Grui and P Wooldridge, Enhancements to the BIS debt securities statistics, BIS
Quarterly Review, December 2012, pp 6376. K Miyajima and I Shim, Asset managers in emerging market economies, BIS Quarterly Review,
September 2014, 1933. Investments in debt securities by non-residents account for roughly one quarter of the stock of global debt
securities, with domestic investors accounting for the remaining 75%. See B Grui and A Schrimpf, Cross-border investments in global debt
markets since the crisis, BIS Quarterly Review, March 2014, pp 1819. For the economies listed, share of outstanding IDS by residence at end-
2013 was around 1% in long-term series on credit to the private non-financial sector. See C Dembiermont, M Drehmann and S Muksakunratana,
How much does the private sector really borrow a new database for total credit to the private non-financial sector, BIS Quarterly Review, March
2013, pp 6581. It is reasonable to assume that large non-financial corporates in the economies tap international debt securities market.
0
30
60
90
120
16 18 20 22 24 26 28 30
US dollar-denominated
Domestic currency-denominated
0
30
60
90
120
16 18 20 22 24 26 28 30
Non-US dollar
foreign currency-denominated
0
150
300
450
600
16 18 20 22 24 26 28 30
Issued on any market
International debt securities

Asset managers in emerging market economies
1

The turbulence in emerging market economies (EMEs) in mid-2013 has reminded policymakers
and investors of the importance of actions by large asset managers for relatively small and
illiquid EME asset markets. The presence of asset managers in EMEs has grown considerably,
and the concentrated use of benchmarks and the directional co-movement of investor flows can
generate correlated investment patterns that may create one-sided markets and exacerbate
price fluctuations. Indeed, we provide evidence showing that, during the past two years, investor
flows to asset managers and EME asset prices have reinforced each others movements.
JEL classification: G11, G15, G23.
In recent years, asset managers have become important players in emerging market
economy (EME) asset markets. This shift has coincided with a prolonged period of
very low interest rates in advanced economies, which has led investors to look for
higher-yielding (but riskier or less liquid) assets in the hope of greater returns. This
has resulted in a sharp rise in bond issuance by EME entities, especially EME
corporations.
Increased foreign investment in EME assets can boost investment and growth
in EMEs and help develop their financial markets. But this comes at a price.
Successive crises have shown that foreign investors may destabilise EME asset
markets accentuating both booms and busts. For example, during the latest
period of turbulence, which started in May 2013, large capital outflows from many
EME bond markets drove up bond yields and led to a sharp depreciation of EME
currencies. This increased the cost to EME borrowers repaying foreign currency debt
and exposed, in particular, EME corporations to refinancing risks (Chui et al (2014)).
In this article, we find that asset managers in EME asset markets tend to behave
in a correlated manner and that investment flows to asset managers and asset
prices amplify each others fluctuations. In particular, we first show that asset
managers have increased their presence in EMEs over the past decade. Second, the
use of common/similar portfolio benchmarks and the directional co-movement of
ultimate investor flows can generate correlated investment decisions by asset
managers. Moreover, we provide evidence showing that, during the past two years,
investor flows to asset managers and EME asset prices reinforced each others

1
The views expressed in this article are those of the authors and do not necessarily reflect those of
the BIS. We are grateful to Claudio Borio, Michael Chui, Herv Hannoun, Soyoung Kim,
Madhusudan Mohanty, David Ng, Frank Packer, Sergio Schmukler, Hyun Song Shin and Christian
Upper for discussion and comments. We also thank Tracy Chan, Steven Kong, Jimmy Shek and
Agne Subelyte for research assistance.
Ken Miyajima
[email protected]
Ilhyock Shim
[email protected]

directional movements. It is therefore important that policymakers understand how
actions by asset managers and their ultimate investors can potentially destabilise
EME asset markets.
Global asset managers in EME asset markets
Global investors are generally interested in holding EME assets but often do not
want to or cannot hold them directly, which leads them to rely on asset managers
to access these markets. In particular, retail investors typically invest in collective
investment vehicles, managed by a small number of large asset management
companies (AMCs). Institutional investors also invest in such vehicles or use
separate (segregated) accounts managed by AMCs. As delegated investors, asset
managers perform an agency function for these investors, who are the ultimate
owners of the assets. Revenues of AMCs largely come from the fees they charge
their clients for investment services.
Collective investment vehicles
2
offer different fund structures and allocation
strategies to investors (Table 1). There are two main fund structures: open-end and
closed-end funds. Open-end funds allow investors to add or redeem investments.
Exchange-traded funds (ETFs) are a form of open-end fund that is traded on
exchanges. Closed-end funds, when set up, issue a fixed number of shares that are
traded on secondary markets. Open-end mutual funds and ETFs are larger than
closed-end funds in terms of both the number of funds and the total amount of
assets under management (AUM).

2
Terminologies differ by country. For instance, open-end funds are called mutual funds in the United
States and UCITS (undertakings for collective investment in transferable securities) in the European
Union. In this article, we use the term mutual funds for collective investment vehicles, which
include traditional open- and closed-end funds managed by AMCs, but not hedge funds.
Types of collective investment vehicles investing in bonds and equities
1

Table 1
Bond funds investing in: Equity funds investing in:
Advanced
economies
Emerging market
economies
Advanced
economies
Emerging market
economies
Fund structure
Open-end mutual funds 90.9 91.5 83.9 72.9
Closed-end mutual funds 2.2 1.1 0.9 1.8
Exchange-traded funds 6.9 7.3 15.2 25.3
Investor
Institutional
2
50.8 55.0 48.4 62.4
Retail 48.6 44.7 50.8 37.2
Strategy
Actively managed 85.7 92.8 69.3 69.5
Passively managed 14.2 7.2 30.7 30.5
1
The share of total net assets as of end-May 2014, in per cent.
2
In the EPFR database, institutional investor funds are defined as
funds targeting institutional investors only or those with the minimum amount of $100,000 per account.
Sources: EPFR; authors calculations.

Asset managers can also be divided into two types by investment strategy. The
first is actively managed funds, whose allocation or investment decisions are not
tied directly to a benchmark index. The second is passively managed funds, which
seek to mirror or closely track the performance of a particular benchmark index.
Table 1 shows that around 7090% of the EME assets managed by collective
investment vehicles have mandates to follow an active investment strategy. That
said, the share of passively managed funds including almost all ETFs has increased
in recent years, particularly following the 2008 financial crisis. Most notably, owing
to their low cost, ETFs have gained popularity among investors interested in EME
assets.
The size and concentration of the global asset management industry have
increased over the past decade. First, as ultimate investors assets continue to grow,
the total amount of AUM by the largest 500 AMCs doubled from $35 trillion in 2002
to almost $70 trillion in 2012 (Graph 1, left-hand panel).
3
Many AMCs saw large
declines in AUM following the collapse of Lehman Brothers in 2008, which some
observers attributed mainly to valuation effects, rather than redemptions. But as
markets rebounded, so did AUM.
Second, the global asset management industry is dominated by a small number
of large players. In 2012, the share of the largest 20 AMCs was about 40% of the
total AUM of the largest 500 companies, or $28 billion (Graph 1, left-hand panel).
Furthermore, the top five accounted for 18% of total AUM, with the largest player
representing nearly 6% of the total.
To gauge the activity of asset managers in EMEs, we rely on the commercial
database of mutual funds and ETFs provided by Emerging Portfolio Fund Research
(EPFR) Global.
4
This is one of the most widely used data sources of foreign fund
flows to EMEs, owing partly to the high frequency of the data. The data are typically
used to track retail and institutional investment flows. That said, the individual
institutional investors represented by the EPFR data are believed to be relatively
small in size compared with those that use the major global custodians.
5
Therefore,
the EPFR institutional flows may not be a very good proxy for the entire universe of
institutional investment flows.
EPFR data indicate that the AUM of dedicated EME funds have grown strongly.
The total AUM of EME equity and bond funds increased from the pre-Lehman peak
of $900 billion in October 2007 to $1.4 trillion in May 2014 (or about 8.5% of total
EME equities and bonds outstanding). In particular, the total AUM of EME equity
funds increased from $702 billion at the end of 2009 to $1.1 trillion at the end of
2013, and those of EME bond funds quadrupled from $88 billion to $340 billion
over the same period.

3
The total includes AMCs which are either independent or owned by banks or insurers. Also included
are some independent hedge funds, which represent only a small share. Of the top 20, nine AMCs
were bank-owned, eight independent and three insurer-owned. In some cases, an institutions total
assets could be larger than those under its asset management function.
4
As of end-May 2014, the total amount of AUM by mutual funds and ETFs in the EPFR database
globally amounted to about $23 trillion. EPFR Global collects data on total net assets and flows by
investor (retail and institutional), country and asset type. The database covers some 11,000 equity
funds and about 4,500 fixed income funds, among others. The coverage of EME assets is generally
not very comprehensive, and particularly weak for EME bonds.
5
EPFR Global defines institutional investor funds as funds targeting institutional investors only or
those with the minimum amount of $100,000 per account.

The picture is more mixed when it comes to the share of EME funds in all
mutual funds and ETFs. The share allocated to EME bonds (foreign currency and
local currency) rose from 4.1% to 6.4% (Graph 1, right-hand panel). By contrast, the
share of EME equity funds out of the total AUM of all equity funds declined from
12.3% to 9.4% over the same period. But these changes are likely to reflect
valuation effects. During 200913, major EME bond indices outperformed advanced
economy (AE) bond indices by about 10% in US dollar terms, accounting for some
of the sharp increase in allocations to EME local currency bonds. Meanwhile, major
EME equity price indices underperformed their counterparts in AEs by about 30% in
US dollar terms. After accounting for these changes, allocations to EME equities may
well have risen.
The large size and concentration of AUM of asset managers in relatively small
and illiquid EME asset markets are a potentially important source of concern. Any
decision by asset managers with large AUM to change portfolio allocation can have
a major impact on EME asset markets that are relatively small. For instance, a
1 percentage point reallocation of AUM of the largest 500 AMCs discussed above,
the total size of which amounts to about $70 trillion, would result in additional
portfolio flows of $700 billion to EMEs.
6
This is larger than the very large amounts
of gross portfolio outflows in 2008 ($246 billion) and inflows in 2012 ($368 billion)
documented in IMF (2014). The implication could be more significant in smaller and
more open economies than in those with larger but less open financial systems.
Correlation of investment in EME assets by asset managers
Whether or not asset managers have a destabilising effect on markets depends in
large part on how correlated their behaviour is. There are two broad channels
through which asset managers might behave very similarly to each other. The first

6
The estimated $700 billion may be considered as an upper bound to the extent that many of the
funds are located in advanced economies and are likely to trade EME assets in offshore markets.
Asset management industry: size, concentration and allocation to EME assets Graph 1
Largest 500 companies
1
EME dedicated mutual funds
2

USD trn

% of total

1
Not all of them are asset managers; see footnote 3 in main text.
2
All EPFR funds.
Sources: EPFR; Towers Watson; authors calculations.
0
15
30
45
60
02 03 04 05 06 07 08 09 10 11 12
Top 500 of which: largest 20
0.0
2.5
5.0
7.5
10.0
2009 2013
Equities Foreign and local currency bonds

channel is described in the first subsection below entitled Use of benchmark
indices, which deals with mechanisms within the asset management industry that
may induce correlated behaviour. We mainly focus on the use of common/similar
benchmarks, while touching upon other key mechanisms, such as relative and short-
horizon performance evaluation. The second channel is treated in the following
subsection entitled Ultimate investor flows, which discusses the behaviour of
ultimate investors, who may commit to or redeem funds simultaneously, thus
forcing asset managers to buy or sell.
Use of benchmark indices
Benchmarks play critical roles for asset managers. Conceptually, there are three
different levels upon which we can consider the impact of benchmarks on
correlations. First, at the individual fund level, a funds management style (active or
passive) given a benchmark matters. The more active a fund is, the less closely the
portfolio weights of the fund follow those of its benchmark. Second, at the level of
the asset management industry, the use of common benchmarks and correlation
between benchmarks is of importance. Asset managers will behave more similarly if
many of the assets are tied to the same benchmark or if the benchmarks themselves
are correlated. Finally, the method used for constructing benchmark indices can be
a source of increased volatility.
7
This subsection focuses on the first and second
aspects, while the third is discussed in the box.
How active are actively managed EME mutual funds?
For passively managed funds, including ETFs, the link between fund allocation and a
benchmark is direct. Although passive managers do not perfectly replicate their
benchmark index because of transaction costs,
8
they have very little leeway to
deviate from their benchmark. As a consequence, the larger the number of passively
managed funds indexed to a particular benchmark, the more correlated these funds
are. When ultimate investors, say, withdraw their money, passively managed funds
will sell portfolios in similar fashions, aggravating directional price movements.
But the use of benchmarks also gives rise to correlated behaviour on the part of
actively managed funds. The managers of such funds tend to be evaluated by
whether the returns of their investments match or exceed those of a particular
benchmark index. Although active managers do not necessarily fully replicate the
portfolio weights of the benchmark, the career risk of short-term underperformance
against their peers can induce them to form similar portfolios or to hug their
benchmarks as tightly as possible. This would increase the correlation of asset
managers portfolio choices. Morris and Shin (2014) and Feroli et al (2014) formalise
this point.

7
Ultimate investors typically use market capitalisation-weighted indices as benchmarks for
evaluating fund manager performance. This follows from the efficient markets hypothesis that
market portfolios constitute optimal, minimum-risk portfolios. But since component securities can
easily be mispriced in reality, investors may be putting most money in overpriced securities and
least in undervalued ones (Woolley and Vayanos (2012)). This practice tends to amplify price
distortions and in turn distort the benchmark itself.
8
Passively managed funds typically aim to track the performance of benchmark indices by using so-
called representative sampling techniques, and thus do not hold all the securities in a benchmark
index.

In order to measure the degree of activism of an EME bond fund, we calculate
the active share as the sum of the absolute value of deviations of the funds country
weights from those of the benchmark, following Cremers and Petajisto (2009):
Acti:c Sorc
=
1
2
_ |w
c
w
c
B
|
N
c=1
.
If a fund replicates a benchmark, the portfolio weight of country c in fund i, w
c
, will
be equal to the weight of that country in the benchmark, w
c
B
, giving an active share
of zero. A funds active share of 10% means a 90% overlap with the benchmark.
We focus on actively managed global and regional EME bond funds which use
JPMorgan bond indices as benchmarks and have monthly data on fund portfolio
country weights available from the EPFR database from January 2011 to June 2014.
We consider 17 EMEs (Brazil, Chile, China, Colombia, Hungary, India, Indonesia,
Malaysia, Mexico, Nigeria, Peru, the Philippines, Poland, Romania, Russia, South
Africa and Turkey) individually, and lump the other countries in a residual category.
Table 2 summarises the sample of bond funds used for our analysis and reports
the average and median value of their active share by type of fund. There are four
types by geographical coverage: global EME (70 funds), Asia ex-Japan (two funds),
emerging Europe (one fund) and Latin America (two funds). The average and
median values of the active share are largest for global EME local currency bond
funds and smallest for regional EME bond funds. Also, global EME local currency
bond funds have a greater degree of activism than global EME hard currency
(ie foreign currency) bond funds.
9

To formally measure the degree of activism of these funds, we conduct panel
regression analysis following the empirical approaches in Raddatz et al (2014). In
addition to considering all 75 EME bond funds, we classify them into the following
three categories based on the active share: closet index funds
10
with an active
share between 0% and 10%, weakly active funds with a share between 10% and
20%, and strongly active funds with a share greater than 20%.

9
The active shares at the security level could be higher. Unfortunately, we were not able to compute
them owing to data limitations.
10
A closet index fund is an asset management fund that claims to actively manage its portfolio but in
reality emulates a benchmark index.
Active share of actively managed global and regional EME bond funds Table 2
Number of
funds
Number of
benchmarks
Average value of
the active share
Median value of
the active share
Global EME local currency bond 28 3 17.7 17.3
Global EME hard currency bond 31 4 15.3 14.1
Global EME mixed currency bond 11 4
1
12.8 10.8
Regional EME bond
2
5 4 9.1 9.6
Total 75 11 15.4 14.2
1
All benchmarks in this category overlap with those for global EME local currency and hard currency bond funds.
2
Includes two Asia
ex-Japan, one emerging Europe and two Latin America funds. In the EPFR database, no currency breakdown is available for these funds.
Sources: EPFR; JPMorgan Chase; authors calculations.

In particular, we consider the following three empirical specifications:
w
ct
= 0
c
+ 0
t
+ o
1
w
ct
B
+ e
ct
(1)

w
ct
= 0
c
+ 0
t
+ o
1
w
ct
B
+ o
2
w
ct
M
+e
ct
(2)

w
ct
= 0
c
+ 0
t
+ o|w
ct-1
B
(R
ct
R
t
B
)] + [E
ct
B
+ e
ct
. (3)
Specification (1) is the baseline specification, which shows how closely w
ct
, a
fund is weight of country c at time t, follows the funds benchmark weight, w
ct
B
.
Specification (2) considers, in addition to benchmark weights, the median peer
weight, w
ct
M
, of each group of funds. The greater the extent to which fund managers
follow their peers due partly to relative performance evaluation, the higher the
coefficient on the median peer weight. Specification (3) is a variation of
specification (1), which gauges the impact of two separate components of the
benchmark weight, w
ct
B
: (i) the buy-and-hold component, w
ct-1
B
(R
ct
R
t
B
), where R
ct

and R
t
B
are the return on a countrys bond and that on the benchmark, respectively;
and (ii) the exogenous component, E
ct
B
, capturing changes in benchmarks due to
new bond issues, upgrades or downgrades, and addition to or deletion from certain
indices on the one hand, and changes in the benchmark construction methodology
on the other.
Table 3 shows that country weights of EME bond funds tend to follow their
benchmark weights quite closely. In particular, when we consider the benchmark
weight only (specification (1)), the estimated coefficient is 0.87 for the whole

Panel regressions of fund country weight on benchmark weight Table 3
Benchmark weight and median peer weight
Total (75) Closet index (16) Weakly active (41) Strongly active (18)
Benchmark
weight
0.87***
(0.04)
0.87***
(0.04)
0.95***
(0.04)
0.96***
(0.03)
0.94***
(0.04)
0.93***
(0.04)
0.65***
(0.09)
0.64***
(0.09)
Median peer
weight

0.45***
(0.06)

0.03
(0.11)

0.42***
(0.07)

0.40***
(0.14)
Observations 35,015 35,015 4,721 4,721 20,499 20,499 9,795 9,795
R-squared 0.80 0.81 0.92 0.92 0.88 0.89 0.66 0.66
Buy-and-hold and exogenous components of benchmark weights
Total (75) Closet index (16) Weakly active (41) Strongly active (18)
Buy-and-hold
component
0.87***
(0.04)
0.95***
(0.04)
0.94***
(0.04)
0.64***
(0.09)
Exogenous
component
0.72***
(0.20)
0.62***
(0.24)
0.83***
(0.29)
0.99***
(0.27)
Observations 34,053 4,628 19,912 9,513
R-squared 0.81 0.93 0.88 0.66
Explanatory variables not reported here include fund-country fixed effects and fund-time fixed effects. Median peer weight is the median
weight for a country at a point in time among all bond funds in each category. Panel estimation does not contain observations where
both fund country weight and benchmark weight are zero. Standard errors are in parentheses. The errors e
ct
are clustered at the
benchmark-time level. */**/*** denotes statistical significance at the 10/5/1% level.
Sources: Raddatz et al (2014) (for regression specifications); EPFR; JPMorgan Chase; authors calculations.

sample, and the coefficient decreases as the degree of activism increases.
11
In
specification (2), the coefficient on the median peer weight is statistically significant
and its size is about half that of the benchmark weight. This is consistent with the
hypothesis of the peer industry effect.
12
Finally, when the buy-and-hold and
exogenous components of the benchmark weight are considered separately
(specification (3)), we find that the coefficient on the buy-and-hold component for
strongly active EME bond funds is smaller than that on the exogenous component.
This result implies that these funds react more strongly to the exogenous changes
in benchmark weights than to market value-driven changes in benchmark weights.
Concentrated use and correlation of benchmarks
At the asset management industry level, the use of common benchmarks by many
EME asset funds and correlation between their benchmarks can lead them to adopt
similar asset allocation strategies. These funds are likely to move in the same
direction and react in similar ways when they face EME-related shocks.
Mutual funds and ETFs investing in EME assets tend to have less diverse
benchmarks than those investing in AE assets, in part because there are fewer
benchmark indices available. Graph 2 shows that the share of AUM linked to the top
five benchmarks is larger for each category of EME funds than for the
corresponding category of AE funds. Graph A in the box shows the list of these
benchmarks for the eight categories in Graph 2.
But even if asset managers benchmark their performance to different indices,
benchmarking may induce correlation if the indices themselves are correlated. This
may be because benchmark providers use similar methodologies in constructing

11
It is difficult to judge whether a coefficient of 0.87 means a high degree of passivity in absolute
terms. However, the coefficient on the benchmark weight in a similar regression analysis conducted
by Raddatz et al (2014) for all global and regional AE and EME bond funds over a different sample
period is greater than that reported in Table 3. Therefore, we can infer that global and regional EME
bond funds tend to follow their benchmarks more closely than global and regional AE bond funds.
12
The coefficient on the median peer weight for closet index funds is statistically insignificant because
the country weight used by the median fund is very similar to that of the other funds in the group.
Share of top five benchmarks used by bond and equity mutual funds and ETFs
As of end-May 2014, as a percentage of total net assets Graph 2
Actively managed funds Passively managed funds

0
20
40
60
Bond Equity
Advanced economy Emerging market economy
0
20
40
60
Bond Equity
Advanced economy Emerging market economy

indices or because benchmark indices tend to comprise the whole range of
investable assets rather than a smaller subset. To formally examine the correlation
across indices, we use the definition of the active share to calculate how similar two
indices from different benchmark providers are between January 2011 and June
2014. Table 4 shows that comparable global EME local currency bond indices
provided by JPMorgan Chase and Barclays Capital have more than an 80% overlap.
The low share of activism, the high concentration in the use of benchmarks and
the strong correlation between benchmarks introduce a high degree of similarity in
the behaviour of asset managers investing in EME assets. In the next subsection, we
show that the behaviour of ultimate investors adds to this correlation.
Ultimate investor flows
The co-movement of inflows from ultimate investors to mutual funds is another
important source of correlation of asset managers investments. Retail investors, in
particular, tend to move their investments in the same direction most of the time.
To arrive at this finding, we count the number of bond funds that faced weekly net
inflows or outflows each week from January 2013 to June 2014 for retail and
institutional investors. We find that retail investors have a tendency to buy (Graph 3,
left-hand panels, red bars) or sell (yellow bars) at the same time.
13
The direction
switched from net inflows to outflows in June 2013, and the degree of co-
movements remained high up to the first quarter of 2014. There appears to be less
co-movement on the part of institutional investors (Graph 3, right-hand panels). A
large number of institutional investors pulled out their funds in June 2013, but the
number of funds experiencing inflows and outflows has since remained relatively
balanced.
The finding that retail investors in particular behave very similarly to each other
is not specific to EME funds, as can be seen from the similarity between the top and
bottom panels of Graph 3. What is different is the volatility of the share of funds
facing inflows or outflows, which is greater for EME funds than for AE funds.

13
We consider global EME, global ex-US, global and US-North America bond funds with investor flow
information over the 78 weeks available in the EPFR database. There are 1,610 retail investor funds
in these four categories of bond funds out of 2,371 retail investor funds, and 1,510 institutional
investor funds in these four categories out of 2,196 institutional investor funds.
Measuring the degree of similarity between two benchmark indices Table 4
Index pair Overlap
1
Index pair Overlap
1
JPMorgan GBI-EM Broad index (20)
2

Barclays Emerging Markets Local Currency
Government Universal Bond index (25)
2

72 JPMorgan GBI-EM Broad index (20)
2

redefined Barclays Emerging Markets Local
Currency Government Universal Bond index (20)
3

81
JPMorgan GBI-EM Global index (17)
2

Barclays Emerging Markets Local Currency
Government Bond index (22)
2

62 JPMorgan GBI-EM Global index (17)
2

redefined Barclays Emerging Markets Local
Currency Government Bond index (17)
3

85
1
In per cent. 0:crlop
t
B1,B2
= 1
1
2
|w
ct
B1
w
ct
B2
|
N
c=1
.
2
Figures in parentheses are the number of countries in each index with at least
one positive value of monthly country weights since January 2011.
3
To make the JPMorgan index and Barclays Capital index
comparable, we exclude five countries not included in the former from the latter, and recalculate the weights for the remaining countries
in the Barclays Capital index.
Sources: Barclays Capital; JPMorgan Chase; authors calculations.

Emerging market benchmarks
To measure concentration in the use of benchmarks of EME asset funds, we identify a benchmark index adopted by
each fund and then aggregate the AUM of all funds using the index. Graph A shows that EME equity and bond
funds are more concentrated than their AE counterparts in their use of benchmarks. For example, two JPMorgan
EMBI Global indices are used for 38% of the total AUM of actively managed EME bond funds in the EPFR database,
while two Barclays Capital bond indices are used for 20% of the total AUM of actively managed AE bond funds.
Given the wide use of a small number of benchmark indices, their construction and composition are crucial.
Major indices have been created by broker-dealers. Indices such as Morgan Stanleys MSCI equity indices, JPMorgan
Chases EMBI and GBI bond indices and Barclays Capitals bond indices are typically constructed on the basis of
tradability, liquidity, credit rating and valuation criteria. Under these criteria, an issuer with a larger amount of
marketable securities receives a larger index weight. Other weighting schemes are also used. For example, Barclays
Capital offers bond indices that weight countries by GDP or fiscal strength.
As an illustration, the EME share in the global equity and bond indices is much smaller than their GDP share. In
particular, in the MSCI All Country World Index (ACWI), which includes 23 major AEs and 21 major EMEs, the EME share
is 10%. This contrasts with the GDP share of 36% of the same 21 EMEs out of the total GDP of the 44 economies
(Table A). The same is true for the JPMorgan GBI Broad index for 27 AEs and the JPMorgan GBI-EM Broad index for
18 EMEs. When we combine these two indices, the EME share is 6%, while their GDP share is 32% (Table B).
Recently, index providers have become progressively more independent. As an example, for the Markit iBoxx
ABF indices, Markit collects prices from multiple sources, controls for quality and then computes an average of the
price on a daily basis. Moreover, in October 2012, the market weights of the Markit iBoxx ABF indices were adjusted
to consider the following three factors: local bond market size (20%), sovereign local debt rating (20%) and the
GEMLOC Investability Indicator (60%). In July 2010, Pacific Investment Management Company (PIMCO) also
launched two GDP-weighted sovereign bond indices (PIMCO Global Advantage Government Bond Index and
European Advantage Government Bond Index) using the individual Markit country components.

AE-EME shares of MSCI indices and nominal GDP Table A
MSCI index market cap
1
Share
2
Nominal GDP
1, 3
Share
2

Advanced economies (23)
4
32,427
5
90 42,881 64
Emerging market economies (21)
6
3,7907
7
10 23,628 36
Total 36,2178
8
100 66,509 100
1
In billions of US dollars, as of 30 April 2014.
2
In per cent.
3
Retrieved from the IMF WEO database as of April 2014.
4
Australia,
Austria, Belgium, Canada, Denmark, Finland, France, Germany, Hong Kong SAR, Ireland, Israel, Italy, Japan, the Netherlands, New Zealand,
Norway, Portugal, Singapore, Spain, Sweden, Switzerland, the United Kingdom and the United States.
5
MSCI World Index.
6
Brazil,
Chile, China, Chinese Taipei, Colombia, the Czech Republic, Egypt, Greece, Hungary, India, Indonesia, Korea, Malaysia, Mexico, Peru, the
Philippines, Poland, Russia, South Africa, Thailand and Turkey.
7
MSCI Emerging Markets Index.
8
MSCI All Country World Index.
Sources: IMF; MSCI; authors calculations.
AE-EME shares of JPMorgan GBI indices and nominal GDP Table B
GBI index market cap
1
Share
2
Nominal GDP
1, 3
Share
2

Advanced economies (27)
4
24,031
5
94 45,397 68
Emerging market economies (18)
6
1,549
7
6 21,681 32
Total 25,580
8
100 67,079
8
100
1
In billions of US dollars, as of 30 April 2014.
2
In per cent.
3
Retrieved from the IMF WEO database as of April 2014.
4
Australia,
Austria, Belgium, Canada, the Czech Republic, Denmark, Finland, France, Germany, Hong Kong SAR, Hungary, Ireland, Israel, Italy, Japan,
Korea, Mexico, the Netherlands, New Zealand, Poland, Portugal, Singapore, Spain, South Africa, Sweden, the United Kingdom and the
United States.
5
GBI Broad index.
6
Brazil, Chile, China, Colombia, Hungary, India, Indonesia, Malaysia, Mexico, Nigeria, Peru, the
Philippines, Poland, Romania, Russia, South Africa, Thailand and Turkey.
7
GBI-EM Broad index.
8
Hungary, Mexico, Poland and South
Africa are included in both the GBI Broad and GBI-EM Broad indices. The total size of the four countries in terms of GBI Broad market
capitalisation is $383 billion (1.5% of the total), and that of the four countries in terms of nominal GDP is $2,257 billion (3.4% of the total).
Sources: IMF; JPMorgan Chase; authors calculations.


Key benchmarks used by bond and equity mutual funds and ETFs
As of end-May 2014, as a percentage of total net assets Graph A
Actively managed funds Passively managed funds
Advanced economy bond funds Advanced economy bond funds

Emerging market economy bond funds Emerging market economy bond funds

Advanced economy equity funds Advanced economy equity funds

Emerging market economy equity funds Emerging market economy equity funds

The GEMLOC Investability Indicator is a measure of accessibility to foreign investors based on a methodology developed by the World
Bank. The indicator scores a market on a set of 14 subfactors that are aggregated to the overall score.
DEX Universe
Bond
Citigroup World
Government Bond
Barclays
Aggregate Bond
Barclays
Municipal Bond
Barclays US
Aggregate Bond
0 5 10 15
Barclays Aggregate Bond
Barclays US
Aggregate Bond
Barclays US 15 year
Government/Credit
Float Adjusted
Barclays US Aggregate
Float Adjusted
0 5 10 15 20 25
Barclays Global Aggregate
ex-USD Float Adjusted RIC
Capped Index (USD hedged)
JPM CEMBI
Broad Diversified
JPM ELMI+
JPM GBI-EM
Global Diversified
JPM EMBI
Global
Diversified
JPM EMBI
Global
0 5 10 15
Barclays Emerging
Markets Local Currency
Liquid Government
Barclays Emerging
Markets Local Currency
Core Government
DB Emerging Market
USD Liquid Balanced
iBoxx ABF Pan-Asia
JPM EMBI
Global Core
0 10 20 30
Russell 1000
Growth
Russell 1000
Value
MSCI World
MSCI EAFE
S&P 500
0 5 10 15
Nikkei 225
MSCI EAFE
FTSE Global
All Cap ex US
CRSP US Total Market
Index
S&P 500
0 5 10 15 20
MSCI AC
Far East ex-Japan
MSCI AC World/
MSCI Emerging
Markets
MSCI AC
Asia ex-Japan
MSCI AC Asia
Pacific ex-Japan
MSCI Emerging
Markets
0 10 20 30 40
Hang Seng
CSI 300
MSCI Select
Emerging Markets
(Custom)
FTSE
Emerging Markets
MSCI Emerging
Markets
0 5 10 15 20 25

We support this conclusion through more formal statistical analysis using a
clustering measure la Lakonishok et al (1992).
14
In particular, we first divide the
number of global AE and EME bond funds facing net inflows in a given week by the
number of funds facing either net inflows or outflows, and use the ratio as the
expected share of funds facing inflows in that week. Then, we calculate the 95%
confidence interval around the expected share under a binomial distribution for that
week.
15
We calculate this interval for each of 78 weeks to control for common

Behaviour of ultimate investors of global AE and EME bond funds
January 2013 to June 2014, weekly data, as a percentage of the total number of funds in each category Graph 3
Retail investor

Institutional investor
1

Global EME bond funds (192) Global EME bond funds (209)

Global AE bond funds
2
(1,418) Global AE bond funds
3
(1,301)

Figures in brackets represent the number of funds in each category.
1
In the EPFR database, institutional investor funds are defined as funds targeting institutional investors only or those with the minimum
amount of $100,000 per account.
2
Global ex-US funds, global funds and US-North America bond funds.

14
As pointed out by Cipriani and Guarino (2014), we present indirect evidence of informational
herding in financial markets through a statistical measure of clustering. This measure cannot
distinguish between spurious herding (eg the result of a common reaction to EME-specific news)
and true herding (ie the decision to disregard ones private information to follow the behaviour of
others).
15
For binomial distribution B(n, p), n is the total number of global EME bond funds facing either net
inflows or outflows, and p the expected share of global AE and EME bond funds facing net inflows.
0
20
40
60
80
100
Q1 13 Q2 13 Q3 13 Q4 13 Q1 14 Q2 14
Share of the number of funds with net inflows
Share of the number of funds with net outflows
0
20
40
60
80
100
Q1 13 Q2 13 Q3 13 Q4 13 Q1 14 Q2 14
Share of the number of funds with zero net inflows
Midpoint of blue bars
0
20
40
60
80
100
Q1 13 Q2 13 Q3 13 Q4 13 Q1 14 Q2 14
0
20
40
60
80
100
Q1 13 Q2 13 Q3 13 Q4 13 Q1 14 Q2 14

shocks on inflows to all global bond funds each week. Finally, we check each week if
the specific share of global EME bond funds lies within this interval.
We find that retail investor flows exhibit clustering more often than institutional
investor flows. In particular, the left-hand panel of Graph 4 shows that the share of
global EME bond funds facing net retail investor inflows falls outside the 95%
confidence band in 49 out of the 78 weeks throughout the sample period. The
right-hand panel shows that institutional investor flows display similar patterns,
although investor clustering is both less frequent (30 out of 78 weeks) and less
severe.
Investor behaviour during recent bouts of market
turbulence
In this section, we analyse the behaviour of ultimate investors during the recent
episodes of financial market volatility. We examine two dimensions of investment
behaviour: whether investment flows to EME assets have differed between retail and
institutional investors, and whether these investment flows and the related asset
returns have been procyclical, ie have reinforced each others movements. It should
be noted that, as discussed in the previous section, correlated behaviour on its own
does not lead to procyclicality.
On the first dimension, the data summarised in Graph 5 suggest that retail and
institutional investors behave differently. The left-hand panel shows that retail
investors have moved their money out of mutual funds and ETFs dedicated to EME
equities since the first quarter of 2013, and out of those dedicated to EME bonds
since May 2013. By contrast, the right-hand panel shows that institutional investors
withdrew less than retail investors, even though they adjusted their net exposure in
a more volatile manner. Moreover, data from the large global custodians suggest
that relatively large institutional investors have held onto, and even increased, their
holdings of EME assets during the same period (IMF (2014)).
Measuring investor clustering in global EME bond funds
Share of the number of funds facing net inflows out of that facing net inflows or outflows Graph 4
Retail Institutional

0.0
0.2
0.4
0.6
Q1 13 Q2 13 Q3 13 Q4 13 Q1 14 Q2 14
Global EME bond funds All global bond funds
0.0
0.2
0.4
0.6
Q1 13 Q2 13 Q3 13 Q4 13 Q1 14 Q2 14
Range of 2 standard deviations

The second dimension is whether investment flows to EMEs can be a source of
cyclical instability. If investors move funds into EME asset markets by chasing
returns, inflows would increase in response to domestic asset price and exchange
rate appreciation. Such inflows could create procyclical flow-price dynamics if, in
response, domestic asset prices and exchange rates appreciated further.
To shed some light on this question, we estimate a small vector autoregressive
(VAR) model. It allows us to capture the dynamic interaction between variables
relying on a minimal number of assumptions about the overall model structure. The
VAR model can be used to compute the typical response over time of one variable
to unforecastable changes (surprises) in other variables.
Our VAR model includes two endogenous variables cumulative net inflows
and US dollar-denominated total returns.
16
The latter variables movements can be
due to those of domestic asset prices and/or exchange rates. We do not attempt to
disentangle the two effects. The model is estimated by investor type (retail and
institutional) for each type of EME asset (equities, foreign currency bonds and local
currency bonds) using weekly data that span the period from June 2012 to early
August 2014.
Our results suggest that both retail and institutional investor flows to EME
assets, and the total returns on these assets in US dollar terms, are generally
procyclical. For instance, a one standard deviation surprise rise in equity returns in
US dollar terms raises both retail and institutional equity flows by a similar
magnitude.
17
The effect is statistically significant. The response of equity returns to
a one standard deviation surprise increase in equity flows is also significantly

16
The data are expressed in terms of weekly percentage changes. Flows are assumed to be slower-
moving than total returns. That is, flows do not react contemporaneously to prices but prices can
react contemporaneously to flows.
17
Equity flows increase by about 0.5% and gradually fall back to the baseline after a few weeks.
Cumulative net inflows to EME equity and bond funds
In billions of US dollars Graph 5
Retail
1
Institutional
1, 2

1
All EPFR funds.
2
EPFR Global defines institutional investor funds as funds targeting institutional investors only or those with the
minimum amount of $100,000 per account.
75
50
25
0
25
50
Q3 12 Q1 13 Q3 13 Q1 14 Q3 14
Equity Bond (foreign currency)
75
50
25
0
25
50
Q3 12 Q1 13 Q3 13 Q1 14 Q3 14
Bond (local currency)

positive for both retail and institutional investors.
18
We obtain broadly similar
results for foreign currency and local currency bond flows.
Taken at face value, these results do not conform to the view that institutional
flows are relatively less procyclical than retail flows because institutional investors
internalise the negative consequence of trading due to their larger market share
(Chen et al (2010)). One possible explanation is that, as mentioned above, the
individual institutional investors represented by EPFR data are relatively small in size
therefore similar to retail investors compared with those that use the major
global custodians.
Conclusion
In this article, we have argued that the asset management industry could potentially
be a source of vulnerabilities for EME asset markets. In particular, there is a higher
degree of concentration in the use of benchmarks by asset managers investing in
EME assets than by those investing in AE assets. Also, we find more clustering in
ultimate investor flows in EME assets than in those in AE assets. These factors can
increase the correlation in the behaviour of asset managers and, under certain
conditions, raise the potential for one-sided markets in EMEs. Indeed, during the
past two years, investor flows to asset managers and EME asset prices reinforced
each others movements.
The possibility that asset managers and their ultimate investors destabilise EME
asset markets is obviously highly relevant for policymakers. The current prudential
regulation of the asset management industry mainly focuses on microprudential
and consumer protection aspects and does not really address any issues that give
rise to correlated behaviour across asset managers or the procyclicality of investor
flows and asset prices. Understanding these macroprudential aspects would be an
important step towards monitoring vulnerabilities created by asset managers and
designing an effective policy response.

18
Returns increase by about 1.5% and revert to the baseline after one week.

References
Chen, Q, I Goldstein and W Jiang (2010): Payoff complementarities and financial
fragility: evidence from mutual fund flows, Journal of Financial Economics, vol 97(2),
pp 23962.
Chui, M, I Fender and V Sushko (2014): Risks related to EME corporate balance
sheets: the role of leverage and currency mismatch, BIS Quarterly Review,
September, pp 3547.
Cipriani, M and A Guarino (2014): Estimating a structural model of herd behaviour
in financial markets, American Economic Review, vol 104(1), pp 22451.
Cremers, M and A Petajisto (2009): How active is your fund manager? A new
measure that predicts performance, Review of Financial Studies, vol 22(9),
pp 332965.
Feroli, M, A Kashyap, K Schoenholtz and H S Shin (2014): Market tantrums and
monetary policy, report prepared for the 2014 US Monetary Policy Forum,
28 February.
International Monetary Fund (2014): How do changes in the investor base and
financial deepening affect emerging market economies?, Global Financial Stability
Report, Chapter 2, April.
Lakonishok, J, A Shleifer and R Vishny (1992): The impact of institutional trading on
stock prices, Journal of Financial Economics, vol 32(1), pp 2344.
Morris, S and H S Shin (2014): Risk-taking channel of monetary policy: a global
game approach, mimeo.
Raddatz, C, S Schmukler and T Williams (2014): International asset allocations and
capital flows: the benchmark effect, World Bank, Policy Research Working Papers,
no 6866.
Woolley, P and D Vayanos (2012): Taming the finance monster, Central Banking
Journal, December, pp 5762.

Risks related to EME corporate balance sheets: the
role of leverage and currency mismatch
1

Corporates in many EMEs have taken advantage of unusually easy global financial conditions
to ramp up their overseas borrowing and leverage. This could expose them to increased interest
rate and currency risks unless these positions are adequately hedged. The key question is
whether EME corporate balance sheets have become more susceptible to shocks. Greater
corporate exposures could, in turn, spill over into vulnerabilities for both local banks and the
financial system more broadly. Shocks to interest or exchange rates could generate damaging
feedback loops if credit risk concerns were to prevent existing bank or bond market funding
from being rolled over.
JEL classification: D21, F31, G32.
Very low yields in advanced countries post-crisis have triggered huge investment
flows into emerging market economies (EMEs), thanks to their brighter growth
prospects. While these capital inflows have brought economic benefits, they could
make EMEs more vulnerable to external shocks if unchecked surges in credit and
asset prices were to raise the spectre of renewed boom-bust cycles (BIS (2014),
Chapter IV). Events in May 2013 and early 2014, for example, suggest that large
cross-border capital movements could cause considerable volatility in EME asset
prices and exchange rates, with implications for growth and financial stability (see
eg Avdjiev and Takts (2014)).
In this environment, the financial exposures of EME non-financial corporations,
in particular, could have wider implications. Debt issuance in foreign currencies
exposes these borrowers to rollover and foreign currency risks. If such risks
materialise, the creditworthiness of some corporations could worsen, pushing up
bond yields. Higher financing costs and tighter funding conditions for firms could
then become a drag on economic growth. Higher bond yields would also inflict
losses on holders of EME corporate debt, which include local banks and other
investors, such as global asset managers. Balance sheet pressure on corporations
could also subject banks and other intermediaries to funding stresses, as firms are
forced to withdraw their deposits. All in all, such developments could generate

1
the BIS or the CGFS. We are grateful to Claudio Borio, Dietrich Domanski, Mathias Drehmann,
Masazumi Hattori, Ulf Lewrick, Hyun Song Shin, Philip Turner, Christian Upper and the participants
of the Joint CGFS FSB-SCAV workshop in Hong Kong SAR on risks from currency mismatches and
leverage on corporate balance sheets for useful comments and discussions, and we thank Branimir
Grui, Mario Morelli and Jhuvesh Sobrun for their expert research assistance.
Michael Chui
[email protected]
Ingo Fender
[email protected]
Vladyslav Sushko
[email protected]

powerful feedback loops in response to exchange rate shocks if credit risk concerns
mean that existing bank or bond market funding is not rolled over.
Against this background, this article examines the risks related to EME
corporate balance sheets and their possible implications for the broader financial
system. To set the scene, the first section below reviews recent patterns in corporate
non-financial sector borrowing and the rising importance of cross-border financing
flows for EME corporates. On this basis, the second section then asks whether
corporate balance sheets have become more vulnerable. The third section discusses
the possible financial stability implications, followed by a short conclusion.
Recent patterns in corporate non-financial sector borrowing
In recent years, EME non-financial corporations have seen growing incentives and
opportunities to increase leverage, by borrowing in both foreign and domestic
currencies. The drivers include low interest rates and compressed term premia,
broad appreciation trends underpinning key emerging market currencies post-crisis,
and better access for EME borrowers to international markets.
2

Developments in cross-border credit are particularly noteworthy. Although
bank claims still account for the largest share of outstanding cross-border credit for

2
In this article, unless otherwise stated, the term EME is to be read as referring to the following 21
major emerging market economies: Argentina, Brazil, Chile, China, Chinese Taipei, Colombia, the
Czech Republic, Hungary, India, Indonesia, Korea, Malaysia, Mexico, Peru, the Philippines, Poland,
Russia, Turkey, South Africa, Thailand and Venezuela. Note that Hong Kong SAR and Singapore are
excluded from this group of EMEs, as many corporates headquartered in developed and other
emerging countries have raised funds there, which could blur the analysis of debt issuance by
residence and nationality in our study.
EME private cross-border bank borrowing and international debt issuance
1
Outstanding amounts Annual changes Issuance of international debt
securities

1
Private non-bank sector. Cross-border bank borrowing (by residence) also includes claims on the household sector and claims on
portfolio debt investment (implying a degree of double-counting), while international debt issuance (by nationality) includes securities
issued by non-bank financials and non-financial corporations; and these securities could be denominated in local or foreign currency.
Source: BIS consolidated banking statistics and international debt securities statistics.
0
300
600
900
1,200
04 06 08 10 12 14
International debt securities
Cross-border bank borrowing
0
50
100
150
200
05 06 07 08 09 10 11 12 13
International debt issuance
Cross-border bank borrowing
0
80
160
240
320
200508 200912
By residence
By nationality

the private non-bank sector (Graph 1, left-hand panel), a key feature of the past few
years has been the strong growth of international debt issuance by non-financial
sector corporates (Graph 1, centre panel). This stands in contrast to the pre-crisis
period (see eg Shin (2013)).
In aggregate, a significant part of the international debt of these EME
corporates is issued through their overseas subsidiaries (Graph 1, right-hand panel).
Issuance data based on issuer nationality (including issuance by the overseas
subsidiaries of the corporations headquartered in a given country) indicate that
private sector borrowers (other than banks) in major EMEs issued international debt
securities worth almost $375 billion in 200912, more than double their issuance in
the four-year period prior to the crisis.
3
Issuance in 2013 was also strong, even
though there were signs late in the year that global bank claims were recovering
too.
The scale and overall importance of recent developments in EME corporate and
wider private sector financing are also apparent from broader indicators of external
financing, such as international investment positions (IIPs). Many EMEs have seen
their net external positions shift considerably since 2008 (Graph 2, left-hand panel).
Comparison of the private sector contributions (Graph 2 right-hand panel) with
country-level IIP changes reveals that the observed decline in net IIP balances was
primarily driven by rising private sector liabilities (including those of corporates),
whereas official sector balances have been stable or rising. Note, however, that IIP

3
The issuer by nationality concept is similar to the consolidated claims concept in the BIS
international banking statistics. It is especially important in the case of EMEs such as Brazil and
China where local corporates have increased their issuance of international debt via overseas
subsidiaries including non-bank financing vehicles. By contrast, the issuer by residence concept
does not include issuance by these overseas subsidiaries, but it does include international debt
issues by other nations subsidiaries residing in the respective country.
Changes in net international investment positions
Between end-2008 and end-2012, in billions of US dollars Graph 2
Country-level Private sector only
1

AR = Argentina; BR = Brazil; CL = Chile; CN = China; CO = Colombia; CZ = Czech Republic; HU = Hungary; ID = Indonesia; IN = India;
KR = Korea; MX = Mexico; MY = Malaysia; PE = Peru; PH = Philippines; PL = Poland; RU = Russia; TH = Thailand; TR = Turkey;
VE = Venezuela; ZA = South Africa.
1
Derived by excluding all official sector (government and central bank) components from the total net international investment position.
Source: IMF, Balance of Payments Statistics.
800
600
400
200
0
200
BR TR ID IN MXRU PL TH KRCOMYCZ PE CL AR ZA PH VEHUCN
deterioration
Debtor country Creditor country
800
600
400
200
0
200
BR TR ID IN MXRU PL TH KRCOMYCZ PE CL AR ZA PH VEHUCN
deterioration
$1.2bn
Debtor country Creditor country

data are derived from residence-based statistics and usually do not include the
gross positions of overseas subsidiaries; nor do they cover domestic positions.
4

Potential risks to the corporate sector
A key question is whether these developments have made EME corporates more
vulnerable for example, to the combined effects of a slowdown in the domestic
economy, currency depreciation and rising interest rates globally. Such risks are
accentuated when leverage starts to loom too large relative to borrowers debt
servicing capacity or when foreign currency assets or revenues are insufficient to
match large foreign currency liabilities. Rising interest rates and depreciating
exchange rates will tend to raise the cost of servicing these debts, denting profits or
depleting capital cushions, unless appropriate hedges are in place.
Unfortunately, data limitations mean that such vulnerabilities are notoriously
hard to assess, especially in a cross-country context. For many EMEs, the lack of
financial accounts data at the national level means that internationally comparable
measures of corporate sector leverage are difficult to obtain. In what follows,
selected metrics are used to provide at least a partial picture.
Corporate leverage
Various measures point to rising leverage on corporate balance sheets. One such
indicator is the debt/earnings ratio as disclosed by individual firms. A recent study,
based on a sample of non-financial corporations from seven large EMEs, suggests a
more or less steady increase in corporate leverage over the last few years (Graph 3,
left-hand panel).
5
Country-level data (based on residence) on corporate debt-to-
GDP ratios appear to confirm this trend, while providing a perspective on broad
leverage levels across jurisdictions. According to this metric, corporate indebtedness
now hovers at around 100% of GDP for some EMEs (Graph 3, centre panel). Yet,
despite recent trend growth, levels vary considerably between countries and remain
modest by international standards.
Borrowing patterns have differed across countries in recent years. While
developments in some economies (eg for corporates in Latin America) appear to
reflect a more general shift from primarily domestic to more internationally
diversified funding sources (Powell (2014)), in others domestic debt rose in tandem
with external borrowing. For example, Chinese corporates (especially property
developers) now appear to be quite highly leveraged, at least in comparison with
their EME peers, and may find it challenging to manage these debt levels in an
environment of slowing growth and tightening profit margins (Bank of America
Merrill Lynch (2014)).

4
Over the past few decades, many Chinese companies have opted for a listing in overseas stock
markets (Hong Kong SAR in particular) to raise capital and hone their corporate governance. As of
end-June 2014, nearly 300 Chinese-owned or affiliated companies were listed on the main board of
the Hong Kong stock exchange with an aggregate market capitalisation of $660 billion.
5
See Bank of America Merrill Lynch (2014). BIS (2014), Chapter VI, provides additional information
based on capitalisation ratios.

Debt/earnings ratios can also reveal how rising leverage may be affecting the
capacity of firms to service their debts. A recent analysis based on firm-level data
finds that corporate debt grew faster than earnings in one third of the sample
economies between 2008 and 2012.
6
For Brazil, China and India, the average firm
required 2.5 to three years of current annual gross earnings to repay its debt in
2012, compared with two to 2.8 years in 2008. In many cases, the deterioration in
debt servicing capacity reflects a combination of rising debt loads and slowing
earnings growth. Furthermore, despite broadly stable and low interest rates over the
past five years, many EMEs have encountered a sharp increase in interest expenses
because of the larger debt loads (Graph 3, right-hand panel).
Asset composition
The nature or quality of assets acquired using the newly borrowed funds may either
strengthen or weaken a firms resilience against external shocks. Evidence on the
use of newly raised corporate funds is mixed. On the one hand, there are signs that
capital expenditure (capex) has been on the rise. Analyst estimates suggest that the
average capex of EME corporates (which includes funds used to upgrade production
capacity and acquire physical assets) has increased by almost one third over the
past few years, based on a sample of 120 EME corporate issuers.
7
In this context,
the stronger earnings prospects associated with capital spending would tend to
offset at least part of the risks associated with rising leverage.

6
See IMF (2014), which compares median corporate debt loads with earnings across 18 EMEs to
gauge the corporate sectors debt servicing capacity.
7
See Bank of America Merrill Lynch (2014).
EME corporate balance sheets: selected metrics Graph 3
Leverage ratio of EME corporations
1
Corporate sector debt in 2013
2
Annual growth rates of interest
expenses
Ratio to earnings % of GDP

Per cent

AR = Argentina; BR = Brazil; CL = Chile; CN = China; CO = Colombia; HU = Hungary; ID = Indonesia; IN = India; MX = Mexico;
MY = Malaysia; PE = Peru; PH = Philippines; PL = Poland; RU =Russia; TH = Thailand; TR = Turkey; ZA = South Africa.
1
Firm-level data from S&P Capital IQ for 900 companies in seven EMEs; simple average across countries; gross leverage = total
debt/earnings; net leverage = (total debt cash)/earnings.
2
External debt includes liabilities from affiliates, direct investments and other
sources.
Sources: IMF, Global Financial Stability Report, April 2014; Morgan Stanley; BIS calculations.
0.0
0.6
1.2
1.8
2.4
2009 2010 2011 2012 2013
Gross leverage Net leverage
0
25
50
75
100
A
R
M
X
P
H
P
E
I
D
I
N
B
R
P
L
T
R
R
U
Z
A
T
H
M
Y
H
U
C
N
Domestic bank debt
Domestic market debt
External debt
40
20
0
20
40
A
R
B
R
M
Y
C
O
T
H
I
D
P
H
I
N
T
R
C
N
R
U
M
X
Z
A
C
L
P
L
2012 Five-year average

On the other hand, due to low volatilities, Sharpe ratio-type risk-adjusted
return metrics (eg interest rate differentials adjusted for exchange rate volatility)
suggest that carry trade incentives are also strong (Graph 4, left-hand panel), which
may have tempted some corporate treasurers into more speculative activities.
One indicator of such activities may be corporate cash holdings, as measured
by the difference between gross and net leverage ratios, which have increased
markedly since 2009 (Graph 3, left-hand panel). Similarly, corporate bank deposits
have grown in a number of banking systems during this period (Graph 4, centre
panel). The fact that the trend has not abated more recently suggests that post-
crisis caution may not be the only reason why firms have increased their cash
holdings. This is in line with reports that corporates in some jurisdictions were
seeking to take advantage of international interest rate differentials by borrowing
overseas and depositing the proceeds in local banks, subscribing to money market
mutual funds or purchasing high-yielding wealth management products.
8
In Korea,
for example, deposits by private non-financial companies in trust companies and
their shares in investment funds rose by a respective 36% and 45% in the two years
to end-2013. In China, reports of over-invoicing by Chinese importers have
emerged, especially for metals and other high value-to-density articles (Graph 4,
right-hand panel). The low-cost funds raised through trade financing for these
imported articles are reportedly being used for both business investment and

Carry trade incentives, corporate deposits and over-invoicing Graph 4
Carry-to-risk ratios
1
Corporate deposits Chinas copper trade with top three
partners
2

Basis points

% of total deposits

USD bn

1
One-month interest rate differentials, adjusted for implied volatility of the respective currency pairs; base currency: US dollar.
2
Bilateral trade of copper and articles thereof (international code: HS74) between China and the worlds top three copper producers:
Australia, Chile and the United States; over-invoicing is defined as the difference between imports and corresponding bilateral
exports.
3
For corporate deposits, business deposits.
Sources: IMF, Balance of Payments Statistics; UN Comtrade database; Bank of America Merrill Lynch; Bloomberg; JPMorgan Chase; BIS
calculations.

8
According to official data, the total balance of banks wealth management products in China rose
from CNY 2.3 trillion in late 2009 to almost CNY 10 trillion in late 2013; see Financial Times (2014).
0
50
100
150
2009 2010 2011 2012 2013 2014
Brazilian real
Mexican peso
Indian rupee
Indonesian rupiah
15
20
25
30
07 08 09 10 11 12 13 14
Korea Thailand
3
1.2
0.0
1.2
2.4
8
0
8
16
93 98 03 08 13
Implied over-invoicing
Lhs:
Chinas reported imports
Partners reported exports
Rhs:

speculation.
9
Yet the overall scale of these activities is difficult to judge, and even
normal treasury operations could well lead to a substantial rise in local currency
deposits at the local banks (eg due to time-to-build and similar constraints).
Increased bond market financing
A related issue concerns the composition of funding sources and, in particular, the
rising share of bond market financing. As highlighted above, strong investor interest
has underpinned EME corporate bond markets in recent years. If investors were to
suffer a significant loss of appetite, issuing firms might face difficulty in rolling over
their outstanding debts, particularly if shifts in risk appetite coincide with a fall-off in
projected earnings.
Many of the recent EME corporate borrowers have gained access to the debt
markets, both domestic and international, for the first time. The willingness of
investors to let these issuers roll over their debt in adverse circumstances is thus
untested. BIS international debt securities data, which exclude domestic as well as
short-term issuance, suggest that the rollover needs of corporates from major EMEs
and their overseas subsidiaries will rise from around $90 billion in 2015 to a peak of
$130 billion in 201718 (Grui et al (2014)).
10
Note that these figures may
underestimate the risk of a sudden retreat by global investors, who may also hold
the domestic debt of EME corporates. For some corporations, rising debt
repayments will be particularly taxing in an environment of US dollar strengthening
(see below) and slowing domestic activity. Also, while domestic banks continue to
be the dominant source of funding for EME corporates, their ability and willingness
to help refinance market debt may be limited, particularly if risk appetite is on the
wane.
Currency mismatch
Given the elevated levels of foreign currency borrowing, currency mismatches
represent another possible source of vulnerability. Recent developments in Ukraine
are a reminder of how abruptly debt sustainability metrics can deteriorate when (in
this case, geopolitical) risks undercut the exchange rate, thus inflating the local
currency value of foreign currency liabilities. This raises the question of how far the
foreign exchange risks of rising foreign currency liabilities at EME corporates are
either financially hedged or naturally matched by foreign currency asset returns and
revenues.
11

9
For example, the World Gold Council (2014) estimates that, by the end of 2013, surplus gold
linked to financial operations in the Chinese shadow banking system could have reached a nominal
value of nearly $40 billion. See also Goldman Sachs (2013) for a detailed exposition of the
mechanics involved in the copper carry trade.
10
International issuance, which is dominated by US dollar-denominated debt, makes up about one
fifth of total debt issuance, with domestic debt accounting for the remainder. Domestic debt will
add to interest rate and rollover risks, but does not usually incur a currency mismatch risk (as
covered in more detail below).
11
Data on country-level foreign currency exposures and on how far they are hedged are generally
unavailable. Australia is an exception in that the Australian Bureau of Statistics conducts a Foreign
Currency Exposure Survey to gauge the countrys net foreign asset position (ie after taking into
account the hedging of foreign currency exposures using financial derivatives) (see Rush et

In the absence of more specific information on natural hedges, issuer sectors
may serve as an important proxy indicator. Commodity producers and manufactures
exporters, for example, earn much of their revenues in foreign currencies and are
thus likely to weather the rising debt service costs associated with currency
depreciation better than would issuers with mostly domestic incomes (eg domestic
telecoms, construction companies and utilities).
On this basis, a cursory examination of firm-level issuance data suggests that
non-financial borrowers from countries such as Brazil, Mexico, Russia and South
Africa would be more likely to have at least partially matching foreign currency
assets and liabilities, given the predominance of commodities producers and
exporters among the largest issuers. In contrast, assets and liabilities are less likely
to be matched at property developers in China or energy and utilities firms in India,
which have been among the more active international debt issuers in recent years,
pointing to possible pockets of risk in these sectors.
Companies can also manage their foreign currency exposures via derivatives.
Again, reliable data on corporate hedging activities are generally scarce, while
incentives to take open interest rate and foreign currency hedging positions have
been relatively strong recently. One issue is hedging cost and, hence, the depth of
the relevant hedging market. This might suggest that corporates from countries
such as Brazil, Korea or Mexico (which are known to have access to liquid domestic
or offshore markets that support financial hedging strategies for both currency and
interest rate risk exposures) are more likely to be hedged than their peers in, say,
China or Indonesia. Indeed, data for Mexico indicate that the volume of exchange
rate derivatives transactions picked up sharply from a monthly average of around
$12 billion in 200708 to more than $25 billion in late 2013, in line with the
observed increase in local corporates international issuance. In countries with less
developed markets, however, mismatches will often go unhedged because markets
may not be deep enough to provide appropriate and cost-effective hedging.
12

The flip side of this argument is that derivatives-related financial exposures can
change the sensitivity of corporate balance sheets in ways that may be unrelated to
what is suggested, say, by the issuers sector. In the early stages of the global
financial crisis, for example, some large corporates in Brazil, Korea and Mexico
experienced significant losses because of largely speculative positions in foreign
exchange derivatives contracts (see box). This experience shows that an abrupt
change in the exchange rate trend can conspire with complex financial exposures to
wreak significant damage on corporate balance sheets even when a firms foreign
exchange liabilities are deemed to be adequately hedged during normal times.
An additional concern is that liquidity in hedging markets can evaporate during
times of market stress. Even longer-term exposures are often hedged with more
liquid short-term contracts with the aim of reducing hedging costs. As the
respective contracts have to be rolled over regularly, this could significantly reduce
the value of financial hedges against large exchange rate fluctuations, since markets
are bound to be at their shallowest when hedging needs are greatest. In this

al (2013)). Yet the Survey is conducted infrequently (once every four years) and is residence-based
(as opposed to nationality-based).
12
For illustration, the 2013 annual report of one large Chinese property developer states: The Group
manages its currency risk by closely monitoring the movements of currency exchange rates. The
Group currently does not have a currency hedging policy [] but will consider hedging significant
currency exposure should the need arise.

Currency derivatives and corporate losses: this time is different?
The Lehman bankruptcy in September 2008 triggered a global shortage of US dollar funding, lifting the US currency.
According to one estimate, the ensuing sharp depreciation of local currencies against the dollar hit 50,000 or more
non-financial corporations with total losses of at least $30 billion, via positions on foreign exchange (FX) derivatives
contracts. This added to the uncertainty in those corporates domestic financial markets, worsening the impact of
the crisis still further. Given that many EME corporations are said to have increased their foreign exchange exposures
significantly in recent years, a key question is how vulnerable such firms are to, possibly abrupt, exchange rate
movements. This box reviews some key features of the derivatives activities of EME corporations in 2008, and
highlights differences between then and now.
One factor behind EME corporates foreign exchange losses in 2008 was the popularity of contracts with a
knock-in, knock-out (KIKO) feature. Heavy use of such contracts meant that many exporters, while insured against
modest exchange rate movements, were exposed to possibly large losses if the local currency depreciated sharply.
In a standard FX option transaction, a company (eg an exporter) with revenues mostly in foreign currency (eg in
US dollars) but with production costs in local currency buys, for a small fee (premium), a put option from a
counterparty (eg a local bank) that gives the exporter the right but not the obligation to sell its dollar income at a
specific strike price at a future time. If the domestic currency spot exchange rate at maturity is stronger than the
agreed rate, the exporter exercises the option and gets a higher income in local currency terms than it would
otherwise get at the spot rate.
Compared with this basic setup, KIKO contracts have two additional features. The first is a call option (knock-in)
held by the bank. If the reference currency (eg the US dollar) strengthens beyond a certain threshold, the knock-in
requires the exporter to sell its dollars at the strike price (ie below market rates). The second, so-called knock-out,
feature dictates that no option can be exercised by either the exporter or the bank if the dollar weakens below a
certain threshold. Both features serve to reduce hedging expenses, albeit at the cost of retaining the tail risk of
stronger currency depreciations.
A third feature is possible acceleration effects. KIKO contracts were quite often leveraged (at, say, 1:2), resulting
in payments that would double the contractual amounts. This resulted in open speculative positions on relatively
stable exchange rates. Furthermore, some EME corporations apparently purchased multiple KIKO contracts with
different banks to bypass each individual banks counterparty limit. As a result, when the US dollar rose sharply
against almost all currencies in late 2008, these corporations suffered unexpected losses owing to the knock-in
feature in their hedging operations.
Given the risk of high potential losses, a key question is why so many EME corporations used KIKO or similar
contracts to hedge their FX exposures prior to 2008. There are a number of possible explanations. By design, KIKO
features lower the premium charged by the contract seller. In that sense, many EME corporations were attracted by
the low hedging costs. This feature was particularly attractive at the time, as the major EME currencies had
experienced a long period of slow but steady appreciation against the US dollar. The resulting false sense of security
was reinforced by most commercial and official forecasts, which, up until 2007, called for this trend to continue in
the near term. Furthermore, local banks were often not the actual seller of the KIKO contracts, but merely acted as
intermediaries for foreign banks and ultimate investors, such as hedge funds. In doing so, banks earned a fee while
passing the exchange rate risk on to the ultimate contract sellers. Under such circumstances, banks may have had an
incentive to sell more contracts to increase their fee income, at least insofar as their client relationships with their
corporate customers were not jeopardised by any losses that their clients might incur.
Against this background, an important difference between now and then is that the recent prolonged period of
relatively low volatility in foreign exchange markets has been punctuated by the two tapering events, in May 2013
and January/February 2014. No major losses from corporate exposures in derivatives markets were revealed in the
aftermath of these episodes. That said, carry trade incentives have since strengthened again, and certain EME
corporations may have incurred exposures via contracts that will generate losses only at a later stage. For example,
there is anecdotal evidence of increased interest from Asian corporates in structured foreign exchange products with
KIKO-like features. In addition, for some EME hedging markets, the sellers of hedging products are often
concentrated and the markets themselves are not very liquid. Again, this tendency could exacerbate any market
reaction once the market changes direction.
See eg Sidaoui et al (2010) and Lee (2009).

context, the May 2013 and early 2014 episodes of sharp currency depreciation in
many EMEs may have served as wake-up calls, by inducing corporate treasurers to
review and trim any open currency exposures. Recent attempts by the Chinese
authorities to introduce more two-way risk into renminbi exchange rates would
seem work in the same direction.
Implications for local banks and the financial system
What are the implications of more vulnerable EME corporate sector balance sheets
for the financial system? Scope for spillovers arises from at least two channels, as
detailed below.
Liability-side exposures
One channel works through the liabilities of banks and, possibly, other financial
institutions.
13
Among these, local institutions are likely to be particularly exposed,
especially if they have come to rely on corporate deposits for part of their wholesale
funding. For deposits that are associated with corporates exploiting the carry
between local and foreign currency interest rates, the unwinding of such positions
when interest rate differentials narrow or volatilities increase will reduce these
funds. Deposits that are denominated in foreign currencies, in turn, are known to be
more procyclical than other types of deposits and may thus be subject to sudden
withdrawals by corporates facing rollover risks (Turner (2014)).
A key factor in the transmission of such effects is the shadow banking system.
In Korea, for example, assets held by non-bank financial institutions have grown at
an annual average rate of 10% since the global financial crisis. Securities companies,
in particular, have seen their assets increase more than twofold during that period.
In this context, it appears that the securities sector in Korea has accumulated
substantial claims on banks and other depository institutions. Securities firms, in
turn, finance themselves with short-term money market instruments held by the
non-financial corporate sector. To the extent that non-financial corporates issue
debt but hold the proceeds as liquid claims, they behave as surrogate
intermediaries channelling funding from global capital markets into the domestic
financial system (Bank of Korea (2014)).
Asset-side exposures
Another, more conventional, channel is the risks embodied in asset-side exposures.
Banks tend to have direct credit exposures to corporates via lending and through
counterparty risk from any derivative positions. While these exposures can be
important internationally, for example vis--vis Asia (Graph 5), local banks, again,
tend to be particularly exposed, with loans to non-banks still accounting for a large
part of domestic loans in many jurisdictions. Furthermore, since larger and more
creditworthy corporates have better access to cross-border borrowing, higher
foreign bank penetration could end up increasing the exposure of local lenders to

13
See Chung et al (2014) for a discussion of how the financial activities of non-financial corporates in
international markets could affect funding conditions and credit availability in home markets.

smaller, possibly less creditworthy, firms. That said, a mitigating factor is that
standard on-balance sheet leverage and capitalisation metrics for EME banks tend
to be rather favourable in the aggregate, which may help to reduce such risks at the
banking system level.
14

Another, less direct, source of credit risk for banks comes from broader
exposure to debt markets, eg via bond holdings. Recently, however, there have
been signs that asset managers and other buy-side investors have increasingly
displaced bank investors in corporate bond markets. This raises questions about
feedback effects if existing positions are not rolled over (see below).
Feedback effects
Working together, both types of channel can give rise to potentially powerful
feedback effects. Currency mismatches, for example, will tend to amplify both
default risk and pressure to deleverage if borrowers are hit by a depreciating local
currency. Combined with uncertainties about the true extent of such mismatches,
concerns about rising default risk could then result in a more widespread rout of
international investors, loss of market access and spillovers into domestic interbank
markets exacerbating the financial and macroeconomic impact of the initial
interest rate or foreign exchange shock.
The duration risk exposures of asset managers and other institutional investors
(the flip side of corporates attempts to issue new debt and term out existing
borrowings) are another potential source of adverse price dynamics. These might be
further amplified by the correlated behaviour of asset managers. Such herding in
bond markets can arise from the reliance on common risk management

14
A possible caveat is that EME and advanced country bank balance sheet metrics may, in fact, be
converging; see CGFS (2014) and BIS (2014), Chapter VI, for details.
Changes in global banks foreign claims on EME non-bank private sector
1

Emerging Asia Latin America Other EMEs

AR = Argentina; BR = Brazil; CL = Chile; CN = China; CO = Colombia; CZ = Czech Republic; HU = Hungary; ID = Indonesia; IN = India;
KR = Korea; MX = Mexico; MY = Malaysia; PE = Peru; PH = Philippines; PL = Poland; RU = Russia; TH = Thailand; TR = Turkey;
VE = Venezuela; ZA = South Africa.
1
Not adjusted for exchange rate movements.
Source: BIS consolidated banking statistics.
25
0
25
50
75
100
CN IN ID MY PH KR TH
2004 to 2007 2009 to 2012
25
0
25
50
75
100
AR BR CL CO MX PE VE
2004 to 2007 2009 to 2012
25
0
25
50
75
100
CZ HU PL RU ZA TR
2004 to 2007 2009 to 2012

technologies, from simultaneous buy and sell decisions due to index tracking, and
from a rush to exit due to concerns about market liquidity.
15

Conclusions
Unusually easy global financial conditions post-crisis and the ubiquitous quest for
yield have encouraged EME non-financial corporations to increase leverage and
overseas borrowing. In many jurisdictions, corporates have opted to lock in low
global interest rates and to sharply increase their international debt issuance. While
cheap funding could boost economic performance if it supports viable investment
projects, it inevitably increases the borrowers interest rate, rollover and currency
risks. Furthermore, some EME corporations may have used borrowed funds for
purely financial (ie speculative) purposes. In other cases, these external positions
may be inadequately hedged, whether through natural offsets or by the use of
financial instruments.
Overall, these factors have increased the risks facing these companies, implying
the existence of pockets of risk in particular sectors and jurisdictions. If these risks
were to materialise, adding to broader EME vulnerabilities (BIS (2014)), stress on
corporate balance sheets could rapidly spill over into other sectors, inflicting losses
on the corporate debt holdings of global asset managers, banks and other financial
institutions. This could be a source of powerful feedback loops in response to
exchange rate and/or interest rate shocks, especially if credit risk concerns
prevented the rollover of existing bank or bond market funding.

15
For a more detailed exploration of the risks arising from the increased participation of global asset
management companies in emerging markets, see Miyajima and Shim (2014). Kamada and
Miura (2014) provide a model and empirical evidence of herding by bond market investors in Japan
due to some of the same factors.

References
Avdjiev, S and E Takts (2014): International bank lending during the taper tantrum:
the role of emerging market fundamentals, BIS Quarterly Review, September,
pp 4960.
Bank for International Settlements (2014): 84th Annual Report, Time to step out of
the shadow of the crisis, Basel, June.
Bank of America Merrill Lynch (2014): Facts, fiction and FX vulnerabilities, Global
Emerging Markets Credit Research, 25 February.
Bank of Korea (2014): Financial Stability Report, April 2014.
Chung, K, J-E Lee, E Loukoianova, H Park and H S Shin (2014): Global liquidity
through the lens of monetary aggregates, IMF Working Papers, no 14/9.
Committee on the Global Financial System (2014): EME banking systems and
regional financial integration, CGFS Papers, no 51, March.
Financial Times (2014): Fall in renminbi sends out carry trade warning, 23 February.
Goldman Sachs (2013): Copper curve ball Chinese financing deals likely to end,
Goldman Sachs Commodities Research, 22 May.
Grui, B, M Hattori and H S Shin (2014): Recent changes in global credit
intermediation and potential risks, BIS Quarterly Review, September, pp 1718.
International Monetary Fund (2014): Making the transition from liquidity- to
growth-driven markets, Global Financial Stability Report, Chapter 1, April.
Kamada, K and K Miura (2014): Confidence erosion and herding behaviour in bond
markets: an essay on central bank communication strategy, Bank of Japan Working
Papers, no 14-E-6, April.
Lee, Y (2009): Korean corporations court bankruptcy with suicidal KIKO options,
Bloomberg, 24 March.
Miyajima, K and I Shim (2014): Asset managers in EM asset markets, BIS Quarterly
Review, September, pp 1934.
Morgan Stanley (2013): EM corporates: rising leverage, rising risk, Morgan Stanley
Research EM Profile, November.
Powell, A (2014): Global recovery and monetary normalisation: escaping a chronicle
foretold?, Inter-American Development Bank, Chapter 4.
Rush, A, D Sadeghian and M Wright (2013): Foreign currency exposure and
hedging in Australia, Reserve Bank of Australia Bulletin, December.
Shin, H S (2013): The second phase of global liquidity and its impact on emerging
economies, remarks at the 2013 Federal Reserve Bank of San Francisco Asia
Economic Policy Conference.
Sidaoui, J, M Ramos-Francia and M Cuadra (2010): The global financial crisis and
policy response in Mexico, BIS Papers, no 54, December.
Turner, P (2014): The global long-term interest rate, financial risks and policy
choices in EMEs, BIS Working Papers, no 441, February.
World Gold Council (2014): Chinas gold market: progress and prospects.

Cross-border bank lending during the taper
tantrum: the role of emerging market fundamentals
1

Cross-border bank lending to emerging markets slowed sharply during the taper tantrum. The
abruptness of this slowdown varied considerably across both lenders and borrowers. We use
newly available data to explain the drivers of this cross-sectional variation. Although the initial
tapering shock originated from advanced economies, EME-specific factors explain the bulk of
the variation in the slowdown across lender-borrower pairs.
JEL classification: F34, G15, G21.
Starting with the Federal Reserves May 2013 hint that it might begin reducing its
bond purchases sooner than previously expected, the taper tantrum triggered
sharp drops in the exchange rates of emerging market economies (EMEs) as well as
in their bond and equity prices. While the sharp reversal in portfolio flows to EMEs
has been widely documented, the behaviour of cross-border bank lending has
received less attention. This is surprising given the importance of cross-border bank
lending for most EMEs: the outstanding stock of cross-border bank claims on EMEs
stood at more than $3.6 trillion at the end of 2013 roughly as large as the stock of
all portfolio investment in EMEs. Thus, notwithstanding the growing importance of
securities market financing, a sudden stop in cross-border bank lending could still
destabilise economies and, in the worst case, lead to a balance of payments crisis.
During the taper tantrum, cross-border bank lending to EMEs slowed sharply.
Its growth rate dropped to 2.5% in the second and third quarters of 2013 from
around 10% over the previous two quarters. Interestingly, the intensity of this
deceleration varied considerably across lender and borrower. This raises important
questions for policymakers: what drove this variation? And were the factors that
drove it mainly on the borrowers or on the lenders side?
We attempt to answer these questions by exploring newly available bank
lending data in a regression framework. The new data contain three dimensions that
are crucial to this analysis the nationality of the lending bank, the location of the
borrower and the currency composition of the claims. These three dimensions allow

1
The authors thank Claudio Borio, Dietrich Domanski, Patrick McGuire, Hyun Song Shin, Christian
Upper and Philip Wooldridge for useful comments and discussions. Bilyana Bogdanova and Emese
Kuruc provided excellent research assistance. The views expressed are those of the authors and do
not necessarily reflect those of the BIS.
Stefan Avdjiev
[email protected]
Eld Takts
[email protected]

us to link the evolution of bilateral exchange rate-adjusted flows to drivers
connected with both the lender banking systems and the borrower EMEs.
We find that EME-specific factors explain most of the slowdown in cross-border
bank lending during the taper tantrum. While factors connected with both the
lender banking system and the borrower EMEs are statistically significant, the latter
group is responsible for roughly 70% of the explained variation. In particular, we
identify the current account balance and the share of cross-border bank lending
denominated in US dollars as significant drivers. One lender banking system
variable, the change in the average bank credit default swap (CDS) spread during
the taper tantrum, accounts for the remaining 30% of explained variation.
The rest of the feature is organised as follows. The first section introduces the
data used for our work, in particular the new cross-border bank lending data. The
second analyses these data in a regression framework, and the final section
concludes with a brief discussion of the main results.
Data
In this section, we first go over the three dimensions of the newly available cross-
border bank lending data needed for our empirical analysis. Second, we describe
how the growth rate of cross-border bank lending changed across the pairs of
lender banking systems and borrower EMEs. Finally, we discuss possible drivers of
the variation in cross-border bank lending flows.
The new cross-border bank lending data
In this study, we utilise the recently implemented Stage 1 Enhancements to the BIS
international banking statistics (IBS).
2
Given that this is the first time the new data
have been used in analytical work, we briefly describe the most relevant features
and make a comparison with the previously available BIS IBS data sets (see CGFS
(2012) for details).
Any attempt to address our question (ie linking fluctuations in cross-border
bank flows both to lender banks and to borrower economy factors) requires three
dimensions of cross-border bank lending data:
A. the nationality of the lending bank;
B. the residence of the borrower; and
C. the currency composition of cross-border claims.
We need the first two dimensions (A and B) to identify factors specific to the
lender banking system and borrower EMEs, respectively. The nationality of the bank
(A) identifies the lender banking system, ie the home of the highest-level banking
entity in the corporate chain which, in turn, is used as a proxy for the decision-
making unit for cross-border bank lending.
3
We need the third dimension (C) to

2
Access to these data is currently limited to central banks and other reporting authorities. Enhanced
data will be made available more widely when their availability and completeness improve.
3
Strictly speaking, the nationality of the lending bank identifies the country of ownership, and not
necessarily that of the decision-making unit. In general, the nationality of ownership and decision-

control for the impact of currency fluctuations on changes in the outstanding stocks
of cross-border bank claims. For instance, a move in the euro/dollar exchange rate
mechanically leads to changes in the US dollar value of euro-denominated claims.
Thus, the changes in BIS IBS claims, which are expressed in US dollars, also reflect
currency movements. The adjustment for currency movements is particularly
relevant, because contractions in cross-border lending tend to coincide with large
exchange rate movements.
4

The Stage 1 enhanced data are the first consistent data set to provide all three
dimensions at the same time (Table 1). Previously, the BIS IBS data provided
information on only two of the above three dimensions. The consolidated data set
identified the nationality of the lending banks (dimension A) and the residence of
the borrower (dimension B), but did not provide a currency breakdown
(dimension C). By contrast, the locational data by residence did reveal the currency
composition of banks cross-border claims (dimension C) as well as the residence of
the borrower (dimension B), but they did not identify the nationality of the lending
bank (dimension A). Finally, the locational data by nationality contained dimensions
A and C, but not dimension B.
The unavailability of all three data dimensions constrained previous studies on
the determinants of foreign bank lending to EMEs. For instance, McGuire and
Tarashev (2008) used the consolidated data to construct the dependent variable in
their model (ie the growth rate of the stock of international claims). As a
consequence, they were able to study how the health of individual national banking
systems affected foreign lending to EMEs, but only at the cost of working with data
that had not been adjusted for exchange rate fluctuations. By contrast,
Takts (2010) used the locational data set in order to construct his dependent

making are more likely to coincide in banking systems that are more centralised (eg in the case of
German and French banks) and, by the same token, they are less likely to do so in more
decentralised banking systems (eg with Spanish banks). However, in the case of cross-border bank
lending, the overlap between the countries of ownership and decision-making is likely to be
stronger than in the case of local lending which makes nationality a reasonable, though imperfect,
proxy for decision-making in this instance. See CGFS (2010) for further discussion of models for
international banking.
4
For instance, when we repeat the full analysis with currency-unadjusted consolidated data, the
results of our benchmark estimation change drastically. For details, see the section on sensitivity
analysis below.
Data availability in the BIS International Banking Statistics
By data dimension Table 1
Nationality of lending
bank (A)
Residence of
borrower (B)
Currency
composition
3
(C)
Consolidated data
1
Yes Yes No
Locational data
2

by residence No Yes Yes
by nationality Yes No Yes
Stage 1 data Yes Yes Yes
1
The BIS consolidated banking statistics group claims according to the nationality of banks (ie according to the location of banks
headquarters), netting out inter-office positions.
2
The BIS locational banking statistics define creditors and debtors according to their
residence, consistently with national accounts and balance of payments principles.
3
In addition to exchange rate fluctuations, the
quarterly flows in the locational datasets are corrected for breaks in the reporting population.

variable (ie the growth rate of the stock of currency-adjusted cross-border claims).
As a result, he was able to work with exchange rate-adjusted cross-border lending
flows, but information on the nationality of the lending banks was lacking, which
prevented a decomposition of the estimated global home country factor into
banking system-specific factors. Avdjiev et al (2012) came closest to using all three
required dimensions by combining the locational data by residence with the
consolidated data. However, this approximation could not have the detail and
precision available in the Stage 1 Enhancements.
Finally, although the new Stage 1 data are not yet complete, they are
nevertheless representative. On aggregate, information on the nationality of lending
banks is available for about two thirds of the total cross-border claims on EMEs. This
ratio varies considerably and tends to be higher for the larger EMEs.
Cross-border bank lending during the taper tantrum
In selecting the sample for our analysis, we aim to include all internationally
significant lender banking systems and borrower EMEs.
In particular, we include all 21 lender national banking systems which had more
than $10 billion of cross-border claims on EMEs in the new Stage 1 data at end-
2013.
5
We also include all 35 EME recipients of cross-border bank lending whose
cross-border borrowing exceeded $10 billion at end-2013: Angola, Argentina,
Bulgaria, Brazil, Chile, China, Chinese Taipei, Colombia, Croatia, the Czech Republic,
Egypt, Hungary, India, Indonesia, Israel, Korea, Kuwait, Lithuania, Malaysia, Mexico,
Morocco, Nigeria, Peru, the Philippines, Poland, Romania, Russia, Saudi Arabia,
South Africa, Thailand, Turkey, Ukraine, the United Arab Emirates, Venezuela and
Vietnam (Graph 1, rows).
6

Across our sample, the new Stage 1 data reveal a substantial degree of
variation during the episode. In particular, we examine the change in the growth
rate of cross-border bank lending to EMEs between the taper tantrum quarters (Q2
and Q3 2013) and the two quarters preceding this episode (Q4 2012 and Q1 2013).
Among the largest individual lender national banking systems, this change ranged
from 13% to +10%. Among the largest borrower EMEs, the change was between
12% and +7%. Smaller lenders and borrowers saw even bigger changes.
The new Stage 1 data also show that the variation is even greater at the level of
bilateral cross-border bank lending flows (ie flows from individual lender banking
systems to individual borrower EMEs). Graph 1 displays changes in the bilateral
growth rates during the taper tantrum between lender banking systems (columns)
and borrower EMEs (rows). For confidentiality reasons, the nationalities of the lender
banking systems are not revealed, and colour-coded interval categories are shown
instead of exact values.

5
In order to ensure that our sample is representative, we exclude lender national banking systems
whose home country had not started providing complete breakdowns for the Stage 1 data as of
Q3 2012 and whose cross-border claims on EMEs are primarily booked in the home country. For
confidentiality reasons, we do not list the lender national banking systems included in our sample.
6
The sample selection for EMEs is based on the locational data by residence. These values may
exceed those obtained from the new Stage 1 data because some BIS reporting countries do not yet
report Stage 1 data. We exclude offshore financial centres from the sample.

Finally, we also undertake a straightforward ANOVA decomposition on the
change in the growth rate of cross-border bank lending. This reveals that the
variation between lender national banking systems (the columns in Graph 1) is of
roughly the same size as the variation between borrower EMEs (the rows in
Changes in growth rate of cross-border bank lending during the taper tantrum
1
By nationality of lending bank (columns)
2
and residence of borrower (rows)
3
Graph 1
AE = United Arab Emirates; AO = Angola; AR = Argentina; BG = Bulgaria; BR = Brazil; CL = Chile; CN = China; CO = Colombia; CZ = Czech
Republic; EG = Egypt; HR = Croatia; HU = Hungary; ID = Indonesia; IL = Israel; IN = India; KR = Korea; KW = Kuwait; LT = Lithuania;
MA = Morocco; MX = Mexico; MY = Malaysia; NG = Nigeria; PE = Peru; PH = Philippines; PL = Poland; RO = Romania; RU = Russia;
SA = Saudi Arabia; TH = Thailand; TR = Turkey; TW = Chinese Taipei; UA = Ukraine; VE = Venezuela; VN = Vietnam; ZA = South Africa.
1
Change in the average growth rate of cross-border bank lending to EMEs between the taper tantrum episode (Q2 and Q3 2013) and the
two quarters immediately preceding it (Q4 2012 and Q1 2013).
2
Lender banking systems, coded in letters, are ordered by the change in
the average bank CDS spread between Q2Q3 2013 and Q4 2012Q1 2013 (from the largest increase to the largest decline).
3
Borrower
EMEs are ordered by the current account balance (from most positive to most negative) in 2012 (expressed as a percentage of GDP).
Sources: IMF, World Economic Outlook; Markit; BIS locational banking statistics by nationality; BIS calculations.
A B C D E F G H I J K L M N O P Q R S T U
KW
SA
AE
TW
AO
NG
MY
VN
KR
RU
VE
PH
CN
HU
IL
HR
AR
LT
TH
BG
MX
CZ
BR
ID
CO
CL
PE
PL
EG
RO
IN
ZA
TR
UA
MA
Less than
50% Between 0% and 10%

Between 50% and 10% Between 10% and 50%
Between 10% and 0% Greater than 50%

Graph 1). The decomposition suggests that we should examine factors relating to
both lender banking systems and borrower EMEs when analysing what drove the
change in cross-border bank lending growth during the taper tantrum.
Possible drivers of cross-border bank lending
Economic theory and past studies of cross-border bank lending suggest a number
of economic variables as possible drivers of cross-border bank lending. Given that
this is the first study to use the new Stage 1 data, we examine a range of these
potential explanatory variables. Furthermore, based on our ANOVA decomposition,
we give equal consideration, as potential drivers, to lender banking system- and
borrower EME-related factors.
We explore five lender banking system variables: the change in the average
bank CDS spread and equity price during the taper tantrum, past real credit and real
deposit growth in the home market of banks, and the share of cross-border claims
denominated in US dollars. A rise in the CDS spread or a decline in the equity price
of the lending bank during the taper tantrum could potentially signal increased
levels of bank stress, which would in turn reduce banks ability to lend. From a
different perspective, rapid credit or weak deposit growth in the home market could
be a precursor of subsequent funding strains. Finally, a high share of cross-border
claims denominated in US dollars could be an indicator of greater reliance on US
dollar funding and, consequently, of greater funding strains experienced during the
taper tantrum.
We also examine five borrowing EME variables: current account balance,
external debt, government budget balance, past real credit growth to the private
sector, and the share of cross-border bank lending denominated in US dollars.
Higher current account and budget deficits and higher levels of external debt all
make borrowers more vulnerable, which typically would reduce banks willingness to
lend. Meanwhile, rapid real credit growth stretches the balance sheets of local
borrowers and increases their susceptibility to external shocks (BIS (2014b)). This
also tends to reduce banks willingness to supply cross-border credit to these
overstretched borrowers. Finally, a large share of cross-border bank lending
denominated in US dollars could indicate, as in the case of lender banking system
factors, higher sensitivity to the evolution of US monetary policy.
7

A preliminary examination of the data suggests that these variables, particularly
the change in banks CDS spreads and the current account balance, could have had
an impact on cross-border bank lending. In Graph 1, lender banking systems are
ranked according to the change in the average CDS spread during the taper
tantrum, while borrower EMEs are ranked according to their current account
balance. Indeed, moving from left to right, ie from rising to declining bank CDS
spreads, one can see fewer instances of a slowdown in cross-border bank lending.
Similarly, moving from top to bottom, ie from current account surplus to deficit, one
can see more instances of deceleration. Of course, these broad trends are
punctuated with many exceptions and should be seen only as an inducement to
undertake a more rigorous regression analysis.

7
The exact definitions of all potential explanatory variables are available upon request.

Analysis
We undertake the analysis in three steps. First, we present our benchmark
regression, ie the regression specification that best fits the cross-border bank
lending data. Second, we discuss the economic implications of the model by
decomposing the slowdown in cross-border bank lending into the contributions of
its main drivers. Finally, we examine the robustness of the benchmark specification.
Benchmark regression
The data structure and the timing of the taper tantrum guide our basic regression
setup. As the BIS bank lending data are reported at a quarterly frequency and the
taper tantrum lasted from May to September 2013, we compare the growth rates in
cross-border bank lending in Q2 and Q3 2013 with their counterparts in the
preceding two quarters (Q4 2012 and Q1 2013).
Furthermore, we weigh each observation by the size of the respective bilateral
stock of outstanding cross-border claims at the end of Q3 2012. More specifically,
the weight that we assign to each observation is equal to the ratio of the respective
bilateral stock to the sum of all bilateral stocks in our sample. Our examination of
the data suggests that smaller volumes tend to be highly volatile and probably
reflect more bank-specific, or even project-specific, factors. As a consequence, the
evolution of larger bilateral cross-border bank lending claims is likely to reflect
changes in the economic environment more accurately. In order to control for
extreme outliers, we also exclude observations for which the value of the dependent
variables is more than 10 standard deviations away from the sample mean.
We select our benchmark explanatory variables through an elimination process.
We start by running a panel regression that includes all 10 candidate explanatory
variables discussed in the previous section. Then we exclude the variable with the
lowest t-statistic. Next, we rerun the regression with the remaining variables. We
continue this iteration until all remaining explanatory variables are statistically
significant at the 5% level.
Our benchmark regression, obtained through the above elimination process,
explains the variation of bilateral cross-border bank lending flows with the degree
of stress experienced by the lender banking system (as proxied by the change in the
average CDS spread) and with the characteristics of the borrower EME (as proxied
by the current account balance and the share of cross-border bank lending
denominated in US dollars). Formally, we estimate the following equation:
o | c A = + A + + +
, , b l l b b b l
XBC c CDS CAB USDS
Our dependent variable AXBC
b,l
represents the change in the average growth
rate of the lender banking system ls cross-border claims on borrower EME b
between the taper tantrum (Q2 and Q3 2013) and the two quarters preceding it
(Q4 2012 and Q1 2013). Formally:
, , , ,
,
, , , ,
2 13 3 13 4 12 1 13
1 1
2 1 13 2 13 2 3 12 4 12
b l b l b l b l
b l
b l b l b l b l
flow Q flow Q flow Q flow Q
XBC
stock Q stock Q stock Q stock Q
A = + +
| | | |
| |
| |
\ . \ .

Our independent variables are defined as follows: c is a constant; ACDS
l
is the
change in the average CDS spread of the lender banking system l between the taper

tantrum (Q2 and Q3 2013) and the two quarters preceding it (Q4 2012 and
Q1 2013); CAB
b
is the current account balance (as a percentage of GDP) in borrower
EME b during 2012; USDS
b
is the share of cross-border bank lending to borrower
EME b denominated in US dollars (as of end-Q3 2012); and c
b,l
is the error term. We
weigh each observation b,l by the share of cross-border claims that the lender
banking system l had on borrower EME b in total cross-border bank lending (across
all borrower-lender pairs) in our sample as of end-September 2012.
The benchmark results show that both lender banking system and borrower
EME factors were statistically and economically significant drivers (Table 2). The
regression fits the data well: it explains close to one eighth of the total variation of a
large and heterogeneous sample. All the coefficient estimates are statistically
significant at the 1% level, well above the 5% significance threshold used to narrow
down the list of variables.
The sign of the estimated coefficient for the lender banking system variable
suggests a straightforward supply effect. The negative sign for the change in the
CDS spread (o) implies that the greater the increase in the stress experienced by a
national banking system during the taper tantrum, the more that banking system
reduced its cross-border bank lending.
The coefficient on the EME current account balance (|) is positive, implying that
a stronger current account position (ie a larger surplus or a smaller deficit) was
associated with more resilient cross-border bank lending growth. Intuitively, EMEs
with larger external financing needs on the eve of the taper tantrum were more
vulnerable to a sudden change in financing conditions. This is consistent with public
discussions during the taper tantrum, which often focused on current account
balances as an important indicator of vulnerability for EMEs.
Finally, a higher share of US dollar-denominated cross-border claims on a given
EME was associated with a sharper slowdown in lending to its residents. Thus, it
appears that the taper tantrum had a larger impact on economies with greater
reliance on cross-border US dollar funding.
Economic significance
We next show that the benchmark regression captures an economically significant
part of the slowdown in cross-border bank lending by plotting predicted and actual
lending slowdowns by lender banking system and by borrower EME.
Regression results Table 2
Variables Coefficient
1
t-statistic Probability
Lending bank CDS spread change (oACDS) 1.36 4.19 0.0000
EME current account balance (|CAB) 2.73 2.83 0.0048
EME USD share of lending (USDS) 134.44 4.20 0.0000
R-squared (in per cent) 11.72
Number of observations 658
1
All coefficient estimates are multiplied by 1,000 to ease representation.
Source: Authors calculations.

In our decomposition, we focus on deviations from means, as in Avdjiev et
al (2012). More specifically, we first create demeaned variables by taking the
difference between the regression variables in our benchmark equation and their
respective means. We then calculate the contributions by multiplying the demeaned
variables by the respective estimated coefficients.
Overall, our benchmark specification explains a large share of the variation in
the data. With respect to lender banking systems (Graph 2, left-hand panel), the
econometric model accurately captures the slowdowns in cross-border lending to
EMEs reported by both advanced economy banks (eg German, Swiss and US banks)
and EME banks (eg Brazilian banks): the distance between actual lending (black
dots) and estimated lending (red dots) is generally small. However, the fit is not
uniformly precise for all banking systems. For example, the model captures neither
the surprising acceleration in French banks lending growth nor the full scale of the
slowdowns reported by Japanese and Dutch banks.
In general, the estimates fit better for larger lender banking systems than for
smaller ones. Technically, this is due to the volume weighting in the benchmark
regression, which puts emphasis on fitting larger bilateral observations.
Economically, this also implies that our benchmark regression more closely follows
the total volumes of cross-border bank lending but that the price paid for doing so
is that it less accurately fits some smaller country-specific growth rates.
The estimates fit even better for borrower EMEs (Graph 2, centre panel). The
benchmark specification performs well not only for the largest borrower EMEs
(eg China, India, Brazil and Turkey) but also for smaller EMEs (eg the Czech Republic
and Peru). This good fit for smaller economies is not necessarily expected, given the
Decomposition of the change in growth rate of cross-border bank lending to EMEs
1
Graph 2
Selected lender banking systems
2
Selected borrower EMEs
2
Share of economic factors
Percentage points

Percentage points

Per cent

BR = Brazil; CA = Canada; CH = Switzerland; CN = China; CZ = Czech Republic; DE = Germany; ES = Spain; FR = France; GB = United
Kingdom; IN = India; JP = Japan; KR = Korea; MX = Mexico; NL = Netherlands; PE = Peru; PL = Poland; RU = Russia; TR = Turkey;
US = United States.
1
Change in the average growth rate of cross-border bank lending to EMEs between Q2Q3 2013 and Q4 2012Q1 2013.
2
The reported
actual and estimated changes in the growth rates for individual banking systems and EMEs represent weighted averages of the respective
bilateral changes, weighted as in the benchmark regression equation (ie by the size of the respective bilateral stock of outstanding cross-
border claims at the end of Q3 2012). The individual changes in the growth rates reported in the graph may differ from the respective
changes obtained from alternative data sources due to the fact that the new Stage I data set is not yet fully complete (see main text for
further details).
3
As defined in the benchmark regression.
Sources: IMF, World Economic Outlook; Markit; BIS locational banking statistics by nationality; BIS calculations.
BR CA CH DE ES FR GB JP NL US
0
5
10
15
5
Actual change in the growth rate
Estimated change in the growth rate
CN BR IN TR KR RU MX PL CZ PE
0
5
10
15
5
Banks CDS spread change
3
0
10
20
30
40
50
Current account balance
3
USD share of lending
3

weighting we used in the regression analysis. The most notable exception is Russia,
where our model does not predict the full extent of the slowdown.
Our estimates imply that factors associated with borrower EMEs accounted for
the bulk of the explained variation in cross-border lending (Graph 2, right-hand
panel). All in all, EME factors account for around 70% of the explained variation.
Among them, the US dollar share of cross-border lending accounted for around
45% and the current account balance for around 25%. Meanwhile, the factor related
to lender banking systems, the change in banks CDS spreads, accounted for around
30% of the explained variation.
The strong explanatory power of EME-related variables in the benchmark
specification potentially explains why our estimates tend to fit borrower EMEs even
better than they do the lender banking systems. Although variation across lender
banking systems and borrower EMEs is roughly similar, the EME-related variables
seem to be able to better capture the EME-specific variation.
In sum, our analysis shows that, even though the initial taper tantrum shock
originated from the advanced economies, it was mostly EME-specific vulnerabilities
that determined how the slowdown in cross-border bank lending would be
distributed across EMEs.
Sensitivity analysis
We examine the robustness of our benchmark results to alternative specifications by
conducting a sensitivity analysis. While for the sake of brevity we do not list the
detailed regression results, they are available on request.
Our benchmark specification is robust to dropping one variable at a time. In
each of the resulting specifications, all estimated coefficients retain their statistical
significance, their respective signs and even their magnitudes.
The benchmark results are also robust to the inclusion of additional explanatory
variables. In particular, we include three additional borrowing EME-linked variables
that might signal vulnerabilities as discussed in BIS (2014a): credit-to-GDP gaps,
inflation and government debt levels, but none of these turns out to be significant
at the 5% level. Similarly, the other potential explanatory variables that were
excluded in the elimination process also remain insignificant when added to the
benchmark regression.
Furthermore, the benchmark results remain robust to the exclusion of
individual borrower EMEs from the sample. The signs of the coefficients remain
unchanged in all cases. In addition, the statistical significance also remains robust in
almost all cases: the three explanatory variables from the benchmark regression
remain significant at the 1% level in 101 out of 105 possible cases. The robustness
of the results is particularly reassuring in the case of China, because it reveals that
the benchmark results are not dominated through the regression weighting by the
large and relatively stable cross-border bank lending flows to China.
The benchmark results also remain robust to the exclusion of individual lender
banking systems from the sample. The sign, size and statistical significance of the
coefficient estimates remain robust in almost all cases. The lender bank CDS spread,
the current account balance and the US dollar share of cross-border bank lending
remain significant at the 5% level in all but one of 63 possible cases.

Finally, we demonstrate that using the new Stage 1 data is critical for our
analysis. In particular, when we run the benchmark regression with the consolidated
data (Table 1, top row) as the dependent variable instead of the new Stage 1 data,
the results become much weaker. The regression loses around half of its
explanatory power. Furthermore, the change in the bank CDS spread loses its
statistical significance and even reverses its sign.
In sum, the sensitivity analysis delivers two key takeaways. First, the benchmark
regression and its coefficient estimates are very robust. Second, it is essential to use
the exchange rate-adjusted bilateral flows from the new Stage 1 data.
Conclusion
Our analysis shows that both lender banking system and borrower EME factors
explain statistically and economically significantly the cross-sectional variation in
cross-border bank lending during the taper tantrum. However, it is the EME-specific
factors that are more relevant from an economic perspective: the US dollar share of
cross-border bank lending and the current account deficit are jointly responsible for
around 70% of the explained cross-sectional variation. The rest is accounted for by
the change in the average bank CDS spread during the taper tantrum.
Given that this is the first analysis of the new Stage 1 data, its findings should
be interpreted with caution. For instance, the result that borrower EME factors
explain more of the cross-sectional variation might partly reflect better data quality
along that dimension. Borrower EMEs are natural objects for national statistical data
collection efforts, whereas data on lender banking systems come from several
sources that may not necessarily be fully harmonised. Future research might shed
further light on such issues.
Finally, any interpretation of these results should take into account that cross-
border bank flows represent only a part, albeit a considerable one, of international
financial flows. Total cross-border bank claims on EMEs are of roughly the same
magnitude as portfolio investments in EMEs. While cross-border bank lending is
important, non-bank cross-border financing has increased even more rapidly over
the past few years, as documented in BIS (2014b) and Chui at al (2014). That said,
cross-border bank lending remains a major source of foreign financing in most
EMEs. As a result, a better understanding of its drivers would help policymakers.

References
Avdjiev, S, Z Kuti and E Takts (2012): The euro area crisis and cross-border bank
lending to emerging markets, BIS Quarterly Review, December, pp 3747.
Bank for International Settlements (2014a): 84th Annual Report, Global financial
markets under the spell of monetary policy, Chapter II, June.
(2014b): 84th Annual Report, Debt and the financial cycle: domestic and
global, Chapter IV, June.
Committee on the Global Financial System (2010): Funding patterns and liquidity
management of internationally active banks, CGFS Papers, no 39, May.
(2012): Improving the BIS international banking statistics, CGFS Papers,
no 47, November.
Chui, M, I Fender and V Sushko (2014): Risks related to EME corporate balance
sheets: the role of leverage and currency mismatch, BIS Quarterly Review,
September, pp 3547.
McGuire, P and N Tarashev (2008): Bank health and lending to emerging markets,
BIS Quarterly Review, December, pp 6780.
Takts, E (2010): Was it credit supply? Cross-border bank lending to emerging
market economies during the financial crisis, BIS Quarterly Review, June, pp 4956.

Residential property price statistics across the
globe
1

Despite their importance in macroeconomic and financial stability analysis, residential property
data are not easily available on a comparable basis. The BIS currently publishes more than
300 price series for 55 countries, among which it has selected one representative series for each
country. For 18 countries, it also publishes series that span the period back to the early 1970s.
House prices can serve as key indicators of financial stability risks, as property booms are often
the source of vulnerabilities that lead to systemic crises.
JEL classification: R30, R31.
Since June 2014, the BIS has published three data sets of residential property prices
that currently cover 55 countries. The unique feature of the first data set, besides its
wide coverage, is that it contains several series per country, focusing on different
segments of the national market. The other two data sets contain one
representative series per country, selected by BIS statisticians to enhance cross-
country comparability. The second set focuses on broad coverage of the selected
series, while the unique feature of the third data set is that it creates long time
series for a subset of countries. In this article, we present these data and highlight
their usefulness.
Households, analysts and policymakers share a keen interest in property price
developments. Buying a house is often the largest single transaction for a
household, and the property its largest asset. It is no surprise that house prices are a
frequent topic in dinner party conversations. Fluctuations in house prices have a
large impact on households net wealth, and their propensity to spend. In addition,
residential property values underpin much of the USD 24 trillion mortgage market
in advanced economies.
2
This makes house markets a central element in the
analysis of trends in aggregate expenditure, the strength of bank balance sheets
and the interactions of macroeconomic and financial stability. Moreover, house
prices are relevant to statisticians who compile macroeconomic series on
households wealth, or consumer price indices.

1
the BIS. We wish to thank Claudio Borio, Dietrich Domanski, Hyun Song Shin and Christian Upper
for constructive comments on earlier drafts.
2
Estimated as the sum of outstanding mortgages in Australia, Canada, the euro area, Japan, the
United Kingdom and the United States in the first quarter of 2014.
Michela Scatigna
[email protected]
Robert Szemere
[email protected]
Kostas Tsatsaronis
[email protected]

Despite their importance, comparable cross-country data on residential
property prices are hard to obtain. The combination of the disaggregated nature of
property transactions and the pronounced diversity of the properties themselves
complicates the compilation of price indices. Moreover, comparability suffers
because of a lack of standardisation and the short time span of many series, a
significant shortcoming in view of the length of property price cycles. The data sets
published by the BIS are aimed at narrowing these gaps. Their unique features in
terms of coverage and comparability should be of use to economists and
policymakers alike.
The article is organised in three sections. The first section discusses various
statistical aspects related to the construction of price indices. The second section
provides an overview of the data published by the BIS. The final section uses the
three data sets to illustrate the use of their unique features. It presents a quick
overview of international market developments, provides cross-country
comparisons of current valuations and highlights the international co-movement of
prices. Finally, it also documents the early warning properties of house price growth
for episodes of systemic banking stress.
Diversity in the characteristics of house price statistics
One of the main objectives of the residential price series published by the BIS is to
provide analysts and researchers with information about house price growth in
various countries. Cross-country analysis places a premium on data comparability,
and from that perspective property price indices present a number of statistical
challenges. In this section, we outline the main factors that can differentiate one
house price index from the other.
Property price indices aim to capture the trend in the average price of dwellings in a
given area. However, the nature of the property market complicates standardisation
in index construction. For one, housing is not a homogeneous good. Properties vary
immensely in several dimensions, some of which are not easy to quantify. While
segmentation of the market in terms of location, size or age of the property may be
straightforward, controlling for quality of construction, customer appeal or level of
comfort is less obvious. In addition, property changes hands through bilateral
transactions and not in centralised markets.
3
This requires mechanisms to collect
dispersed price information in order to create a representative index, as indices with
different characteristics provide different messages even for the same country (see
box). These practical difficulties also explain why in many cases statistics have been
gathered (at least initially) by private sector firms and associations with specific
commercial interest in the housing market while public sector efforts have lagged
behind. This pattern is even more pronounced in the case of commercial property,
where heterogeneity and dispersion of transactions are even greater and where
hardly any official statistics exist.
4

3
Central property registries can function as surrogate centralised marketplaces for the purpose of
compiling price indices. Unfortunately, these registries are not always organised in a way that
makes it easy to extract useful information on price and property characteristics.
4
BIS researchers have used commercial property price indices, mostly sourced from proprietary
commercial compilers, for research and analysis, including for various editions of the BIS Annual
Report.

Property price indicators differ in three principal dimensions: their geographical
coverage, the source of information on property values, and the approach to
controlling for differences in property characteristics.
The old adage Location, location, location! points to the key driver of real
estate value. Prices of properties in different locations can diverge persistently even
within the same region. Demographics, tastes and demand from foreign buyers may
be pushing up the price of houses in highly desirable spots at the same time as
properties in other areas are languishing due to lack of demand or a decline in local
economic activity. During the 200709 crisis, properties in Manhattan and central
London maintained their value while in the rest of the United States and the United
Kingdom average home values were plummeting. Aggregate price indices that are
skewed towards one or the other segment may convey a distorted picture of
average developments. That said, for some purposes it is useful to focus on market
segments. For instance, given the wider economic disruptions created when a large
stock of outstanding debt defaults, the assessment of the macro-financial risks of
rapid growth in mortgage credit would naturally focus on the booming segment of
the housing market. This is particularly true for very diverse countries such as the
United States, and often coincides with properties in major cities. In practice, the
majority of the published series refer to country-wide averages, but for about
13 countries the data cover only prices in major cities or the capital alone (Table 1,
left-hand set of columns).
Indices differ also in terms of the primary source of price information (Table 1,
middle set of columns). The ideal source would be actual transaction prices for a
representative set of properties. Comprehensive sets of transaction prices can be
collected from land registries, where these exist. However, widespread tax
avoidance in some countries can introduce bias in recorded transaction prices
compared with real market property values. Moreover, depending on the available
resources, registries may keep only incomplete records of the characteristics of
dwellings, hence complicating the quality adjustment of price indicators. Similar
factors explain the time lag between the publication of official indicators and less
comprehensive indicators compiled from private sources.
5

Tools that facilitate the workings of the real estate and mortgage markets may
also be a source of house price information, as are advertised prices, collected from
the internet, newspapers or the databases of real estate agents. Advertised prices
may capture trend shifts even before transactions take place. However, their
usability is limited by systematic differences vis--vis actual transaction prices (for
example, owners listing a high asking price or failing to sell the property), which
may also be time-varying (for instance, close to turning points in demand or supply
cycles). Alternatively, in countries with developed mortgage markets, yet another
source of information is appraisals made as part of the loan approval process.
Experts valuations are likely to be closer to final transaction prices than advertised
prices but may also be subject to biases if they are primarily driven by lenders
desire to ensure adequate cover for their exposure.
The data source or the purpose of the indicator also influences its coverage. For
example, appraisal-based indicators usually do not incorporate price data from cash
purchases; and owner-occupied property price indices measure only developments
in the prices of dwellings owned by the households that live in them. They exclude

5
For more details, see Eurostat (2013), Chapter 9.

Characteristics of series included in the three data sets
Summary of attributes for all series available for each country Table 1
Country
Number
of series
Geographical
coverage
Source of price
information
Quality
adjustment
Series starting
year
All Cities Other
Trans-
actions
Appraisal
Adver-
tised
None
Size
only
Sophis-
ticated
1

Selected Long
Australia 6

2003 1970
Austria 16

2000
Belgium 9

2005 1970
Brazil 1

2001
Bulgaria 1

1993
Canada 3

1970 1970
China 3

2011
Croatia 3

1997
Cyprus 3

2002
Czech Republic 7

2008
Denmark 7

1971 1970
Estonia 7

2005
Euro area 1

1980
Finland 17

2005 1970
France 10

1996 1970
Germany 48

2003 1970
Greece 18

2006
Hong Kong SAR 3

1993
Hungary 5

2007
Iceland 1

2000
India 1

2009
Indonesia 5

2002
Ireland 8

2005 1970
Israel 1

2001
Italy 1

1990 1971
Japan 12

1955 1970
Korea 1

1986
Latvia 3

2006
Lithuania 1

1998
Luxembourg 5

2007
Macedonia 1

2005
Malaysia 4

1999
Malta 1

2000
Mexico 1

2005
1
Includes all other types of quality adjustment methods (hedonic regression, repeat sales, stratification, etc).
Source: BIS.

- Continued - Table 1
Country
Number
of series
Geographical
coverage
Source of price
information
Quality
adjustment
Series starting
year
All Cities Other
Trans-
actions
Appraisal
Adver-
tised
None
Size
only
Sophis-
ticated
1

Selected Long
Morocco 5

2006
Netherlands 6

1995 1970
New Zealand 3

1979 1970
Norway 9

1992 1970
Peru 1

1998
Philippines 2

2008
Poland 14

2006
Portugal 2

1988
Romania 6

2009
Russia 2

2001
Singapore 4

1998
Slovakia 1

2005
Slovenia 9

2007
South Africa 4

1966
Spain 3

1995 1971
Sweden 2

1986 1970
Switzerland 5

1970 1970
Thailand 4

2008
Turkey 2

2010
United Kingdom 5

1968 1970
United States 3

1975 1970
1
Includes all other types of quality adjustment methods (hedonic regression, repeat sales, stratification, etc).
Source: BIS.

sales of properties purchased with the intention to let or resell, a potentially
important share of the market in countries with low home ownership rates
(eg Germany) or in periods of intense speculation in residential property.
Finally, a price index must separate price movements from changes in the
characteristics of the underlying properties. Index compilers have adopted methods
of varying sophistication to perform this quality adjustment (Table 1, right-hand set
of columns). The quality adjustment process is more challenging in the case of
property price indices than for consumer or producer price indices, and the choice is
partly driven by data availability. The most basic approach uses the average house
price in a given area. This is the only possibility when the main characteristics of the
dwellings are not collected, or cannot be processed by the compiler. It can produce
misleading messages if the distribution of characteristics in the cross section of
properties surveyed varies over time. Indicators based on price per square metre are
calculated when only the size is available. More elaborate quality adjustments take
into account other characteristics and follow methods such as hedonic regression,

stratification, and repeated sales or appraisals.
6
Some of these methodologies are
better at adjusting for general trends in quality over time (eg hedonic regressions or
indices based on repeat sales), while others are more geared towards controlling for
cross-sectional differences in quality at a given point in time (eg stratification or the
ratio of sales price to appraisal value).
The demand for comparable price statistics led to Eurostats publication of the
Handbook on Residential Property Prices Indices (HRPP) last year. The HRPP gathers
recommendations on best practices for the compilation of property price indices in
the context of different user needs. The HRPP built on work undertaken by a
number of international organisations (including the BIS) to identify the user
requirements from the point of view of economic and policy analysis.
The implementation of HRPP recommendations runs into constraints related to
the availability of data and resources. Generally, public sector compilers are more
likely to follow its recommendations than their private sector counterparts.
Residential property price indices based on the recommendations of the handbook
have been constructed in many European countries, but progress is slower
elsewhere.
The BIS property price statistics
In 2009, the G20 asked the BIS and its member central banks to collect and publish
residential property prices. The request was made in the context of the Data Gaps
Initiative, which aims to improve the availability and comparability of economic and
financial data across countries, and follows long-standing efforts by the BIS to
collect and use cross-country data on property prices.
The BIS work on cross-country house price data dates back to the late 1980s
(BIS (1989)). An expanding set of price series have been collected from various
national sources (both public and private) and cover both residential and
commercial properties. The information has been extensively used in analysis and
reported in publications, such as several editions of the BIS Annual Report. BIS
analysis has focused on the determinants of property prices (Tsatsaronis and
Zhu (2004), gert and Mihaljek (2007), Glindro et al (2010), Takts (2012)) and the
link between property prices and credit (Davis and Zhu (2005)) as well as the
information content of real estate prices for financial stability (Borio and
Drehmann (2009)). Borio et al (1994) constructed an indicator of aggregate asset
prices combining equity and property prices, and explored its link with money,
credit, interest rates, output and inflation.
The residential property price series that are published by the BIS are organised
in three data sets: the Detailed data set that includes all the original series collected
for each country, the Selected series data set that includes one representative price
series per country, and the Long series data set that presents a price series with a
long time span which also satisfies some minimum comparability criteria across
countries. These three data sets are discussed in more detail below. Table 1
provides a succinct overview of the range of series in the three data sets for each
country.

6
See Eurostat (2013), Chapters 47.

The Detailed data set: several series per country
The BIS started to publish monthly residential property prices in July 2010. The
coverage of these statistics has expanded from an initial set of 37 countries to
55 today, including 18 G20 countries and all 28 EU members. The total number of
series published currently exceeds 300, as, for several countries, more than one
series is available. Irrespective of the original compiler, all residential property price
series are reported to the BIS by central banks, which also monitor their quality.
The series published on the BIS website differ from country to country in terms
of type of property, area covered, property age, priced unit, compilation method
and seasonal adjustment. For many countries, residential property price indices
cover the entire market, encompassing all types of dwelling throughout the country
concerned in both new and existing dwelling markets. For other countries, the
coverage of the compiled series is more limited, either in terms of types of dwelling,
from a geographical perspective, or regarding the market. In general, the BIS has
made an effort to collect price indices that have been compiled following the HRPP
recommendations, and publishes extensive descriptive information regarding
source, coverage and compilation method alongside each series.
The Selected series data set: one series per country
Faced with multiple series for a country, an analyst must often confront the issue of
representativeness. The Selected Representative series data set, published by the
BIS, provides a possible answer by including only one house price indicator per
country.
As far as possible, the selected series cover all types of dwelling (flats and
houses, new and existing) throughout the country. Indicators with such a broad
coverage help assess aggregate household wealth and are thus useful for the
analysis of financial vulnerabilities linked to household indebtedness.
7
In addition,
such a broad indicator can be the best to use for international comparisons, and for
this reason the selection also took into account the HRPP criteria.
8
The resulting
Selected series data set is as homogeneous as feasible without, however, fully
eliminating all discrepancies.
To assist analysts who are interested in ready-to-use comparable price series,
these selected series are of a quarterly frequency and are updated once in each
quarter.
9
They are rebased, deflated by the CPI, and presented as both levels and
year-on-year growth rates.
10

7
This is, of course, subject to the caveat raised earlier about diverging trends in major segments of
the property market.
8
International comparisons of less aggregated data (for example, based on cities or specific types of
dwelling) are complicated because of the diversity in the definition of these segments as well as the
unavailability of the relevant series in many countries. Moreover, the importance of each segment
varies considerably across countries.
9
The series in the Detailed data set are presented at the same frequency as originally published by
the compiler.
10
A note on residential property price developments accompanies the quarterly publication,
analysing the recent evolutions of these indicators.

Diversity of residential property price statistics: the German case
The Detailed data set (see main text) includes 48 indicators for Germany covering different segments of the market
and following different methodologies. The four most representative series are all reported at least quarterly and are
quality-adjusted, but differ in many other aspects (Table A). Two indices are compiled by the private sector and the
other two by public sector agencies. More importantly, the indicators vary in their regional coverage, refer to
different types of property, use different approaches for the quality adjustment and have different starting dates as
well as reporting lags.
Characteristics of four residential property price indicators Table A
Compiling institution
Sector of
compiler
Data coverage
Quality
adjustment
Starting
date
Reporting lag
Statistisches Bundesamt
(Federal Statistical Office)
Public All dwellings (transactions) Hedonic
regression
Q1 2000 9 months
Verband deutscher
Pfandbriefbanken (vdp)
Private Mortgage-financed owner-occupied
dwellings (appraisals)
Hedonic
regression
Q1 2003 1 months
Bundesbank (based on
bulwiengesa AG)
Private Flats in the 7 largest cities (transactions
and appraisals)
Method of
typical cases
Q1 2008 3 weeks
Hypoport (EUROPACE) Private Dwellings (based on mortgages on
EUROPACE platform for brokerage)
Hedonic
regression
Q1 2005 18 days
The official indicator compiled by the Federal Statistical Office covers most of the housing market, but it is only
available with a long reporting delay. Analysts interested in the most recent developments might prefer the Verband
deutscher Pfandbriefbanken (vdp) indicator. The Hypoport (EUROPACE) indicator is even more timely but covers just
15% of overall transactions, and even though it has a monthly frequency, the data represent moving quarterly
averages. The Bundesbank indicator, based on data from bulwiengesa AG, is the only one focusing solely on urban
properties. The Bundesbank compiles an indicator that covers the entire country (not listed in the table) that is
available only at an annual frequency.
The four price indicators convey somewhat different messages both about longer-term trends and about
shorter-term developments. Graph A plots the recent history of the four indicators. Looking at longer-term trends
(Graph A, left-hand panel), the three country-wide indicators convey a similar picture, placing the cumulative growth
rate since the beginning of 2008 in the range of 11% (vdp) to 16% (Hypoport). However, only the Bundesbank
indicator shows the pronounced price increase registered for flats in major cities, where prices have climbed by a
cumulative 45% since 2008. The picture is even more mixed in quarter-to-quarter movements (Graph A, right-hand
panel). House price growth estimates calculated on the basis of the Bundesbank and vdp indices differ substantially,
but they are both much less volatile than that based on the Hypoport series, arguably reflecting the shifting sample
of the properties listed on the online brokerage that supplies the price information. The graph also illustrates the
reporting lag of the Federal Statistical Office indicator.
The BIS has chosen the vdp index for inclusion in the Selected data set (see main text) because of its broad
coverage of properties and its short reporting lag.
The Long data set series (see main text) for Germany also uses the vdp index for the period starting in 2003.
Prior to that date, the long series is constructed on the basis of price indices referring to properties in western
Germany. Going backwards in time, the long series uses the historical time series of the Bundesbank on western
German dwelling prices for the period 19752002 and sales price data in four cities from a private sector source
(Ring Deutscher Makler) for the period 197274. For the first two years (197071) and in the absence of house price
data, the series uses construction cost data as a proxy indicator. For the years where only annual data were available
(ie prior to 2003), the quarterly series was constructed using the Chow-Lin (Chow and Lin (1971)) interpolation
procedure based on the quarterly pattern of the construction cost index.

Long series data set: stretching the time dimension
Many questions can only be answered by analysing long time series of property
prices.
11
The BIS also publishes a Long series data set on residential property prices,
which extends the Selected series for 18 advanced economies with historical data
that go as far back as 1970 or 1971 on a quarterly basis.
This work has been driven by the BIS in close coordination with national
authorities, based on existing data. These data originate from various sources such
as central banks, national statistics offices, research institutes, private companies
and academic studies. They rely on different methodologies and can cover
heterogeneous types of geographical area and types of dwelling. The resulting long
series for each country start from the series in the selected data set, extended
backwards with information from alternative sources, using standard statistical
techniques to fill any gaps.
12

Ongoing improvements
The three data sets continue to expand in three ways. The first goal is to keep
increasing the number of countries included in the data sets, for instance by adding
price series for the remaining two G20 countries (Argentina and Saudi Arabia).

11
See eg Borio et al (1994) and Drehmann et al (2012).
12
The documentation accompanying the long series data set provides a full explanation of the
sources and methodologies used (http://www.bis.org/statistics/pp_long_documentation.pdf). For
instance, in order to interpolate annual values and construct quarterly series, the Chow-Lin
procedure (Chow and Lin (1971)) was used with construction costs or the housing component of
the CPI as reference series (or a combination of the two).

Nominal house price indices in Germany Graph A
Index, Q1 2008 = 100 Year-on-year growth rates, in per cent

Sources: See graph legend and Table A.

90
100
110
120
130
140
2008 2009 2010 2011 2012 2013 2014
Deutsche Bundesbank
vdp
Hypoport (EUROPACE)
0
2
4
6
8
10
Q2 12 Q4 12 Q2 13 Q4 13 Q2 14
Deutsche Bundesbank
vdp
Hypoport (EUROPACE)

The second goal is to broaden the collection of quality-adjusted price indices
that cover the whole residential property market (all types of dwelling and all
locations in a given country). For the moment, such series are available for about
two thirds of the 55 countries in the data set (Table 1).
The third goal is to enrich the coverage of the long series data set by adding
more countries with price series that meet the comparability standards, and by
extending the existing series back in time, whenever this is possible.
House prices as input to economic analysis
This section highlights the usefulness of the three BIS residential property price data
sets for different types of analysis that relies on comparisons across countries and
time. These range from obtaining an overview of recent price developments and
trends to the identification of common drivers in price dynamics and stylised facts
in their relationship with macroeconomic and financial stability.
House price trends internationally
Lately, house price developments have displayed greater diversity across countries
than they did in the immediate aftermath of the crisis. Analysis of the selected price
series shows that in some advanced economies, over the past year, real house
prices halted their decline and even staged a recovery while in others they
continued their downward trend. In contrast, in most emerging market economies
outside Europe, house prices generally rose (Graph 1).
Year-on-year residential property prices, deflated by CPI, rose by 9.5% in the
United States and 6% in the United Kingdom. Real house prices also grew, by 7% in
Canada, 7.7% in Australia and 2.2% in Switzerland, three countries that were less
affected by the crisis, as well as in some countries that were severely affected by the
crisis, such as Ireland (+7.2%) and Iceland (+6.4%).
Real price growth remained in negative territory in Japan (2.6%) and was
generally weak or negative in continental Europe. Prices rose in Germany (+1.2%)
and the Nordic countries (+1.7% in Denmark and +4.8% in Sweden), but continued
to fall in the euro areas southern periphery (Italy, 5%; Spain, 3.8%; Portugal,
1.2%; and Greece, 6%).
House prices generally grew in emerging regions outside Europe. In Asia, year-
on-year growth rates remained high in a number of countries in the first quarter of
2014 for instance, prices increased in real terms in China (+13%), the Philippines
(+13%) and Malaysia (+5%). In Latin American countries, the increase in real
residential property prices was more moderate. Prices in Brazil increased by 3.9% in
the first quarter of 2014, whereas in Mexico real prices were mostly stable compared
with one year ago. Price developments were mostly negative among central and
eastern European countries, but prices in the Baltic countries rebounded sharply.
Sustainability of house prices: valuation benchmarks
Are current house prices sustainable? Property valuations are a subject of intense
debate given the inherent subjectivity of the assessment and the volatility in the

demand trends. Comparisons of recent developments across countries and
benchmarking against historical valuation guides can provide additional insights
with which to assess the sustainability of recent house valuations. Two such
valuation guides widely used among economists and practitioners are the ratio of
rental income to the purchase price of the property (the housing asset equivalent to
the dividend ratio for stocks) and the ratio of house prices to household income (a
metric of affordability).
Graph 2 juxtaposes recent house price growth rates with the deviation of
current valuation ratios from their longer-term average values for the countries in
the database with sufficiently long series.
13
The horizontal axis in both panels

13
We require that we have at least 15 years of data for both ratios. The choice is somewhat arbitrary,
but consistent with the observation that credit and property cycles are of average duration in
excess of 15 years (Drehmann et al (2012)). We use the Selected series for all countries and, where
available, the long series to calculate the historical average of the ratio in each panel.
International house price indices
1

Full sample = 100 Graph 1
Advanced economies Advanced economies

Emerging Asia Emerging Europe Latin America

1
Figures are CPI-deflated and seasonally adjusted.
Sources: National data; BIS long series and selected property price database.
0
50
100
150
200
74 79 84 89 94 99 04 09 14
United States
United Kingdom
Australia
Spain
Ireland
France
Italy
0
50
100
150
200
74 79 84 89 94 99 04 09 14
Germany
Japan
Switzerland
Sweden
Netherlands
Denmark
Norway
50
75
100
125
150
175
87 90 93 96 99 02 05 08 11 14
China
Korea
Hong Kong
Thailand
Malaysia
India
50
75
100
125
150
175
93 96 99 02 05 08 11 14
Czech Republic
Lithuania
Bulgaria
Croatia
Hungary
Poland
50
75
100
125
150
175
99 02 05 08 11 14
Brazil
Mexico
Peru

measures the three-year growth rate of residential property prices for each country
on the basis of the selected price series. The graph clearly shows the very different
recent experience across countries, with declines of more than one quarter in some
countries and price hikes in excess of one third in others. Are these movements
indicative of market corrections or, instead, of shocks that suggest overshooting in
one direction or the other?
The vertical axis in the top panel shows the most recent value of the ratio of
house prices to rents, rebased using its historical average. A value of 110 indicates
that the current ratio is 10% higher than the historical average. Countries are quite
dispersed with respect to this benchmark. While for most countries the current ratio
implies that price movements are not diverging from rental values in ways that
imply unsustainability, for a number of other countries current property prices are
much higher than those implied by the historical relationship to rents. A priori, this
could be a reason to expect a price correction in the future. Interestingly, some of
these countries have experienced only moderate price growth recently (eg Canada,
Norway and Sweden), or even price declines (eg Australia, Belgium and France).
Relative benchmarks for house prices Graph 2

AU = Australia; AT = Austria; BE = Belgium; BG = Bulgaria; CA = Canada; CH = Switzerland; CZ = Czech
Republic; DE = Germany; DK = Denmark; EE = Estonia; ES = Spain; FI = Finland; FR = France; GB = United
Kingdom; HU = Hungary; IE = Ireland; IS = Island; IT = Italy; JP = Japan; KR = Korea; LT = Lithuania;
MY = Malaysia; NL = Netherlands; NO = Norway; NZ = New Zealand; PT = Portugal; SE = Sweden;
SG = Singapore; US = United States.
1
In Q1 2014; full sample = 100.
Sources: OECD; AMECO; national data; BIS long series and selected property price database; BIS estimates.
AT
AU
BE
BG
CA
CH
CZ
DE
DK
EE
ES
FI
FR
GB
HU
IE
IS
IT
JP
KR
LT
MY
NL
NO
NZ
PT
SE
SG
US
40
60
80
100
120
140
160
180
20 15 10 5 0 5 10 15 20 25 30 35 40
P
r
i
c
e
-
t
o
-
r
e
n
t

r
a
t
i
o
1
Real property price growth rate in the last three years, in %
AT
AU
BE
BG
CA
CH CZ
DE
DK
EE
ES
FI
FR
GB
HU
IE
IS
IT
JP
KR
LT
MY
NL
NO
NZ
PT
SE
SG
US
40
60
80
100
120
140
160
180
20 15 10 5 0 5 10 15 20 25 30 35 40
Real property price growth rate in the last three years, in %
P
r
i
c
e
-
t
o
-
i
n
c
o
m
e

r
a
t
i
o
1

The vertical axis in the bottom panel shows the most recent value of the
affordability benchmark: the ratio of prices to disposable income per capita, also
rebased using its historical average. The graph suggests that for most countries the
current ratio is not too far from this benchmark, suggesting that prices have not
diverged dramatically from income trends. For a number of countries, however,
current property prices are at least one fifth higher than those implied by the
historical relationship to incomes, suggesting potential downward pressures on real
house prices. This might lead to a reversal or moderation of recent growth (eg in
Canada) or a further sliding of prices (eg in Belgium and France). This argument
would be more compelling for markets where prices have grown rapidly in the
recent past, and where income growth is projected to be rather moderate.
House price cycles and international co-movement
Long time series are also useful for understanding the dynamics of house prices.
Research has pointed to long cycles in property prices that are associated with
similar cycles in the availability of credit (Drehmann et al (2012)). Graph 3 shows the
average growth rate of real house prices across all countries in the Selected and
Long series databases over five-year rolling time windows (red line). The series
shows the remarkable regularity of a global cycle with length in excess of 10 years
and of increasing amplitude over time.
Interestingly, the intensity of house price co-movement varies over time (blue
bars). It was generally high for the years between the mid-1980s and late 1990s,
whereas it declined markedly during the boom years of the first decade of this
century, only to increase again in the aftermath of the global financial crisis. Closer
inspection suggests that while most of the time the intensity of co-movement
mirrors the direction of average real property prices, there are periods when the two
move in opposite directions (198993, 199799, 200008).
House prices as early warning indicators for financial instability
How can property prices contribute to financial stability analysis? Work at the BIS
Asymmetric co-movement of house prices across countries
1
Graph 3
Per cent
1
Calculated using CPI-deflated residential property prices across all countries in the database; mean values.
Sources: BIS long series and selected property price database; authors calculations.
40
20
0
20
40
0.4
0.2
0.0
0.2
0.4
80 82 84 86 88 90 92 94 96 98 00 02 04 06 08 10 12 14
Five-year cumulative growth
Lhs:
Five-year rolling correlations of y-o-y growth rates
Rhs:

has pointed to the early warning indicator properties of real estate prices. Leverage-
fuelled housing booms that turn into busts have so often been at the very heart of
episodes of systemic distress. Historical experience has demonstrated that the
interactions between rapidly growing house prices and excessive credit expansion
are a tell-tale sign of the build-up of vulnerabilities in the household sector and the
source of future losses for banks (Borio and Drehmann (2009)).
We follow Drehmann and Juselius (2014) to illustrate the statistical relationship
between residential price booms and systemic banking crises in the form of the
AUC statistic.
14
In short, the AUC measures the success rate of a particular indicator
in correctly predicting future occurrences (in this case, a systemic crisis) while at the
same time minimising false positive signals (ie flagging a potential crisis without
one occurring). The closer to unity the value of the AUC for a given indicator at a
specific horizon, the more informative the indicator.
The left-hand panel of Graph 4 shows the AUC metric for real property price
growth over forecast horizons that range from 20 quarters to one quarter prior to a
banking crisis.
15
The AUC for the house price indicator is clearly above 0.5, the
threshold for being informative (see footnote 14), for several years prior to the
crisis. However, the quality of its signal drops significantly in the five quarters prior
to the crisis date. This is because property price declines are in many instances a
hallmark of the early stages of a financial bust. However, up to that point the signal
they emit is quite stable.

14
The area under the curve (AUC) captures the trade-off between true positives and false positives for
the full range of policymaker preferences (see Box A in Drehmann and Tsatsaronis (2014) for a
description). A completely uninformative indicator has an AUC of 0.5: it is no better than tossing a
coin. The greater the difference of the AUC from 0.5, the better the indicator balances the trade-off
between missed crises and false positives. The perfect indicator has an AUC of 1 or zero, depending
when higher or lower values of the indicator predict crises.
15
The analysis is based on approximately 50 crisis episodes (see Drehmann and Juselius (2014) for
details).
Property prices, credit booms and systemic crises
AUCs for different forecast horizons Graph 4
Property price growth Property price growth and credit-to-GDP gap

The horizontal axis denotes the forecast horizons in quarters before crises (the timing of which is denoted by the zero point). The vertical
axis denotes the AUC. The horizontal line at 0.5 corresponds to the AUC value of a completely uninformative indicator. The blue line shows
the AUC of the indicator for the given horizon.
Source: Drehmann and Juselius (2014).
0.1
0.3
0.5
0.7
0.9
20 15 10 5 0
0.1
0.3
0.5
0.7
0.9
20 15 10 5 0

The performance of property prices dramatically improves when their signals
are combined with those of other early warning indicators, and in particular
indicators more directly linked to the behaviour of credit. The right-hand panel of
Graph 4 shows the AUC of a signal combining real property growth and the credit-
to-GDP gap, defined as the difference of the credit-to-GDP ratio from its historical
trend.
16
The AUC of the combined indicator is much higher than that of real
property price growth, and most importantly it stays high throughout the forecast
horizon up to the crisis date.
Conclusions
House prices are important inputs to conjunctural analysis of macroeconomic and
financial stability risks as well as to research into the interactions of the real and
financial sectors. The three sets of residential price data published by the BIS should
provide analysts of international housing markets with a wealth of easily accessible
and comparable data on price developments. This should stimulate further analysis,
including in the form of cross-country benchmarking of valuations and the
commonalities in house price dynamics. In addition, the availability of price series
with long time spans should stimulate research on property price cycles and
interactions with other macroeconomic and financial variables.
The BIS residential price series could potentially also act as a catalyst in
improving the availability of property price statistics. For one, it may stimulate
efforts towards greater standardisation of the statistical processes underpinning the
national compilation of these data. Importantly, this could possibly extend to the
compilation of comparable price indices for commercial property prices. Recalling
that exposure to frothy commercial real estate markets was at the root of many an
episode of banking system stress, this would greatly enhance our analytical tools.
Data on commercial property markets are very patchy and hard to come by, at the
national level and especially across countries.

16
The indicators are combined by considering that a signal has been emitted if either of the two is
signalling (Drehmann and Juselius (2014)).

References
Bank for International Settlements (1989): 59rd Annual Report, June.
Borio, C and M Drehmann (2009): Assessing the risk of banking crises revisited,
BIS Quarterly Review, March, pp 2946.
Borio, C, N Kennedy and S Prowse (1994): Exploring aggregate asset price
fluctuations across countries: measurement, determinants and monetary policy
implications, BIS Economic Papers, no 40, April.
Chow, G and A Lin (1971): Best linear unbiased interpolation, distribution and
extrapolation of time series by related series, The Review of Economics and
Statistics, vol 53, no 4, November, pp 3725.
Davis, E and H Zhu (2005): Commercial property prices and bank performance, BIS
Working Papers, no 175, April.
Drehmann, M, C Borio and K Tsatsaronis (2012): Characterising the financial cycle:
dont lose sight of the medium term!, BIS Working Papers, no 380, June.
Drehmann, M and M Juselius (2014): Evaluating early warning indicators of banking
crises: satisfying policy requirements, International Journal of Forecasting,
forthcoming (also available as BIS Working Paper no 421, August 2013).
Drehmann, M and K Tsatsaronis (2014): Countercyclical capital buffers and the
credit-to-GDP gap: questions and answers, BIS Quarterly Review, March, pp 5573.
gert, B and D Mihaljek (2007): Determinants of house prices in central and eastern
Europe, BIS Working Papers, no 236, September.
Eurostat (2013): Handbook on Residential Property Prices Indices (RPPIs),
Methodologies and Working papers, available at http://epp.eurostat.ec.europa.eu/
portal/page/portal/hicp/methodology/hps/rppi_handbook.
Glindro, E, T Subhanij, J Szeto and H Zhu (2010): Determinants of house prices in
nine Asia-Pacific economies, BIS Papers, no 52, July.
Takts, E (2012): Aging and house prices, Journal of Housing Economics, vol 21,
no 2, June, pp 13141.
Tsatsaronis, K and H Zhu (2004): What drives housing price dynamics: cross country
evidence, BIS Quarterly Review, March, pp 6578.

Bis q3-14

Uploaded by

Copyright:

Available Formats

Bis q3-14

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Bis q3-14

Uploaded by

Copyright:

Available Formats

BIS Quarterly Review

50% Between 0% and 10%

You might also like