determinants of financial stress in emerging market economies

in

uyon

of 2ss iroue exansatesSIsionaresu

demonthougho econncial c06; Morkets in lateal syste

prices and capital ows during times of crises. Earlier studies clas-sied the causes of nancial contagion into two broad categories(Calvo and Reinhart, 1996; Dornbusch et al., 2000; Kaminsky andReinhart, 1999, 2000; Moser, 2003). First, nancial asset pricesand capital ows can move similarly when economies share simi-lar fundamentals and have strong macroeconomic interdepen-dence through trade and nancial linkages. Similar fundamentals

uthors devKeeton (20

the Federal Reserve Bank of Kansas City; Hollo et al. (20European markets; Misina and Tkacz (2009) for selected adeconomies; and Yiu et al. (2010) for Hong Kong MoAuthority.

The use of FSI has far-reaching benets for monetary authoritiesand nancial regulatory and supervisory agencies. First, FSI com-bines various nancial market indicators into an aggregate indexto measure nancial market stress, hence eliminating the depen-dency on one or few indicators in the measurement of nancialstress. Second, FSI allows for a measure of nancial stress to cap-ture the degree and severity of nancial stress on a continuous

Corresponding author. Tel.: +63 2 632 4444.

Journal of Banking & Finance xxx (2013) xxxxxx

Contents lists availab

k

w.E-mail address: [email protected] (C.-Y. Park).substantial body of economic literature. Cross-border transmissionof nancial crises is often manifested in co-movements of asset

expectations of loss in nancial markets other atheir own versions of FSI, including Hakkio and0378-4266/$ - see front matter 2013 Elsevier B.V. All rights reserved.http://dx.doi.org/10.1016/j.jbankn.2013.09.018

Please cite this article in press as: Park, C.-Y., Mercado Jr., R.V. Determinants of nancial stress in emerging market economies. J. Bank Financehttp://dx.doi.org/10.1016/j.jbankn.2013.09.018eloped09) for12) forvancednetarykets are rapidly integrated into global and regional markets, theorigin of nancial stress is also becoming ubiquitous and the im-pact felt borderless. For example, the nancial crisis which startedin Thailand in 1997 quickly spread to the rest of East Asia, and thento the Russia Federation and Brazil.

The transmission of nancial crises has been the subject of a

stress index (FSI) as a continuum and contemporaneous measureof the severity of nancial crises. Using FSI shows that nancialstress intensies due to greater fragility in the nancial systemsand exogenous shocks. Since the pioneering work of Illing andLiu (2006) who dened nancial stress as episodes where eco-nomic agents are subjected to extreme uncertainty and varying1. Introduction

The recent global nancial crisisfects of nancial globalization. Albrings direct and indirect benets tthe countries vulnerability to nawhere (Stulz, 2005; Kose et al., 20stance, the freezing of the credit maparticularly the United States (US), iturmoil in emerging market nancistrates the adverse ef-nancial integration

omies, it may increaserises originating else-shirian, 2008). For in-n advanced economies,2008 caused signicantms. As emerging mar-

may induce similar response to a shock, which leads to strongco-movements in asset prices and capital ows. Second, the co-movements may also result from herding behaviors and/or certaindecisions of investors, which affect different countries simulta-neously. For example, a crisis in one country may prompt investorsto withdraw from all emerging market countries.

While the impact of nancial crisis is often devastating espe-cially in emerging market economies, it has not been easy to mon-itor the buildup of a full-blown nancial crisis and to trace itsspread across borders. A number of studies have used a nancialFinancial contagionEmerging market economies

restrictions indicate that although a domestic nancial shock still accounts for most of the variation indomestic FSI, regional shocks play an important role in emerging Asia.

2013 Elsevier B.V. All rights reserved.Determinants of nancial stress in emerg

Cyn-Young Park , Rogelio V. Mercado Jr.Economics and Research Department, Asian Development Bank, 6 ADB Avenue, Mandal

a r t i c l e i n f o

Article history:Available online xxxx

JEL classication:F30G01G15

Keywords:Financial stress index

a b s t r a c t

The global nancial crisistrigger severe nancial stretagion show the linkages thkets. This paper extends ththe channels of nancial trour panel regression estimgional emerging market Fthat there is a common regEurope. Furthermore, the

Journal of Ban

journal homepage: wwg market economies

g City 1550, Philippines

0082009 illustrates how nancial turmoil in advanced economies couldn emerging markets. Previous studies dealing with nancial crises and con-gh which nancial stress are transmitted from advanced to emerging mar-isting literature on the use of nancial stress index (FSI) in understandingmission in emerging market economies. Using FSI of 25 emerging markets,show that not only advanced economies FSI, but also regional and nonre-

signicantly increase domestic nancial stress. Our ndings also suggestl factor signicantly affecting domestic FSI in emerging Asia and emerginglts from a structural vector autoregression model with contemporaneous

le at ScienceDirect

ing & Finance

elsevier .com/locate / jbf(2013),

basis such that spikes in the stress index correspond to the periods market nancial shocks. This will help assess whether or not theimpact of a nancial shock on domestic FSI would differ by the ori-gins of the shock such as different economic (advanced versus

structs nancial stress index. Section 3 presents the determinants

2 C.-Y. Park, R.V. Mercado Jr. / Journal of Banking & Finance xxx (2013) xxxxxxof severe nancial stress. It also allows fast, reliable identicationof crises or stressful periods. Third, FSI offers an aggregate measureof nancial stability that is much simpler than other measures ofsystemic nancial risks, without the complications of micro-levelassumptions of other measures.1 Nevertheless, FSI is found veryuseful in alerting systemic risk conditions by its signaling propertiesfor identifying systemic stress episodes, grade thresholds, and corre-sponding probability of systemic stress (Oet et al., 2011). Of course,FSI has its own shortcomings; particularly pertaining to its construc-tion including choice variables, aggregation, and frequency choice.

Existing studies dealing with nancial stress, however, offer lit-tle insight on various channels for transmission of nancial stressemanating from advanced economies and emerging market econo-mies (either from the same region or from different regions) todomestic nancial markets. Balakrishnan et al. (2009, 2011) ex-plores the issue of nancial transmission from advanced and otheremerging economies to individual emerging market economies,but paid little attention to geographical proximity in nancialtransmission and did not distinguish between regional and nonre-gional markets. Fernandez (2007) studied the impact of instabilityin the Middle East to regional and nonregional emerging stockmarkets; however, she focused on political stability as the sourceof nancial market turmoil. This paper addresses the gap in the lit-erature. Specically, it aims to examine the determinants of nan-cial stress in emerging market economies and to assess thetransmission of nancial shocks emanating from advanced andother regional and nonregional emerging market economies toindividual emerging market economies. This paper adds to the pre-vious literature in the following aspects.

First, it covers a longer period by extending the observationsfrom 1992 to 2012 to include a number of episodes of emergingmarket crises in the early to mid-1990s as well as the latest crisisepisode that affected the global nancial markets in 20082009.Extending the sample period with more crisis episodes will allowfor the analysis to provide more reliable results regarding thetransmission of a nancial shock.

Second, this paper employs two methodologies for constructingdomestic FSI for each emerging market in the sample one is thevariance-equal weights and the other is the principal componentanalysis. This will allow for robustness checks on the overall pat-terns of individual FSIs.

Third, this study assesses the impact of external nancial shockson domestic FSI by differentiating their economic and geographicorigins, such as advanced versus emerging market economies aswell as regional versus nonregional emerging market economies.The analysis specically focuses on whether or not a shock origi-nating from emerging market economies would exert inuenceon the FSI of an individual emerging market economy in additionto a shock from advanced economies. It will also assess the effectof a common regional factor in domestic FSI for emerging markets.The signicance of a common regional factor would help explainthe vulnerability of emerging market countries to regional nan-cial contagion.

Fourth, following the panel regression analysis on the magni-tude and signicance of advanced and other regional and nonre-gional emerging market FSI on domestic FSI, we employ impulseresponse functions and variance decompositions to assess the im-pact of a nancial shock coming from advanced and other emerg-ing market economies on individual emerging market economiesFSI. A nancial shock generated from other emerging market econ-omies are decomposed into regional and nonregional emerging1 See Arnold et al. (2012) for discussion on the challenges in monitoring bankingsystemic risks; and Allen et al. (2012) for discussion on the micro-level systemic riskmeasures.

Please cite this article in press as: Park, C.-Y., Mercado Jr., R.V. Determinantshttp://dx.doi.org/10.1016/j.jbankn.2013.09.018of FSI and provides empirical specication for the panel regressionwhich will determine the signicance of advanced and otheremerging market nancial stress on domestic FSI, as well as theimportance of a common-regional factor. Section 4 provides thestructural vector autoregression specication and presents the im-pulse response functions and variance decompositions of FSIshocks from advanced and other emerging market economies ondomestic emerging market FSI. Summary and some policy implica-tion follow in Section 5.

2. Financial stress index

2.1. Literature review

There is abundant literature investigating the occurrence anddeterminants of currency, banking, and sovereign debt crises in ad-vanced and emerging economies; however, this literature failed toaccount for the proper dating and intensity of said crises. For in-stance, Laeven and Valencia (2008) developed a database on thetiming and frequency of banking, currency, and sovereign debt cri-ses for both advanced and emerging markets. Eichengreen et al.(2004) looked into currency crises by developing an index of for-eign exchange market pressure which incorporates foreign ex-change depreciation and changes in international reserves.Reinhart and Rogoff (2008) studied sovereign debt defaults andfound that crises usually emanate from nancial centers; and wereoften accompanied by other crises like currency and banking cri-ses. However, these studies devote little attention in dealing withthe contemporaneous severity of nancial crises. This comes fromthe fact that most studies measure the occurrence of crises as asimple binary variable, i.e., no crisis takes the value of zero (0)and presence of crisis takes a value of one (1). As pointed out by

2 Emerging Asia includes the Peoples Republic of China; Hong Kong, China; India;Indonesia; the Republic of Korea; Malaysia; the Philippines; Singapore; Taipei,China;and Thailand. Emerging Americas includes Argentina, Brazil, Chile, Colombia, Mexico,emerging market economies) and geographic (regional versus non-regional) groupings.

To carry out the empirical analysis of this study, aggregatedomestic FSI are constructed, drawing on the methodology usedby Cardarelli et al. (2011) and Balakrishnan et al. (2011), for a sam-ple of emerging market economies using variance-equal weightsand principal component analysis as aggregation technique. Toverify the importance of global, country-specic, other countriesnancial stress, and regional factors in explaining domestic FSI, apanel regression model involving 25 emerging markets from vari-ous regions including emerging Asia, emerging Americas, emergingEurope, and other emerging countries is employed using quarterlydata from Q1 1992 to Q4 2012.2 Specically, it aims to determinewhich factors including common regional factors contribute tothe increase of nancial stress in developing economies. Knowingthe signicance of advanced and other emerging market FSI in deter-mining domestic FSI, we examine the impact of a nancial shockemanating from advanced and other emerging economies on indi-vidual domestic FSI using a structural vector regression approachwith contemporaneous restrictions. This will allow us to determinethe magnitude and persistence of the effects of advanced and otheremerging market nancial shocks on individual domestic FSI.

This paper proceeds as follows. Section 2 discusses and con-and Peru. Emerging Europe includes Czech Republic, Hungary, Poland, RussianFederation, and Romania. Other emerging countries include Egypt, Israel, SouthAfrica, and Turkey.

of nancial stress in emerging market economies. J. Bank Finance (2013),

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12ARG_PC ARG_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12ATR_PC ATR_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12AUS_PC AUS_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12BRA_PC BRA_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12

CAN_PC CAN_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12

CHI_PC CHI_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12

COL_PC COL_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12

CZE_PC CZE_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12

DEN_PC DEN_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12EGY_PC EGY_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12

FRA_PC FRA_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12

GER_PC GER_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12

HKG_PC HKG_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12

HUN_PC HUN_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12

IND_PC IND_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12

INO_PC INO_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12

ISR_PC ISR_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12

ITA_PC ITA_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12

JPN_PC JPN_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12

KOR_PC KOR_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12

MAL_PC MAL_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12

MEX_PC MEX_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12

NET_PC NET_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12

NOR_PC NOR_SUM

Fig. 1. Individual country nancial stress index. PC = principal components; SUM = variance-equal weights. Note: The computed FSI corresponds to the aggregated valuesshown in Eq. (6). Monthly average data on banking sector price index and the benchmark stock price index were taken from DataStream. The data were converted to year-on-year returns by taking the difference between current period and last years price index both in natural logarithm form. Monthly data for both foreign exchange and foreignreserves are taken from the International Financial Statistics of the International Monetary Fund. Sovereign debt data refers to yield differentials between long-term domesticgovernment and US Treasury bonds in basis points. Monthly average data on sovereign debt spreads were taken from national sources accessed through CEIC Database.However, for some countries where data started in the late 1990s, government treasury yield spreads of comparable tenure was used. In cases where data is unavailable forcertain months of the year, available data was extended to cover the whole year. For some countries with unavailable treasury yield data in the early 1990s, policy rates wereused. Source: Authors estimate.

C.-Y. Park, R.V. Mercado Jr. / Journal of Banking & Finance xxx (2013) xxxxxx 3

Please cite this article in press as: Park, C.-Y., Mercado Jr., R.V. Determinants of nancial stress in emerging market economies. J. Bank Finance (2013),http://dx.doi.org/10.1016/j.jbankn.2013.09.018

812

16

12

16

2-1

-

-

12

16

8

12

16

-1

-

-

1

1

1

1

4 C.-Y. Park, R.V. Mercado Jr. / Journal of Banking & Finance xxx (2013) xxxxxx-12

-8

-4

0

4

92 94 96 98 00 02 04 06 08 10 12

PER_PC PER_SUM

-12

-8

-4

0

4

8

92 94 96 98 00 02 04 06 08 10 1

PHI_PC PHI_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12ROM_PC ROM_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12RUS_PC RUS_SUM

-4

0

4

8

12

16

-4

0

4

8

12

16Balakrishnan et al. (2011) and Illing and Liu (2006), the use of bin-ary variable for crisis occurrence and dating does not provide ameasure of intensity of crisis and near-miss events.3 However,some studies used a sector-specic index to measure sector-specicintensity of crisis. For instance, Hanschel and Monnin (2005) deriveda banking stress index for Switzerland, although stress from othernancial sectors is not considered. Furthermore, most of these stud-ies do not include crises that emanate from the equity markets.

Against this backdrop, several authors proposed an FSI to ad-dress the weaknesses of previous literature in dating and measur-ing the severity of nancial crises. Several research worksdeveloped FSI by capturing key features of nancial stress to iden-tify a buildup in nancial stress and measure the intensity of nan-cial crisis. Illing and Liu (2006) created an index of nancial stressfor the Canadian nancial system, employing a continuous variablewith a spectrum of values where extreme values correspond toperiods of nancial crises. Their method was adapted and renedby Cardarelli et al. (2008, 2011) which was used by the Interna-tional Monetary Fund in their World Economic Outlook 2008; whileBalakrishnan et al. (2009, 2011) developed a similar index for 18

-12

-8

92 94 96 98 00 02 04 06 08 10 12SPA_PC SPA_SUM

-12

-8

92 94 96 98 00 02 04 06 08 10 12SWE_PC SWE_SUM

-1

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12THA_PC THA_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12TUR_PC TUR_SUM

-1

-

-

1

1

Fig. 1 (cont

3 Some periods of heightened nancial market stress do not evolve into full-blownnancial crisis. For example, the emerging market equity sell-off in June 2006 hadlittle macroeconomic impact, although it raised asset price volatility in somecountries.

Please cite this article in press as: Park, C.-Y., Mercado Jr., R.V. Determinantshttp://dx.doi.org/10.1016/j.jbankn.2013.09.0182

8

4

0

4

8

92 94 96 98 00 02 04 06 08 10 12POL_PC POL_SUM

-12

-8

-4

0

4

92 94 96 98 00 02 04 06 08 10 12

PRC_PC PRC_SUM

2

8

4

0

4

8

2

6

92 94 96 98 00 02 04 06 08 10 12SAF_PC SAF_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12SNG_PC SNG_SUM

-4

0

4

8

2

6

-4

0

4

8

12

16emerging economies and used at the subsequent issue of IMFsWorld Economic Outlook 2009.4

The use of FSI as a means of dating the duration and assessingthe severity of nancial crises has also gained popularity amongmonetary authorities and nancial regulatory institutions. Forexample, the Federal Reserve Bank of Cleveland, Kansas, and St.Louis post their respective FSI on their website. Other researcherscovering advanced and emerging economies employ the said in-dex, albeit with different component nancial variables to suitdomestic nancial market characteristics. For example, unlike theuse of foreign exchange market pressure index, Yiu et al. (2010)used at-the-market implied volatility of Hong Kong dollar per USdollar for the exchange rate component of the FSI. Van Roye(2011) used several indicators to measure nancial stress in Ger-manys banking sector, including Treasury bill and Eurodollar fu-ture contract (TED) spread, money market spread, and bankingbeta.

Another branch of literature on FSI examines the link betweennancial stress and economic activity. For example, Davig and Hak-kio (2010) found the US economy uctuates between episodes of

2

-8

92 94 96 98 00 02 04 06 08 10 12SWI_PC SWI_SUM

-12

-8

92 94 96 98 00 02 04 06 08 10 12

TAP_PC TAP_SUM

2

8

4

0

4

8

2

6

92 94 96 98 00 02 04 06 08 10 12UKD_PC UKD_SUM

-12

-8

-4

0

4

8

12

16

92 94 96 98 00 02 04 06 08 10 12USA_PC USA_SUM

inued)

4 The main difference between the advanced and emerging economies FSI used bythe IMF is the inclusion of indicators such as corporate bond spreads, inverted termspread, and TED spread for advanced economies, which is inapplicable to emergingeconomies given the low issuance of corporate bonds and data unavailability inemerging economies.


f BaC.-Y. Park, R.V. Mercado Jr. / Journal olow nancial stress and high economic activity; and high nancialstress and low economic activity. Other papers in this eld dealswith the contribution of nancial stress index to improving fore-casts on economic activity. Ng (2011) showed FSI improves fore-casting performance at horizons of 24 quarters for the USeconomy. Afonso et al. (2011), Dovern and van Roye (2013), andvan Roye (2011) found that nancial stress lowers economic out-put and worsens scal positions. Cardarelli et al. (2011) identiedepisodes of nancial turmoil in advanced economies using FSI andassessed the impact of nancial stress on the real economy. Theyfound that nancial turmoil characterized by banking distress ishighly associated with severe and protracted downturns thanstress originating from securities or currency markets. In addition,they also argued that economies with more arms-length nancialsystems appear to be particularly vulnerable to sharp contractions.

This paper follows the eld of literature set by Balakrishnanet al. (2011) which employs FSI to examine cross-border transmis-

Fig. 2b. Advanced and emerging markets nancial stress index and stress episodes (byunweighted averages. Aggregate emerging market nancial stress periods are indentieregional trend, where the regional trend is computed as the average of individual emerg

Fig. 2a. Advanced and emerging markets nancial stress index an

Please cite this article in press as: Park, C.-Y., Mercado Jr., R.V. Determinantshttp://dx.doi.org/10.1016/j.jbankn.2013.09.018nking & Finance xxx (2013) xxxxxx 5sion of nancial stress. Balakrishnan et al. (2011) created FSI foremerging economies using the same methodology as Cardarelliet al. (2011). They argued that domestic nancial stress index inan emerging economy is inuenced by nancial stress in advancedeconomies as well as common factors like global gross domesticproduct (GDP) growth and interest rates; and country-specic fac-tors like degree of nancial and trade linkages and other domesticmacroeconomic vulnerabilities. Their ndings suggest that nan-cial crises in advanced economies pass-through strongly to emerg-ing economies; and that the depth of nancial linkages betweenthe two determines the extent of pass-through.

While the previous literature focused on the impact of nancialtransmission from advanced to emerging markets, this paper ar-gues that a shock emanating from emerging markets is also impor-tant as emerging markets gain increasing presence in the globalnancial system. This paper also examines evidence of regionalnancial contagion by focusing on the transmission of a regional

region, using principal component analysis). Note: Aggregate and regional FSIs ared based on the difference between the unweighted regional FSI and its unweighteding market trend using Hodrick-Prescott lter method. Source: Authors calculation.

d stress episodes (by region, using variance-equal weights).


f Ba6 C.-Y. Park, R.V. Mercado Jr. / Journal oshock. We look into the signicance of a common regional factor asanother key determinant of domestic nancial stress index. Ifdomestic FSI is signicantly affected by a common regional factor,we may conclude that there is evidence of regional nancial conta-gion. In this regard, the paper also assesses the effect of regionaland nonregional nancial shocks on domestic FSI.

2.2. Constructing emerging market nancial stress index

This paper follows the denition of nancial stress for emergingmarkets as suggested by Balakrishnan et al. (2011). They denenancial stress as episodes when the nancial system is understrain and its ability to intermediate is impaired. It is usually asso-ciated with the following: (1) large shifts in asset prices; (2) abruptincrease in risk or uncertainty; (3) illiquidity of the nancial sys-tem; and (4) concerns about the health of the banking system.

Fig. 3b. Emerging markets nancial stress index and stress episodes (by country, usingequal weights. Source: Authors calculation.

Fig. 3a. Emerging markets nancial stress index and stress

Please cite this article in press as: Park, C.-Y., Mercado Jr., R.V. Determinantshttp://dx.doi.org/10.1016/j.jbankn.2013.09.018nking & Finance xxx (2013) xxxxxxWhile we try to construct an index to capture the aforemen-tioned conditions of nancial stress, there are three key issues.First is to identify components of the index to cover key nancialsectors. Second is to choose right variables to represent each com-ponent. And, third, what weighting scheme to use to aggregateeach component to a single FSI. Each is discussed accordingly, withan explanation on how we construct an FSI for each of the emerg-ing markets in this paper.

2.2.1. Components and variable choicesWe construct an FSI for each of the 25 emerging economies and

15 advanced economies following Balakrishnan et al. (2009, 2011),Cardarelli et al. (2008, 2011), and Yiu et al. (2010). As in the previ-ous studies, the composite FSI for each economy covers the fourmajor nancial sectors of the economy, which include:

principal component analysis). Note: PC = principal components; SUM = variance-

episodes (by country, using variance-equal weights).


f Ba(a) Emerging Economies

C.-Y. Park, R.V. Mercado Jr. / Journal oBanking sector: The lack of suitable data and institutional differ-ences across countries make it hard to have a clear denition ofwhat constitutes a banking crisis. Some studies use ad hoc coun-try-specic events to dene banking crisis. Others rely more on acombination of qualitative and quantitative approach. For instance,Kunt and Detragiache (1996) dene banking crisis as a situationwhere any of the following conditions holds: (i) non-performingloans is greater than 10%; (ii) the cost of bank rescue is at least2% of GDP; (iii) banking problems result to large scale nationaliza-tion of banks; and (iv) extensive bank runs lead to emergency mea-sures. Furthermore, some studies rely on quantitative methodsusing aggregate balance sheet data of banks.

In constructing FSI, we include a measure of banking stresscalled banking sector b as in Balakrishnan et al. (2011). This mea-

(b) Advanced Economies

(c) Emerging Asia

Fig. 4. Components of nancial stress index. Note: Aggregate components of FSIs are unwweights. Source: Authors calculation.

Please cite this article in press as: Park, C.-Y., Mercado Jr., R.V. Determinantshttp://dx.doi.org/10.1016/j.jbankn.2013.09.018(d) Emerging Americas

nking & Finance xxx (2013) xxxxxx 7sure involves the ratio of bank share prices to total share prices.It provides a stationary measure of relative equity-return volatilityand isolates banking sector-specic shocks. The banking sector b isgiven by:

b covr;mvarm 1

where r and m are the returns to the banking sector stock price in-dex and the overall stock price index, respectively. If b is larger than1, then the banking sector is relatively risky as the volatility of re-turns on bank shares is greater than the volatility of returns forthe overall market. The higher the banking sector b, the greaterthe banking sectors stress. It must be noted that banking sector

(e) Emerging Europe

(f) Other Emerging Markets

eighted averages of individual country. components computed using variance-equal


were converted to month-on-month returns by taking the differ-ence between current and previous months stock price index bothin natural logarithm form.

Debt markets: Illing and Liu (2006) denes debt crisis as theinability of sovereign nations or the private sector to service its for-eign debts. Earlier literature on debt crises deals with a group ofemerging economies that were exposed to severe external indebt-edness in the early-1980s. However, the occurrence of emergingeconomies debt crises was identied mainly based on qualitativeinformation. The most common indicator of debt crises has beenthe spread between risky and risk-free bond yields as a functionof expected losses. Spreads will widen if expectations of futurelosses increase, or if greater uncertainty leads to lower condence,implying a higher probable loss. Both factors are indicative ofstress.

This paper uses sovereign debt spreads to measure sovereigndebt stress. Data refers to yield differentials between long-term(10-year) local government bonds and US Treasuries in basispoints. Monthly average data on sovereign debt spreads were ta-ken from national sources accessed through CEIC Database. How-ever, for some countries where data started in the late 1990s,government treasury yield spreads of comparable tenure was used.In cases where data is unavailable for certain months of the year,available data was extended to cover the whole year. Furthermore,

Table 1Principal component analysis: eigenvalue components.

5 The estimated variance equation for the equity returns following GARCH(1,1)shows that the estimated lagged squared residual and lagged variance are mostly

f Banking & Finance xxx (2013) xxxxxxbeta is not a measure of co-movement between the two variables.What it provides is a measure of how the banking sector returnsare more volatile than the overall stock price returns.

Some studies, including Cardarelli et al. (2008, 2011), van Roye(2011), and Yiu et al. (2010), use risk spreads such as TED spreadand inverted term spread as a proxy for banking sector stress.However, including other banking sector variables may pose as aconstraint for constructing FSI for each emerging market econo-mies in the sample as most economies have relatively short timeseries data for said indicators. For this reason, this paper will onlyuse banking sector b as a measure of banking stress in emergingmarkets.

Monthly average data on the banking sector price index and thebenchmark stock price index were taken from DataStream. Thedata were converted to year-on-year returns by taking the differ-ence between current period and last years price index both innatural logarithm form. Twelve-month rolling covariance and var-iance of returns were used to compute for the banking sector b. Tobetter capture banking sector stress, the derived series takes onlypositive values exceeding a threshold of one and zero otherwise.

Foreign exchange market: Currency crises are dened as periodsof signicant devaluations, losses in foreign exchange reserves,and/or defensive interest rate hikes. This study utilizes exchangemarket pressure index (EMPI) as proposed by Eichengreen et al.(2004) and used in Balakrishnan et al. (2011). The EMPI capturesthe depreciation of the local currency with respect of US dollarand the reduction in foreign exchange reserves. It is dened as:

EMPIi;t Dei;t li;De

ri;De DRESi;t li;DRES

ri;DRES2

where De and DRES denote month-on-month percent changes inthe foreign exchange rate of local currency per US dollar and foreignexchange reserves; while r and l are standard deviation and mean,respectively. Monthly data for both foreign exchange and foreignreserves are taken from the International Financial Statistics of theInternational Monetary Fund. Other methods of measuring foreignexchange stress include hybrid volatility-loss approach such asthe CMAX calculation as used by Illing and Liu (2006) and foreignexchange volatility following GARCH (1,1) as in Bollerslev et al.(1992). Eicher et al. (2009) developed an indicator of currency crisisrisk using price spreads between American Depositary Receipts(ADR) and their underlying. They found ADR investors perceivehigher currency crisis risk when export commodity prices decline,sovereign yield spreads increase, trading partners currencies depre-ciate, and interest rate spreads widen. This paper utilizes only theEMPI as a measure of foreign exchange market stress.

Equity market: Previous studies dene equity crises as a sharpdecline in the overall stock price index. The drop suggests greaterexpected loss, higher risk, or increased uncertainty about rms fu-ture prots. The simplest measure of equity crisis is the use of aGARCH (1,1) process to take into account time-varyingcharacteristics of movements in equity returns, followingBollerslev et al. (1992). The volatility following a GARCH (1,1) pro-cess is given by:

r2t x /1e2t1 /2r2t1 3where r2 refers to the variance and e the error term in the regres-sion given by:

yt ai;t byt1 ei;t 4where yt is the current periods equity return and yt1 is the previ-ous periods equity returns. Balakrishnan et al. (2009, 2011), Cardar-

8 C.-Y. Park, R.V. Mercado Jr. / Journal oelli et al. (2008, 2011), and Yiu et al. (2010) used the same approachin constructing equity market crisis index. This study will also usetime-varying volatility of stock returns as a measure of equity

Please cite this article in press as: Park, C.-Y., Mercado Jr., R.V. Determinantshttp://dx.doi.org/10.1016/j.jbankn.2013.09.018market stress.5 Monthly average data on benchmark stock price in-dex were taken from Datastream. The data were converted tomonth-on-month returns by taking the difference between currentand previous months stock price index both in natural logarithmform.

Aside from the stock market volatility measure, we also includestock market returns as a component for equity market stress. Thestock market returns are computed from monthly average data onbenchmark stock price index availed from DataStream. The data

Eigenvalue components 1 2 3 4 5 PC1PC3

Argentina 0.383 0.234 0.163 0.142 0.077 0.781Brazil 0.412 0.215 0.181 0.113 0.079 0.808Chile 0.363 0.233 0.183 0.124 0.097 0.779Colombia 0.339 0.231 0.182 0.155 0.093 0.752Czech Republic 0.345 0.238 0.180 0.146 0.091 0.763Egypt 0.317 0.258 0.181 0.143 0.101 0.756Hong Kong, China 0.371 0.232 0.177 0.141 0.079 0.780Hungary 0.342 0.207 0.200 0.182 0.069 0.749India 0.366 0.207 0.181 0.142 0.105 0.754Indonesia 0.432 0.217 0.180 0.126 0.045 0.829Israel 0.284 0.223 0.183 0.159 0.150 0.690Rep. of Korea 0.333 0.256 0.166 0.134 0.112 0.754Malaysia 0.318 0.232 0.206 0.135 0.109 0.756Mexico 0.372 0.215 0.195 0.143 0.075 0.782Peru 0.409 0.202 0.170 0.129 0.091 0.781Philippines 0.367 0.241 0.165 0.114 0.113 0.773Poland 0.394 0.244 0.182 0.145 0.035 0.819Peoples Rep. of China 0.336 0.229 0.199 0.150 0.086 0.765Romania 0.323 0.258 0.191 0.133 0.096 0.771Russian Federation 0.397 0.257 0.177 0.094 0.074 0.832Singapore 0.302 0.219 0.210 0.148 0.121 0.731South Africa 0.332 0.206 0.199 0.144 0.119 0.737Taipei, China 0.301 0.245 0.209 0.155 0.091 0.755Thailand 0.347 0.208 0.182 0.156 0.107 0.737Turkey 0.448 0.213 0.174 0.108 0.058 0.835

Source: Authors calculations.signicant for all countries in the sample. The estimated coefcients offer strongsupport on the persistence of volatility of stock returns as the lagged variance issignicant and greater than the lagged squared residual.


mp

5)

.449

.05

FSI (excl country).727

0.10 0.07Emerging economies (+) 0.434*

f BaFSI (excl region)0.12

Regional FSI (excl (+) 0.420*Table 2Full sample panel estimates.

Dependen variable:emerging country FSI

Expectedsign

Variance-equal weights Principal co

(1) (2) (3) (4) (

Advanced economiesFSI

(+) 0.566* 0.329* 0.346* 0.545* 0

0.05 0.05 0.05 0.05 0Emerging economies (+) 0.856* 0

C.-Y. Park, R.V. Mercado Jr. / Journal ofor some countries with unavailable treasury yield data in the early1990s, policy rates were used.

2.2.2. Weighting schemeThe choice of a weighting scheme or how to combine the vari-

ous components of nancial stress into one index is perhaps thekey to constructing an FSI. The difculty arises from the lack of areference series upon which meaningful weights can be derivedand tested. Hence, various weighting techniques are considered.

The most common method is the use of variance-equal weights.With this approach, a nancial stress index is generated by givingequal importance to each component variables. The variables areassumed to be normally distributed and the series is demeanedand standardized. The mean is subtracted from each variable be-

country)0.12

LIBOR (3-month) (+) 0.279* 0.262* 0.265* 0.106** 0.1230.07 0.06 0.06 0.04 0.04

Global GDP growth () 0.235* 0.115** 0.098*** 0.119 0.0130.05 0.05 0.05 0.07 0.07

Global commodityprice change

() 0.033* 0.028* 0.029* 0.020* 0.0

0.00 0.00 0.00 0.00 0.00Financial openness

(t 1)(+) 0.002* 0.002* 0.002* 0.002* 0.002

0.00 0.00 0.00 0.00 0.00Trade openness (t 1) () 0.011 0.004 0.003 0.018* 0.0

0.01 0.01 0.01 0.01 0.01Current account

(t 1)() 0.035 0.019 0.015 0.052** 0.0

0.03 0.02 0.02 0.02 0.02Fiscal balance (t 1) () 0.065** 0.047* 0.047** 0.039** 0.0

0.02 0.02 0.02 0.02 0.02DForeign exchange

reserves (t 1)() 0.012 0.009 0.009 0.027 0.0

0.02 0.02 0.02 0.02 0.01Dummy emerging Asia (+)

Dummy emergingAmericas

(+)

Dummy emergingEurope

(+)

Constant 0.938 0.506 0.587 0.452 0.3250.73 0.62 0.63 0.44 0.42

R-squared (overall) 0.257 0.418 0.419 0.118 0.248Observations 1845 1845 1837 1840 1840Country 25 25 25 25 25Fixed effects Yes Yes Yes Yes Yes

Note: Specications (1) to (6) do not include regional dummy variables and are estimatedare estimated using random-effects. Specications (1) to (3) and (7) to (9) use FSI aggrecomponents. Specications (1), (4), (7), and (10) only include advanced economies FSI; (2FSI; (3), (6), (9), and (12) include emerging market FSI excluding the region and regional FSource: Authors estimate.* Signicant at 0.01.** Signicant at 0.05.*** Signicant at 0.10.

Please cite this article in press as: Park, C.-Y., Mercado Jr., R.V. Determinantshttp://dx.doi.org/10.1016/j.jbankn.2013.09.018onents Variance-equal weights Principal components

(6) (7) (8) (9) (10) (11) (12)

* 0.447* 0.574* 0.336* 0.353* 0.544* 0.445* 0.444*

0.05 0.04 0.05 0.05 0.05 0.05 0.05* 0.853* 0.764*

0.09 0.070.277* 0.438* 0.301*

0.06 0.12 0.070.407* 0.413* 0.417*

nking & Finance xxx (2013) xxxxxx 9fore it is divided by its standard deviation, hence, the term vari-ance-equal weights. Each component variable is computed as:

yt xt xr

5

where yt is the demeaned and standardized series, x the mean of theseries, and r is the standard deviation of the series. The demeanedand standardized components are then rebased from 0 to 100 (with100 having the historically highest value) and averaged as done byCardarelli et al. (2008, 2011), and Yiu et al. (2010) or simply added(without rebasing) as in Balakrishnan et al. (2009, 2011). Theadvantage of this approach is that it is easily implemented andapplicable for cross-country comparisons. On the other hand, thedisadvantage is that it assumes that the demeaned and standard-ized series follows a normal distribution.

0.07 0.11 0.07** 0.121** 0.284* 0.252* 0.253* 0.119** 0.125* 0.123**

0.05 0.06 0.06 0.06 0.04 0.04 0.040.013 0.236* 0.118** 0.101** 0.112 0.001 0.0020.06 0.05 0.05 0.05 0.07 0.06 0.06

20 0.020* 0.034* 0.028* 0.029* 0.022* 0.020* 0.020*

0.00 0.00 0.00 0.00 0.00 0.00 0.00* 0.002* 0.001 0.001 0.001 0.001** 0.001** 0.001**

0.00 0.00 0.00 0.00 0.00 0.00 0.0013** 0.013** 0.002 0.002 0.002 0.004** 0.005* 0.005*

0.00 0.00 0.00 0.00 0.00 0.00 0.0030 0.032 0.007 0.005 0.008 0.022 0.006 0.005

0.02 0.01 0.01 0.01 0.01 0.01 0.0136** 0.037** 0.047** 0.035** 0.035** 0.025*** 0.026*** 0.025***

0.02 0.02 0.02 0.02 0.01 0.01 0.0115 0.016 0.015 0.007 0.007 0.029*** 0.015 0.015

0.01 0.02 0.02 0.02 0.02 0.01 0.010.014 0.099 0.078 0.250** 0.164 0.181**0.18 0.16 0.16 0.11 0.10 0.090.156 0.067 0.090 0.078 0.074 0.066

0.21 0.18 0.18 0.10 0.09 0.090.237 0.224 0.109 0.221 0.335** 0.254***

0.22 0.20 0.22 0.14 0.13 0.150.309 1.406* 0.286 0.352 0.570** 0.208 0.1970.42 0.31 0.27 0.28 0.24 0.23 0.24

0.256 0.293 0.436 0.439 0.215 0.300 0.3071832 1845 1845 1837 1840 1840 183225 25 25 25 25 25 25Yes No No No No No No

using xed-effects. Specications (7) to (12) include regional dummy variables andgated using variance-equal weights; while (4) to (6) and (10) to (12) use principal), (5), (8), and (11) include other emerging market FSI excluding the specic countrySI excluding the country. Robust standard errors are used and are reported in italics.


Table 3Emerging Asia, xed effects panel estimates.

Dependent variable: emerging Asia country FSI Expected sign Variance-

(1)

Advanced economies FSI (+) 0.482*0.05

Emerging economies FSI (excl country) (+)

Emerging economies FSI (excl region) (+)

Regional FSI (excl country) (+)

**

**

2**

***

4

5

0

5

10 C.-Y. Park, R.V. Mercado Jr. / Journal of BaLIBOR (3-month) (+) 0.2790.10

Global GDP growth () 0.2790.10

Global commodity price change () 0.040.01

Financial openness (t 1) (+) 0.0020.00

Trade openness (t 1) () 0.000.01

Current account (t 1) () 0.030.04

Fiscal balance (t 1) () 0.040.03

DForeign exchange reserves (t 1) () 0.0070.02

Constant 1.321.30Another popular approach is the use of principal componentanalysis. The main idea behind using the principal componentanalysis is to represent each component of the nancial stress in-dex into a single variable by forming linear combinations of eachcomponent. Through this approach, the resulting stress index cap-tures the most common information from all components. Theresulting index is derived from the rst-three principal compo-nents, which refers to the coefcients of the linear combinationthat maximizes the variance of the resulting composite nancialstress index.6 Other weighting techniques used in the literature in-clude credit aggregate-based weights, and transformations of thevariables using their sample cumulative distribution functions.

In this paper, the ve components of the nancial stress index,given by:

R-squared (overall) 0.231Observations 804Country 10Fixed effects Yes

Note: Specications (1) to (3) use FSI aggregated using variance-equal weights; while (4economies FSI; (2) and (5) include other emerging market FSI excluding the specic countexcluding the country. All specications were estimated using xed-effects to account foSource: Authors estimate.* Signicant at 0.01.** Signicant at 0.05.*** Signicant at 0.10.

6 The rst three principal components are used to derive the individual country FSIas it captures around 7080% of information available from each component. 7080%Eigenvalue proportion is widely used in empirical literature utilizing principalcomponent analysis. Table 1 presents the proportion of eigenvalue components. Inthe earlier version of this paper, we used rst two principal components, as FSI usingthe rst two principal components could identify more crisis periods. However, therst two principal components captured only about 5060% of available information.As we use the rst three principal components, the number of identied crisis periodsdeclined somewhat, but the ndings from panel regression and structural vectorautoregression remain the same as those using rst two principal components. TheFSI for advance economies is constructed following the same method as that foremerging market FSI to have comparable components and results. Advancedeconomies include Australia, Austria, Canada, Denmark, France, Germany, Italy,Japan, the Netherlands, Norway, Spain, Sweden, Switzerland, the United Kingdom,and the United States.

Please cite this article in press as: Park, C.-Y., Mercado Jr., R.V. Determinantshttp://dx.doi.org/10.1016/j.jbankn.2013.09.018equal weights Principal components

(2) (3) (4) (5) (6)

0.254* 0.297* 0.476* 0.405* 0.415*0.07 0.05 0.07 0.06 0.071.024* 0.716*0.16 0.09

0.334*** 0.1370.15 0.080.636** 0.568*0.22 0.10

0.256** 0.263** 0.067 0.079 0.1050.09 0.10 0.06 0.06 0.070.076 0.083 0.079 0.002 0.0060.08 0.08 0.14 0.14 0.140.036* 0.034 0.023 0.024* 0.022**0.01 0.01 0.01 0.01 0.010.002*** 0.002** 0.002* 0.002* 0.002*0.00 0.00 0.00 0.00 0.000.000 0.001 0.011 0.008 0.0100.01 0.01 0.01 0.01 0.010.020 0.020 0.060 0.035 0.0330.04 0.04 0.03 0.03 0.030.031 0.034 0.022 0.022 0.0270.02 0.02 0.01 0.01 0.020.022 0.019 0.022 0.013 0.0160.01 0.01 0.02 0.02 0.020.952 0.814 0.654 0.367 0.4411.00 0.99 0.77 0.76 0.79

nking & Finance xxx (2013) xxxxxxEMFSI b Stockreturns Stockvolatility Debtspreads EMPI6

are aggregated to a composite nancial stress index using the vari-ance-equal weights and principal component analysis. FollowingBalakrishnan et al. (2009, 2011), all components are demeanedand standardized before adding for the variance-equal weights.For the principal component analysis, the rst three componentsare added and used as the emerging market FSI. In case of seriesbreaks, monthly FSIs were computed using the average of precedingand succeeding monthly values. Fig. 1 presents the computed FSI foreach country covered in this study using variance-equal weightsand principal component analysis; including those for selected ad-vanced economies. The gures illustrate that both methods ofaggregate FSI lead to comparable pattern of stressful and calm epi-sodes. However, it can be noticed that the variance-equal weightsmethod lead to more erratic or volatile pattern than the one usingprincipal component analysis as aggregating technique.

2.2.3. Identifying episodes of nancial stressSeveral approaches have been used in identifying episodes of

nancial stress based on the composite FSI. The simplest is througha graphical inspection of the composite index done by Yiu et al.(2010). Periods when the composite FSI peaks are consideredhighly stressful episodes, while troughs are relatively calm periods.Balakrishnan et al. (2009, 2011) and Cardarelli et al. (2008, 2011),used a more rigorous approach. They identied episodes of nan-cial stress when the composite FSI reaches 1.01.5 standard devi-ation above trend. Illing and Liu (2006) used an event study

0.376 0.392 0.157 0.261 0.266804 804 804 804 80410 10 10 10 10Yes Yes Yes Yes Yes

) to (6) use principal components. Specications (1) and (4) only include advancedry FSI; (3) and (6) include emerging market FSI excluding the region and regional FSIr country heterogeneity. Robust standard errors are used and are reported in italics.


Table 4Emerging Americas xed effects panel estimates.

Dependent variable: emerging Americas country FSI Expected sign Varianc

(1)

Advanced economies FSI (+) 0.651*0.11

Emerging Economies FSI (excl country) (+)

37**6280.024114*0.0371.0787.1246.1158

C.-Y. Park, R.V. Mercado Jr. / Journal of Banking & Finance xxx (2013) xxxxxx 11Emerging economies FSI (excl region) (+)


LIBOR (3-month) (+) 0.40.1


Global commodity price change () 00.0


Trade openness (t 1) () 00.0

Current account (t 1) () 00.0

Fiscal balance (t 1) () 00.0

DForeign exchange reserves (t 1) () 00.0approach for Canada, where stressful events were drawn from an-nual and monetary policy reports of the Bank of Canada.

This study identies periods of nancial stress when the nan-cial stress indices exceed its long-run trend by one point for thevariance-equal weights and 0.7 point for the principal componentanalysis.7 Given monthly structure of the dataset, stressful episodesalso include periods (months) in-between identied stressful peri-ods. For instance, FebruaryApril 1998 are also counted as stressfulperiods since they are in-between months of high stress levels(December 1997January 1998; and June 1998March 1999). Aggre-gate advanced and emerging (regional and nonregional) markets,and regional FSIs refer to unweighted average of individual countryFSI.8 To identify aggregate emerging market nancial stress periods,the difference between the unweighted regional FSI and its un-weighted regional trend are used. The identied stressful periodsare generally consistent with those of Balakrishnan et al. (2009) spe-cically for 1990s; and they capture the recent global nancial crisis.

Constant 1.8541.18



7 The trend was derived using the HodrickPrescott method where the smoothingparameter k is set to 1600. Since the rst three principal component account for morethan 70% of variation, 0.7 deviation from the long-trend was used as criteria toidentify stressful episodes for the FSIs computed using principal component analysis.Using one point rule as in the variance-equal weights method will fail to capturesignicant episodes of emerging market nancial stress.

8 Unweighted average is used so that individual country weight will not affect theaggregate nancial stress index. For example, if the average FSI for emerging Asia isweighted, then the impact of the Asian nancial crisis will be muted because of thehuge weight of Peoples Republic of China.

Please cite this article in press as: Park, C.-Y., Mercado Jr., R.V. Determinantshttp://dx.doi.org/10.1016/j.jbankn.2013.09.018e-equal weights Principal Components

(2) (3) (4) (5) (6)

0.433** 0.461** 0.494** 0.414*** 0.423***0.15 0.16 0.14 0.16 0.170.908** 0.901*0.19 0.14

0.629*** 0.576**0.30 0.210.256*** 0.278**0.11 0.07

0.441** 0.423** 0.188** 0.234** 0.210**0.14 0.13 0.07 0.06 0.050.222*** 0.202 0.013 0.105 0.1070.10 0.12 0.09 0.09 0.08

** 0.022** 0.022** 0.014** 0.017** 0.015**0.01 0.00 0.00 0.00 0.000.011** 0.011** 0.009** 0.008** 0.009***0.00 0.00 0.00 0.00 0.00

*** 0.027 0.035 0.030** 0.018 0.0320.02 0.03 0.01 0.01 0.020.055 0.050 0.091 0.001 0.0080.06 0.07 0.06 0.05 0.05

*** 0.047 0.027 0.072 0.052 0.0210.05 0.05 0.04 0.04 0.040.028 0.030 0.042 0.024 0.0190.08 0.07 0.03 0.04 0.03However, several points are noted on the identied stressful periods.First, since the FSI is an ex post measure of nancial instability, itwould be inappropriate to use it in an ex ante context such that itwould be used to assess whether nancial system is fragile or nottoday.9 Second, the identied periods are based on unweighted aver-age values and, hence, may not capture stressful periods experiencedby one or relatively few countries in the sample. Third, the identiedperiods are presented to illustrate the performance of the con-structed FSI. They are not used in estimation in this paper.

2.3. Patterns of advanced and emerging markets FSI

Based on the computed composite domestic FSI for both ad-vanced and emerging economies, several observations are notedon the general patterns of nancial stress. First, episodes of nan-cial stress in emerging markets closely track those in advancedeconomies (Figs. 2a and 2b). This pattern is clearly seen in the late1998 and 20082009 nancial crises, where the crisis in advancedcountries instigated or aggravated emerging market nancialstress. This observation is consistent with those from Balakrishnanet al. (2011). For the episodes in the early 1990s, nancial stresshas been more pronounced in advanced than emerging economiesas northern Europe and Japan confronted banking sector woes.Nonetheless, emerging market FSI showed a spike. For the late

0.762 0.484 0.529 0.592 0.0590.78 0.83 0.93 0.87 0.90



9 See Borio and Drehmann (2009) on the discussion on the issues pertaining tomeasures of nancial stability.


ce-

*

**

9**

*

2**

6

*

f BaTable 5Emerging Europe xed effects panel estimates.

Dependent variable: emerging Europe country FSI Expected sign Varian

(1)

Advanced economies FSI (+) 0.6430.11




LIBOR (3-month) (+) 0.1850.05








Constant 2.245

12 C.-Y. Park, R.V. Mercado Jr. / Journal o1998 episode, nancial crises in Latin America and the RussiaFederation compounded the nancial strain in the US due to thecollapse of Long Term Capital Management, causing both advancedand emerging market FSIs to increase. Finally, during the 20082009 global nancial meltdown, advanced economies FSI reacheda new high and followed by emerging market FSIs. Emerging mar-ket FSI also tends to be larger than advanced countries FSI partic-ularly in the 1990s. This may reect the fact that emergingmarket economies experienced more nancial crises and thereforegreater nancial stress in the 1990s.

Second, although FSIs computed using variance-equal weightsand principal components exhibit a similar pattern during bothcalm and stressful periods, there is a clear difference in their com-puted magnitudes. FSIs derived using variance-equal weights tendto have greater/lesser magnitudes than those computed usingprincipal components, implying that the variance-equal weightscan better capture episodes of severe nancial stress and near-miss events. In addition, there are also differences between thestressful period identied using the variance-equal weights andprincipal components. Variance-equal weight tends to capturemore stressful periods than principal components analysis.

Third, emerging market FSIs exhibit co-movement such thatindividual country nancial stress index increases during periodsof great nancial market turmoil in emerging economies (Figs. 3aand 3b). However, the peak of individual country FSIs can varyacross countries during episodes of nancial stress. For instance,only few emerging market countries experienced severe nancialstrain in 1995 and 2002 compared to 19971998 and 20082009,

0.98




(2) (3) (4) (5) (6)

0.352** 0.355* 0.696* 0.491** 0.459**0.06 0.07 0.09 0.10 0.110.666** 0.861*0.17 0.11

0.480** 0.543**0.14 0.160.233** 0.351**0.08 0.12

0.224** 0.231** 0.097 0.170 0.1670.05 0.05 0.08 0.09 0.090.127 0.125 0.270** 0.091 0.0530.13 0.12 0.08 0.07 0.06

* 0.019** 0.020** 0.017*** 0.020*** 0.019***0.01 0.01 0.01 0.01 0.010.007* 0.007* 0.003 0.004*** 0.0030.00 0.00 0.00 0.00 0.000.040** 0.036** 0.053** 0.044*** 0.0330.01 0.01 0.01 0.02 0.020.011 0.009 0.043 0.003 0.0030.02 0.03 0.04 0.04 0.050.028 0.029 0.031 0.036 0.0360.02 0.02 0.05 0.04 0.040.010 0.016 0.017 0.007 0.0080.03 0.03 0.04 0.03 0.031.677*** 1.435 3.200** 2.653 1.885

nking & Finance xxx (2013) xxxxxxwhere almost all emerging market FSIs rose. In addition, thereare episodes of individual market stress that are specic to a coun-try and not to emerging markets in general. For example, the spikein Brazils FSI in late 2005 backs emerging market trend.

Fourth, the constructed FSIs seem to capture moments of stressin emerging market nancial systems very well. Fig. 4af presentsthe component breakdown of the unweighted average of advancedand emerging economies FSIs, along with their regional groupings.For the emerging economies as a whole (Fig. 4a), it can be observedthat banking, equity, currency, and debt markets were all under se-vere strain during the 1997/1998 crises; while equity and currencymarkets were under strain during the 2008/2009 global nancialcrisis, reecting emergingmarkets healthy banking and scal posi-tions. For the advanced economies (Fig. 4b), debt markets played ahuge role during the stress episode in the early 1990s, while almostall components were under stress during the 2008/2009 crisis.Interestingly, banking sector stress has been elevated since the re-cent global nancial crisis. Across emerging market regions(Fig. 4cf), equity market stress dominates all nancial crisisepisodes. Nonetheless, there are differences in components perepisode across regions. Currency market stress spiked in late19971998 for emerging Asia; while debt market stress was moresevere in emerging America during the same period.

3. Determinants of domestic emerging market FSI

Understanding the determinants of domestic FSI is importantfor policymakers to ensure nancial stability. Given the increasing

0.78 0.75 1.18 1.28 1.61




nce-

**

**

45**

74*

77**

29

f BaTable 6Fixed effects panel estimates for 1Q19924Q1999.

Dependent variable: emerging Europe country FSI Expected sign Varia

(1)





LIBOR (3-month) (+) 0.9370.35







DForeign exchange reserves (t 1) () 0.0

C.-Y. Park, R.V. Mercado Jr. / Journal odegree of nancial integration, it is also crucial to know whetherthere is nancial contagion and, if so, to what extent a nancialshock originating elsewhere affects domestic nancial condition.In this section, we employ a panel regression analysis to assessthe effects of nancial stress from various sources such as global,country-specic, and other emerging market on domestic nancialstress, as well as the role of global and domestic factors in explain-ing domestic FSI. We also include a dummy variable for region toevaluate the effect of region-specic factors.

3.1. Data and methodology

The dataset includes quarterly data for 25 emerging marketeconomies. Data on individual emerging market FSI is taken fromthe previous section, while the advanced, emerging excludingcountry, emerging excluding region, and region excluding countryFSI are unweighted average of individual FSI in Section 2. Data forthe global GDP growth and scal balance (% of GDP) are taken fromthe Oxford Economics. Data for London interbank offered rate (LI-BOR), trade openness (exports plus imports as % of GDP), currentaccount (% of GDP), and foreign exchange reserves are sourcedfrom the International Financial Statistics and World Economic Out-look Database of the IMF and national sources accessed throughCEIC. Data for nancial openness is taken from the External Wealthof Nations Database and extended using the International Invest-ment Position Database of the IMF. Monthly data are converted

0.06Constant 3.905**

2.00




(2) (3) (4) (5) (6)

0.272** 0.281** 0.234*** 0.240*** 0.263***0.11 0.11 0.13 0.13 0.130.322 0.0240.24 0.15

0.034 0.1790.19 0.110.337*** 0.279**0.17 0.11

0.661** 0.658** 0.359 0.352 0.3690.27 0.27 0.29 0.28 0.300.546 0.556 0.536** 0.526*** 0.477***0.38 0.39 0.22 0.26 0.260.058* 0.059* 0.033*** 0.033** 0.032***0.02 0.02 0.02 0.01 0.020.005 0.006 0.001 0.001 0.0010.01 0.01 0.00 0.00 0.000.018 0.016 0.004 0.004 0.0070.03 0.03 0.01 0.01 0.010.038 0.031 0.028 0.028 0.0260.03 0.04 0.02 0.03 0.03

* 0.069 0.071 0.059*** 0.059*** 0.060***0.04 0.04 0.03 0.03 0.030.021 0.029 0.005 0.005 0.008

nking & Finance xxx (2013) xxxxxx 13to quarterly series beginning Q1 1992Q4 2012 using the averageof 3 months of a quarter. In cases where quarterly data is unavail-able, annual data is converted to quarterly series or the average an-nual value is used to ll the missing observations.

As mentioned in the previous section, FSIs computed usingprincipal component analysis have smaller values than those com-puted using variance-equal weights. Among the global indicators,the change in global commodity price has the highest mean andstandard deviation compared to global GDP growth and LIBOR.For openness indicators, de facto nancial integration has greatercross-country variation compared to trade openness. For thedomestic indicators, foreign exchange reserves has the higheststandard deviation compared to current account and scal balance,implying that cross-country differences in emerging market for-eign exchange reserve holdings is relatively high.

Panel unit root test of the Augmented Dickey Fuller (ADF)type was used to check for stationarity of all variables. The re-sults reject the null hypothesis that all variables in the panelcontain unit root, in favor of the alternative hypothesis of nounit root. However, for foreign exchange reserves, the nullhypothesis cannot be rejected, therefore the lagged value ofthe rst-differenced foreign exchange reserves was used inthe estimation. For the global indicators, a similar test wasconducted using the same procedure and specication for timeseries data. The results show that the indicators do not containunit root at 10% level of signicance.

0.06 0.06 0.05 0.05 0.05* 3.130 3.030 0.914 0.905 0.745

1.91 2.11 1.17 1.20 1.37




nce-equal weights Principal components

*

*

5*

1

1

1**

2**

***

f Banking & Finance xxx (2013) xxxxxxTable 7Fixed effects panel estimates for 1Q20004Q2007.

Dependent variable: emerging Europe country FSI Expected sign Varia

(1)





LIBOR (3-month) (+) 0.3540.06








Constant 1.3460.64

14 C.-Y. Park, R.V. Mercado Jr. / Journal o3.2. Panel least squares regression

The model specication is as follows:

EMFSIi;t ai b1AEFSIt RjbjEMFSIXi;t RjbjGlobalt RjbjDomestici;t1 RjbjDumi ei;t 7

where EMFSIi,t is the individual country FSI computed using vari-ance-equal weights and principal component analysis; AEFSIt isthe unweighted average of advanced economies nancial stress in-dex;

PjbjEMFSIXi,t refers to measures of other emerging market

nancial stress indexEMFSI excluding country, EMFSI excludingregion, and regional FSI excluding country, where each index iscomputed as residual using unweighted average of other emergingmarket FSIs as dependent variables; and advanced economies FSIand global indicators as the regressors.

PjbjGlobalt includes indica-

tors such as global interest rates, global output growth, and globalcommodity price increase.

PjbjDomestici.t1 refers to lagged value

of country-specic factors including nancial and trade openness,current account balance, scal balance, and change (rst difference)in foreign exchange reserves.

PjbjDumt refers to regional dummy

variables for emerging Asia, emerging Americas, and emergingEurope, where the value takes one (1) if the country belongs tothe region and zero (0) otherwise.

Eq. (7) is rst estimated without regional dummy variables(P

jbjDumt) using xed-effects ordinary least squares estimation.Based on the results of the Hausman test, country-specic effectsare adequately modeled by xed-effects regression. However, sincewe also want to test the signicance of a common regional factor,



Please cite this article in press as: Park, C.-Y., Mercado Jr., R.V. Determinantshttp://dx.doi.org/10.1016/j.jbankn.2013.09.018(2) (3) (4) (5) (6)

0.311* 0.323* 0.435* 0.389* 0.386*0.07 0.07 0.06 0.06 0.060.817* 0.752*0.11 0.12

0.440* 0.605*0.14 0.140.381* 0.1180.09 0.13

0.257 0.261* 0.188* 0.136** 0.140**0.05 0.05 0.05 0.05 0.050.249*** 0.257*** 0.433* 0.138 0.1510.14 0.14 0.13 0.14 0.150.022* 0.020* 0.003 0.015* 0.012**0.01 0.01 0.00 0.00 0.000.001 0.001 0.000 0.000 0.0000.00 0.00 0.00 0.00 0.000.001 0.002 0.011 0.005 0.0040.01 0.01 0.01 0.01 0.01

* 0.038 0.034 0.023 0.012 0.0130.02 0.02 0.02 0.02 0.020.043*** 0.044*** 0.058** 0.044** 0.045**0.02 0.02 0.02 0.02 0.020.041 0.042*** 0.024 0.040*** 0.041***0.02 0.02 0.02 0.02 0.020.141 0.233 1.124*** 0.252 0.1810.61 0.61 0.53 0.61 0.64Eq. (7) is estimated including three regional dummy variablesusing random-effects generalized least squares estimation. Usingxed-effects with regional dummy variables for Eq. (7) would beinappropriate since cross-country heterogeneity is already cap-tured by the regional dummy variables.10

To avoid possible endogeneity, lagged values of country-specicfactors were used in the estimation. Simple pairwise correlation ofresiduals reveals weak correlation between the estimated residualsand independent variables, implying that endogeneity is not a con-cern under the current specications. To address possible heter-oskedasticity, robust standard errors are used. To conductrobustness checks, Eq. (7) is estimated for each region by droppingcountries that are not member of the region. For example, Eq. (7) isestimated using xed-effects only for emerging Asia countries andthe same is conducted for the other regions. To test the evolution ofnancial contagion or transmission from advanced and otheremerging market (regional and nonregional) economies to an indi-vidual emerging market FSI over time, the sample period was splitinto three. The rst covers 1Q19924Q1999 or the 1990s; the sec-ond covers 1Q20004Q2007 or before the global nancial crisisperiod (pre-GFC); and the third includes 1Q20084Q2012 or afterthe global nancial crisis period (post-GFC).



10 The three regional dummy variables represent emerging Asia, emerging Americas,and emerging Europe, respectively. The intercept for the random-effects generalizedleast squares estimation for Eq. (7) corresponds to the common factor for otheremerging market economies including Egypt, Israel, South Africa, and Turkey.


3.3. Empirical results

Balakrishnan et al. (2011) argued that there are several factorsinuencing emerging market nancial stress. First, the nancialturmoil in advanced economies tends to increase the nancialstress in emerging economies. This reects the nancial contagionfrom advanced economies to emerging economies. Second, domes-tic nancial stress can be also exacerbated by common global fac-tors such as changes in GDP growth, commodity prices, andinterest rates. Third, country-specic factors such as the degreeof openness (nancial and trade) and macroeconomic vulnerabili-ties (current account, scal balance, and foreign exchange reserves)seem to affect the individual emerging market FSIs.

Previous literature notes that nancial transmission can becaused by some common factors, which would affect individualemerging market FSIs simultaneously. Such common factors caninclude global shocks and may manifest through investors herdingbehavior, cross-country contagion, and common credit conditions.Adding to the existing studies, this paper argues that there may beshocks that are region-specic, in line with the observed regionalnancial integration. These shocks may explain the rapid transmis-sion of a nancial shock in a particular region, such as regionalnancial contagion that was witnessed in emerging Asia duringthe nancial crisis of 19971998 and emerging Americas in the1990s and early 2000s.

Table 2 presents the panel estimates on the determinants ofemerging market FSI. The estimates show that specicationsincluding other emerging market FSI as a determinant of domestic

FSI have better t than those that include advanced economies FSIonly, as shown by their higher overall R-squared. For robustnesschecks, dataset for each region is estimated separately and periodcoverage is divided into three. Tables 35 present results foremerging Asia, emerging Americas, and emerging Europe, respec-tively. Similar to the estimates in Table 2, specications includingother emerging market FSI as a determinant of domestic FSI havebetter t than those that include advanced economies FSI only.Tables 68 show estimates for the subsample periods, namely1Q19924Q1999 (1990s); 1Q20004Q2007 (pre-GFC period); and1Q20084Q2012 (post-GFC period).

The estimates presented in Table 2 offer several key ndings.First, nancial stress from both advanced and emerging marketeconomies (excluding a country), signicantly increases domesticFSI for the (excluded) emerging market economy. This nding isconsistent with the earlier studies including Aizenman and Pasri-cha (2010) and Balakrishnan et al. (2011). The estimates in this pa-per also show that emerging market (excluding the particularregion) and regional FSIs (excluding the particular country) signif-icantly increases domestic FSI as seen in specications (3), (6), (9),and (12). These ndings support the view that nancial contagioncould originate from both advanced and other emergingeconomies.

Second, both global and domestic factors signicantly inuencedomestic FSI. Higher global interest rates tend to increase domesticnancial stress, suggesting tightening conditions in internationalcredit markets can have adverse effects on the domestic nancialcondition. Higher global GDP growth reduces domestic nancial

Table 8Fixed effects panel estimates for 1Q20084Q2012.

Dependent variable: emerging Europe country FSI Expected sign Variance-equal weights Principal components

(1) (2) (3) (4) (5) (6)

3*

3

0*

11*

2

4

86*

02

02

15*

8

le (4ountnt fo

C.-Y. Park, R.V. Mercado Jr. / Journal of Banking & Finance xxx (2013) xxxxxx 15Advanced economies FSI (+) 0.530.06




LIBOR (3-month) (+) 0.160.13








Constant 1.90.61


Note: Specications (1) to (3) use FSI aggregated using variance-equal weights; whieconomies FSI; (2) and (5) include other emerging market FSI excluding the specic cexcluding the country. All specications were estimated using xed-effects to accouSource: Authors estimate.

* Signicant at 0.01.** Signicant at 0.05.*** Signicant at 0.10.

Please cite this article in press as: Park, C.-Y., Mercado Jr., R.V. Determinantshttp://dx.doi.org/10.1016/j.jbankn.2013.09.0180.347* 0.352* 0.527* 0.447* 0.432*0.06 0.06 0.08 0.07 0.070.745* 0.629*0.20 0.18

0.602** 0.675*0.23 0.180.158 0.0240.13 0.16

0.273** 0.290** 0.067 0.117 0.1430.13 0.13 0.12 0.12 0.120.086 0.080 0.130** 0.013 0.0040.06 0.05 0.06 0.05 0.05

* 0.024* 0.024* 0.008** 0.016* 0.015*0.01 0.01 0.00 0.00 0.000.004* 0.004* 0.003 0.004*** 0.004***0.00 0.00 0.00 0.00 0.000.005 0.005 0.012** 0.006 0.0050.01 0.01 0.01 0.01 0.010.073* 0.072* 0.046** 0.037*** 0.036***0.02 0.02 0.02 0.02 0.020.008 0.008 0.006 0.009 0.0110.02 0.01 0.01 0.01 0.010.014 0.013 0.011 0.005 0.0010.02 0.02 0.02 0.02 0.011.018 1.045 2.338* 1.955** 1.855**0.67 0.65 0.66 0.70 0.68


) to (6) use principal components. Specications (1) and (4) only include advancedry FSI; (3) and (6) include emerging market FSI excluding the region and regional FSIr country heterogeneity. Robust standard errors are used and are reported in italics.of nancial stress in emerging market economies. J. Bank Finance (2013),

(b) Brazil(a) Argentina

(d) Peoples Rep.ofChina

(f) Czech Republic(e) Colombia

(h) Hong Kong,China

-2

-1

0

1

2

2 4 6 8 10 12

Response of DOM_SUM to Shock1

-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


Response to Structural One S.D. Innovations 2 S.E.

-2

-1

0

1

2

2 4 6 8 10 12

Response of DOM_PC to Shock1

-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



(c) Chile

(g) Egypt

Fig. 5. Impulse responses of domestic FSI to advanced economies, other emerging markets, and domestic nancial shock (response to structural one standard deviation).

16 C.-Y. Park, R.V. Mercado Jr. / Journal of Banking & Finance xxx (2013) xxxxxx

Please cite this article in press as: Park, C.-Y., Mercado Jr., R.V. Determinants of nancial stress in emerging market economies. J. Bank Finance (2013),http://dx.doi.org/10.1016/j.jbankn.2013.09.018

(j) India(i) Hungary

(l) Israel(k) Indonesia

(n) Malaysia(m) Rep. of Korea

(p) Peru (o) Mexico

-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12



-2

-1

0

1

2

2 4 6 8 10 12


-2

-1

0

1

2

2 4 6 8 10 12

Response of DOM_PC to

determinants of financial stress in emerging market economies

Documents