Correlación, Contagio, y evidencia asiática
Correlación, Contagio, y evidencia asiática

Mardi Dungey
Cambridge Endowment for
Research in Finance
Judge Business School
University of Cambridge
Trumpington Street
Cambridge, Reino Unido
CB2 1AG
M.dungey@cerf.cam.ac.uk
y
Centre for Applied
Macroeconomic Analysis
Australian National University

Renée Fry
Centre for Applied
Macroeconomic Analysis
Australian National University
Canberra, ACT, Australia
0200
renee.fry@anu.edu.au
y
Cambridge Endowment for
Research in Finance
University of Cambridge

Vance L. Martín
Departamento de Economía
Universidad de Melbourne
Parkville, VIC, Australia
3010
vance@unimelb.edu.au

Correlación, Contagio, y
Asian Evidence*

Abstracto
This paper examines the empirical literature on financial market
contagion in Asia during the 1997–98 financial crises with respect
to existing tests of contagion. Empirical evidence shows that con-
tagion affects both developed and emerging markets and does
not seem to vary with the relative fundamental economic health
or trade and financial linkages of the Asian economies. Contagio
occurs across both asset types and geographical borders and
tends to have larger effects in equity markets than in currency and
bond markets. There is evidence to support the hypothesis that
contagion is regional and transmitted through developed markets.
A discussion of the behavior of correlation coefficients in the
presence of contagion and financial crises suggests that they are
not a reliable metric for detecting contagion.

1. Introducción

Despite a relatively large amount of empirical literature
addressing the issue of the existence of contagion between
countries and markets during ªnancial crises, there is little
consensus on the results. Of course this creates difªculties
for policymakers and researchers in assessing whether or
not contagion exists and whether it is truly a problem
upon which policy should or could be focused. This paper
canvases related issues across the literature on contagion
in the context of the Asian ªnancial crisis in 1997–98. El
Asian crisis focused market participants, Responsables políticos,

* Paper prepared for the Asian Economic Panel hosted by the

Lowy Institute for International Policy and the ANU Centre for
Applied Macroeconomic Analysis, Sídney, 13–14 October 2005.
We would like to thank our discussant Chan Hyun Sohn and
participants in the Asian Economic Panel for helpful comments.
This project was funded under ARC Discovery grant
DP0343418.

Asian Economic Papers 5:2

© 2006 The Earth Institute at Columbia University and the Massachusetts

Institute of Technology

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Correlación, Contagio, y evidencia asiática

and researchers on the issue of contagion, even though the transmission of ªnancial
crises to other countries, como el 1987 stock market collapse (King and
Wadhwani 1990) and the 1994–95 Mexican peso crisis, which spawned the so-called
tequila effect (Sachs, Tornell, and Velasco 1996), occurred earlier.

The deªnition of contagion in the academic literature has been an evolving concept,
which has served as an obstacle to reaching a consensus on its relevance in explain-
ing crises. During the Asian ªnancial crisis, contagion was used commonly in the
popular press to describe the dramatic spread of the crisis, while academics had
many competing deªnitions. In the last few years, a terminology has developed to
distinguish the categories of contagion that are recognized in the existing frame-
works for conducting empirical analysis. The major distinction is between funda-
mentals-based contagion and pure contagion (see Dornbusch, Parque, and Claessens
2000; Kaminsky and Reinhart 2000). Fundamentals-based contagion refers to the
transmission of shocks between countries or markets routed through real links such
as trade, macroeconomic similarities, ªnancial links such as banking linkages and
capital ºows, or linkages that could have been anticipated ex ante the shock to asset
markets. Fundamentals-based contagion reºects that a crisis in one country will
spill over to other countries based on linkages as a result of normal levels of
ªnancial and economic integration.

Pure contagion refers to the transmission of shocks over and above that which oc-
curs through fundamental linkages. In periods of crisis, markets are suddenly more
integrated compared to tranquil times.1 These hypothesized additional linkages in-
clude competitive devaluations in which terms-of-trade effects hasten rapid devalu-
ations between economies but are mainly focused on the role of investor behavior.
The importance of investor behavior in transmitting contagion across countries is
explained in concepts such as wake-up calls, in which market participants focus
their attention on countries with fundamental indicators comparable to those in cri-
sis countries; portfolio adjustment or rebalancing, in which accounting and market
structures provide incentives for individual portfolio managers to make similar
moves at the same time, thus propagating crises between countries (Kodres and
Pritsker 2002); herd behavior, in which uncertainty and the possibility of informa-
tion asymmetries cause investors to follow the crowd (Calvo and Mendoza 2000);

1 In the extreme case of two markets being fully integrated in tranquil times, if one market ex-
periences a crisis, then the second market will also experience a crisis. In this case there is no
role for contagion, as it is expected that both markets will decline. Alternativamente, if there is
usually no integration between two markets in tranquil times, and a crisis occurs in one mar-
ket, it is not expected that the second country will experience a crisis. If this second country
also experiences a crisis, this suggests contagion.

33

Asian Economic Papers

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Correlación, Contagio, y evidencia asiática

and wealth effects, which arise from the cumulative effect of investors making losses
in one area rushing for liquidity from other areas of their portfolios (Kyle and Xiong
2001).2 Dungey, Fry, González-Hermosillo, and Martin (2005a) show that many of
the distinctions in the deªnitions of pure contagion in empirical work are nested.
During the Asian crisis, the term contagion was fairly broadly deªned and could
mean any or all of these things.

Earlier literature tends to use the term contagion to refer to both fundamentals-based
and pure contagion effects, such as Eichengreen, Rose, and Wyplosz (1995, 1996).
More recently, contagion refers principally to pure contagion effects, and fundamen-
tals-based contagion is considered separately and often labeled as spillovers, a term
due to Masson (1999). The focus on contagion in policymaking and empirical litera-
ture revolves around the importance of understanding linkages between ªnancial
markets during times of crisis and a recognition that such linkages may change in
unexpected ways during periods of turmoil. The optimal policy response to a crisis
should differ depending on the relative importance of fundamentals-based or pure
contagion, as the underlying causes are presumably quite different. This paper as-
sesses the evidence in the literature as to whether contagion is important in the ªrst
lugar, its characteristics, and whether or not a reaction to it should occur.

Most theoretical models seek to explain why markets appear to be more correlated
during a crisis period. The problem is that theoretical concepts explaining contagion
are difªcult if not impossible to measure empirically, so appealing to the data for
some understanding of the underlying reasons for pure contagion is quite challeng-
En g. Most empirical models of contagion instead look for contagion by examining
the evidence for strengthened linkages during a crisis period. Often this is associ-
ated with some form of test for increased correlation among asset returns, a pesar de
some authors point out that increasing correlation is not necessarily an indication of
contagion (Corsetti, Pericoli, and Sbracia 2001, 2005; Forbes and Rigobon 2002;
Bekaert, harvey, and Ng 2005; Dungey, Fry, González-Hermosillo, and Martin
2006b). A number of important overviews of empirical literature exist, incluido
Dornbusch, Parque, and Claessens (2000) and Pericoli and Sbracia (2003), with Bekaert
and Harvey (2003) providing a particular focus on emerging markets.

Empirical literature, in some instances supported by theory, has raised a number of
important propositions about contagion during ªnancial crises. This is augmented
by agendas arising from international bodies, como el 1998 Bank for Interna-
tional Settlements (BIS) survey of market participants following the Russian and

2 For overviews of these concepts, see Goldstein (1998), Lowell, nuevo, and Tong (1998), y

Dungey and Tambakis (2005).

34

Asian Economic Papers

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Correlación, Contagio, y evidencia asiática

long-term capital management (LTCM) crises. These propositions provide a frame-
work for understanding potential policy responses to contagion. Seven such propo-
sitions are synthesized below.

1. Strong fundamentals imply immunity to contagion.
2. Trade and ªnancial linkages between countries are associated with contagion

transmission.

3. Regional proximity is important in transmitting/receiving contagion effects.
4. Emerging markets experience greater contagion than developed markets.
5. Developed markets operate as a conduit for contagion effects between regions.
6. Contagion effects differ by asset market.
7. Contagion occurs across both asset market and country borders.

The rest of this paper proceeds as follows. Sección 2 presents data relating to the
Asian ªnancial crisis along with a brief background of events. Sección 3 examines is-
sues arising from using changes in correlation as the measure to distinguish crises
and contagion in a formal model. The paper then canvases evidence for contagion
during the Asian ªnancial crisis from the empirical literature across different asset
classes in section 4. Sección 5 turns to the seven propositions previously listed to ex-
amine evidence as to whether pure contagion is a relatively important inºuence on
asset returns, or alternatively, whether domestic or world/regional inºuences, en-
cluding fundamentals-based contagion, are of greater importance. Sección 6 pro-
vides some concluding comments.

2. The crisis in Asia

The events of the 1997–98 Asian crisis have been examined in great detail. Some of
the key events of this crisis include the devaluation of the Thai baht in July 1997; el
Hong Kong speculative attack of October 1997; the devaluation and ºoat of other
Asian currencies during the period; the coordinated rollover of Korea’s short-term
debt by commercial banks to avoid a debt moratorium in December 1997; the clos-
ing or restructuring of ªnancial institutions in most crisis countries including Indo-
nesia, Korea, Malasia, y Tailandia; the imposition of capital controls in Malaysia
en 1998; the change of political leadership in all countries excluding Malaysia (Alabama-
though there the deputy prime minister, who also served as the ªnance minister,
was ªred); and the dramatic reduction of sovereign ratings for all crisis countries.
Indonesia, Tailandia, and Korea all sought IMF assistance packages, and many in-
dustrial countries pledged ªnancial assistance. For chronologies of the crisis, ver
Bank for International Settlements (1998), IMF (1999), Kaminsky and Schmukler
(1999), Athukorala (2001), and Baur and Fry (2006).

35

Asian Economic Papers

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Correlación, Contagio, y evidencia asiática

Figures 1 y 2 show daily currency and equity returns for a selection of economies
most affected by the crisis over the period. The currency returns are presented from
2 Junio 1997 a 30 Junio 1998. Many of the Asian currencies were pegged prior to this
fecha. A longer period is covered in the ªgures on equity returns, which are shown
de 2 Enero 1996 a 31 Julio 1998. The ªgures highlight the speed with which the
Asian crisis spread across different asset classes in the Asian economies.

The implications of the lack of consensus on the transmission process in the Asian
crisis are evident in the response of national and international policymaking bodies.
There were numerous calls for the reform of international ªnancial infrastructure to
halt the potential transmission of crises via contagion; Por ejemplo, Claessens and
Forbes (2001), Goldstein (1998), Eichengreen (1999), and Vines and Gilbert (2004).
Many of the original proposals focused on reforms to national economic systems
through transparency, domestic regulation, improved national economic manage-
mento, and development of domestic ªnancial markets. This agenda has not been
universally popular with the crisis countries, as it puts the burden of adjustment on
those countries that have already borne the greatest cost. Por ejemplo, Radelet and
Sachs (1998) and Germain (2002) argue that the crucial change needed post-crisis is
to reshape international ªnance as inclusive of emerging market economies, cual
is a political agenda rather than a purely economic one.

3. Preliminary tests of contagion using correlations

The preliminary work on testing for contagion involved testing for a change in the
correlation of asset markets between crisis and noncrisis periods. Early work was by
King and Wadhwani (1990) and Baig and Goldfajn (1999). Forbes and Rigobon
(2002) are associated with a heteroskedasticity-adjusted correlation test that fa-
mously ªnds little evidence of contagion during a number of ªnancial crises, incluir-
ing from Hong Kong equity markets in 1998.

Cifra 3 presents selected rolling correlations between daily equity returns for Hong
kong, Indonesia, Korea, Malasia, and Thailand surrounding four crisis events. El
window width is one month. The four rows of the ªgure correspond to the four
events selected:

1. the Thai baht devaluation of July 1997, with the panel covering 2 June to 31 Julio

1997;

2. the Hong Kong speculative attack from 1 Octubre 1997 a 14 Noviembre 1997;
3. the period surrounding the successful avoidance of the Korean debt moratorium

en 24 December 1997, with the ªgure covering the month of December;

4. the turmoil in Indonesia from 1 January to 27 Febrero 1998.

36

Asian Economic Papers

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

8
9
9
1

mi
norte
tu
j
0
3
oh
t
7
9
9
1
mi
norte
tu
j

2
d
oh
i
r
mi
pag
mi
h
t

r
oh
F
d
norte
a
yo
i
a
h
t
d
norte
a

,
a
i
s
y
a
yo
a
METRO

,
a
mi
r
oh
k

,
a
i
s
mi
norte
oh
d
norte
I

r
oh
F

s
r
a
yo
yo
oh
d

.

.

S
Ud.

Correlación, Contagio, y evidencia asiática

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

t
s
norte
i
a
gramo
a

s
norte
r
tu
t
mi
r
y
C
norte
mi
r
r
tu
C
–
yo
a
C
oh
l

.
1

mi
r
tu
gramo
i
F

37

Asian Economic Papers

Correlación, Contagio, y evidencia asiática

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

8
9
9
1
y
yo
tu
j

1
3
oh
t

6
9
9
1
y
r
a
tu
norte
a
j
2
d
oh
i
r
mi
pag
mi
h
t

r
oh
F
d
norte
a
yo
i
a
h
t
d
norte
a

,
a
mi
r
oh
k

,
a
i
s
mi
norte
oh
d
norte
I

,

gramo
norte
oh
k
gramo
norte
oh
h

r
oh
F

a
t
a
d
s
norte
r
tu
t
mi
r
y
t
i
tu
q
mi

.
2

mi
r
tu
gramo
i
F

38

Asian Economic Papers

Correlación, Contagio, y evidencia asiática

Cifra 3. Rolling correlations of daily equity returns calculated over a monthly window
around four key events

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

Nota: First column based on actual data; second column based on the residuals from a VAR containing all equity returns with one lag.

Panel (a) correlation with Thailand, 2 June–31 July 1997; (b) correlation with Hong Kong, 1 October–14 November 1997; (C) correlation

with Korea, 1–31 December 1997; (d) correlation with Indonesia, 1 January–27 February 1998.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

39

Asian Economic Papers

Correlación, Contagio, y evidencia asiática

The ªrst column in the ªgure presents correlation coefªcients calculated on equity
returns, and the second presents correlation coefªcients calculated on the residuals
of a vector autoregression (VAR) model containing all returns with one lag. The cor-
relations of the residuals of the VAR are reported, as tests for contagion such as that
proposed by Forbes and Rigobon (2002) attempt to capture fundamental relation-
ships through a VAR, so contagion is identiªed through the modeling of outliers.

From inspection of Figure 3, it is immediately obvious that the correlations are not
stable, either before or after fundamentals are controlled for. More pertinently, es
not clear that correlations necessarily increase after a shock: the Thai baht devalua-
tion is an example in which correlations increase, whereas the Hong Kong shock is
an example in which they decrease.

3.1 A model of asset returns
To better assess what the correlation analysis presented above might represent in
identifying contagion, a ªnancial market model with its origins in the factor models
of arbitrage pricing theory is speciªed (see Sharpe 1964; Solnik 1974). Two periods
are distinguished in the model. The ªrst is a factor model in which assets are priced
based on “normal” shocks during a noncrisis, or tranquil, período. The second is a
factor model which extends the tranquil-period factor model by allowing for addi-
tional linkages arising from contagion and various types of structural breaks that
potentially arise during ªnancial crises.

A model for tranquil times The model in a noncrisis period in the case of three
assets consists of a one-factor model in which returns (xi,t) are speciªed as a function
of a common factor (peso) and an idiosyncratic component (ui,t):

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

where for simplicity

xi,t

(cid:2) (cid:3)

iwt

(cid:4) (cid:5)

iui,t,

i (cid:2) 1, 2, 3,

peso

(cid:6) norte(0,1),

ui,t

(cid:6) norte(0,1),

i (cid:2) 1, 2, 3,

(1)

(2)

(3)

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

are assumed to be independent. Por simplicidad, all returns are assumed to be de-
meaned, in which case there is no need for a constant term in (1). The common fac-
tor captures market fundamentals or systematic risk that has an impact on asset re-
turns with a loading of (cid:3)
i. This factor is also interpreted as a measure of integration.

40

Asian Economic Papers

Correlación, Contagio, y evidencia asiática

(cid:2) (cid:3)

The idiosyncratic components capture those periods in which returns deviate from
their market fundamentals and have an impact on asset returns with a loading of (cid:5)
i.
(cid:2) 0, the markets are segmented with vola-
In the special case in which (cid:3)
3
tility in asset returns, which are driven entirely by their respective idiosyncratic
componentes. The assumption that the world and idiosyncratics are distributed as
norte(0,1) can be relaxed by including autocorrelation and conditional volatility in the
form of generalized autoregressive conditional heteroskedasticity (GARCH), as in
Dungey, Fry, and Martin (2003).

(cid:2) (cid:3)

1

2

Estimation issues The assumption that the common factor (peso) and the idiosyn-
cratic factors (ui,t) are independent implies that the volatility of returns as measured
by the variance is obtained by squaring both sides of (1) and taking expectations:

mi [

2 (cid:2) li
]xi,t

2 (cid:4) fi
2,

i (cid:2) 1, 2, 3.

(4)

This expression shows that the variance of returns can be conveniently decomposed
into two components: the contribution from the common factor (li
bution from the idiosyncratic factor (fi
is that it makes it possible to provide an estimate of the average relative contribu-
tions of the two factors to the volatility of returns during tranquil periods.

2) and the contri-
2). An important advantage of this expression

Similarmente, the covariance between returns during periods of tranquility is obtained
de (1) by multiplying xi,t by xj,t and taking expectations:

mi [xi,txj,t] (cid:2) (cid:3)

(cid:3)
i

j, (cid:2)i (cid:7) j.

(5)

During tranquil periods, co-movements in returns are solely determined by the
common factors, with no role played by the idiosyncratic factors. For asset returns
that are affected by the common factor in the same direction, eso es, for which the
loading parameters have the same sign, the returns on these assets move in the
same direction on average. For asset returns that are affected by the common factor
in opposite directions, eso es, for which the loading parameters are not the same
sign, the returns move in opposite directions on average.

The expressions for the variance in (4) and covariance in (5) suggest that a natural
way to estimate the six parameters of the tranquil model,

{(cid:3)

1, (cid:3)

2, (cid:3)

3, (cid:5)

1, (cid:5)

2, (cid:5)

3},

(6)

41

Asian Economic Papers

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Correlación, Contagio, y evidencia asiática

is to equate the theoretical moments in (4) y (5) with the respective empirical mo-
mentos. For a sample of size T, the pertinent set of equations that needs to be solved
es

(cid:2) (cid:2)yo 1

2 (cid:4) (cid:2)F 1
2 ,

s11

(cid:2) (cid:2)yo 2

2 (cid:4) (cid:2)f2
2,

s22

(cid:2) (cid:2)l3

2 (cid:4) (cid:2)f3
2,

s33

(cid:2) (cid:2)

(cid:2)l l1

2,

s12

(cid:2) (cid:2)

(cid:2)l l1

3,

s13

(cid:2) (cid:2) (cid:2)l l2

3,

s23

where a ^ denotes an estimator and

sij

(cid:2) 1

t

(cid:4)
1t
(cid:3)
t

x xi t
,

j t
,

(7)

(8)

represents the sample covariance (empirical moment) between assets i and j. Esto es
a system of six equations and six unknowns that can be solved to provide estimates
of the parameters. Formalmente, estimation proceeds using a generalized method of mo-
mentos (GMM) estimator. As the system is in this case just identiªed, as the number
of unknown parameters matches the number of equations, it is possible to derive
the following analytical expressions for the estimators:

2

(cid:2)yo 1

2

(cid:2)F 1

,

(cid:3) s s
12 13
s
23
s s
11 23

(cid:3)

(cid:5)

s s
12 13

s
23

,

(cid:3) s s
12 23
s
13
s s
22 13

(cid:3)

2

(cid:2)yo 2

2

(cid:2)f2

(cid:5)

s s
12 23

s

13

,

2

(cid:2)l3

2

(cid:2)f3

,

(cid:3) s s
13 23
s
12
s s
33 12

(cid:3)

(cid:5)

s s
13 23

s

12

.

(9)

A model for crisis times Crisis period relationships can be accounted for by ex-
tending the noncrisis model in (1) a (3) by allowing for structural breaks in the
world factor and idiosyncratic factors, as well as for increases in asset return volatil-
ity from an additional propagation mechanism from one country to another arising
from contagion. To distinguish the crisis period from the noncrisis period, returns
for the former period are denoted as yi,t. As before, these returns are also assumed to
be demeaned. The factor structure during the crisis period for a trivariate system is
speciªed as

y1,t

(cid:2) (cid:3)

1peso

(cid:4) (cid:5)

1u1,t,

y2,t

(cid:2) (cid:3)

2peso

(cid:4) (cid:5)

2u2,t

(cid:4) (cid:8)

y3,t

(cid:2) (cid:3)

3peso

(cid:4) (cid:5)

3u3,t

(cid:4) (cid:8)

(cid:5)

1u1,t,

(cid:5)

1u1,t,

2

3

(10)

(11)

(12)

42

Asian Economic Papers

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Correlación, Contagio, y evidencia asiática

dónde

peso

(cid:6) norte(0,(cid:9)2),

t

ui,t

(cid:6) norte(0,ki

2),

i (cid:2) 1, 2, 3.

(13)

(14)

Structural breaks
If the common factor is interpreted as representing market fun-
damentals, a change in the pricing of assets during the crisis period as a result of a
change in wt is captured by the parameter (cid:9)2 en (13). Para (cid:9)2 (cid:10) 1, this leads to an in-
crease in volatility of all asset returns during the crisis period. An alternative form
of a structural break, which increases asset market volatility, occurs when there are
2 (cid:10) 1. It is more
changes in the idiosyncratic components in the crisis period, with ki
typical to model the structural break in the country directly experiencing the crisis,
otherwise known as the source country. Sin embargo, for the crisis to spread from the
source country to the other countries in this situation, it is necessary for contagion to
existir.

Contagion Contagion is deªned as shocks originating in the source country over
and above the inºuence of the common factor (peso) which impact the asset returns of
the remaining countries. In the case of the crisis model in (10) a (14), country 1 rep-
resents the source country, with the strength of contagion to countries 2 y 3 estafa-
trolled by the parameters (cid:8)
3, respectivamente.

2 y (cid:8)

Estimation issues Estimation of the parameters of the crisis model proceeds along
the lines of that for the tranquil model by adopting a GMM approach whereby the
theoretical moments are matched with the empirical moments. As the parameters in
(6) are uniquely identiªed by the tranquil moments, with the solutions given in (9),
the remaining parameters in the crisis model,

(15)

2 ,
1

(16)

2 (cid:4) (cid:2) (cid:2) (cid:2)
d f k3

3

1

{(cid:8)

2, (cid:8)

3, (cid:9), (cid:11)

1, (cid:11)

2, (cid:11)

3},

are estimated from the following set of equations:

(cid:12)

(cid:12)

11

12

(cid:2) (cid:2)

(cid:2) (cid:2)

2

2 (cid:6)(cid:7) (cid:4) (cid:2) (cid:2)f k1
(cid:2)yo 1
(cid:2) (cid:2)

l l1

2

(cid:6)(cid:7) (cid:4) (cid:2) (cid:2) (cid:2)
d f k2

2
1

2 ,
1

(cid:12)
22

2 , (cid:12)

1

13

2 (cid:4) (cid:2) (cid:2) (cid:2)
d f k2

2

1

2

(cid:2) (cid:2) (cid:2)l2
(cid:2) (cid:2)

2 (cid:6)(cid:7) (cid:4) (cid:2) (cid:2)f k2
(cid:2) (cid:2)

l l1

3

(cid:6)(cid:7) (cid:4) (cid:2) (cid:2) (cid:2)
d f k3

2
1

2 ,
1

2 ,
1

(cid:12)
33

(cid:12)
23

(cid:2) (cid:2) (cid:2)l3
(cid:2) (cid:2) (cid:2) (cid:2)
l l2

3

2 (cid:6)(cid:7) (cid:4) (cid:2) (cid:2)f k3

2

(cid:6)(cid:7) (cid:4) (cid:2) (cid:2) (cid:2) (cid:2)
d d f k
2 3

2
1

2 ,
1

dónde, como antes, a ^ denotes an estimator, y

(cid:12)

ij

(cid:2) 1

t

(cid:4)
1t
(cid:3)
t

y yi t
,

j t
,

(17)

43

Asian Economic Papers

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Correlación, Contagio, y evidencia asiática

represents the sample covariance (empirical moment) between assets i and j during
the crisis period. This is a just-identiªed system, as there are six equations and six
unknowns.

Changes in correlations The factor model given above has the advantage that it
provides convenient expressions for the correlations between asset returns. Usando
ecuaciones (1) a (3), the correlation between asset 1 and asset 2 during a noncrisis
period is

r
X

X

1

,

t

2

,

t

(cid:3)

E x x
1

[

t

,

]

2

,

t

(cid:3)

2
E x E x
] [
t
,
1

[

2
2

,

t

]

l l
2
1
2
f l
1

.

2(cid:8) F

2

2
2

yo

(cid:8)

2
1

From equations (10) a (14) the corresponding correlation in the crisis period is

r
y

y

1

,

t

2

,

t

(cid:3)

E y y
1

[

t

,

]

2

,

t

(cid:3)

E y
[

2
1

,

t

E y
] [

2
2

,

t

]

2

l l
1
2
f k
1

2
1

(cid:7)
(cid:6)

(cid:8)

yo

2
1

(cid:8)(cid:6)(cid:7)

2
2
d f
k
1
1
2
(cid:7)
(cid:8)
(cid:6)
2
f k
2

2
yo
2

.

(cid:8)

2
2

2
2
d f k
1
2

2
1

(18)

(19)

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

As a number of theoretical and empirical deªnitions of contagion involve the obser-
vation of an increase in correlation coefªcients between the noncrisis and crisis peri-
probabilidades, it is natural to look at the change in correlation by subtracting (18) de (19):

r
y

y

1

,

t

2

,

t

(cid:5)

r

X

X

1

,

t

2

,

t

(cid:3)

2

l l
1
2
f k
1

2
1

(cid:6)
2

(cid:8)

yo

2
1

(cid:6)
2

(cid:8)

2
2
d f k
1
2
1
(cid:8)
2
f k
2

(cid:8)

2
2

2
2
d f k
2
1

2
1

(cid:5)

l l
1
2
2
2
f l
1
2

.

(cid:8)

2
F
2

yo

(cid:8)

2
1

(20)

(cid:6)2

2
yo
2

There are several parameters that govern the difference in correlations across the
2, (cid:9)2, and ki
2, which are now explored. This analysis is
two subperiods. These are di
closely related to that of Corsetti, Pericoli, and Sbracia (2001, 2005).

3.2 Understanding changes in correlations
To understand the underlying relationships governing changes in correlations,
equation (20) is computed for various parameterizations. This analysis is based on a
simpliªed version of the model in Dungey, Fry, González-Hermosillo, and Martin
(2006b). The parameterized system is

y1,t

(cid:2) 4peso

(cid:4) 2u1,t,

y2,t

(cid:2) 2peso

(cid:4) 3u2,t

(cid:4) (cid:8)

22u1,t,

y2,t

(cid:2) 3peso

(cid:4) 4u2,t

(cid:4) (cid:8)

32u1,t,

(21)

(22)

(23)

44

Asian Economic Papers

Correlación, Contagio, y evidencia asiática

2 (cid:2) 1, (cid:2) i. In the noncrisis period, the data-generating process for xi,t

and in this case there are no structural breaks in the common or idiosyncratic fac-
tores, entonces (cid:9)2 (cid:2) ki
(cid:2) 0 and redeªnes yi,t as xi,t.
is the same as equations (21) a (23) if one sets (cid:8)
2
Cifra 4 presents the results of the experiments for the difference in correlations,
with all parameters held constant except for one. Panels (a) y (b) explore the be-
havior of the difference in the correlation coefªcients between assets 1 y 2 and as-
(cid:2) (cid:8) aumenta. Panels (C) y
conjuntos 2 y 3, respectivamente, as the strength of contagion (cid:8)
(d) consider the case in which there is no contagion, but a structural break in the
common factor ((cid:9)2).

(cid:2) (cid:8)

3

i

Changes in correlation due to contagion
Asset 1 to asset 3 Panel (a) of Figure 4 shows the change in correlation between y1,t
and y3,t for increasing values of contagion from (cid:8) (cid:2) 0 a (cid:8) (cid:2) 20, based on equation
(20) con (cid:9)2 (cid:2) (cid:11)2 (cid:2) 1. The correlation initially rises but hits a peak and eventually
falls and in the limit becomes negative. The initial increase in correlation is consis-
tent with much of the empirical literature, wherein an increase in correlation indi-
cates evidence of contagion. Sin embargo, the fall in correlation for strong levels of con-
tagion, eso es, higher values of (cid:8), contradicts the premise that increasing correlation
is associated only with contagion. This result casts doubt on empirical evidence that
ªnds no contagion based on nonincreasing correlation.3

Asset 2 to asset 3 Panel (b) of Figure 4 repeats the analysis for y2,t and y3,t. Este
panel appears to provide support for the hypothesis that higher correlation is associ-
ated with higher levels of contagion. Sin embargo, this result is purely spurious, como el
increase in correlation is fully generated by a common component, a saber, u1,t. Por
construction, there are no contagious linkages between assets 2 y 3.

Changes in correlation due to systemic structural breaks The effects of increas-
ing systemic shocks ((cid:9)2) on the change in correlations between y1,t and y3,t and be-
tween y2,t and y3,t is given in panel (C) of Figure 4. En este caso, there is no contagion
((cid:8) (cid:2) 0) and no idiosyncratic structural break ((cid:11)2 (cid:2) 1). Despite the fact that the true
process does not involve contagion, there is still an increase in correlation between
the two periods. In the limit, the correlation in the crisis period approaches unity,

3 Dungey, Fry, González-Hermosillo, and Martin (2006b) muestra esa (20) yields a fall in correla-

tion for high levels of contagion if

(cid:8)

1

(cid:9)
(cid:11)(cid:11)
(cid:10)

F
yo

2

2

2

(cid:12)
(cid:14)(cid:14) (cid:15)
(cid:13)

yo
F

1

1

.

Corsetti, Pericoli, and Sbracia (2001, 2005) also discuss the problem of measuring contagion
via correlation coefªcients from this angle in some depth.

45

Asian Economic Papers

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Correlación, Contagio, y evidencia asiática

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

s
norte
oh
i
t
pag
metro
tu
s
s
a
k
a
mi
r
b
yo
a
r
tu
t
C
tu
r
t
s
d
norte
a
norte
oh
i
gramo
a
t
norte
oh
C

s
tu
oh
i
r
a
v
r
mi
d
norte
tu
s
t
norte
mi
i
C
ª
F
mi
oh
C
norte
oh
i
t
a
yo
mi
r
r
oh
C
norte
i

mi
gramo
norte
a
h
C

.
4

mi
r
tu
gramo
i
F

46

Asian Economic Papers

Correlación, Contagio, y evidencia asiática

reºecting the total domination of the asset returns by the common factor (peso). Este
result further highlights the danger of misdiagnosing increasing correlation as evi-
dence of contagion. Most applications for tests of contagion speciªcally assume that
the appropriate process describing the data is the presence of contagion and not a
structural change in the common factor: Por ejemplo, Forbes and Rigobon (2002)
and Dungey and Martin (2004). Forbes and Rigobon (2001) speciªcally designate the
situation of a structural change in the common factor as shift contagion.

Relationship with regression analysis The occurrence of a structural break in the
common factor brings out an important relationship between correlation and regres-
sion analysis in testing for contagion. To highlight this relationship, consider a spe-
(cid:2) 0, so country 1
cial case of the noncrisis and crisis models in (1) a (14) en el cual (cid:5)
becomes the common factor. De (1) y (10) the asset returns can be expressed in
terms of the ªrst asset return. Por ejemplo, the second asset return equations for the
noncrisis and crisis periods are, respectivamente,

1

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

x2,t

y2,t

(cid:2) yo
yo
(cid:2) yo
yo

2

1

2

1

x1,t

(cid:4) (cid:5)

2u2,t

(cid:2) (cid:13)x1,t

(cid:4) (cid:5)

2u2,t,

(24)

y1,t

(cid:4) (cid:5)

2u2,t

(cid:2) (cid:13)y1,t

(cid:4) (cid:5)

2u2,t,

2 /(cid:3)

1, where it is still assumed that there is no contagion ((cid:8)

con (cid:13) (cid:2) (cid:3)
(cid:2) (cid:8) (cid:2) 0) y
2 (cid:2) 0). The strength of the trans-
that there are no idiosyncratic structural breaks (ki
mission mechanism between countries 1 y 2 is given by the parameter (cid:13). As this
parameter is the same over both noncrisis and crisis periods, it can be estimated by
simply regressing the returns of country 2 on the returns of country 1 using the full
sample of data, despite the structural break in the common factor.

i

The fact that the parameter (cid:13) en (24) is constant over the two sample periods does
(cid:2) 0,
not imply that the correlations are also constant. To see this, en (20), colocar (cid:5)
1
(cid:2) 1, so that the change in the correlation between the crisis and noncrisis pe-
(cid:11)

(cid:2) (cid:11)

(cid:2) (cid:8)

2

1

2

riods reduces to

r
y

y

1

,t

2

,

t

(cid:5)

r

X

X

1

,t

2

,

t

(cid:3)

1

(cid:5)

(cid:8)

1

2

(cid:9)
(cid:11)(cid:11)
(cid:10)

F
2
(cid:6)
yo

2

(cid:12)
(cid:14)(cid:14)
(cid:13)

.

2

1

(cid:9)
(cid:11)(cid:11)
(cid:10)

(cid:12)
(cid:14)(cid:14)
(cid:13)

F
2
yo

2

(cid:8)

1

This expression is positive for (cid:9)2 (cid:10) 1, whereby the structural break in the common
factor results in an increase in correlation even though the regression parameter in
(24) is constant. This point was ªrst made by Loretan and English (2000) and Forbes

47

Asian Economic Papers

Correlación, Contagio, y evidencia asiática

and Rigobon (2002). This result suggests that the stability of the relationship be-
tween countries 1 y 2 can be assessed by performing a test of parameter con-
stancy. If the null hypothesis of constant parameters is rejected, this will be evidence
of an additional transmission mechanism during the crisis period and hence will
constitute contagion. This forms the basis of the regression-based test of contagion
suggested by Dungey, Fry, González-Hermosillo, and Martin (2005a).

4. A review of the empirical evidence for Asian ªnancial markets

Although there are limitations in many of the contagion tests in the suite, as demon-
strated by the correlation coefªcient example above, some generalities can be ex-
tracted from the evidence presented in the empirical literature. Apart from tech-
niques based on correlation analysis, there are several other methodologies by
which researchers test empirically for contagion; these methodologies are reviewed
in Dungey, Fry, González-Hermosillo, and Martin (2005a). The main empirical tests
include those based on modeling outliers from a VAR system, such as that in Favero
and Giavazzi (2002); dummy variable-based tests such as that in Pesaran and Pick
(2006); and the probability-based measure of Bae, Karolyi, and Stulz (2003), cual es
related to the previous work of Eichengreen, Rose, and Wyplosz (1995, 1996). Tests
based on latent-factor models, such as those examined by the current authors,
Corsetti, Pericoli, and Sbracia (2001), and Bekaert, harvey, and Ng (2005), are also
commonly used. Aquí, evidence is organized by a particular asset market during the
crisis period.

4.1 Currency markets
Contagion during the Asian ªnancial crisis is widely believed to have originated
with the ºoat and depreciation of the Thai baht on 2 Julio 1997. This date is often
used in empirical applications to mark the start of the crisis despite the fact that
some claim that turmoil was evident in equity markets earlier; see McKibbin and
Martín (1998). The problems in dating ªnancial crises are quite pronounced and rep-
resent one of the issues which make comparisons of results across empirical studies
difªcult. Despite the importance of currency markets in the crisis, the literature on
testing for contagion in Asian currencies is limited, reºecting the fact that these mar-
kets were largely operating under ªxed-exchange-rate regimes prior to 1997, hacer-
ing pre-crisis-period volatility comparisons difªcult.

Under the alternative hypothesis that the contemporaneous spread of currency cri-
ses is consistent with contagion, Glick and Rose (1999), using a large panel of ªve
Asian countries, ªnd that the most important economic linkages explaining these
transmissions are trade linkages. Van Rijckeghem and Weder (2001), using a similar

48

Asian Economic Papers

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Correlación, Contagio, y evidencia asiática

methodology, ªnd that ªnancial links are more important, although they also note
that as a result of the close relationship between ªnancial and trade linkages it is
difªcult to disentangle the two.

Baig and Goldfajn (1999) show a pronounced increase in correlation coefªcients dur-
ing the crisis period as evidence of contagion. Sin embargo, the effects of this contagion
on developed markets in the region is muted; notably both Debelle and Ellis (2005)
and Dungey, Fry, and Martin (2004), using an alternative statistical framework,
show small effects of contagion to Australia and New Zealand, and the latter paper
to Japan.

Substantial contagion effects within the Asian region are revealed in Dungey and
Martín (2004). In investigating the links among Malaysia, Indonesia, Korea, y
Thailand during 1997–98, Dungey and Martin ªnd statistically signiªcant contagion
effects of up to 46 percent of total volatility for Korea, mostly sourced from Thai-
based shocks. Sin embargo, although Thailand is the major source of shocks for Korea
and Malaysia, the majority of the effects on Indonesia are transmitted indirectly
through Malaysia.

4.2 Equity markets
Equity markets have probably received the most attention in empirical applications
of contagion tests in East Asian markets. A number of these applications have exam-
ined the effects of contagion during the period associated with the turmoil in the
Hong Kong equity market in late October 1997, including Baig and Goldfajn (1999),
Forbes and Rigobon (2002), Baur and Schulze (2005), Baur and Fry (2006), Vínculo,
Dungey, and Fry (2006), and Dungey, Fry, González-Hermosillo, and Martin (2005b).
De estos, Forbes and Rigobon (2002), Baur and Schulze (2005), and Bond, Dungey,
and Fry (2006) focus on Hong Kong as the identiªed source of the potentially conta-
gious shocks, whereas the others look more generally at contagion in the region.
Kleimeier, Lehnert, and Verschoor (2003) and Baur and Schulze (2005) focus on
Thailand as the source country for the potentially contagious equity market shock.

Studies are divided in their results on the importance of contagion effects in Asian
equity markets. Por un lado, Forbes and Rigobon (2002) and Kleimeier,
Lehnert, and Verschoor (2003) ªnd little or no evidence of contagion using tests
based on a statistically signiªcant increase in the correlation coefªcient between re-
turns in these markets. Both papers use the same methodology, although Kleimeier,
Lehnert, and Verschoor control precisely for the timing of the observations on the
various equity markets, whereas other studies use market-closing observations, re-
sulting in differences in the effective times across the day. Por otro lado, a num-

49

Asian Economic Papers

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Correlación, Contagio, y evidencia asiática

ber of papers do ªnd signiªcant contagion effects in East Asia. Caporale, Cipollini,
and Spagnolo (2005) ªnd evidence for contagion in almost all pairs of eight Asian
economies using conditional correlation analysis.4

Vínculo, Dungey, and Fry (2006) show that during the period of 1997–98 associated
with turmoil in Hong Kong, equity and real estate markets behave somewhat differ-
ently, but both indicate statistically signiªcant contagion effects. This paper also em-
phasizes the role of the Japanese and Singaporean markets in helping to transmit the
contagion effects around the region. Contagion from Hong Kong is found to be sig-
niªcant to Singapore and the Philippines by Corsetti, Pericoli, and Sbracia (2001)
and from Hong Kong to a wide range of Asian economies by Baur and Schulze
(2005). Baur and Fry (2006) ªnd that contagion is signiªcant across 11 countries dur-
En g 8 percent of days over the period of the Asian crisis, with Hong Kong being an
important factor.5

Bekaert, harvey, and Ng (2005) and Wongswan (2003) also ªnd evidence of signiª-
cant contagion effects within Asia, but not from Asia to other countries or regions.
Wongswan (2003) controls for common effects by ªtting a capital asset pricing
modelo, including GARCH conditional variances, and examines the correlations
among the residuals across countries and regions. Baig and Goldfajn (1999) uncover
mixed evidence of contagion in equity markets. They ªnd evidence for bivariate
contagion between Korea and each of Indonesia, Malasia, the Philippines, y
Thailand and between Indonesia and Malaysia and Indonesia and Thailand, pero no
between other pairings of these countries. There is also evidence of contagion from
both Korean and Thai equity markets to Indonesian equities in Cerra and Saxena
(2002), which uses a Markov switching modeling approach.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

The effect of a change in the propagation of the common factor during the Asian cri-
sis as an alternative explanation is explored in Rigobon (2003) using his determinant
of change in covariance (DCC) prueba. The test does not suggest any break in the prop-
agation in Asia during the Asian crisis but provides some limited evidence that
other countries experienced a change in their propagation coefªcients associated
with the crisis in Thailand during 1997–98, notably India, South Africa, and Russia.
Other evidence on structural breaks by Yang, Kolari, and Min (2003) suggests that

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

4 The countries considered are Hong Kong, Indonesia, Korea, Malasia, the Philippines, Singa-

pore, Taiwán, y Tailandia.

5 The countries considered are China, Hong Kong, India, Indonesia, Japón, Korea, Malasia,

the Philippines, Singapur, Taiwán, y Tailandia.

50

Asian Economic Papers

Correlación, Contagio, y evidencia asiática

there was some increase in the intensity of the integration among the Asian equity
markets during the crisis period.

4.3 Fixed-income markets
Studies of contagion in ªxed-income markets for Asia are far less common than for
other asset types. This partly arises from the relative stability in these markets up
until the events of August–September 1998, when the Russian government sus-
pended payment and subsequently defaulted on sovereign bond payments, creating
substantial volatility in international bond markets; see the analysis of the period in
the BIS report by the Committee on the Global Financial System (1999).

Only three studies have given particular attention to bond markets with East Asian
markets as the focus. Baig and Goldfajn (1999) conduct a correlation analysis for
bond spreads similar to that for the currency and equity markets and ªnd evidence
of contagion among the Asian sovereign bond markets. Debelle and Ellis (2005) estafa-
sider links from East Asia to Australia and New Zealand through bond markets and
ªnd evidence of signiªcant but small contagion effects. Sander and Kleimeier (2003)
view contagion as a change in cross-market interdependencies, which they assess
using Granger causality tests during pre-crisis and crisis periods. They ªnd evi-
dence for crisis-related changes in the short-term propagation mechanism, cual
they denote as contagion, a saber, from the Philippines to Malaysia and Korea; de
Indonesia to the Philippines, Tailandia, and Malaysia; from Malaysia to Thailand;
and from Korea to Indonesia. They attribute their ªnding of no contagion from
Thailand as suggesting that Korea was a more important inºuence. Sin embargo, este
may well relate to the fact that the crisis in Thailand erupted in currency and equity
markets and was not particularly visible in sovereign debt markets.

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

4.4 Cross-country linkages and cross-market linkages
En algunos casos, a particular asset market is hit by an identiªable shock, but this does
not result in unusual levels of volatility in the returns for the asset. The clearest ex-
ample of this phenomenon is the successful defense of managed or ªxed-exchange-
rate arrangements, such as those conducted by the currency board in Hong Kong
over a number of instances in 1997–98. Sin embargo, it would be incorrect to conclude
that it is possible to isolate economies from the effects of such shocks, as the volatil-
ity was transferred to alternative asset types. En octubre 1997 and January, Junio, y
Agosto 1998, speculators attacked the Hong Kong currency board. En cada caso, el
currency was successfully defended, but substantial volatility emerged in equity
markets.6 The general principle is that ªnancial markets are closely interrelated and

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

6 See Goodhart and Dai (2003) for a detailed review of the Hong Kong Monetary Authority’s

subsequent defense of its currency through equity market intervention.

51

Asian Economic Papers

Correlación, Contagio, y evidencia asiática

that studying the transmission in crises across one asset market type in isolation
may not yield the appropriate conclusions, particularly for policymakers who need
to consider the impact on all asset markets and the economy in general.7

Given the well-known linkages between markets and economies in Asia, it is some-
what surprising that contagion in multiple markets has tended to be understudied.
One possible reason is the obvious difªculties in combining models of the different
markets. Sin embargo, there are some developments in this vein of the literature. A few
papers consider modeling linkages among different asset classes during a crisis
within a particular country: Granjero, Huang, and Yang (2000), Fang and Miller
(2002), Hartman, Straetmans, and de Vries (2004), and recently, for pairs of asset
types across geographical borders, Hartman, Straetmans, and de Vries (2004).
Vínculo, Dungey, and Fry (2006) attempt to differentiate transmissions among real es-
tate market instruments from those in the related equity markets by using orthogo-
nal shocks obtained from regressing equity market returns on real estate returns.
Some papers consider different asset types during the same crisis, but not the poten-
tial interactions among them; Baig and Goldfajn (1999) and Debelle and Ellis (2005)
consider a range of markets during the Asian crisis of 1997–98.

More relevant to the current discussion are models that consider transmissions
among multiple markets across potentially different asset types and a number of
economías. Ito and Hashimoto (2005) use a Tobit model on extreme returns to test
for contagion across equity markets and currency markets during the Asian crisis of
1997–98. Khalid and Kawai (2003) conduct Granger causality tests on the residuals
of a VAR across equity, currency, and bond markets. Kaminsky and Reinhart (2002)
identify the commonalities in equity, bond, and currency markets across a range of
countries using principal-components methods.

These papers almost always ªnd evidence of statistically signiªcant contagion
across different asset market types. Dungey and Martin (2006) ªnd that equity mar-
ket shocks are contagiously transmitted to currency markets and vice versa. Esto es
supported by the Granger causality tests in Granger, Huang, and Yang (2000), cual
ªnds feedback between equity and currency markets for Malaysia, Singapur, Thai-
land, and Taiwan during the period of 1 Junio 1997 a 16 Junio 1998.8 The exception to
the ªnding of cross-market, cross-country linkages is Khalid and Kawai (2003),
which concludes that the evidence for signiªcant contagion is limited.

7 The seminal work on crises by Kindelberger (1996) discusses the importance of cross-market

ªnancial interlinkages well before the Asian crisis.

8 For other countries in the study, they ªnd Granger causality from equity to currency markets

for Hong Kong and the Philippines and from currency to equity markets for Korea.

52

Asian Economic Papers

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Correlación, Contagio, y evidencia asiática

A further important conclusion to be drawn from comparing both cross-market and
cross-country evidence is that crises seem to propagate through various asset classes
differently. This is shown to be the case for equity markets and currency markets in
Dungey and Martin (2006), for equity markets and real estate instruments in Bond,
Dungey, and Fry (2006), and for equity and bonds in Hartmann, Straetmans, y
de Vries (2004).

5. Propositions and evidence

This section addresses, Sucesivamente, the seven propositions in regard to contagion out-
lined in the introduction. These propositions are investigated by drawing on nine
papers of the authors, using versions of the factor decomposition approach. An ad-
vantage of this approach is that it is possible to decompose volatility into a set of la-
tent factors, including a common factor, an idiosyncratic factor, and contagion fac-
tores, and thereby quantify the magnitude of contagion as a proportion of overall
volatility. Other speciªcations include some combination of regional factors, asym-
metric contagion factors to account for positive or negative contagion effects,
multiple-asset markets, subperiods of contagion within a crisis, and in some appli-
cations, the simultaneous modeling of multiple crises. This factor structure is related
to the model in equations (1) a (14) de Sección 3. These papers are summarized in
Mesa 1 in terms of the crisis addressed, the asset market investigated, the sample of
countries considered, and the factor structure speciªed in the paper.9

5.1 Strong fundamentals imply immunity to contagion
Sachs, Tornell, and Velasco (1996) suggest that countries with strong fundamentals
experience less contagion than other countries. Athukorala and Warr (2002) go fur-
ther by saying that contagion cannot occur in countries that do not have some un-
derlying “vulnerability,” in terms of some fundamental weakness; see also
Kaminsky and Reinhart (1999). The evidence for Asia is mixed in this respect. Fur-
man and Stiglitz (1998) suggest that the standard macroeconomic fundamentals in
the main crisis countries showed little evidence of weakness ahead of the crisis. En
contrast, Athukorala and Warr (2002) examine the ªnancial fragility, reserve ade-
quacy, and real exchange rate misalignment of the key crisis countries in compari-
son to a set of noncrisis countries and ªnd the opposite result. Using the method of
Frankel and Rose (1996), Furman and Stiglitz (1998) predict a less than 7 por ciento
probability of a currency crisis in 1997 for Malaysia, Indonesia, Tailandia, y el

9 For precise information on the estimation of each model, consult the papers in Table 1. Modelo
identiªcation and estimation is discussed in general terms for both the tranquil period and
the crisis period in section 3.1.

53

Asian Economic Papers

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Correlación, Contagio, y evidencia asiática

t
norte
oh
C

,

gramo
mi
r

,

oh
i
d

i

,

metro
tu
norte

,

metro
oh
C

t
norte
oh
C

,

oh
i
d

i

,

metro
oh
C

t
norte
oh
C

,

y
t
norte
C

,
t
k
metro

,

oh
i
d

i

,

metro
oh
C

t
norte
oh
C

,

gramo
mi
r

,

oh
i
d

i

,

metro
oh
C

t
norte
oh
C

,

oh
i
d

i

,

metro
oh
C

t
norte
oh
C

,

oh
i
d

i

,

metro
oh
C

t
norte
oh
C

,

gramo
mi
r

,

oh
i
d

i

,

metro
oh
C

t
norte
oh
C

,

gramo
mi
r

,

oh
i
d

i

,

metro
oh
C

t
norte
oh
C

,

oh
i
d

i

,

metro
oh
C

t
norte
oh
C

,

oh
i
d

i

,

metro
oh
C

t
norte
oh
C

,

oh
i
d

i

,

metro
oh
C

,

S
Ud.
k
Ud.
norte
,
mi
METRO

,

,

,
r
B
,
r
A
R
PAG
B
t
k

,

,

,

,

,
I

,

t
METRO
k

,

,
I
,

k
h
A

,

,

t
METRO
k

,

,
I

,

S
Ud.
k
Ud.

,
j
,

,

,

METRO
D
R
PAG
t
k
h

,

,

,
r
B
,
r
A

,

S
Ud.
t
METRO

,

,
I
,

,

k
A

S
Ud.

,

,
j
,
S
k
h
A

,

,

METRO
k
k
h

,

,

mi
METRO
C
,
r
A

C
,
r
B
,
r
A

.
1

.
2

.
3

,

S
Ud.
k
Ud.
norte
,
mi
METRO

,

,

,
r
B
,
r
A
R
PAG
B
t
k

,

,

,

,

,
I

z
norte

,
j
,

,

,

A
t
METRO
k

,

,
I

s
r
oh
t
C
a
F

s
mi
i
r
t
norte
tu
oh
C

y
yo
t
norte

i
oh

j

d
mi
yo
mi
d
oh
metro
y
C
norte
mi
r
r
tu
C
/
y
t
i
tu
q
mi

y
t
i
tu
q
mi
d
norte
a
mi
t
a
t
s
mi

yo
a
mi
R

t
mi
k
r
a
METRO

y
C
norte
mi
r
r
tu
C

y
C
norte
mi
r
r
tu
C

y
t
i
tu
q
mi

d
norte
oh
B

METRO
C
t
l
/
a
i
s
s
tu
R

)
2
0
0
2
(
norte
i
t
r
a
METRO
d
norte
a

,

oh
yo
yo
i
s
oh
metro
r
mi
h
–
z
mi
yo
á
z
norte
oh
GRAMO

,

y
r
F

,

y
mi
gramo
norte
tu
D

s
i
s
i
r
C

a

r
mi
pag
a
PAG

a
i
s
A

a
i
s
A

a
i
s
A

a
i
s
A

a
i
s
A

)
3
0
0
2
(
norte
i
t
r
a
METRO
d
norte
a

,

y
r
F

,

y
mi
gramo
norte
tu
D

)
4
0
0
2
(
norte
i
t
r
a
METRO
d
norte
a

,

y
r
F

,

y
mi
gramo
norte
tu
D

)
6
0
0
2
(
y
r
F
d
norte
a

,

y
mi
gramo
norte
tu
D

,

d
norte
oh
B

)
4
0
0
2
(
norte
i
t
r
a
METRO
d
norte
a
y
mi
gramo
norte
tu
D

)
6
0
0
2
(
norte
i
t
r
a
METRO
d
norte
a
y
mi
gramo
norte
tu
D

s
yo
mi
d
oh
metro

r
oh
t
C
a
F
gramo
norte
i
s
tu
s
r
oh
h
t
tu
a
y
b
norte
oh
i
gramo
a
t
norte
oh
C
norte
oh
s
r
mi
pag
a
pag
F
oh
y
r
a
metro
metro
tu
S

.
1

mi
yo
b
a
t

y
t
i
tu
q
mi

y
t
i
tu
q
mi

y
t
i
tu
q
mi

y
t
i
tu
q
mi

d
norte
oh
B

mi
norte

i
t
norte
mi
gramo
r
A

a
yo
i

tu
q
mi
t

a
i
s
A

.
1

.

2

.

3

METRO
C
t
l
/
a
i
s
s
tu
R

METRO
C
t
l
/
a
i
s
s
tu
R

)
b
5
0
0
2
(
norte
i
t
r
a
METRO
d
norte
a
,

oh
yo
yo
i
s
oh
metro
r
mi
h
–
z
mi
yo
á
z
norte
oh
GRAMO

,

y
r
F

,

y
mi
gramo
norte
tu
D

)
C
5
0
0
2
(
norte
i
t
r
a
METRO
d
norte
a
,

oh
yo
yo
i
s
oh
metro
r
mi
h
–
z
mi
yo
á
z
norte
oh
GRAMO

,

y
r
F

,

y
mi
gramo
norte
tu
D

)
a
6
0
0
2
(
norte
i
t
r
a
METRO
d
norte
a
,

oh
yo
yo
i
s
oh
metro
r
mi
h
–
z
mi
yo
á
z
norte
oh
GRAMO

,

y
r
F

,

y
mi
gramo
norte
tu
D

,
s
d
norte
a
yo
r
mi
h
t
mi
norte

norte

,
oh
C
i
X
mi
METRO

mi

METRO

,
a
i
s
y
a
yo
a

METRO

METRO

,
a
mi
r
oh
k

k

,

norte
a
pag
a
j

j

,
a
i
s
mi
norte
oh
d
norte
I

I

,
gramo
norte
oh
k
gramo
norte
oh
h

k
h

,
y
norte
a
metro

r
mi
GRAMO

METRO
D

,
mi
yo
i
h
C

C

,
yo
i
z
a
r
B

r
B

,
a
i
r
a
gramo
yo
tu
B

B

,
a
norte
i
t
norte
mi
gramo
r
A

r
A

,
a
i
yo
a
r
t
s
tu
A

A

:
mi
t
oh
norte

t
k
metro

,
yo
a
norte
oh
i
gramo
mi
r

gramo
mi
r

,

norte
oh
i
gramo
a
t
norte
oh
C

t
norte
oh
C

,
C
i
t
a
r
C
norte
y
s
oh
i
d
i

oh
i
d
i

,

norte
oh
metro
metro
oh
C

metro
oh
C

;
s
mi
t
a
t
S
d
mi
t
i
norte
Ud.

S
Ud.

,

metro
oh
d
gramo
norte
i
k
d
mi
t
i
norte
Ud.

k
Ud.

,
d
norte
a
yo
i
a
h
t

t

,
mi
r
oh
pag
a
gramo
norte
i
S

S

,
a
i
s
s
tu
R

R

,
d
norte
a
yo
oh
PAG

PAG

,
d
norte
a
yo
a
mi
z
w
mi
norte

z
norte

.
t
norte
mi

metro

mi
gramo
a
norte
a
metro

yo
a
t
i
pag
a
C
metro
r
mi
t
–
gramo
norte
oh
yo

METRO
C
t
l

;
mi
r
i
a
r
mi
metro
tu
norte

metro
tu
norte

,
y
r
t
norte
tu
oh
C

y
t
norte
C

,
t
mi
k
r
a
metro

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

54

Asian Economic Papers

Correlación, Contagio, y evidencia asiática

Philippines. A diferencia de 17 other countries are found to have higher probabilities of
crisis.

The focus of existing studies has been on whether fundamentals in Asia were strong
or weak relative to those in other countries that have experienced ªnancial crises, como
in the comparisons with the Mexican peso crisis presented in Sachs, Tornell, y
Velasco (1996), Furman and Stiglitz (1998), and Kaminsky and Reinhart (1999).
Sin embargo, there may also be insights to be gained by comparing the relative strength
of fundamentals within the crisis-affected countries in conjunction with the degree
of contagion they experienced. The top panel of Table 2 provides details on the rela-
tive values of a selection of fundamental macroeconomic variables for Indonesia,
Korea, Malasia, y Tailandia. These variables reºect those found to be important
in the ªnancial crisis literature represented by Sachs, Tornell, and Velasco (1996),
Furman and Stiglitz (1998), and Frankel and Rose (1996).

A good set of economic fundamentals in Table 2 would be composed of a high GDP
per capita, strong GDP growth, a low short-term-debt-to-reserves ratio, low and sta-
ble inºation, a low rate of nonperforming loans, and an exchange rate near its equi-
librium value.10 More difªcult to interpret is the role of openness and export growth.
Typically, openness is seen as a requirement for development, and a high degree of
openness is desirable. Export growth is also generally seen as important to an econ-
omy, indicating strong domestic growth, although it may also be associated with
overheating as in the case of Thailand; see Athukorala and Suphachalasai (2004).

The country with the strongest fundamentals in Table 2 is Korea. It has higher GDP
per capita and higher average GDP growth in the run-up to the crisis, a low rate of
nonperforming loans, and an exchange rate close to equilibrium. Sin embargo, también
has the highest short-term-debt-to-reserves ratio of the four countries. Its openness
and export growth are only slightly higher than those of Indonesia, which is clearly
the worst-performing country in the sample on the basis of these indicators. Nota
that Kenward (1999) describes the lack of macroeconomic indicators of weakness in

10 The real exchange rate misalignment in Table 2 is computed relative to an equilibrium

measure based on the predicted value of the real exchange rate in 1996 estimated with levels
of real GDP per capita taking into account the Balassa-Samuelson effect. Furman and
Stiglitz (1998) give four alternate measures, of which this is one, and the extent (e incluso
sign) of the deviation from equilibrium differs across the measures. Two of the measures
given in Furman and Stiglitz agree with the relative ordering of the extent of misalignment
aquí, eso es, that Malaysia is the most misaligned and Korea the least. The other two mea-
sures suggest that Korea is the most misaligned and Indonesia the least. Por eso, the results
vary enormously with the measure chosen, and undue weight should not be placed on the
indicator adopted in Table 2.

55

Asian Economic Papers

yo

D
oh
w
norte
oh
a
d
mi
d

F
r
oh
metro
h

t
t

pag

:
/
/

d
i
r
mi
C
t
.

metro

i
t
.

/

mi
d
tu
a
s
mi
pag
a
r
t
i
C
mi
–
pag
d

/

yo

F
/

/

/

/

/

5
2
3
2
1
6
8
2
2
8
0
a
s
mi
pag
2
0
0
6
5
2
3
2
pag
d

.

.

.

.

.

F

b
y
gramo
tu
mi
s
t

t

oh
norte
0
8
S
mi
pag
mi
metro
b
mi
r
2
0
2
3

Correlación, Contagio, y evidencia asiática

d
norte
a
yo
i
a
h
t

a
i
s
y
a
yo
a
METRO

1
5
0
7

,

9
4
5

.

5
4
1

.

7
3
4

.

.

.

a
norte

6
1
4
8

.

8
1
(cid:14)

8
2
5

3
A

A

7
8
4
0
1

,

4
0
4

,

1

9
4
8

9
8
6
,
9

4
5
.
4

1
6
.
0

1
0
.
1

3

6
2
.
8
7
1

1
4
(cid:14)

1
7
4

(cid:4)
A

1
A

4
2
8
,
3
1

8
9
2
,
1

7
2
9
,
7

a
mi
r
oh
k

4
0
7
,
4
1

3
6
.
6

7
0
.
2

7
3
.
4

8
.
0

3
6
.
7
6

4
7
2

1

(cid:14)
A
A

1
A

5
3
5
,
4
1

5
4
3
,
2
(cid:14)

2
5
7

a
i
s
mi
norte
oh
d
norte
I

7
3
1
,
4

6
0
.
5

7
6
.
4

7
.
1

8
.
8

4
0
.
6
5

6
1
(cid:14)

6
3
2

3
a
a
B

B
B
B

1
4
8
,
9

6
5
9

4
9
5
,
5

d
norte
a
yo
i
a
h
t

a
i
s
y
a
yo
a
METRO

a
mi
r
oh
k

4
1
9
2

.

0
0
1
1

.

8
7

.

7

9
5
.
9

2
9
.
9
4

5
6
.
2
2

1
3
.
6
4

5
4
.
4
7

1
1
.
0
1

2
3
.
5

a
i
s
mi
norte
oh
d
norte
I

5
7
.
9
1

8
6
.
0

)

%

(
y
t
i
yo
i
t
a
yo
oh
v
F
oh
norte
oh
i
t
r
oh
pag
oh
r
pag
norte
oh
i
gramo
a
t
norte
oh
C

mi
C
r
tu
oh
S

a
i
s
A

gramo
norte
oh
k
gramo
norte
oh
h

d
norte
a

a
i
s
s
tu
R

a
i
s
y
a
yo
a
METRO

a
mi
r
oh
k

.

.

S
Ud.

d
norte
a

a
i
s
s
tu
R

.

.

S
Ud.

i

)
norte
oh
i
yo
yo
i

metro
$ S U ( 6 9 9 1 , t n e m t s e v n i t c e r i d . . S U i ) n o i l l i m $
S
Ud.

(

6
9
9
1

,
s
w
oh
yo
F

I

D
F
t
mi
norte

norte
oh
i
gramo
a
t
norte
oh
C
oh
t
mi
tu
d
y
t
i
yo
i
t
a
yo
oh
v
F
oh
t
norte
mi
t
X
mi

)
b
5
0
0
2
(
norte
i
t
r
a
METRO
d
norte
a

,

oh
yo
yo
i
s
oh
metro
r
mi
h

–
z
mi
yo
á
z
norte
oh
GRAMO

,

y
r
F

,

y
mi
gramo
norte
tu
D

)
4
0
0
2
(
norte
i
t
r
a
METRO
d
norte
a
y
mi
gramo
norte
tu
D

t
mi
k
r
a
metro
y
C
norte
mi
r
r
tu
C

t
mi
k
r
a
metro
y
t
i
tu
q
mi

)
C
5
0
0
2
(
norte
i
t
r
a
METRO
d
norte
a

,

oh
yo
yo
i
s
oh
metro
r
mi
h

–
z
mi
yo
á
z
norte
oh
GRAMO

,

y
r
F

,

y
mi
gramo
norte
tu
D

t
mi
k
r
a
metro
d
norte
oh
B

)
a
6
0
0
2
(
norte
i
t
r
a
METRO
d
norte
a

,

oh
yo
yo
i
s
oh
metro
r
mi
h

–
z
mi
yo
á
z
norte
oh
GRAMO

,

y
r
F

,

y
mi
gramo
norte
tu
D

t
mi
k
r
a
METRO

i

)
norte
oh
i
yo
yo
i

metro
$ S U ( 6 9 9 1 , s w o l f l a t i p a c t e N b 7 9 9 1 e n u J , o i t a r s e v r e s e r – o t – t b e d – m r e t – t r o h S d c 6 9 9 1 , ) l a t o t f o % ( i s n a o l g n m r o f r e p n o N 7 9 9 1 , ) % ( n o i t a l f n i e c i r p r e m u s n o C a 6 9 – 0 7 9 1 , ) % ( h t w o r g P D G e g a r e v A a ) $
S
Ud.

(

7
9
9
1
,
a
t
i
pag
a
C

r
mi
pag
PAG
D
GRAMO

s
yo
a
t
norte
mi
metro
a
d
norte
tu
F

F

)

%

(

6
9
9
1
,

mi

norte
oh
i
t
a
tu
yo
a
v
r
mi
v
oh
mi
t
a
r

mi
gramo
norte
a
h
C
X
mi

)

%

(

6
9
–
6
8
9
1
,

h
t
w
oh
r
gramo
t
r
oh
pag
X
mi

Y
/
)
METRO
(cid:4)
X

(

a

s
s
mi
norte
norte
mi
pag
oh

h
gramo
norte
i
t
a
r

s
'
r
oh
oh
PAG
d
norte
a
d
r
a
d
norte
a
t
S

s
k
norte
i
yo

yo
a
i
C
norte
a
norte
i
F

gramo
gramo
norte
i
t
a
r

s
'

y
d
oh
oh
METRO

y
r
t
norte
tu
oh
C
y
b
d
mi
C
norte
mi
i
r
mi
pag
X
mi
norte
oh
i
gramo
a
t
norte
oh
C

F
oh
t
norte
mi
t
X
mi
d
norte
a

s
r
oh
t
a
C
i
d
norte
i

yo
a
t
norte
mi
metro
a
d
norte
tu
F

.
2

mi
yo
b
a
t

mi
norte
tu
j
–
y
r
a
tu
norte
a
j

r
oh
F

mi
t
a
r
norte
oh
i
t
a
yo
F

norte

i

yo
a
tu
norte
norte
a
y
yo
h
t
norte
oh
metro

mi
gramo
a
r
mi
v
a

;
s
norte
oh
i
t
a
yo
tu
C
yo
a
C

'
s
r
oh
h
t
tu
a

,
s
C
i
t
s
i
t
a
t
s
S
F
I
F
METRO

I

.
C

;
3

mi
yo
b
a
t

,
)
8
9
9
1
(

s
h
C
a
S
d
norte
a

t
mi
yo
mi
d
a
R

.
b

;
mi
t
i
s
b
mi
w
R
mi
B
norte
a
i
v

d
mi
s
s
mi
C
C
a

1
.
6
v
s
mi
yo
b
a
t
d
yo
r
oh
W.
norte
norte
mi
PAG

.
a

:
s
mi
C
r
tu
oh
S