Diagnosing the Source of Financial Market Shocks: An Application to the Asian, Subprime and European Financial Crises
| Published date | 01 December 2018 |
| DOI | http://doi.org/10.1111/1468-0106.12162 |
| Date | 01 December 2018 |
| Author | Marina Marinkov,Johannes W. Fedderke |
DIAGNOSING THE SOURCE OF FINANCIAL MARKET
SHOCKS: AN APPLICATION TO THE ASIAN, SUBPRIME
AND EUROPEAN FINANCIAL CRISES
JOHANNES W. FEDDERKE*Pennsylvania State University, Economic Research
Southern Africa and University of the Witwatersrand
MARINA MARINKOV International Monetary Fund
Abstract. This paper presents a test diagnostic that determines whether financial shocks are due to the
propagation of idiosyncratic shocks originating in a single source country (or group of countries), or
a reflection of market interdependence due to factors common across markets. The test is given by
the ratio, λ, of the unconditional to the conditional correlation coefficient between markets. We dem-
onstrate analytically that the test statistic is robust to heteroscedasticity due to conditional market
volatility, to the impact of omitted variables (particularly important in the event that shocks may
be transmitted between any two markets via a third ‘intermediate’market) and to the impact of
endogeneity between markets. Size and power characteristics of the test are strong. An application
to the Asian financial crisis of 1997–1998, the subprime crisis of 2007 and the European crisis of
2009 demonstrates its empirical tractability. For the Asian and the subprime crises, the λ-test sug-
gests that propagation of shocks was predominantly due to common fundamentals: in the European
crisis shock propagation by contrast is indicated to be due to idiosyncratic shocks centred on Cyprus,
Greece and Latvia.
1. INTRODUCTION: DIAGNOSING THE SOURCE OF FINANCIAL SHOCKS
The persistence and severity of financial crises into the 21st century has placed
renewed emphasis on the need for clear diagnostic tools that serve to identify
the nature of transmission channels between financial markets.
In this paper we present a diagnostic statistic that serves to identify whether
the transmission of a shock to financial markets is due to contagion, or due to
the fact that the markets are interdependent. The virtue of the test is twofold.
First, it is very simple to implement. Second, we demonstrate analytically that
the test is robust to a range of the difficulties that have been identified as
attaching to preexisting diagnostic tests in the literature in a clearly defined
manner. We also clearly specify the limits of the test.
The literature on financial crisis propagation has employed a range of alter-
native conceptions of contagion.1Our analysis employs the definition of conta-
gion provided by Forbes and Rigobon (2002), as a significant increase in cross-
*Address for Correspondence: Katz, Pennsylvania State University, State College, PA 16802, USA.
E-mail: jwf15@psu.edu. The financial research support of Economic Research Southern Africa is ac-
knowledged. Our thanks to Hao Deng for valuable assistance with some of the core derivations of the
paper. Vimal Ranchod and Yang Liu provided assistance in preparing the data set for this study. The
views expressed in this paper are those of the authors alone, and do not necessarily reflect those of the
International Monetary Fund.
1See the discussion in Forbes and Rigobon (2001) on the range of definitional alternatives and con-
troversies surrounding contagion.
Pacific Economic Review,••:•• (2016)
doi: 10.1111/1468-0106.12162
© 2016 John Wiley & Sons Australia, Ltd
market linkages after a shock to one country, measured as an increase in cross-
market correlations. By contrast, persistent high levels of correlation across mar-
kets, including in the absence of a shock, under this definition of contagion is
termed interdependence of markets.2
The definition is narrow, but has a number of clear advantages. First, it is pre-
cise in terms of what constitutes contagion, and differentiates contagion from
market interdependence. Second, the definition has the virtue of simplicity in
avoiding the need to identify which of a range of theoretically feasible transmis-
sion mechanisms and channels are operational in any specific crisis. This con-
trasts with a range of alternative structural approaches to contagion that rely
on concrete transmission mechanisms. For example, Kodres and Pritsker
(2002) and Erdorf and Heinrichs (2011) claim contagion to be driven by funda-
mentals, while others view it to result from overreactions in markets (Goldstein
and Pauzner, 2004; Broner et al., 2006). Channels of transmission have been
identified as trade (Glick and Rose, 1999), financial (van Rickenghem and
Weder, 2001; Kunieda and Shibata, 2013; Bremus and Bruch, 2015; Lane,
2015), similarity between economies (Eichengreen et al., 1996), and policy coor-
dination or geographical proximity (Bayoumi et al., 2003, 2007). Rose and
Spiegel (2010) allow for both a financial and a real channel, depending on
whether the link between markets is due to the holding of securities or trade.
See also the review discussions in Claessens et al. (2001) and Forbes and
Rigobon (2001). Such an agnosticism as to transmission channels has the very
significant advantage of avoiding the identification challenge of isolating
changes in investor behaviour or specific real transmission mechanisms, all of
which are econometrically challenging.
Correlation-based tests of contagion have a long history,3and face their own
measurement and methodological challenges. Of particular concern is that sea-
sonality, outliers and conditional heteroscedasticity affect correlation-based
tests by introducing size distortion (see Beine and Hecq, 1999; Hecq, 1998).
Forbes and Rigobon (2002) demonstrate analytically that tests based on correla-
tions are biased unless they correct for heteroscedasticity resulting from volatil-
ity conditional on crisis events: increased conditional volatility will bias
correlation-based measures upward, providing false positive diagnoses of conta-
gion (hence the poor size properties of correlation based tests). They further
show that omitted variables affecting cross-market linkages, or endogeneity
between markets, also bias correlation-based test statistics, requiring additional
corrections to the testing procedure for contagion.
2Increased correlations across markets due to globalization have been documented both for equity
(see e.g. Berben and Jansen, 2005; Morana and Beltratti, 2008) and bond markets (see e.g. Hunter
and Simon, 2004).
3See, for example, King and Wadhwani (1990) on the US stock market crash of 1987, Calvo and
Reinhart (1995) on the Mexican peso crisis, and Baig and Goldfajn (1998) on the 1997 Asian crisis.
Forbes and Rigobon (2001, 2002) provide a discussion and review of the literature, but see Billio and
Caporin (2010), Corsetti et al. (2005), Hossein and Nossman (2011) for more recent
implementations. Note, however, that correlations after crises tend to increase within asset classes,
but decrease between asset classes, most likely due to a ‘flight to quality’during crises; see the discus-
sion in Brière et al. (2012).
JOHANNES W. FEDDERKE AND MARINA MARINKOV2
© 2016 John Wiley & Sons Australia, Ltd
Pacific Economic Review
, 23: 5 (2018) pp. 742–777
doi:10.1111/1468-0106.12162
© 2016 John Wiley & Sons Australia, Ltd
DIAGNOSING THE SOURCE OF FINANCIAL MARKET
SHOCKS: AN APPLICATION TO THE ASIAN, SUBPRIME
AND EUROPEAN FINANCIAL CRISES
JOHANNES W. FEDDERKE*Pennsylvania State University, Economic Research
Southern Africa and University of the Witwatersrand
MARINA MARINKOV International Monetary Fund
Abstract. This paper presents a test diagnostic that determines whether financial shocks are due to the
propagation of idiosyncratic shocks originating in a single source country (or group of countries), or
a reflection of market interdependence due to factors common across markets. The test is given by
the ratio, λ, of the unconditional to the conditional correlation coefficient between markets. We dem-
onstrate analytically that the test statistic is robust to heteroscedasticity due to conditional market
volatility, to the impact of omitted variables (particularly important in the event that shocks may
be transmitted between any two markets via a third ‘intermediate’market) and to the impact of
endogeneity between markets. Size and power characteristics of the test are strong. An application
to the Asian financial crisis of 1997–1998, the subprime crisis of 2007 and the European crisis of
2009 demonstrates its empirical tractability. For the Asian and the subprime crises, the λ-test sug-
gests that propagation of shocks was predominantly due to common fundamentals: in the European
crisis shock propagation by contrast is indicated to be due to idiosyncratic shocks centred on Cyprus,
Greece and Latvia.
1. INTRODUCTION: DIAGNOSING THE SOURCE OF FINANCIAL SHOCKS
The persistence and severity of financial crises into the 21st century has placed
renewed emphasis on the need for clear diagnostic tools that serve to identify
the nature of transmission channels between financial markets.
In this paper we present a diagnostic statistic that serves to identify whether
the transmission of a shock to financial markets is due to contagion, or due to
the fact that the markets are interdependent. The virtue of the test is twofold.
First, it is very simple to implement. Second, we demonstrate analytically that
the test is robust to a range of the difficulties that have been identified as
attaching to preexisting diagnostic tests in the literature in a clearly defined
manner. We also clearly specify the limits of the test.
The literature on financial crisis propagation has employed a range of alter-
native conceptions of contagion.1Our analysis employs the definition of conta-
gion provided by Forbes and Rigobon (2002), as a significant increase in cross-
*Address for Correspondence: Katz, Pennsylvania State University, State College, PA 16802, USA.
E-mail: jwf15@psu.edu. The financial research support of Economic Research Southern Africa is ac-
knowledged. Our thanks to Hao Deng for valuable assistance with some of the core derivations of the
paper. Vimal Ranchod and Yang Liu provided assistance in preparing the data set for this study. The
views expressed in this paper are those of the authors alone, and do not necessarily reflect those of the
International Monetary Fund.
1See the discussion in Forbes and Rigobon (2001) on the range of definitional alternatives and con-
troversies surrounding contagion.
Pacific Economic Review,••:•• (2016)
doi: 10.1111/1468-0106.12162
© 2016 John Wiley & Sons Australia, Ltd
market linkages after a shock to one country, measured as an increase in cross-
market correlations. By contrast, persistent high levels of correlation across mar-
kets, including in the absence of a shock, under this definition of contagion is
termed interdependence of markets.2
The definition is narrow, but has a number of clear advantages. First, it is pre-
cise in terms of what constitutes contagion, and differentiates contagion from
market interdependence. Second, the definition has the virtue of simplicity in
avoiding the need to identify which of a range of theoretically feasible transmis-
sion mechanisms and channels are operational in any specific crisis. This con-
trasts with a range of alternative structural approaches to contagion that rely
on concrete transmission mechanisms. For example, Kodres and Pritsker
(2002) and Erdorf and Heinrichs (2011) claim contagion to be driven by funda-
mentals, while others view it to result from overreactions in markets (Goldstein
and Pauzner, 2004; Broner et al., 2006). Channels of transmission have been
identified as trade (Glick and Rose, 1999), financial (van Rickenghem and
Weder, 2001; Kunieda and Shibata, 2013; Bremus and Bruch, 2015; Lane,
2015), similarity between economies (Eichengreen et al., 1996), and policy coor-
dination or geographical proximity (Bayoumi et al., 2003, 2007). Rose and
Spiegel (2010) allow for both a financial and a real channel, depending on
whether the link between markets is due to the holding of securities or trade.
See also the review discussions in Claessens et al. (2001) and Forbes and
Rigobon (2001). Such an agnosticism as to transmission channels has the very
significant advantage of avoiding the identification challenge of isolating
changes in investor behaviour or specific real transmission mechanisms, all of
which are econometrically challenging.
Correlation-based tests of contagion have a long history,3and face their own
measurement and methodological challenges. Of particular concern is that sea-
sonality, outliers and conditional heteroscedasticity affect correlation-based
tests by introducing size distortion (see Beine and Hecq, 1999; Hecq, 1998).
Forbes and Rigobon (2002) demonstrate analytically that tests based on correla-
tions are biased unless they correct for heteroscedasticity resulting from volatil-
ity conditional on crisis events: increased conditional volatility will bias
correlation-based measures upward, providing false positive diagnoses of conta-
gion (hence the poor size properties of correlation based tests). They further
show that omitted variables affecting cross-market linkages, or endogeneity
between markets, also bias correlation-based test statistics, requiring additional
corrections to the testing procedure for contagion.
2Increased correlations across markets due to globalization have been documented both for equity
(see e.g. Berben and Jansen, 2005; Morana and Beltratti, 2008) and bond markets (see e.g. Hunter
and Simon, 2004).
3See, for example, King and Wadhwani (1990) on the US stock market crash of 1987, Calvo and
Reinhart (1995) on the Mexican peso crisis, and Baig and Goldfajn (1998) on the 1997 Asian crisis.
Forbes and Rigobon (2001, 2002) provide a discussion and review of the literature, but see Billio and
Caporin (2010), Corsetti et al. (2005), Hossein and Nossman (2011) for more recent
implementations. Note, however, that correlations after crises tend to increase within asset classes,
but decrease between asset classes, most likely due to a ‘flight to quality’during crises; see the discus-
sion in Brière et al. (2012).
JOHANNES W. FEDDERKE AND MARINA MARINKOV2
© 2016 John Wiley & Sons Australia, Ltd © 2016 John Wiley & Sons Australia, Ltd
DIAGNOSING THE SOURCE OF FINANCIAL MARKET SHOCKS 743
The test proposed in this paper is an explicit extension of the approach pro-
posed by Forbes and Rigobon (2002), and deals with all three of the methodo-
logical issues noted in the preceding discussion. While the structure of the test
has similarities to Corsetti et al. (2005), Forbes and Rigobon (2002), Pericoli
and Sbracia (2003), Dungey et al. (2004) and Dungey et al. (2005), it employs
a simpler test structure. Our approach contrasts with that proposed by Candelon
et al. (2005), which also extends Forbes and Rigobon (2002) in order to deal with
the same set of methodological concerns by means of a cointegration frame-
work, in that we continue to employ a ratio of correlation coefficients.
Specifically, the test is given by the ratio, λ, of the unconditional to the condi-
tional correlation coefficient between markets. We demonstrate that where the
correlation between markets is affected by shocks idiosyncratic to a single mar-
ket, the test ratio strictly assumes the value λ>1, while if shocks originate in fac-
tors common across markets, the test ratio λ<1 strictly. Thus, the test clearly
differentiates between the Forbes–Rigobon definition of contagion (increased
market linkages post-crisis under λ>1) and market interdependence (under
λ<1).4
We then demonstrate analytically that the test statistic is robust to the range
of methodological concerns that the literature has identified. First, the construc-
tion of the test explicitly renders it robust to heteroscedasticity due to
conditional market volatility.
Second, we demonstrate analytically that the test is further robust to the im-
pact of omitted variables. Where the test statistic gives a clear diagnostic signal
either in terms of contagion (idiosyncratic shocks affecting other markets λ>1)
or market interdependence due to common factors (λ<1), it would certainly do
so if the impact of the omitted variables were fully accounted for. This form of
robustness is particularly important in the event that shocks may be transmitted
between any two markets via a third ‘intermediate’market, that is not controlled
for in correlation-based tests that by construction are bivariate in nature.5
Third, we also show analytically that the test is robust to the impact of
endogeneity between markets, in the sense that where the test statistic gives a
clear diagnostic signal either in terms of contagion (idiosyncratic shocks affect-
ing other markets λ>1), it will certainly do so in the presence of endogeneity
between markets. While the analytical proof also shows that our proposed test
is undefined in the case of market interdependence due to common factors
4Such an approach which tests simultaneously whether contagion is due to idiosyncratic or to com-
mon factors, has affinities with the approach adopted by Flavin and Panopoulou (2010), who test si-
multaneously for ‘shift’and ‘pure’contagion. A number of additional studies distinguish between
common and idiosyncratic shocks (and model these two types of shocks simultaneously). Fry et al.
(2010) use coskewness tests of contagion, while Dungey and Martin (2007) use a latent factor ap-
proach to modelling the financial crises. Applying their test to 1997 Hong Kong crisis, Fry et al.
(2010) find that contagion is more pervasive across asset markets through higher moments. Dungey
and Martin (2007) find that for East Asian crisis of 1997-98, over half of the observed volatility was
the result of either country factors or idiosyncratic factors.
5See for example the discussion in Bodart and Candelon (2005, 2009) regarding the importance of
shock transmission via ‘third’markets.
DIAGNOSING THE SOURCE OF FINANCIAL MARKET SHOCKS 3
© 2016 John Wiley & Sons Australia, Ltd
(λ<1), nonetheless it is strictly true that where the simple test we propose signals
the presence of contagion, this is robust to the presence of endogeneity.
Therefore, the principal attraction of the λ-test we propose is threefold. First,
it provides a clear diagnostic to distinguish between shocks due to unique events
in one progenitor country and shocks due to common fundamentals. Second, its
implementation is extremely simple. Third, its simplicity does not compromise
its robustness to a range of methodological challenges (heteroscedasticity, omit-
ted variables and endogeneity).
Nonetheless, limitations remain. We have already noted that in the presence
of endogeneity between markets, the test is not defined for market interdepen-
dence. In addition, the test does not explicitly exploit information from
autoregressive conditional heteroscedasticity.6The test also does not facilitate
a decomposition of shocks into permanent and transitory components.7A fur-
ther issue concerns timing. Provided that markets are fully efficient, news will be
reflected simultaneously in all markets, with the implication that using data sep-
arated in time (e.g. closing values across markets) potentially biases correlation-
based tests (although note that the evidence that this is of practical significance is
not conclusive).8Moreover, in most correlation-based tests the crisis period is
defined by reference to a priori information concerning the unfolding of the crisis,
with an associated possibility of selection bias.9Alternative tests for contagion to
deal with some of these concerns have been proposed by means of employing cau-
sality tests in the frequency domain,10 outlier tests,11 nonlinearities,12 co-
exceedance tests,13 regime-switching models14 and threshold tests.15 Nonetheless,
the clarity, simplicity and robustness of the test proposed here suggest its practical
usefulness as a diagnostic.
In Section 2 of the paper we present the proposed test diagnostic, including its
size and power characteristics. Section 4 presents evidence on the robustness of
the proposed test to some alternative propagation mechanisms of financial
crises, while Section 5 examines associated empirical evidence from the Asian
financial crisis of 1997–1998. We conclude in Section 5
6As in Hong (2001), Cho and Parhizgari (2008) and Dungey and Yalama (2010)..
7For this see the separate literature which has proposed univariate decompositions of shocks into
permanent and transitory components (e.g. Beveridge and Nelson, 1981; Stock and Watson, 1988;
Gonzalo and Granger, 1995).
8See the deomnstration in Martens and Poon (2001) and the discussion in Dungey and Yalama
(2010), but note also that Kleimeier et al. (2008) show that Forbes and Rigobon type tests, while
changing the magnitude of the calculated test value, do not change their implied inference (contagion
vs. no contagion).
9See the discussion in Djacman (2013). On the other hand, in some instances (e.g. the Asian crisis
considered here), the unfolding of the crisis is well studied, and thus the likelihood of subjective bias
is minimized.
10 See Candelon et al. (2005), Breitung and Candelon (2006) and Bodart and Candelon (2005, 2009).
11 See Favero and Giavazzi (2002).
12 See Manner and Candelon (2010).
13 See Bae et al. (2003).
14 See Flavin and Panopoulou (2010) and Chang et al. (2013).
15 See Pesaran and Pick (2007).
JOHANNES W. FEDDERKE AND MARINA MARINKOV4
© 2016 John Wiley & Sons Australia, Ltd
© 2016 John Wiley & Sons Australia, Ltd
JOHANNES W. FEDDERKE AND MARINA MARINKOV
744
To continue reading
Request your trial