See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/318208488

Predicting Financial Crises: A Study of Asian Economies

Article in Global Business Review · October 2017

DOI: 10.1177/0972150917710330

CITATIONS READS
3 312

1 author:

Priyadarshi Dash
Research and Information System for Developing Countries
32 PUBLICATIONS 53 CITATIONS

SEE PROFILE

All content following this page was uploaded by Priyadarshi Dash on 14 February 2019.

The user has requested enhancement of the downloaded file.

Article

Predicting Financial Crises: Global Business Review

18(5) 1262–1277
A Study of Asian Economies © 2017 IMI
SAGE Publications
sagepub.in/home.nav
DOI: 10.1177/0972150917710330
http://gbr.sagepub.com

Priyadarshi Dash1

Abstract
Financial crises entail heavy adjustment costs on the global economy in the form of growth decelera-
tion, weak external sector, rising prices and high unemployment. Moreover, the affected economies
witness distressed policy reversals which sometimes prove detrimental to the home economy in high

se
growth periods, typical of East Asian tigers during the 1997 financial crisis. These negative externalities
inflicted upon the society in event of such crises warrant the need for credible effective crisis detec-

lu
tion and monitoring mechanisms. This article contributes to that debate by identifying the presence of
synchronicity and contagion risk in Asia in a panel cointegration model involving select financial sector

a
variables. Empirical findings of the study interestingly reveal a strong existence of cointegration among
ci
these set of variables for the full period, that is, 1990: Q1 to 2011: Q2 as well as for various sub-periods
suggesting high risk of contagion in the region. Further, the article presents a literature review on the
er
role and effectiveness of early warning systems (EWSs) with a view to examine whether this synchro-
m

nicity was adequately captured by those models or not. Regardless of the econometric techniques, the
predictions of EWSs have rarely reflected the real crisis episodes.
om

Keywords
rc

Financial crisis, panel cointegration, early warning

fo

Introduction
ot

Crisis is a blessing in disguise if it is understood properly. Each time a financial crisis1 occurs, the press
N

gets flooded with the post-mortem analyses covering regrets of excesses and hopes of possible correction
primarily in the direction of strengthening detection and monitoring mechanisms. Financial crises in the
1990s and the recent global financial crisis during 2007–2009 which, to many, parallels the Great
Depression of the 1930s reiterated the necessity of expanding the scope of diagnosis in order to avoid the
catastrophic consequences of any such crisis in future. While large fraction of policy effort was devoted

1
Research Associate, Research and Information System for Developing Countries (RIS), India Habitat Centre, New Delhi, India.

Corresponding author:
Priyadarshi Dash, Research Associate, Research and Information System for Developing Countries (RIS), Core IV-B, Fourth Floor,
India Habitat Centre, Lodhi Road, New Delhi-110003, India.
E-mails: [email protected]; [email protected]
Dash 1263

to ensure smooth flow of credit to the industrial sectors and restoring buoyant demand conditions in the
post-crisis period, the concern for proper analysis of the crisis symptoms, factors and effects reached its
peak after the present global financial crisis. This post-crisis cacophony covered a range of issues like
effectiveness of early warning systems (EWSs), efficacy of International Monetary Fund (IMF) crisis
resolution packages, and distorted incentives associated with ill-placed institutional mechanisms.
Regardless of the epicentres, these financial crises have inflicted enormous economic costs on national
exchequers and on the people at large, especially in an era that is marked by reduction in poverty,
improved standard of living and rapid spread of liberal democratic values.
Following the conventional sequence, the policy measures that the crisis-affected economies often
follow include containing short-run foreign currency exposure, rationalizing fiscal deficits and resorting
to reserve accumulation. Although the short-run effectiveness of these measures in restoring external
stability is not disputed, the need for establishing suitable crisis prevention frameworks remains a distant
goal. This view buys merit in presence of growing awareness in the Asian economies about correcting

se
economic distortions like exchange rate misalignment, excessive leverage in financial institutions, asset
price bubble and so on. At the same time, there exists a growing recognition of complementary regional

lu
level policies in early detection, amicable resolution and ensuring harmonized practices for crisis
mitigation. Apart from domestic orientation in fiscal and monetary policies, the policy choices that

a
worries most are the identification of various forms of financial vulnerabilities rooted in currency and
ci
maturity mismatch, distressed outflows of regional savings to the USA and other mature markets, bank-
dominated financial systems, shallow financial markets and absence of proper monitoring and supervi-
er
sory processes. All these deficiencies noted above require certain preemptive steps to build effective
m

crisis prevention systems, of which the EWS is considered the most feasible one. It is believed that a new
vintage of EWS models could yield robust predictions of crisis signals and thereby fast-track the policies
om

in response to those symptoms.

Against this backdrop, this article attempts to analyze the patterns, synchronicity and dependence in
the evolution of financial sector indicators in the sample Asian economies. This exercise is undertaken
rc

with a view to identify the contagion risk present in the region from the empirical results and suggest the
ways to revamp the crisis prevention mechanisms. The article is organized as the following. The second
fo

section reviews the EWS models with regard to their efficacy in identifying crisis episodes at various
points of time. The third section presents the level of financial development in the regional economies
ot

and the stylized trends and patterns exhibited in various financial variables over the period 1990–2011.
The fourth section expands the descriptive analysis to an empirical testing of pro-cyclicality and comove-
N

ments observed in those variables over the full period as well as different sub-periods during the study
period. Last but not the least, the article ends with conclusion and the roadmap for expediting coopera-
tion in these aspects of financial cooperation.

Review of Literature
Crisis prediction is a contentious area in macroeconomics. EWSs are widely used in this field with
varying degrees of success with respect to their effectiveness and timing in predicting a financial crisis
ex-ante. EWS, in general, produces predictions from the macroeconomic models maintained by the IMF
and the private financial institutions (see Berg, Borensztein & Pattillo, 2005). A standard EWS identifies
the abnormal features in behaviour of macroeconomic and financial indicators that cause financial crisis
and generate a numerical measure of likelihood of crisis (Peng & Bajona, 2008). These probability
scores produced over a defined time horizon captures the patterns in the behaviour of standard economic
1264 Global Business Review 18(5)

variables such as exchange rate, bond spread, sovereign credit ratings, export growth, reserve loss, short-
term debt-to-reserves, broad money growth, domestic credit, stock market indices and so on. Despite the
utility of EWS in detecting the crisis episodes, financial crises continued to rock the world at various
points of time. In most cases, EWS predictions resulted in several false signals and missed crisis epi-
sodes. Each time a crisis occurs; the relevance of EWS inputs is questioned, particularly with respect to
robustness and accuracy of predictions. In this context, it is imperative to know whether the existing
EWS framework plays any role in predicting crisis signals. The applicability of EWS models is purely
an empirical question and subject to the choice of econometric approaches. Empirical literature on its
role in identifying crisis periods is mixed, whereas it remains a primary tool for routine surveillance and
monitoring of macroeconomic conditions in the economy. Berg, Borensztein and Pattillo (2005) under-
take a comparative analysis of in-the-sample and out-of-sample properties of four EWS models, namely,
DCSD, KLR, GS and CSFB. The study finds that EWS models are not accurate in anticipating crises.
Fuertes and Kalotychou (2007) observe that EWS models paid least importance to forecasting, design

se
and validation issues that adequately address the decision maker’s preferences. And forecasts of the KLR
model is found more robust in predicting actual crises than other models. Applying an improvised

lu
version of the KLR model for China, Peng and Bajona (2008) find partial success of the EWS model in
locating crisis episodes. While the model correctly predicts the financial crisis in July 1994, it fails to

a
anticipate the events in the case of the East Asian crisis due to weak and continuously deteriorating
ci
fundamentals. In contrast, Alvarez-Plata and Schrooten (2004) observe the ineffectiveness of KLR
model in predicting the Argentinean crisis in 2002. Neither single indicators nor the composite indicators
er
anticipated any crisis signals before outbreak of the crisis.
m

Rose and Spiegel (2012) present a wide coverage of early warning indicators. Their paper is built
upon the assumption that real estate prices, bank leverage and stock market indices are some of those
om

segments that show the signs of departure from normalcy in a Multiple Indicator Multiple Cause
(MIMIC) model. However, the study lacks any common explanation on the causes of the current global
financial crisis and fails to generate any credible link between the causes and its incidence by concluding
rc

that the probable factors causing the crisis could be idiosyncratic. In this regard, Ali and Kestens (2006)
provide interesting findings. The results of a Geographically Weighted Regression (GWR) reveal that
fo

speculative pressure originating from trade competition, and/or access to capital markets and/or financial
and macroeconomic imbalances vary significantly across countries. It requires a proper modelling of
ot

local factors in building regional warning systems. Based on Salant and Henderson (1978) and Krugman
(1979), Ozlale and Metin-Ozcan (2007) claim that the currency crises were more predictable before the
N

1990s than the recent period. In that period, fixed exchange rate policy and expansionary pre-crisis fun-
damentals were the major factors responsible for the financial crises. Bordo et al. (2001) find support for
this strand by observing that the choice of exchange rate regime could be one of the factors that explain
the cause of a financial crisis. To them, banking crises are more likely in presence of pegged exchange
rates. However, they do not seem to favour a restrictive exchange rate regime as well. Unlike that, the
behaviour of macroeconomic variables in the post-1990 period is very uncertain. This ambiguity perhaps
leads to a proper diagnosis of country fundamentals, contagion risks and cross-country risk transmission
channels for instituting efficient EWS systems. Frankel and Saravelos (2012) present a consolidated
viewpoint emerging from the available literature on EWSs covering the period from 1950 to 2002.
The article identifies a good number of variables as robust predictors of the past currency crises. Using
these variables in an empirical model, the study attributes the causes of the global financial crisis during
2007–2009 to two key indicators: international reserves and real exchange rate overvaluation.
The other potential factors that explain the crisis include lower past credit growth, larger current
accounts, saving rates and lower external and short-term debt. Extending Rose and Spiegel (2012),
Dash 1265

Acosta-Gonzalez, Fernandez-Rodriguez and Sosvilla-Rivero (2012) attempt to select a few robust crisis
indicators from around 60 possible factors used in previous empirical studies. Of these, two financial
factors such as bank claims on private sector over deposits and domestic credit to the private sector
appear as credible warning signals of the current crisis. Overall, the factors like credit expansion, dete-
rioration in credit quality and constraints in supply of credit in the year 2006 could be listed as symptoms
of the current financial crisis. In this connection, Beckmann, Menkhoff and Sawischlewski (2006)
provide interesting insights. They profess a much broader definition of crisis to capture the fundamental
origins of the crisis. Further, they suggest inclusion of a regional contagion variable to account for the
effects of the contagion in the international financial markets. By analyzing stock prices in India, Chittedi
(2015) notices existence of a strong contagion between India and USA markets. As mentioned above, a
host of variables could be examined for generating inputs for crisis prediction. However, none of those
could suffice as the paradigms and approaches of ex-ante forecasting of financial crisis are quite diverse.
While both real and financial variables are invariably chosen in different EWS and other models, Alesi

se
and Detken (2011) find more information for crisis in the financial variables. Moreover, global financial
indicators are superior in terms of performance compared to the domestic ones. Interestingly, short-term

lu
debt, the usual suspect for the East Asian financial crisis in 1997, is not a clearly identifiable crisis vari-
able as it is viewed in the literature. Benmelech and Dvir (2013) refute this conventional wisdom and

a
rather view short-term debt as a symptom of weakness in the financial institutions than as a factor that
amplified the severity of the crisis in East Asia. ci
As mentioned before, the predictions on warning signals differ depending upon the econometric tech-
er
nique used. For instance, the study by Fuertes and Kalotychou (2012) on the forecasting performance of
m

logit regression and k-means clustering in analyzing sovereign debt crises in the emerging and develop-
ing economies observe that EWS with regard to sovereign debt are incomplete from the point of view of
om

econometric models and evaluation of their end use. Whether to rely on individual forecasts generated
by different methods separately or on combined forecasts is quite contingent upon the decision makers’
preferences. Whether to choose logit models for predicting crisis or not, Davis and Karim (2008) find
rc

evidence for the effectiveness of logit models for global EWS and signal extraction for country-specific
EWS. Similarly, Cipollini and Kapetanios (2009) ascertain the superiority of principal component analy-
fo

sis in forecasting performance vis-à-vis other models.

The above analysis suggests that crisis prediction is mostly the outcome of the interplay of a complex
ot

set of macroeconomic and financial variables in a dynamic setting. From historical experiences, it is
observed that the origin of crisis and the consequent adjustment path to equilibrium are dramatically
N

different from crisis to crisis which make prediction altogether difficult. Whether the robustness
of EWS predictions holds true or not, the onus rests on explaining the evolution of core economic
fundamentals on real time basis to capture the disorderly trends.

State of the Financial Sector in Asia

The success of any regional scheme for financial cooperation in predicting and mitigating crises depends
on a fair degree of convergence in domestic financial sector development in the region. If regional
economies follow divergent paths towards financial development, then the scope for faster harmoniza-
tion and smoother adoption of regional level policies may appear opaque and time consuming. In general,
most economies, both the crisis-affected and the resilient ones, in the developing world underwent
sweeping reforms in their financial sectors following the contagious effects of financial crises in the
1990s and early 2000s. In particular, Asian economies have undertaken widespread reforms in order to
1266 Global Business Review 18(5)

promote preventive measures and attaining financial depth. A crude look at the financial indicators such
as domestic credit to private sector, broad money, net foreign direct investment inflows, stock market
capitalization and access to international capital markets shows steady progress in regional financial
development over the period 2000–2010. Stock market capitalization that captures the level of sophistica-
tion and depth in financial markets shows a strong upward trend in the region over this period. For the East
Asia and Pacific (EAP) (developing) and South Asia regions, the ratio improved from 46.9 per cent and
26.1 per cent in 2000 to 80.4 per cent and 83.4 per cent in 2010, respectively. Likewise, a significant pro-
gress is visible in the domestic firms’ access to international capital markets signifying financial diversi-
fication in the region. With regard to the size and depth of the financial sector, Asia seems to be on the
right track. Domestic credit to the private sector has grown significantly in both the EAP (developing) and
South Asian regions, and in the economies such as China, India, Korea, Indonesia and Hong Kong.
A similar picture emerges from the trends in broad money figures. Broad money that measures liquidity
in the economy indicates encouraging trends in China, India, Korea, Malaysia, Singapore and Hong Kong.

se
Investment regimes in the Asian economies have become quite liberal with rising FDI inflows.
As the study emphasizes more on financial flows, a detailed analysis of the behaviour of financial

lu
variables is undertaken. A consistent rise in cross-border investment flows is noticed over the study
period. The volume of direct and portfolio investments continued to rise in the 2000s for most economies

a
in the Asian region. For instance, direct investment flows increased from US$44.4 billion in 2000–2005
ci
to US$ 11.5 billion in 2005–2010. Even though foreign liabilities outpaced foreign assets in BOP terms,
the greater movement of capital across borders indicates growing engagement of Asian economies in
er
financial globalization. While direct investment balance of India turned out to be positive throughout the
m

period from 1995 to 2010, Korea experienced negative balance in direct investment account during
2005–2010. The two other economies those registered negative net direct investment during 2005–2010
om

were Malaysia and Hong Kong. Besides three large markets such as China, India and Korea, the trends
in other investments especially in portfolio investments and bank flows in rest economies in the sample
substantiate a healthy pace of financial liberalization in the regional economies. Sharp reversals in port-
rc

folio investment flows, bank transactions and international reserves and large oscillations in exchange
rate movements during the past two financial crises—1997–1998 and 2007–2009—expose a greater
fo

degree uncertainty in predicting the future evolution of the financial variables.

Following the basic premise on which we envisage financial cooperation in the region, domestic
ot

financial sector development needs to be augmented by mutually reinforcing financial integration within
the region. From the Asian Development Bank (ADB) data, it is evident that financial integration has
N

assumed momentum in Asia and between its various sub-regions. Cross-border intra-regional and inter-
sub-regional direct and portfolio investment flows have deepened in the recent years. Barring FDI,
a discernible pattern of market integration is on the card. Portfolio debt and equity securities market
witnessed magnificent growth in the 2000s, from US$103 billion and US$51 billion in 2001 to US$ 258
billion and US$265 billion in 2008, respectively (Table 1).
From the above analysis, it is clear that Asian economies have deepened financial reforms in the
2000s and sincerely pushed a vigorous agenda of regional economic integration. Although there are
many possible ways in which the extent of financial integration could be measured, correlation among
regional stock indices presents the extent of pro-cyclicality and thereby the risk of transmission. High
correlation coefficients reveal that the regional stock markets have become highly integrated in the post-
1997 period compared to the pre-crisis period (Table 2). Prior to 1997, stock market integration in East
Asia was at a low level and confined to the sub-region itself. In that period, the large Asian economies
such as China, India and Korea displayed negative correlation with the East Asian economies. Unlike the
1997 crisis, Asian stock markets show a relatively high level of synchronicity both within the region and
Table 1. Intra-regional and Sub-regional Financial Integration in Asia, 2001–2010 (US$ Billion)

Portfolio Investment Assets

N Long-term Debt Short-term Total Debt Total Equity
Region/ FDI Inflows Securities Debt Securities Securities Securities
Sub-region Partner 2001 2005 2001 2008 2001 2008 2001 2008 2001 2008
ot
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Asia Asia 44.5 62.4 69.8 183.6 33.9 68.1 102.9 258.1 50.6 265.0
fo
ASEAN 8.6 7.7 14.0 24.8 1.8 6.7 15.8 31.8 13.3 25.0
ASEAN+3 26.9 39.4 36.7 75.6 12.3 47.4 48.8 128.3 31.9 182.9
South Asia 0.03 0.05 0.5 5.8 0.1 6.0 0.6 11.8 0.3 14.1
rc
ASEAN Asia 7.1 7.1 13.3 45.2 15.0 20.7 28.4 65.9 17.1 66.7
ASEAN 2.9 2.5 2.9 9.4 1.1 4.4 4.0 13.7 8.2 12.5
om
ASEAN+3 7.0 6.4 8.6 27.0 7.1 11.7 15.7 38.7 11.9 36.0
South Asia 0.03 0.03 0.5 3.9 0.1 3.6 0.6 7.5 0.3 12.4
m
ASEAN+3 Asia 34.9 44.3 43.6 125.8 18.2 22.8 61.8 148.6 28.9 128.8
ASEAN 6.1 6.3 9.0 17.6 1.5 4.9 10.5 22.5 10.2 18.6
ASEAN+3 17.3 23.4 21.2 47.3 7.5 12.3 28.7 59.6 15.1 64.8
er
South Asia 0.03 0.04 0.5 3.9 0.1 3.6 0.6 7.5 0.3 12.4
South Asia Asia 0.2 0.6 – 0.02 – – – 0.02 – 0.1
ci
ASEAN 0.1 0.3 – 0.01 –
a – – 0.01 – 0.03
ASEAN+3 0.2 0.5 – 0.01 – – – 0.01 – 0.1
lu
South Asia 0.002 0.01 – 0.01 – – – 0.01 – 0.001
Source: ADB, Integration Indicators Database (www.aric.adb.org).
se
1268 Global Business Review 18(5)

Table 2. Correlation of Stock Market Indices of Asian Economies (High Correlation, r $ 0.6)

Asia (East Asia + Advanced Economies

India, China, (USA, UK and
Country East Asia Pakistan, Sri Lanka) USA Germany)
(1) (2) (3) (4) (5)
Pre-1997 period
China High negative High positive Low Low
India High negative High Very high positive Very high positive
Korea High negative Low Very high positive Very high positive
Indonesia Low Low Very high positive Low
Thailand High Very high negative Very high negative Very high negative
Malaysia High High negative Low High negative
Philippines High High High Very high negative

se
During 1997 crisis
China Low Low Very low Low

lu
India Low Low High positive High positive
Korea Very high positive Low Low Low

a
Indonesia High positive Low Low Low
Thailand
Malaysia
Very high positive
Very high positive
Low
Low
ci Very high negative
Very high negative
Very high negative
Very high negative
er
Philippines Low Low Low Low
Post-1997 period
m

China High positive High positive High positive High positive

om

India Very high positive Very high positive High positive Low
Korea Very high positive Very high positive High positive Low
Indonesia Very high positive Very high positive Low Low
Thailand Very high positive Very high positive High positive Low
rc

Malaysia Very high positive Very high positive High positive Low
Philippines High positive High positive Low Low
fo

During 2007–2009 crisis

China Very high positive Very high positive Very high positive Very high positive
ot

India Very high positive Very high positive Very high positive Very high positive
Korea Very high positive Very high positive Very high positive Very high positive
N

Indonesia Very high positive Very high positive Very high positive Very high positive
Thailand Very high positive Very high positive Very high positive Very high positive
Malaysia Very high positive Very high positive Very high positive Very high positive
Philippines Low Low Low Low
Source: Author’s own findings based on data from IMF-IFS, September 2011.
Note: ‘High’ means mixed sign of correlation coefficients. ‘Low’ indicates insignificant value of correlation coefficients.

with the USA and other advanced economies. Apparently, this period was marked by greater degree of
capital account liberalization, harmonization of trade and investment regimes, and intense financial
transactions within the region. This implies the presence of stronger symptoms of coupling and synchro-
nicity in the financial sectors of the Asian economies. This connotes a very high risk of cross-border
spillovers of financial dislocations in both the tranquil and crisis periods.
Dash 1269

Empirical Analysis
The extent of synchronicity and interdependence among the Asian financial sectors exhibited in the
previous section needs an empirical validation so as to enable a better understanding of the behaviour of
financial variables with respect to time and the direction of causality evolving among those. Both
univariate and multivariate models are used to decipher their behaviour. Univariate analysis, particularly
unit root tests, addresses the evolution of financial series independently without taking into account the
possible influence of other variables. This requires a cross-country dimension to examine whether the
observed pattern in one economy holds in other economies or not? In this regard, panel cointegration
techniques are the most preferred econometric models. Although Pedroni (1999, 2004) is the widely used
method in the context of panel cointegration, this study employs the Westerlund (2007) error-correction
method for detecting cointegration across the panels among the five financial variables such as direct
investment, portfolio investment, bank intermediated financial flows, exchange rate and stock indices.

se
lu
Data and Econometric Methodology

a
The empirical study is based on a dataset consisting of panels representing eight sample economies from
ci
the Asian region and the time period spanning from 1990:Q1 to 2011:Q2. The econometric methodology
involves two steps. In the first stage, the time series filters, namely, Hodrik–Prescottt (HP) (see Hodrick
er
& Prescott, 1997) and Butterworth rational signal wave, are used to identify cyclical fluctuations in the
m

variables. Graphically, the smoothed series generated by these filters exhibit substantial degree of
cyclical movements in the indicators over the study period. As trend in a series captures the outcome of
om

long-term growth factors, the series including trend may not reveal the effects of transient factors that
might reflect warning signals. It, therefore, requires detrending the series before testing for cointegration
among them. In that spirit, the second stage of analysis involves application of the Westerlund error
rc

correction panel cointegration method to the residual series generated by the filters. Filters are applied
separately to all the five series for the full period. Both the filters show substantial volatility in the sample
fo

series around trend. It substantiates our presumption that macroeconomic indicators idiosyncratically
portray very high magnitude of variability over time. It prompts us to extend the argument a step further
ot

by examining whether those observed fluctuations are correlated across the economies in the region.
This exercise is expected to shed information on the probable risk of contagion and the suitable pre-
N

emptive policy measures need to be undertaken in event of such disequilibriating tendency. As discussed
later, there exists cointegration among all the five variables and their various combinations for the full as
well as various sub-periods over the period 1990: Q1 to 2011: Q2.
Features of Filtering Techniques
HP filter is based on the prior knowledge that the growth component varies smoothly over time. The
nature of the comovements of the cyclical components of macroeconomic time series is very different
from the comovements of the slowing varying components of the corresponding variables. The concep-
tual framework of the filter is as the following. A given time series is represented as the sum of the
growth component and the cyclical component:

yt = gt + ct

where gt = growth component and ct = cyclical component.

1270 Global Business Review 18(5)

This filtering technique assumes that the smoothness of the ‘gt’ path is the sum of the squares of its
second difference and ‘ct’ is the deviation from ‘gt’ and the average of ‘ct’ is zero over long period.
Accordingly, the growth component is determined by the following minimization programme:
T T
Min " g ,Tt =- 1 ( | c 2t + m | 6^ g t - g t - 1h - ^ g t - 1 - g t - 2h@2 2(1)
i=1 t=1

Butterworth rational wave filter is a linear filtering technique adapted with the signal processing tech-
niques. As observed in Pollock (2000), HP filter possesses only a single adjustable parameter that shows
both the rate of transition and the location of the nominal cut-off point. It fails to impede the powerful
low-frequency components from passing into the detrended series, and is likely to induce spurious cycles
in the smoothed series. In contrast, Butterworth rational wave filter is a model-based method that postu-
lates a trend which is the product of an integrated moving average process of a lower order which has no

se
firm upper limit frequency range. The mechanism of the Butterworth filter is represented as follows.
If y(t) = {yt ; t = 0, ±1, ±2,…..} represent a discrete time signal, then the two filtering operations can

lu
be described by the following equations:

a
c (L) q (t) = d (L) y (t)(2a)
ci
c (F) x (t) = d (F) q (t)(2b)
er
where q(t) = intermediate output from the forward pass of the filter; x(t) = final output generated in the
m

backward pass; γ(L) and δ(L) = polynomials of the lag operator; γ(F) and δ(F) = inverse forward shifts
om

operator.
The filter must satisfy the condition of stability that requires roots of the polynomial equation to lie
outside the unit circle. The two filters can be combined to form a symmetric two-sided rational filter:
rc

d ^ F h d ^ Lh
W ^ Lh = (3)
fo

c ^ F h c ^ Lh
ot

Westerlund Error Correction Model

Westerlund (2007) proposes four new panel tests those are based on structural rather than residual
N

dynamics. Two tests are designed to test the alternative hypothesis that the panel is cointegrated as a
whole, while the other two test the alternative that there is at least one individual among the panel units
that is cointegrated.
Given the following data generating process

y it = z 1i + z 2i t + z it(4a)

x it = x it - 1 + v it(4b)

where t = time index n = cross-sectional units.

From this, the conditional error correction model for ‘yit’ is given as

a i (L) Dy it = d 1i + d 2i t + a i (y it - 1 - bli x it - 1) + c i (L)l v it + e it(5)

Dash 1271

1. If ai < 0, there is cointegration between ‘yit’ and ‘xit’, error correction is present.
2. If ai = 0, there is no cointegration between ‘yit’ and ‘xit’, error correction is absent.

Equation (5) can be rewritten as:

pi pi

Dy it = dli d t + a i ^ y it - 1 - bli x it - 1h + | a ij Dy it - j + | c ij Dx it - j + e it(6)

j=1 j=0

where dt holds the deterministic components.

The parameter of interest to estimate is ‘ai’ which can be estimated in several ways. One way to esti-
mate ‘ai’ is by assuming that ‘bi’ are known. However, these tests are not similar and depend on nuisance
parameters. Alternatively, Equation (6) can be represented as:
pi pi

Dy it = dli d t + a i y it - 1 + mli x it - 1 + | a ij Dy it - j + | c ij Dx it - j + e it(7)

se
j=1 j=0

Estimating Equation (7) by least squares, the four statistics is computed as follows:

lu
1 N at i
| (8a)

a
N i = 1 SE ^at ih

1 N Tat i
ci
er
|
N i = 1 at i ^1h
(8b)
m

at
om

PT = (8c)
SE ^at h
rc

Pa = Tat (8d)
fo

Panel Unit Root Results

ot

Results of the panel unit root tests clearly show that four out of the five variables are integrated of order
N

one except the exchange rate series. Since cointegration tests are applied to the full period as well as for
various sub-periods, the unit root tests are carried out for sub-periods also. Interestingly, exchange rate
series appears non-stationary in level for the sub-periods 2001–2006, 2007–2011 and 2001–2011.
Barring exchange rate, divergent unit root results are noticed for different sub-periods for bank transac-
tions and stock indices (Table 3). In addition, some variables are found to have higher unit roots. Even
though contradictory results are observed for bank flows and stock indices, the results for the exchange
rate series warrant attention. Exchange rate series is found stationary for the full period is considered and
non-stationary for certain sub-periods.

Discussion of Empirical Results

Panel cointegration tests conducted on different combinations of the four economic series such as direct
investment, portfolio investment, bank transactions and stock indices for the full period and various
1272 Global Business Review 18(5)

Table 3. Panel Unit Root Results

Im–Pesaran–Shin
Fisher Type (ADF) (Lag–2) (IPS) (Lag–2)
Inverse Chi– Inverse Inverse Modified Inverse
Variable squared (P) Normal (Z) Logit t (L*) Chi–squared (Pm) W–t–bar
(1) (2) (3) (4) (5) (6)
di2 5.7 1.8 1.8 –1.3 1.7
Ddi2 33.1*** –3.2*** –3.4*** 4.3*** –3.1***
port2 14.6 –0.1 0.2 0.5 0.1
Dport2 39.8*** –4.1*** –4.4*** 5.7*** –3.9***
bank2 13.5** 0.02 0.3 0.3 0.2
Dbank2 40.2*** –3.4*** –3.7*** 5.7*** –3.1***
ex2 29.8*** –1.8** –2.5** 3.6*** –1.9**

se
ex2 (2001–06) 3.2 3.5 3.7 –1.8 3.1
Dex2 (2001–06) 15.1 0.1 0.2 0.6 0.004

lu
D2ex2 (2001–06) 30.9*** –1.5* –1.99** 3.9*** –1.4*
ex2 (2007–11)a 8.7 0.7 0.7 –0.7 0.3

a
ex2 (2001–11) 7.9 1.99 2.2 –0.8 2.0
Dex2(2001–11)
stock2
27.9***
0.2
–2.6***
5.6
ci–2.8***
6.6
3.2**
–2.2
–2.5***
7.1
er
Dstock2 20.4** –2.4** –2.3** 2.3** –2.3**
Source: Author’s own findings.
m

Notes: Prefix ‘D’ and ‘D2’ stand for variables in first and second differences, respectively. ‘***’, ‘**’ and ‘*’ denote 1 per cent,
om

5 per cent and 10 per cent level of significance, respectively. Consistent results are observed for sub-samples: 1990–
1996, 1997–2000, 2001–2006, 2007–2011, 1997–2011 and 2001–2011 barring these exceptions: ex2 (non-stationary
for the periods 2001–2006, 2007–2011 & 2001–2011 (both Fisher-ADF and IPS)), bank2 (stationary for the period
1997–2000) and stock2 (stationary for the periods 1997–2000 and 2007–2011 (both Fisher-ADF and IPS). anon-station-
rc

ary after second difference also.

fo

sub-periods uniformly suggest comovement among the variables. The results hold well even by includ-
ing exchange rate as one of the cointegrating variables for the sub-periods 2001–2006, 2007–2011 and
ot

2001–2011. For the full period, all the four statistics—Gt, Ga, Pt and Pa—are found highly significant
showing strong presence of cointegration in the panels. Unlike the full sample, the values of these four
N

statistics are divergent for the sub-period estimations substantiating our hypothesis of regime change
during the study period. It implies that the regional financial sectors remained less integrated in the pre-
crisis (1997) period. This is evident from the cointegration results for the period 1990–1996. Except in
one case, that is, the combination of portfolio investment, bank flows and stock indices, the statistics are
found significant for all other variable combinations in this period (Tables 4 and 5).
As observed, the gradual improvement in coefficient significance of the four test statistics rightly
mimics the slower process of sequencing of financial reform measures those were implemented in
regional economies over the past two decades. Unlike the period 2001–2006, the evidence of cointegra-
tion across panels is weak in the period 2007–2011. Since almost all test statistics are highly significant
during the period 2001–2011, it can be safely generalized that financial integration in the region in the
post-crisis period especially after 2000 present encouraging trends. This implies that the transmission of
risks and shocks has become faster in the 2000s than the pre-crisis period. This is in conformity with the
number of financial crises occurred in different parts of the world in the post-2000 period and its varying
origins and severity.
Dash 1273

Table 4. Panel Cointegration Results (Westerlund (2007) Error Correction Model)

Specification Lags, Leads 1990–2011 1990–1996 1997–2000 2001–2006 2007–2011 1997–2011 2001–2011
(1) (2) (3) (4) (5) (6) (7) (8) (9)
di, port, bank, 1,1 √ × × Gt, Pt × Gt, Pt Gt, Pt
stock
di, port, bank, 1,0 √ × Gt, Pt Gt, Pt × Gt, Pt, Pa Gt, Pt, Pa
stock
di, bank, stock 1,1 √ × Gt, Pt Gt, Pt Gt, Pt Gt, Pt, Pa Gt, Pt, Pa
di, bank, stock 1,0 √ × Gt, Pt Gt, Pt Pt √ Gt, Pt, Pa
port, bank, stock 1,1 √ × Gt Pt × √ Gt
port, bank, stock 1,0 √ Gt Gt × √ Gt, Ga
di, port, bank, 1,1 – Gt – × × – Gt, Pt
ex, stock

se
di, port, bank, 1,0 – – – Gt × – Gt, Pt
ex, stock

lu
di, bank, ex, stock 1,1 – – – Gt × – Gt, Pt, Pa
di, bank, ex, stock 1,0 – – – Gt, Pt × – Gt, Pt, Pa

a
port, bank, ex, 1,1 – – – Gt Gt – Gt, Pt
stock
port, bank, ex, 1,0 – – – ci Gt Gt – Pt, Pa
er
stock
Source: Author’s own findings.
m

Notes: ‘√’ represents that all the four statistics are significant. ‘X’ denotes absence of cointegration in the panels. As exchange
om

rate (ex2) was found non-stationary only for the three sub-periods (2001–2006, 2007–2011 and 2001–2011), cointegra-
tion test was attempted for these three periods only.
rc

Table 5. Panel Cointegration Estimates

fo

Group Mean Statistics Panel Statistics

Lags, Gt Ga Pt Pa
ot

Specification Leads Value Z–Value Value Z–Value Value Z–Value Value Z–Value
N

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Full Period (1990–2011)
di, port, bank, stock 1,1 –4.4 –5.0*** –15.9 –1.6* –9.4 –4.8*** –13.6 –2.1**
di, port, bank, stock 1,0 –5.6 –8.0*** –16.8 –1.8** –10.4 –5.8*** –15.8 –2.8***
di, bank, stock 1,1 –3.5 –3.7*** –16.6 –2.7*** –9.5 –5.5*** –15.0 –3.7***
di, bank, stock 1,0 –4.3 –5.5*** –18.9 –3.5*** –10.5 –6.5*** –18.0 –4.9***
port, bank, stock 1,1 –4.2 –5.4*** –19.6 –3.7*** –8.8 –4.8*** –16.0 –4.0***
port, bank, stock 1,0 –4.4 –5.8*** –21.7 –4.5*** –8.5 –4.5*** –17.6 –4.7***
Sub–period (2001–2006)
di, port, bank, stock 1,1 –7.6 –12.8*** –2.7 –2.6 –10.8 –6.2*** –2.0 –1.9
di, port, bank, stock 1,0 –8.1 –13.9*** –6.0 –1.6 –12.3 –7.6*** –6.6 –0.3
di, bank, stock 1,1 –6.7 –11.2*** –4.1 –1.8 –19.3 –15.0*** –4.6 –0.5
di, bank, stock 1,0 –5.7 –8.8*** –6.8 –0.8 –13.1 –8.9*** –6.8 –0.4
port, bank, stock 1,1 –2.4 –0.9 –2.8 2.3 –6.3 –2.4*** –3.5 0.9
(Table 5 continued)
1274 Global Business Review 18(5)

(Table 5 continued)

Group Mean Statistics Panel Statistics

Lags, Gt Ga Pt Pa
Specification Leads Value Z–Value Value Z–Value Value Z–Value Value Z–Value
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
port, bank, stock 1,0 –2.9 –2.0** –5.0 1.5 –4.1 –0.2 –5.1 0.3
di, port, bank, ex, stock 1,1 –1.4 2.4 –0.4 3.7 –1.8 2.8 –0.4 2.6
di, port, bank, ex, stock 1,0 –4.4 –4.6*** –3.1 2.9 –3.7 1.0 –2.9 1.9
di, bank, ex, stock 1,1 –6.9 –11.0*** –0.9 3.2 –0.9 3.3 –0.5 2.4
di, bank, ex, stock 1,0 –4.9 –6.2*** –5.9 1.6 –11.3 –6.6*** –6.1 0.5
port, bank, ex, stock 1,1 –12.0 –23.2*** –2.1 2.8 –4.8 –0.4 –1.8 1.9
port, bank, ex, stock 1,0 –5.1 –6.7*** –4.6 2.0 –1.3 2.9 –3.6 1.3

se
Sub–period (2001–2011)
di, port, bank, stock 1,1 –5.6 –8.0*** –9.0 0.6 –11.5 –6.8*** –8.7 –0.4

lu
di, port, bank, stock 1,0 –6.3 –9.6*** –11.4 –0.1 –9.4 –4.8*** –11.3 –1.3*
di, bank, stock 1,1 –3.4 –3.4*** –9.3 –0.1 –7.1 –3.2*** –9.6 –1.5

a
di, bank, stock 1,0 –4.7 –6.4*** –11.4 –0.8 –8.1 –4.2*** –11.6 –2.3**
port, bank, stock 1,1 –3.3 –3.0*** –10.8 –0.6 –4.9 –1.1 –8.5 –1.0
port, bank, stock 1,0 –3.4 –3.4*** –13.3ci –1.5* –4.7 –0.8 –7.7 –0.7
er
di, port, bank, ex, stock 1,1 –5.7 –7.7*** –8.1 1.4 –12.2 –6.8*** –7.9 0.4
di, port, bank, ex, stock 1,0 –5.3 –6.6*** 11.3 0.5 –11.8 –6.5*** –11.1 –0.5
m

di, bank, ex, stock 1,1 –4.8 –6.1*** –11.3 –0.1 –8.8 –4.3*** –11.4 –1.4*
di, bank, ex, stock 1,0 –4.5 –5.5*** –12.0 –0.3 –8.4 –3.8*** –12.5 –1.7**
om

port, bank, ex, stock 1,1 –3.1 –2.2** –10.5 0.2 –6.2 –1.8** –9.2 –0.6
port, bank, ex, stock 1,0 –2.5 –0.6 –10.5 0.1 –8.4 –3.9** 13.4 –2.0**
Source: Author’s own findings.
rc

Notes: Based on the robustness of the results these three periods are chosen. ‘***’, ‘**’ and ‘*’ show significance at 1 per cent,
5 per cent and 10 per cent level of significance.
fo

Conclusion
ot

Whether financial crises are predictable or not, the endeavour to predict and to correct the observed flaws
N

in crisis prevention mechanisms has surfaced prominently in the post-crisis literature. This article makes
a small contribution to the discourse by reviewing literature on crisis prediction with an empirical note
on the extent of synchronicity observed in the eight Asian economies over the period from 1990 to 2011.
Empirical findings show the presence of panel cointegration among the five financial variables such as
direct investment, portfolio investment, bank transactions, exchange rate and stock index. It suggests
that financial integration in Asia remained shallow in the period prior to the East Asian crisis in 1997.
With deepening of trade integration in the 2000s, the region witnessed relatively higher pace of financial
integration which was reflected in a stronger degree of synchronicity in the post-crisis period especially
during 2001–2011. These findings seem to be consistent with the gradualist experimental path followed
towards financial liberalization in the sample economies over the past two decades. Evidence of comove-
ment in the behaviour of financial indicators signals contagion risk in Asia and establish the case for
effective EWSs which will aid the policymakers for timely detection of crisis signals. Further, early
warning models experimented in the past have played a crucial role in producing key information on the
macroeconomic and financial state of the economies. However, these models largely failed in predicting
Dash 1275

the actual crisis episodes. Given this limitation, there lies sufficient economic rationale for developing
suitable EWS in the region, presumably under a regional policymaking supervisory framework.
The finding of this article has implications for policy and businesses as the risk of contagion is present
in the financial sectors of the region. Investors having multi-country exposure are expected to behave
like herds in the event of financial disruptions in any of the integrated markets in the region. It implies
that outflow of capital in one country would tempt the investors to pull out of other connected markets
in the region as economic fundamentals in those economies exhibit extraordinary synchronicity.
Likewise, the central bankers would face liquidity crunch as banks and equity markets would respond
equally to the news of financial crisis. Firms need to hedge their exposure to foreign markets in view of
the growing interdependence among the financial markets in the region.
Despite substantive contribution to explain the interrelationship among certain macroeconomic vari-
ables, the article has some limitations. The article does not take into real variables. All the model varia-
bles are mostly financial sector variables. By including real variables, the article could have deciphered

se
the complex interactions among the variables for yielding better signals about crises. Other models of
EWS types can be tried for producing more meaningful results about crises.

lu
Acknowledgement

a
ci
The author is grateful to the anonymous referees of the journal for their extremely useful suggestions to improve the
quality of the article. Usual disclaimers apply.
er
Note
m

1. A financial crisis occurs if at least any of these four elements or combinations(s) of those such as sharp deprecia-
om

tion of currency, large drop in stock market indices, disruptions in financial systems and reversal of GDP growth
takes place (see Bustelo, 2000).
rc

Appendix
fo

Data and Variable Definitions

ot

Primary Transformed Data Series

N

Variable Definition Frequency Variable Definition Source Code

(1) (2) (3) (4) (5) (6) (7)
port Net portfolio Quarterly port2 Residual series generated IFS 78BFDZF
investment by subtracting smoothed
flows ‘Butterworth port series’
from ‘port’
bank Net bank- Quarterly bank2 Residual series generated IFS 78BQDZF
intermediated by subtracting smoothed
flows ‘Butterworth bank series’
from ‘bank’
ex National Quarterly ex2 Residual series generated IFS AE.ZF
exchange rate by subtracting smoothed
per US dollar ‘Butterworth ex series’
(end of period) from ‘ex’
(Appendix continued)
1276 Global Business Review 18(5)

(Appendix continued)

Primary Transformed Data Series

Variable Definition Frequency Variable Definition Source Code
di Net direct Quarterly di2 Residual series generated IFS 78BDDZF
investment by subtracting smoothed
flows ‘Butterworth di series’
from ‘di’
stock Share price Quarterly stock2 Residual series generated IFS 62ZF
indices by subtracting smoothed
‘Butterworth stock series’
from ‘stock’
Source: Author’s own analysis.

se
References
Acosta-Gonzalez, E., Fernandez-Rodriguez, F., & Sosvilla-Rivero, S. (2012). On factors explaining the 2008 finan-

lu
cial crisis. Economics Letters, 115(2), 215–217.
ADB. (n.d). Integration Indicators Database. Retrieved from www.aric.adb.org

a
Alesi, L., & Detken, C. (2011). Quasi real time early warning indicators for costly asset price boom/bust cycles:
ci
A role for global liquidity. European Journal of Political Economy, 27(3), 520–533.
Ali, L., & Kestens, Y. (2006). Contagion and crises clusters: Toward a regional warning system? Review of World
er
Economics, 142(4), 814–839.
Alvarez-Plata, P., & Schrooten, M. (2004). Misleading indicators? The Argentinean currency crisis. Journal of
m

Policy Modeling, 26(5), 587–603.

Beckmann, D., Menkhoff, L., & Sawischlewski, K. (2006). Robust lessons about practical early warning systems.
om

Journal of Policy Modeling, 28(2), 163–193.

Benmelech, E., & Dvir, E. (2012). Does short-term debt increase vulnerability to crisis? Evidence from the East
Asian financial crisis. Journal of International Economics, 89(2), 485–494.
rc

Berg, A., Borensztein, E., & Pattillo, C. (2005). Assessing early warning systems: How have they worked in prac-
tice. IMF Staff Papers, 52(3), 462–502.
fo

Bordo, M., Eichengreen, B., Klingebiel, D., & Martinez-Peria, M.S. (2001). Is the crisis problem growing more
severe? Economic Policy, 16(32), 53–82.
Bustelo, P. (2000). Novelties of financial crises in the 1990s and the search for new indicators. Emerging Markets
ot

Review, 1(3), 229–251.

N

Chittedi, K.R. (2015). Financial crisis and contagion effects to Indian stock market: ‘DCC–GARCH’ analysis.
Global Business Review, 16(1), 50–60.
Cipollini, A., & Kapetanios, G. (2009). Forecasting financial crises and contagion in Asia using dynamic factor
analysis. Journal of Empirical Finance, 16(2), 188–200.
Davis, E.P., & Karim, D. (2008). Comparing early warning systems for banking crises. Journal of Financial Stability,
4(2), 89–120.
Frankel, J., & Saravelos, G. (2012). Can leading indicators assess country vulnerability? Evidence from the 2008–09
global financial crisis. Journal of International Economics, 87(2), 216–231.
Fuertes, Ana-Maria, & Kalotychou, E. (2007). Optimal design of early warning systems for sovereign debt crises.
International Journal of Forecasting, 23(1), 85–100.
Hodrick, R.J., & Prescott, E.C. (1997). Postwar U.S. business cycles: An empirical investigation. Journal of Money,
Credit and Banking. 29(1), 1–16.
IMF. (2011). International Financial Statistics (CD-ROM).
Krugman, P. (1979). A model of balance-of-payments crises. Journal of Money, Credit and Banking, 11(3),
311–325.
Ozlale, U., & Metin-Ozcan, K. (2007). An alternative method to measure the likelihood of a financial crisis in an
emerging market. Physica A, 381, 329–337.
 Dash 1277

Pedroni, P. (1999). Critical values for cointegration tests in heterogeneous panels with multiple regressors. Oxford
Bulletin of Economics and Statistics, 61(S1), 653–670.
Pedroni, P. (2004). Panel cointegration: Asymptotic and finite sample properties of pooled time series tests with an
application to the PPP hypothesis. Econometric Theory, 20, 597–625.
Peng, D., & Bajona, C. (2008). China’s vulnerability to currency crisis: A KLR signals approach. China Economic
Review, 19(2), 138–151.
Pollock, D.S.G. (2000). Trend estimation and de-trending via rational square-wave filters. Journal of Econometrics,
99(2), 317–334.
Rose, A.K., & Spiegel, M.M. (2012). Cross-country causes and consequences of the 2008 crisis: Early warning.
Japan and the World Economy, 24(1), 1–16.
Salant, S., & Hendersen, D. (1978). Market anticipations of government gold policies and the price of gold. Journal
of Political Economy, 86(4), 627–648.
Westerlund, J. (2007). Testing for error correction in panel data. Oxford Bulletin of Economics and Statistics, 69(6),
709–748.

se
a lu
ci
er
m
om
rc
fo
ot
N

View publication stats