Academy of Economic Studies
Doctoral School of Finance and Banking
Volatility Spillovers and Financial
Contagion in the CEE Stock Markets
MSc. Student: Țânțaru Mihai
Supervisor: Professor PhD. Moisă Altăr
Summary






Introduction
Methodology
Data description
Estimation results
Conclusions
References
Introduction




The spread of crises throughout the financial system at the
global or regional level has been (loosely) defined as contagion.
Despite the large interest in the subject, there is no generally
accepted definition for contagion.
The implications of contagion in the pricing of risk and for
financial regulators are of outmost importance.
The methodologies employed in the scientific literature vary with
the definitions for contagion:


Spillovers in return and volatility across financial markets – modeled with
simple GARCH models in Engle et al. (1988), Hamao et al. (1990), or
multivariate GARCH models as in Beirne et al. (2008).
Restrictive definition – change in the cross-market shock transmission
mechanism that takes place during crises – study of cross-market
correlation coefficients: King and Wadhwani (1990), Forbes and Rigobon
(2002), Dungey et al. (2005).
Introduction


In the light of Bekaert, Harvey and Ng (2005), this study adopts
the restrictive definition of contagion as “correlation over and
above what one could expect from economic fundamentals”.
Motivations of this study:



To develop a model that correctly accounts for the cross-market
fundamental linkages, and therefore, gives an accurate description of the
cross-market volatility transmission mechanism.
To verify to what extent does the model choice influence contagion test
results.
I construct a two-factor spillover model for the CEE stock
markets, with global (US) and regional (European) risk loadings:


It distinguishes between regional and global market integration.
It outperforms the one-factor model in modeling cross-market
correlations – Bekaert et al. (2008).
Methodology
1. The Bivariate Global – Regional Specification

The framework for the joint process of US and EU returns:
 r  μ ε
rEU ,t ) - return vector
t
t
t with rt  (rUS ,t
 US ,t 
 μ 
 - expected mean: lagged information variables, US and EU
t


 EU ,t  returns.
  US ,t 

 ~ N (0, H t ) - vector of unexpected returns.
 ε |
t
t 1  

  EU ,t 
 H  CC  Aε ε A  BH B - joint conditional variance-covariance
t
t 1 t 1
t 1
process specified by Engle and
Kroner’s bivariate BEKK(1,1).
 The orthogonalization process to obtain the US and EU idiosyncratic shocks:
2
  US
0 
 εUS ,t   1 0  eUS ,t 
,t


  K t 1e t, with et | t 1 ~ N (0, Σt ) , Σ t  

ε t  
2





 EU ,t 
  EU ,t   k t 1 1  eEU ,t 
 0
kt 1  hUS ,EU ,t / hUS ,t
Methodology
2. The Univariate Volatility Spillover Model


General model for the return of CEE stock market index i, at time t:
 ri ,t  i ,t   i ,t , with  i ,t | t 1 ~ N (0, hi ,t )

i,t - conditional mean: (lagged) US return or local dividend yield.

EU
- unexpected return composed of global,
ˆ
ˆ
 i,t  iUS
,t eUS ,t  i ,t eEU ,t  ei ,t
regional and local idiosyncratic
shocks.
( EU the
)
US ( EU )
The restricted models
for
risk factor exposure or ‘beta’:
iUS
 global/regional
,t
i ,0
 Constant ‘beta’ :


( EU )
US ( EU )
US ( EU )
iUS


X
,t
i
i ,t
( EU )
X iUS
,t
Structural ‘beta’ :
, with
- a trade integration
measure as in Bekaert et al. (2005).
( EU )
US ( EU )
Si ,t
iUS


(Si,t )
,t
i ,0
Regime-switching ‘beta’:
, with
- a latent regime
variable as in Baele (2005).
Methodology
2. The Univariate Volatility Spillover Model

This study employs the flexible ‘beta’ specification as in Baele et al. (2010):
( EU )
( EU )
( EU )
iUS
 iUS
(Si,t )  iUS ( EU ) X iUS
,t
,0
,t
where:



X
US ( EU )
i ,t

IMPi ,US ( EU ),t  EXPi ,US ( EU ),t
( IMPi ,t  EXPi ,t )TOTAL
US ( EU )


, S i ,t  1
i
, S ,1

US ( EU )
 i ,S
( S i ,t )   US ( EU )

  i , S , 2 , S i ,t  2
in
- structural economic instrument that
reflects time-varying integration measure.
-
regime-switching component that
reflects temporary fluctuations
financial markets conditions.
The latent regime variable Si ,t follows a Markov chain process with constant
transition probabilities:
Pi  prob(Si,t  1| Si,t 1  1) and Qi  prob(Si ,t  2 | Si,t 1  2.)
Methodology
2. The Univariate Volatility Spillover Model

When the spillover model for the individual market i:
EU
ˆ
ˆ
ri,t  i,t  iUS
ei ,t | t 1 ~ N (0,  i2,t )
,t eUS ,t  i ,t eEU ,t  ei ,t
entertains regime-switching component in the market ‘betas’, then:
2

N
(
0
,

i , S ,1 ), S i ,t  1

2
2
2
 Case 1: ei ,t ~ 
and


p


(
1

p
)

i
,
t
i
,
1
,
t
i
,
S
,
1
i
,
1
,
t
i
,S , 2
2
N
(
0
,

),
S

2

i ,S , 2
i ,t

pi ,1,t  prob(Si ,t  1| t 1 )
Case 2:  i2,t  i  i ei2,t 1   i i2,t 1 - GARCH(1,1) variance process.
The estimation of the regime-switching specification is done through the
maximization of the sample log-likelihood function:


T
log f (ri ,1 , ri , 2 ,...,ri ,T | ri ,0 ; i )   log f (ri ,t | t 1 ; i ) 
t 1


  log prob(Si ,t  k | Si ,t 1  l ) * prob(Si ,t 1  l | t 1 ; i ) * f (ri ,t | Si ,t  k , t 1 ; i )
t 1
 k 1 l 1

T
2
2
Methodology
3. Variance Ratios and Conditional Correlations


The depicted models are complete with the assumption: E(ei ,t eUS ( EU ),t )  0
of zero correlation between the local idiosyncratic shocks and US/EU
specific innovations.
The total conditional variance of market i can be decomposed:
2 2
EU 2 2
2
E( i2,t | t 1 )  hi2,t  (iUS
,t )  US ,t  ( i ,t )  EU ,t   i ,t
2
E( i,t eˆUS ,t | t 1 )  hi,US ,t  iUS
,t US ,t
2
E( i,t eˆEU ,t | t 1 )  hi,EU ,t  iEU
,t  EU ,t

Variance ratios and conditional correlations are given by:
US 2
2
(

)

i ,t
US ,t
US 2

VRiUS



,t
i ,t 
2
hi ,t
VR iEU
,t 
2
2
(  iEU
,t )  EU ,t
hi2,t

  iEU
,t

2
Methodology
4. The Contagion Test


An unconditional correlation ˆi ,US ( EU )  0 (over the full sample) does not
guarantee that there has not been contagion across some episodes of time.
The following specification is estimated to test for any remaining correlation,
separately for each market and through a panel regression:
eˆi ,t  wi  vUS ,t eˆUS ,t  vEU ,t eˆEU ,t  ui,t
where:
vUS ,t  vUS ,0  vUS ,1 * Dt
vEU ,t  vEU ,0  vEU ,1 * Dt

Dt represents a dummy to account for crisis (high volatility) periods in the
global/regional equity markets.

Significant vUS ,1 , vEU ,1 parameters signal contagion.
Data description

All data spans between Jan 2005 – Mar 2010, 262 weekly (Tue) observations.

Equity market data: returns of the S&P500 for the global market, MSCI Europe for
the regional market and of the most liquid stock market indices in Romania (BET),
Hungary (BUX), Poland (WIG) and Czech Republic (PX) for the CEE markets.

Information variables : CDS prices for CEE 5Y sovereign debt and EUR/CEE
currencies exchange rates; (first difference of) US default spread, TED spread, US 10Y
Treasury Bond yield, local dividend yields.

Structural data : the sum of imports and exports between an individual country and
US/EU divided by the sum of the total imports and exports for that country.

The crisis dummy equals 1 during periods:
 the peak of the recent global economic and financial crisis between Sep 2008 and
the beginning of May 2009, when VIX volatility index was more than 1 std. dev.
above the sample mean;
 when both the S&P and MSCI returns were 1 std. dev. below the sample mean.
Estimation Results
1.The US and EU joint specification
The BEKK(1,1) model results:

S&P (i=1)
Coefficient p - value
Mean Equations
C
S&P(-1)
MSCI(-1)
Div Yield MSCI
Default Spread
Treasury 10 Y
Variance Equations
C1i
C2i


MSCI E (i=2)
Coefficient p - value
0.0001
-0.1258
-0.0721
0.0234
0.9473
0.0487
0.0016
0.0542
0.0004
0.1385
-0.0980
-0.1450
-
0.6394
0.0079
0.0424
0.0000
-
0.0015
0.0055
0.5268
0.0008
0.0000
0.9985
A1i
A2i
-0.1995
-0.2683
0.2041
0.4322
0.1663
-0.4722
0.0309
0.0012
B1i
B2i
0.9502
-0.0105
0.0000
0.9562
0.0524
0.7837
0.0402
0.0000
Specification tests:
S&P
Q -stat
p - value
Standardized residuals
lag 6
2.11
0.91
lag 12
9.21
0.69
Squared standardized residuals
lag 6
6.77
0.34
lag 12
8.64
0.73

MSCI E
Q -stat
p - value
5.45
16.71
0.49
0.16
1.89
7.99
0.93
0.79
The Ljung-Box tests find that
no autocorrelation remains in
the (squared) standardized
residuals of BEKK(1,1)
model.
There are significant unidirectional news and volatility spillovers from the US
market to the aggregate European equity market.
Estimation Results
2. The Dynamic Factor Regime-Switching Models

The orthogonalized US and European residuals are plugged as components
in the unexpected returns of the individual CEE indices.

The various specifications of market ‘betas’ are tested for statistical
significance:



For all CEE indices, the model with constant ‘betas’ and the model with timevarying structural ‘betas’ are statistically valid.
When the most flexible ‘beta’ specification as proposed by Baele and Inghelbrecht
(2010) does not fit the data, less-complex specifications are employed, at least
one factor loading involving a regime-switching component.
The specification tests on the models with regime-switching are Ljung-Box
tests on the generalized (regime-independent) residuals as in Smith (2007).
The Hansen (1992, 1996) standardized LR test is employed for the general
validity of the switching hypothesis.
Estimation Results
2. The Dynamic Factor Regime-Switching Models

Romania BET index
Coefficient
Mean parameters
S&P(-1)
Beta EU - constant
Beta US State 1
Beta US State 2
Regime probabilities
P
Q
Regime variances
Sigma State 1
Sigma State 2

p - value
0.4586
0.6391
0.9592
0.6765
0.00
0.03
0.00
0.00
0.9938
0.9494
0.00
0.09
0.0012
0.0052
0.00
0.04
Specification tests
BET
Statistic
p - value
Ljung-Box for Standardized residuals
lag 6
4.2
0.65
lag 12
11.64
0.48
Ljung-Box for Squared standardized residuals
lag 6
6.17
0.40
lag 12
9.79
0.63
Hansen Standardized LR
Newey-West lag 0
2.88
0.03

EU
EU
EU market ‘beta’:  BET


,t
BET , 0

US
US
US market ‘beta’: BET
,t   BET, S (SBET,t )
Estimation Results
2. The Dynamic Factor Regime-Switching Models

Poland WIG index
Mean parameters
Div yield WIG
Beta EU - constant
Beta US - structural
Beta US State 1
Beta US State 2
Regime probabilities
P
Q
Regime variances
Sigma State 1
Sigma State 2

Coefficient
p - value
-0.0335
1.0135
-1.2655
2.0868
1.8515
0.00
0.00
0.08
0.00
0.01
0.9589
0.8557
0.00
0.00
0.0005
0.0025
0.00
0.00
Specification tests
WIG
Statistic
p - value
Ljung-Box for Standardized residuals
lag 6
5.79
0.45
lag 12
14.16
0.29
Ljung-Box for Squared standardized residuals
lag 6
5.71
0.46
lag 12
11.02
0.53
Hansen Standardized LR
Newey-West lag 0
2.77
0.04

EU
EU
EU market ‘beta’: WIG


,t
WIG , 0

US
US
US
US
US market ‘beta’: WIG
,t  WIG , S (Si ,t )  WIG X POL ,t
Estimation Results
2. The Dynamic Factor Regime-Switching Models

Czech Republic PX index
Coefficient
Mean parameters
S&P(-1)
Beta EU - structural
Beta US State 1
Beta US State 2
Regime probabilities
P
Q
Regime variances
Sigma State 1
Sigma State 2

p - value
0.2893
0.8906
0.7774
1.2112
0.00
0.00
0.00
0.00
0.9753
0.7540
0.00
0.00
0.0004
0.0094
0.00
0.00

EU
EU
EU
EU market ‘beta’: PX
,t   PX X CZ ,t

US
US
US market ‘beta’:  PX
,t   PX , S (S PX ,t )
Specification tests
PX
Statistic
p - value
Ljung-Box for Standardized residuals
lag 6
11.65
0.07
lag 12
13.78
0.32
Ljung-Box for Squared standardized residuals
lag 6
2.65
0.85
lag 12
4.93
0.96
Hansen Standardized LR
Newey-West lag 0
5.67
0.00
Estimation Results
2. The Dynamic Factor Regime-Switching Models

Hungary BUX index
Mean parameters
Div yield BUX
Beta EU - structural
Beta US State 1
Beta US State 2
Regime probabilities
P
Q
Variance parameters
C
ARCH(1)
GARCH(1)

Coefficient
p - value
-0.0400
0.6820
0.4535
1.6414
0.01
0.00
0.00
0.00
0.9695
0.9404
0.00
0.00
0.0007
0.2911
0.1538
0.00
0.00
0.07
Specification tests
BUX
Statistic
p - value
Ljung-Box for Standardized residuals
lag 6
3.9
0.69
lag 12
8.59
0.74
Ljung-Box for Squared standardized residuals
lag 6
0.94
0.99
lag 12
4.88
0.96
Hansen Standardized LR
Newey-West lag 0
2.46
0.09

EU
EU
EU
EU market ‘beta’:  BUX
,t   BUX X HU ,t

US
US
US market ‘beta’:  BUX
,t   BUX , S (S BUX ,t )
Estimation Results
2. The Dynamic Factor Regime-Switching Models
Graphs of smoothed probabilities of being in the low volatility regime

Smoothed Probability for State 1 in WIG model
Smoothed Probability for State 1 in BET model
1.0
1.0
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0.0
0.0
2005
2006
2007
2008
2005
2009
Smoothed Probability for State 1 in PX model
2007
2008
2009
Smoothed Probability for State 1 in BUX model
1.0
1.0
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0.0
2006
0.0
2005
2006
2007
2008
2009
2005
2006
2007
2008
2009
Estimation Results
2. The Dynamic Factor Regime-Switching Models





The models involving regime-switching are the best-fitted by the
measure of Hansen’s test – the null of one state is rejected at
90% confidence level for all the CEE indices.
The generalized residual-based tests find no evidence of linear
dependence or ARCH type effects for residuals from RS models.
The switching component pertains only to the US spillover
effects for all the CEE markets.
Poland and Czech equity markets are more integrated at the
regional level, while US shock spillovers are prevalent for the
Romanian and Hungarian equity markets.
The high local volatility states coincide with the peak of the
recent global financial crisis in 2008 -2009.
Estimation Results
3. Economic determinants of switching between states

The logit regressions for switching states
C
CDS
Exchange rate volatility
Default Spread
TED Spread
R2 McFadden




BET
Coefficient p - value
-3.8897
0.00
0.6023
0.00
4481.011
0.00
0.0447
0.09
0.015
0.06
38%
PX
Coefficient
p - value
-4.1924
0.00
0.0137
0.06
1184.674
0.72
0.0971
0.00
0.0217
0.04
38%
WIG
Coefficient p - value
-1.8937
0.00
1.2498
0.00
-912.5602
0.09
0.0276
0.02
0.0088
0.04
33%
Dependent variable in the regressions is a binary dummy: equals 1 if smoothed
probability of high volatility regime is greater than 50%, 0 otherwise.
The CDS price entertains a positive effect for switching from low to high local
volatility state for all markets.
EUR/RON conditional volatility positively influences switching to a high volatility
state for the BET index returns.
Higher US default spread and TED spread turn CEE equity markets turbulent.
Estimation Results
4. Variance ratios and Conditional Correlations

Average variance ratios and conditional correlations from best-fitted models
BET
Correlation
Period
VR
Full sample
Crisis
0.0216
0.0241
0.1449
0.1518
Full sample
Crisis
0.2082
0.2248
0.4421
0.4635



BUX
VR
Correlation
EU spillover effects
0.0340
0.1783
0.0452
0.1961
US spillover effects
0.2286
0.4500
0.3193
0.5348
VR
PX
Correlation
VR
WIG
Correlation
0.0776
0.0639
0.2743
0.2441
0.0892
0.0988
0.2949
0.3066
0.2982
0.3985
0.5326
0.6149
0.243
0.3218
0.4737
0.5526
Over full sample, US volatility spillovers explain cross-sectional approx. 25%
of the variance of CEE indices returns; the average EU variance ratio is 5%.
During crisis periods, volatility spillovers from US market account for about
30% cross-sectional average for the CEE markets volatility, while EU
spillover effects only increase to 6% on average.
The increase of conditional correlations during crisis periods is not evidence
for contagion, but an effect of the natural interdependence between markets.
Estimation Results
5. Contagion Tests – individual markets

Romania
BET

Constant spillover
Coefficient
p - value
Structural model
Coefficient
p - value
Regime Switching model
Coefficient
p - value
v EU,0
-0.1853
0.55
-0.1741
0.58
-0.0363
0.91
v EU,1
0.5913
0.47
0.5432
0.50
0.5400
0.51
v US,0
-0.1396
0.44
-0.1368
0.46
-0.2028
0.26
v US, 1
0.1278
0.61
0.3954
0.12
0.2739
0.26
Poland
WIG
Constant spillover
Coefficient
p - value
Structural model
Coefficient
p - value
Regime Switching model
Coefficient
p - value
v EU,0
-0.0358
0.88
-0.0392
0.87
-0.0874
0.71
v EU,1
0.3086
0.54
0.3107
0.53
0.3153
0.53
v US,0
-0.1264
0.33
-0.0971
0.47
-0.0780
0.54
v US, 1
0.0995
0.62
0.1515
0.45
0.0816
0.67
Estimation Results
5. Contagion Tests – individual markets

Czech Republic
PX

Constant spillover
Coefficient
p - value
Structural model
Coefficient
p - value
Regime Switching model
Coefficient
p - value
v EU,0
0.0759
0.70
0.0992
0.62
0.0909
0.64
v EU,1
0.6701
0.42
0.6621
0.42
0.7654
0.35
v US,0
-0.0954
0.35
-0.1700
0.09
-0.1631
0.11
v US, 1
0.4179
0.04
0.5322
0.01
0.2906
0.16
Hungary
BUX
Constant spillover
Coefficient
p - value
Structural model
Coefficient
p - value
Regime Switching model
Coefficient
p - value
v EU,0
-0.3101
0.17
-0.2406
0.28
-0.2307
0.31
v EU,1
0.9741
0.06
0.8647
0.08
0.774
0.10
v US,0
-0.1294
0.29
-0.2677
0.03
-0.2047
0.09
v US, 1
0.3027
0.14
0.3971
0.04
0.2869
0.12
Estimation Results
5. Contagion Tests – CEE markets group

The panel tests of contagion
CEE




Constant spillover
Coefficient p - value
Structural model
Coefficient p - value
Regime Switching model
Coefficient
p - value
v EU,0
-0.1136
0.48
-0.0887
0.59
-0.1181
0.49
v EU,1
0.6362
0.26
0.5952
0.29
0.5431
0.28
v US,0
-0.1228
0.20
-0.1679
0.09
-0.1618
0.13
v US, 1
0.2373
0.17
0.3690
0.03
0.2141
0.21
There is no evidence for contagion to the Romanian and Polish equity markets,
regardless of the cross-market linkages model employed.
Contagion from the global or regional level is identified during crisis periods to the
Czech and Hungarian equity markets, except for the RS model.
Testing on residuals from regime-switching models gives the same conclusion of
no contagion at both individual market levels and to the CEE as a group.
The panel test of contagion indicates excess exposure to the US effects for the
CEE equity markets when structural ‘beta’ model is employed.
Conclusions





Shock-spillover from the global level are larger than those from
the regional level to the group of CEE equity markets.
The US market volatility is the dominating influence on CEE
equity market variation.
Higher risk of local CEE sovereign default, higher currency
volatility and worsening financial conditions in the world
economy (US) lead to higher CEE stock market volatility.
Contagion tests results depend upon the model of volatility
spillovers.
I find no contagion when using the best-fitted models. The
results of the study come in line with the findings in the
literature on contagion which employs similar methodologies.
References








Baele, L. (2005), “Volatility Spillover Effects in European Equity markets”, Journal of
Quantitative Analysis, 40, 373 – 401
Baele, L. and K. Inghelbrecht (2010), “Time-varying integration, interdependence
and contagion”, Journal of International Money and Finance, 1–28
Beirne, J., G. M. Caporale, M. Schulze-Ghattas, and N. Spagnolo (2008),
“Volatility Spillovers and Contagion from Mature to Emerging Stock Markets”, IMF
WP/08/286
Bekaert, G. and C. Harvey (1997), “Emerging equity market volatility”, Journal of
Financial Economics, 43, 29 – 77
Bekaert, G., C. Harvey, and A. Ng (2005), “Market Integration and Contagion”,
Journal of Business, 78, 39 – 69
Bekaert, G., R. Hodrick, and X. Zhang (2008), “International stock return
comovements”, ECB WP NO 931
Bohl, M. T. and D. Serwa (2005), “Financial contagion vulnerability and resistance:
A comparison of European stock markets”, Economic Systems, 29, 344 – 362
Chen, N. and F. Zhang (1997), “Correlations, trades and stock returns of the
Pacific-Basin markets”, Pacific-Basin Finance Journal, 5, 559 – 577
References








Dornbusch, R., Y. C. Park, and S. Claessens (2000), “Contagion: Understanding
How It Spreads”, The World Bank Research Observer, 15, 179 – 197
Dungey, M., R. Fry, B. Gonzalez-Hermosillo, and V. L. Martin (2005),
“Empirical modelling of contagion: A review of methodologies”, Quantitative
Finance, 5, 9 – 24
Engel, C. and J. Hamilton (1990), “Long Swings in the Dollar: Are They in the Data
and Do Markets Know It?”, American Economic Review, 80, 689 – 713
Engle, R., T. Ito, and W. Lin (1988), “Meteor showers or heat waves?
Heteroskedastic intra-daily volatility in the foreign exchange market”, NBER Working
Paper No. 2609
Engle, R. and K. Kroner (1995), “Multivariate Simultaneous Generalized ARCH”,
Econometric Theory, 11, 122 – 150
Forbes, K. J. and R. Rigobon (2002), “No Contagion, Only Interdependence”, The
Journal of Finance, 57, 2223 – 2261
Franses, P. H. and D. van Dijk (2003), “Nonlinear Time Series Models in Empirical
Finance”, Cambridge University Press
Hamao, Y., R. Masulis, and V. Ng (1990), “Correlations in Price Changes and
Volatility across International Stock Markets”, The Review of Financial Studies, 3, 281 –
307
References







Hamilton, J. D. (1990), “Analysis of Time Series Subject to Changes in Regime”,
Journal of Econometrics, 45, 39-70
Hamilton, J. D. (1994), “Time Series Analysis”, Princeton University Press, NJ
Princeton
Hamilton, J. D. (1996), “Specification testing in Markov-Switching time-series
models”, Journal of Econometrics, 70, 127 – 157
Hansen, B. E. (1992), “The Likelihood Ratio Test under Nonstandard Conditions:
Testing the Markov Switching Model of GNP”, Journal of Applied Econometrics, 7, S61 –
S82
Hansen, B. E. (1996), “Erratum: The Likelihood Ratio Test under Nonstandard
Conditions: Testing the Markov Switching Model of GNP”, Journal of Applied
Econometrics, 11, 195 – 198
Karolyi, G. A. (1995), “A Multivariate GARCH Model of International Transmissions
of Stock Returns and Volatility: The Case of the United States and Canada”, Journal of
Business & Economic Statistics, 13, 11 – 25
King, M. A. and S. Wadhwani (1990), “Transmission of Volatility between Stock
Markets”, The Review of Financial Studies, 3, 5 – 33
References






Li, H. and E. Majerowska (2007), “Testing stock market linkages for Poland and
Hungary: A multivariate GARCH approach”, Research in International Business and Finance
Maheu, J.M., and T.H. McCurdy (2000), “Identifying Bull and Bear Markets in
Stock Returns”, Journal of Business and Economic Statistics, 18, 100 – 112
Ng, A. (2000), “Volatility spillover effects from Japan and the US to the Pacific–
Basin”, Journal of International Money and Finance, 19, 207 – 233
Pericoli, M. and M. Sbracia (2001), “A primer on financial contagion”, Banca
d’Italia, Temi di discussione, 407/2001
Perlin, M. (2009), “MS_Regress - A Package for Markov Regime Switching Models in
Matlab”, Available at:
http://www.mathworks.com/matlabcentral/fileexchange/authors/21596
Smith, D. R. (2007), “Evaluating Specification Tests for Markov-Switching Time
Series Models”, WP available at SSRN: http://ssrn.com/abstract=976385