Available online at www.sciencedirect.

com

ScienceDirect

Journal of Policy Modeling 37 (2015) 1–13

Value at Risk of the main stock market indexes in the

European Union (2000–2012)夽
Emma M. Iglesias ∗
University of A Coruña, Spain
Received 31 July 2014; received in revised form 4 November 2014; accepted 4 January 2015
Available online 24 January 2015

Abstract
We analyze extreme movements of the main stocks market indexes in the European Union. We find that
the Sweden and UK markets are the preferred ones for risk averse investors since they present the best
risk-return performance. Moreover, the UK market is found to have a very low dependence with the rest of
the European financial cycles, being the best one (in terms of risk-return) available for investors among the
ones studied in this paper. Greece and Holland have the worst performance in terms of risk-return. Austria
has the highest average return although the VaR is also high. Moreover, all markets are found to be linked:
Austria, Belgium, Germany, Ireland and UK are the markets that are less dependent; while France, Greece,
Holland, Italy, Spain and Sweden are more dependent on the rest of the European financial cycles. We find a
very strong dependence of France from Belgium. Our results have very important policy implications with
respect to the appropriate monetary policy that countries should adopt. In special, countries that experience
unstable financial markets should consider similar macroeconomic policies to the UK and Sweden.
© 2015 Society for Policy Modeling. Published by Elsevier Inc. All rights reserved.

JEL classification: C12; C13; C22; G01; G11; G15; G32

Keywords: Value-at-Risk; Extreme value theory; Heavy tails; Stock market indexes; Eurozone

夽 Financial support from the Spanish Ministry of Economy and Competitiveness, through Grant ECO2012-31459, and

two referees for helpful comments are gratefully acknowledged.

∗ Correspondence to: Departamento de Economía Aplicada II, Facultad de Economía y Empresa, Universidad de A

Coruña, Campus de Elviña, 15071, Spain. Tel.: +34 981167000.

E-mail address: emma.iglesias@udc.es

http://dx.doi.org/10.1016/j.jpolmod.2015.01.006
0161-8938/© 2015 Society for Policy Modeling. Published by Elsevier Inc. All rights reserved.
2 E.M. Iglesias / Journal of Policy Modeling 37 (2015) 1–13

1. Introduction

In the last years, an increasing number of firms and countries have suffered important losses
mainly while being exposed to extreme events. Although diversification in the portfolio is a well-
known strategy to reduce risk, this is not a guarantee to avoid losses specially in crisis periods.
All this motivates the need to have rigorous measures of risk, and in special, during extreme
events. Many studies focus on studying different aspects of international stock market indexes.
For example, Edwards, Gómez-Biscarri, and Pérez-de-Gracia (2003) focus on Latin American
and Asian countries and how they have been affected by a specific event: the liberalization.
Bessler and Yang (2003) search for interdependencies in international stock markets. Su (2011)
looks for bidirectional non-linear causalities between the real state market and the stock market
of western European countries and Diamandis and Drakos (2011) find multidirectional Granger-
causalities between stock prices and exchange rates for four Latin-American countries. More
recently Majumder (2013) studies efficiency issues in the Indian market and Cevik, Dibooglu,
and Kenc (2013) measure the financial stress in Turkey. Our objective in this paper is to go further
and to study in more detail the most representative European financial markets in the current
crisis and which of them present the best risk-return performance from the investors point of
view. Although knowledge of the past does not guarantee future events, we show the performance
of average return versus Value at Risk (VaR) of the main European stock market indexes and this
can help future investors to choose their portfolios in the Eurozone. Iglesias and Lagoa-Varela
(2012) did not find significant differences by country when analyzing the companies included in
the Eurostoxx50 in the 2000s decade. Our objective is to study now the most representative stock
market indexes.
Both academics and practitioners have been trying to find determinants of extreme behav-
iors. For example, Cutler, Poterba, and Summers (1989) in fact conclude that extreme values for
returns happen during periods where there are no news with special relevance. The concept of
financial risk is directly linked with the one of losses; however, there are different measures of
risk. VaR has its origin in Riskmetrics, that was developed in the 80 s and 90 s by JP Morgan.
Moreover, VaR became a very important measure of risk since the Basel Committee on banking
supervision declared that banks should be able to cover losses in their portfolios for horizons
of 10 days with a confidence level of 99 per cent. We use several alternative estimators to pro-
vide a good estimate of the VaR in small samples. There are two distinct VaR measures, one
dealing with the unconditional distribution and one with the conditional distribution. The former
provides risk managers with information on different worst case scenarios dealing with market
risk occurring over long periods, for example ten years. In contrast, the latter measure details
the present risk facing investment managers conditional on the present risk and return environ-
ment of a futures contract. In this paper we are interested in providing measures of unconditional
risk, and moreover, since we have a not very large sample size, we need a method that can pro-
vide reasonable VaR estimates (see Jansen and de Vries, 1991; Kearns and Pagan, 1997; Iglesias
and Linton, 2009; Iglesias, 2012 as empirical examples where unconditional VaR estimates are
provided).
Many alternative estimators of unconditional VaR can be found. The most traditional one
was proposed by Hill (1975), where the existence of generalized autoregressive and conditional
heteroskedastic (GARCH) effects (very important when modelling financial returns; see e.g. Engle
(1982) and Bollerslev (1986) for more details) can be implicitly acknowledged. Hill (2010) has
 E.M. Iglesias / Journal of Policy Modeling 37 (2015) 1–13 3

shown that the Hill (1975) estimator is robust to the existence of GARCH effects; however, there
exists clear evidence in the literature of its poor finite sample properties (see e.g. Kearns and
Pagan, 1997; Wagner and Marsh, 2005). In this paper, we use also an alternative estimator that
is shown to have improved finite sample properties under some assumptions and it is based on
the work of Stărică and Pictet (1997), Berkes, Horváth, and Kokoszka (2003) and it has been
generalized to the case of the GJR-GARCH model of Glosten, Jagannathan, and Runkle (1993)
by Iglesias and Linton (2009). The reason for the improved finite sample √ properties of the Iglesias
and Linton (2009) estimator of tail thickness is that it converges at rate T to a normal distribution
(where T is the sample size). Following Iglesias and Linton (2009), we present Hausman type
specification tests in the empirical results section to provide evidence of the better performance
of the new estimator. Measures of VaR are also presented.
In this paper, we find that clearly, at aggregate level (i.e., using the main stock market indexes
of each country) the UK and Sweden stock markets are the preferred ones for risk averse investors
and they present the best risk-return performance. Moreover, the UK market is found to have a
very low dependence with the rest of the European financial cycles and this result complements
the study in Iglesias (2012) that finds that the British pound has also the smaller VaR estimate.
Therefore the UK market is found to be the best one available nowadays for investors among the
ones studied in this paper. Belgium also presents a low VaR estimate. Greece is the country with
the worst performance in terms of risk-return. Holland presents a very high VaR estimate and a
negative average return in the analyzed period, being also not attractive for investors. Austria is
the country where we can find higher average returns although the VaR is high. We do not find
very clear differences when analyzing Germany, Ireland, Italy, France and Spain. They all have
very similar high VaR estimates and negative average returns (Germany has a very low average
return). Moreover, all markets are found to be linked: Austria, Belgium, Germany, Ireland and UK
are the markets that are less dependent on the rest of the European market financial cycles (since
only one lag is statistically significant for each of these countries). France, Greece, Holland, Italy,
Spain and Sweden are more dependent on the rest of the European financial cycles (they present
more than one statistically significant relationships). We find a very strong relationship of France
from Belgium.
The German stock market index (DAX) is seen to have a very high volatility in Graph 1
specially in 2008. Baláž (2009) studies the DAX’s January 2008 crash that lost 15% of its value
in three days and they were days with very high volatility. They coincided with the bad news
coming from the USA economy and in Asia and the pressure of selling participations in Societe
Generale, a company with very high relevance. There was also another important crash in the
German DAX in October 2008 due to the nationalization of the banks in Island and in November
2008 due to the rejection of the US senate to help the three big brands of cars due to the crisis
(Ford, General motors and Chrysler). In this paper we show that Germany has had a very high
volatility similar to Spain in the analyzed period 2000–2012 on average (both the standard error
and the VaR estimate). The volatility levels where very similar; however, in Germany they were
mainly due to the bad news coming from other countries (such as the USA and Asia) while in
Spain they usually are related to bad news coming from the own Spanish economy.
The plan of the paper is as follows. In Section 2 we present the data. Section 3 shows the model
and the estimators and tests that are used, while Section 4 provides the empirical results. Section
5 finally concludes.
4 E.M. Iglesias / Journal of Policy Modeling 37 (2015) 1–13

Graph 1. Returns of each of the countries.

2. Data

We use daily data1 of returns of eleven European countries from January 2000 until December
20122 (we use the value at the moment of the closing of each daily session). We use all the
available dataset in all countries except in Italy, since the base value of the Italian FTSE MIB
Index was re-set at the level of the MIB 30 Index at the close of trading on October 31, 2003.
Therefore we analyze Italy only from November 2003 onwards in order to avoid structural breaks.
If we denote pt the stock market index of country p at period t, then we construct the return of
country p at period t (Rpt ) as
 
pt
Rpt = log
pt−1
where t = 1, . . ., T. We construct the returns for the eleven companies and we test for the stationarity
of the series by using standard Augmented Dickey–Fuller (1979) test and Phillips–Perron (1988)
tests. We also computed the GLS-detrended Dickey–Fuller (Elliott, Rothenberg, & Stock, 1996),
Kwiatkowski, Phillips, Schmidt, and Shin (1992), Elliott, Rothenberg, and Stock Point Optimal

1 The data has been obtained from the finance-yahoo database: http://es.finance.yahoo.com. We use the German DAX,

Austrian ATX, Belgium Bel20, Spanish IBEX35, French CAC40, Greek GD.AT, Dutch AEX, Irish ISEQ, British FTSE100
and the Swedish OMX index.
2 Since January 2013, many structural changes have been made in Europe. For example, in the Spanish economy,

housing cannot be deducted in the personal income tax and an important increase in the Value Added Tax (VAT) has been
made. Also many economic reforms initiated by the Spanish conservative party in 2012 (such as in the labour and financial
markets) have changed significantly Spain. This affects importantly the European system, and therefore, we decide to
stop our analysis in December 2012 to avoid structural breaks.
 E.M. Iglesias / Journal of Policy Modeling 37 (2015) 1–13 5

Table 1
Summary of descriptive statistics of returns.
Raustria Rbelgium Rfrance Rgermany Rgreece Rholland Rireland

Mean 0.0002 −8.75E−05 −0.0002 3.62E−05 −0.0006 −0.0002 −0.0001

Standard deviation 0.0153 0.0135 0.0157 0.0162 0.0175 0.0156 0.0148
Minimum −0.1025 −0.0832 −0.0947 −0.0743 −0.1021 −0.0959 −0.1396
Maximum 0.1202 0.0933 0.1060 0.1080 0.1343 0.1003 0.0973

Ritaly Rspain Rsweden Ruk

Mean −0.0002 −0.0001 5.56E−05 −3.72E−05

Standard deviation 0.0155 0.0157 0.0148 0.0129
Minimum −0.0860 −0.0959 −0.0807 −0.0927
Maximum 0.1087 0.1348 0.0863 0.0938

(1996), and Ng and Perron (2001) tests and we reject at all standard significant levels the null
hypothesis of the unit root.3 Table 1 shows a summary of some descriptive statistics.

3. Model and theory for the VaR

Hill (1975, 2010) provides an estimator (κ̂+ ) of the tail index parameter κ that it does
not use the GARCH structure at all. Ordering the data as r1:T ≤ r2:T ≤ · · · ≤ rT :T , define
m−1

κ̂+ = m1 ((log rT −j:T − log rT −m:T ). Here, m = m(T) is a smoothing parameter that satisfies
j=0
m → ∞ and √ m/T → 0 κ̂+ is consistent and asymptotically normal for iid data with κ > 0 and
+
it satisfies m(κ̂ − κ) ⇒ N(0, Φ)as m = m(T ) → ∞ for some Φ. Hill (2010) has shown that
under the traditional GARCH(1,1) specification, the asymptotic variance Φ = κ2 is as if the
process were iid with the same marginal distribution.
An alternative new estimator to Hill (1975) was proposed originally in Stărică and Pictet
(1997), denoted by κ̂. Berkes et al. (2003) provided the formal theory of the asymptotic normality
of the estimator under the presence of GARCH effects and it was generalized to the case of the
GJR-GARCH model of Glosten et al. (1993) by Iglesias and Linton (2009).
A traditional GARCH model is given for each return series by
rt = ε t σ t (1a)
σt2 = ω + γrt−1
2
+ βσt−1
2
(1b)

where εt is an iid process with conditional variance given by σt2 .

If (1a)–(1b) are correctly specified, Mikosch and Stărică (2000) have shown that there exists
a positive constant C0 and tail index parameter κ such as
Pr(σt > x)∼c0 x−κ , as x → ∞ (2)

3 Results are available from the author upon request.

6 E.M. Iglesias / Journal of Policy Modeling 37 (2015) 1–13

Pr(|rt | > x)∼ < E[|εt |κ ]Pr(σt > x), as x → ∞ (3)

Then, the Iglesias√and Linton (2009) alternative estimator is constructed as follows. Let
 
ω̂, γ̂, β̂ be some T consistent estimator of (ω, γ, β) , such as the quasi-maximum like-
lihood estimator, and define the standardized residuals ε̂t = rt /σ̂t , where σ̂t2 = ω̂ + γ̂rt−1
2 +

β̂σt−1
2 , t = 2, . . ., T , with σ12 some initial value. Define

Ât = γ̂ ε̂2t + β̂. (4)

The estimator is any solution κ̂ to

ψ̂T (κ̂) = 0, (5)

T
1  κ/2
ψ̂T (κ) = Ât − 1. (6)
T
t=1

This can be computed by grid search over some suitable range. Berkes et al. (2003) show
that if the model is correctly specified as (1a)–(1b), then κ̂ is consistent and asymptotically
normally distributed. This result has been generalized to the case that the correct model is the GJR-
GARCH model of Glosten et al. (1993) by Iglesias and Linton (2009). The asymptotically normal
distribution theory can be used to provide a specification test of the underlying GJR-GARCH
model based on a Hausman test. Under the traditional GJR-GARCH specification
κ̂ − κ̂+
√ ⇒ N(0, 1). (7)
κ̂+ / m
and one can reject for large or small values of this statistic.
Also, as a final application of the two estimators to the computation of VaR, Iglesias and Linton
(2009) show that the VaR is for small α (denoted Varα )
α = Pr(rt > Varα ) ≡ c0 Varα−κ ,
which gives
 c 1/κ
0
Varα = .
α
Iglesias and Linton (2009) also show an estimator for c0 and hence of Varα given by
T 1 t=1 κ̂/2 κ̂/2
1  κ̂ [(ω̂+ Ât σ̂t2 ) − (Ât σ̂t2 ) ]
ĉ = |ε̂t | T
T  (8)
κ̂/2
2T
t=1 (κ̂/2T ) t=1
T (Ât ln Ât )
 1/κ̂
α = ĉ
Var . (9)
α

When using the Hill estimator, it implies the alternative estimator of c0 and hence of Varα
m
1  κ̂+
ĉ+ = rT −i:T i. (10)
Tm
i=1
 E.M. Iglesias / Journal of Policy Modeling 37 (2015) 1–13 7

 1/κ̂+
+ ĉ
Var α = . (11)
α
as is known in the literature. These estimates both converge at the same rate as κ̂+ .

4. Empirical results

We first provide the VaR estimates and later we focus on the interrelations among the financial
markets.

4.1. VaR estimates

First we apply to our eleven returns the Hill (1975) estimator of κ. The results are given in
Table 3. Later, we also apply the Iglesias and Linton (2009) alternative estimator. In order to do
that, we first fit the standard GJR-GARCH model proposed by Glosten et al. (1993) where εt is
an iid process with conditional variance given by σt2 and
Rpt = εt σt = ut (12a)
σt2 =ω + γu2t−1 + δu2t−1 1{ut−1 < 0} + βσt−1
2
(12b)
We cannot reject the null hypothesis of absence of neglected serial correlation with the diag-
nostic tests of Ljung–Box (1978) in all (GJR)-GARCH models that were fitted, at 5% significant
level after including an autoregressive of order 1 (AR(1)) model for Austria, Belgium and Ireland
in the mean equation. For the rest of the countries, we cannot fit any (GJR)-GARCH model
that can pass all diagnostic tests of neglected serial correlation. Table 2 shows the results of the
parametric estimation of the volatility models when using quasi-maximum likelihood for each of

Table 2
Parametric estimation. Standard errors given in parenthesis.
Raustria Rbelgium Rfrance Rgermany Rgreece Rholland Rireland

ω 3.19E−06 2.38E−06 – – – – 3.14E−06

(4.92E−07) (2.66E−07) – – – – (3.09E−07)
γ 0.0262 0.0118 – – – – 0.0364
(0.0110) (0.0078) – – – – (0.0078)
δ 0.1309 0.1793 – – – – 0.1096
(0.0149) (0.0140) – – – – (0.0118)
β 0.8873 0.8833 – – – – 0.8915
(0.0099) (0.0074) – – – – (0.0082)

Ritaly Rspain Rsweden Ruk

ω – – – –
– – – –
γ – – – –
– – – –
δ – – – –
– – – –
β – – – –
– – – –
8 E.M. Iglesias / Journal of Policy Modeling 37 (2015) 1–13

Table 3
Estimated values of κ and Varα for daily returns when α = 0.001.
Raustria Rbelgium Rfrance Rgermany Rgreece Rholland Rireland

κ̂+ , hill estimatora 2.31 2.19 2.34 2.31 2.07 2.05 2.14
κ̂ 0 0 – – – – 0
ĉ 0 0 – – – – 0
ĉ+ , hill estimatora 2.51E−05 2.98E−05 2.44E−05 2.92E−05 6.87E−05 6.23E−05 3.66E−05
Hausmanb – – – – – – –
α in (9)
Var – – – – – – –
+
α in (11)
Var 0.2029 0.2011 0.2046 0.2166 0.2743 0.2582 0.2132

Ritaly Rspain Rsweden Ruk

κ̂+ , hill estimatora 2.17 2.38 2.39 2.25

κ̂ – – – –
ĉ – – – –
ĉ+ , hill estimatora 3.21E−05 2.21E−05 1.78E−05 2.08E−05
Hausmanb – – – –
α in (9)
Var – – – –
+
α in (11)
Var 0.2050 0.2015 0.1853 0.1788
a We use m = 0.1T.
b Hausman statistic given in (7).

the returns. Note that following Jensen and Rahbek (2004a, 2004b), we can allow for GARCH
parameter estimates where strict stationarity does not hold. When we compute the Iglesias and
Linton (2009) estimator, however, it does not find any significant value different from zero, and
therefore being misspecified. Therefore, we rely in all cases on the Hill (1975) estimator. We have
several findings (summarized in Table 4): we find that the Sweden and UK markets present the
smallest risk both analyzing the traditional standard deviation as well as its Value at Risk (VaR)
and they also provide one of the best available European average returns, being the most attractive
places for investors. Belgium also presents a low VaR estimate. Greece is the country with the
worst performance in terms of risk-return. Holland also presents a very high VaR estimate and
a negative average return in the analyzed period, being also not attractive for investors. Finally,
Austria is the country where we can find higher average returns although the VaR is high. We do
not find very clear differences when analyzing Germany, Ireland, Italy, France and Spain. They
all have very similar VaR estimates and average returns, excluding the case of a positive return for
Austria. Germany has had periods of very high volatility mainly in 2008 similar to those in Spain
at the end of the analyzed period (see Tables 1 and 4) producing that on average, they present in
total similar risk in the analyzed period.

4.2. Causality relationships

4.2.1. Preliminary analysis: bivariate Granger causalities

There are many papers that look for causal relationships between international stock markets.
For example Malliaris and Urrutia (1992) investigate the Granger causalities of six stock market
indices before, during, and after the October 1987 crash to identify the origin of the crisis.
 E.M. Iglesias / Journal of Policy Modeling 37 (2015) 1–13 9

Table 4
Ranking of returns according to estimated VaR when α = 0.001 (note that the rank by the estimated VaR is not the same
according to the average return).
Returns Estimated VaR Average return Std. error Ranking
Average return Estimated VaR Std. error

Rgreece 0.2743 −0.0006 0.0175 1 1 1

Rholland 0.2582 −0.0002 0.0156 2 2 5
Rgermany 0.2166 3.62E−05 0.0162 9 3 2
Rireland 0.2132 −0.0001 0.0148 5 4 8
Ritaly 0.2050 −0.0002 0.0155 2 5 6
Rfrance 0.2046 −0.0002 0.0157 2 6 3
Raustria 0.2029 0.0002 0.0153 11 7 7
Rspain 0.2015 −0.0001 0.0157 5 8 3
Rbelgium 0.2011 −8.75E−05 0.0135 7 9 8
Rsweden 0.1853 5.56E−05 0.0148 10 10 10
Ruk 0.1788 −3.72E−05 0.0129 8 11 11

No significant lead–lag relationship is found in the pre- or post-crash period. Kwan, Sim, and
Cotsomitis (1995) employ a similar methodology and examine the efficient markets hypothesis
in its weak form. Again, no dominant market is identified, so they reject their hypothesis. More
recent papers such as Masih and Masih (2001) document the leading role of the US market in the
international stock market.
We look for Granger causality relationships among the eleven main European stock market
indexes analyzed previously. We analyze several lag orders and Table 4 presents the results for lag
7 (to allow for the daily relationships up to lag 7). The results are robust to considering different lag
orders. Table 5 shows only the p-value that allows to reject the null hypothesis that a country does
not Granger cause another country. We report only the results that are statistically significant at
1% to focus only on the stronger relationships. After analyzing all possible cases we find evidence
of the relationships that are reported in Table 5. All markets are found to be linked: in special,
Sweden is affected mainly by fluctuations in the German, Greek and Dutch markets while the UK
is affected mainly by fluctuations in the Spanish, Irish and Dutch markets. Finally, Austria, the
country where we can find higher average returns although the VaR is high, it is linked mainly to
the Irish financial market.
Note that standard Granger causality relationships are only considering explicitly bivariate
Vector Autoregression (VAR) models, and therefore it has important limitations. In order to
overcome this issue, we proceed now to analyze a full VAR model with the 11 stock market
returns.

4.2.2. Vector autoregression

Since when we fit univariate GARCH models to each of the series by themselves we do not find
any statistically significant evidence of conditional heteroskedasticity, we proceed to model jointly
the eleven time series jointly as a VAR without allowing for conditional heteroskedasticity (we
estimated a VAR model with GARCH effects as in Iglesias and Phillips (2003) but we did not find
statistical significant evidence of GARCH effects). We estimated the VAR model up to seven lags
(due to the daily characteristics of our dataset) and we chose, using the AIC (Akaike Information
10 E.M. Iglesias / Journal of Policy Modeling 37 (2015) 1–13

Table 5
Granger casuality relationships.
Null hypothesis p-Value

Rireland does not Granger cause Raustria 4.9E−06

Raustria does not Granger cause Rireland 0.00132
Rfrance does not Granger cause Rbelgium 0.00153
Rbelgium does not Granger cause Rfrance 0.00000
Rgermany does not Granger cause Rbelgium 0.00000
Rbelgium does not Granger cause Rgermany 7.0E−15
Rireland does not Granger cause Rbelgium 0.00157
Rspain does not Granger cause Rbelgium 3.3E−06
Rbelgium does not Granger cause Rsweden 0.00039
Ruk does not Granger cause Rbelgium 6.6E−12
Rgermany does not Granger cause Rfrance 1.0E−06
Rfrance does not Granger cause Rgermany 6.4E−07
Rireland does not Granger cause Rfrance 0.00599
Rspain does not Granger cause Rfrance 0.00094
Rfrance does not Granger cause Rsweden 0.00263
Ruk does not Granger cause Rfrance 5.3E−11
Rgermany does not Granger cause Rholland 0.00027
Rspain does not Granger cause Rgermany 4.8E−05
Rgermany does not Granger cause Rsweden 0.00013
Ruk does not Granger cause Rgermany 0.00000
Rgreece does not Granger cause Rholland 3.2E−05
Ritaly does not Granger cause Rgreece 9.3E−07
Rgreece does not Granger cause Ritaly 0.00172
Rsweden does not Granger cause Rgreece 0.00039
Rgreece does not Granger cause Rsweden 0.00000
Ritaly does not Granger cause Rholland 7.8E−13
Rholland does not Granger cause Ritaly 0.00000
Rsweden does not Granger cause Rholland 0.00000
Rholland does not Granger cause Rsweden 0.00740
Rholland does not Granger cause Ruk 0.00270
Rspain does not Granger cause Rireland 1.3E−09
Rireland does not Granger cause Rspain 0.00088
Ruk does not Granger cause Rireland 0.00385
Rireland does not Granger cause Ruk 2.3E−16
Rsweden does not Granger cause Ritaly 0.00325
Ruk does not Granger cause Rspain 8.7E−06
Rspain does not Granger cause Ruk 0.00000

Criterion) and the SBC (Schwarz’s Bayesian Criterion), the VAR(2) given in Table 6. Again we
only show the t-statistics that are statistically significant at the 1% level to look for the stronger
relationships. We see how (as it happened with the Granger-causality relationships) all markets are
linked since all returns depend on some previous lag(s). However, the relationships that we find
are different than using bivariate-VARs. We obtain now that Austria, Belgium, Germany, Ireland
and UK are the markets that are less dependent on the rest of the European market financial cycles
(since only one lag is statistically significant for each of these countries). France, Greece, Holland,
Italy, Spain and Sweden are more dependent on the rest of the European financial cycles (they
present more than one statistically significant relationships). We find a very strong relationship
for France from Belgium.
 E.M. Iglesias / Journal of Policy Modeling 37 (2015) 1–13 11

Table 6
VAR relationships (t-statistics are shown for the statistically significant relationships).
Lags Raustria Rbelgium Rfrance Rgermany Rgreece Rholland

Raustria(−1) 3.1553
Raustria(−2)
Rbelgium(−1) −4.6332
Rbelgium(−2) 93.0596
Rfrance(−1) −2.7096 −3.4982
Rfrance(−2) −2.9515
Rgermany(−1)
Rgermany(−2)
Rgreece(−1) 2.6886
Rgreece(−2) −3.4211 4.4759
Rholland(−1)
Rholland(−2)
Rireland(−1)
Rireland(−2)
Ritaly(−1) 5.0553 5.2083
Ritaly(−2) 3.3204
Rspain(−1)
Rspain(−2)
Rsweden(−1) 3.2670
Rsweden(−2) 2.6232
Ruk(−1)
Ruk(−2)

Lags Rireland Ritaly Rspain Rsweden Ruk

Raustria(−1) −2.6599
Raustria(−2)
Rbelgium(−1)
Rbelgium(−2)
Rfrance(−1) 2.6936
Rfrance(−2) −2.7165
Rgermany(−1)
Rgermany(−2) 2.8136
Rgreece(−1)
Rgreece(−2)
Rholland(−1)
Rholland(−2) 14.3200
Rireland(−1)
Rireland(−2)
Ritaly(−1)
Ritaly(−2) −3.1405 3.8740
Rspain(−1)
Rspain(−2)
Rsweden(−1)
Rsweden(−2)
Ruk(−1) −2.8382
Ruk(−2) 4.4481
12 E.M. Iglesias / Journal of Policy Modeling 37 (2015) 1–13

5. Policy implications and conclusions

We analyze extreme movements of the main stocks market indexes in the European Union in
the period 2000–2012. We find that the Sweden and UK markets present the smallest risk both
analyzing the traditional standard deviation as well as its VaR and they also provide one of the
best available European average returns, being the most attractive places for investors. This result
complements the study in Iglesias (2012) that finds that the British pound has also the smaller
VaR estimate. Belgium also presents a low VaR estimate.
We use several alternative estimators to provide a good estimate of the VaR in small samples.
Greece is the country with the worst performance in terms of risk-return. Holland also presents
a very high VaR estimate and a negative average return in the analyzed period, being also not
attractive for investors. Austria is the country where we can find higher average returns although
the VaR is high. Germany, Ireland, Italy, France and Spain have very similar VaR estimates and
average returns. Germany has had periods of very high volatility mainly in 2008 similar to those
in Spain at the end of the analyzed period, producing that on average, they present in total similar
risk in the analyzed period. Moreover, all markets are found to be linked: Austria, Belgium,
Germany, Ireland and UK are the markets that are less dependent on the rest of the European
market financial cycles (since only one lag is statistically significant for each of these countries).
France, Greece, Holland, Italy, Spain and Sweden are more dependent on the rest of the European
financial cycles (they present more than one statistically significant relationships). We find a very
strong relationship for France from Belgium.
Our results (the UK and Sweden are found to be the most attractive places to invest and also the
British pound – see Iglesias (2012) – has the smaller VaR estimate) have very important policy
implications with respect to the appropriate monetary and exchange rate policy that countries
should adopt. In special, countries that experience unstable financial markets should consider
similar macroeconomic policies to the UK and Sweden. Moreover, our results should be of interest
to investors and portfolio managers who are active in financial markets for risk management
purposes.
In this paper we only study the European financial markets at aggregate level through the main
stock market indexes. Future research will extend the analysis to each of the companies that are
the components of the stock market indexes.

References

Baláž, M. (2009). DAX’s January 2008 crash: A routine correction or outright panic? Comenius University in Bratislava.
Available at http://diplomovka.sme.sk/zdroj/3495.pdf
Berkes, I., Horváth, L., & Kokoszka, P. (2003). Estimation of the maximal moment exponent of a GARCH(1,1) sequence.
Econometric Theory, 19, 565–586.
Bessler, D. A., & Yang, J. (2003). The structure of interdependence in international stock markets. Journal of International
Money and Finance, 22, 261–287.
Bollerslev, T. (1986). Generalized autoregressive conditional heteroscedasticity. Journal of Econometrics, 31, 307–327.
Cevik, E. I., Dibooglu, S., & Kenc, T. (2013). Measuring financial stress in Turkey. Journal of Policy Modeling, 35(2),
370–383.
Cutler, D. M., Poterba, J. M., & Summers, L. H. (1989). What moves stock prices? NBER working papers 2538. National
Bureau of Economic Research, Inc.
Diamandis, P. F., & Drakos, A. A. (2011). Financial liberalization, exchange rates and stock prices: Exogenous shocks in
four Latin America countries. Journal of Policy Modeling, 33(3), 381–394.
Dickey, D. A., & Fuller, W. A. (1979). Distribution of the estimators for autoregressive time series with a unit root. Journal
of the American Statistical Association, 74, 427–431.
 E.M. Iglesias / Journal of Policy Modeling 37 (2015) 1–13 13

Edwards, S., Gómez-Biscarri, J., & Pérez-de-Gracia, F. (2003). Stock market cycles, financial liberalization and volatility.
Journal of International Money and Finance, 22, 925–955.
Engle, R. F. (1982). Autoregressive conditional heteroscedasticity with estimates of the variance of united kingdom
inflation. Econometrica, 50, 987–1007.
Elliott, G., Rothenberg, T. J., & Stock, J. H. (1996). Efficient tests for an autoregressive unit root. Econometrica, 64(4),
813–836.
Glosten, L. R., Jagannathan, R., & Runkle, D. E. (1993). On the relationship between the expected value and the volatility
of the nominal excess returns on stocks. Journal of Finance, 48, 1779–1801.
Hill, B. M. (1975). A simple general approach to inference about the tail of a distribution. Annals of Statistics, 3, 1163–1174.
Hill, J. B. (2010). On tail index estimation for dependent, heterogenous data. Econometric Theory, 26, 1398–1436.
Iglesias, E. M. (2012). An analysis of extreme movements of exchange rates of the main currencies traded in the foreign
exchange market. Applied Economics, 44(35), 4631–4637.
Iglesias, E. M., & Lagoa-Varela, D. (2012). Extreme movements of the main stocks traded in the Eurozone: an analysis
by sectors in the 2000’s decade. Applied Financial Economics, 22(24), 2085–2100.
Iglesias, E. M., & Linton, O. B. (2009). Estimation of tail thickness parameters from GARCH models Working paper.
Michigan State University and University Carlos III.
Iglesias, E. M., & Phillips, G. D. A. (2003). Another look about the evolution of the risk premium: a VAR-GARCH-M
model. Economic Modelling, 20(4), 777–789.
Jansen, D. W., & de Vries, C. G. (1991). On the frequency of large stock returns: putting booms and busts into perspective.
Review of Economics and Statistics, 73, 18–24.
Jensen, S. T., & Rahbek, A. (2004a). Asymptotic normality of the QML estimator of ARCH in the nonstationary case.
Econometrica, 72(2), 641–646.
Jensen, S. T., & Rahbek, A. (2004b). Asymptotic inference for nonstationary GARCH. Econometric Theory, 20(6),
1203–1226.
Kearns, P., & Pagan, A. (1997). Estimating the density tail index for financial time series. Review of Economics and
Statistics, 79(2), 171–175.
Kwan, A., Sim, A., & Cotsomitis, J. A. (1995). The causal relationships between equity indices on world exchange.
Applied Economics, 27(1), 33–37.
Kwiatkowski, D., Phillips, P. C. B., Schmidt, P., & Shin, Y. (1992). Testing the null hypothesis of stationarity against the
alternative of a unit root. Journal of Econometrics, 54, 159–178.
Ljung, G. M., & Box, G. E. P. (1978). On a measure of a lack of fit in time series models. Biometrika, 65, 297–303.
Majumder, D. (2013). Towards an efficient stock market: Empirical evidence from the Indian market. Journal of Policy
Modeling, 35(4), 572–587.
Malliaris, A. G., & Urrutia, J. L. (1992). The international crash of October 1987: Causality test. Journal of Financial
and Quantitative Analysis, 27(3), 353–364.
Masih, R., & Masih, A. M. M. (2001). Long and short term dynamic causal transmission amongst international stock
markets. Journal of International Money and Finance, 20, 563–587.
Mikosch, T., & Stărică, C. (2000). Limit theory for the sample autocorrelations and extremes of a GARCH(1,1) process.
Annals of Statistics, 28(5), 1427–1451.
Ng, S., & Perron, P. (2001). Lag length selection and the construction of unit root tests with good size and power.
Econometrica, 69, 1519–1554.
Phillips, P. C. B., & Perron, P. (1988). Testing for a unit root in time series regression. Biometrika, 75, 335–346.
Stărică, C., & Pictet, O. (1997). The tales the tails of GARCH processes tell Unpublished Working paper.
Su, C.-W. (2011). Non-linear causality between the stock and real estate markets of Western European countries: Evidence
from rank tests. Economic Modelling, 28(3), 845–851.
Wagner, N., & Marsh, T. A. (2005). Measuring tail thickness under GARCH and an application to extreme exchange rate
changes. Journal of Empirical Finance, 12, 165–185.