Master's thesis - Student Positive Awards

Master’s thesis
The impact of the Belgian political crisis
on Belgian company results:
evidence from stock market reaction
Supervisor: Pr Dr Gani Aldashev
Reader: Pr Dr Bertrand Candelon
Thesis presented by Thomas Jeegers
in order to obtain the titles of
Master 120 en Sciences Economiques
Orientation Générale - Finalité spécialisée
and
Master’s 60 in International Economics Studies
ACADEMIC YEAR 2011-2012
Economics School of Louvain/UCL · Place Montesquieu 3 · 1348 Louvain-la-Neuve
Economics School of Louvain/FUNDP · Rempart de la Vierge 8 · 5000 Namur
School of Business and Economics/UM · P.O. Box 616 · 6200 MD Maastricht

Acknowledgements
It is a sincere pleasure to thank all those who made this thesis possible.
First and foremost, I would like to thank Professor Gani Aldashev, my thesis supervisor,
for his expert advice on the different topics covered in the coming pages. His support during
the months needed to write this thesis, his encouraging comments and his suggestions were
of priceless value. I would also like to thank Professor Bertrand Candelon, not only as the
reader of this thesis, but also as my former Professor at Maastricht University, for having given
me the burning desire to analyse further many macroeconomic issues. With regard to this
thesis, his sharp understanding of the macroeconomic processes at work and his suggestions for
improvement provided me with expert guidance and enabled me to reach this level of precision
for the different econometric regressions.
In addition, I am grateful to Professor Pierre Giot for his helpful advice concerning the
financial aspects of the analysis driving this thesis and to Professor Paul Wynants for his detailed
and clear scrutiny of the Belgian political crisis and without whose help this work would not
have reached such relevancy. Their recommendations certainly contributed significantly to the
completion of this thesis and to the insights that were able to be drawn from it.
Moreover, I would like to thank my friends for their comments, for the time they spent on
reading the previous versions of this thesis and for their unflinching motivation in the many
projects we undertook together. Their contributions were so varied that I could not rank them,
so I will simply list them in alphabetical order. Grateful thanks to Bastien, Joris, Julien, Kevin,
Maxime, Stéphane, Xavier and last but not least, to Stefanie, my endless source of inspiration.
In addition to that, special thanks to my parents for daily challenging my perspectives for
the future and for their unconditional support over so many years.
Finally, I would like to thank all the people I have met during my travels around the world,
for having broadened my mind on global issues. It gave me the opportunity to take a step back
and think about the Belgian political difficulties from a different angle. I owe them intellectually
much more than I could possibly acknowledge. I hope that I will meet them again and be able
to bring them as much as they brought me.
Thomas Jeegers
ii

Abstract
In 2010-2011, Belgium experienced the longest political crisis of recent history. But how did the
Belgian stock market react to this crisis? Theory suggests that, under the assumption of market
efficiency, any political event should be anticipated and should have no impact on stock prices.
But are the markets really efficient in a period of crisis? Or do they respond to political events
once these events have actually occurred? These are the questions that we will be investigating
in the coming pages.
The Belgian 2010-2011 political crisis 1 gives us a unique opportunity to analyse the economic
impacts of political events in the context of a crisis. This analysis consists of an event study
built upon the Belgian political occurrences relevant to the outcome of the crisis. Unlike some
other similar approaches, we do not analyse the impact of isolated individual events. Instead, we
take advantage of the uninterrupted sequence of events that have shaken the Belgian political
scene by investigating how a series of political events impacts the main indicator of the stock
market of Belgium: the BEL20.
As was underlined by MacKinlay (1997), event studies are particularly useful for such kinds
of dissection of a period, since they reflect the impact of different types of events immediately in
stock and security prices. Abnormal stock returns are, therefore, usually central to most event
studies. However, it is not our purpose to focus on abnormal returns. The objective, instead, is
put on the process of building a relevant political indicator of the Belgian political crisis and on
the analysis of its effect on the daily returns of the BEL20 index, in a well-specified model, free
of endogeneity, of anticipatory movements and robust to heteroskedastic residuals.
Evidence of significant and robust results is provided. Methodical and punctilious, we
tried as much as we could to provide a thorough analysis reflecting the complex reality of the
political world. We place the Belgian political crisis of 2010-2011 in context, give details about
the selection of the events necessary to build the political indicator and justify why we take
particular events into account if their inclusion may be considered subjective. With regard to
the econometric regressions, intermediate steps and results are provided, most of the common
problems for least squares regressions are tested and corrected if necessary and different tests of
robustness are conducted. Finally, we extend the analysis to more sophisticated models in order
to fit the reality of the stock market as closely as possible.
Keywords: financial economics, political crisis, stock market reaction, BEL20, event study.
1 And still on-going at the time this thesis was begun.
iii

Contents
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
List of tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
List of figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
List of abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Introduction 1
1 Literature review 4
2 The Belgian context 8
2.1 Political aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.1 Insights into the Belgian political system . . . . . . . . . . . . . . . . . . 8
2.1.2 Brief review of Belgian political history . . . . . . . . . . . . . . . . . . . 9
2.1.3 Political risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Economic aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.1 Brief review of recent Belgian economic history . . . . . . . . . . . . . . . 12
2.2.2 Macroeconomic situation: some key features . . . . . . . . . . . . . . . . . 13
2.2.3 The BEL20 index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Insights into the 2010-2011 crisis . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3 Setting up the indicator 16
3.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2 Timespan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.3 The different political indicators . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.4 Computing the returns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4 Significant events 23
4.1 Politically significant events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.2 Rating agencies’ outlooks and downgrades . . . . . . . . . . . . . . . . . . . . . . 28
4.2.1 Standard & Poor’s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.2.2 Moody’s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.2.3 Fitch Ratings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.2.4 Conclusion about the ratings . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.3 Summary of the events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
iv

v
5 Impact of political events on the Brussels stock market 31
5.1 Theoretical background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.1.1 Efficient market hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . 32
5.1.2 Capital asset pricing model . . . . . . . . . . . . . . . . . . . . . . . . . . 32
5.1.3 Arbitrage pricing theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.2 Multiple regression analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.2.1 Simplistic return-analysis model . . . . . . . . . . . . . . . . . . . . . . . 34
5.2.2 Adding the main political indicator, free of endogeneity, to the model . . 36
5.2.3 Adding the control variables to the model . . . . . . . . . . . . . . . . . . 38
5.2.4 Heteroskedasticity: verification and correction . . . . . . . . . . . . . . . . 46
5.2.5 Autocorrelation: verification for both the returns and the residuals . . . . 48
5.2.6 Anticipation and late reaction effects . . . . . . . . . . . . . . . . . . . . . 49
5.2.7 Robustness of the political indicator: dropping the events, each in turn . 50
5.2.8 Robustness of the political indicator: alternative indicators . . . . . . . . 51
5.2.9 Robustness of the rating dummy: alternative indicator . . . . . . . . . . . 53
5.2.10 Summary of the main model . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.3 Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.3.1 Testing for unit root . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.3.2 Error correction model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.3.3 Generalised autoregressive conditional heteroskedasticity model . . . . . . 60
5.3.4 Exponential generalised autoregressive conditional heteroskedasticity model 63
6 Discussion 65
7 Conclusion 70
Appendices 73
Bibliography 84
Books . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
Articles in journals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
Online newspapers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Databases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
Others . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

List of Tables
5.1 Regression of the BEL20 daily returns, controlling for interdependence with the
European stock market index Euro Stoxx 50. . . . . . . . . . . . . . . . . . . . . 35
5.2 Instrumental variables (2SLS) regression of the BEL20 daily returns, using an ad
hoc estimator of the political indicator as the instrument. . . . . . . . . . . . . . 38
5.3 Regression of the BEL20 daily returns, adding the square of the daily returns of
the Euro Stoxx 50 index as a regressor. . . . . . . . . . . . . . . . . . . . . . . . 39
5.4 Regression of the BEL20 daily returns, adding an indicator of rating agencies’
modifications in outlook and downgrades as a regressor. . . . . . . . . . . . . . . 40
5.5 Regression of the BEL20 daily returns, adding as a regressor the daily change in
the yield of Belgian government bonds on the secondary market, with one-year
maturity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.6 Regression of the BEL20 daily returns, adding the predictability of the political
events as a regressor, measured as a three-day lagged estimation of the variance. 42
5.7 Regression of the BEL20 daily returns, adding the change in the interbank interest
rate (one-year maturity Euribor) as a regressor. . . . . . . . . . . . . . . . . . . . 43
5.8 Regression of the BEL20 daily returns, adding as a regressor the monthly growth
of M3 (free of endogeneity) as an estimator for inflation. . . . . . . . . . . . . . . 44
5.9 Regression of the BEL20 daily returns, on all selected control variables and adding
the squared residuals as a regressor for an heteroskedasticity check. . . . . . . . . 47
5.10 Regression of the BEL20 daily returns, on all selected control variables, with the
residuals robust to heteroskedasticity. . . . . . . . . . . . . . . . . . . . . . . . . 47
5.11 Regression of the BEL20 daily returns, using a one-day lagged version of the main
political indicator as a regressor. . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.12 Regression of the BEL20 daily returns, using a one-day anticipated version of the
main political indicator as a regressor. . . . . . . . . . . . . . . . . . . . . . . . . 50
5.13 Regression of the BEL20 daily returns, on all selected control variables, with the
residuals robust to heteroskedasticity and with separate indicators for positive
and negative events. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.14 Regression of the BEL20 daily returns, on all selected control variables, robust to
heteroskedasticity, using more political events in the political indicator. . . . . . 52
5.15 Regression of the BEL20 daily returns, on all selected control variables, with the
residuals robust to heteroskedasticity and using the “main political indicator 2”
purged from endogeneity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.16 Regression of the BEL20 daily returns on all selected control variables, with
the political indicator purged from endogeneity and the residuals robust to
heteroskedasticity, replacing the rating dummy by a more sophisticated indicator. 54
5.17 Vector error correction model, with the residuals corrected for heteroskedasticity. 59
5.18 Generalised autoregressive conditional heteroskedasticity regression. . . . . . . . 62
5.19 Exponential generalised autoregressive conditional heteroskedasticity regression. . 64
vi

A.1 Final results of the main model, corrected for endogeneity and heteroskedasticity:
2SLS regression, using a generated instrumental variable for the political indicator. 83
A.2 Final results of the main model, corrected only for heteroskedasticity: basic OLS
regression, robust to an arbitrary form of heteroskedasticity. . . . . . . . . . . . . 83
vii

List of Figures
2.1 Belgian Public Debt (1981 - 2010). . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.1 Main Belgian political events against the BEL20 and Euro Stoxx 50 indices daily
returns, April 19, 2010 - December 31, 2010 . . . . . . . . . . . . . . . . . . . . . 30
4.2 Main Belgian political events against the BEL20 and Euro Stoxx 50 indices daily
returns, January 01, 2011 - December 13, 2011 . . . . . . . . . . . . . . . . . . . 30
A.1 Constituency Brussels-Halle-Vilvoorde. . . . . . . . . . . . . . . . . . . . . . . . . 81
A.2 Frequency distribution of the daily returns of the BEL20 over the period: April
19, 2010 - December 13, 2011. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
viii

List of abbreviations
2SLS:
Two-stage least squares
APT:
Arbitrage pricing theory
ARCH:
Autoregressive conditional heteroskedasticity
BHV:
Brussels-Halle-Vilvoorde
CAPM:
Capital asset pricing model
CD&V:
Christen-Democratisch en Vlaams
cdH:
Centre Démocrate Humaniste
CRSP:
Center for Research in Security Prices
CVP:
Christelijke Volkspartij
DF-GLS: Dickey–Fuller generalised least squares
DW:
Durbin-Watson
ECM:
Error correction model
EGARCH: Exponential generalised autoregressive conditional heteroskedasticity
EMH:
Efficient market hypothesis
Eonia:
Euro overnight index average
Euribor: Euro interbank offered rate
FDF: Front Démocratique des Francophones (before January 2010)
FDF: Fédéralistes Démocrates Francophones (since January 2010)
FFR:
Federal fund rate
GARCH: Generalised autoregressive conditional heteroskedasticity
IID:
Independent and identically distributed
IV:
Instrumental variable
Libor:
London interbank offered rate
MR:
Mouvement Réformateur
MLE:
Maximum likelihood estimate
N-VA:
Nieuw-Vlaamse Alliantie
OLS:
Ordinary least squares
Open Vld: Open Vlaamse Liberalen en Democraten
PS:
Parti Socialiste
RW:
Rassemblement Wallon
S&P:
Standard & Poor’s
sp.a:
Socialistische Partij Anders
Std. Err.: Standard error
VB: Vlaams Block (before November 2004)
VB: Vlaams Belang (since November 2004)
VECM:
Vector error correction model
WLS:
Weighted least squares
ix

Introduction
“The goal in the end is not to win elections. The goal is to change society.”
Paul Krugman, 2012
In 2011, Belgium became “world champion” as the country with the most extended political
crisis, when it took from Iraq the record of the longest period in times of peace without a
federal government. From April 26, 2010, the resignation date of the Leterme II government,
to December 6, 2011, when Elio Di Rupo took the oath as the new Prime Minister, the 589
days 2 during which Belgium lacked a government made of this period a unique, harmful and
interesting political crisis to analyse.
The behaviour of stock markets in difficult political periods or times of war has received a lot
of attention in economic and financial literature over the past 20 years. As identified by Fama
(1990) and evidenced over a longer period by Schwert (1990), more than 40% of the variations
in price on the stock market cannot be explained by real economic activity. Many authors have,
therefore, tried to identify the other variables that can explain the variations in stock prices,
within which the political events are a substantial part. In this paper, we add to this literature
by covering an issue which, to the best of our knowledge, has not been analysed before, i.e. the
impact of the Belgian political crisis (2010-2011) on the Brussels stock exchange main index;
the BEL20.
A stock market approach to the behaviour of a government concedes that the policies it
implements are essential in determining the environment in which investors will conduct their
transactions. These transactions will in turn collectively influence the price, and thus the return,
of a firm’s stock. In the Belgian case, a temporary government was in place during the 2010-2011
crisis, but its purpose was to take care of current and urgent affairs and it was not supposed
to deal with important topics that had been stalemated while the previous government was in
power. Therefore, the environment for investors was, at best, unstable and likely to provoke
uncertainties. Empirical results confirm this instability since we find evidence for significant
reaction in the stock prices concurrent to major political events, during the crisis.
The logic behind event studies 3 is often based on the efficient market hypothesis (EMH),
according to which all available information and expectations are reflected by the prices of the
market 4 . A political event study consists of analysing the reaction of the market (typically the
stock market) after one or several political events that are likely to have an impact on stock
prices. These reactions are typically evaluated through the analysis of abnormal returns within
the period of interest. A literature review on the topic seems to indicate that political event
studies reach more significant results when the political environment is unstable (i.e. during
2 If we consider the federal election of June 13, 2010 as the starting date, as many sources do, then the length
of the crisis amounts to 541 days.
3 See Wells (2004).
4 For further information, see section 5.1.1 or Fama (1970).
1

2
political crises, in highly politically risky environments or in emerging markets). However,
depending on the periods and the markets analysed, evidence is found for both significant effects
of political events on the stock markets and non-significant effects.
The “event window”, as defined by MacKinlay (1997) for an event study, is “the period
over which the security prices of the firms involved [...] will be examined”. In our case, the
key beginning and ending dates are April 26, 2010 and December 6, 2011, as this is the period
characterised by the absence of a government in Belgium. As pointed out by Zach (2003),
not much literature shows direct evidence of a link between political events and stock market
fluctuations. Nevertheless, this is mainly due to the fact that it is typically hard to quantify
political change. This literature, though, can be split up into short-window event studies and
long-horizon studies. Our analysis falls within the first of these groups. Further categorising
our approach, we build an ad hoc political indicator, slightly more sophisticated than a dummy
variable. This approach has not been extensively covered in literature but seems to fit the case
at hand very well.
The aim of this thesis is dual. On the one hand, we want to make the thesis conclusions
useful for any possible future political crisis. On the other hand, the procedure can be reiterated
for another country, in order to evaluate to what extent the stock market depends on the political
situation. Such information is particularly interesting for potential investors since, as mentioned
in the literature review, some countries are more subject than others to strong variations in
stock returns after significant political events.
The procedure consists of a multi-factor least squares approach, fitting a regression of the
daily returns 5 of the BEL20 index on an indicator of political events. We use several control
variables in the process, in order to explain a bigger proportion of the BEL20 fluctuations and,
therefore, to decrease the variance of the political indicator. We find interesting and somewhat
counterintuitive results. The negative political events (not in favour of the formation of a
government) have a positive impact on the stock market, on the day of the event, and the
positive events have a negative impact. These results are robust to the specification of the
model and remain valid even when we drop some of the political events. Our results show a high
level of significance (within a 2% threshold) and are relatively free of endogeneity, anticipatory
movements and heteroskedasticity.
An empirical work in financial econometrics is never easy to conduct. Every econometric
model has some drawbacks and some weaknesses. While simple models have the benefits of clarity,
which led to their extensive use in literature, their limitations have often been reported. It made
econometricians establish more sophisticated models, which are an interesting complement to the
basic approaches, but they often need many assumptions to fit the complex reality of the political
and financial worlds. In addition, econometrics is a discipline in which a causal relationship can
never be found. We can only draw conclusions on causal inference, that is, a likelihood that
a causal relationship between two or more variables exist. However, these weaknesses do not
discard the econometric approach, as techniques exist to release the implausible assumptions, to
correct for possible bias and ultimately to adequately fit reality.
The source of the data for stock prices is Datastream (2012); the different indicators that we
use as control variables come from the National Bank of Belgium (2012) or from Datastream
(2012); the source of the data for the political events is Faniel (2011), completed by personal
researches and with the help and supervision of Professor Paul Wynants, who teaches Belgian
5 We decided to restrict our attention to an analysis of the daily returns of this index, in order to keep the
focus on one single idea and to conduct a proper investigation into it. We initially also investigated parallel topics,
such as the impact of the political events on the volatility of the BEL20 or on the daily volume of transactions.
But we finally disregarded these additions, because they are of little added value while excluding them allowed us
to keep the light shed on one key aspect of the financial market.

3
Politics at the University of Namur (Belgium).
In chapter 1, we start off by reviewing the state of the art concerning political event studies
and dummy variable event studies, whose combination is the methodology used in this thesis.
Chapter 2 provides the reader with the necessary background knowledge about Belgium and its
political system and history, leading to a better understanding of the crisis, of the events that
precipitated it and ultimately of the reason why their impact on the stock market is significant.
In chapter 3, the methodological and practical aspects of the procedure are covered. The detailed
analysis of the political events to be investigated is done in chapter 4. Finally, the econometrical
work is performed in chapter 5. This chapter covers the theoretical foundations of stock return
analyses for the model that we investigate and also covers some extensions to this model. The
complexity of the econometric work is progressive: we start with an overly simplistic model that
gets more and more sophisticated as we go along. We add one by one seven control variables
and justify their addition to the model. We correct for endogeneity of the political indicator
and for heteroskedasticity of the residuals. We also test the robustness of our results according
to different approaches. Finally, the extensions include the application of an error correction
model and of autoregressive conditional heteroskedasticity models. We finish the thesis with
a discussion about the relevancy of our methodology. This is done through a series of short
reviews, each concerning a particular topic raised in this study and by establishing the basis for
possible further extensions of the current work.

Chapter 1
Literature review
Event studies are a common procedure in economic, accounting and finance literature, measuring
the impact of an event of interest on some economic variables. The first event study that can be
found in literature probably dates back to the interwar period with Dolley (1933), analysing the
effect of stock splits on stock prices. Over the following three decades, the procedure of event
studies became slightly more sophisticated, including control variables in the analyses. The event
study literature during this period was characterised by authors such as Myers and Bakay (1948),
Barker (1958) and Ashley (1962) 6 . In the late 1960s, event studies started to be commonly used
in relevant literature, with Fama et al. (1969) 7 sparking this habit (Pynnönen (2005)). In the
following few years, the aim of this kind of studies was focused on the fluctuations of security
prices (Binder (1998)). Later, other variables of interest were used in this type of literature,
such as the exchange or interest rates. Event studies were often used to measure the effect of
corporate events on wealth or to test the efficient market hypothesis 8 and they blossomed in
such literature, with at least 565 papers in relevant finance journals between 1974 and 2000
(Aktas et al. (2007)). The results of previous event studies have shown that stock prices respond
to news about corporate control and about policies and macroeconomic conditions that are
likely to affect fundamentals of the economy.
The link between political events (exclusively) and share prices was left aside in literature
until the late 1980s. In the same period, other fields started to be analysed by event studies,
such as insurance literature (Henderson (1990)). Some of the political event studies have been
performed with positive results, under some circumstances and during certain periods. In
this chapter, we review some intriguing papers considering political event studies (mainly the
short-window ones), since this is the procedure that we follow in the remainder of the paper.
Other authors, their procedure and their results will also be introduced in further chapters
(especially in chapter 5) of this thesis, where we investigate sharper financial and econometrical
models 9 . Some papers in the coming list are not strictly restricted to political events, but all of
6 Cited by MacKinlay (1997), page 14 and by Campbell et al. (1997), page 150.
7 Fama et al. (1969) used the term “event study” for the first time, describing a new methodology for measuring
the adjustment process of stock prices. Their methodology is very similar to contemporary event studies and
their paper was cited more than 500 times in the following 25 years, showing that their trailblazing approach was
the basis behind the many following event studies.
8 They are called “market efficiency studies” and evaluate the rapidity and accuracy of market reactions to new
information. The EMH states that all available information and expectations are reflected by market prices, and
that any news is integrated without delay into these prices. We investigate the EMH into more details in section
5.1.1.
9 This choice was made for two reasons; firstly it allows us to keep the focus on the link between politics and
the stock market throughout the first four chapters, rather than entering financial and econometrical models
before alluding to the context or explaining our procedure. Secondly it enables the reader to link theory with
empirical evidence of the case under scrutiny, in chapter 5.
4

5
them include at least political events as a potential cause for stock price fluctuations. Moreover,
some studies are not directly applicable to Belgium, as they focus on the impact on the stock
market of elections and of the party holding the power. We will not cover that kind of analysis in
this thesis, because Belgium is a particracy (see section 2.1.1 for further details), which implies
that no party fully controls the political scene. Therefore, the approach of using either the
impact of elections or the changes in the main party in the country makes less sense in the
case of Belgium, since the impact is likely to be more limited in a country with this political
structure. However, the results of this literature are worth mentioning, since they evidence the
existence of a relationship between politics and the stock markets. As political event studies
do not show a strong tendency of being cumulative 10 , we cannot build a proper “history” of
this art, i.e. how the procedure evolved over the years. However, most papers rely on the basic
approach of Fama et al. (1969), which has become the most accepted standard in event studies.
Since the cumulative approach is not possible for this topic, we present the existing papers on
the topic in a chronological way.
Niederhoffer (1971) analyses how world events, including some political events, affect the
movements of the stocks in the S&P 500 index. These events are selected according to the
magnitude of their headlines in the New York Times, measured according to the number of
columns in the newspaper. His results show that on days following world events, the returns are
more extreme than on other days. It evidences that the selected events have an impact on the
fluctuations of the stocks included in the S&P 500.
Huang (1985) shows that the party affiliation of the U.S. president yields a difference of
annual rates of return on stocks, concluding that those returns are higher under Democratic
presidents over the period 1929-1980. He also discovered a pattern of four-year cycles in stock
returns in the United States, corresponding to the presidential terms. His result was confirmed
over the period 1888-2001 by Siegel (2002) 11 .
Cutler et al. (1989) investigate the impact of political and world events on the Dow Jones
stock index, using monthly stock returns from 1926 to 1985. They discovered a small and
non-significant impact of main events on stock returns, but their results show a higher volatility
of the returns on days characterised by a major event, on average, than on other days.
Kim and Mei (1994) measure the impact of political risk on asset prices in Hong Kong, over
the period from 1989 to 1993. They show that adding a political risk dummy “significantly
increases the explanatory power of return regressions” and that China’s political policy has an
effect on the stock market of Hong Kong. They also model the fluctuations in volatility as a
function of political events.
Chan and Wei (1996) investigate further this work, focusing on the impact of occurences
of political events on the stock market of Hong Kong. They find an impact on the volatility
of the stocks of several indices. The impact on the returns of the stocks is not significant for
all indices, however. For the index for which they find significance, the direction of the shock
after political events is opposed to what we found for Belgium: they find a positive impact after
favourable events and negative impacts after unfavourable events.
Pantzalis et al. (2000) analyse, from 1974 to 1995 and through an event study, how the
stock market reacts before and after political election dates. Using indices of 33 countries, they
conclude that the stock returns in the weeks preceding the political elections are abnormally
higher than usual. They also find more significant results for countries for which the press has
less freedom.
10 “Cumulativeness” in literature is the fact that each author can rely upon his predecessors and build his
work upon existing literature. Hedges (1987) argues, for example, that literature in physical science is usually
cumulative, while the cumulativeness in social sciences is usually much smaller.
11 Cited in Jones and Banning (2009).

6
Zach (2003) finds evidence that in Israel, returns on stocks are more extreme on days that
follow political events than on days that do not. The fact that Israeli stocks traded only in the
United States do not show such extreme returns make it even more likely that it is indeed the
political events that are the cause of the extreme stock returns movements in Israel.
Abadie and Gardeazabal (2003) try to measure the economic cost of political conflicts
(terrorist attacks) in the Basque country, notably through the stock market. They find significant
and robust evidence that the stocks of the Basque country comparatively underperform during
unstable periods, while they outperformed the non-Basque stocks during the cease-fire period of
1998-1999.
Leblang and Mukherjee (2004) refer to the financial impact of electoral systems, elections,
partisanship and political uncertainty. They analyse how the stock market responds to public
signals. Another point of their argumentation is the fact that when an electoral victory of a
left-wing party is anticipated, then the volatility of stock prices increases 12 .
Aquino (2004) reviews the impact of major events in the Philippines during the period from
July 1987 to May 2004. His results confirm that the stock market is influenced by political and
economic events, and even by natural phenomena.
Chen et al. (2005) analyse further emerging markets, with the case of Taiwan. Using simple
least squares models to analyse the impact of political events on the stock market, they find
significant abnormal returns following these events but using a multivariate regression model,
this evidence is attenuated (except for some important events).
Jensen and Schmith (2005) study the effect of the election of a politician on the stock market
in Brasil. One result they find, which is an expected one, is that electoral uncertainty increases
stock market volatility.
Aktas and Oncu (2006) test the efficient market hypothesis. Their test consists of measuring
the returns of different portfolios of the Turkish stock market after main events in Turkey,
including political events. Their results clearly show no overreaction or underreaction from the
investors. The assumptions of the EMH are, therefore, not violated with this study.
Romero-Meza et al. (2007) investigate the cause of the nonlinearities in the stock market
index in Chili. They call it a “reverse form of event study”, since they search for economic
and political events explaining the stock market index fluctuations. They find that the events
explaining the abnormal fluctuations are mainly regulatory changes and international events.
Benton (2008), along the same lines as Jensen and Schmith (2005), analyse the impact of
campaign commitments on the stock market in Mexico. They find that investors react negatively
to electoral uncertainty, while no significant move is found after electoral commitments of the
candidates.
Ali et al. (2010) assess the short-run stock overreaction in Malaysia. They use political events
to represent domestic events because of the strong political involvement in Malaysia, implying
interconnections between business and politics. Their findings confirm that the Malaysian stock
market overreacts significantly to economic events, which is coherent with the results of Bilson
et al. (2002), showing that, in emerging markets, political risk is an important explanatory factor
in the returns’ fluctuations.
Milyo (2012) summarises previous event studies on possible political corruption, finding
evidence that firms located in the neighbourhood of powerful political actors in the US are
subject to favoritism and are, therefore, more likely to show abnormal returns.
Junqué de Fortuny et al. (2012) conduct a different type of event study. They analyse the link
between political events and political news during the Belgian political crisis 2010-2011. Their
approach is based on an analysis of the opinions contained within the articles of the Flemish
12 Established notably by Herron (2000), Cohen (1993), Alesina et al. (1997), Freeman et al. (2000) and Gemmill
and Saflekos (2000).

newspapers, over the 541 days of the crisis (June 13, 2010 - December 6, 2011). Detecting
the sentiments of 68 000 articles through fully computerised “opinion mining”, they find that
newspapers show peaks of negative or positive opinion, during the hottest moments of the
negotiations. Their approach is thus very different from ours, but their conclusions somehow
complement ours, since they find a correlation between newspapers opinions and the key moments
of the crisis. Since we mainly use newspapers as references for the establishment of a political
indicator (see chapter 4), the work of Junqué de Fortuny et al. (2012) seems to confirm that
newspapers are indeed a reasonably good reflection of the political process in progress.
As one can notice, most of these analyses concern politically unstable or emerging countries.
It is worth mentioning it, since the political instability of the Belgian political scene might cast
doubts on the efficiency of its stock market. Now that we have reviewed the results of authors
on the existing relationship between politics and the stock market, we will dig into the details
of the Belgian structure, both economically and politically. Investigating the particularities of
Belgium will enable the reader to recognise the importance of this relationship for the Belgian
case.
7

Chapter 2
The Belgian context
“There are in Belgium, the Walloon and the Flemish; there are no Belgians.” 13
Jules Destrée (1963)
To gain an understanding of the events of the Belgian political crisis, some background
knowledge about this particular country is needed. In this chapter, we review in turn the political
background in which the longest governmental crisis of history emerged, some key fundamentals
of the Belgian economy and finally, some insights into the political stalemate of 2010-2011.
2.1 Political aspects
In order to correctly understand the political crisis that shook Belgium from April 2010 to
December 2011, one needs to understand how Belgium is ruled and especially, the particular
history of the country. This section is heavily based on Delwit et al. (1999) and Deschouwer
(2009), which are two of the few references giving the opportunity to understand how the Belgian
political system has been forged over the years. We first briefly review the Belgian political
system, then we draw attention to some milestones in the political history of Belgium. In no
way do we claim to be exhaustive about any of the topics raised in this section; the objective is
rather to provide the reader with the necessary background knowledge about Belgian politics to
fully understand the extent and the interpretation of the analysis at hand.
2.1.1 Insights into the Belgian political system
“In Belgium, the political system reflects, as elsewhere, the social complexity.”
Pascal Delwit et al. (1999)
Understanding the underlying reasons for the political crisis in Belgium is a tough task if one
does not know how complicated the Belgian political structure is. Based on a federal parliamentary
democracy and under a constitutional monarchy, Belgium has one central government in
addition to which there are three regions, each with a government responsible for many essential
matters, such as education, environment or public health. These regions are the Walloon Region,
the Flemish Region and the Region of Brussels-Capital. A federal state is based on the rule of
equity. In Belgium, this rule of equity exists, but only within each community or each region
(Delwit et al. (1999)).
13 Originally in French: “Il y a en Belgique, des Wallons et des Flamands; il n’y a pas de Belges.”
8

9
There are three communities; Flemish, French and German-speaking (one for each of the three
national languages; Dutch, French and German), but in each of these communities, municipalities
offering their public services also in some or all of the other national languages exist. To be
exact, 25 municipalities benefit from these language facilities 14 . In addition to the three regions
and the three communities, Belgium is also divided into four language areas; the Dutch language
area, the French one, the German one and Brussels, which is bilingual (French/Dutch).
The form of government in Belgium is based on group representation, a typical form for very
divided societies, called “consociationalism” (Deschouwer (2009)). This scheme is designed to
avoid conflicts and requires the parties involved to agree or to face a standstill for the matters
concerning them all. Yet, the parties are usually given a large autonomy. Belgian federalism can
be termed confrontational federalism 15 . Indeed, the two big linguistic communities make it hard
to manage the equilibrium. Belgium is, therefore, subject to continuous identity and community
crises.
The proportional representation characterising the Belgian political scene has led to a
“particracy” system, a government where different parties have a significant influence on the
political process. In Belgium, the parties are the most influential political actors (Deschouwer
(2009)). Finally, three historical political pillars exist; Social Christian, Socialist and Liberal 16 .
2.1.2 Brief review of Belgian political history
“When Belgian independence was declared, language reflected social-class differences more than a
cultural cleavage.”
Robert Mnookin (2007)
Since its foundation, Belgium had a composite population. Built as a buffer between different
rival countries and not according to a nationalistic movement as was the case for Italy or Germany,
it has long been considered as divided between the Flemish and the Walloons. Belgium has a
long history of internal conflicts, however these never became violent.
Broadly speaking, during the first century of its existence, the elite in Belgium were all
French-speaking, no matter whether they were Flemish, Walloon or inhabitants of Brussels.
During the 20 th century however, the Flemish grew in importance, as Flanders rose. In 1970,
the federalisation of the State took place with the first State reform, establishing cultural
communities, at the behest of the Flemish.
The political dualisation of Belgian society dates from the late 19 th century 17 , when two
political structures developed; one Catholic, the other socialist. It was only in 1954 that a party,
the Christelijke Vlaamse Volksunie (fast abandoning the Christian denomination), was built
around a federalist structure. They revendicated a two-sided federal State, with Brussels as
federal district. Its growth was important in the 1960s and the early 1970s while another party
grew in parallel; the Front démocratique des Bruxellois (later renamed Front démocratique des
Francophones (FDF), then Fédéralistes Démocrates Francophones in January 2010), fighting for
the rights of the French speakers in Brussels. The third regional party to rise during this period
(in addition to the VU, in Flanders, and to the FDF, in Brussels) is the Rassemblement Wallon
(RW), in Wallonia. The RW was established in 1965 to protect the interests of a declining
Wallonia, but this party no longer exists today. There was a turning point in the late 1960s,
14 Figure A.1 on page 81 illustrates the proportion of native French-speakers in the Brussels-Halle-Vilvoorde
(BHV) constituency. One can notice on this figure how diverse the population in the area around Brussels is.
15 Originally “fédéralisme de confrontation” by Delwit et al. (1999), p.57.
16 The parties representing these pillars notably established together the consociational logic ruling the country,
after World War I (Deschouwer (2009)).
17 The coming paragraph is based on Delwit et al. (1999), pages 113 to 153.

10
when the decline of the historical parties started, leaving room for another type of dualisation on
the Belgian political scene. The Christian Social Party 18 was the first one to break up because
of internal disagreements between the Flemish and French-speaking sides. The move of the
French-speaking section of the Catholic University of Louvain from Flanders to Wallonia, in
1968, was major evidence of the split. The political environment of the 1970s saw another
important change; the existence of a different dominant party in each of the three regions (the
Socialist Party in Wallonia, the FDF in Brussels and the CVP in Flanders).
In the early 1980s, two ecologist parties were founded in Belgium; Agalev (today known
as Groen!) and Ecolo. In addition to these new ones, at about the same period, Belgium
also saw the establishment of far-right parties. In Flanders, in 1979, it was the Vlaams Block
(VB), built from a cartel between two Flemish parties; the Vlaams Nationale Partij and the
Vlaamse Volkspartij. Distinguishing features of the Vlaams Block were notably an elitism based
on nationality and a strong Flemish patriotism. This party quickly rose in Antwerp, then in the
whole of Flanders. In the French-speaking Community, the Front National is another far-right
party, founded during the same period. It grew over the following few years, but it never held a
substantial share of the government.
The 1990s, and in particular the elections of 1991, fragmented the Belgian political scene.
Many small parties rose, while the three historical Belgian political families lost a significant
share of their voters. This fragmentation increases all along the decade (Delwit mentions,
in Jaumain (1997), a record number of candidates for the elections of 1995) and reaches its
maximum at the turn of the century (Delwit (2003)). The years 1996-1997 know a deep evolution
of the Belgian structure, as a consequence to the shocks and traumatisms of Belgian society
during this period (Mabille (2000)). A new era begins for the country, implying a different way
to look at the political scene.
The tensions between the French-speaking and Flemish Communities did not decrease in
the 2000s. In October 2001, a new centre-right Flemish nationalist party, the Nieuw-Vlaamse
Alliantie (N-VA), was replaced from its predecessor, the Volksunie. The N-VA craves an
independent Flanders, which would be part of the European Union. In 2004, the Vlaams Block
renamed itself Vlaams Belang (Erk (2005)). Together, the VB and the N-VA form the main
secessionist right-wing Flemish parties, wishing for Flanders to be directly represented in the
European Union (Rochtus (2010)). With the rise of the N-VA, a new page has been turned,
since a democratic Flemish-nationalist party enjoys success based on its secessionist objectives,
as Rochtus (2010) states, “contrary to the Vlaams Belang which is assumed to thank its previous
successes to its xenophobic slogans”. The leading role of the N-VA on the political scene in the
late 2000s is largely attributed to its main charismatic figure and leader: Bart De Wever. He
also played, as we will see, a meaningful role in the 2010-2011 Belgian political crisis.
Since its establishment, the country has undergone six State reforms, each of which took place
within the last 42 years; in 1970, 1980, 1988-1989, 1993, 2001 and 2011. Let us briefly overview
the main modifications resulting from each of these reforms 19 . In 1970, the three “Cultural
Communities” were established, corresponding to the current Communities of the country.
The National Parliament was divided into two language groups and a French-speaking/Dutchspeaking
parity was imposed on the Council of Ministers (with the exception of the Prime
Minister). Each language group could also, from this reform on, resort to an “alarm bell” if
a proposed law would hurt its interest or disturb the good functioning of the country 20 . In
1980, the power of the two main regions was enhanced, allowing their decrees to have the same
18 “Christelijke Volkspartij” (CVP) and “Parti Social Chrétien”, respectively in Dutch and in French.
19 We are mainly based on Rochtus (2010) for this part of the historical analysis.
20 In april 2010, this procedure was used three days after the fall of the government (see section 4.1 for more
details).

11
value as federal law. The “Cultural” denomination of the Communities was abandoned, as their
competences increased. In 1993, the Constitution was amended, giving more freedom to the
Regions and Communities. The reforms of 2001 and 2011 gave even more power to these Regions
and Communities and the recent reform also established the split of the BHV constituency.
From 1991 to 2007, four governments reached the term of their four-years mandate, testifying
to a relative political stability (Deschouwer (2009)). From June 2007 to December 2007, the
country was shaken by another political crisis. This one was based on the question of whether,
and to which extent, another State reform was necessary. The two opposing views on the topic
were held by the Flemish, on the one hand, claiming that reforming the State would lead to a
better governance of the country, and the French-speaking parties, on the other hand, fearing an
eventual split of the country (Rochtus (2010)). The years following this period were characterised
by political instability, as several governments (including some “interim governments”) were
in place: Verhofstadt III, Leterme I, Van Rompuy I, Leterme II. Doubts have been expressed
about the relevancy of Belgian federalism, notably because of the series of state reforms and the
continuous Belgian political instability. It is beyond our scope to analyse whether these doubts
are justified. Instead, we will investigate the impact of this political instability on the stock
market. But before doing so, we investigate Belgium’s political risk and economic background,
to which we now turn.
2.1.3 Political risk
Though this complex structure and wavering history might darken the functioning of the country,
the political risk associated with the country is actually in the same category as other West-
European countries. According to the PRS Group (2011), establishing yearly country reports
for a hundred countries all over the world, the political risk of Belgium is not especially high. In
particular, in 2010, Belgium had the twentieth lowest political risk in the world, halfway between
its neighbours the Netherlands and France (respectively the world’s 5 th and 34 th lowest), and
equivalent to the one of Germany. The PRS Group forecasts the risk of turmoil in international
business, which is based on probable regime scenarios and on types of government interventions
and it has estimated Belgium as a “low risk” country.
Rating agencies play a large role in estimating a country’s risk. Those agencies are of even
greater interest for the purpose of this thesis, since agencies’ notations have a direct impact
on the price of a country’s company stocks. The three main rating agencies are Standard and
Poor’s, Moody’s and Fitch. As an example, in mid-December 2010, Standard & Poor’s warned
the Belgian interim regime that it might downgrade Belgium if the setting up of a government
continued to be delayed (L’Echo (May 7, 2011)). This threat remained pending for almost a
year but was finally executed at the end of November 2011; Belgium was downgraded from
“AA+” to “AA” in S&P’s notation (L’Echo (November 26, 2011)). Such kind of events will be
included in our analyses of the stock market (see section 4.2 for further details).
2.2 Economic aspects
In this section, we review some macroeconomic indicators of the state of the Belgian economy,
at the beginning of the crisis (2010). We also introduce some background about the economic
history of the country, in order to better appraise the current situation.

12
2.2.1 Brief review of recent Belgian economic history
The context leading to the multiple crises and reforms of the past few decades in Belgium is
intrinsically linked with the economic background of the country. Let us briefly review some of
the main features of it, with a focus on the regional development of Belgium.
Generally speaking 21 , after World War II, Belgium experienced thirty years of growth with
low employment. This rise suddenly stopped at the end of the 1970s, as a consequence of
the oil crisis and the increase in prices. The change was also characterised by the end of the
Bretton-Woods era, and therefore the end of the convertibility of the American dollar into gold
(August 15, 1971). The oil crisis led to a situation of stagflation in Belgium during the mid-1970s,
notably because of its foreign dependence in the energy sector. This crisis significantly shook the
Belgian economy, and is a considerable break in its history. In the 1980s, a social era began in
Belgium, bringing with it higher wage demands and therefore inflation. In face of this rampant
inflation, a minimum level of unemployment was used by companies, as a way to cap wage
demands. Unemployment rose from 2% to 11% from 1970 to 1982. In 1982 and 1984, different
mechanisms were used to help the economy to bounce back, including a devaluation of the
Belgian franc. As a result, Belgium recovered some competitiveness but the imposed austerity
resulted in an increase of the public-sector debt while investment fell. The drop in investment
was also due to the sudden rise of interest rates, giving incentives to the owners of capital
not to invest in the real economy but in monetary markets. In the 1980s, as a consequence
to the baby-boom, a strong intensification of unemployment arose. The following years are
characterised by a Europeanisation process and more generally by globalisation, allowing the
country’s economy to bounce back. With regard to Belgian monetary policy, it was relinquished
to the European System on June 1, 1998 and the Euro was adopted on January 1, 1999.
If we turn to the picture within the country 22 , we can notice that Wallonia was dominant (in
terms of regional GDP) until the 1960. In this period, simultaneously Flanders caught up with
Wallonia and the industrial sector, mostly constituted of coal mining and the metal industry
and mainly present in Wallonia, started declining. This fall, affecting most industrial regions of
the European Union, brought about many job losses. Flemish politicians were claiming more
autonomy for the Regions, in order for Flanders to stop subventionning Wallonia. From the
mid-1980s to the mid-1990s, the growth of Flanders was higher than that of Wallonia and it
resulted in a much higher GDP in the northern part of the country. This period represented a
significant modification in the development of the country; in particular, Flanders previous to
the 1950s was known to be mostly agrarian, while Wallonia was already more industrialised.
From the end of the 1990s to the end of the 2000s, the growth rate of both regions did not
differ widely, but their share in the Belgian GDP showed a steady gap. Over the same period,
the share of Wallonia in the Belgian GDP fluctuated from 23% to 24%, while that of Flanders
was between 57% and about 58% 23 . For the year 2005, if we normalise the EU27 PIB/capita
average at 100, Flanders lies at 120.1 while Wallonia reaches only 87.5 24 . Finally, Quevit (2010)
concludes that the regional transfers and the opinion of Wallonia firmly established in Flanders
are largely exaggerated and do not correctly reflect the reality.
21 The coming paragraph mainly relies on Savage (2004) and Maes (2010).
22 The coming paragraph mainly relies on Quevit (2010) and Rochtus (2010).
23 Source: own computation based on Eurostat database (European Commission (2012)).
24 Source: Eurostat 2007, cited in Quevit (2010).

13
2.2.2 Macroeconomic situation: some key features
Globally speaking, the Belgian economy performs well, but different negative trends imply
potential risks for Belgium (European Commission (2012)). Let us investigate some key macroeconomic
features of the country, that are of interest to help us comprehend the situation of
Belgium for the coming analysis. We review first the public-sector debt of the country, which is
key to understanding the reasons behind the downgrades of the main rating agencies. Then, we
turn to a problem that might cast some doubts on the future growth of Belgium: its decreasing
competitiveness due to its labour market. Finally, we review an important element of Belgium’s
international positioning: its trade openness. As for the section about the political system
of the country, we do not aim to be exhaustive, but just to briefly cover some aspects of its
macroeconomic position, in order to be able to grasp the in-depth analysis of the following
chapters.
Considerable public-sector debt
Peeking at the 2010 financial situation of Belgium, one will notice a public-sector debt burden
problem. The European Commission identifies it as the main source of internal imbalances for
Belgium (European Commission (2012)). The IMF remind us that at the beginning of 2011,
Belgium had the third largest public debt-to-GDP ratio in the Eurozone, just after Greece and
Italy (International Monetary Fund (2011)). At the end of the year it became the fifth largest
(with Ireland and Portugal coming ahead) (European Commission (2012)). The Belgian public
debt has lain above 88% of GDP for 30 years (with the exception of 2007).
Figure 2.1: Belgian Public Debt (1981 - 2010).
Source: own computation based on National Bank of Belgium data.
We can see on figure 2.1 that public debt reached its peak at 137.8% of GDP in 1993 and
decreased regularly from then until 2007, when the tendency was inverted. It then climbed back
to reach 95.9% in 2010 (and 98.2% in 2011). It is also worth noting that only 3% of this debt is
denominated in foreign currencies and that the largest share (about 90%) of this burden is held
by the federal government (International Monetary Fund (2011)).
Even though its public sector debt is considerable, Belgium was well rated by the main
rating agencies, before the 2010-2011 crisis: just one grade under the top AAA-rating. As we
will see, this grade was reviewed by all three rating agencies in 2011 and 2012.

14
Weaknesses in its external competitiveness
Over the decade before the financial crisis of the late 2000s, Belgium progressively lost an
important share of its external competitiveness (European Commission (2012)). In 2011-2012,
Belgium is ranked 15 th out of 142 by the Global Competitiveness Index, but has some weaknesses
with regards to its labour market efficiency. In particular, it is ranked 129 th for its flexibility
of wage determination and 131 th for its hiring and firing practices (World Economic Forum
(2011)). The European Commission reports that these weaknesses are due to the restrictive
labour regulation of Belgium: a rigid labour market, combined with a “highly structured wage
bargaining system” (European Commission (2012)). The IMF testifies that Belgium has lost,
like other developed countries, market share (due to its lack of competitiveness), but this is not
particular to the Belgian case (International Monetary Fund (2011)). However, the IMF also
establishes that the loss of Belgian competitiveness was caused by high labour costs.
Trade openness
After World War II, Belgium greatly expanded its trade relations beyond the national border
(Maes (2010)) and became a largely open economy. It explains why “[t]he Belgian economy was
harder hit by the international economic crisis than some other countries (Buyst et al. (2005),
p.216)” 25 . It is worth mentioning, though, that the Belgian share in total exports have been
globally declining for the past few decades, in line with trends in many Eurozone’s countries
(International Monetary Fund (2011)), but it does not change its status of open economy.
Since a substantial part of the activity of the BEL20 companies is pursued cross-border, the
political crisis is not supposed to have a significant impact on their results.
2.2.3 The BEL20 index
The BEL20 is a real-time index, constituted of 20 leading firms in Belgium, i.e. (at the beginning
of the crisis) AB Inbev, Ackermans & van Haaren, Ageas (formerly known as Fortis), Befimmo-
Sicafi, Bekaert, Belgacom, Cofinimmo-Sicafi, Colruyt, Delhaize Group, Dexia, GBL, GDF Suez,
KBC, Mobistar, Nyrstar, Omega Pharma, Solvay, Telenet Group, UCB and Umicore 26 . According
to Euronext, these are the most liquid Belgian shares on Brussels stock market. The firms in
this index must never be more than 20 in number. They have a different weight in the value of
the BEL20, but none of them can weigh more than 15% of the total.
The original value of the index was 1000, on December 30, 1990. It was 2665.34 on April 26,
2010 (date retained as the beginning of the political crisis) and 2086.05 on December 6, 2011
(date retained as the end of the political crisis).
25 Cited by Maes (2010).
26 It is worth noting that this constitution of the BEL20 is different from the current one, as since March 19,
2012, Dexia and Omega Pharma left the main Belgian stock index and they were replaced by Elia and D’Ieteren.

15
2.3 Insights into the 2010-2011 crisis
“It is one of the key precepts of the ‘political economy of reform’ that every crisis is an opportunity
to be exploited to drive reform forward.”
Paul Cammack (2009)
We will now review some principles that ruled the 2010-2011 political crisis and that can help us
better understand why the crisis took place and how it could last so long.
Before the crisis even began, a reform had already been envisaged by some, including Kris
Peeters, Minister President of Flanders. He clearly stated that he was looking forward to more
competences relinquished from the federal state to the regions (Quevit (2010)). The Flemish
objective was not exclusively a split of Belgium, as Professor Bruno De Wever states (cited by
Quevit (2010)), but if the Flemish leaders had to choose between the state interests or their
linguistic community’s interest, the community would prevail.
In early 2010, the lack of consensus concerning the Brussels-Halle-Vilvoorde constituency 27
prompted the Open Flemish Liberal and Democrats party (Open Vld) to quit the government
on April 22, 2010. Immediately afterwards, the Prime Minister Yves Leterme handed in his
resignation to King Albert II, but it took effect only four days later. The electorate was then
invited to elect new members to the House of Representatives and the Senate. The previous
election at the time had been held on June 10, 2007. The Belgian political crisis of 2010-2011
thus began after the resignation of the government on April 26 and was confirmed by the Belgian
legislative federal elections on June 13, 2010. From April 26, 2010 to December 6, 2011, a
caretaker government was in charge of dealing with the day-to-day business of state.
Since Belgium is a particracy, there was a large number of different negotiating parties. Their
opposing interests were not easy to conciliate. In total, nine parties took part in the negotiations
but, unable to find a solution, some were removed from the negotiation table. Another feature
that played a crucial role in the problem of setting up a government was the fact that when a
party is offered a first suggestion and accepts it, then it is hard for it to accept a subsequent less
favourable one. The large number of mediators (eight in total) implied that many suggestions
were made and, therefore, the likelihood of finding one that would be accepted by everybody
was low.
In 2011, at the end of the crisis, even more competencies were transferred to the regions and
the BHV constituency was split 28 .
27 In the appendices, figure A.1 illustrates how plurilinguistic this constituency is, which gives an idea of why
finding a solution suiting both the minority of French speakers on Flemish soil (which are the majority in some of
the municipalities concerned) and the Flemish was an uphill struggle.
28 To be exact, the split only came into effect in 2012, whereas the agreement was made in 2011.

Chapter 3
Setting up the indicator
“If a business does well, the stock eventually follows.”
Warren Buffett
The impact of significant political events in times of political crisis on the stocks of Belgium’s
main stock index can be measured in different ways. The most conventional procedure in event
studies consist of analysing the abnormal returns, after a computation of what the “normal”
returns should be (see Brown and Warner (1985), Campbell et al. (1997) or Binder (1998) for a
detailed analysis of classical procedures). However, as Henderson (1990) explains, event studies
exist under different forms and it is not exceptional to see authors using a combination of several
approaches. We believe that the procedure that we use, and that is presented in this chapter, is
the most adapted for the case of analysis and we will strive to justify it as we go along, with
every choice we make.
This chapter starts with a description of our methodology, then lists the different procedures
that we investigate in order to set up the political indicator. We also weigh their respective
strengths and weaknesses, as well as their interests and limitations for the case at hand. This
should, therefore, legitimise why we proceed to a particular regression for a particular analysis
in the chapter 5.
3.1 Methodology
The dummy-variable regression model for event studies is the procedure on which we base our
approach. This model is opposed to classical event studies, more frequently found in literature,
consisting of (i) computing the event window, (ii) measuring the normal returns of a stock
(index) though an “estimation window”, (iii) computing abnormal returns and finally (iv) testing
significance 29 . We chose not to strictly follow the conventional models because the Belgian
political crisis is a unique period and, therefore, the use of an out-of-sample estimation window,
as is the case in most conventional event studies, may not be appropriate. Indeed, we believe that
the political uncertainty due to the absence of a government had an impact on the determination
of the stock prices. Therefore, the estimation of the “normal returns” of the BEL20 index, during
the political crisis, cannot be correctly measured by an estimation window prior to April 2010.
The dummy-variable regression model for event studies was first introduced by Karafiath
(1988), then re-used by other authors (notably Aktas et al. (2007)). Basically, this model is
applied in two steps. Primarily, a dummy variable is built, taking value 1 for days belonging to
29 The most detailed procedure of a classical event studies that we found is available in Campbell et al. (1997),
chapter 4.
16

17
the event window (i.e. either solely the days of the event, or some days before and after the
event as well, depending on the context) and taking value 0 otherwise. For our analysis, we
explain how we build the indicator in section 3.3 and how we select the events in chapter 4.
Once we have built the dummy variable, the second step consists of measuring the impact of this
variable on the returns of a firm, free of the movements of a previously selected market-portfolio.
Formally, we regress the following equation:
where
• R i,t is the return of firm i, at time t,
• R m,t is the return of the market-portfolio m at time t,
• D t is the dummy variable,
R i,t = α i + β i R m,t + γ i D t + ɛ i,t (3.1)
• α i , β i and γ i are the coefficients estimated by the model to fit the regression and
• ɛ i,t is the residual of the regression, for firm i and at each point in time.
As Aktas et al. (2007) establish when they present the same model, γ i can be used as a measure
of “abnormal returns”. This model usually sets as standard assumptions the independence and
identical distribution of the disturbance ɛ i,t . We verify this assumption in section 5.2.5.
For our analysis, we slightly adapt this method, taking the returns of an index (the BEL20)
instead of the returns of a firm as the dependent variable. And as equivalent to the market
portfolio, we take the returns of another index (the Euro Stoxx 50).
As Brown and Warner (1985) state, the model does not need further consideration. In
particular, they establish that “[d]aily data generally present few difficulties for event studies.
Standard procedures are typically well-specified even when special daily data characteristics are
ignored”. However, over the past 27 years (since the publication of Brown and Warner (1985)),
advances have been made on the topic. We will, therefore, add several control variables and
apply procedures to the model in order to correct for endogeneity and heteroskedasticity. We
will also use autoregressive models in order to obtain the best possible fit to our data.
3.2 Timespan
There are 589 days separating April 26, 2010 and December 6, 2011, respectively the fall of
Leterme II’s government and the establishment of Di Rupo I’s government. If we add one week
before this period and one week after this period, in order to take these two important events
into account, the starting and ending dates are April 19, 2010 and December 13, 2011. This is
our “event-window”. We arbitrarily chose one week, because we believe that taking a longer
period would “drown” the political events in a long series 30 and taking a shorter period might
not let us have had enough flexibility when we lag a variable of several days. One week seemed
to be a good compromise between the two.
However, the stock market is open only five days a week and was closed during Easter
Monday and Easter Friday 2011. These two days and the week-ends were the only closed days
30 In other words, a longer period would have meant even more “normal” days for only few days with a significant
political event.

18
over the period of analysis 31 . Subtracting these stock-market free days, we obtain 430 days
relevant to our analysis.
If an event happened when the stock market was closed, we chose to attribute the event to
the next open day. This choice was made in order to correctly assign the impact of the event to
the first moment when the stock market was able react to this event. The same applies if an
event occurred during the night between two days (as was the case for the agreements on BHV).
It is worth noting that we do not take into account a period before the event window as most
of the event studies do. It is not the only approach that differs from conventional event studies.
The justification for this is that we are analysing a period of crisis. Taking a pre-event-window
for computing the “normal return” of the market (frequently used procedure in event studies)
would be an arguable decision. Indeed, it seems possible that the returns in times of political
crisis are different than in times of political stability. We, therefore, chose to use only the period
of consideration (no pre- or post event window period) as we believe that using another approach
could bias the results.
3.3 The different political indicators
In this section, we explain how we build the different political indicators that we will use to
assess the impact of the political events on Brussels stock market. In particular, we will primarily
use (in chapter 5) one main indicator, and then we will test our results with seven alternative
versions of this indicator. In order to better understand the procedure used for the construction
of the different versions of the indicators, refer to the appendices which list the selected dates
day by day and the corresponding value of each indicator.
The easiest way to understand how we built the different indicators is to start with the two
most simple ones. In order to understand the impact of the selected political events on the
BEL20 returns, we would like to differentiate between the events with a negative connotation
for the stability of the country (i.e. events that are likely to push the stock market downward),
from events with a positive connotation (i.e. events that are likely to pull the market upward).
The two basic dummy variables are:
1. “Significant positive event” takes value 1 every time a significantly positive event takes
place (e.g. Johan Vande Lanotte is designated as mediator, the CD&V agrees to start
negotiating without the N-VA, a collective agreement on BHV is reached). Otherwise, the
variable takes value 0.
2. “Significant negative event” is a variable taking value 1 every time a significantly
negative event takes place (e.g. Alexander de Croo leaves the government, the official
resignation of the government, Johan Vande Lanotte resigns from his mission of forming a
government). Otherwise, the variable takes value 0.
These two variables are meant to be used together in the econometric regressions, as they are
complementary. There are 10 days for which significant positive event takes value 1 and 13 days
for which significant negative event does. Therefore, it is respectively 2.33% and 3.02% of the
period we analyse (430 days).
The indicator that we will use the most in our analysis is:
3. “Main political indicator”. This indicator is a combination of significant positive event
and significant negative event. In particular, it takes value 1 when significant positive event
31 Apart from these days, the BEL20 index fluctuated on every day of the period, including on the national day
and for Christmas.

19
has value 1 and takes value −1 when significant negative event has value 1. It takes value
0 when neither of them is positive. Formally:
Main political indicator =
Significant positive event − Significant negative event (3.2)
Since no positive and negative event took place on the same day, over the period we
analysed, this variable has 5.35% of non-zero values.
It is worth justifying the choice of building an indicator that can take only three values
for the analysis at hand. More sophisticated indicators can be easily thought of, such as
indicators that would measure on a scale (from −4 to 4 for example) how good or how
bad a political event is, for the stability of the country. An indicator able to measure
the relative importance of the political events of the crisis would be very interesting, but
extremely arbitrary as well and therefore subject to easy manipulations. In order to limit
as much as possible the “arbitrary” component in the selection of the events, we have used
an indicator that distinguishes only between “significant positive”, “significant negative”
or “non-significant”.
Another interesting indicator would consist of verifying if the significant political events,
regardless whether they are positive or negative, have an impact on the stock market. For this
reason, we computed the indicator:
4. “Significant event dummy”. This indicator is another type of combination of significant
positive event and significant negative event. In particular, it takes value 1 whenever
significant positive event or significant negative event has value 1. It takes value 0 when
neither of them is positive. In other words, it is the absolute value of the main political
indicator. Formally:
Significant event dummy =
Significant positive event + Significant negative event (3.3)
Just as for the main political indicator, this variable has 5.35% of non-zero values.
In order to verify if our arbitrary selection of events (see chapter 4 for more details about it)
is justified, we created a larger version of main political indicator:
5. “All political events”. This indicator includes more events than just those that are very
significant. Instead of having 23 events, we therefore have 40, which corresponds to 9.30%
of non-zero values.
Making inference when the event date is uncertain may be a source of trouble to our analysis.
By uncertain, we mean that newspapers release a piece of information at time t, but the
event might have been known by some investors holding private information at time t − 1.
Fortunately, a procedure exists to solve this problem, as presented by Campbell et al. (1997).
This procedure consists of enlarging the event window, so that it covers two days (day 0 and
day 1). This procedure is double-edged since on the one hand the risk of missing events is
reduced, while on the other hand, the significance may decrease because many days included in
the event window do not correspond to any political events. This procedure has been analysed
by Ball and Torous (1988), who compare it with more sophisticated models including maximum
likelihood estimations. They find that the simple process of enlarging the event window yields
approximately the same results as more sophisticated models. The approach, therefore, seems
worth trying. Following this procedure, we created the following three indicators:

20
6. “Main political indicator 2 ” takes value 1 or −1 when the main political indicator
has this value. But, in addition, main political indicator 2 takes the same value the day
following the event considered.
7. “Significant event dummy 2 ” takes value 1 when the significant event dummy has this
value. But, in addition, significant event dummy 2 takes the same value the day following
the event considered.
8. “All political events 2 ” takes value 1 or −1, as all political events does. But, in addition,
all political events 2 takes the same value the day following the event considered 32 .
Thanks to these different variables, we have relevant building blocks for the analyses of
chapter 5. Some variables, such as significant positive event and significant negative event, share
the qualities of simplicity and clarity, while other variables, slightly less simple, such as main
political indicator 2 avoid the problem of having too few non-zero-valued entries and, therefore,
will not be drowned among too large a number of zero-valued dates.
Let us note, finally, that the different tests that we performed, as a way to “explore” the
topic, included some other variables, which were combinations or slight modifications of the
aforementioned possibilities. We have decided to mention here only the most relevant variables
and to leave aside the recurrent results and the non-informative ones.
As previously mentioned, a summary of the indicators presented in this chapter, and of the
value they take for each of the selected political event, is available in the appendices, at the end
of this thesis.
3.4 Computing the returns
As Campbell et al. (1997) put it, focusing on the returns in finance-related topics is justified
since “the return is a complete and scale-free summary of the investment opportunity”. However,
the return on stock prices can be computed according to different processes. The process
chosen for such computation is important because the results of the analysis will depend on
it 33 . Ultsch (2009) summarises three processes used for computing daily returns, and points out
the advantages and the weaknesses of each of them. The simplest one is the arithmetic return
rate. The most used one in recent literature is the logarithm of the ratio of successive prices
or logarithmic return rate. Finally, Ultsch (2009) establishes that these first two ratios have
weaknesses, in particular with regards to series taking occasionally extreme values (compared
to the rest of the series). He, therefore, sets up an alternative way to compute the returns,
called the relative difference, which does not bear the same weaknesses as the other previously
mentioned ratios. This is particularly useful when extreme values are identified in the series.
Let us have a deeper look at each of them.
32 Two precisions are required for the arbitrary computation of this variable. Firstly, when two events follow
each other, e.g. a positive event at time t 1 and another positive event at time t 2, then the indicator will take
value 1 for the time t 1,2,3. In particular, in time t 2, the value of the indicator is only 1 (and not 2 as some might
have thought). However, as there are only four cases like this, for the 40 events selected, the implication of this
choice is not drastic. Secondly, when a negative event (at time t 1) is followed by a positive event (at time t 2), the
values of the indicator are the following: t 1 = −1, t 2 = 1 and t 3 = 1.
33 Even if, as we will see, the difference is negligible for our data.

21
Arithmetic return rate
This first measurement of ratios is done by following the formula:
Arithmetic return t = P t − P t−1
P t−1
(3.4)
Where P t is the closing value of the stock index, on day t. The choice of taking the closing
value, rather than the mean value or the opening one, is done simply to stick to the standard in
financial literature.
This return is also called “simple net return” by Campbell et al. (1997), as opposed to the
“simple gross return” which is the simple net return to which we add 1.
Logarithmic return rate
The most common measure in literature is computed by the following formula:
( ) Pt
Logarithmic return t = ln
P t−1
The logarithmic return is an approximation of the arithmetic one. They do not differ much
from each other for small values. However, when the return is large (i.e. tens of percents, in
absolute value), then using the logarithmic approach allows us to restrain the value of the return.
Moreover, as Cootner (1964) 34 evidenced, the logarithmic return for a stock price follows a
random walk.
Relative difference
The relative difference consists of comparing the price difference from one day to another to the
average of these prices. The exact formula is the following:
Relative difference t =
It can also be expressed more simply, as follows.
1
2
(3.5)
P t − P t−1
( ) (3.6)
Pt + P t−1
Relative difference t = 2 × P t − P t−1
P t + P t−1
(3.7)
The returns computed by their relative difference have the desirable property that they are
contained within a finite range: [−200%, 200%].
Abnormal returns
“Abnormal returns” is a concept that occurs over and over again in literature about event studies
regarding stock returns. Probably one of most influential papers on this topic was the one of
MacKinlay (1997) in which he defines what abnormal returns are and how to calculate them.
Basically, he explains a statistical procedure consisting of defining a market model to measure
the normal returns calculated out of the sample (calculated just before the sample begins for
example). Then, the abnormal return is the disturbance term of this market model when we
apply it to the sample. Formally, the model is set as the following equation:
Abnormal return t = return t − ̂α − ̂β return mt (3.8)
34 Cited in Parkinson (1980).

22
Where
• return t is the return of the stock within the sample,
• return mt is the return of the stock according to the market model,
• ̂α and ̂β are the coefficients estimated by the model.
As Pynnönen (2005) shows, “the traditional event study with non-overlapping event windows
is equivalent to estimating dummy variable regressions over the combined sample and event
windows” and continues “[t]he dummy variable coefficients corresponds to the abnormal returns”.
We will not conduct an exhaustive analysis of abnormal returns for two reasons. Firstly because,
as just mentioned, it is equivalent to our dummy variable approach. Secondly, because doing
this analysis in parallel and comparing it with the dummy variable methodology exceeds the
scope of this thesis, but such a work is a possible extension of our work.
Conclusion about the computation of the return
For the coming analyses, we use the logarithmic return as a reference, since it is widely accepted
as the standard in event studies. Moreover, it is worth mentioning that the weaknesses that
it bears (and that the relative difference addresses) are not crucial for the case at hand, since
no value of the BEL20 index can be considered as “extreme” (no value exceeds 10% return
in absolute value on a day-to-day basis and only three values exceed 5% return in absolute
value and on the same basis 35 ). The logarithmic approach, therefore, seems the most suitable
approach.
Let us now digress from finance and econometry for one chapter and investigate in more
detail what the milestones of the Belgian political crisis were.
35 These three values are May 10, 2010 (8.9550%), August 12, 2011 (5.3539%) and September 22, 2011 (−5.4925%).
And even for these values, the relative difference method ends up with very similar results: 8.9490%, 5.3526% and
5.4911% respectively, for each of these three dates.

Chapter 4
Significant events
“One thing is sure. We have to do something. We have to do the best we know how at the moment.
If it doesn’t turn out right, we can modify it as we go along.”
Franklin D. Roosevelt
We will now review all politically significant events that have come up since the beginning of
the Belgian political crisis. The chapter is divided into different sections, within which the events
are classified chronologically. We briefly investigate in turn the events related to the formation
of government then the downgrades of the rating agencies. The dates that will be retained as
a basis for forming our main political indicator of the Belgian political crisis (established in
section 3) are indicated in bold. The selection of these "significant" dates among the elements
listed hereunder has been done according to the importance of the events and under the advice
and supervision of Pr P. Wynants 36 , based on Faniel (2011). In the appendices, each event and
its impact on each of our different political indicators is summarised.
For each event mentioned, we ascribe a relevant source, to which the interested reader can
refer. These references are exclusively quality newspapers (or official sources), a choice made
because newspapers convey one of the main building blocks of public opinion 37 and therefore
are one of the bases for stock investors’ reaction. In order to be as politically-neutral as possible,
we chose Belgian reference newspapers both in French and in Dutch 38 and also added some
international newspapers to complete the analysis. The newspapers are also an interesting source
for the period of analysis since, as Junqué de Fortuny et al. (2012) put it, “Belgium has seen a
unique governmental crisis in 2007-2010 during which both political parties and politicians have
had wide media coverage”.
4.1 Politically significant events
During the major part of the crisis, seven parties were negotiating the formation of the government
(politically from left to right, French-speaking/Dutch-speaking); PS/sp.a, Ecolo/Groen!,
cdH/CD&V, N-VA, i.e. three French-speaking and four Dutch-speaking parties.
From the beginning of the crisis and up to December 6, 2011 eight mediators were
mandated by the King with the ultimate aim of forming a permanent government. Specific titles
36 Teacher of Belgian politics at the University of Namur.
37 It has been established, notably by Savigny (2002), that mass media, and especially the Internet, play a
crucial role in the formation of public opinion. Let us also remember that Junqué de Fortuny et al. (2012) have
found that the Flemish newspaper articles show peaks of positive or negative public opinions concomitantly with
the milestones of the Belgian political crisis 2010-2011.
38 In the same proportion.
23

24
were given to them, in the following order; Bart De Wever as informateur 39 , Elio di Rupo as
préformateur, Danny Pieters and André Flahaut as mediators, Bart De Wever as clarifier, Johan
Vande Lanotte as mediator, Didier Reynders as informateur, Wouter Beke as negotiator, Elio di
Rupo as formateur (La Libre (May 17, 2011)). Various other events in the political arena had a
significant influence and were thus added between these milestones of the Belgian political crisis
of 2010-2011. We also include some less meaningful events, in order to provide a comprehensive
analysis of the progress of the negotiations.
Open Vld’s departure from government
On April 22, 2010, Alexander de Croo, leader of the Flemish liberal party Open Vld, decides
to leave the government after months of unsuccessful negotiations on the BHV issue (Le Soir
(April 22, 2010)). Two days earlier, Jean-Luc Dehaene (Belgian Prime Minister from 1992 to
1999), considers his mission finished, handing in recommendations which could have been the
basis for an agreement between the different parties (Le Soir (April 20, 2010)). Five months
earlier, he had been tasked by the King Albert II, to make a suggestion on the institutional
problems of Belgium, with a particular focus on BHV. In the days following the event, Didier
Reynders is put in charge of finding a solution while the French newspaper Le Monde (April 23,
2010) carries the headline “La Belgique est morte le 22 avril 2010”, “Belgium died on April 22,
2010”.
Fall of the government
On April 26, 2010, the King accepts the government’s resignation (De Standaard (April 26,
2011)). Leterme II’s government had been in charge since November 25, 2009 (Service Public
Fédéral Belge (November 25, 2009)) and now becomes a caretaker government, in charge of
dealing with the day-to-day business of the state but not empowered to do major reforms (The
Guardian (April 26, 2010)).
Massive reaction on the legislative proposal about BHV
On April 29, 2010, almost all French-speaking members of Parliament file a motion triggering
the alarm on the legislative proposal about the splitting of BHV (De Standaard (April 29, 2010)).
Federal legislative elections
The Federal legislative elections take place on June 13, 2010 40 .
Bart De Wever as informateur
On June 17, 2010, the winner of the elections Bart De Wever, leader of the New Flemish
Alliance 41 , is mandated by the King to identify the different possible coalitions (De Tijd (June
18, 2011)).
39 His title, “informateur”, is written in French in most articles about the topic written in English; similarly for
the other titles written in French, such as “pré-formateur” or “formateur”.
40 The results are available at Federal Public Service Home Affairs (June 13, 2010).
41 Originally in Dutch: Nieuw-Vlaamse Alliancie (N-VA)

25
Elio di Rupo as préformateur
On July 8, 2010, after the period given to Bart De Wever for gauging the temperature, Elio
di Rupo is put in charge of forming a government. His main tasks concern notably the public
finances and the extent of the State reform, including the BHV file (Le Soir (July 9, 2010)).
Officially, he truly believes that a solution is possible until September 3, when he will openly
accept his failure (L’Echo (May 18, 2011)).
Refusal of di Rupo’s resignation by the King
On August 29, 2010, the King refuses di Rupo’s resignation (De Morgen (August 29, 2010)).
Danny Pieters and André Flahaut as mediators
On September 4, 2010, King Albert II finally accepts di Rupo’s resignation and mandates
André Flahaut (president of the Chamber of Representatives) and Danny Pieters (president of
the Senate) as mediators (La Libre (September 4, 2010)).
Bart De Wever as clarifier
On October 4, the N-VA holds a conference in which it unilaterally interrupts the negotiations
with seven parties (De Morgen (December 10, 2011)). The next day, the King relieves Flahaut
and Pieters of their mission (De Standaard (October 5, 2010)).
On October 8, 2010, Albert II puts Bart De Wever in charge of the different files. Now
being clarifier, De Wever has ten days to hand in a report based on meetings with the seven
negotiating parties and with the ultimate objective of bridging the differences between each
perspective (Le Soir (October 9, 2010)).
On October 18, 2010, the leader of the N-VA declares "Fabula acta est", literally "the
story is over", quoting the Roman emperor Augustus during his very last days. He expresses a
strong disappointment about the reaction of the three French-speaking parties. On the same
day, the King agrees to relieve him from his mission (BBC News (October 18, 2010)).
Johan Vande Lanotte as mediator
Designated on October 21, 2010, Johan Vande Lanotte gives great hope to the country, as he
is more likely to reach an agreement than the previous mediators (De Tijd (October 21, 2010)).
Unfortunately, he resigns for the first time on January 6 (La Libre (January 6, 2011)) but
finally accepts, under the pressure of the King, to make one more attempt, before definitely
resigning on January 26, 2011 (De Morgen (January 26, 2011)).
Belgian record of the longest political crisis
On December 25, 2010, Belgium beats its own record for the longest duration without any
federal government: 194 days (2007) (La Libre (December 23, 2010)).
European record of the longest political crisis
On January 8, 2011, the European record, previously held by the Netherlands (1977), is taken
by Belgium: 208 days without a government (L’Echo (January 7, 2011)).

26
Empowerment of the temporary government
On February 2, 2011, the King asks Yves Leterme, Prime Minister, to present the budget
guidelines and estimates for 2011, and to take any necessary measure to respond to the European
requests with regard to budgetary policies and structural reforms (De Morgen (February 2,
2011)).
Didier Reynders as informateur
On February 2, 2011, Didier Reynders, former Finance Minister, is named informateur by the
King and has two weeks to report to the King whether there is a possibility of reaching an
agreement (La Libre (February 2, 2011)). A month after his nomination he hands the file over
to his successor Wouter Beke (Le Monde (March 2, 2011)).
Wouter Beke as negotiator
On March 2, 2011, Wouter Beke is entrusted with the mission of negotiating the agreement
on the State reform (Le Soir (March 2, 2011)). Even though he has no official deadline, the
N-VA want to impose one, end of April, but the other Flemish parties are opposed to it (De
Standaard (March 5, 2011)). Some weeks later, on May 12, 2011, he hands in his “note” and
waits for the next person that will be in charge: most likely Bart De Wever or Elio Di Rupo (De
Standaard (May 12, 2011)).
World record of the longest political crisis
On March 30, 2011, after 294 days, Belgium now holds the world’s record for the longest
period without a federal government in times of peace (previously held by Iraq (2009)). The
Flemish newspaper De Tijd contradicts it, claiming that Cambodia holds this record (De Tijd
(March 30, 2011)) even though it has been widely recognised and even attested by the Guinness
World Record Book (De Morgen (April 19, 2011)).
Elio di Rupo as formateur
On May 16, 2011, the King accepts the note of Wouter Beke and his request to relieve him. In
an eighth attempt, the King tasks Elio Di Rupo with forming a government (De Standaard (May
16, 2011)). On July 4, the French-speaking socialist president hands in a 111-page long note,
based on his contacts with the nine parties (in addition to the seven parties implied, he had
contacts with the French-speaking and Dutch-speaking liberals, MR and Open Vld respectively,
which now also have a seat at the table of negotiation). Three days later, the N-VA refuses to
start negotiating on the basis of this note (De Morgen (July 7, 2011)). The CD&V asks for
some modifications to the note, while the other seven parties agree to start the negotiation (De
Morgen (July 19, 2011)).
Eight parties around the table
On July 21, 2011, the CD&V finally agrees to start negotiating without the presence of the
N-VA, which is a new start, unblocking the situation (Le Soir (July 22, 2011)).
First agreement on BHV
During the night between September 14 and 15, 2011, an agreement of the eight parties on
the electoral district BHV is reached (De Tijd (September 15, 2011)).

27
Agreement on the financing law
On September 24, 2011, a second agreement is reached. This one concerns the law about the
financing of the Regions and the Communities, about fiscal autonomy and about the refinancing
of Brussels (De Tijd (September 26, 2011)).
Second agreement on BHV
During the night preceding October 5, 2011, a similar agreement is reached concerning the
judiciary district BHV (De Standaard (October 5, 2011)).
Global agreement
On October 8, 2011, a global agreement on the whole constitutional file is reached, after
530 days of crisis (482 days since the elections). The entire agreement is published in Le Soir
(December 2, 2011).
Sixth reform of the State
On October 11, 2011, the sixth reform of the State is designed and agreed upon by the parties.
The reform includes the decisions reached in the previous recent agreements (Le Soir (October
11, 2011)).
Six parties left
On October 13, 2011, the ecologist parties, Ecolo and Groen!, are forced to quit the negotiating
table. Six parties are left (De Standaard (October 13, 2011)).
Political agreement on the budget
On November 26, 2011, after 18 hours of negotiations, the budget for 2012 is agreed (Le Soir
(November 26, 2011)). The downgrade of Belgium by S&P the day before created a pressure
from the markets, which was probably a decisive variable in this agreement (The New York
Times (December 14, 2011)).
Conclusion of the governmental agreement
On November 30, 2011, an agreement is reached on the formation of a government which
would be led by Elio Di Rupo (Washington Times (November 30, 2011)). He would be the first
French-speaking Prime Minister for more than three decades and the first socialist to hold that
position since 1974 (BBC News (December 6, 2011)).
Formation of a government
On December 6, 2011, Elio di Rupo takes the oath and becomes Prime Minister of the new
government. The government is constituted of the Social Democrats (PS/sp.a), the Christian
Democrats (cdH/CD&V) and the Liberals (MR/Open Vld), but excludes the N-VA (Le Soir
(December 5, 2011)). This date marks the end of the world’s longest political crisis.

28
4.2 Rating agencies’ outlooks and downgrades
We now move on to another kind of event which has an impact on the stock market. As for
the previous section (4.1), we bold the dates that are important and that we will use in the
econometric regressions in the next chapter. We list hereunder the different events related to
the main rating agencies (Standard & Poor’s, Moody’s and Fitch). As evidenced by data, the
threat of downgrading the rating of Belgium by the agencies or the actual downgrade of this
rating have an impact on the daily returns of the BEL20 index. We, therefore, need to take
these parameters into account, in order for our political indicator to be more accurate. In other
words, if a movement in the stock market is due to an announcement of a rating agency and
not to a political event taking place at the same moment, and if we do not take into account
the rating agencies’ impact, then an ordinary econometric regression will wrongly attribute this
movement to the political event. In order to avoid this kind of inaccuracies, we will create a new
dummy variable, called “Rating”, taking value 1 when a rating agency announces that it starts
reviewing Belgium for a potential downgrade or when this downgrade takes place. Let us now
go through each of the three major rating agencies one by one.
4.2.1 Standard & Poor’s
Negative outlook
On December 14, 2010, S&P threatens Belgium with a potential downgrade. Their outlook
for Belgian debt turns to “negative”, as one of the Eurozone’s most indebted countries might
encounter higher borrowing costs (Reuters (December 14, 2010)). Quoting Reuters; “[i]f the
country’s inability to form a government threatened deficit- and debt-reduction goals, S&P
said Belgium’s AA+ rating could be downgraded within six months”. Let us note that “AA+”
corresponds to the second best grade, in S&P’s ranking, out of a list of 22 grades.
S&P’s downgrade
On November 25, 2011, Standard & Poor’s decides to apply its one-year-old threat of
downgrading Belgium. The Belgian grade thus goes down from “AA+” to “AA” (third best
grade), and “negative perspectives” were bound to this new grade. According to the economic
newspaper L’Echo (November 26, 2011), this will have an unquestionable impact on the real
economy. The decision was motivated by a pressure on financing and market risk, increasing the
likelihood for complementary state support. In the same article, a relevant question for this
thesis is addressed to Alexandre De Groote (CEO of Petercam Institutional Bonds); “should we
fear a sudden rise of the returns after this downgrade?” He believes that the market fluctuation
already incorporated this information. Data however are less clear about it: the BEL20 index
did not decrease on that day, it even increased slightly but much less than the Euro Stoxx 50
index. Therefore, it seems that this event was not totally forecast and incorporated by the
market.
4.2.2 Moody’s
Moody’s reviews Belgium’s “Aa1”
On October 07, 2011, the high level of Belgian debt combined with the “long-term funding
risk on the euro-area sovereign” encouraged Moody’s to review Belgium for a possible downgrade
in the future (Moody’s (October 7, 2011)).

29
Moody’s downgrade
On December 16, 2011, Belgium’s credit rating is downgraded by Moody’s from Aa1 to Aa3
(from second best to fourth best grade, in a list of 21 different grades), which corresponds to a
“negative outlook” (Moody’s (December 16, 2011)). However, this event happens after the end
of the period we analyse here. We cannot, therefore, take it into account.
4.2.3 Fitch Ratings
Negative outlook
On May 23, 2011, the outlook of Fitch on the sovereign rating of Belgium changes from “stable”
to “negative”.
Rating watch and downgrade
On December 16, 2011, as Moody’s downgrades Belgium, the rating watch of Fitch turns to
negative. The actual downgrade from Fitch takes place on January 27, 2012 (Fitch Ratings
(2012)).
4.2.4 Conclusion about the ratings
Even if only four relevant rating-related events are highlighted during the political crisis 2010-
2011, using this data enables us to capture some fluctuations of the BEL20 daily returns. It
frees our political indicator of some possible inaccuracies.
4.3 Summary of the events
In order to provide a global view of the Belgian crisis to the reader, we established a timeline
(on page 30), summarising the aforementioned events. The previous analysis appears to provide
23 key dates relating to the Belgian political crisis 2010-2011, and 17 less important dates. With
regard to the “rating dummy”, we identified four key events. We include the most important
events chronologically on the following timeline, which is split into two parts; the first one for
2010 and the second one for 2011. We also added to this summary the value of the BEL20 and
the Euro Stoxx 50, as a percentage compared to their value on April 19, 2010 (just before the
resignation of Leterme II’s government). As these figures show, it is hard, if at all possible, to
notice an impact of the political events on the daily return of the BEL20. However, we can
clearly identify that the BEL20 and the Euro Stoxx 50 move closely together and respond to
the same shocks.

Figure 4.1: Main Belgian political events against the BEL20 and Euro Stoxx 50 indices
daily returns, April 19, 2010 - December 31, 2010
30
Figure 4.2: Main Belgian political events against the BEL20 and Euro Stoxx 50 indices
daily returns, January 01, 2011 - December 13, 2011

Chapter 5
Impact of political events on the
Brussels stock market
“As we realise that more and more things have global impact, I think we’re going to get people
increasingly wanting to get away from a purely national interest.”
Peter Singer 42 .
As the literature review has shown, the stock market is closely linked with political conditions,
and in particular with political events. We have seen that it was especially the case when we
considered emerging markets and politically unstable countries. Belgium, during the 2010-
2011 political crisis, might be considered to fall into this second category. We investigate,
in this chapter, the impact of the politically significant events highlighted in the previous
chapter on the evolution of the return of the BEL20 index. Significant results are found,
evidencing the connection between the political and economic worlds in Belgium. We first start
by reviewing some basics of financial econometrics; we then introduce a simple static model that
we progressively sophisticate as we go along. In the extensions, we end with an error correction
model and with autoregressive conditional heteroskedasticity models. Some topics with regard
to our approach for this chapter are raised in the next chapter - Discussion - as they do not
concern one subsection in particular.
5.1 Theoretical background
Two main theoretical models are of interest for our analysis. The first one is the capital asset
pricing model (CAPM), introduced by the Nobel laureate William Sharpe (1964) and by Lintner
(1965). This is probably the most widely used model for the evaluation and management of
financial portfolios. Even though the CAPM was a theoretical success, empirical evidence seems
not to validate it (Fama and French (2004)). The second model of interest is the arbitrage pricing
theory (APT), established by Ross (1976), determining the expected return of a stock (or more
generally of an asset) as a function of multiple factors. The CAPM was the standard in event
studies during the 1970s (Campbell et al. (1997)) but later studies preferred using multi-factor
models derived from the APT because it prevents one from using incorrect assumptions on the
expected returns.
In this section, we review the efficient market hypothesis (EMH) at the basis of these two
models, then we briefly introduce the foundation of each of them.
42 Australian Inspiration (2010)
31

32
5.1.1 Efficient market hypothesis
Financial econometrics is based on the uncertainty faced by investors (Campbell et al. (1997)).
Anticipation and expectations are, therefore, key to understand how the stock markets react.
Since Samuelson (1965), who proved that “properly anticipated prices fluctuate randomly”, the
EMH (summarised by Fama (1970)) has often been challenged in literature. Basically, it states
that stock markets are supposed to “fully reflect” all available information, or at least to reflect
correctly the relevant stock-related information. The EMH has a direct implication, which is
unfortunate for greedy investors: that stock returns must be random. It should thus not be
possible for an investor to make profit based on the available information.
Having perfect market efficiency is unlikely, not only because the expectations are not
always rational, but also because some of the relevant information comes at a price. However,
challenging the hypothesis of efficiency of the markets is always tricky. Indeed, if the hypothesis
is rejected, it can mean either that the market is inefficient or that the model that has been
used is not properly defined (Campbell et al. (1997)).
For these reasons, the debate about the EMH remains open in the context of the capital
asset pricing model and the arbitrage pricing theory. We now turn to the rationale of these
models.
5.1.2 Capital asset pricing model
The CAPM establishes what the optimal point between the risk and the return of a portfolio
is. One that lies on this point is called “mean-variance efficient” (Campbell et al. (1997)). The
CAPM is formulated as follows:
(
)
E(R i ) = R f + β im E(R m ) − R f (5.1)
where
• E(x) is the expected value of x,
• R i is the return on an asset i,
• R f is the return of a risk-free asset (the model assumes that there is such asset),
• R m is the return of the market portfolio and
• β im is a measure of the risk of an asset. This measure can be defined as follows:
β im = Cov(R i, R m )
V ar(R m )
(5.2)
where
• Cov(R i , R m ) is the covariance of the respective returns of the asset i and the market
portfolio and
• V ar(R m ) is the variance of the market portfolio.
Putting words on formula 5.2, one of the main findings of the CAPM is that a stock’s
expected return is a linear function of its covariance with the market.
The CAPM is sometimes used to measure the expected return of a stock in event studies. It
is worth noting that several weaknesses and inaccuracies have been debated in the literature

33
about the CAPM. In particular, Campbell et al. (1997) mention that some CAPM frameworks
consider that it holds only conditionally: only period per period or only conditionally to the
state of the economy, for example. Other reasons for this debate include statistical evidence (in
the US market) against the CAPM, or the fact that the CAPM should be replaced by other
sources of risk.
5.1.3 Arbitrage pricing theory
The APT, introduced by Ross (1976), is a general theory about the pricing of assets, stating
that the return on an asset is a linear function of the expected return on that asset and of a set
of relevant variables. Formally, it is expressed as follows:
where
• r i is the return on asset i,
• E i is the expected return on asset i,
r i = E i + β i1 X 1 + ... + β ik X k + ɛ i (5.3)
• the X k are the k zero-mean factors termed X, whose subscript goes from 1 to k,
• to each factor, is joined a coefficient β, measuring the sensitivity of asset i to the factor in
consideration,
• the n different assets are identified by their subscript i, going from 1 to n and finally,
• ɛ i is the residual (also called “noise” or “disturbance” 43 ) of the asset i.
Basic assumptions are made, notably (i) that ɛ i is an unsystematic idiosyncratic risk
component such that E(ɛ i |X j ) = 0, (ii) monotonicity and (iii) concavity. It follows the same
intuition as the CAPM (Roll and Ross (1980)), since it also considers a linear process for
estimating the return. One distinction between these two models is the fact that the APT does
not rely on any market portfolio 44 . Another major difference resides in the fact that the APT
allows the expected return to be a function of several other variables. Literature evidencing the
fact that several factors are likely to affect the process of return generating can be found in
Rosenberg and Marathe (1975).
Shanken (1982) claims that the APT is more likely to be validated empirically than the
CAPM, but no clear-cut answer has been found as to which of these models better reflects reality.
As he presents it (based on the argumentation of Roll (1977)), testing the CAPM is tricky since
it relies on an assumption hardly ever met: the existence of an efficient market portfolio, which
should be used in the test of the CAPM.
In the coming analyses, we mainly follow the intuition of the APT, building a model explaining
the fluctuations of the BEL20 index by the fluctuations of several other variables. However,
we also keep the “market” trend approach of the CAPM, since we use the Euro Stoxx 50 as a
regressor in our model. The Euro Stoxx 50 can indeed be considered as the equivalent to the
market portfolio for the BEL20 index. As we mention in the following paragraphs, the strong
correlation between the BEL20 index and the Euro Stoxx 50 as well as the significance of the
different control variables analysed confirm that our approach is justified.
43 In the coming pages, we will indifferently use any of these denominations.
44 As Roll and Ross (1980) put it: “[i]n general, the market portfolio plays no special role whatsoever in the
APT, unlike its pivotal role in the CAPM”.

34
5.2 Multiple regression analysis
In this section, we start by adopting a simplistic model, that gets more and more sophisticated as
we go along. We do not extensively interpret the results before the end of the section, since the
coefficients of the different variables used and their significance change as the model gets better
specified. Section 5.2.10 brings together the conclusions of each subsection of the main model.
It is a useful summary for the reader who does not feel confortable with the econometric details
presented hereunder. However, the reader may be interested to observe how the coefficients of
the various regressors evolve, as we sharpen the analysis.
5.2.1 Simplistic return-analysis model
The basic estimation of the impact of political events, taken all together, on the BEL20 index
returns is:
where
{BEL20 return} t = β 0 + β 1 {P olitical indicator} t + β 2 X t + ɛ t (5.4)
• {BEL20 return} t is the return of the BEL20’s index, on day t,
• {P olitical indicator} t is the value of the main political indicator built in section 3.3, on
day t,
• X t is a vector of exogenous variables 45 ,
• the βs are the coefficients estimated by the model for each regressor and for the intercept
(β 0 ). They capture the sensitivity of their respective regressor with regards to the dependent
variable 46 . And
• ɛ t is the residual, or disturbance, of the regression for the observation t 46 .
With ordinary least squares (OLS) regressions, several assumptions have to be made. In
particular, we assume that the residuals ɛ t :
1. are normally distributed: ɛ t ∼ N(·, ·),
2. have a zero mean: ɛ t ∼ N(0, ·),
3. have a constant variance: ɛ t ∼ N(0, σ 2 ),
4. are not serially correlated and
5. are not correlated with the explanatory variables.
These assumptions can be considered as strong ones because usually: “security returns are
not normally distributed, [...] [t]here is evidence of slight serial correlation in security returns,
[...][t]here is evidence that variance shifts sometimes are associated with financial events” and
“[t]here is evidence that the residuals are correlated with values of the independent variable[s]”
(Henderson (1990)). Techniques exist to deal with these problems in event studies. We will,
45 Here, as in the following equations, we indicate vectors by bolded letters.
46 In the remainder of the thesis, we will not explicitly state what these β coefficients are, or whichever Greek
letter we attribute to them. The same applies to the disturbances ɛ.

35
therefore, release some of these assumptions in further sections. In particular, we will deal with
endogeneity of some regressors, with the heteroskedasticity of the residuals and analyse the
serial correlation present within these residuals.
The null hypothesis associated with this model is that the stock market is efficient, anticipates
every political move and, therefore, does not react to political events. Significance for the political
indicator means that the null hypothesis is rejected (using a two-sided t-test). In other words,
that the coefficient is statistically different from zero. The significance is accompanied by a
threshold value 47 , measuring how certain it is that the null hypothesis is rejected. As a way to
put it simply, we will often just refer to the threshold value to describe the level of significance
of an estimated coefficient 48 .
Interdependence on other stock markets
Forbes and Rigobon (2002) show that major events in stock markets can have a significant
impact on other stock markets across the globe, even if these markets differ in size and in
structure. With regard to our analysis, the period that we investigate is characterised, at best,
by instability of the different stock markets around the world, due to the financial crisis. Using a
more global stock index as a control variable, we are able to free the BEL20 from the fluctuations
due to global shocks 49 . Two such indices seem to be particularly appropriate for our analysis;
the Euro Stoxx 50 and the S&P 500. The Euro Stoxx 50 is the index covering the leading
companies in the main sectors of the Eurozone. The S&P 500 is the most commonly used index
for American companies; it includes 500 U.S. leading companies within the main industries in
the American economy.
We use the European index Euro Stoxx 50 instead of the S&P 500, because the European
index responds to some events and variables that the American one does not consider. In
particular, it considers all monetary-policy related events and variables, because the Eurozone
has a unique and common monetary policy whose impact will be obviously much bigger on
the Euro Stoxx 50 than on the S&P 500. Putting numbers to this intuition: ordinary least
squares (OLS) regressions show that the fraction of the fluctuations in the BEL20 returns that
is explained by the S&P 500 returns is 45.90%, while if we use the Euro Stoxx 50, we reach
89.05% 50 . In other words, the European index fluctuations explain about twice as much as the
American one. We consider this as a sufficient reason for using the Euro Stoxx 50.
As Table 5.1 shows 51 , this control variable is highly significant, which means that the Belgian
stock market largely fluctuates in tandem with the European one. It confirms the trend that we
already saw in figure 4.1.
Table 5.1: Regression of the BEL20 daily returns, controlling
for interdependence with the European stock
market index Euro Stoxx 50.
Variable Coefficient (Std. Err.)
Euro Stoxx 50 return 0.811 ∗∗ (0.014)
Intercept -0.014 (0.024)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%
47 This threshold depends on the p-value for the test of the null hypothesis.
48 The lower the threshold, the more significant the coefficient is.
49 It will also integrate seasonality effects, such as the “January effect” and the “week-end effect”, well known in
financial time series.
50 The data comes from Datastream (2012).
51 In this table, as in the following ones, we indicate the latest added regressor in bold.

36
It should be noted, however, that the BEL20 index is part of the Euro Stoxx 50 index and
therefore, it is not surprising that they are, to some extent, correlated. Though, it should be
noted that the European index is principally constituted of French companies (36.2%) and
German ones (31.5%) while Belgium has a very weak weight in it (3.4%) 52 . Therefore, using
this index does not seem inappropriate.
The Euro Stoxx 50 explains the major part of the fluctuations of the BEL20 index. However,
we will add several other macroeconomic variables in order to have a model as well-specified as
possible. This approach seems justified, as Campbell et al. (1997) explain about the procedure
to be used in event studies, since “in principle, [...] increases in R 2 could be achieved by using a
multi-factor model”. But it is worth mentioning that “[i]n practice, however, the gains in R 2
from adding additional factors are usually small”, which is confirmed with our data.
The positivity of the coefficient is a coherent result, since we expected these two indices
to move in tandem. We believe that a similar correlation can be expected for the volatilities
of these stocks. Indeed, as Black (1976) 53 mentions while talking about the correspondence
between a stock and its market portfolio: “[i]n general, it seems fair to say that when stock
volatilities change they all tend to change in the same direction”.
5.2.2 Adding the main political indicator, free of endogeneity, to the model
Endogeneity is a serious problem. If our regressions are subject to endogeneity for the political
indicator, i.e. if both the returns have an effect on political events and political events have an
effect on the returns, then one of the most basic assumptions made in our OLS regressions is
broken. This assumption is that errors are not correlated with any of the dependent variables.
Endogeneity can result in biased and inconsistent estimators. We, therefore, have to check it for
the political indicator, because it is likely that we observe some endogeneity, which would bias
our interpretations. The addition of the main political indicator to the model, after having freed
it of endogeneity with the BEL20 index return, is the purpose of this subsection.
Dealing with endogeneity: building the instrumental variable
We follow a procedure that has been extensively used previously in literature to control for
possible endogeneity between stock market returns and political actions: the instrumental
variable approach 54 .
As investigated by Rigobon and Sack (2001) and pointed out by D’Amico and Farka (2002),
“[f]or the majority of cases [...], the stock market is allowed to instrument itself through its
own lagged values”. However, since the stock market is likely to anticipate political events, the
bidirectional causal relationship may remain, implying that the lagged values of the stock market
are not a valid instrument. It is also hard to conceive instruments totally independent from
political events but determining stock market fluctuations. As a solution to this problem, we
can construct our own instrument, based on the BEL20 return and on the indicator of political
events. This choice may cause a problem of generated regressor, investigated further in this
section, but seems appropriate since the resulting instrument would bear the desirable properties
of a good instrument. This simple approach is based on the work of D’Amico and Farka (2002)
but is highly intuitive.
52 Source: Stoxx Limited (June 29, 2012).
53 Cited by Schwert (1990).
54 As cited by Angrist and Pischke (2009), the first author to use this method is Wright (1928), while the term
“instrumental variable” is due to Reiersøl (1941). For a good summary of the theory about this method, refer to
Wooldridge (2009).

37
We start by establishing a system that on the one hand is likely to characterise the stock
return function and on the other hand shows some possibility for political events to be influenced
by the stock market.
where
• RBel is the daily return on the BEL20 index,
• P OL is the political indicator,
• X is the set of different control variables and
• the ε are the residuals for each equations.
RBel t = βP OL t + θX t + ε RBel
t (5.5)
P OL t = αRBel t + δX t + ε P OL
t (5.6)
We will now build a proper instrumental variable, addressing the possible endogeneity
problem by eliminating the correlation between the residuals of the Belgian stock market daily
returns (ε RBel
t ) and the indicator of political events (P OL t ).
The construction of the instrument is done by regressing our political indicator on the returns
of the BEL20 index, then using the resulting residuals as the new political indicator in the
initial equation (equation (5.4)). Similar procedures using a generated regressor as instrumental
variable have been used by several authors in literature, notably Barro (1977), Barro (1979) and
Pagan (1984). Oxley and McAleer (1993) provide a summary of the danger of using generated
regressors if one does not correct the standard errors. Fortunately, using a two-stage least
squares (2SLS) approach, as we do, corrects these standard errors and seems thus perfectly
appropriate. Murphy and Topel (1985) are usually seen as a reference for two-step econometric
models, yielding asymptotically correct standard errors.
With regard to our case, we proceed to:
By construction, the residuals of equation (5.7):
P OL t = γ 0 + γ 1 RBel t + e IV
t (5.7)
ê IV = P OL t − ̂P OL t (5.8)
ê IV = P OL t − (̂γ 0 + ̂γ 1 RBel t ) (5.9)
are the share of P OL t that is not correlated with ε RBel
t , because êIV contains all the information
of the political indicator that is not explained by the Belgian stock market main index.
These residuals bear two attributes of interest: first, they are highly correlated 55 with our
initial political indicator P OL t ; second, they are not correlated 56 with the returns of the BEL20
index RBel t . We can, therefore, use these residuals as a valid instrument for the political
indicator. Since this indicator is now purged from ε RBel
t , we have eliminated any possible
endogeneity.
55 To be exact, the correlation is 99.43%.
56 Correlation of 0.00%.

38
Testing the results, free of endogeneity
Proceeding to this methodology and replacing the political indicator by êIV in equation 5.4,
we obtain the results of table 5.2. As expected, the model defined by equation 5.4 showed
endogeneity between the political indicator and the daily returns of the BEL20 57 .
Table 5.2: Instrumental variables (2SLS) regression of the
BEL20 daily returns, using an ad hoc estimator of the
political indicator as the instrument.
Variable Coefficient (Std. Err.)
Main political indicator -0.206 ∗ (0.104)
Euro Stoxx 50 return 0.814 ∗∗ (0.014)
Intercept -0.015 (0.024)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%
As the table shows, all regressors are significant. In particular, for êIV , referred to as “main
political indicator” in table 5.2, the significance threshold for the test of the null hypothesis that
political events do not cause fluctuations in the return of the BEL20 index is 4.7% 58 .
Its coefficient suggests that for every positive political event, on the day of the event, the
return of the BEL20 decreases by 0.2 percentage points 59 , and for every negative event, it
increases by the same amount, on average.
However, both the coefficient and the significance of the regressors of the above regression
can be inexact because of missing control variables. Fortunately, extensive literature about
the variables that can have an impact on the stock market exists. Adding the correct missing
independent variables to the model will give us a coefficient that is more accurate for the political
indicator. Now, let us turn to the explanatory variables that should be included in the X t of
equation 5.4, i.e. the variables that would help us better specify our model.
5.2.3 Adding the control variables to the model
Several other variables may have an impact on stock prices and therefore on stock market returns.
Not including them in our regression might result in what is called “an omitted variable bias”.
To be more precise, we have to distinguish between equation 5.4 and equation 5.10:
{BEL20 return} t = β 0 + β 1 {P olitical indicator} t + ɛ t (5.10)
Indeed, if a control variable included in equation 5.4 and not in equation 5.10 is correlated
with the political indicator, then the coefficient β 1 will be biased in equation 5.10 (Angrist and
Pischke (2009)). The procedure of adding control variables will therefore yield a more precise
estimate of the effect of the political indicator on the fluctuations of the BEL20 index daily
return. Angrist and Pischke (2009) also establish that “a longer regression - one with controls
[...] - has a causal interpretation while a shorter regression does not”. Since establishing if such
a causal relationship exists is exactly what we are looking for, adding control variables seems
the best procedure to follow.
57 In the appendices, we indicate the results of our final model if we do not control for endogeneity, which
evidences that the main political indicator in the initial model was indeed endogenous.
58 For the sake of clarity, in the following paragraphs we will loosely express it as “the significance threshold is
x%”.
59 This coefficient is only approximate, so far, since we still need to add the control variables.

39
This section therefore explains in detail which variable we add to our regression and how the
results are affected by the addition of each of these variables 60 . We also provide the reader with
some theoretical background about the use of each of them, by briefly reviewing some of key
literature about the topic. In particular, potential control variables might be the changes in
rating from rating agencies, the change in the interest rate of the government bonds, the change
in the inter-bank overnight interest rate (Eonia), the monetary growth, but also the change in
the sovereign debt risk, the fluctuations in the price of other assets (such as gold) or the state of
the economy. These variables, if they can be correctly measured, if regularly released data can
be obtained and if they participate in the explanation of the fluctuations of the BEL20 index,
should be used as control variables. This is done in order to purge our analysis from possible
over- or under-estimations due to variations wrongly attributed to political events, instead of
attributing them to these neglected macroeconomic variables. In this section, we briefly mention
the authors who have analysed these variables, their results and explain why we believe that
including them in our regression is relevant for the analysis at hand.
We will review possible control variables one by one and keep the ones that increase the
fraction of the BEL20 which is explained by the regressors (econometrically speaking, it is the
coefficient of multiple determination or “adjusted R 2 ” 61 ). In other words, we will keep the ones
that are significant or close to being significant, i.e. the variables with which the model is better
specified.
Quadratic interdependence on other stock markets
Several tests with the data reveal that the BEL20 index and the Euro Stoxx 50 may not be
solely correlated linearly. The square of the daily returns of the Euro Stoxx 50 should be used as
well as a control variable, because in addition to a linear correlation, the Belgian stock market
seems to react quadratically to this index. As table 5.3 indicates, using these squared returns
is not only a significant control variable, but it also improves the significance of the political
indicator 62 . The coefficient of the square of the Euro Stoxx 50 return is not large, but it is
significant at a 5% threshold, evidencing that it captures some part of the BEL20 fluctuations.
Table 5.3: Regression of the BEL20 daily returns, adding
the square of the daily returns of the Euro Stoxx 50 index
as a regressor.
Variable Coefficient (Std. Err.)
Main political indicator -0.234 ∗ (0.104)
Euro Stoxx 50 return 0.813 ∗∗ (0.014)
Euro Stoxx 50 return 2 0.008 ∗ (0.003)
Intercept -0.040 (0.026)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%
60 Some of the data is not available daily but only monthly or quarterly. In order to solve the missing data
problem due to different frequencies in the publication of data, we considered that every day between two valued
dates (e.g. between the first of each month, for monthly data) take the same value as the previously valued date.
Similarly for the return, we considered that it remained the same for every day of these non-valued periods.
61 The adjusted R 2 , contrary to the R 2 , takes into account the number of independent variables in the specification
of the regression. In other words, while the R 2 can only increase if a regressor is added to a regression, the
adjusted R 2 can also decrease if this regressor does not significantly improve the specification of the equation.
62 Its coefficient now reaches the 2.4% threshold.

40
Rating dummy
The changes in the outlook or in the rating given to a country by the main rating agencies are
likely to have an impact on the Belgian stock market. Indeed, Kaminsky and Schmukler (2002)
found that ratings have an impact on stock returns, and also have a spillover effect across the
financial markets. Their analysis goes even further; “[i]nterestingly, rating changes trigger more
widespread market instability during times of turmoil, suggesting that rating changes may act
like a wake-up call or a signal that coordinates investors towards a bad equilibrium”. It seems,
therefore, particularly relevant to add this variable for our analysis.
Through adding the dummy variable about the changes in the outlook or in the rating of a
country by the main rating agencies 63 , we obtain the results reported in table 5.4. In particular,
when an outlook is modified to “negative” or when Belgium is downgraded by one of the three
main rating agencies, then the daily return on the BEL20 index seems to decrease.
Table 5.4: Regression of the BEL20 daily returns, adding
an indicator of rating agencies’ modifications in outlook
and downgrades as a regressor.
Variable Coefficient (Std. Err.)
Main political indicator -0.232 ∗ (0.103)
Euro Stoxx 50 return 0.813 ∗∗ (0.014)
Euro Stoxx 50 return 2 0.008 ∗ (0.003)
Rating dummy -0.511 ∗ (0.245)
Intercept -0.035 (0.026)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%
One can notice that with this specification of the model, our rating dummy is significant at
a 5% threshold. The significance of the political indicator remains unchanged compared to the
regression with only the Euro Stoxx 50 daily returns and their square as control variables.
In section 5.2.9, we test the robustness of this “rating” indicator, by replacing it by an
alternative measure of it.
Changes in the yield of Belgian government bonds on the secondary market
As Bernhard and Leblang (2006) conclude, the interest rate on government bonds is likely to have
an impact on political outcomes. Should this rate be high, it will be harder for a government to
manage fiscal policy (implementation of new programmes or tax cuts for example). It, therefore,
seems to be a relevant control variable for our analyses.
In line with the predictions of Bernhard and Leblang (2006) about the potential impact
of a change in the yield of a country’s government bonds on the stock market, we find almost
significant results evidencing this link. For our data, we used the one-year maturity yield of
Belgian government bonds, exchanged on the secondary market 64 . Our results show a negative
link between this variable and the BEL20 daily returns, even if this link is rather small. In
particular, when the yield increases, then the return of the BEL20 seems to slightly decrease, on
average 65 .
63 This is the dummy variable that we built in section 4.2.
64 The data comes from the National Bank of Belgium (2012).
65 This control variable has a significance threshold of 11.1%, and very slightly affects the measure of the political
indicator, whose coefficient (in absolute value) increases and standard error decreases. The significance is therefore
bigger.

41
Table 5.5: Regression of the BEL20 daily returns, adding as a
regressor the daily change in the yield of Belgian government
bonds on the secondary market, with one-year maturity.
Variable Coefficient (Std. Err.)
Main political indicator -0.237 ∗ (0.103)
Euro Stoxx 50 return 0.813 ∗∗ (0.014)
Euro Stoxx 50 return 2 0.008 ∗ (0.003)
Rating dummy -0.491 ∗ (0.244)
Government bonds yield -0.008 (0.005)
Intercept -0.031 (0.026)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%
It is worth mentioning that over the period of interest in this analysis, there are 13 dates
missing with respect to the stock market days, i.e. there are 13 days for which there is a value
for the stock market, but no value for the interest rate on government bonds. These dates are in
2010: May 13, May 14, May 24, July 21, November 1 and November 11, and in 2011: June 2,
June 3, June 13, July 21, August 15, November 1 and November 11. We considered that the
return for these days is simply zero, because the value of the stock remains the same as for the
previous day.
We also tried a similar regression, using the yield on Belgian bonds as control variable, with
every different maturity for which the National Bank of Belgium (2012) gives the data, i.e. two
years, three years, four years, five years and “six years and more”. The results are that the
“government bonds yield” indicator is not even close to being significant for the other maturities.
Contrary to the one-year maturity yield, they do not improve the explanation of the BEL20’s
fluctuations. We will therefore keep only the one-year maturity yield in further regressions.
Predictability
The predictability of political events should have a strong impact on our results. Indeed the
more predictable an event is, the less strong the impact on stock prices will be, and vice versa.
Predictability includes not only the outcome of similar past events, but also the opinion given in
the main newspapers about the likelihood that the event will end up with a successful outcome.
According to Bernhard and Leblang (2006), this predictability may be evaluated by the volatility
of the stock market before the event; indeed the higher the volatility, the less certain stock price
forecasts are.
As a way to measure the volatility, we refer to the most cited source on the topic, which is
Parkinson (1980) who provides a good estimate of the variance for stock prices. This estimation
is easily computed as:
( ) Hight
Est t = ln
Low t
(5.11)
where
• Est t is the estimation of the variance at time t,
• High t is the highest price reached by the stock, over the period of consideration, here the
day t, and

42
• Low t is the lowest price reached by the stock, for the same period.
We, therefore, compute this estimated variance for each day of the period of interest. Then,
we lag the results from one, two, three, four or five periods in order to fit the estimator of
predictability established by Bernhard and Leblang (2006). Adding each of these lags of the
estimated variance separately to our regression shows that the resulting variable is the most
significant when we use a three-day lag on the estimator of the variance. The results of this
addition can be seen in table 5.6.
Table 5.6: Regression of the BEL20 daily returns, adding the
predictability of the political events as a regressor, measured
as a three-day lagged estimation of the variance.
Variable Coefficient (Std. Err.)
Main political indicator -0.244 ∗ (0.103)
Euro Stoxx 50 return 0.813 ∗∗ (0.014)
Euro Stoxx 50 return 2 0.006 † (0.004)
Rating dummy -0.493 ∗ (0.243)
Government bonds yield -0.008 (0.005)
Predictability indicator 0.057 ∗ (0.024)
Intercept -0.116 ∗ (0.045)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%
One can see on this table that this estimator of predictability is largely significant (1.9%
threshold) and that some other regressors are impacted as well. In particular, firstly, the change
in the yield of Belgian government bonds on the secondary market and the square of the return
of the Euro Stoxx 50 are both less significant than before 66 . Secondly, the political indicator’s
significance is also impacted but positively 67 .
Therefore, it is important to consider the predictability, as we see that there is an anticipatory
movement of the market prior to political events. This movement is significant for all three days
preceding the event and is the most significant in t − 3, meaning that on average, it is three
days before a political event happens that the market fluctuates the most.
Changes in the Eonia and in the Euribor
Bernanke and Kuttner (2005) find that an unexpected 25-basis-point rate cut of the federal
funds rate (FFR) would typically imply an increase of about 1% in stock prices of the CRSP
value-weighted index 68 . Even if the context is different, the approach brings some pieces of
interesting information about a possible control variable that could have an impact on stock
prices: the federal fund rate.
However, this rate is a good control variable for the American market; indeed Bernanke and
Kuttner (2005) worked with an American index. For Europe, fortunately, there is an equivalent
short-maturity rate to the FFR, which is the Eonia 69 and which is the average rate at which
European banks lend to each other overnight.
66 Their thresholds are 9.5% and 12.4% respectively (instead of 2.8% and 11.1%), so they are still worth
consideration.
67 It is now at the 1.8% threshold, with a coefficient that is larger in absolute value.
68 The CRSP (Center for Research in Security Prices) indices are considered as a benchmark for both academic
research and investment. They are based on the NYSE, Amex, NASDAQ and Arca markets and are, therefore,
an alternative to the S&P 500.
69 Euro OverNight Index Average

43
There also exists another European benchmark, very similar to the Eonia, which is the
Euribor 70 , the Eurozone-equivalent of the Libor 71 (Euribor-rates.eu (2012)).
Using the change in the Eonia does not show any significance, but using that of the Euribor
(one-year maturity) does. It lies within the 10% significance upper bound and contributes to the
explanation of the BEL20 index fluctuations, we will, therefore, keep it as a control variable. Its
inclusion gives us table 5.7.
Table 5.7: Regression of the BEL20 daily returns, adding
the change in the interbank interest rate (one-year maturity
Euribor) as a regressor.
Variable Coefficient (Std. Err.)
Main political indicator -0.244 ∗ (0.103)
Euro Stoxx 50 return 0.811 ∗∗ (0.014)
Euro Stoxx 50 return 2 0.007 † (0.004)
Rating dummy -0.490 ∗ (0.243)
Government bonds yield -0.008 (0.005)
Predictability indicator 0.060 ∗ (0.024)
Euribor change 0.074 † (0.044)
Intercept -0.131 ∗∗ (0.046)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%
National inflation and money growth
Feldstein (1980) shows that for the U.S. market, increased inflation had an adverse effect on
stock prices. What is of particular interest to us is that national inflation can have an impact
on the stock market. Similarly, monetary growth has been identified by Rozeff (1974) to be
in favour of an increase in stock returns. The paper of Thorbecke (1997) about the effects of
monetary policy on stock market returns backs up the view of Rozeff, at least in the short term.
His findings support the hypothesis that the causality is due to firms’ access to credit. It seems
thus that national inflation is an interesting variable, worth including in our model.
Testing for this relationship, our first approach is to use the monthly change in the Consumer
Price Index in Belgium 72 as an estimator for inflation. Adding it as a regressor shows that this
variable is not significant. Worse, it reduces the adjusted R 2 of the model. Therefore, we cannot
use it.
As alternative measures of inflation, we verify the relevancy of the change in M1, then M2
then M3 73 . These three measures estimate the quantity of money in the Belgian economy going
from the most liquid money (M1, principally constituted of demand deposit accounts), then M2
(constituted of M1, plus saving deposits and slightly less liquid assets) and finally M3 (including
longer-time deposits as well). When we use M1 or M2, the result does not show any closeness to
significance. However, the control variable M3 is close to being significant and improves the
adjusted R 2 . To be precise, the significance threshold reached by this variable is 16.6%. In
literature, M3, or the sum of the liabilities of the financial intermediaries, was found by King and
Levine (1993) to have an impact on economic growth. Therefore, it confirms that its inclusion
in our model makes sense.
70 EURo InterBank Offered Rate.
71 London InterBank Offered Rate.
72 Computed by the Belgian National Institute of Statistics.
73 The data comes from Datastream and has one-month interval.

44
However, several authors show that some problems arise if we want to use this variable, as it
is identified as being largely subject to endogeneity. As was extensively established in previous
literature, the causality effect goes in both ways between stock market returns and monetary
growth. In particular, the link between the two has been investigated by Boudoukh et al. (1994),
Bernanke and Gertler (2000), and D’Amico and Farka (2002), through analyses of monetary
policies. D’Amico and Farka state that “[s]tock market fluctuations are likely to be an important
determinant of monetary policy decisions because of their potential impact on macroeconomy”.
Bernanke and Gertler (2000) mention that stock returns should be included as a regressor to
estimate the impact of asset price volatility on monetary policy, because monetary policy could
be pursuing other objectives, besides stabilisation of output and expected inflation 74 .
These discussions brought us to verify a possible endogeneity between monetary growth and
the return of the BEL20 index over the period of interest. We find that these variables are
indeed endogenous. After correcting this endogeneity, as we did for the political indicator in
section 5.2.2, we obtain the results reported in table 5.8.
Table 5.8: Regression of the BEL20 daily returns, adding as
a regressor the monthly growth of M3 (free of endogeneity)
as an estimator for inflation.
Variable Coefficient (Std. Err.)
Main political indicator -0.240 ∗ (0.102)
Euro Stoxx 50 return 0.812 ∗∗ (0.014)
Euro Stoxx 50 return 2 0.007 † (0.004)
Rating dummy -0.515 ∗ (0.242)
Government bonds yield -0.009 † (0.005)
Predictability indicator 0.058 ∗ (0.024)
Euribor change 0.084 † (0.044)
Money growth -0.030 † (0.016)
Intercept -0.128 ∗∗ (0.045)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%
We can see from this table that money growth, when corrected for endogeneity, is significant
(6.4% threshold) and that including it in the model increases the absolute value of the yield on
Belgian government bonds, making it significant at a 10% threshold.
One might wonder if using more than one instrumented variables is relevant, since it is not
common in literature 75 . Wooldridge (2001), seen as a reference in econometrics 76 , uses several
instrumented variables (with as many instruments) for estimating the impact of education and
skin colour on wages. We therefore believe that it is relevant to apply a similar procedure:
correcting more than one variable for endogeneity by using several instrumented variables and
several instruments.
State of the economy
Romer and Romer (2007) use interesting control variables, not only to free the regressions from
endogeneity and anticipatory movements, but also to evaluate the robustness of their paper.
74 It is worth mentioning that some authors, such as Boudoukh et al. (1994), do not completely agree with this
possible endogeneity, considering that the potential causal effects between monetary growth and the real economy
are a question that is still open.
75 To the best of our knowledge.
76 Many thousands of sources cite his books and papers.

45
Notably, they use the state of the economy, as measured by the lagged GDP growth (quarterly
data). Some major differences remain, however, between their analyses and ours: i.e. firstly the
topic varies greatly since they evaluate the macroeconomic effects of tax changes; secondly, the
timeframe is substantially different since their analysis is made over half a century.
GDP data is available for Belgium 77 , but only as quarterly data. We tried to add the
lagged GDP of Belgium to our regression, and found that the lagged GDP is significant at a 5%
threshold if we lag it of 30 to 55 days and even at the 1% level if we lag it of 55 days to 65 days.
To be precise, it is 60-trading days (thus 12 weeks), which corresponds to about three months,
and therefore the lag corresponds to a quarter.
However, we will not include this lagged GDP growth as a regressor in the coming regressions
for multiple reasons. First and foremost, since our timeframe is not even two years long, using
this variable as a control would not be relevant for the analysis at hand, because only seven
different measures can be made (because we have only quarterly data). Another reason, which
should be sufficient for this decision not to include the lagged GDP growth, is that its addition
seriously reduces the adjusted R 2 . It implies that if we add it, then a bigger fraction of the
BEL20 daily returns is not explained by the model, even though the coefficient is significant.
This variable seems, therefore, inappropriate for the specification of our model.
Output
As Blanchard (1981) mentions, “the effect of output on the stock market is ambiguous”, but he
does not discard the possibility that output has an impact on the stock market. We therefore
verify if including this variable as a regressor is relevant in our analysis.
Using measures of the output such as the unemployment rate 78 or the net exports 79 show
that for Belgium and over the period of interest, the output has no significant impact on the
variations of the BEL20 index. The inclusion of these variables actually worsens the model,
given the number of regressors. We, therefore, consider the output as irrelevant for the case at
hand and we do not include it in the model.
The reasons for this non-significance is likely to be that the Euro Stoxx 50 already captures
the change in output. Indeed, as Belgium is such an open country (see section 2.2.2), its output
is largely correlated with the output of other countries.
Fluctuations in gold price
Gold is often considered as a refuge currency against inflation. It seems therefore possible that
in difficult times, more investors turn to gold instead of to the stock market. The return on gold
is thus likely to affect the stock market.
However, proceeding to the addition of the daily return on gold as a regressor 80 shows that
this variable is not significant in our analysis. Thus, we will not take it into account.
Sovereign debt risk
The report about financial stability of the National Bank of Belgium (2011) mentions an
intensification of market concerns with regard to the sovereign debt risk (these concerns were
already present in 2010). The same report underlines that an increase of this risk has strong
77 The data comes from the National Bank of Belgium (2012).
78 The data comes from Datastream and were verified with the data coming from the National Bank of Belgium
(2012) (monthly data).
79 The data comes from Datastream (monthly data).
80 The data comes from the National Bank of Belgium (2012).

46
negative consequences for a country, since it would increase the financing cost of that country.
An indicator of this risk would be useful for our analysis, since a stock market full of concerns
seems less likely to generate positive returns than an optimistic stock market.
Unable to find such an indicator for Belgium, we use an estimator of sovereign debt risk:
the amount of the government’s debt itself 81 . We tried both for central government’s debt and
for general government’s debt and the results are similar: non-significance. Thus, the debt of
the Belgian government does not explain the fluctuations in the BEL20 index over the period
of analysis. The reason behind it may be the same as for the state of the economy: that the
data about the debt are only available quarterly, while our event-window is relatively short.
Alternatively, it could simply be that the amount of the debt has no direct 82 impact on the
stock market. Finally, it is also possible that the “rating” variable captures part of the sovereign
debt risk, since the rating agencies use this risk to determine the countries rating. In any case,
we will not use an explicit measure of the sovereign debt risk for our analysis.
Conclusion about the control variables
To summarise this section, the formal expression of our regression, with the control variables
that we use, is the following:
{BEL20 return} t = β 0 + β 1 {Main P olitical indicator} t
+ β 2 {Euro Stoxx 50 return} t
+ β 3 {Euro Stoxx 50 return} 2 t
+ β 4 {Rating dummy} t
+ β 5 {Y ield on government bonds} t
+ β 6 {P redictability indicator} t
+ β 7 {Euribor change} t
+ β 8 {Money growth} t + u t (5.12)
The results of this regression are contained in table 5.8, on page 44. However, this model
may suffer from a weakness that we are about to verify: its residuals may not be constant over
time. This common problem for financial econometrics is called heteroskedasticity. We cannot
interpret the data before verifying if our model is subject to it.
5.2.4 Heteroskedasticity: verification and correction
A possible problem in least squares regressions is heteroskedasticity. We are in presence of
heteroskedasticity when the residuals of a regression are not constant over time. Such a problem
would result in a biased variance of the estimators, thereby directly affecting the standard errors
and resulting in biased confidence intervals and significance thresholds.
Harrington and Schrider (2002) deal with the problem of heteroskedasticity in event studies.
They finish by advising to complete the OLS regressions with a Weighted Least Squares (WLS)
approach, robust to arbitrary forms of heteroskedasticity. In this section, we start by verifying
that we are in presence of heteroskedasticity, then we follow the procedure advised by Harrington
and Schrider.
81 The data comes from Datastream.
82 We have seen that it has an indirect impact: a high debt drives the rating agencies to downgrade a country,
which has a significant impact on the stock market.

47
In order to check if the residuals of our model are heteroskedastic, we conduct a Breusch-
Pagan test. First, we compute the residuals of equation 5.12, then we add the square of these
residuals as a regressor in the model, which gives us the results of table 5.9.
Table 5.9: Regression of the BEL20 daily returns, on all selected
control variables and adding the squared residuals as a regressor for an
heteroskedasticity check.
Variable Coefficient (Std. Err.)
Main political indicator -0.231 ∗ (0.101)
Squared residuals of equation 5.12 0.185 ∗∗ (0.054)
Euro Stoxx 50 return 0.815 ∗∗ (0.014)
Euro Stoxx 50 return 2 0.006 † (0.003)
Rating dummy -0.516 ∗ (0.239)
Government bonds yield -0.008 (0.005)
Predictability indicator 0.034 (0.025)
Euribor change 0.090 ∗ (0.043)
Money growth -0.030 † (0.016)
Intercept -0.130 ∗∗ (0.450)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%
As one can notice, our regression is largely subject to heteroskedasticity (even at a 1%
threshold, as the variable “squared residuals of equation 5.12” shows). We must, therefore,
correct it before being able to draw any conclusion from the results. Easily-computed methods
exist to correct heteroskedasticity. The one we use is based on a WLS approach, giving a
different weight to observations depending on how well they behave (i.e., the “robust” tool of
Stata). This yields standard errors that are asymptotically exact (Angrist and Pischke (2009)).
The results of our model, robust to heteroskedasticity, are contained in table 5.10.
Table 5.10: Regression of the BEL20 daily returns, on
all selected control variables, with the residuals robust
to heteroskedasticity.
Variable Coefficient (Std. Err.)
Main political indicator -0.240 ∗ (0.098)
Euro Stoxx 50 return 0.812 ∗∗ (0.018)
Euro Stoxx 50 return 2 0.007 (0.005)
Rating dummy -0.515 ∗ (0.232)
Government bonds yield -0.009 ∗ (0.004)
Predictability indicator 0.058 (0.039)
Euribor change 0.084 ∗ (0.040)
Money growth -0.030 † (0.016)
Intercept -0.128 ∗ (0.058)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%
As table 5.10 contains the final results of our model, we interpret it in section 5.2.10. What
we can already notice, as a comparison with our non-robust model, is that the standard error
of the square of the returns of the Euro Stoxx 50 and the standard error of the estimator for
the predictability have increased. Their respective coefficients do not lie in the 10% significance
area anymore.

48
5.2.5 Autocorrelation: verification for both the returns and the residuals
Another possible problem with financial time series is that the returns are autocorrelated. In
other words, the return in time t is influenced by the return in time t − 1. Autocorrelation is a
problem because it would contradict one of the assumptions of the least squares models: that
the returns are independent and identically distributed (IID). We start with a verification of a
simple linear form of autocorrelation of the returns, then also verify if the disturbances of the
model are serially correlated.
Autocorrelation: verification for the BEL20 daily returns
The simple linear form is formally expressed as follows:
where
• r t is the return considered at time t and
r t = α + βr t−1 + ɛ t (5.13)
• α, β and ɛ are respectively the intercept, the coefficient of the lagged returns and the
residuals of the model.
Fitting equation 5.13 to the BEL20 daily returns, no evidence of autocorrelation is found.
The BEL20 index returns are therefore IID. We also find that, as predicted by the theory, they
have a zero-mean.
This test of autocorrelation corresponds to the simplest approach to testing for unit root, i.e.
testing that after a shock the series has a tendency to come back to the trend. Testing for unit
root can be better specified, by using a dynamically complete model. We do it in the extensions:
section 5.3.1.
Autocorrelation: verification for the disturbances of the model
In our regression model, not only do we have to verify a possible autocorrelation of the returns
but we must also consider whether the disturbances of the model are serially correlated, i.e.
if they are not IID. Probably the most well-known test of serial correlation, constructed with
the residuals, is the Durbin-Watson (DW) statistics, based on Durbin and Watson (1950) and
Durbin and Watson (1951). The statistic is computed as follows:
DW =
∑ Tt=2
(û t − û t−1 ) 2
∑ Tt=1 û 2 t
(5.14)
where
• û t is the disturbance of the least squares regression at time t.
The DW statistic lies within the [0, 4] interval and if it differs significantly from value 2, then
we can assume that the residuals of the least squares regression are autocorrelated. With our
model, the DW statistic yields a value of 0, indicating that the residuals are certainly not IID.
In order to correctly fit the model, given this result, we need to use an autoregressive process.
We proceed to such process in further sections (5.3.3 and 5.3.4), however since it requires a
significantly more advanced knowledge of econometrics, we introduced it as an extension.
The fortunate thing, however, is that even though the residuals are serially correlated, it
does not imply bias of the least squares estimators computed above. The coefficients previously
presented are therefore still consistent.

49
5.2.6 Anticipation and late reaction effects
Glascock et al. (1987) find that daily stock prices react significantly to anouncements of Moody’s,
but this adjustment is lagged. This result convinces us that it is possible that the effect of the
political events on the stock market is lagged, i.e. that a fraction of the investors have a delayed
reaction to the events. The other way round, we believe that it is possible that the market
anticipates events. These two verifications are the purpose of this section. For each, we proceed
to an analysis of the results with a lag or an anticipation of one, two or three days. We do not
go further than three days, because it is unlikely that the market needs more than three days to
adjust or that it significantly anticipates more than three days in advance 83 .
Late reaction: lagging the indicator
In order to have a well-specified model for the lagged indicator analysis, we have to re-consider
an assumption that we made: endogeneity. Previously, we believed that political events have
an impact on the stock market and that the stock market has an impact on the occurence of
the political events. Now that we lag the indicator, we have to envisage that it might not be
endogenous anymore. It is indeed likely that political events have an impact on the stock market
on the days following the events. However, it is not possible that the returns of the BEL20 index
on the days following a political event have an impact on the occurence of this event. In other
words, an event cannot possibly be influenced by the daily return of a future date. Therefore,
no endogeneity should be assumed for this section.
The results of the regressions using a lagged version of the main political indicator 84 are
clear: there is no “late reaction effect”, as the lagged indicator is not significant in the model.
We present the results of a one-day lagged version of the main political indicator in table 5.11.
We can see on this table that the coefficients, standard errors and significances of all other
regressors are not very different than previously, evidencing robustness of the model.
Table 5.11: Regression of the BEL20 daily returns, using a one-day lagged
version of the main political indicator as a regressor.
Variable Coefficient (Std. Err.)
Main political indicator, lagged of one day -0.023 (0.101)
Euro Stoxx 50 return 0.807 ∗∗ (0.014)
Euro Stoxx 50 return 2 0.006 † (0.004)
Rating dummy -0.519 ∗ (0.243)
Government bonds yield -0.008 † (0.005)
Predictability indicator 0.056 ∗ (0.025)
Euribor change 0.084 † (0.044)
Money growth -0.030 † (0.016)
Intercept -0.122 ∗∗ (0.046)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%
The hypothesis that the coefficient of the one-day lagged version of the main political
indicator is different from zero cannot be rejected. The result is similar for the different lags
83 Except for one particular event: if a piece of news makes the occurence of the event very likely, then the
market could shift abruptly, several days in advance. However, as we consider an indicator made of a set of
multiple events, our assumption seems reasonable.
84 To be more precise, we keep the same control variables as for our main model, we do not correct for endogeneity
of the lagged political indicator and the results do not vary whether we correct for heteroskedasticity or not.

50
that we used, the entire reaction of the stock market to the political events seems therefore to
occur no later than on the day of the events.
Anticipating the events
As for the lagged-indicator analysis, we have to consider if the presence of endogeneity is likely
or not for this new specification of the model. With an “anticipation analysis”, firstly, we wonder
if the stock market a few days before the political event has an impact on the occurence of this
event. We believe that it is indeed likely that events are somehow a response to good or bad
conditions in the stock market. Secondly, we wonder if a political event might have an influence
on the stock market a few days before this event. Since it is likely that investors anticipate
political events, we believe that this influence indeed exists. This two-way causality relationship
convince us to use an endogeneity-corrected anticipated indicator for this section.
As for the lagged-indicator analysis, the results that we find by advancing the political
indicator by one or more days show non-significance, no matter how many days of anticipation
we use. In particular, table 5.12 show the results with a one-day anticipation of the main political
indicator.
Table 5.12: Regression of the BEL20 daily returns, using a one-day anticipated
version of the main political indicator as a regressor.
Variable Coefficient (Std. Err.)
Main political indicator, advanced by one day 0.052 (0.106)
Euro Stoxx 50 return 0.807 ∗∗ (0.014)
Euro Stoxx 50 return 2 0.006 † (0.004)
Rating dummy -0.545 ∗ (0.249)
Government bonds yield -0.008 † (0.005)
Predictability indicator 0.053 ∗ (0.025)
Euribor change 0.083 † (0.044)
Money growth -0.030 † (0.016)
Intercept -0.118 ∗ (0.046)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%
One can notice, however, that even if the coefficient of the advanced political indicator is
insignificant, there seems to be a slight positive tendency. We believe that it is possible that the
market anticipates an event, even over-anticipates it, then adjusts (in the opposite direction) on
the day of the event (as we will argue in the conclusions of the model). This positive tendency for
the one-day advanced political indicator seems in line with this possibility. But, let us mention
it one more time, since the coefficient is insignificant, we cannot build a strong theory on it.
5.2.7 Robustness of the political indicator: dropping the events, each in turn
To verify that our results are robust, one possible check consists of dropping key events one by
one, and verifying if the coefficient of the indicator is unwavering or if it fluctuates significantly.
If the coefficient remains steady, then it proves that the political indicator is robust, since it is
not a particular event that biasses the whole indicator. This is the first robustness check that
we investigate.
Fitting 23 new least squares regressions, each time dropping a significant event, for each of
the 23 non-zero values of the main political indicator, we find that the indicator is robust to a
certain degree. In particular, the coefficient of the political indicator fluctuates between −0.2020

51
and −0.2688. The gap is relatively important, but this result does not fundamentally change
the conclusions previously reached: even at its least significant value, the political indicator still
lies within the 3% significance threshold, testifying to its robustness.
If we compute the mean of these coefficients, we obtain −0.23990, while the coefficient of
the regression with all events is −0.23988. The fact that these two values are so close evidences
robustness of our main political indicator.
5.2.8 Robustness of the political indicator: alternative indicators
The second robustness check consists of replacing the main political indicator that we have used
so far by the other versions of the indicator, that we introduced in section 3.3. We use the same
control variables as previously and follow the same approach with regard to endogeneity and
heteroskedasticity. That is, we fit equation 5.12, just replacing the main political indicator by
each alternative version in turn, using endogeneity-corrected indicators and residuals robust to
an arbitrary form of heteroskedasticity.
Regressing with these other versions of the political indicator is very useful, as it confirms
the intuitive conclusions that we could draw from previous regressions and it gives even more
information about the impact of political events on the returns of the BEL20 index. Let us
now investigate one by one each of these alternative versions of the political indicator that we
established.
Separate indicators for positive and negative events
The main political indicator that we used so far was built on the combination of a dummy
variable for the events favourable to the formation of a government and of another dummy
variable for the events unfavourable to the formation of a government. Our results with the
main political indicator (see table 5.10) indicated a negative coefficient and therefore, on average
a positive reaction of the market after negative events and reversely, a negative reaction of the
market after positive events. The estimated coefficient was about −0.24. We will now be able to
verify if the reactions to negative and to positive events are indeed in the reverse directions and
if one of the two types of event drives the coefficient of the main political indicator or if both
are significant.
Table 5.13: Regression of the BEL20 daily returns, on all selected
control variables, with the residuals robust to heteroskedasticity and
with separate indicators for positive and negative events.
Variable Coefficient (Std. Err.)
Significant positive event dummy -0.402 ∗ (0.162)
Significant negative event dummy 0.118 (0.123)
Euro Stoxx 50 return 0.808 ∗∗ (0.018)
Euro Stoxx 50 return 2 0.007 (0.005)
Rating dummy -0.519 ∗ (0.232)
Government bonds yield -0.008 † (0.004)
Predictability indicator 0.056 (0.039)
Euribor change 0.083 ∗ (0.039)
Money growth -0.030 † (0.016)
Intercept -0.125 ∗ (0.058)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%

52
As table 5.13 shows, our intuition for the sign of the coefficients was correct. However several
comments are worth mentioning. Firstly, the absolute value of the coefficient of the significant
positive event dummy is much bigger than that of the significant negative event dummy (0.402
versus 0.118). It seems, therefore, that the market reacts much more to positive events than
to negative events. Secondly, the standard errors of both coefficients are bigger than for the
main political indicator. It is likely to be due to the smaller number of non-zero values for
each of these two indicators. Thirdly, as the consequence of the smaller coefficient and larger
standard errors, the significance of the negative event dummy has plummeted, such that we
can no longer reject the hypothesis that its coefficient is equal to zero. However, we can still
observe a positive tendency. With regards to the significance of the positive event dummy, it
almost reaches the 1% threshold and is even more significant than the main political indicator
presented earlier. Finally, the coefficients and standard errors of the other variables have not
meaningfully changed, evidencing their robustness in this model.
Significant event dummy
With the results we have found so far, mixing positive and negative events in one indicator taking
only value 1 or 0, no matter what kind of event occurs, is highly unlikely to give significant
results. Proceeding to a regression with the significant event dummy instead of the main political
indicator confirms this intuition: evidence of non-significance is found. We do not include the
related table, since it does not contain any other relevant new information.
All political events
The next alternative indicator consists of selecting not only the most significant political events,
but also the less significant ones (the appendices show the detail of which event is selected). The
results yielded by the regression with this indicator are summarised in table 5.14
Table 5.14: Regression of the BEL20 daily returns, on all selected
control variables, robust to heteroskedasticity, using more
political events in the political indicator.
Variable Coefficient (Std. Err.)
All political events indicator -0.138 † (0.079)
Euro Stoxx 50 return 0.810 ∗∗ (0.018)
Euro Stoxx 50 return 2 0.006 (0.005)
Rating dummy -0.516 ∗ (0.232)
Government bonds yield -0.008 † (0.004)
Predictability indicator 0.059 (0.040)
Euribor change 0.086 ∗ (0.039)
Money growth -0.030 † (0.016)
Intercept -0.129 ∗ (0.059)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%
As we can see, the indicator’s significance is smaller than the significance of the main political
indicator. It shows that the stock market reacts less (if at all) to the less meaningful political
events. The standard error of the new indicator is also smaller, which allows the coefficient to
be still significant at the 10% threshold.

53
Main political indicator 2
Using the indicator taking value 1 or −1 not only on the day of the most important political
events, but also on the days following these events, i.e. the indicator that we called “main
political indicator 2”, we obtain the results reported in table 5.15.
Table 5.15: Regression of the BEL20 daily returns, on all
selected control variables, with the residuals robust to heteroskedasticity
and using the “main political indicator 2”
purged from endogeneity.
Variable Coefficient (Std. Err.)
Main political indicator 2 -0.153 ∗ (0.067)
Euro Stoxx 50 return 0.811 ∗∗ (0.018)
Euro Stoxx 50 return 2 0.007 (0.005)
Rating dummy -0.514 ∗ (0.233)
Government bonds yield -0.009 ∗ (0.004)
Predictability indicator 0.059 (0.040)
Euribor change 0.084 ∗ (0.040)
Money growth -0.030 † (0.016)
Intercept -0.131 ∗ (0.059)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%
As the table shows, the coefficient of the political indicator has dropped by 0.09 percentage
point, compared to table 5.10. Even though the standard errors are smaller as well, the indicator
is less significant than before, but still lies within the 5% threshold. This loss of significance
indicates that the effect of an event happens mainly on the day of the event and not on the day
following the event. This result confirms what we found when investigating the late reaction
effect of the stock market, in section 5.2.6.
Significant event dummy 2
As we found no significance for the variable “significant event dummy” and a decrease in the
significance of the main political indicator when we add value 1 on the days following the events
as well, we do not have great expectations of significance for the variable “significant event
dummy 2”. The data confirms this intuition.
All political events 2
As for the variable “all political events”, the negative tendency of the indicator “all political
events 2” is still present, but the statistic test cannot significantly reject the hypothesis that the
coefficient is equal to zero. Its coefficient drops to −0.067 with a standard error of 0.055.
5.2.9 Robustness of the rating dummy: alternative indicator
Finally, we verify the robustness of our “rating dummy” as well, by replacing this dummy by
another indicator, built upon the same events but taking an incremental value for each event. In
particular, instead of using a dummy variable taking value 1 only on the day of the modification
to a negative outlook or the day of the downgrade, we can build a variable whose value increases
for each of these dates. This approach makes sense from an economic point of view, since a

54
downgrade is not an event that is likely to affect the BEL20 during one period only, but that is
likely to affect it continuously (i.e. it is a persistent shock).
We have, therefore, built this indicator, standardised for our period and thus starting at
value 0 at the beginning of the period of interest. We increment this variable on four dates; on
December 14, 2010, on May 23, 2011, on October 7, 2011 and on November 25, 2011 (section
4.2 gives more detail about these particular events).
If we replace the previously used rating dummy by this “rating indicator”, we obtain the
results summarised in table 5.16.
Table 5.16: Regression of the BEL20 daily returns on
all selected control variables, with the political indicator
purged from endogeneity and the residuals robust
to heteroskedasticity, replacing the rating dummy by
a more sophisticated indicator.
Variable Coefficient (Std. Err.)
Main political indicator -0.218 ∗ (0.092)
Euro Stoxx 50 return 0.812 ∗∗ (0.018)
Euro Stoxx 50 return 2 0.007 (0.005)
Rating indicator -0.048 ∗ (0.023)
Government bonds yield -0.009 ∗ (0.004)
Predictability indicator 0.070 † (0.040)
Euribor change 0.070 (0.045)
Money growth -0.025 (0.016)
Intercept -0.101 † (0.060)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%
The coefficient of the new indicator is much smaller than that of the previously used dummy,
but its significance is similar. We could expect such change in the coefficient, since many more
observations are now non-zero valued and the value of some is also bigger. We can interpret these
results as follows: when one of the main rating companies downgrades Belgium or places it under
a negative outlook, then the daily returns of the BEL20 index is significantly and continuously
negatively affected. In other words, it does not just cause a one-day fall then bounces back
and fully recovers, it is, instead, a persistent shock. However, it is worth remembering, as the
previously used rating dummy showed, that the impact on the day of the event is much bigger,
which makes sense since this change might appear as unexpected for some investors.
5.2.10 Summary of the main model
Let us briefly summarise the results reached in this section about the impact of political events
on the daily returns of the BEL20 index.
First and foremost, the results of the main model are reported on page 47, in table 5.10,
which puts together the results of the analysis after correcting for both endogeneity of the
political indicator and heteroskedasticity of the residuals. This model uses as control variables
the daily returns of the Euro Stoxx 50, their square, an indicator for the changes in rating, the
changes in the yield on Belgian government bonds, an indicator of the predictability, the changes
in the Euribor and the growth of money. The two control variables that do not reach the 10%
threshold in this model are not far from being significant, as they are only a few percents above
this threshold. Each control variable used is useful, since dropping any of them decreases the
coefficient of determination (the adjusted R 2 ), i.e. the share of the fluctuations of the BEL20

55
index daily returns that is explained by the regressors 85 . One can observe that the political
indicator is almost significant at a 1% threshold 86 , meaning that the political events that we
identified have had a meaningful effect on the BEL20 index returns. To be more precise, the
coefficient indicates that, all other things being equal, when a positive event took place, the
BEL20 index daily return decreased on average. And when a negative event happened, the
daily return increased, on average. These movements are significant on the particular day of
the event and the average fluctuation has an absolute value of about 0.24 percentage points.
The negative sign leads us to believe that the Belgian stock market anticipates events and even
over-anticipates their effect, then readjusts when the event actually takes place.
Secondly, even though we expected that the market would partially anticipate the outcome
of the events and would partially react with a lag to these events, no such evidence can be found.
It seems therefore that if there is some anticipation of the outcome of political events, this
anticipation is slightly absorbed by the market over several days, such that there is no sudden
fluctuation due to this event before the day when the event actually takes place. Similarly, the
market seems to have fully absorbed the news on this particular day, such that no significant
lagged reaction can be found. Though, even if the anticipation is not significant, there seems
to be a positive tendency on the day before the event (the stock market yields higher returns
before positive events and smaller before negative event). This tendency backs up our theory of
over-anticipation of the market.
Thirdly, we also identified that we face a relative robustness of the political indicator.
Dropping some events may change the coefficient of this indicator, dropping to about −0.20,
but even in this case, the political indicator is still significant at a 3% threshold, testifying from
robustness.
Fourthly, we found that positive events have on average a bigger (negative) net impact on
the BEL20 than negative events (positive impact). The impact of positive events is about −0.40
percentage point and is very significant, while that of negative events is about 0.12 percentage
point but this result is not significant at a 10% threshold.
Fifthly, data shows that the market reacts to a lesser extent to less important political events
than to the most meaningful ones.
Sixthly, the rating dummy used in the analysis is robust as well, since replacing it by an
alternative indicator does not fundamentally change its significance, nor does it substantially
affect the coefficient of the other regressors of the model.
Last but not least, we find that the BEL20 index daily return over the period of analysis
bears the properties of financial time series found in similar previous literature. In particular,
(i) the daily returns are independent and identically distributed with a zero-mean and (ii) the
disturbance of the simple least squares models used show significant serial correlation, which
implies that an autoregressive process would better fit the data. We now turn to the extensions,
where we use such a process to model the volatility of the BEL20 daily returns.
85 Let us remember that the adjusted R 2 , contrary to the R 2 , takes into account the number of regressors. In
other words, if an independent variable added as a regressor does not significantly improve the explanation of the
fluctuations of the dependent variable, then the adusted R 2 would decrease (while the R 2 can only increase).
86 The significance threshold of this indicator is 1.4%.

56
5.3 Extensions
“A politician needs the ability to foretell what is going to happen tomorrow, next week, next month,
and next year. And to have the ability afterwards to explain why it didn’t happen.”
Winston Churchill, 1965
This section enlarges the scope of the thesis in different respects. We analyse topics that
are more advanced than the simple previous regressions, which mainly consisted of using the
basic econometric tools through a dummy variable event study. This section is an interesting
complement to the analysis conducted so far, as it reaches slightly different conclusions, by
using more sophisticated tools. We start by verifying if our data has a unit root, a necessary
step before being able to investigate the three models of interest for these extensions. These
models, that we review in turn, are the error correction model, and two augmented versions of
the autoregressive conditional heteroskedasticity model: one linear in its parameters and the
other logarithmic, allowing thereby more flexibility of the fit.
5.3.1 Testing for unit root
In section 5.2.5, we briefly mentioned the rationale for unit root testing. Let us investigate it
here in more detail. We first introduce the necessary concepts for a complete understanding of
the model; then we proceed to an augmented version of a Dickey-Fuller (DF) test.
Stationarity, for a time series, is the attribute of remaining constant over time. In econometrics,
a time series is stationary if its mean, its variance and its autocorrelation do not change
when they are shifted in time.
If a time series is not stationary, then it can be differentiated until it is made approximately
stationary. The degree of integration (introduced by Granger (1981)) can be understood as follows;
if a time series needs to be differentiated d times to be made approximately stationary, then its
degree of integration is d. In order to check for cointegration and to apply an autoregressive
conditional heteroskedasticity model, it first needs to be verified that our data is stationary or if
it needs to be integrated.
Testing for unit root consists of testing if a shock at time t has an impact not only at time t,
but also in the following periods. In other words, it tests if the series takes some time to adjust
(if it ever does). For example, the daily closing price of the BEL20 index over the period we
analyse in this thesis is found to have a unit root 87 , in other words, the price of the index is
largely dependent on its value the day before and does not bounce back after a shock, which
seems very intuitive.
A series has a unit root if and only if β = 1 in equation 5.13 (page 48). We have already seen
that under the simple specification of equation 5.13, the return on the BEL20 index does not
have a unit root. However, this model was not dynamically complete, since it did not take into
account the values of every previous return in the series. A dynamically complete model would
follow a Dickey-Fuller distribution and unit root for this distribution can be tested through a
Dickey-Fuller test, as established by Dickey and Fuller (1979).
We proceed to a more recent version of this test, called an augmented DF-GLS 88 test, a
model that was suggested by Elliott et al. (1996). The DF-GLS test is applied with 1 to k
lags, where k is selected by the Schwert Criterion (see Schwert (1989)). This test consists of
estimating a regression of the following equation:
∆y t = α + βy t−1 + δt + ζ 1 ∆y t−1 + ... + ζ k ∆y t−k + u t (5.15)
87 To be exact, we cannot reject the hypothesis of the presence of a unit root.
88 Acronym for “Dickey–Fuller Generalised Least Squares”.

57
where ∆ represents the change from one period to the next and y is the variable whose unit root
we want to test. Then, we test the null hypothesis that y t has a unit root against the alternative
hypothesis that β = δ = ζ 1 = ... = ζ k = 0, in other words, that y t follows a random walk.
Proceeding to this test with our data shows that the results are highly significant (1%
threshold) for each of the 1 to 17 lags selected by the Schwert criterion, which means that the
null hypothesis is rejected. Therefore, even a dynamically complete model evidences that the
daily return on the BEL20 index does not have a unit root. In other words, it follows a random
walk.
This absence of a unit root confirms that the data do not need to be differentiated before
applying the autoregressive models of the coming sections. Understanding stationarity is also
key for the application of the error correction model, to which we now turn.
5.3.2 Error correction model
An error correction model allows us to identify both the long-term and the short-term trends of
a time series. As a complement to the analysis of section 5.2, it would be interesting to have
statistical evidence suggesting whether the impact of political events has a long-term effect or if
it is only short-term. The model of this section was first established for two time series that are
cointegrated, but it can also be applied to stationary time series (according to authors that we
will cite), and to more than two series. In this section, we review in turn the original model of
error correction, then argue why it can be applied for stationary data and finally, we present an
augmented version of the model: the vector error correction model (VECM), which consists of
using more than one independent variable. Ultimately we are able to gain some insights into the
short-term/long-term components of our independent variables.
Error correction model for cointegrated time series (original model)
As Engle and Granger (1987) established, for two or more time series x t , if it is possible to find
a linear combination α ′ x t that is stationary, then “the time series x t are said to be cointegrated
with cointegrating vector α”. In other words, if these series have a degree of integration 1
(i.e. they are “first-order integrated”) while there exists a vector of coefficents that can form
a stationary linear combination of them, then the series are cointegrated. And cointegration
implies, in short-term dynamics, the need to use an error correction model.
In their paper, Engle and Granger (1987) suggested a model of error correction for cointegrated
time series. This Engle-Granger model is defined in two steps and needs all data to be previously
made stationary. The first step consists of a simple regression of the control variable (denoted
X in the following equations), for which we isolate the residuals:
The residuals are given by:
Y t = α + βX t + ɛ t (5.16)
ɛ t = Y t − βX t − α (5.17)
The second step consists of a regression of the change in Y on the lagged change in X and the
lagged errors:
where
∆Y t = β 0 + β 1 ∆X t−1 − γɛ t−1 + u t (5.18)

58
• β 1 captures the short-term effect of X t−1 on Y t . In other words, the short term effect of
the regressor one period before t on the dependent variable in period t.
• γ captures the rate of adjustment of the system, in response to a shock. Since it measures
how fast the errors of the model are corrected over time (which gave its name to the
model), it can thus be understood as the “speed of error correction”.
• u t are the disturbances of the model.
Error correction model for stationary data
Econometricians using stationary data may want to know what the short-term and long-term
effects of their independent variables are and therefore they may want to use an error correction
model for stationary data (and not just for integrated data). Whether the error correction
model can be applied to stationary data has been at the core of many debates (see Beck (1992),
Beck (1993) and Durr (1993) 89 ). Keele and De Boef (2005) 90 devote particular attention to the
topic and find empirical evidence that this model remains valid even for stationary data. We
follow his approach: we proceed to an error correction model with our data and assume that
the results of this model are correct, even though the current debate on the topic is not clear
whether this approach is justified.
Vector error correction model
The vector error correction model (VECM) is based on the approach of Engle and Granger
(1987), but uses a vector of independent variables instead of only one regressor. This approach
has one main drawback; since there is only one long-term coefficient, this coefficient reflects
the long-term effect of all independent variables taken together. The long-term coefficient
can therefore not always be interpreted properly. However, the significance of the short-term
coefficient of each independent variable indicates if the corresponding variable has any significant
short-term effect or not on the dependent variable. If it does not, then we can assume that the
impact of the regressor is mainly long-term. If it does, then we can assume that it has some
effect in the short-term but we cannot say anything about its long-term effect.
Adapting the approach of Engle and Granger (1987), the first step (equation 5.16) needs
only to be adjusted for a vector:
where
Y t = α + βX t + ɛ t (5.19)
ɛ t = Y t − βX t − α (5.20)
• X t represents the vector of all the independent variables of the model.
For our analysis, we first isolate the residuals of equation 5.12 (page 46).
89 It is worth mentioning that all of these papers have been published in the same journal: Political Analysis.
90 It is worth mentioning that this reference is a working paper and has, therefore, never been published.

59
where
The second step, using our independent variables, is formalised as follows:
∆{BEL20 return} t = β 0 +β 1 ∆{Main P olitical indicator} t−1
+β 2 ∆{Euro Stoxx 50 return} t−1
+β 3 ∆{Euro Stoxx 50 return} 2 t−1
+β 4 ∆{Rating dummy} t−1
+β 5 ∆{Y ield on government bonds} t−1
+β 6 ∆{P redictability indicator} t−1
+β 7 ∆{Euribor change} t−1
+β 8 ∆{Money growth} t−1
+γɛ t−1 + u t (5.21)
• β i captures the short-term effect of the corresponding i’s regressor.
• γ captures the long-term effect of all regressors combined.
The results of equation 5.21 are reported in table 5.17 91 .
Table 5.17: Vector error correction model, with the residuals corrected for heteroskedasticity.
Variable Coefficient (Std. Err.)
∆{Main P olitical indicator} t−1 0.512 ∗ (0.233)
∆{Euro Stoxx 50 return} t−1 -0.312 ∗∗ (0.054)
∆{Euro Stoxx 50 return} 2 t−1 -0.013 (0.018)
∆{Rating dummy} t−1 1.458 ∗ (0.633)
∆{Y ield on government bonds} t−1 0.042 ∗∗ (0.015)
∆{P redictabilityindicator} t−1 -0.038 (0.124)
∆{Euribor change} t−1 0.028 (0.129)
∆{Money growth} t−1 0.063 (0.216)
Lagged residuals of equation 5.19 -1.366 ∗∗ (0.261)
Intercept -0.004 (0.087)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%
What table 5.17 suggests is primarily that some of the regressors have long-term effects
(since the coefficient for the lagged residuals is highly significant). Secondly, it also indicates that
the square of the Euro Stoxx 50 daily returns, the estimator of the predictability, the changes in
the Euribor and the growth of M3 seem not to have significant short-term effects. Finally, it
suggests that political events, the returns of the Euro Stoxx 50, the change in rating of the main
rating companies and the changes in the yield on government bonds have short-term effects on
the return of the BEL20. It does not, however, erradicate the possibility that these variables
also have a long-term effect: we just cannot draw any conclusions about it from this model.
91 This regression, initially largely subject to heteroskedasticity of the residuals, has been corrected and is robust
to any arbitrary form of heteroskedasticity.

60
5.3.3 Generalised autoregressive conditional heteroskedasticity model
Variance is a measure of volatility. As Engle (1982) mentions, “[i]n financial theory, the variance
as well as the mean of the rate of return are determinants of portfolio decisions”. Moreover, if
the disturbances of a standard regression at time t are found to be correlated with those at time
t − 1, then we are in a situation of autocorrelation. In such a case, the standard errors (and
thus the variance as well) are likely to be biased downward, thus affecting the significance of the
results. Correctly identifying the variance is, therefore, crucial.
Rationale of the GARCH model
Traditional econometric models, such as those we have used above (ordinary least squares and
two-stage least squares) assume that the disturbances are independent and identically distributed
(IID). It implies that the variance of the regression is constant over time (i.e. the residuals
are homoskedastic). However, this assumption was found to be “very strongly” rejected by
Hsieh (1991) for stock returns. In the 1980s, models releasing this IID implausible assumption
were established. In particular the autoregressive conditional heteroskedasticity model (ARCH),
introduced by Engle (1982), enables a fraction of the variance to change over time, as a function
of its past disturbances. This fraction is called the conditional variance. The rest of the variance,
the unconditional variance, remains constant. With an ARCH model, the variance can, therefore,
learn from its history and adjust over time 92 . Bollerslev (1986) generalised the approach of
Engle (introducing the generalised ARCH or GARCH model), by allowing the variance to be
not only a function of the past disturbances, but also a function of its own past values.
We believe that it is likely that our data are subject to conditional heteroskedasticity because,
several authors find evidence of such heteroskedasticity in daily asset price returns (notably
Bollerslev (1986) and Hsieh (1991) but also 93 Theodossiou (1994), Koutmos and Theodossiou
(1994) and Lobo and Tufte (1998)). In addition, Pynnönen (2005) establishes that ARCH-like
models (in particular the GARCH model) are a better approach to represent the volatility of
stock returns 94 . Indeed, these models are white noise processes, implying non-autocorrelated
residuals (E(ɛ t |ɛ t−1 ) = 0).
It seems also worth mentioning two other interesting qualities of ARCH processes for our
analysis. Firstly, the daily returns of stocks are often subject to leptokurtic distributions (see
the frequence distribution of the daily BEL20 returns over the period of analysis on figure A.2,
page 82), i.e. fat-tailed data, common for stock returns, that an ARCH-like process can correct
(Hsieh (1991)). Secondly, ARCH-like models also capture the effects of variables that would
have been omitted in the estimated model.
In this section, we apply a GARCH model to our results in order to have a relevant measure
of the volatility.
Application of a GARCH model
As we saw previously, if the returns are stationary, then a GARCH process correctly generates
data. We know, since section 5.3.1, that the daily returns of the BEL20 index is stationary. We
can therefore apply the GARCH model without further consideration. This model will reveal if
the variance of the residuals of the models used in section 5.2 were correctly specified. Let us
92 This model, therefore, solves the problem of volatility clustering found in financial data by Mandelbrot (1963).
As he claimed: “large changes tend to be followed by large changes, of either sign, and small changes tend to be
followed by small changes”.
93 The following authors are cited by Bernhard and Leblang (2006), page 185.
94 However, the model also has limitations. As Gregory and Reeves (2010) note, ARCH models do not
appropriately capture extreme movements of data, such as crashes of the stock market.

61
start with a definition of autoregressive models and the theoretical specification of the model
before applying it.
Box and Jenkins (1970) define an autoregressive model in the following way: it is the
expression of the value of a process at time t as a linear function of its previous values and of a
shock. Formally, it can be expressed as:
where
ỹ t = φ 1 ỹ t−1 + φ 2 ỹ t−2 + ... + φ p ỹ t−p + a t (5.22)
• ỹ is the deviation of the dependent variable y from its mean µ: ỹ = y − µ,
• a t is an exogenous shock happening at time t,
• φ v is the estimated coefficient attributed to the lagged deviation of the dependent variable
at time t − v.
The basic generalised autoregressive conditional heteroskedasticity model estimates the
following equation:
Where
{BEL20 return} t = βX t + ɛ t (5.23)
V ar(ɛ t ) = σ 2 t = γ 0 + A(σ t , ɛ t ) + B(σ t , ɛ t ) 2 (5.24)
• βX t is the vector of control variables and their respective coefficient,
• γ 0 is a constant in the model of the residuals’ variance,
• A() and B() are functions that are added to an OLS regression 95 and that depend on σ t
and ɛ t ,
• ɛ t is the vector of residuals of equation 5.23 and
• σ t is the vector of the standard deviations of the residuals ɛ t , at time t.
This regression on our data gives us the results reported in table 5.18. The upper part of the
table is the estimation of the impact of each regressor on the BEL20 index return. It corresponds
to the analysis we made in 5.2, but using a different measure of the volatility. This volatility
is given by the last three lines of the table, which fit the conditional variance estimate, as a
function of its own lag and of a lag of the disturbance. It can be summarised as the following
formula:
where
• ω is the intercept,
• q is the order of the ARCH term ɛ 2 ,
q∑
p∑
σt 2 = ω + α i ɛ 2 t−1 + β i σt−1 2 (5.25)
i=1
i=1
• α i is the coefficient of the lagged squared disturbance of equation 5.23, which corresponds
to L.arch in table 5.18,
95 If A() = B() = 0, then the model is a simple linear regression.

62
• p is the order of the GARCH term σ 2 and
• β i is the coefficient of the lagged variance, which corresponds to L.garch in table 5.18.
“GARCH(1,1) [i.e. the model used here] has been found to capture adequately stock return
volatility” (Pynnönen (2005)).
Table 5.18: Generalised autoregressive conditional heteroskedasticity regression.
Variable Coefficient (Std. Err.)
Equation 1 : BEL20 return
Main political indicator -0.153 (0.117)
Euro Stoxx 50 return 0.799 ∗∗ (0.012)
Euro Stoxx 50 return 2 0.002 (0.003)
Rating variable -0.497 ∗ (0.243)
Government bonds yield -0.009 † (0.005)
Predictability indicator 0.040 † (0.021)
Euribor change 0.074 † (0.039)
Money growth -0.020 (0.014)
Intercept -0.076 † (0.041)
Equation 2 : ARCH
L.arch 0.243 ∗∗ (0.063)
L.garch 0.442 ∗∗ (0.133)
Intercept 0.072 ∗∗ (0.024)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%
The results of table 5.18 can be interpreted as follows. The conditional variance is given by
equation 5.26, whose estimated coefficients (given by the second equation), the intercept, the
ARCH term and the GARCH term, have been added to the theoretical equation 5.25.
σ 2 t = 0.072 + 0.243ɛ 2 t−1 + 0.442σ 2 t−1 (5.26)
All three coefficients of this equation are highly significant, as table 5.18 shows. The
conditional variance therefore increases with time, which evidences that our previous models (the
simple OLS and 2SLS used in section 5.2) were too simplistic. In these previously used models,
the variance was incorrectly specified, which implies that the significance of the coefficient might
have been exaggerated.
Now that we let the residuals change over time, the coefficients of the regressors (in the upper
part of table 5.18) have changed. In particular, the impact of political events on the BEL20
index is smaller, in absolute value, than before, while its standard error has increased. As a
consequence, the significance of the indicator has decreased. The sign of the coefficient remains
the same, indicating that, even though our previous analyses are likely to have overestimated
the impact of political events on the BEL20 returns, the qualitative conclusion that we drew
remains valid.
Even though the flexibility allowed by this model can enhance the relevancy of the analysis, it
is worth noting that the GARCH model suffers from some drawbacks. As Nelson (1991) presents,
the GARCH model’s assumptions are sometimes violated and analysing the long-term effect of
shocks with this model is not easy, since the usual measures of persistence do not agree. We will
now turn to a non-linear model that, according to Nelson (1991), “meets these objections”.

5.3.4 Exponential generalised autoregressive conditional heteroskedasticity
model
Hsieh (1991) investigates why stock market returns are found not to follow independent,
identically distributed processes. After reviewing exogenous changes in the environment and
chaotic dynamics, he finds that the reason behind this rejection is conditional heteroskedasticity.
However, even though we proceeded to a GARCH model on our data, it seems not to be enough
to fully capture the non-linearity of stock returns. Hsieh (1991) uses a model more flexible than
the ARCH or GARCH models and which accommodates the nonlinearity of stock returns better:
the exponential generalised autoregressive conditional heteroskedasticity model (or EGARCH).
Rationale of the EGARCH
This model is attributed to Nelson (1991) 96 , who replaced the linear dependence of the GARCH
model by a logarithmic dependence. Doing so, the EGARCH allows to better fit asymmetric
responses of the variance of stock returns to the direction of the dependent variable. Simply put,
it allows the variance to be bigger when the return is negative than when the return is positive.
This is the reason why the EGARCH model is sometimes called a “leverage” or “asymmetric”
volatility model (Engle and Viktor (1993)).
Formally, the conditional variance estimated by the GARCH model (equation 5.25) is now
modified to fit:
q∑
p∑
log(σt 2 ) = ω + α i ɛ 2 t−1 + β i log(σt−1) 2 (5.27)
i=1
where the only modification to equation 5.25 is the logarithmic fit of the variance σ 2 and of its
lagged value.
Using an EGARCH model seems therefore an even more flexible solution than the GARCH
that we used above. This intuition is confirmed by the comparison “GARCH vs. EGARCH” of
Lee and Brorsen (1997). However, Engle and Viktor (1993) find evidence that the variability of
the conditional variance might be overly high when using the EGARCH model. It is not clear
whether the GARCH or the EGARCH model would better reflect the reality of the Belgian stock
market during the 2010-2011 political crisis. Thereafter, we present the results of an EGARCH
model, using the same specifications as for the previous model.
Application of an EGARCH model
Table 5.19 reports the result of the application of an EGARCH model.
The estimation of the volatility, according to the EGARCH model, results in this equation
for the logarithmic variance of the model:
i=1
log(σ 2 t ) = 0.785 + 0.539ɛ 2 t−1 + 0.576log(σ 2 t−1) (5.28)
As we can see, only little difference is found between the GARCH and EGARCH models,
respectively table 5.18 and table 5.19. The coefficients and standard errors of the regressors are
not drastically different. However, we can observe that the political indicator is now slightly more
significant than with the GARCH model (it now reaches the 11% threshold), as the coefficient
is bigger in absolute value and the standard error smaller. Another difference resides in the
indicator of predictability 97 , which is now bigger and more significant than before.
96 Nelson mentions, however, that authors before him (Pantula and Geweke) used similar models, called
log-GARCH models.
97 Which is a three-day lagged estimation of the variance, computed according to Parkinson (1980).
63

64
Table 5.19: Exponential generalised autoregressive conditional heteroskedasticity regression.
Variable Coefficient (Std. Err.)
Equation 1 : BEL20 return
Main political indicator -0.174 (0.109)
Euro Stoxx 50 return 0.801 ∗∗ (0.012)
Euro Stoxx 50 return 2 0.002 (0.003)
Rating variable -0.508 ∗ (0.242)
Government bonds yield -0.008 (0.005)
Predictability indicator 0.053 ∗ (0.021)
Euribor change 0.066 † (0.037)
Money growth -0.019 (0.014)
Intercept -0.096 ∗ (0.043)
Equation 2 : ARCH
L.arch 0.539 ∗∗ (0.162)
L.egarch 0.576 ∗∗ (0.135)
Intercept -0.785 ∗∗ (0.239)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%
Before summarising the results of the different models and tests investigated so far and
establishing our conclusions, let us turn to some discussion about related topics that are worth
mentioning and investigate some aspects that have not been raised so far.

Chapter 6
Discussion
In this chapter, we investigate various topics that did not belong to any of the previous sections
of this paper. We dig slightly further with regard to our analysis; we give the basis for possible
further investigations of our work and we invite the reader to establish his or her own critical
assessment of our conclusions.
Rationale for the selection of daily data
One choice that we have not discussed so far is the time interval between two consecutive
observations in our procedure. We chose to use daily data, but this is not the shortest interval
that can be used. Indeed, Barclay and Litzenberger (1988) use intradaily observations for event
studies and discuss the relevancy of using such intervals. They find that using intradaily data
increases the power of statistic tests 98 . However they highlight that several methodological
questions should be raised by such an approach. Campbell et al. (1997) conclude that it remains
unclear whether using intradaily data is beneficial. The relevancy of our choice of using daily
data therefore seems untouched.
Possible bias due to uneven intervals between observations
The source data that we selected is the closing price of the daily BEL20 index. The closing
price of a stock reflects the price at which the last transaction of this stock was traded on a
particular day. As Campbell et al. (1997) remind us, it is a common mistake to consider that
the series derived from the closing price has observations that are evenly spaced (24-hour gap).
In fact, if it is not the case (most of the time), then we observe a “nonsynchronous trading
effect”, which results in a bias for the computation of the returns. This effect has been heavily
reviewed in literature 99 but mostly concerns the analysis of one stock at a time. In the case
at hand, however, the possibility for nonsynchronous trading effect is much less likely than
in this literature, because we consider an index instead of a stock. The BEL20 is a real-time
index, computed every 15 seconds, and including the change in the price of any of the 20 stocks
composing it. Given this high regularity, it is most unlikely that our analysis suffers from a
significant bias due to nonsynchronous trading effect.
98 Which means that it increases the probability of rejecting the null hypothesis if this hypothesis is indeed
wrong.
99 See Campbell et al. (1997), page 85, for relevant literature about the topic. Box and Jenkins (1970) insist on
this equal spacing in time condition for the GARCH model notably.
65

66
The normality assumption
The analyses in section 5.2 are based on a model assuming that the returns are jointly normally
distributed (the APT). This assumption may not be empirically verified 100 , which would imply
that the results of the analysis are only asymptotic. However, as Brown and Warner (1985)
discuss, in most cases, this is not considered as an issue since the statistic tests converge rapidly
to their asymptotic distribution. Therefore, our results remain relevant even when we release
the implausible normality assumption.
The same issue has also been raised by Brown and Warner (1985) and by Berry et al. (1990),
but they provide another reason why we should not be worried about it: even if the daily returns
are most of the time not normally distributed, the residuals usually are. And even when the
null hypothesis of normality has to be rejected, they do not find any gain in power of using test
statistics that are distribution-free.
Cross-sectional dependence due to clustered events
One of the possible problems of event studies is cross-sectional dependence, i.e. when events
are “clustered” (issue analysed by Brown and Warner (1980)) 101 . This dependency problem
disappears when events happen randomly within the event-window (Binder (1998)). In our
analysis, one can see (on figure 4.2 or in the appendices) that, during the last quarter of the
period of analysis, the events seem to be somehow clustered. Indeed, the agreements were
reached one after the other during this period. At the same time, Moody’s placed the Belgian
grade under review and S&P downgraded the country. As Brown and Warner (1980) mention,
this clustering 102 affects the randomness in the selection of securities performing well. However,
this is something very different to what we are doing (since we do not rely on our data to select
particular securities) and therefore it does not challenge the relevancy of our approach. We
considered that it was worth mentioning, though.
Inconsistent coefficient estimates due to a model driven by the choices of agents
The paper of Acharya (1993) about latent information in event studies mentions that “coefficient
estimators in a dummy variable model are inconsistent when data on the dependent variable
(stock return) are limited by the choices of agents (decision rules that trigger events)”. It seems,
therefore, natural to consider whether our analysis suffers from inconsistency of the coefficient
estimates, because political events may be somehow driven by the choices of the political actors.
We believe that it is not the case for two reasons. Firstly, Acharya focuses on the returns of
particular firms, after company-specific events, such as the announcement of a dividend. In this
case, the event is indeed a choice, while for our analysis, the events that we consider are more the
consequence of a process (attempts to the formation of a government). Secondly, the example of
Acharya concerns one or few agents (the board of the company), while in our case, there is a
multitude of agents commonly responsible for the occurence of the events. In particular there
are at least the nine parties which had a role in the negotiations during the political crisis, as
well as the King, the rating agencies and most likely many other institutions with an influence
on the political scene. Therefore, as for the previous potential problems considered, our analysis
seems not to suffer from this weakness.
100 Even though, as figure A.2 (in the appendices) shows, the returns of the BEL20 index over the period of
consideration seem to show an approximately normal distribution. However, one can also notice that kurtosis
seem to be present in the data.
101 The clustering problem is also sometimes called “Moulton problem” (Angrist and Pischke (2009)).
102 Called “calendar clustering” or “event clustering” by Henderson (1990).

67
Bias of the instrumental variables
As Angrist and Pischke (2009) remind us, instrumental variables can sometimes be a source
of bias, especially in small samples, which would imply that the estimated coefficients are
not correct. The source of this common bias in empirical studies using 2SLS is due to weak
instruments, that is, instruments that are not very correlated with the parameter they estimate.
However, we should not worry about such a bias in our case since, as we presented earlier, the
correlation between the instrumental variable and the instrumented variable is above 99%. The
resulting coefficient of the regression can, therefore, be assumed to be consistent.
Another way to verify the validity of our instrumental variable is to proceed to a Sargan test.
It consists of testing the overidentifying restrictions, or in other words to testing if the instruments
are correlated to the residuals and therefore if we can rely on them as good instruments. To
proceed to this test, we (i) isolate the residuals ɛ t of equation 5.12, (ii) fit the same regression
but replacing BEL20 return t by ɛ t and (iii) verify if any of the regressors are significant.
Doing so indicates that no regressor significantly explains the residuals. It therefore suggests
that our instruments are correct and thus that the estimated coefficients are unbiased.
Possible endogeneity of the rating variable with other regressors
One problem we have not raised so far is the possibility that the rating variable has an influence
on the other independent variables of the model. In other words, this dummy variable may be
endogenous with the other regressors. The implication of this possibility would be that the
coefficient of the dummy and of the regressors with which it is endogenous would be biased.
However, as long as the rating variable is not endogenous with the political indicator, it is
not a major problem for our analysis. What is of interest to us is to measure the impact of
the political indicator on the fluctuations of the BEL20 daily returns, not the impact of the
other variables on it, therefore, the presence of a bias due to endogeneity between two control
variables should not change the conclusions of our model. After verification, the rating dummy
variable is not a significant explanatory variable of the variations in our political indicator. We
can, therefore, consider that the inclusion of the rating dummy variable is perfectly justified and
does not bias our conclusions.
Not unique maximum likelihood estimate (MLE) of the coefficient, due to the
presence of dummy variables
Doornik and Ooms (2008) have investigated GARCH and EGARCH models using dummy
variables and find that this procedure might result in a “multimodality problem”. In such a case,
the coefficient of the dummy variable, as computed by the MLE, might not be unique.
What can be done to solve that problem is to simplify the specification of the problem,
eliminating (at least some of) the dummy variables and to re-estimate the model. Following this
procedure, we fitted other GARCH and EGARCH models, according to the same specification
as in the extensions of this thesis, but without the rating dummy variable. Doing so, we may
lose some accuracy (since a relevant control variable is omitted), but according to Doornik and
Ooms (2008), at least our indicators would be convergent. The results obtained are interesting:
the coefficient of the political indicator (β 1 ) increases and its standard error decreases, making it
more significant for both the GARCH and the EGARCH model. The coefficient of the political
indicator becomes 0.172 and 0.184, respectively for each of these models, and is even significant
at the 10% threshold for the EGARCH model.
One can notice that following this procedure yields results that are closer to the conventional
least squares model. However, since we could not find many authors applying this relatively

68
recent procedure, we simply indicated these results as a discussion and we did not replace the
models used in the extensions. Further investigations of the paper of Doornik and Ooms (2008)
are probably necessary to provide a proper summary of the relevancy of this procedure.
Alternative versions of the EGARCH model
The original EGARCH model established by Nelson (1991), and used in section 5.3.4, can
be modified in several respects, in order to better accommodate the possible asymmetries of
exogenous shocks. In particular, separate coefficients can be used for small and for large shocks.
This procedure has been investigated by Wu and Xiao (2002), who use seven different versions
of the EGARCH model to find the best possible fit for the conditional volatility of returns
subject to shocks. Their results show that, indeed, doing so enabled them to better capture
the asymmetric effect of shocks on returns. Reaching such level of precision for the conditional
volatility of our model is beyond the scope of this thesis but a more advanced study of the topic
could investigate these alternative EGARCH models. As we found that the impact of positive
and negative political events are different from each other, the volatility is likely to respond
differently as well. This approach constitutes, therefore, a possible continuation to our work.
Let us now turn to other possibilities of extension of this analysis.
Possible further work
The simple model of this paper (section 5.2) seems rather exhaustive, since it takes into account
(to the best of our knowledge) most important control variables, it corrects the weaknesses of the
data and its robustness is tested in several respects. However, further work could complete this
thesis. Firstly, our models do not take into account the news about the future macroeconomic
situation that is known by the public but not contained within the variables used at the time
of the analysis. Analysing that kind of events was not our purpose, but their inclusion in the
analysis is clearly a first possible broadening of our work.
Secondly, we introduced relevant extensions to this simple model, but the analysis that we
provided in section 5.3, extensions, is not exhaustive. Indeed, other models (even if we believe
that they are less relevant than those we used) could provide a complementary analysis of the
topic. Verifying whether their results are in line with what we found would be another way to
evaluate the robustness of our conclusions.
Thirdly, as we have mentioned in the literature review, most papers finding significant results
about the topic were done in periods of political instability. As another way to check the
robustness of this paper, it would be interesting to see how much political events affect the stock
market in periods of political stability. Such additional work would allow us to derive conclusions
about whether the causality relationship between political events and the stock market was
affected by the fact that the country had no federal government during the period of analysis.
Fourthly, the robustness of our findings could also be tested further by analysing other
political crises in Belgium. Fortunately for the researcher interested in the topic, Belgium
provides many other political crises to analyse, even though they are not as long as the 2010-2011
one. In particular, in 2007, 194 days were necessary to form a government in Belgium. Proceeding
to a similar analysis, but over this period, would be a good way to confirm (or not) the relevance
of our results.
Fifthly, as Pynnönen (2005) showed, an “abnormal return” analysis is equivalent to a dummy
variable event study, under certain conditions (non-overlapping event windows). Analysing the
Belgian political crisis of 2010-2011 following the classical abnormal return analysis could be
interesting to test further the robustness of our analysis and to verify if the two methods are
indeed similar, as Pynnönen (2005) claims theoretically.

As a final word about these possible further works, we would like to mention that even
though some more results could certainly be found from these extensions, it is likely that they
would be of only little added-value compared to the conclusions that we reached in this thesis.
Let us now present them.
69

Chapter 7
Conclusion
The political history of Belgium has shown how two very different communities could be
held together, by means of agreements and reforms. As Flanders prospered, the claims of its
inhabitants became increasingly more constraining, requiring successive governments to modify
the political structure of Belgium, which eventually became exceedingly complex. Craving more
independence over the past decades and taking advantage of its recently acquired economic
prosperity, Flanders successfully obtained larger competences, draining the power from the
federal state to the regional level. At the same time, the Europeanisation process weakened the
Belgian federal state as other competences were relinquished to the supranational level. The
national issues grew in importance, casting doubts on the existence of a Belgian identity for the
inhabitants. The governments stalemated one after the other on the increasingly problematic
community disagreements. It is against this background that the Belgian political crisis 2010-2011
took place. Eventually, Belgium endured the world’s longest period in recent history as a state
without a federal government. It was our purpose to analyse this period through a stock market
approach.
Our political event study was based on multiple models, very common in financial econometrics,
which showed interesting and significant causal relationships between the political
events which drove the Belgian political crisis and the fluctuations of the stock market over this
period. We detailed each model and provided the results of intermediary steps, so as to let the
reader evaluate the relevancy of each component of it. We started with a basic ordinary least
squares model, which we corrected for endogeneity of the political indicator, therefore applying
a two-stage least squares procedure. Once augmented with relevant macroeconomic variables 103 ,
the model was corrected for the heteroskedasticity of its residuals. Different robustness tests
were performed in order to challenge the relevancy of the significant results found. Then, we
extended the procedure to more sophisticated models, including error correction and autoregressive
conditional heteroskedasticity models 104 . The plurality of models within the analysis was
interesting as it let us take some distance with the topic by analysing different views of the same
data. Hereafter, we summarise our conclusions, but we remain open to the possibility of further
improvement of the models used and therefore of the results. Let us also remember that these
conclusions are context-specific. We do not claim that they are applicable to any political crisis
situation, even though we believe, given the robustness of the indicators, that similar contexts
are likely to yield, to some extent, similar results.
In our main model, using a two-stage least squares procedure, we gave robust evidence that
103 The problem of the presence of an economic crisis simultaneously with the political crisis was solved through
the addition of these variables to the model.
104 In particular, we proceed to a VECM, to a GARCH model and to an EGARCH model.
70

71
the main political events had a highly significant impact 105 on the fluctuations of the main
Belgian stock index, the BEL20. This result is relatively free of endogeneity, of anticipatory
movements and of heteroskedasticity of the residuals. As we extensively argued, every control
variable used seems relevant to the analysis, therefore enhancing the accuracy of the political
indicator. The political indicator shows a negative relationship between the political events and
the stock market fluctuations: it reacts positively to negative events and negatively to positive
events. By “positive event”, we mean events that are favourable to the establishment of a fully
empowered government and therefore to the end of the political crisis. Similarly, by “negative
event”, we mean events that are unfavourable to it. The average reaction of the stock market
to a meaningful political event is about 0.24 percentage points, but the reaction is different
depending on the type of event. For positive events, the average (negative) reaction amounts
about 0.40 percentage points while for negative events, the (positive) reaction reaches only about
0.12 percentage points and is therefore much less significant.
Further specifications of the model showed that dropping some of the key-events of the
crisis did not make the indicator lose its high significance, which testified to its robustness.
We also found that the market seemed to fully react to a political event on the exact date of
the event: no significant anticipatory or lagged movement could be found. In addition, we
provided evidence that the market reacts more importantly to more meaningful political events
than to less meaningful ones. As an extension to our main model, we investigated a vector
error correction model which seemed to indicate that the events had a significant short-term
impact on the Belgian stock returns, but we cannot eliminate the possibility that part of the
impact has long-term consequences. As the time series of the BEL20 daily returns seemed to
be characterised by the same shape and features as usual financial time series, we used models
designed to better fit such series. The autoregressive conditional heteroskedasticity models used
showed that the coefficient and significance of our indicator might have been exagerated by the
more conventional models, but the direction of the shocks (the negative sign) was validated.
Finally, let us now suggest a possible interpretation of these results. Based on the evidence
we found, we believe that the stock market progressively and smoothly anticipates the political
events, and even overanticipates them. Then, on the day of the event, the market adjustment is
negative, as a response to the overanticipation. The results of the many tests and regressions to
which we proceeded are in line with this interpretation. However, despite the significance and
robustness of the results reached, we cannot argue that these conclusions apply to countries other
than Belgium. As we saw in the literature review, authors find different results for other parts
of the world and during different periods. The Belgian case seems, therefore, very particular, if
not unique.
A topic such as this one can never be fully investigated. Further work is always possible to
understand it more completely. However, we believe that we have provided a relevant analysis,
addressing the main problems of the subject matter and using appropriated approaches to model
it. We have sketched some ideas for possible further investigations, in the discussions of this
thesis.
Political event studies are, unfortunately, underexploited in econometric literature. “Standing
on the shoulders of giant” is only possible to a limited extent in this field, because these studies
do not seem to be built on one another, but only on a restrained set of papers. We believe,
though, in the high potential value of political event studies, as it can help investors to refine
their expectations of the financial markets.
We would like to end with a word about the political crisis. As Gordon Brown puts it when
referring to the recent financial crisis: “[s]ometimes it’s a crisis that forces change. The world
105 The coefficient of our political indicator reaches almost the 1% threshold, implying that the hypothesis that
the coefficient is equal to zero is strongly rejected.

72
that emerges out of the economic and financial crisis of 2007-2010 won’t be the same. The
banking and finance system will be based on sounder principles. There is a huge opportunity
over the next 10 or 20 years (to improve things)” 106 . We believe that a similar school of thought
applies to the Belgian political crisis. The reform resulting from the crisis has certainly given
more power to the federated entities of Belgium. But the objective is not holding a bigger share
of power, the objective is to efficientely share it, in order to have a country “based on sounder
principles”. The 2010-2011 crisis was far from being the first political crisis of Belgium and it
will probably not be the last. However, we hope that this crisis raised awareness for all Belgians,
about the common outlook for the future that they all share.
106 Cited in Benedikter (2011).

Appendices
73

Number of non-zero observations: 4 10 13 23 23 40 46 46 75
Proportion of non-zero observations: 0.93% 2.33% 3.02% 5.35% 5.35% 9.30% 10.70% 10.70% 17.44%
74
Day
Event
Value
Mon 19 Apr 10 1 0 0 0 0 0 0 0 0 0
Tue 20 Apr 10 April 20: J.-L. Dehaene considers his mission finished. 2 0 0 0 0 0 -1 0 0 -1
Wed 21 Apr 10 3 0 0 0 0 0 0 0 0 -1
April 22: Alexander de Croo ( Open VLD) leaves the
Thu 22 Apr 10
government.
4 0 0 1 -1 1 -1 -1 1 -1
Fri 23 Apr 10 5 0 0 0 0 0 0 -1 1 -1
April 26: the King Albert II accepts the government's
Mon 26 Apr 10
resignation.
6 0 0 1 -1 1 -1 -1 1 -1
Tue 27 Apr 10 7 0 0 0 0 0 0 -1 1 -1
Wed 28 Apr 10 8 0 0 0 0 0 0 0 0 0
April 29: the Flemish parties start examining BHV. A
Thu 29 Apr 10 motion is filed by French-speaking parties, triggering 9 0 0 1 -1 1 -1 -1 1 -1
the alarm.
Fri 30 Apr 10 10 0 0 0 0 0 0 -1 1 -1
Mon 3 May 10 11 0 0 0 0 0 0 0 0 0
Tue 4 May 10 12 0 0 0 0 0 0 0 0 0
Wed 5 May 10 13 0 0 0 0 0 0 0 0 0
Thu 6 May 10 14 0 0 0 0 0 0 0 0 0
Fri 7 May 10 15 0 0 0 0 0 0 0 0 0
Mon 10 May 10 16 0 0 0 0 0 0 0 0 0
Tue 11 May 10 17 0 0 0 0 0 0 0 0 0
Wed 12 May 10 18 0 0 0 0 0 0 0 0 0
Thu 13 May 10 19 0 0 0 0 0 0 0 0 0
Fri 14 May 10 20 0 0 0 0 0 0 0 0 0
Mon 17 May 10 21 0 0 0 0 0 0 0 0 0
Tue 18 May 10 22 0 0 0 0 0 0 0 0 0
Wed 19 May 10 23 0 0 0 0 0 0 0 0 0
Thu 20 May 10 24 0 0 0 0 0 0 0 0 0
Fri 21 May 10 25 0 0 0 0 0 0 0 0 0
Mon 24 May 10 26 0 0 0 0 0 0 0 0 0
Tue 25 May 10 27 0 0 0 0 0 0 0 0 0
Wed 26 May 10 28 0 0 0 0 0 0 0 0 0
Thu 27 May 10 29 0 0 0 0 0 0 0 0 0
Fri 28 May 10 30 0 0 0 0 0 0 0 0 0
Mon 31 May 10 31 0 0 0 0 0 0 0 0 0
Tue 1 Jun 10 32 0 0 0 0 0 0 0 0 0
Wed 2 Jun 10 33 0 0 0 0 0 0 0 0 0
Thu 3 Jun 10 34 0 0 0 0 0 0 0 0 0
Fri 4 Jun 10 35 0 0 0 0 0 0 0 0 0
Mon 7 Jun 10 36 0 0 0 0 0 0 0 0 0
Tue 8 Jun 10 37 0 0 0 0 0 0 0 0 0
Wed 9 Jun 10 38 0 0 0 0 0 0 0 0 0
Thu 10 Jun 10 39 0 0 0 0 0 0 0 0 0
Fri 11 Jun 10 40 0 0 0 0 0 0 0 0 0
Mon 14 Jun 10 June 13: federal legislative elections. 41 0 0 1 -1 1 -1 -1 1 -1
Tue 15 Jun 10 42 0 0 0 0 0 0 -1 1 -1
Wed 16 Jun 10 43 0 0 0 0 0 0 0 0 0
June 17: Bart De Wever wins the elections. He is
Thu 17 Jun 10
appointed as informateur.
44 0 0 0 0 0 -1 0 0 -1
Fri 18 Jun 10 45 0 0 0 0 0 0 0 0 -1
Mon 21 Jun 10 46 0 0 0 0 0 0 0 0 0
Tue 22 Jun 10 47 0 0 0 0 0 0 0 0 0
Wed 23 Jun 10 48 0 0 0 0 0 0 0 0 0
Thu 24 Jun 10 49 0 0 0 0 0 0 0 0 0
Fri 25 Jun 10 50 0 0 0 0 0 0 0 0 0
Mon 28 Jun 10 51 0 0 0 0 0 0 0 0 0
Tue 29 Jun 10 52 0 0 0 0 0 0 0 0 0
Wed 30 Jun 10 53 0 0 0 0 0 0 0 0 0
Rating dummy
Significant positive event
Significant negative event
Main political indicator
Significant event dummy
All political events
Main political indicator 2
Significant event dummy 2
All political events 2

75
Thu 1 Jul 10 54 0 0 0 0 0 0 0 0 0
Fri 2 Jul 10 55 0 0 0 0 0 0 0 0 0
Mon 5 Jul 10 56 0 0 0 0 0 0 0 0 0
Tue 6 Jul 10 57 0 0 0 0 0 0 0 0 0
Wed 7 Jul 10 58 0 0 0 0 0 0 0 0 0
July 8: Bart De Wever gives up. Elio Di Rupo is put in
Thu 8 Jul 10
charge of forming a government.
59 0 0 0 0 0 1 0 0 1
Fri 9 Jul 10 60 0 0 0 0 0 0 0 0 1
Mon 12 Jul 10 61 0 0 0 0 0 0 0 0 0
Tue 13 Jul 10 62 0 0 0 0 0 0 0 0 0
Wed 14 Jul 10 63 0 0 0 0 0 0 0 0 0
Thu 15 Jul 10 64 0 0 0 0 0 0 0 0 0
Fri 16 Jul 10 65 0 0 0 0 0 0 0 0 0
Mon 19 Jul 10 66 0 0 0 0 0 0 0 0 0
Tue 20 Jul 10 67 0 0 0 0 0 0 0 0 0
Wed 21 Jul 10 68 0 0 0 0 0 0 0 0 0
Thu 22 Jul 10 69 0 0 0 0 0 0 0 0 0
Fri 23 Jul 10 70 0 0 0 0 0 0 0 0 0
Mon 26 Jul 10 71 0 0 0 0 0 0 0 0 0
Tue 27 Jul 10 72 0 0 0 0 0 0 0 0 0
Wed 28 Jul 10 73 0 0 0 0 0 0 0 0 0
Thu 29 Jul 10 74 0 0 0 0 0 0 0 0 0
Fri 30 Jul 10 75 0 0 0 0 0 0 0 0 0
Mon 2 Aug 10 76 0 0 0 0 0 0 0 0 0
Tue 3 Aug 10 77 0 0 0 0 0 0 0 0 0
Wed 4 Aug 10 78 0 0 0 0 0 0 0 0 0
Thu 5 Aug 10 79 0 0 0 0 0 0 0 0 0
Fri 6 Aug 10 80 0 0 0 0 0 0 0 0 0
Mon 9 Aug 10 81 0 0 0 0 0 0 0 0 0
Tue 10 Aug 10 82 0 0 0 0 0 0 0 0 0
Wed 11 Aug 10 83 0 0 0 0 0 0 0 0 0
Thu 12 Aug 10 84 0 0 0 0 0 0 0 0 0
Fri 13 Aug 10 85 0 0 0 0 0 0 0 0 0
Mon 16 Aug 10 86 0 0 0 0 0 0 0 0 0
Tue 17 Aug 10 87 0 0 0 0 0 0 0 0 0
Wed 18 Aug 10 88 0 0 0 0 0 0 0 0 0
Thu 19 Aug 10 89 0 0 0 0 0 0 0 0 0
Fri 20 Aug 10 90 0 0 0 0 0 0 0 0 0
Mon 23 Aug 10 91 0 0 0 0 0 0 0 0 0
Tue 24 Aug 10 92 0 0 0 0 0 0 0 0 0
Wed 25 Aug 10 93 0 0 0 0 0 0 0 0 0
Thu 26 Aug 10 94 0 0 0 0 0 0 0 0 0
Fri 27 Aug 10 95 0 0 0 0 0 0 0 0 0
August 29: refusal of Di Rupo's resignation by the
Mon 30 Aug 10
King.
96 0 0 1 -1 1 -1 -1 1 -1
Tue 31 Aug 10 97 0 0 0 0 0 0 -1 1 -1
Wed 1 Sep 10 98 0 0 0 0 0 0 0 0 0
Thu 2 Sep 10 99 0 0 0 0 0 0 0 0 0
Fri 3 Sep 10 September 3: Di Rupo openly accepts his failure. 100 0 0 0 0 0 -1 0 0 -1
September 4: the King accepts Di Rupo's
Mon 6 Sep 10 101 0 0 1 -1 1 -1 -1 1 -1
resignation. Pieters and Flahaut are put in charge.
Tue 7 Sep 10 102 0 0 0 0 0 0 -1 1 -1
Wed 8 Sep 10 103 0 0 0 0 0 0 0 0 0
Thu 9 Sep 10 104 0 0 0 0 0 0 0 0 0
Fri 10 Sep 10 105 0 0 0 0 0 0 0 0 0
Mon 13 Sep 10 106 0 0 0 0 0 0 0 0 0
Tue 14 Sep 10 107 0 0 0 0 0 0 0 0 0
Wed 15 Sep 10 108 0 0 0 0 0 0 0 0 0
Thu 16 Sep 10 109 0 0 0 0 0 0 0 0 0
Fri 17 Sep 10 110 0 0 0 0 0 0 0 0 0
Mon 20 Sep 10 111 0 0 0 0 0 0 0 0 0
Tue 21 Sep 10 112 0 0 0 0 0 0 0 0 0
Wed 22 Sep 10 113 0 0 0 0 0 0 0 0 0
Thu 23 Sep 10 114 0 0 0 0 0 0 0 0 0
Fri 24 Sep 10 115 0 0 0 0 0 0 0 0 0
Mon 27 Sep 10 116 0 0 0 0 0 0 0 0 0
Tue 28 Sep 10 117 0 0 0 0 0 0 0 0 0
Wed 29 Sep 10 118 0 0 0 0 0 0 0 0 0
Thu 30 Sep 10 119 0 0 0 0 0 0 0 0 0

Fri 1 Oct 10 120 0 0 0 0 0 0 0 0 0
October 4: press conference of the N-VA, unilaterally
Mon 4 Oct 10 121 0 0 0 0 0 -1 0 0 -1
interrupting the negotiations with seven parties.
October 5: the King relieves Flahaut and Pieters from
Tue 5 Oct 10
their mission.
122 0 0 0 0 0 0 0 0 -1
Wed 6 Oct 10 123 0 0 0 0 0 0 0 0 0
Thu 7 Oct 10 124 0 0 0 0 0 0 0 0 0
Fri 8 Oct 10 October 8: De Wever becomes clarifier.
125 0 0 0 0 0 0 0 0 0
Mon 11 Oct 10 126 0 0 0 0 0 0 0 0 0
Tue 12 Oct 10 127 0 0 0 0 0 0 0 0 0
Wed 13 Oct 10 128 0 0 0 0 0 0 0 0 0
Thu 14 Oct 10 129 0 0 0 0 0 0 0 0 0
Fri 15 Oct 10 130 0 0 0 0 0 0 0 0 0
Mon 18 Oct 10 October 18: Bart De Wever: “fabula acta est”. 131 0 0 1 -1 1 -1 -1 1 -1
Tue 19 Oct 10 132 0 0 0 0 0 0 -1 1 -1
Wed 20 Oct 10 133 0 0 0 0 0 0 0 0 0
October 21: Johan Vande Lanotte is designated as
Thu 21 Oct 10
mediator.
134 0 1 0 1 1 1 1 1 1
Fri 22 Oct 10 135 0 0 0 0 0 0 1 1 1
Mon 25 Oct 10 136 0 0 0 0 0 0 0 0 0
Tue 26 Oct 10 137 0 0 0 0 0 0 0 0 0
Wed 27 Oct 10 138 0 0 0 0 0 0 0 0 0
Thu 28 Oct 10 139 0 0 0 0 0 0 0 0 0
Fri 29 Oct 10 140 0 0 0 0 0 0 0 0 0
Mon 1 Nov 10 141 0 0 0 0 0 0 0 0 0
Tue 2 Nov 10 142 0 0 0 0 0 0 0 0 0
Wed 3 Nov 10 143 0 0 0 0 0 0 0 0 0
Thu 4 Nov 10 144 0 0 0 0 0 0 0 0 0
Fri 5 Nov 10 145 0 0 0 0 0 0 0 0 0
Mon 8 Nov 10 146 0 0 0 0 0 0 0 0 0
Tue 9 Nov 10 147 0 0 0 0 0 0 0 0 0
Wed 10 Nov 10 148 0 0 0 0 0 0 0 0 0
Thu 11 Nov 10 149 0 0 0 0 0 0 0 0 0
Fri 12 Nov 10 150 0 0 0 0 0 0 0 0 0
Mon 15 Nov 10 151 0 0 0 0 0 0 0 0 0
Tue 16 Nov 10 152 0 0 0 0 0 0 0 0 0
Wed 17 Nov 10 153 0 0 0 0 0 0 0 0 0
Thu 18 Nov 10 154 0 0 0 0 0 0 0 0 0
Fri 19 Nov 10 155 0 0 0 0 0 0 0 0 0
Mon 22 Nov 10 156 0 0 0 0 0 0 0 0 0
Tue 23 Nov 10 157 0 0 0 0 0 0 0 0 0
Wed 24 Nov 10 158 0 0 0 0 0 0 0 0 0
Thu 25 Nov 10 159 0 0 0 0 0 0 0 0 0
Fri 26 Nov 10 160 0 0 0 0 0 0 0 0 0
Mon 29 Nov 10 161 0 0 0 0 0 0 0 0 0
Tue 30 Nov 10 162 0 0 0 0 0 0 0 0 0
Wed 1 Dec 10 163 0 0 0 0 0 0 0 0 0
Thu 2 Dec 10 164 0 0 0 0 0 0 0 0 0
Fri 3 Dec 10 165 0 0 0 0 0 0 0 0 0
Mon 6 Dec 10 166 0 0 0 0 0 0 0 0 0
Tue 7 Dec 10 167 0 0 0 0 0 0 0 0 0
Wed 8 Dec 10 168 0 0 0 0 0 0 0 0 0
Thu 9 Dec 10 169 0 0 0 0 0 0 0 0 0
Fri 10 Dec 10 170 0 0 0 0 0 0 0 0 0
Mon 13 Dec 10 171 0 0 0 0 0 0 0 0 0
December 14: S&P place Belgium under negative
Tue 14 Dec 10
outlook and threaten it with a downgrade.
172 1 0 0 0 0 0 0 0 0
Wed 15 Dec 10 173 0 0 0 0 0 0 0 0 0
Thu 16 Dec 10 174 0 0 0 0 0 0 0 0 0
Fri 17 Dec 10 175 0 0 0 0 0 0 0 0 0
Mon 20 Dec 10 176 0 0 0 0 0 0 0 0 0
Tue 21 Dec 10 177 0 0 0 0 0 0 0 0 0
Wed 22 Dec 10 178 0 0 0 0 0 0 0 0 0
Thu 23 Dec 10 179 0 0 0 0 0 0 0 0 0
Fri 24 Dec 10 180 0 0 0 0 0 0 0 0 0
December 25: Belgian record of the longest political
Mon 27 Dec 10
crisis.
181 0 0 1 -1 1 -1 -1 1 -1
Tue 28 Dec 10 182 0 0 0 0 0 0 -1 1 -1
Wed 29 Dec 10 183 0 0 0 0 0 0 0 0 0
Thu 30 Dec 10 184 0 0 0 0 0 0 0 0 0
Fri 31 Dec 10 185 0 0 0 0 0 0 0 0 0
76

Mon 3 Jan 11 186 0 0 0 0 0 0 0 0 0
Tue 4 Jan 11 187 0 0 0 0 0 0 0 0 0
Wed 5 Jan 11 188 0 0 0 0 0 0 0 0 0
Thu 6 Jan 11 January 6: Vande Lanotte resigns.
189 0 0 1 -1 1 -1 -1 1 -1
Fri 7 Jan 11 190 0 0 0 0 0 0 -1 1 -1
January 8: European record of the longest political
Mon 10 Jan 11
crisis.
191 0 0 1 -1 1 -1 -1 1 -1
January 11: the King refuses the resignation of Johan
Tue 11 Jan 11
Vande Lanotte.
192 0 0 0 0 0 -1 -1 1 -1
Wed 12 Jan 11 193 0 0 0 0 0 0 0 0 -1
Thu 13 Jan 11 194 0 0 0 0 0 0 0 0 0
Fri 14 Jan 11 195 0 0 0 0 0 0 0 0 0
Mon 17 Jan 11 196 0 0 0 0 0 0 0 0 0
Tue 18 Jan 11 197 0 0 0 0 0 0 0 0 0
Wed 19 Jan 11 198 0 0 0 0 0 0 0 0 0
Thu 20 Jan 11 199 0 0 0 0 0 0 0 0 0
Fri 21 Jan 11 200 0 0 0 0 0 0 0 0 0
Mon 24 Jan 11 201 0 0 0 0 0 0 0 0 0
Tue 25 Jan 11 202 0 0 0 0 0 0 0 0 0
January 26: Vande Lanotte definitely resigns. The
Wed 26 Jan 11
King accepts his resignation.
203 0 0 1 -1 1 -1 -1 1 -1
Thu 27 Jan 11 204 0 0 0 0 0 0 -1 1 -1
Fri 28 Jan 11 205 0 0 0 0 0 0 0 0 0
Mon 31 Jan 11 206 0 0 0 0 0 0 0 0 0
Tue 1 Feb 11 207 0 0 0 0 0 0 0 0 0
February 2: the King increases the power of the
Wed 2 Feb 11 caretaker government. Didier Reynders is named 208 0 0 0 0 0 1 0 0 1
informateur.
Thu 3 Feb 11 209 0 0 0 0 0 0 0 0 1
Fri 4 Feb 11 210 0 0 0 0 0 0 0 0 0
Mon 7 Feb 11 211 0 0 0 0 0 0 0 0 0
Tue 8 Feb 11 212 0 0 0 0 0 0 0 0 0
Wed 9 Feb 11 213 0 0 0 0 0 0 0 0 0
Thu 10 Feb 11 214 0 0 0 0 0 0 0 0 0
Fri 11 Feb 11 215 0 0 0 0 0 0 0 0 0
Mon 14 Feb 11 216 0 0 0 0 0 0 0 0 0
Tue 15 Feb 11 217 0 0 0 0 0 0 0 0 0
February 16: the King increases the length of Reynders'
Wed 16 Feb 11 218 0 0 0 0 0 1 0 0 1
mission, allowing him to hand in a more detailed report.
Thu 17 Feb 11 219 0 0 0 0 0 0 0 0 1
Fri 18 Feb 11 220 0 0 0 0 0 0 0 0 0
Mon 21 Feb 11 221 0 0 0 0 0 0 0 0 0
Tue 22 Feb 11 222 0 0 0 0 0 0 0 0 0
Wed 23 Feb 11 223 0 0 0 0 0 0 0 0 0
Thu 24 Feb 11 224 0 0 0 0 0 0 0 0 0
Fri 25 Feb 11 225 0 0 0 0 0 0 0 0 0
Mon 28 Feb 11 226 0 0 0 0 0 0 0 0 0
Tue 1 Mar 11 March 1: Reynders is relieved from his mission
227 0 0 0 0 0 -1 0 0 -1
Wed 2 Mar 11 March 2: Wouter Beke is designated as negotiator. 228 0 0 0 0 0 1 0 0 1
Thu 3 Mar 11 229 0 0 0 0 0 0 0 0 1
Fri 4 Mar 11 230 0 0 0 0 0 0 0 0 0
Mon 7 Mar 11 231 0 0 0 0 0 0 0 0 0
Tue 8 Mar 11 232 0 0 0 0 0 0 0 0 0
Wed 9 Mar 11 233 0 0 0 0 0 0 0 0 0
Thu 10 Mar 11 234 0 0 0 0 0 0 0 0 0
Fri 11 Mar 11 235 0 0 0 0 0 0 0 0 0
Mon 14 Mar 11 236 0 0 0 0 0 0 0 0 0
Tue 15 Mar 11 237 0 0 0 0 0 0 0 0 0
Wed 16 Mar 11 238 0 0 0 0 0 0 0 0 0
Thu 17 Mar 11 239 0 0 0 0 0 0 0 0 0
Fri 18 Mar 11 240 0 0 0 0 0 0 0 0 0
Mon 21 Mar 11 241 0 0 0 0 0 0 0 0 0
Tue 22 Mar 11 242 0 0 0 0 0 0 0 0 0
Wed 23 Mar 11 243 0 0 0 0 0 0 0 0 0
Thu 24 Mar 11 244 0 0 0 0 0 0 0 0 0
Fri 25 Mar 11 245 0 0 0 0 0 0 0 0 0
Mon 28 Mar 11 246 0 0 0 0 0 0 0 0 0
Tue 29 Mar 11 247 0 0 0 0 0 0 0 0 0
Wed 30 Mar 11 March 30: world record of the longest political crisis. 248 0 0 1 -1 1 -1 -1 1 -1
Thu 31 Mar 11 249 0 0 0 0 0 0 -1 1 -1
77

78
Fri 1 Apr 11 250 0 0 0 0 0 0 0 0 0
Mon 4 Apr 11 251 0 0 0 0 0 0 0 0 0
Tue 5 Apr 11 252 0 0 0 0 0 0 0 0 0
Wed 6 Apr 11 253 0 0 0 0 0 0 0 0 0
Thu 7 Apr 11 254 0 0 0 0 0 0 0 0 0
Fri 8 Apr 11 255 0 0 0 0 0 0 0 0 0
Mon 11 Apr 11 256 0 0 0 0 0 0 0 0 0
Tue 12 Apr 11 257 0 0 0 0 0 0 0 0 0
Wed 13 Apr 11 258 0 0 0 0 0 0 0 0 0
Thu 14 Apr 11 259 0 0 0 0 0 0 0 0 0
Fri 15 Apr 11 260 0 0 0 0 0 0 0 0 0
Mon 18 Apr 11 261 0 0 0 0 0 0 0 0 0
Tue 19 Apr 11 262 0 0 0 0 0 0 0 0 0
Wed 20 Apr 11 263 0 0 0 0 0 0 0 0 0
Thu 21 Apr 11 264 0 0 0 0 0 0 0 0 0
April 22: anniversary of the fall of Leterme II's
Tue 26 Apr 11 government. April 26: one year without a fully
265 0 0 0 0 0 -1 0 0 -1
empowered government.
Wed 27 Apr 11 266 0 0 0 0 0 0 0 0 -1
Thu 28 Apr 11 267 0 0 0 0 0 0 0 0 0
Fri 29 Apr 11 268 0 0 0 0 0 0 0 0 0
Mon 2 May 11 269 0 0 0 0 0 0 0 0 0
Tue 3 May 11 270 0 0 0 0 0 0 0 0 0
Wed 4 May 11 271 0 0 0 0 0 0 0 0 0
Thu 5 May 11 272 0 0 0 0 0 0 0 0 0
Fri 6 May 11 273 0 0 0 0 0 0 0 0 0
Mon 9 May 11 274 0 0 0 0 0 0 0 0 0
Tue 10 May 11 275 0 0 0 0 0 0 0 0 0
Wed 11 May 11 276 0 0 0 0 0 0 0 0 0
May 12: Wouter Beke asks the King to relieve him from
Thu 12 May 11
his duties.
277 0 0 1 -1 1 -1 -1 1 -1
Fri 13 May 11 278 0 0 0 0 0 0 -1 1 -1
May 16: the King accepts the request of Wouter Beke,
after having consulted all nine parties. Di Rupo is tasked
Mon 16 May 11
to be formateur with the request to take any necessary
279 0 1 0 1 1 1 1 1 1
measures.
Tue 17 May 11 280 0 0 0 0 0 0 1 1 1
Wed 18 May 11 281 0 0 0 0 0 0 0 0 0
Thu 19 May 11 282 0 0 0 0 0 0 0 0 0
Fri 20 May 11 283 0 0 0 0 0 0 0 0 0
May 23: Fitch's outlook changes from “stable” to
Mon 23 May 11
“negative”.
284 1 0 0 0 0 0 0 0 0
Tue 24 May 11 285 0 0 0 0 0 0 0 0 0
Wed 25 May 11 286 0 0 0 0 0 0 0 0 0
Thu 26 May 11 287 0 0 0 0 0 0 0 0 0
Fri 27 May 11 288 0 0 0 0 0 0 0 0 0
Mon 30 May 11 289 0 0 0 0 0 0 0 0 0
Tue 31 May 11 290 0 0 0 0 0 0 0 0 0
Wed 1 Jun 11 291 0 0 0 0 0 0 0 0 0
Thu 2 Jun 11 292 0 0 0 0 0 0 0 0 0
Fri 3 Jun 11 293 0 0 0 0 0 0 0 0 0
Mon 6 Jun 11 294 0 0 0 0 0 0 0 0 0
Tue 7 Jun 11 295 0 0 0 0 0 0 0 0 0
Wed 8 Jun 11 296 0 0 0 0 0 0 0 0 0
Thu 9 Jun 11 297 0 0 0 0 0 0 0 0 0
Fri 10 Jun 11 298 0 0 0 0 0 0 0 0 0
Mon 13 Jun 11 June 13: Belgium voted a year ago. 299 0 0 0 0 0 -1 0 0 -1
Tue 14 Jun 11 300 0 0 0 0 0 0 0 0 -1
Wed 15 Jun 11 301 0 0 0 0 0 0 0 0 0
Thu 16 Jun 11 302 0 0 0 0 0 0 0 0 0
Fri 17 Jun 11 303 0 0 0 0 0 0 0 0 0
Mon 20 Jun 11 304 0 0 0 0 0 0 0 0 0
Tue 21 Jun 11 305 0 0 0 0 0 0 0 0 0
Wed 22 Jun 11 306 0 0 0 0 0 0 0 0 0
Thu 23 Jun 11 307 0 0 0 0 0 0 0 0 0
Fri 24 Jun 11 308 0 0 0 0 0 0 0 0 0
Mon 27 Jun 11 309 0 0 0 0 0 0 0 0 0
Tue 28 Jun 11 310 0 0 0 0 0 0 0 0 0
Wed 29 Jun 11 311 0 0 0 0 0 0 0 0 0
Thu 30 Jun 11 312 0 0 0 0 0 0 0 0 0

79
Fri 1 Jul 11 313 0 0 0 0 0 0 0 0 0
July 4: Di Rupo hands in a 111-page long note, based on
Mon 4 Jul 11 314
his meetings with all nine parties.
0 0 0 0 0 1 0 0 1
Tue 5 Jul 11 315 0 0 0 0 0 0 0 0 1
Wed 6 Jul 11 316 0 0 0 0 0 0 0 0 0
Thu 7 Jul 11 July 7: N-VA refuses to start negotiating from this note. 317 0 0 0 0 0 -1 0 0 -1
July 8: In reply to this refusal, Di Rupo asks the King for
Fri 8 Jul 11
permission to resign.
318 0 0 0 0 0 -1 0 0 -1
Mon 11 Jul 11 319 0 0 0 0 0 0 0 0 -1
Tue 12 Jul 11 320 0 0 0 0 0 0 0 0 0
Wed 13 Jul 11 321 0 0 0 0 0 0 0 0 0
Thu 14 Jul 11 322 0 0 0 0 0 0 0 0 0
Fri 15 Jul 11 323 0 0 0 0 0 0 0 0 0
Mon 18 Jul 11 324 0 0 0 0 0 0 0 0 0
Tue 19 Jul 11 325 0 0 0 0 0 0 0 0 0
Wed 20 Jul 11 326 0 0 0 0 0 0 0 0 0
July 21: The CD&V agrees to start negotiating
Thu 21 Jul 11
without the N-VA.
327 0 1 0 1 1 1 1 1 1
Fri 22 Jul 11 328 0 0 0 0 0 0 1 1 1
Mon 25 Jul 11 329 0 0 0 0 0 0 0 0 0
Tue 26 Jul 11 330 0 0 0 0 0 0 0 0 0
Wed 27 Jul 11 331 0 0 0 0 0 0 0 0 0
Thu 28 Jul 11 332 0 0 0 0 0 0 0 0 0
Fri 29 Jul 11 333 0 0 0 0 0 0 0 0 0
Mon 1 Aug 11 334 0 0 0 0 0 0 0 0 0
Tue 2 Aug 11 335 0 0 0 0 0 0 0 0 0
Wed 3 Aug 11 336 0 0 0 0 0 0 0 0 0
Thu 4 Aug 11 337 0 0 0 0 0 0 0 0 0
Fri 5 Aug 11 338 0 0 0 0 0 0 0 0 0
Mon 8 Aug 11 339 0 0 0 0 0 0 0 0 0
Tue 9 Aug 11 340 0 0 0 0 0 0 0 0 0
Wed 10 Aug 11 341 0 0 0 0 0 0 0 0 0
Thu 11 Aug 11 342 0 0 0 0 0 0 0 0 0
Fri 12 Aug 11 343 0 0 0 0 0 0 0 0 0
Mon 15 Aug 11 344 0 0 0 0 0 0 0 0 0
Tue 16 Aug 11 345 0 0 0 0 0 0 0 0 0
Wed 17 Aug 11 346 0 0 0 0 0 0 0 0 0
Thu 18 Aug 11 347 0 0 0 0 0 0 0 0 0
Fri 19 Aug 11 348 0 0 0 0 0 0 0 0 0
Mon 22 Aug 11 349 0 0 0 0 0 0 0 0 0
Tue 23 Aug 11 350 0 0 0 0 0 0 0 0 0
Wed 24 Aug 11 351 0 0 0 0 0 0 0 0 0
Thu 25 Aug 11 352 0 0 0 0 0 0 0 0 0
Fri 26 Aug 11 353 0 0 0 0 0 0 0 0 0
Mon 29 Aug 11 354 0 0 0 0 0 0 0 0 0
Tue 30 Aug 11 355 0 0 0 0 0 0 0 0 0
Wed 31 Aug 11 356 0 0 0 0 0 0 0 0 0
Thu 1 Sep 11 357 0 0 0 0 0 0 0 0 0
Fri 2 Sep 11 358 0 0 0 0 0 0 0 0 0
Mon 5 Sep 11 359 0 0 0 0 0 0 0 0 0
Tue 6 Sep 11 360 0 0 0 0 0 0 0 0 0
Wed 7 Sep 11 361 0 0 0 0 0 0 0 0 0
Thu 8 Sep 11 362 0 0 0 0 0 0 0 0 0
Fri 9 Sep 11 363 0 0 0 0 0 0 0 0 0
Mon 12 Sep 11 364 0 0 0 0 0 0 0 0 0
Tue 13 Sep 11 365 0 0 0 0 0 0 0 0 0
Wed 14 Sep 11 366 0 0 0 0 0 0 0 0 0
Night 14-15: agreement on the electoral district of
Thu 15 Sep 11
BHV.
367 0 1 0 1 1 1 1 1 1
Fri 16 Sep 11 368 0 0 0 0 0 0 1 1 1
Mon 19 Sep 11 369 0 0 0 0 0 0 0 0 0
Tue 20 Sep 11 370 0 0 0 0 0 0 0 0 0
Wed 21 Sep 11 371 0 0 0 0 0 0 0 0 0
Thu 22 Sep 11 372 0 0 0 0 0 0 0 0 0
Fri 23 Sep 11 373 0 0 0 0 0 0 0 0 0
Mon 26 Sep 11 September 24: agreement on the financing law. 374 0 1 0 1 1 1 1 1 1
Tue 27 Sep 11 375 0 0 0 0 0 0 1 1 1
Wed 28 Sep 11 376 0 0 0 0 0 0 0 0 0
Thu 29 Sep 11 377 0 0 0 0 0 0 0 0 0
Fri 30 Sep 11 378 0 0 0 0 0 0 0 0 0

80
Mon 3 Oct 11 379 0 0 0 0 0 0 0 0 0
Tue 4 Oct 11 380 0 0 0 0 0 0 0 0 0
October 5: agreement on the judiciary district of
Wed 5 Oct 11
BHV.
381 0 1 0 1 1 1 1 1 1
Thu 6 Oct 11 382 0 0 0 0 0 0 1 1 1
October 7: Moody's decision to place Belgium's Aa1
Fri 7 Oct 11
rating on review for possible downgrade.
383 1 0 0 0 0 0 0 0 0
Mon 10 Oct 11 October 8: global agreement. 384 0 1 0 1 1 1 1 1 1
Tue 11 Oct 11 October 11: the sixth reform of the state is done. 385 0 0 0 0 0 1 1 1 1
Wed 12 Oct 11 386 0 0 0 0 0 0 0 0 1
October 13: Ecolo and Groen! are removed from the
Thu 13 Oct 11 387 0 0 0 0 0 1 0 0 1
negotiating table. Six parties are left to negotiate.
Fri 14 Oct 11 388 0 0 0 0 0 0 0 0 1
Mon 17 Oct 11 389 0 0 0 0 0 0 0 0 0
Tue 18 Oct 11 390 0 0 0 0 0 0 0 0 0
Wed 19 Oct 11 391 0 0 0 0 0 0 0 0 0
Thu 20 Oct 11 392 0 0 0 0 0 0 0 0 0
Fri 21 Oct 11 393 0 0 0 0 0 0 0 0 0
Mon 24 Oct 11 394 0 0 0 0 0 0 0 0 0
Tue 25 Oct 11 395 0 0 0 0 0 0 0 0 0
Wed 26 Oct 11 396 0 0 0 0 0 0 0 0 0
Thu 27 Oct 11 397 0 0 0 0 0 0 0 0 0
Fri 28 Oct 11 398 0 0 0 0 0 0 0 0 0
Mon 31 Oct 11 399 0 0 0 0 0 0 0 0 0
Tue 1 Nov 11 400 0 0 0 0 0 0 0 0 0
Wed 2 Nov 11 401 0 0 0 0 0 0 0 0 0
Thu 3 Nov 11 402 0 0 0 0 0 0 0 0 0
Fri 4 Nov 11 403 0 0 0 0 0 0 0 0 0
Mon 7 Nov 11 404 0 0 0 0 0 0 0 0 0
Tue 8 Nov 11 405 0 0 0 0 0 0 0 0 0
Wed 9 Nov 11 406 0 0 0 0 0 0 0 0 0
Thu 10 Nov 11 407 0 0 0 0 0 0 0 0 0
Fri 11 Nov 11 408 0 0 0 0 0 0 0 0 0
Mon 14 Nov 11 409 0 0 0 0 0 0 0 0 0
Tue 15 Nov 11 410 0 0 0 0 0 0 0 0 0
Wed 16 Nov 11 411 0 0 0 0 0 0 0 0 0
Thu 17 Nov 11 412 0 0 0 0 0 0 0 0 0
Fri 18 Nov 11 413 0 0 0 0 0 0 0 0 0
Mon 21 Nov 11 414 0 0 0 0 0 0 0 0 0
Tue 22 Nov 11 415 0 0 0 0 0 0 0 0 0
Wed 23 Nov 11 416 0 0 0 0 0 0 0 0 0
Thu 24 Nov 11 417 0 0 0 0 0 0 0 0 0
Fri 25 Nov 11 November 25: S&P downgrade Belgium. 418 1 0 0 0 0 0 0 0 0
Mon 28 Nov 11 November 26: political agreement on the budget. 419 0 1 0 1 1 1 1 1 1
Tue 29 Nov 11 420 0 0 0 0 0 0 1 1 1
November 30: conclusion of the governmental
Wed 30 Nov 11
agreement.
421 0 1 0 1 1 1 1 1 1
Thu 1 Dec 11 422 0 0 0 0 0 0 1 1 1
Fri 2 Dec 11 423 0 0 0 0 0 0 0 0 0
Mon 5 Dec 11 424 0 0 0 0 0 0 0 0 0
Tue 6 Dec 11 December 6: Elio Di Rupo takes the oath.
425 0 1 0 1 1 1 1 1 1
Wed 7 Dec 11 426 0 0 0 0 0 0 1 1 1
Thu 8 Dec 11 427 0 0 0 0 0 0 0 0 0
Fri 9 Dec 11 428 0 0 0 0 0 0 0 0 0
Mon 12 Dec 11 429 0 0 0 0 0 0 0 0 0
Tue 13 Dec 11 430 0 0 0 0 0 0 0 0 0

Figure A.1: Constituency Brussels-Halle-Vilvoorde.
Source: La Libre Belgique (in de Coorebyter (2011)).

82
Figure A.2: Frequency distribution of the daily returns of the BEL20
over the period: April 19, 2010 - December 13, 2011.
Source: own computation, based on Datastream data.
The returns on financial assets usually do not follow a normal distribution. Akgiray
and Booth (1987) investigated the topic in more detail 107 , reporting that stock returns,
as other financial assets, are leptokurtic (higher peaks around the mean and larger tails).
Similar results can be found in Mandelbrot (1963) and Fama (1965). This kurtosis
problem can be partially solved by ARCH-like processes, as Hsieh (1991) develops, which
explains why these models represent better financial time series.
107 Cited in Cheung and Lai (1993).

83
Table A.1: Final results of the main model, corrected for endogeneity and
heteroskedasticity: 2SLS regression, using a generated instrumental variable for
the political indicator.
Variable Coefficient (Std. Err.)
Main political indicator, free of endogeneity -0.240 ∗ (0.098)
Euro Stoxx 50 return 0.812 ∗∗ (0.018)
Euro Stoxx 50 return 2 0.007 (0.005)
Rating dummy -0.515 ∗ (0.232)
Government bonds yield -0.009 ∗ (0.004)
Predictability indicator 0.058 (0.039)
Euribor change 0.084 ∗ (0.040)
Money growth -0.030 † (0.016)
Intercept -0.128 ∗ (0.058)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%
Table A.2: Final results of the main model, corrected only for heteroskedasticity:
basic OLS regression, robust to an arbitrary form of
heteroskedasticity.
Variable Coefficient (Std. Err.)
Main political indicator, endogeneous -0.163 † (0.095)
Euro Stoxx 50 return 0.810 ∗∗ (0.018)
Euro Stoxx 50 return 2 0.007 (0.005)
Rating dummy -0.516 ∗ (0.231)
Government bonds yield -0.008 † (0.004)
Predictability indicator 0.057 (0.039)
Euribor change 0.084 ∗ (0.040)
Money growth -0.030 † (0.016)
Intercept -0.126 ∗ (0.058)
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%
The absence of correction for endogeneity of the political indicator substantially
modifies its significance, as the comparison between tables A.1 and A.2 shows. The
coefficients, standard errors and significance of the other regressors remain almost
unchanged.

Bibliography
Books
Alesina, A., Roubini, N. and Cohen, G. (1997), Political Cycles and the Macroeconomy, MIT
Press, Cambridge, Massachusetts.
Angrist, J. and Pischke, J. (2009), Mostly Harmless Econometrics, Princeton University Press,
Princeton, New Jersey.
Benedikter, R. (2011), Social Banking and Social Finance: Answer to the Economic Crisis,
Springer, New York.
Bernhard, W. and Leblang, D. (2006), Democratic Processes and Financial Markets - Pricing
Politics, Cambridge University Press, New York.
Box, G. and Jenkins, G. (1970), Time Series Analysis: Forecasting and Control, Holden-Day,
San Francisco, California.
Buyst, E., Maes, I., Pluym, W. and Danneel, M. (2005), The Bank, the Franc and the Euro, A
History of the National Bank of Belgium, Lannoo, Tielt.
Campbell, J., Lo, A. and MacKinlay, A. (1997), The Econometrics of Financial Markets,
Princeton University Press, Princeton, New Jersey.
Cootner, P. (1964), The Random Character of Stock Market Prices, M.I.T. Press, Boston,
Massachusetts.
Delwit, P. (2003), Composition, Décomposition et Recomposition du Paysage Politique en
Belgique, Labor, Bruxelles.
Delwit, P., De Waele, J. and Magnette, P. (1999), Gouverner la Belgique, Presses universitaires
de France, Paris.
Deschouwer, K. (2009), The Politics of Belgium: Governing a Divided Society, Palgrave
Macmillan, Basingstoke (1 st edition).
Destrée, J. (1963), La Lettre au Roi, Edition de la Wallonie libre, Gembloux.
Jaumain, S. (1997), La Réforme de l’État... et Après?, Éditions de l’Université de Bruxelles,
Brussels.
Mabille, X. (2000), Histoire Politique de la Belgique, Facteurs et Acteurs de Changement, CRISP,
Brussels.
Maes, I. (2010), A Century of Macroeconomic and Monetary Thought at the National Bank of
Belgium, National Bank of Belgium, Brussels.
National Bank of Belgium, (2011), Rapport 2011: Stabilité Financière et Contrôle Prudentiel,
National Bank of Belgium, Brussels.
84

85
Quevit, M. (2010), Flandre-Wallonie, Quelle Solidarité?, Couleur livres, Brussels.
Savage, R. (2004), Histoire Inédite de l’Économie en Belgique, Couleur livres, Charleroi.
Wooldridge, J. (2009), Introductory Econometrics: A Modern Approach, South-Western
Publishing, Canada (4 th edition).
Wright, P. (1928), The Tariff on Animal and Vegetable Oils, MacMillan, New York.

86
Articles in journals
Abadie, A. and Gardeazabal, J. (2003), « The Economic Costs of Conflict: A Case Study of the
Basque Country», The American Economic Review, 93(1), 113–132.
Acharya, S. (1993), « Value of Latent Information: Alternative Event Study Methods», The
Journal of Finance, 48(1), 363–385.
Akgiray, V. and Booth, G. (1987), « Compound Distribution Models of Stock Returns: An
Empirical Comparison», Journal of Financial Research, 10(3), 269–280.
Aktas, H. and Oncu, S. (2006), « The Stock Market Reaction to Extreme Events: The Evidence
from Turkey», International Research Journal of Finance and Economics, (6), 78–85.
Ali, N., Nassir, A., Hassan, T. and Abidin, S. (2010), « Short Run Stock Overreaction: Evidence
from Bursa Malaysia», Journal of Economics and Management, 4(2), 319–333.
Ashley, J. (1962), « Stock Prices and Changes in Earnings and Dividends: Some Empirical
Results», Journal of Political Economy, 70(1), 82–85.
Ball, C. and Torous, W. (1988), « Investigating Security-Price Performance in the Presence of
Event-Date Uncertainty», Journal of Financial Economics, 22, 123–153.
Barclay, M. and Litzenberger, R. (1988), « Announcement Effect of New Equity Issues and the
Use of Intraday Price Data», Journal of Financial Economics, 21, 71–100.
Barker, C. (1958), « Evaluation of Stock Dividends», Harvard Business Review, 36, 101–106.
Barro, B. (1977), « Unanticipated Money Growth and Unemployment in the United States»,
American Economic Review, 67(2), 101–115.
Barro, B. (1979), « Unanticipated Money Growth and Unemployment in the United States:
Reply», American Economic Review, 69(5), 101–115.
Beck, N. (1992), « Comparing Dynamic Specifications: the Case of Presidential Approval»,
Political Analysis, 3, 27–50.
Beck, N. (1993), « The Methodology of Cointegration», Political Analysis, 4, 237–248.
Benton, A. (2008), « Do Investors Assess the Credibility of Campaign Commitments? : The
Case of Mexico’s 2006 Presidential Race», Political Research Quarterly, 61(3), 403–418.
Bernanke, B. and Kuttner, K. (2005), « What Explains the Stock Market’s Reaction to Federal
Reserve Policy», The Journal of Finance, 60(3), 1221–1257.
Berry, M., Gallinger, G. and Henderson, G. (1990), « Using Daily Stock Returns in Event Studies
and the Choice of Parametric versus Nonparametric Test Statistics», Quarterly Journal of
Business and Economics, 29(1), 70–85.
Bilson, C., Brailsford, T. and Hooper, V. (2002), « The Explanatory Power of Political Risk in
Emerging Market», International Review of Financial Analysis, 11(1), 1–27.
Binder, J. (1998), « The event study methodology since 1969», Review of Quantitative Finance
and Accounting, 11, 111–137.

87
Blanchard, O. (1981), « Output, the stock market and interest rates», The American Economic
Review, 71(1), 132–143.
Bollerslev, T. (1986), « Generalized Autoregressive Conditional Heteroskedasticity», Journal of
Econometrics, 31, 307–327.
Boudoukh, J., Richardson, M. and Whitelaw, R. (1994), « A Tale of Three Schools: Insights
on Autocorrelation of Short Term Stock Returns», The Review of Financial Studies, 7(3),
539–573.
Brown, S. and Warner, J. (1980), « Measuring Security Price Performance», Journal of Financial
Economics, 8, 205–258.
Brown, S. and Warner, J. (1985), « Using Daily Stock Returns: The Case of Event Studies»,
Journal of Financial Economics, 14, 3–31.
Cammack, P. (2009), « All Power to Global Capital!», Papers in the Politics of Global Competitiveness,
July(10), 1–17.
Chan, Y. and Wei, J. (1996), « Political Risk and Stock Price Volatility: The Case of Hong
Kong», Pacific-Basin Finance Journal, 4(2-3), 259–275.
Chen, D., Bin, F. and C., Chen (2005), « The Impacts of Political Events on Foreign Institutional
Investors and Stock Returns: Emerging Market Evidence from Taiwan», International Journal
of Business, 10(2), 165–188.
Cheung, Y. and Lai, K. (1993), « Do Gold Market Returns Have Long Memory», The Financial
Review, 26(2), 181–202.
de Coorebyter, V. (2011), « Principaux Partis et Dernières Élections en Belgique», Pouvoirs,
136, 5–8.
Dickey, D. and Fuller, W. (1979), « Distribution of the Estimators for Autoregressive Time
Series With a Unit Root», Journal of the American Statistical Association, 74, 427–431.
Dolley, J. (1933), « Characteristics and Procedures of Common Stock Split-Ups», Harvard
Business Review, 11, 316–327.
Doornik, J. and Ooms, M. (2008), « Multimodality in GARCH Regression Models», International
Journal of Forecasting, 24, 432–448.
Durbin, J. and Watson, G. (1950), « Testing for Serial Correlation in Least Squares Regression:
I», Biometrika, 37, 409–428.
Durbin, J. and Watson, G. (1951), « Testing for Serial Correlation in Least Squares Regression.
II», Biometrika, 38, 159–177.
Durr, R. (1993), « An Essay on Cointegration and Error Correction Models», Political Analysis,
4, 185–228.
Elliott, G., Rothenberg, T. and Stock, J. (1996), « Efficient Tests for an Autoregressive Unit
Root», Econometrica, 64(4), 813–836.
Engle, R. (1982), « Autoregressive Conditional Heteroscedasticity with Estimates of the Variance
of United Kingdom Inflation», Econometrica, 50(4), 987–1008.

88
Engle, R. and Granger, C. (1987), « Co-integration and Error Correction: Representation,
Estimation and Testing», Econometrica, 55(2), 251–276.
Engle, R. and Viktor, K. (1993), « Measuring and Testing the Impact of News on Volatility»,
The Journal of Finance, 48(5), 1749–1778.
Erk, J. (2005), « From Vlaams Blok to Vlaams Belang: The Belgian Far-Right Renames Itself»,
West European Politics, 28(3), 493–502.
Fama, E. (1965), « The Behavior of Stock-Market Prices», The Journal of Business, 38(1),
34–105.
Fama, E. (1970), « Efficient Capital Markets: A Review of Theory and Empirical Work», The
Journal of Finance, 25(2), 383–417.
Fama, E. (1990), « Stock Returns, Expected Returns, and Real Activity», The Journal of
Finance, 45(4), 1089–1108.
Fama, E. and French, K. (2004), « The Capital Asset Pricing Model: Theory and Evidence»,
Journal of Economic Perspectives, 18(3), 25–46.
Fama, E., Fischer, L., Jensen, M. and Roll, R. (1969), « The Adjustment of Stock Prices to New
Information», International Economic Review, 10, 1–27.
Feldstein, M. (1980), « Inflation and the Stock Market», American Economic Review, 70,
839–847.
Forbes, K. and Rigobon, R. (2002), « No Contagion, Only Interdependence: Measuring Stock
Market Co-movements», The Journal of Finance, 57(5), 2223–2261.
Freeman, J., Hays, C. and Stix, H. (2000), « Democracy and Markets: The Case of Exchange
Rates», American Journal of Political Science, 44(3), 449–468.
Gemmill, G. and Saflekos, A. (2000), « How Useful are Implied Distributions? Evidence From
Stock Index Options?», The Journal of Derivatives, Spring 2000, 1–17.
Glascock, J., Davidson, W. and Henderson, G. (1987), « Announcement Effects of Moody’s
Bond Rating Changes on Equity Returns», Quarterly Journal of Business and Economics, 26
(3), 67–78.
Granger, C. (1981), « Some Properties of Time Series Data and their Use in Econometric Model
Specification», Journal of Econometrics, 16, 121–130.
Gregory, A. and Reeves, J. (2010), « Estimation and Inference in ARCH Models in the Presence
of Outliers», Journal of Financial Econometrics, 8(4), 547–549.
Hedges, L. (1987), « How Hard Is Hard Science, How Soft Is Soft Science?»,
Psychologist, 42(2), 443–455.
American
Henderson, G. (1990), « Problems and Solutions in Conducting Event Studies», Journal of Risk
and Insurance, 57, 282–306.
Herron, M. (2000), « Estimating the Economic Impact of Political Party Competition in the
1992 British Election», American Journal of Political Science, 44(3), 326–337.

89
Hsieh, D. (1991), « Chaos and Nonlinear Dynamics: Application to Financial Markets», The
Journal of Finance, 46(5), 1839–1877.
Huang, R. (1985), « Common Stock Returns and Presidential Elections», Financial Analysts
Journal, 41(2), 58–61.
Jensen, N. and Schmith, S. (2005), « Market Responses to Politics: The Rise of Lula and the
Decline of the Brazilian Stock Market», Comparative Political Studies, 38(10), 1245–1270.
Jones, S. and Banning, K. (2009), « US Elections and Monthly Stock Market Returns», Journal
of Economics and Finance, 33(3), 273–287.
Junqué de Fortuny, E., De Smedt, T., D., Martens and W., Daelemans (2012), « Media Coverage
in Times of Political Crisis: A Text Mining Approach», Expert Systems with Applications, 39
(14), 11616–11622.
Kaminsky, G. and Schmukler, S. (2002), « Emerging Markets Instability: Do Sovereign Ratings
Affect Country Risk and Stock Returns?», World Bank Economic Review, 16(2), 171–195.
Karafiath, I. (1988), « Using Dummy Variables in the Event Methodology», Financial Review,
23(3), 351–357.
King, R. and Levine, R. (1993), « Finance and Growth: Schumpeter Might Be Right», Journal
of Economics, 108(3), 717–737.
Koutmos, G. and Theodossiou, P. (1994), « Time-Series Properties and Predictability of Greek
Exchange Rates», Managerial and Decision Economics, 15, 159–167.
Leblang, D. and Mukherjee, B. (2004), « Presidential Elections and the Stock Market: Comparing
Markov-Switching and (FIE)GARCH Models of Stock Volatility», Political Analysis, 12(3),
1–56.
Lee, H. and Brorsen, W. (1997), « A Non-Nested Test of GARCH vs. EGARCH Models»,
Applied Economics Letters, 4(12), 765–768.
Lintner, J. (1965), « The Valuation of Risk Assets and the Selection of Risky Investments in
Stock Portfolios and Capital Budgets», Review of Economics and Statistics, 47(1), 13–37.
Lobo, B. and Tufte, D. (1998), « Exchange Rate Volatility: Does Politics Matter?», Journal of
Macroeconomics, 20(2), 351–365.
MacKinlay, C. (1997), « Event Studies in Economics and Finance»,
Literature, 35(1), 13–39.
Journal of Economic
Mandelbrot, B. (1963), « The Variation of Certain Speculative Prices», The Journal of Business,
36(4), 394–419.
Mnookin, R. (2007), « Ethnic Conflicts: Flemings and Walloons, Palestinians and Israelis»,
Daedalus, 136(1), 103–119.
Murphy, K. and Topel, R. (1985), « Estimation and Inference in Two-Step Econometric Models»,
Journal of Business and Econometric Statistics, 3(4), 88–97.
Myers, J. and Bakay, A. (1948), « Influence of Stock Split-ups on Market Prices», Harvard
Business Review, 26, 251–265.

90
Nelson, D. (1991), « Conditional Heteroskedasticity in Asset Returns: A New Approach»,
Econometrica, 59(2), 347–370.
Niederhoffer, V. (1971), « The Analysis of World Events and Stock Prices», The Journal of
Business, 44(2), 193–219.
Oxley, L. and McAleer, M. (1993), « Econometric Issues in Macroeconomic Models with
Generated Regressors», Journal of Economic Surveys, 1(1), 1–40.
Pagan, A. (1984), « Econometric Issues in the Analysis of Regressions with Generated Regressors»,
International Economic Review, 25(1), 221–247.
Pantzalis, C., Stangeland, D. and Turle, H. (2000), « Political Elections and the Resolution of
Uncertainty: The International Evidence», Journal of Banking & Finance, 24, 1575–1604.
Parkinson, M. (1980), « The Extreme Value Method for Estimating the Variance of the Rate of
Return», The Journal of Business, 53(1), 61–65.
Pynnönen, S. (2005), « On Regression Based Event Study», Acta Wasensia, 143, 327–354.
Reiersøl, O. (1941), « Confluence Analysis By Means of Lag Moments and Other Methods of
Confluence Analysis», Econometrica, 9(1), 1–24.
Roll, R. (1977), « A Critique of the Asset Pricing Theory’s Tests: Part I: On Past and Potential
Testability of the Theory», The Journal of Financial Economics, 4(2), 129–176.
Roll, R. and Ross, S. (1980), « An Empirical Investigation of the Arbitrage Pricing Theory»,
The Journal of Finance, 35(5), 1073–1103.
Romero-Meza, R., Bonilla, C. and M., Hinich (2007), « Nonlinear Event Detection in the Chilean
Stock Market», Applied Economics Letters, 14, 987–991.
Ross, S. (1976), « The Arbitrage Theory of Capital Asset Pricing», Journal of Economic Theory,
13, 341–360.
Rozeff, M. (1974), « Money and Stock Prices: Market Efficiency and the Lag in Effect of
Monetary Policy», Journal of Financial Economics, 1(3), 245–302.
Samuelson, P. (1965), « Proof that Properly Anticipated Prices Fluctuate Randomly», Industrial
Management Review, 6, 41–49.
Savigny, H. (2002), « Public Opinion, Political Communication and the Internet», Politics, 22
(1), 1–8.
Schwert, G. (1989), « Margin Requirements and Stock Volatility», Journal of Financial Services
Research, 3, 153–164.
Schwert, G. (1990), « Stock Return and Real Activity: a Century of Evidence», The Journal of
Finance, 45(4), 1237–1257.
Shanken, J. (1982), « The Arbitrage Pricing Theory: is it Testable?», The Journal of Finance,
37(5), 1129–1140.
Sharpe, W. (1964), « Capital Asset Prices: A Theory of Market Equilibrium under Conditions
of Risk», The Journal of Finance, 19(3), 425–442.

91
Theodossiou, P. (1994), « The Stochastic Properties of Major Canadian Exchange Rates»,
Financial Review, 29(2), 193–221.
Thorbecke, W. (1997), « On Stock Market Returns and Monetary Policy», The Journal of
Finance, 52(2), 635–654.
Ultsch, A. (2009), « Is Log Ratio a Good Value for Measuring Return in Stock Investments», In:
A. Fink, B. Lausen, W. Seidel & A. Ultsch (Eds.), Advances in Data Analysis, Data Handling
and Business Intelligence, 505–511. Springer.
Wells, W. (2004), « A Beginner’s Guide to Event Studies», Journal of Insurance Regulation, 22
(4), 61–70.
Wooldridge, J. (2001), « Applications of Generalized Method of Moments Estimation», Journal
of Economic Perspectives, 15(4), 87–100.
Wu, G. and Xiao, Z. (2002), « A Generalized Partially Linear Model of Asymmetric Volatility»,
Journal of Empirical Finance, 9, 287–319.
Zach, T. (2003), « Political Events and the Stock Market - Evidence from Israel», International
Journal of Business, 8(3), 243–266.

92
Online newspapers
Australian Inspiration, Philosophy & Ethics, 2010. Available at: http://www.
australianinspiration.com.au/Quotes/Wisdom/Philosophy.aspx. Consulted on June 8,
2012.
BBC News, Belgium swears in new government headed by Elio Di Rupo, December 6, 2011.
Available at: http://www.bbc.co.uk/news/world-europe-16042750. Consulted on August
7, 2012.
BBC News, Belgium fails to break political deadlock, October 18, 2010. Available at: http:
//www.bbc.co.uk/news/world-europe-11568174. Consulted on August 7, 2012.
De Morgen, Belgisch wereldrecord regeringsvorming erkend door Guinness
World Records, April 19, 2011. Available at: http://www.
demorgen.be/dm/nl/5036/Wetstraat/article/detail/1252745/2011/04/19/
Belgisch-wereldrecord-regeringsvorming-erkend-door-Guinness-World-Records.
dhtml. Consulted on August 8, 2012.
De Morgen, Chronologie van 207 dagen politieke stilstand, August 29, 2010. Available at:
http://www.demorgen.be/dm/nl/6956/Verkiezingen-2010/article/detail/1203964/
2011/01/06/Chronologie-van-207-dagen-politieke-stilstand.dhtml. Consulted on
August 7, 2012.
De Morgen, Regeringsonderhandelingen: een calvarie van 545 dagen, December 10, 2011. Available
at: http://www.demorgen.be/dm/nl/5036/Wetstraat/article/detail/1361080/
2011/12/10/Regeringsonderhandelingen-een-calvarie-van-545-dagen.dhtml. Consulted
on August 7, 2012.
De Morgen, Reynders krijgt twee weken, February 2, 2011. Available at: http:
//www.demorgen.be/dm/nl/5036/Wetstraat/article/detail/1216935/2011/02/02/
Reynders-krijgt-twee-weken.dhtml. Consulted on August 8, 2012.
De Morgen, Vande Lanotte gooit de handdoek, January 26, 2011. Available at:
http://www.demorgen.be/dm/nl/5036/Wetstraat/article/detail/1213689/2011/
01/26/Vande-Lanotte-gooit-de-handdoek.dhtml. Consulted on August 8, 2012.
De Morgen, "Nota aanpassen kan ik niet maken tegenover de rest", July 19, 2011. Available
at: http://www.demorgen.be/dm/nl/3625/De-Formatie/article/detail/1294496/
2011/07/19/CD-V-en-zeven-partijen-vergaderen-vanmiddag.dhtml. Consulted on August
8, 2012.
De Morgen, N-VA maakt brandhout van nota, July 7, 2011. Available at: http:
//www.demorgen.be/dm/nl/989/Binnenland/article/detail/1289137/2011/07/07/
N-VA-maakt-brandhout-van-nota.dhtml. Consulted on August 8, 2012.
De Standaard, Koning aanvaardt ontslag regering, April 26, 2011. Available at: http://www.
standaard.be/artikel/detail.aspx?artikelid=DMF20100426_036. Consulted on August
8, 2012.
De Standaard, BHV van de baan: Franstaligen luiden alarmbel, April 29, 2010. Available at: http:
//www.standaard.be/artikel/detail.aspx?artikelid=DMF20100429_058. Consulted on
August 7, 2012.

93
De Standaard, CD&V: ’Beke heeft geen deadline’, March 5, 2011. Available at: http://www.
standaard.be/artikel/detail.aspx?artikelid=DMF20110305_011. Consulted on August
8, 2012.
De Standaard, Wouter Beke wil landing inzetten, May 12, 2011. Available at: http://www.
standaard.be/artikel/detail.aspx?artikelid=5U3A23J1. Consulted on August 8, 2012.
De Standaard, Di Rupo wordt formateur, May 16, 2011. Available at: http://www.standaard.
be/artikel/detail.aspx?artikelid=DMF20110516_168. Consulted on August 8, 2012.
De Standaard, LETTERLIJK. Reactie van Groen! op keuze van formateur, October 13, 2011.
Available at: http://www.standaard.be/artikel/detail.aspx?artikelid=DMF20111013_
150. Consulted on August 8, 2012.
De Standaard, Pieters en Flahaut niet langer bemiddelaar, October 5, 2010. Available at: http:
//www.standaard.be/artikel/detail.aspx?artikelid=DMF20101005_012. Consulted on
August 7, 2012.
De Standaard, Akkoord over gerechtelijke splitsing BHV is rond, October 5, 2011. Available
at: http://www.standaard.be/artikel/detail.aspx?artikelid=DMF20111005_002. Consulted
on August 8, 2012.
De Tijd, Bart De Wever wordt informateur, June 18, 2011. Available at: http://www.tijd.be/
nieuws/politiek_economie_belgie/Bart_De_Wever_wordt_informateur.8929079-3136.
art. Consulted on August 8, 2012.
De Tijd, België toch geen wereldrecordhouder, March 30, 2011. Available at:
http://www.tijd.be/nieuws/politiek_economie_belgie/Belgie_toch_geen_
wereldrecordhouder.9039246-3136.art. Consulted on August 8, 2012.
De Tijd, BIO Johan Vande Lanotte, October 21, 2010. Available at: http://www.tijd.be/
nieuws/politiek_economie_belgie/BIO_Johan_Vande_Lanotte.8977268-3136.art. Consulted
on August 8, 2012.
De Tijd, Akkoord over BHV, September 15, 2011. Available at: http://www.tijd.be/nieuws/
politiek_economie_belgie/Akkoord_over_BHV.9104022-3136.art. Consulted on August
8, 2012.
De Tijd, De financieringswet van A tot Z, September 26, 2011. Available at:
http://www.tijd.be/nieuws/politiek_economie_belgie/De_financieringswet_van_
A_tot_Z.9108160-3136.art. Consulted on August 8, 2012.
La Libre, La crise politique la plus longue, December 23, 2010. Available
at: http://www.lalibre.be/actu/politique-belge/article/631652/
la-crise-politique-la-plus-longue.html. Consulted on August 7, 2012.
La Libre, Le roi a chargé Didier Reynders d’une mission de 2 semaines, February 2,
2011. Available at: http://www.lalibre.be/actu/politique-belge/article/640137/
le-roi-a-charge-didier-reynders-d-une-mission-de-2-semaines.html. Consulted on
August 7, 2012.
La Libre, Vande Lanotte ne veut plus de sa mission, January 6, 2011. Available
at: http://www.lalibre.be/actu/politique-belge/article/634143/
vande-lanotte-ne-veut-plus-de-sa-mission.html. Consulted on August 8, 2012.

94
La Libre, Di Rupo entre dans le champ de mines, May 17, 2011. Available
at: http://www.lalibre.be/actu/politique-belge/article/661203/
di-rupo-entre-dans-le-champ-de-mines.html. Consulted on August 8, 2012.
La Libre, Flahaut et Pieters désignés médiateurs, Di Rupo déchargé, September 4,
2010. Available at: http://www.lalibre.be/actu/politique-belge/article/607140/
flahaut-et-pieters-designes-mediateurs-di-rupo-decharge.html. Consulted on August
7, 2012.
Le Monde, La Belgique est morte le 22 avril 2010, April 23, 2010. Available
at: http://www.lemonde.fr/europe/article_interactif/2010/04/23/
la-belgique-est-morte-le-22-avril-2010_1341870_3214.html. Consulted on November
6, 2011.
Le Monde, Crise belge : d’un médiateur à l’autre, March 2, 2011. Available at: http://www.
lemonde.fr/europe/portfolio/2011/03/02/crise-belge-d-un-mediateur-a-l-autre_
1487504_3214.html. Consulted on November 6, 2011.
Le Soir, Jean-Luc Dehaene refile BHV à Yves Leterme, April 20, 2010.
Available at: http://blog.lesoir.be/jour-apres-jour/2010/04/20/
la-proposition-de-dehaene-pour-les-habitants-de-la-peripherie.../. Consulted
on February 2, 2012.
Le Soir, Crise sur BHV : la chronologie de la journée, April 22, 2010.
Available at: http://www.lesoir.be/actualite/belgique/2010-04-22/
crise-sur-bhv-la-chronologie-de-la-journee-765871.php. Consulted on April
27, 2012.
Le Soir, Voici l’accord de gouvernement en intégralité, December 2, 2011. Available
at: http://www.lesoir.be/actualite/belgique/elections_2010/2011-12-02/
voici-l-accord-de-gouvernement-en-integralite-881093.php. Consulted on April 27,
2012.
Le Soir, Le Roi a nommé Elio Di Rupo Premier ministre, December 5, 2011.
Available at: http://www.lesoir.be/actualite/belgique/elections_2010/2011-12-05/
le-roi-a-nomme-elio-di-rupo-premier-ministre-881543.php. Consulted on April 27,
2012.
Le Soir, Le CD&V lâche la N-VA et accepte de négocier, July 22, 2011. Available at:
http://archives.lesoir.be/le-cd-amp-v-lache-la-n-va-et-accepte-de-negocier_
t-20110722-01HAHX.html. Consulted on August 8, 2012.
Le Soir, Di Rupo nommé "préformateur", July 9, 2010. Available at:
http://www.lesoir.be/actualite/belgique/elections_2010/2010-07-08/
di-rupo-nomme-preformateur-780791.php. Consulted on August 7, 2012.
Le Soir, Wouter Beke, jeune président peu connu, March 2, 2011. Available
at: http://www.lesoir.be/actualite/belgique/elections_2010/2011-03-02/
wouter-beke-jeune-president-peu-connu-825532.php. Consulted on August 8, 2012.
Le Soir, En route vers un gouvernement, November 26, 2011. Available at:
http://www.lesoir.be/actualite/belgique/elections_2010/2011-11-26/
en-route-vers-un-gouvernement-879570.php. Consulted on June 2, 2012.

95
Le Soir, La réforme de l’Etat, qu’est-ce que ca change ?, October 11, 2011.
Available at: http://www.lesoir.be/actualite/belgique/elections_2010/2011-10-11/
la-reforme-de-l-etat-qu-est-ce-que-ca-change-869012.php. Consulted on June 2,
2012.
Le Soir, De Wever nommé "clarificateur", October 9, 2010. Available at:
http://www.lesoir.be/actualite/belgique/elections_2010/2010-10-08/
de-wever-nomme-clarificateur-797291.php. Consulted on April 27, 2012.
L’Echo, Record d’Europe de la crise politique, January 7, 2011. Available at:
http://www.lecho.be/actualite/economie_politique_belgique/Record_d_Europe_
de_la_crise_politique.9007132-3154.art. Consulted on August 8, 2012.
L’Echo, 8 juillet 2010: le bonheur est dans le pré, May 18, 2011. Available at: http://blogs.
lecho.be/humeurs/2011/05/8-juillet-2010-le-bonheur-est-dans-le-pré-.html.
Consulted on April 27, 2012.
L’Echo, S&P menace toujours la Belgique, May 7, 2011. Available at: http:
//www.lecho.be/actualite/economie_politique_belgique/S_P_menace_toujours_
la_Belgique.9055204-3154.art. Consulted on November 20, 2011.
L’Echo, La Belgique au pied du mur: sa note est dégradée par S&P, November 26, 2011. Available
at: http://www.lecho.be/actualite/economie_politique_belgique/La_Belgique_
au_pied_du_mur_sa_note_est_degradee_par_S_P.9131225-3154.art. Consulted on July
11, 2012.
Moody’s, Moody’s downgrades Belgium’s credit ratings to Aa3, negative outlook,
December 16, 2011. Available at: http://www.moodys.com/research/
Moodys-downgrades-Belgiums-credit-ratings-to-Aa3-negative-outlook--PR_
233667. Consulted on June 3, 2012.
Moody’s, Key drivers of Moody’s decision to place Belgium’s Aa1 rating on review for
possible downgrade, October 7, 2011. Available at: http://www.moodys.com/research/
Key-Drivers-of-Moodys-Decision-to-Place-Belgiums-Aa1-Rating--PBC_136680. Consulted
on June 3, 2012.
Reuters, Belgium faces threat of S&P downgrade, December 14, 2010. Available at: http://www.
reuters.com/article/2010/12/14/us-belgium-sp-downgrade-idUSTRE6BD1UB20101214.
Consulted on June 3, 2012.
Service Public Fédéral Belge, Nouveau gouvernement Leterme II, November 25, 2009.
Available at: http://www.belgium.be/fr/actualites/2009/news_nieuwe_federale_
regering_leterme_2.jsp. Consulted on June 8, 2012.
The Guardian, Belgium’s five-party coalition government collapses, April
26, 2010. Available at: http://www.guardian.co.uk/world/2010/apr/26/
belgian-government-collapses-leterme-resigns. Consulted on August 7, 2012.
The New York Times, Belgium, December 14, 2011. Available at: http://topics.nytimes.com/
top/news/international/countriesandterritories/belgium/index.html. Consulted on
August 8, 2012.

Washington Times, At long last, Belgium nearing a government, November
30, 2011. Available at: http://www.washingtontimes.com/news/2011/nov/30/
at-long-last-belgium-nearing-a-government/?page=all. Consulted on August 7,
2012.
96

97
Databases
Datastream, Thomson Financial, 2012.
European Commission, Eurostat, 2012.
National Bank of Belgium, Belgostat, 2012.

98
Others
Aktas, N., de Bodt, E and Cousin, J. (2007), « Assessing the Power and Size of the Event Study
Method Through the Decades », 1–23, working paper.
Aquino, R. (2004), « News, Noise and Stock Price Movements», Philippine Management Review,
(0404), discussion paper.
Bernanke, B. and Gertler, M. (2000), « Monetary Policy and Asset Price Volatility», National
Bureau of Economic Research, (7559), 1–74, working paper.
Cohen, G. (1993), Pre and Post-Electoral Macroeconomic Fluctuations, PhD thesis, Harvard
University, Cambridge, MA.
Cutler, D., Poterba, J. and Summers, L. (1989), « What moves stock prices?», National Bureau
of Economic Research, (2538), 1–27, working paper.
D’Amico, S. and Farka, M. (2002), « The Fed and Stock Market: a Proxy and Instrumental
Variable Identification», Manuscript, Columbia University, 1–33.
Euribor-rates.eu, Euribor, 2012. Available at: http://www.euribor-rates.eu. Consulted on
July 14, 2012.
European Commission, (2012), « In-Depth Review for Belgium », May,
working document.
Commission Staff
Faniel, J. (CRISP) (2011), « Chronologie de la Crise Politique Belge 2007-2011 », 1–4, working
document.
Federal Public Service Home Affairs, Elections, June 13, 2010, June 13, 2010. Available at:
http://www.polling2010.belgium.be/en/. Consulted on August 7, 2012.
Fitch Ratings, Fitch - Complete Sovereign Rating History, 2012. Available at: http:
//www.fitchratings.com/web_content/ratings/sovereign_ratings_history.xls. Consulted
on June 3, 2012.
Harrington, S. and Schrider, D. (2002), « All Events Induce Variance: Analyzing Abnormal
Returns When Effects Vary Across Firms » SSRN, 1–41, working paper.
International Monetary Fund, (2011), « Belgium: Selected Issues Paper », 1–41, IMF Country
Report No. 11/82, Washington, D.C.
Keele, L. and De Boef, S. (2005), « Not Just for Cointegration: Error Correction Models with
Stationary Data » Nuffield College, Oxford University, (2005-W7), 1–19, working paper in
politics.
Kim, H. and Mei, J. (1994), « Political Risk and Stock Returns: the Case of Hong Kong » New
York University, (FD-94-39), 5–8, working paper.
Milyo, J. (2012), « Stock Market Reactions to Political Events: What Can We Learn about the
Efficacy of Political Connections? », 1–18, working paper.
PRS Group, (2011), « Belgium » Political Risk Yearbook: Belgium Country Report, 1–32.
Rigobon, R. and Sack, B. (2001), « Measuring the Reaction of Monetary Policy to the Stock
Market » National Bureau of Economic Research, (8350), 1–32, working paper.

99
Rochtus, D. (2010), « Is the Renascence of the Flemish Nationalism Due to the Charismatic
Authority of N-VA Chairman Bart De Wever? » ASEN Conference ‘Nation and Charisma’,
(London 13-15 April 2010), 1–19, Lessius University College/Association KU Leuven, Belgium.
Romer, C. and Romer, D. (2007), « The Macroeconomic Effects of Tax Changes: Estimates
Based on a New Measure of Fiscal Shocks», National Bureau of Economic Research, (13264),
1–47, working paper.
Rosenberg, B. and Marathe, V. (1975), « Test of Capital Asset Pricing Model Hypotheses »,
(32), 1–154, working paper.
Stoxx Limited, Euro Stoxx 50 Index, June 29, 2012. Available at: www.stoxx.com/download/
indices/factsheets/sx5e_fs.pdf. Consulted on June 3, 2012.
World Economic Forum, (2011), The Global Competitiveness Report 2011-2012, Klaus Schwab,
Geneva.