Quantifying contagion among European banks from the Great Recession to the Pandemic crisis using Network Analysis

Introduction
In the last 20-years, two economic crises have affected the European financial system:
the Great Recession and the Pandemic crisis. The first was an economic crisis that lasted
from 2007 to 2013 with a broad range of causes ranging from the strong trade imbal-
ances between China and the US to the financial bubble that burst in the US real estate
market until the excessive use of complex derivative instruments such as CDOs or CDO
squares Corden [2009].
The second crisis was caused by the outbreak of the Covid-19 pandemic that has forced
governments around the world to introduce measures of social containment as well as a
limitation to economic activity.
The two crises are comparable for the overall size of the actors involved as well as for
the strong negative shock to which the real economies around the world have been sub-
jected.
The source of the shock is the first main difference between the two crises.
In the first case, the causes of the crisis were endogenous to the economic system. The
secondoriginatedfroma"pure"blackswancomingoutsidetheeconomyDanielssonand
Shin [2002] and Danielsson [2020]. The Great Recession was a financial and economic
crisis while the nature of the Covid 19 pandemic has severely affected the real economy.
Thelinkbetweentherealeconomyandthefinancialsectorisveryclose. Thetwonatures
ofthe economic system are nothingmore than two sidesof thesame coin. Thecomplex-
ityofthemoderneconomicsystemaswellasthegreatuncertaintythatstillcharacterizes
the Pandemic crisis cannot fail to think that the crisis from the real economy turns into a
financial one.
INTRODUCTION
The most important channel of union between the real economy and the financial econ-
omy is the banking sector. It has played a key role during the Great Recession for the
exposure of the US banking sector and some European banking sector such as Ireland
to the burst of the financial bubble in the real estate market that has led to an erosion
of bank capital in the financial markets as well as the massive presence of non-perming
loans in the bank balance sheet.
The losses suffered by the former caused domino effects in which third economic agents
not directly exposed to the bubble were involved. The spread of losses due to intercon-
nectionsinthemodernfinancialsystemhasdeterminedthesystemicnatureofthiscrisis.
Inresponsetothecrisis,severalmeasureshavebeenputinplace,includingfiscalexpan-
sion measures. In Europe, the issuance of new debt, as well as the structural fragility of
the economies of the South of the Euro area contributed to the burst of the European
debt crisis Schwendner et al. [2015].
Inthecurrentcrisis,thepossibleincreaseindefaultinthecorporatesectorcouldtransfer
the crisis from the real economy to the financial economy. Possible negative effects on
financialmarketscouldcauseanewwaveofbankcapitalerosionduetothevaluationat
market values imposed by banking regulation.
The domino effects of shock transmission, as well as interconnections, are typical con-
cepts of systemic risk more precisely contagion risk. Contagion risk can in general be
defined as the probability that the instability of one institution will spread to other parts
of the financial system with negative effects.
A clearer definition of contagion in the stock market is the increase of co-movements of
stock prices Horta et al. [2009].
This thesis aims to study the contagion in the European banking stock market between
2004 and 2020. Studying contagion risk means:
• Map the paths with which shocks can be transmitted in the system.
• Quantify the shock transmission force.
The large number of actors involved in the two crises requires the use of analytical
2
INTRODUCTION
methodologies capable of dealing with high-dimensional data-sets. To investigate the
relationships and interconnections present in the system, Network analysis has been
identified as a technique able to reach this target.
Its use with the Graphical models, for the robustness of their estimates, allows taking a
picture of the fragility of the financial system in a precise historical moment.
The work is organized into 5 chapters.
In the first chapter, the concept of systemic is presented. The attention on systemic risk
grew later Subprime crisis due to the default of Lehman Brothers USA Smaga [2014].
The concept of systemic risk is no easy to be defined in a single definition. Even today
there is no consensus in providing an unambiguous definition Smaga [2014].
However, the lack of a single definition has probably encouraged the development of
several techniques for its analysis. In the first chapter, after a brief review of the main
features and some definitions of systemic risk, the Network analysis is explained.
The use of Network analysis requires the estimation of the network of interconnections
betweenbanks. Avarietyofmethodologieshavebeenproposedintheliteraturetoreach
this goal. They range from the use of bilateral exposures from bank supervisory as data
provided by EBA (European Banking Authority) Cont and Schaanning [2017] or models
using financial market data Torri et al. [2018] and Puliga et al. [2014].
The network reconstruction procedure used in this paper is based on Graphical mod-
els. The Graphical models are a family of statistical models that attempt to link Graph
theory with inferential statistics Torri et al. [2018]. This family of models is nowadays
having great success in the field of machine learning.
The estimation technique used is the T-Lasso model Torri et al. [2018]. It is a Graphi-
cal model in which the underlying multivariate distribution is a multivariate T Students
distribution. The term Lasso indicates the regularization mode used to have a sparser
network. It removes less relevant interconnections thanks to feature selection.
The T-Lasso model is described in the second chapter. It is structured into two parts:
• In the first part (sections 2.1 and 2.2) a brief introduction on Graphical models is
3
INTRODUCTION
presented. That section introduces in the first part the Gaussian Graphical model
and the Gaussian Graphical Lasso model while concludes with the explanation of
the T-Lasso model as presented in Torri et al. [2018].
• In the second part of the second chapter (section 2.3), the data set used in the em-
piricalestimationisdescribed. Thedatasetiscomposedof75equitypriceseriesof
European banks from 19 countries, Euro area and not. The period of the analysis
is from 2004 to September 2020. The section presents pre-analysis results for the
choice of the model between the G-Lasso and the T-Lasso with tests on univariate
andbivariatedistributionsgiventhecomplexityandcomputationalissuedtocheck
the statistical property of high dimensional multivariate distributions Torri et al.
[2018]. The tests on the univariates are carried out with the Kolmogorov-Smirnov,
Jarque-Bera tests, and 2
goodness of fit test.
The bivariates are estimated with the copula estimator with different model speci-
fications. The Bayesian information criteria is applied for the model selection.
The robustness between the Graphical Lasso and the T-Lasso in responding to new
outliers is performed through the rolling windows of the Frobenius norm. It is cal-
culated as the distance measure between the estimated partial correlation matrix
in t and t-1.
Empirical results are described in Chapters 3, 4, and 5.
Chapters 3 and 4 are constructed identically. They are a classic example of static com-
parison of networks. In Chapter 3 the Great Recession is presented, while in chapter 4
the Pandemic crisis is presented.
In the mentioned chapters, three snapshots of the network at three different times are
presented in an attempt to understand network structure during the crisis.
The analysis is structured in two parts:
• Centrality analysis.
It analyzes the mesomorphic structure of the network with a focus on the central
measuresandthebehaviorofthenodestoestablishreportsontheabovemeasures.
4
INTRODUCTION
Theattentionismoreontheformnetworkthanapunctualanalysisofthecentrality
of each node.
• Community detection.
We have tried to answer a double question:
– What are the most central banking groups in the network?
– What are the links between national banking systems in Europe?
In chapter 5 the dynamic Network analysis is presented. Using a rolling window, an
analysis of the evolution of the system over time is presented. The main indicators
identified in the previous two chapters are used to study the dynamic behavior of the
network over time. This chapter :
• Triestoovercomethelimitsofastaticanalysisinwhichthetimefactorisnottaken
into account.
• Tries to understand how changes in economic and financial conditions impact the
network structure.
5
Chapter 1
Systemic risk
1.1 What is systemic risk?
Thefinancialcrisisof2008hasdeeplychangedhowthemoderneconomytriestomodel
the world. From 2008 a new attention was given to the interconnections in the financial
system that increased in the last 40 years Fund [2010]. The raising of interconnections
shown some limits of the main financial models Dicembrino and Scandizzo [2012].
For instance, in the Modern Portfolio Theory of Markowitz, diversification can be used
to reduce the portfolio risk thanks to the speculars correlation in its assets. It is the base
of the CAPM in which an efficient allocation from an individual micro-level brings to an
efficient allocation at the aggregate level. However, some researchers have argued that
this is not true Dicembrino and Scandizzo [2012].
One possible explanation is that excessive diversification leads to increased intercon-
nections among financial institutions which, in the presence of strong negative events
affecting individual securities can lead to systemic losses due to the sharing of the same
asset by the different financial institutions Cont and Schaanning [2017], Aymanns et al.
[2017] and Roncoroni et al. [2019].
This is particularly increased during the financial crises for the tendency of correlations
among assets to become positive during the bear market phase. During this phase, the
financial links turn from being a means of risk diversification to channels for the prop-
agation of risk across financial institutions Caccioli et al. [2018] and Roncoroni et al.
Chapter1.Systemicrisk 1.1. WHATISSYSTEMICRISK?
[2019].
An economic explanation of these stylized facts is the spillover effect. It is the situation
in which the volatility of shock is transmitted at other financial institutions having the
same or similar assets Agénor and Pereira da Silva [2018].
The increased risk arising from the interconnection of financial agents’ exposures is one
of the ways to figure out the systemic risk. Even today, there is no clear definition of
systemic risk in the literature because of the countless characteristics that distinguish it
Galati and R. [2010].
In a literary review of the concept of Systemic risk, Paul Swaga, analyzing different def-
initions of systemic risk, has outlined a theoretical framework in which the concept of
systemic risk can be contained Smaga [2014]:
• It is frequently emphasized that systemic risk concerns a large part of the financial
system or a significant number of financial institutions. It can disrupt the perfor-
mance of the financial system and its functions, such as financial intermediation.
• Akeyelementofsystemicriskisthetransmissionofdisturbances(shocks)between
interconnected elements of the system, which may ultimately hurt the real econ-
omy.
• In the literature, systemic risk definitions began to appear in the mid-‘90s of the
XX century, but their “creation” has intensified after the outbreak of the Global
financial crisis. The focus of research has gone from an analysis of contagion to
impacts on the functions performed by the financial system.
The lack of consensus to have one singular definition has also given the ability to the
researchers to analyze it from various points of view.
According to Swaga Smaga [2014], it is possible to analyze the systemic risk for:
• Macro or Micro dimension.
– A Micro dimension describes the importance of one institution for the system.
Micro-systemic risk arises when the failure of an individual institution harms
the financial system as a whole. The concept of micro-dimension of systemic
Chapter 1. Systemic risk 7