Accepted Manuscript

Title: Banks’ systemic risk in the Tunisian context: Measures

and Determinants

Authors: Wided Khiari, Jamila Nachnouchi

PII: S0275-5319(17)30483-X
DOI: http://dx.doi.org/doi:10.1016/j.ribaf.2017.07.181
Reference: RIBAF 871

To appear in: Research in International Business and Finance

Received date: 20-5-2016

Revised date: 17-11-2016
Accepted date: 6-7-2017

Please cite this article as: Khiari, Wided, Nachnouchi, Jamila, Banks’ systemic risk in
the Tunisian context: Measures and Determinants.Research in International Business
and Finance http://dx.doi.org/10.1016/j.ribaf.2017.07.181

This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.
<AT>Banks’ systemic risk in the Tunisian context: Measures and Determinants

<AU>Wided Khiaria ##Email##[email protected]##/Email##, Jamila Nachnouchib

##Email##[email protected]##/Email##
<PA>a
Corresponding author, AssistantProfessor, University of Tunis, Institut Supérieur de
Gestion of Tunis, Department: of Finance, GEF-2A Laboratory-Tunisia bPh.D Student,
University of Tunis, Department of Finance.
<ABS-Head><ABS-HEAD>Graphical abstract
<ABS-P>
<ABS-P><xps:span class="xps_Image">fx1</xps:span>

<ABS-HEAD>Abstract
<ABS-P>This paper proposes a systemic Risk cartography for Tunisian banks in order to
shed the light on the key drivers of the Tunisian banks’ involvement in systemic risk. The aim
of this analysis is to better understand the determining factors of the systemic risk levels. We
intend to find the most influential vital few variables that are impacting the degree of systemic
risk implications.In this perspective, we will develop a proper statistical framework based on
a muli-level approach in order to explore, assess and explain the systemic risk incurred by the
Tunisian banks. As a first step, we will provide a systemic risk cartography for the Tunisian
banks using, and for the first time, a unified approach that combines CoES and Multi
Dimensional Scaling (MDS) techniques. The recovered map revealed that public banks are
taking the lead as the most involved banks in the systemic risk, followed closely by the two
most important private banks BIAT and UBCI. As a second step, we will indicate how the use
of a Systemic Risk Implication Composite Index (SRICI) provides the key drivers analysis of
banks in systemic risk implication and shows that the implication of the Tunisian banks is
highly dependent on the size of the financial institution, its liquidity, its technical efficiency,
and its direct exposure to the interbank lending market.
<ABS-P>The present research offers two main insights. First, the ranking of the banks in
terms of systemic implication, and second, the factors from which stems systemic importance.
<ABS-P>The first insight could complement the micro prudential scope of the regulation by
providing a broader perspective that includes contagion and subsequently formulates a macro
prudential vision and reinforces the regulation policy. The second one deserves higher
attention on on-going basis in any systemic risk analysis. Supervisors could detect potentially
systemic banks by keeping track of these factors and imposing a close monitoring for this type
of institution.
<KWD>Keywords: Systemic Risk; Cartography; Value at Risk; Expected shortfall;
CoVaR; CoES; MDS; PLS regression; Systemic Risk Implication Composite Index

Acronyms

AB: Amen Bank

ATB: Arab Tunisian Bank
ATTIJARI: Attijari Bank
BH: Banque de l’Habitat
BIAT: Banque Internationale Arabe de Tunisie
BNA: Banque Nationale Agricole

1
BT: Bank of Tunisia
BTE: Banque de Tunisie and Emirates
CoES: Co-Expected Shortfall
CoVaR: Conditional Value-at-Risk
ES: Expected Shortfall
EVT: Extreme Value Theory
MDS: Multi Dimensional Scaling
PLS: Partial Latent Structure Regression
PROXSCAL: PROXimity SCALing
SRICI: Systemic Risk Implication Composite Index
STB: Société Tunisienne des Banques
UBCI: Union Bancaire pour le Commerce et l’Industrie
UIB: Union Inernationale des Banques
VaR: Value-at-Risk

<H1>1. Introduction
In the wake of the sub primes crises which showed how a shock that originated from a single
institution, country or asset can quickly spread to other institutions and markets putting into
jeopardy the entire financial system. This recent showcase of large spillovers, later identified
as a materialization of systemic risk, has revealed the need for a better understanding of this
risk.
In the academia, it appears that, despite the continuing endeavor for this risk, it is not yet
possible to advance a consensus definition for it. This concept which is qualified as ``different
from other risks, multiform, mutant...and unpredictable<xps:span
class="xps_endnote">1</xps:span>'' doesn’t have a clear and widely accepted definition.
This ambiguity surrounding the systemic risk is mainly attributed to the fact that there is a
quite range of perspectives from which it can be tackled: how it originates, how it is
transmitted and how it affects different institutions, markets and economies.
One perspective is to identify systemic risk as the risk of experiencing a major shock that
simultaneously and significantly affects all financial market’s participants (De Bandt and
Hartmann, 2000 and Chakrabarty, 2012). Viewed from another perspective, systemic risk can
be akin to the risk of contagion that causes the transmission of an isolated incident occurring
in one market to the other market’s players like a domino effect (Kaufman, 1995; Schwarcz
2008 and Moussa 2011).
In many recent researches, the systemic risk is always linked with the vulnerability of the
banking sector. Indeed, Brunnermeier and Pederson (2009) plead that banks are major actors
in maintaining the stability of the economy as their role is to provide the necessary funds to its
proper functioning. Hence, crisis within the banking system are more likely to cause major
disruptions and significant costs to the economy (Barth and Caprio, 2006 and Dermirguc-
Kunt et al., 2009). Moreover, systemic risk is often considered to be the modern counterpart
of bank failure triggered by liquidity problems (Hansen, 2013). Consequently, the anxiety
over systemic risk is greatest among the bank executives.

2
The theoretical debate, which initially tackled the understanding and the definition of the
prevailing systemic risk in the financial system, had raised questions about how to
operationalize this concept, and more importantly how to manage it.
Researchers are now trying to issue appropriate empirical measures to better calculate the
systemic risk measures as they consider the widely used ones, the Beta and the Value at Risk,
weren’t efficient to deal with the global contagion as forcefully demonstrated by the recent
financial crisis. In fact, the most well-known risk measure among risk managers, namely the
Value-at-Risk (𝑉𝑎𝑅) has been criticized by many as incapable of capturing the systemic
nature of risk as it focuses on just one institution. This latter does not take into account the
fact that an institution is a part of a complex system that can generate new risks (Danielsson et
al., 2011). Consequently, it does not consider the negative spillovers associated with the
decisions made by other institutions.
Thus, a host of new quantification measures have emerged in academic literature post crisis. It
is worthwhile to investigate the large number of systemic risk measures as there is not yet a
consensus between academics and regulators bodies over a commonly used tool to estimate it
accurately. The related researches can be divided into two main streams. The first one
analyses the channels through which systemic risk spread from one institution to the other
parts of the system (Pritsker, 2000 and Forbes and Rigobon, 2001). The second one focuses
mainly on individual measures of systemic risk. This stream of literature evaluates the co-
dependence of financial institutions in the financial system during times of distress. In this
literature, we have the prominent Conditional Value-at-Risk (CoVaR) advanced by Adrian
and Brunnermeier (2011). The researchers have adapted the VaR to measure systemic risk
through introducing a new conceptual framework based on the tail covariation between
financial institutions’ returns. They defined it as the𝑉𝑎𝑅of institution i conditional on a
financially distressed institution j .Danielsson et al. (2011) underline that such a measure is
rugged tool as it takes into account the spillovers effects. Moreover, the researchers adopted
the CoVaRto another attractive risk measure, the Expected Shortfall (ES) which they calledthe
co-expected shortfall (CoES). Mainik and Schaanning (2012) consider the work of Adrian and
Brunnermeier as « one of the major threads in the current regulatory and scientific discussion
of systemic risk».
Adrian and Brunnermeier (2011) presented the CoVaR and the CoES in order to manage
efficiently the financial instability. These two measures would be very useful applied to the
Tunisian banking system, which is now going through a rough patch, and being the principal
cause of the economic and financial systems vulnerabilities. The Tunisian case is one example
among many others where the systemic risk has become, in the recent years, one of the major
concerns of the financial authorities in the country. After the Revolution of January 14, 2011,
Tunisians have experienced an unprecedented period of insecurity through its struggle with
terrorism and strikes which deepened instability and delayed economic recovery. Tunisia's
economic difficulties become even more important with the debt crisis of its main economic
partner, namely the European Union. Furthermore, with the Libyan revolution, our direct
neighbor, the stock market indexes and the cash transactions have dropped sharply and the
Tunisian dinar has dangerously declined. Tunisia has also experienced persistently high rates
of unemployment, especially for the educated youth, and an increasing degree of labor market
segmentation that created a widespread sense of economic disenfranchisement as well as
substantial loss of potential growth. Moreover, the former regime and its allies engaged in
conspicuous and predatory corruption that undermined investment and entrepreneurial
incentives and sowed indignation over the lack of free and fair competition in the market
place. For instance, poor governance of the banking sector has led to deep dysfunctions which
made the banking system inefficient, highly vulnerable, characterized by low profitability,

3
and especially unable to optimally allocate resources toward the productive activities resulting
in weak economic performance and insufficient jobs creation.
These national and international hostile conditions have deeply crippled all Tunisian sectors,
especially the banking sector, as banks have become unable to cope with the financial
instability. This means that Tunisian Banks are now extremely vulnerable to any adverse
shocks. Knowing that the Tunisian financial sector is dominated by banks, with assets equal
to about 115<xps:span class="xps_endnote">2</xps:span> percent of GDP, any agitation that
originates within this particular sector is of great impact on the stability of the Tunisian
economy. Although small and fragmented, this sector dominates the nonbank financial sector
as this latter represented only about 20 percent of all financial system assets in 2011.
According to the IMF, the Banking sector vulnerabilities are likely to be much higher
compared to what is declared in the officially reported balance-sheet data. Indeed,
reclassifying<xps:span class="xps_endnote">3</xps:span> some rescheduled loans as
nonperforming would add about 5 percent to the overall NPL ratio which is around 13, 3% in
2011. It is important to note that the State owned banks hold the major part of this NPL
compared to the private banks. Well aware of this fact the government has launched reforms
to modernize and strengthen these banks starting with a full audit before proceeding with their
recapitalization (World Bank, 2016). The authorities are also working on the improvement of
the legal framework to deal with the inadequate bankruptcy law and a weak judicial system
that do not allow banks to quickly use collateral to recoup its losses from delinquent loans
which make them reluctance to lend.
This paper adds a different approach to the growing literature related to systemic risk in the
banking sector. We propose an appropriate muli-level approach in order to explore, assess and
explain, and for the first time, the systemic risk incurred by the Tunisian banks.
The use of a combined approach associating the CoES and MDS methodologies allows the
reconciliation of two different stands of literature. Firstly, our work can be associated with the
large field of literature that relies on the information included in market prices with the use of
stock prices as in put data for the calculation of the ES and the CoES. Giving that we don’t
mix market and balance sheets data, in this work the CoES could be qualified as
homogeneous measure. Secondly, this work can be related to the literature through the
network analysis. Indeed, using an MDS techniques we attempted to address the cross
sectional dimension of systemic risk (which refers to the increasing interconnectedness among
financial institutions) within the banking framework. We actually seek to analyze the bilateral
exposures between entities of the banking sector while taking into consideration the intricate
structure of the modern financial system. This particular feature of the banking sector has
participated in the collapse of the market assets as revealed by the recent financial crisis.
Moreover, it is a well-documented fact that at the core of the systemic risk there is acontagion
which is exasperated by the intertwined nature of the financial system (Sankaran et al., 2011
and Danielsson et al., 2011).The network analysis allows investigating the types of linkages
that exist between different financial institutions and tries to assess their degree of
interconnectedness.
The remainder of the paper is organized as follows. Section 2 will present the methodology
used to explore, assess and explain the systemic risk incurred by the Tunisian banks and will
include a brief description of the data used for our analysis. Our main empirical results are
presented in Section 3.The final section will cover the conclusion.
<H1>2. Methodology
The purpose of this section is to develop an appropriate statistical framework to better
understand the systemic risk in the Tunisian banking sector. In this perspective, we adopt a
muli-level approach in order to explore, assess and explain the systemic risk incurred by the
Tunisian banks. First, we provide systemic risk cartography for Tunisian banks using, and for

4
the first time, a unified approach combining CoES and Multi Dimensional Scaling (MDS)
techniques. Second, this same technique that allowed putting forward a Systemic Risk
Implication (exposure and contribution) Composite Index (SRICI). Using this index we
undertook a Partial Latent Structure regression (PLS) that implicates multiple sources of
systemic risk in order to analyze the key drivers of the banks' involvement in the systemic
risk.
The first sub section will include a brief description of the data used for our analysis. The
second one will present the Multi-Dimensional Scaling approach. The third sub section will
focus on the major steps to calculate index values. The last section we will put the accent on
the exploration of the major drivers of systemic risk using a Partial Latent Structure
regression (PLS).
<H2>2.1. Presentation of Tunisian banking sector and data

Tunisia’s banking sector is dominated by public-controlled banks. In 2012, the sector

comprised five state-owned commercial banks (representing 39% of total banking assets in
June 2011), ten private commercial banks (33% of assets), and six foreign banks (28% of
assets). The three largest state-owned banks (STB, BNA and BH) represented 37% of the
assets of the banking sector. Similarly the three large private domestic banks (BIAT, BT and
AMEN) represented28% of the total assets. Three of the largest foreign banks (from France,
Jordan, and Morocco) are former state-owned banks, but only one of them appeared to have
completed its restructuring. There are five small development banks, partially with funds from
the Gulf States, which enjoy universal banking licenses. No major changes in the number of
market players have occurred during the last five years except the implementation of a second
Islamic bank (Zitouna Bank).
This division in the market shares is demonstrated by the fact that no institution, in the Middle
East and North Africa region (MENA), has a market share greater than 14% of total assets or
loans and 16% of deposits, which is unusual as these percentages are generally much higher.
The three largest banks—BIAT, BNA and STB—have almost 50% of the total assets with
approximately equal weight. Contrary, in Morocco for example the top three banks granted
62% of loans to the economy while the first five recorded 81% in 2012 (IMF and WB, 2012).
Poor governance of the banking sector has also led to many several issues. One of these is the
abnormally high proportion of non-performing loans. Though the size of non-performing
loans compared to total assets has decreased from a recent peak of 24.2%, it remains high at
13%. This practice has left banks overly exposed to certain sector that did not perform well,
which represents 20% of all bad loans.
Average bank capitalization ratios are broadly in line with international standards, but in some
cases they are not yet adequate to maintain solvency given these high non-performing loans.
In 2009, banks’ average Capital Adequacy Ratio stood at 12.4% with only 10.9% associated
to the state‐owned banks. This is only slightly higher than the minimum ratio set by the Basel
Accord and may slip further. The largest state‐operated bank, STB, is in a particularly severe
financial situation. In 2011, its capital adequacy ratio fell below the regulatory threshold of 8
percent and the quality of its loan portfolio deteriorated. Due to the STB’s size and
importance in the banking sector, its financial difficulties may have far‐reaching economic
effects.
These issues include difficulties that can sometimes encounter banks when managing their
risks. In fact, Tunisia is now facing several major economic, social, and political challenges
since it is trying to create a more democratic and accountable political system in order to
attain higher levels of prosperity. There are important risks, especially in the banking sector,
that must be managed carefully so that they will not undermine the economic and social
progress already achieved.

5
In this research, we will especially focus on the publicly traded Tunisian banks. Currently, the
Tunisian banking sector includes 11 deposit Banks listed in the Tunisian Stock Exchange.
Three of these banks are public (STB, BNA, BH) since the government’s participation in their
capital is higher than 36%. Hence, our panel contains the banking sector index and the 11
banks. We started our research sample in January 2009 and ended it in October 2014. Thus,
the data recovered also involves consequences of the Tunisian revolution of 14th January
2011. The daily data obtained from Tunis Stock Exchange, which are used to generate
market-valued assets returns, were changed to weekly data. The banking sector index and
every bank included in our final estimation sample have 300 non-missing market valued
assets returns.
Table 1 shows the different statistics of financial time series of the 11 Tunisian banks of our
panel.
Most of the banks have negative means except for three banks namely BIAT, ATTIJARI and
AB. This negative trend mirrors perfectly the financial difficulties encountered by the
Tunisian banks. Considering the fact that this statistical measure of central tendency is very
sensitive to extreme values and that it can be heavily contaminated by outliers, we decided to
calculate the adjusted mean. This latter compensates for this measure by omitting a pre-
specified percentage of values on the tails (the largest along with the smallest values)and
calculates the mean using the remaining observations.
By comparing the mean and the adjusted mean we find that the values of these two statistical
indicators are different, indicating that the banks’ returns distribution is skewed. The results of
these higher moments lend support to the conclusion already noted regarding the asymmetry
of the returns’ distributions. Indeed, four of these banks namely BIAT, BH, UIB, and BTE are
skewed to the right as they have a positive skewness. However, the BT, UBCI, TIJARI, AB,
STB, ATB and BNA, have negative Skewness. Hence, they are skewed to the left. These
banks tend to have negative extreme values. All banks have high kurtosis values of returns is
an evidence of the non-normality of their returns’ series. An intriguing observation is related
to three specific banks which are UBCI, AB and BT. A further look to their standard
deviation, their skewness and their kurtosis reveals that these banks have the highest values
for these three indicators. Hence they are the most skewed ones and they represent the largest
range between the lowest and highest returns. More generally we can conclude that their
returns are scattered far and wide from the mean.
This observation shows importance of the third and the fourth moments of the distributions.
Due to these considerations, we were tempting to bypass the restricted use of VaR and
CoVaR and move further to calculate the Expected Shortfall and CoES as they give more
information about the distribution of returns in the tail.
<H2>2. 2. The Multi Dimensional Scaling Approach
Multidimensional scaling, in its most basic form, states a set of elements (banks) in a low-
dimensionality space. The objective is to obtain a spatial map corresponding to the input data,
with the smallest number of dimensions. In our case, the input data is a square matrix of order
11 in which all banks are inventoried in rows and columns: the elements we are trying to
represent in the multidimensional space. This matrix called P is shown in Table 2. What we
have as a data is actually a matrix of interdependencies in terms of systemic risk between all
bank pairs. More explicitly, each cell located at the intersection of row i and column j,
contains the mean ∆𝐶𝑜𝐸𝑆𝑞𝑖 and consequently indicates the effect of bank i on the bank j. There
aretwo different possible readings of the matrix P depending on the direction chosen. Each
column corresponds to the systemic risk vulnerability profile for the bank j when other banks
fail. Similarly every row gives the systemic risk impact of a bank i over the other banks.

6
 𝑠𝑦𝑠 𝑖
The last column and row of table 2 shows respectively the ∆𝐶𝑜𝐸𝑆𝑞,𝑡 and the ∆𝐶𝑜𝐸𝑆𝑞,𝑡 <
𝐱𝐞𝐧 > 𝟒 </𝐱𝐞𝐧 >. Each of these two terms expresses, for each bank i, its levels of
vulnerability and contribution to the distress of the banking sector, respectively.
In order to obtain a common space by simultaneously visualizing the positioning of banks
according to their levels of exposure and contribution in terms of systematic risk, w e use the
matrices P and its transpose P’ as input data for MDS technique. In our case, each column of
the matrix P is retained to reflect the vulnerability profile of a particular bank to the
difficulties of the other banks. In contrast, each column of the matrix P’ is supposed to set the
nuisance effect of a given bank upon the others.
We had to deal with the fact that the diagonal of the matrixes P and P’ were empty. Hence, the
base line idea of the choice of the CoES over the CoVaR is the need of coherent elements to
fill these diagonals. Our innovative approach uses ∆ES for this purpose. This latter is
calculated as follow:
∆𝐸𝑆𝑞𝑖 = 𝐸[𝑋 𝑖 |𝑋 𝑖 ≤ ̅̅̅̅̅̅
𝑉𝑎𝑅𝑞𝑖 ] − 𝐸[𝑋 𝑖 |𝑋 𝑖 ≤ ̅̅̅̅̅̅
𝑉𝑎𝑅50% 𝑖
] (1)
What makes the CoES a better candidate than the CoVaR is that ∆𝐸𝑆𝑞𝑖 aims at quantifying the
loss registered by a bank i, if this latter goes from a median state to a state of financial crisis.
In the literature, such approach, which is based on multiple input matrices, is known as a
disaggregated MDS. This technique takes the tow input matrices data and generates a relative
image of the input information such the distances {𝑑𝑖𝑗 } between these points correspond as
closely as possible to these proximities (Kruskal, 1964, Torgerson, 1958). Here, the
greater∆𝐶𝑜𝐸𝑆𝑞𝑖 will be the proximities between banks, and in addition the smaller will be the
distance in common multidimensional space.
Multidimensional scaling programs project proximities as distances in space. This is done by
solving the following objective function
𝑀𝐼𝑁{𝑑𝑖𝑗} σ = ∑[ 𝑑𝑖𝑗 − f(𝛿𝑖𝑗 ) ]2 (2)
Where f( ) is an hypothesized proximity function and when σ is close to 0 reveals a good fit
between proximities and distances.
There are different available methods of MDS. In this present work we choose to apply the
PROXimity SCALing (PROXSCAL) technique (Commandeur and Heiser, 1993)
implemented in IBM-SPSS 21. This particular choice is motivated by the fact that the
PROXSCAL improve the efficiency of the solution given by MDS analysis. Indeed, instead of
solving Equation (9) as any other MDS model, the PROXSCAL procedure implements an
Iterative Majorization (IM) algorithm (Commandeur and Heiser,1993). For our research, we
used the weighted model with ratio transformation as it appears at first glance the most
suitable for our work.
To establish the map, two fictitious banks are incorporated in the analysis. The first associates
the highest ∆𝐶𝑜𝐸𝑆𝑞𝑖 value for each bank that comes out of the correspondent column of the
matrix P. It symbolizes an extreme situation of implication in terms of systematic risk.
Conversely, the second refers to the lowest ∆𝐶𝑜𝐸𝑆𝑞𝑖 value and indicates a situation of
insignificant implication in the systemic risk. This approach led us to a space in which each
bank is indicated by a point, and where all the points are represented by a direction from a
reference point I (insignificant implication in the systemic risk) to a reference point II
(extreme implication in the systemic risk). In this representation model, called ``point-vector
model'', the banks’ involvement order in systemic risk is given by the order of the projections
(orthogonal) of the different points in that direction.
From the MDS technique we recovered a two-dimensional map represented in Figure 1.

7
<H2>2. 3. Systemic Risk Implication Composite Index: SRICI
In the following, we proposed a Systemic Risk Implication Composite Index (SRICI) based
on the results of the implication’s dimension recovered from an MDS analysis.
Methodologically speaking, we have established a scaling of banks that goes from the most
involved banks in systemic risk to the least implicated ones. Using the ``points vector model''
settings, we will be able to recover a ranking of banks based on their implication in systemic
risk by proceeding to the projection (orthogonal) of each point representing a bank on the line
joining the two references (as shown in Figure 1).
For every point, we calculated, on one hand, the ratio between the distance that
separates each bank of the first reference and, on the other hand, the second reference.
2
𝑑 (𝐵𝑖 ,reference1)
𝑆𝑅𝐼𝐶𝐼𝑖 =ln( 2 (3)
𝑑 (𝐵𝑖 ,reference2)
Where: 𝑑 2 (𝐵𝑖 , reference1): distance between a bank i and reference 1
𝑑 2 (𝐵𝑖 , reference 2): distance between a bank i and reference 2
A positive value of the index indicates a higher implication in systemic risk. In the
same line, a negative value mirrors a situation of a less important systemic involvement.
<H2>2. 4. Determinants of systemic risk implication: PLS regression approach
In the following subsection, we shed the light on the key drivers of the implication of
Tunisian banking institutions in systemic risk. The aim of this analysis is to gain a better
understanding of the determining factors of the systemic risk levels. We want to find the most
influential vital variables on the degree of systemic risk implication. To this respect, the
Investigation is carried out by means of multivariate statistical technique. Although there is a
wide array of methods that can be used for this purpose, they often perform poorly when the
multi co-linearity exists among variables. For our research, we used the projection on latent
structures analysis (Known as PLS) as it is a well suited method that deals with few
observations on a large set of highly-correlated variables (11 observations for each variable).
This non parametric multivariate method is a recent technique that combines features,
generalizes principal component analysis (PCA) and multiple linear regressions. Using this
method, we can predict a set of dependent variables from a set of independent variables or
predictors with no need to stringent assumptions regarding the variables and the residual
distributions. This is achieved by extracting from the independent variables a set of
orthogonal factors called latent variables as performed in the PCA but with a slight difference.
In fact, orthogonality of the principal components eliminates the problem of multico-linearity.
This technique is used to establish a mathematical relationship between the numerous set of
factors of systemic implication (referred to as independent variables), represented by a matrix
X and the Systemic Risk Implication Composite Index (SRICI), represented by a matrix Y
(Procopio et al., 2013).
In our research, the PLS regression analysis was carried out with a large set of variables that
were taken to capture different aspects of the banks’ activities. We recovered over 30 different
variables inherent to the risk’ management strategy (risk related cost, solvability ratio, level of
classified loans…), liquidity (depth ratio…), the positioning over the market (market
capitalization, number of exchange…), size (number of branches, total assets ….) and
interbank market (interbank claims...)…. Data used in this study were collected from the
banks’ balance sheets, and their annual financial reports along with the reports published by
the Tunisian market of exchange and by different Tunisian brokers during the period from
2009 to 2014.

8
After conducting the PLS regression, we proceed to a selection of the most relevant
predictors. This identification is usually achieved using a strategy based on the VIP score of
predictor which stands for the importance of the projections found in latent variables. The
basis of this PLS-VIP methods is to use ``the greater than on rule'' to judge the relevance of
the selected variables. Wold et al. (1993) stated that for a variable to be excluded, it should
have a small VIP. Based on the result recovered from the PLS regression, we only kept 11
variables that register a VIP higher than one. These variables are stated as follow: number of
Branches, total assets, Bank revenue, operating expenses, number of exchange, market
capitalization, share of credit market, interbank deposits and assets, interbank claims,
operating ratio, number of securities treated and net earnings.
<H1>3. Results and discussions

Let us recall, at this level, that the main objective of this paper is to provide a mutli-level
approach allowing the exploration, assessment and explanation of the systemic risk incurred
by the Tunisian banks. First, we provide systemic risk cartography for Tunisian banks using,
and for the first time, a unified approach combining CoES and Multi Dimensional Scaling
(MDS) techniques. Second, we construct a Systemic Risk Implication Composite Index
(SRICI) using the results of the implication’s dimension recovered from the MDS analysis.
The outlined index is used to investigate the dominant determinants explaining the
implication of Tunisian Banks into systemic risk based on a PLS regression.
<H2>3.1. Results of the MDS approach

The results from the MDS technique provided a cartography (Figure 1) that allowed us to
establish the ranking of banks based on their involvement in systemic risk. As shown in this
figure, concerned with the involvement in systemic risk, public banks namely STB, BNA and
BH occupy the top three places as they are closer to the maximum point. They are closely
followed by two of the major private banks in Tunisia: UBCI and BIAT. These five banks are
the major systemic actors inthe Tunisian banking sector. It appears that they are the most
vulnerable ones to the other banks financial difficulties and the biggest contributors to the
distress of the existing banks. Then comes the remaining private banks respectively UIB,
ATB, ATTIJARI AB and BT which have a less important systemic implication. Notice also
that BTE is the less systemically involved bank as it is the bank situated the most to the left.
A second reading of the cartography derives from the identification of dimensions. According
to figure 2, it seems that the first dimension indicates the systemic potential of the bank.
Based on this observation, we argue that this dimension account for the banks ‘contribution in
systemic risk. As for the second dimension, it seems to be related to the sensitivity of banks to
systemic risk.
It appears that banks which are located further to the right are perceived as the most systemic
risk producer. This implies that they are bigger contributor to systemic risk than the ones on
the left. Using the second dimension, one can conclude that banks that are located lower down
faces fewer losses in case other banks fail, than those being on the higher up. This particular
reading allows the introduction of the domination notion. Indeed, the graph shows two
separated group of banks. The first class in green is dominating the second class in red. Banks
situated bellow the diagonal of the cartography imposes important systemic risk to the banks
over the diagonal.
A third reading that could be derived from our graph is related to the fact that banks situated
in the higher part of the graph, and more precisely over the line of projection relating the
maximum and minimum points, have interestingly the higher standard deviation, skewness
and Kurtosis. Namely the UBCI, BT and AB as noticed earlier are those who are the most
fragile to internal financial deficiencies.

9
In the MDS literature, the goodness-of-fit of the derived solution is a very important step as it
provides the degree to which the configuration generated by the program fits the ideal one.
There are usually two types of indicators. The first type belongs to the normalized raw stress,
for which zero represents a perfect fit and generally a value less than 0.100 is acceptable
(Kruskal and Wish, 1978). The second type of indicators such as Tucker’s congruence
coefficient where a value of 1 represents the perfect fit
The goodness-of-fit’ indices offered by the PROXSCAL are presented in Table 3 below.
From Table 3 one can reaches the conclusion that the fit of the model with the two dimensions
is very good.
The conclusions advanced in the upper part of the article are in line with those obtained
𝑠𝑦𝑠
studying ∆𝐶𝑜𝐸𝑆𝑞𝑖 and the ∆𝐶𝑜𝐸𝑆𝑞 .
Table 4 provides a comparison between the banks' ranking based on the systemic implication
recovered from the MDS that combines two aspect of systemic risk (exposure and
contribution to systemic risk) against dealing only with contribution (∆𝐶𝑜𝐸𝑆𝑞𝑖 )or exposure
𝑠𝑦𝑠
(∆𝐶𝑜𝐸𝑆𝑞 ) separately.
We then examine the relationship between these different rankings by conducting a
spearman’s rank correlation. The results of this analysis are reported in a single summary
table. It lays out the name of each risk measure and the way in which their rankings are
correlated with MDS ranking in terms of systemic implication.
Results provided in Table 5 show that the ranking based on the involvement in systemic risk
seems to be highly correlated with both rankings, although it appears to be more correlated
with the ranking of contribution. In this sense, it is more related to the banks’ contribution to
the financial distress more than their sensitiveness to it. This observation could explain the
ranking of UBCI which is very divergent. Indeed this bank occupies the fifth place based on
the implication criteria but the first and the eighth place based on the exposure and the
contribution to the systemic’ risk. Respectively,

<H2>3.2. Results of the SRICI construction

This SRICI identifies the most involved banks in systemic risk. According to Figure 1, there
is a huge presence of points on the second reference’s side. This trend reflects a widespread
situation of systemic implication among Tunisian Banks as indicated by larger number of
positive values of the SRICI. Results in the following Table 6 support the above assumptions
as the mean and the median of the index are positive.
The purpose of the following sub section is to investigate the consistency of our SRICI and its
coherence with other risk measures. To this end, reflections initially conducted over the
correlation between different rankings are here provided by a factor analysis (Principal
Components Analysis: PCA).
𝑆𝑦𝑠
Table 7 shows that the two risk measures, ∆𝐶𝑜𝐸𝑆𝑞𝑖 and∆𝐶𝑜𝐸𝑆𝑞 , that indicates two different
aspects of systemic risk respectively systemic potential and systemic exposure are not
correlated. However, our SRICI seems to be highly correlated with both measures. This
demonstrates its multidimensional character which grants it the status of composite index.
This conclusion is further verified by the results of the following table which indicates that
these measures can be grouped in a unique dimension. This process generated only one factor
reflecting the systemic implication of a bank in systemic risk.
Inter-correlations among the sub-indexes included in this unique factor are analysed using the
Cronbach's alpha. Indeed, this measure of internal consistency indicates the degree to which a
set of measures the same single underlying construct. Thus it reflects homogeneity since it

10
describes how closely linked a set of measures are as a group. It is considered reliable if it
exceeds 0, 7. Table 8 shows that our Cronbach's alpha is higher than the latter threshold:
0,727. We can claim that these measures are describing the very same latent construct which
is the implication of a bank in the systemic risk (its vulnerability as well as its contribution to
systemic risk).
It should be noted that our SRICI holds the higher score of representation of systemic
importance. These results validate the fact that our index is the best indicator of systemic
involvement.
<H2>3.3. Results of the PLS regression

Results of the PLS regression are presented in Table 9. As shown in this table, the quality of
fit of the model is acceptable as R2 is equal to 0,969 and the adjusted R2 is equal to 0.939.
These remaining variables judged as most significant were combined over a reduced number
of components. Affecting each variable to the appropriate factor is traditionally achieved
using the sign and the magnitude of the weight of the variable, which represents their relative
contribution to factors. However, using solely this criterion to judge the importance of a
variable is subject to serious criticism as it is typically unstable particularly in instances of
multicollinearity. Based on our results the presence of multicollinearity is proven through the
opposite signs assigned to the loadings and the weights of some variables.
This inherent deficiency is for the sole use of the weight has led us to base our interpretation
over an increasingly used criterion namely the loading which is considered more valid than
the former one. This latter reflects the simple linear correlation between the variable and the
factor. The larger loading, the more important it is in deriving the concerned factor. Based on
the loadings contained is Table 9, we were able to affect each variable to the most appropriate
factor. As such, the first factor contains number of the Branches, total assets and share of
credit market. The second one has the number of securities treated, number of exchange and
market capitalization. As for the third and the forth factors we count respectively interbank
deposits and assets, interbank claims and Bank revenue, net earnings, and operating expenses.
Looking at these variables in greater details is crucial to better understand the determining
factors of banks’involvement in systemic risk. Indeed, identifying these factors was the most
challenging as it required a careful investigation and a closer inspection of variables.
Following our results, factor one seems to be related to the size of the banks. In the literature
on systemic risk size is identified as one of the dominant determinant of systemic risk. So not
surprisingly, we find that a bank with a bigger size is more involved in systemic risk. The
second factor is rather oriented towards the activity of the bank on the stock market. We
referred to it as liquidity. An increasing number of papers have illustrated that the inability of
banks to roll over or to acquire a new short term funding, and this represented a key aspect on
the worsening of the Banking crisis of 2007-2009. Due to the important shortcomings in
banking liquidity practices, central banks had to step in and distribute liquidity in the system
to prevent a total breakdown of the banking sector. The third factor reflects the interbank
commitments. It traces the interconnected architecture between banks and reflects the
interbank channels of contagion. Similar to the size,this factor was heavily investigated in the
literature as an important amplifier of initial shocks. Our results shows that higher
interconnectedness can lead to higher banks’ systemic risk implication as a growing level of
interbank linkages can render the network more fragile to systemic shocks. The last factor
relates to technical efficiency. The systemic risk implication strongly depends on the
efficiency of the bank. A mediocre efficiency translates into a higher vulnerability to systemic
shocks and systemic risk implication rises accordingly.
<H1>4. Conclusion

11
The international uncertainties that emerged due to the recent crises and the national
instability attributed to the Tunisian revolution have prompted an increasing interest to
systemic risk. Tunisian banks’ executives are now more than ever aware of the potential of
one bank’s failures to spillover to the other institutions and consequently put into jeopardy the
whole financial system. Nonetheless, the risk of spillovers among these banks has not been
examined yet from a systemic risk’s angle.
This paper proposes a systemic Risk cartography for Tunisian banks in order to display the
key drivers of the Tunisian banks’ involvement in the systemic risk. Using stock market
information, we compute VaR, ES, CoVaR and CoES estimates. The use of CoES allows
dealing with the problem that CoVaR exhibits when responding to idiosyncratic risk by
introducing the∆ES. In a second step, these latter estimates were confined in an
interdependency matrix. Then they were used to conduct an MDS process that generates a
systemic’ risk cartography. This map provides information about the bank’s implication in
systemic risk and the extent to which each of them is affected by the internal financial
distress.
The MDS technique allows to put forward our SRICI using a point vector model applied to
the MDS output. This latter was put in place in order to investigate the dominant determinants
of systemic risk implication based on a PLS regression technique. This regression revealed
four major drivers of systemic involvement namely the size, banks’ liquidity, interconnections
with other banks and technical efficiency.
Results suggest that public banks respectively STB, BNA, and BH along with the two most
important private banks UBCI and BIAT hold the leading positions in the systemic risk
rankings. What emerges from this exercise is that those banks are the largest systemic players
among the Tunisian banks. It is also important to highlight that BTE is the least involved bank
into systemic risk. These results involves among other things, a rethinking of risk
management practices of these banks.
The present research offers two main insights. First, the ranking of the banks in terms of
systemic implication, and second, the factors from which stems systemic importance.
The first insight could complement the micro prudential scope of the regulation by providing
a broader perspective that includes contagion and subsequently formulates a macro prudential
vision and reinforces the regulation policy. The second one deserves higher attention on on-
going basis in any systemic risk analysis. Supervisors could detect potentially systemic banks
by keeping track of these factors and imposing a close monitoring for this type of institution.
Nonetheless, it is important to mention that there are some methodological limitations that can
be addressed to our methodology due to the use of quantile regression. Indeed, its estimation
error grows considerably in the extreme quantiles of the distribution. Moreover it is
impossible to measure VaR with precision which makes the COVaR’s estimates less precise.
Another main difficulty encountered during this work is that data on bilateral exposures are
not publiclyavailable andonly revealed to national supervisors. The scarcity of data is also due
to the fact that some of the information is not published on a systematic basis.
The method implemented in this article can easily be extended to wider range of data.
Another interesting avenue for further research is to refine this method. Imporvements can lay
in the use of more sophisticated models in the calculation of VaR, CoVaR,ES and CoES.
<ACK>Acknowledgement
The authors would like to thank the Editor in Research International Business and Finance.
Authors are grateful to anonymous reviewer for helpful comments.
Appendix A

12
The literature on systemic risk measures, and more specifically on the CoVaR, abounds of
different methodologies to estimate the CoVaR. Giraldi et al. (2013), for instance, estimated
CoVaR based on a multivariate GARCH. Moreover Hakwaetal. (2012)used a copula
approach. In our context as the calculation of CoVaR is only a mean to an end (the sole
purpose of its estimation is to calculate the CoES), we followed the seminal work of Adrian
and Brunnermeier (2011) and opted for quantile regression. Several key building blocks of the
proposed approach are explicated in this section. First, we conduct the𝑉𝑎𝑅 measures using
Filtered Historical Simulations. Second, we run the ES’s estimates.
Indeed as advanced by Adrian and Brunnermeier (2010), an adequate technique to estimate
the CoVaR is through quantile regression since it provides a more extensive analysis than
ordinary least squares. This technique is also known for its simplicity and its robustness when
exploring relationships between variables evaluated in extreme quantiles. It takes into account
the non-linearity of the dependencies between the returns. In addition, quantile regression
does not require a hypothesis regarding the distribution of variables, as it is a non-parametric
method. Hence, it avoids the bias inherent to the distribution’s assumptions of the parametric
methods.
Formally, the CoVaR is a «dependence adjusted version of VaR» (Mainik and Schaanning,
2012). Put differently it is a VaR conditioned on a bad event (a financial institution being at its
VaR)like it is shown in the equation (1’) below

CoVaR α (Y| X) = VaR α [Y| X = VaR α (𝑋)] (1’)

As stated in the work of Adrian and Brunnermeier (2011) we consider the predicted
𝑖,𝑗
value of a quantile regression of a bank i 𝑋̂𝑞 on a particular bank j for the quantile q
𝑖,𝑗 𝑗 𝑗
𝑋̂𝑞 = 𝛼̂𝑞 + 𝛽̂𝑞 𝑋𝑗 (2’)
In quantile regression settings, the CoVaRwhich estimates the loss of the bank i when
a bank j faces an extreme event is given by
𝑖|𝑋𝑗 =𝑉𝑎𝑅𝑞 𝑗 𝑗 𝑗 𝑗
𝐶𝑜𝑉𝑎𝑅𝑞 =𝑉𝑎𝑅𝑞𝑖 |𝑉𝑎𝑅𝑞 = 𝛼̂𝑞 + 𝛽̂𝑞 𝑉𝑎𝑅𝑞 (3’)
Using these estimates, we conducted the CoES measures using the following
definition provided by Adrian and Brunnermeier (2011)
𝐶𝑜𝐸𝑆𝑞𝑖 = 𝐸[𝑋 𝑠𝑦𝑠𝑡𝑒𝑚 |𝑋 𝑠𝑦𝑠𝑡𝑒𝑚 ≤ 𝐶𝑜𝑉𝑎𝑅𝑞𝑖 ](4’)
That is, the CoESqi corresponds to the expectation over the q-tail of the conditional probability
distribution.
The specific choice of this tool as a systemic risk measure is mainly motivated by the
particular in formativeness of its methodology regarding the systemic profile of a particular
bank. In fact, it allows studying the bilateral relationship between two financial institutions.
Using CoESwe can measure not only the contribution of the bank j to the risk of a bank i
(∆𝐶𝑜𝐸𝑆𝑞𝑖 ) as calculated using Equation (5’) but also its vulnerability to its financial distress
𝑗
(∆𝐶𝑜𝐸𝑆𝑞 ) as shown in Equation (7).
̅̅̅̅̅̅̅̅̅𝑞𝑖 ] − 𝐸[𝑋𝑗 |𝑋𝑗 ≤ 𝐶𝑜𝑉𝑎𝑅
∆𝐶𝑜𝐸𝑆𝑞𝑖 = 𝐸[𝑋𝑗 |𝑋𝑗 ≤ 𝐶𝑜𝑉𝑎𝑅 ̅̅̅̅̅̅̅̅̅50%
𝑖
](5’)
𝑗 𝑗
̅̅̅̅̅̅̅̅̅𝑞 ] − 𝐸[𝑋 𝑖 |𝑋 𝑖 ≤ 𝐶𝑜𝑉𝑎𝑅 𝑗
̅̅̅̅̅̅̅̅̅50% ](6’)
∆𝐶𝑜𝐸𝑆𝑞 = 𝐸[𝑋 𝑖 |𝑋 𝑖 ≤ 𝐶𝑜𝑉𝑎𝑅

Knowing that for our case, we used themean ̅̅̅̅̅̅̅̅̅̅̅̅̅̅

j j|i
∆COVaR q,t (the average of ∆COVaR q,t over the
observation period) to calculate the CoES. We can further consider the case where the
𝑆𝑦𝑠
institution j is the banking sector. Applying (4’) and (5’) we recover respectively ∆𝐶𝑜𝐸𝑆𝑞

13
and ∆𝐶𝑜𝐸𝑆𝑞𝑖 . The first establishes the sensitiveness of banks to the distress of the banking
sector and the latter reflects their capacity to trigger extreme events known as ``systemic
importance''.
<REF>References
<BIBL>
Adrian, T. and Brunnermeier, M.,;1; 2011. COVaR. Federal Reserve Bank of New York Staff
Reports, N° 348.
Barone-Adesi, G., Giannopoulos, K. and Vosper, L.,;1; 1999, VaR without correlations for
portfolios of derivative securities. Journal of Futures Markets. Vol. 19 No 5, pp. 583-602.
Barth, J. and Caprio, G.,;1; 2006. Rethinking Bank Regulation:Till Angels Govern.
<PL>Cambridge </PL> <PN>University Press</PN>, <PL>Cambridge</PL>.
Bhattacharyya, M. and Ritolia, G.,;1; 2008. Conditional VaR using EVT – Towards a planned
margin scheme. International Review of Financial Analysis, Vol. 17 No. 2, pp. 382-395.
Brunnermeier, M., and Pedersen, L.,;1; 2009. Market liquidity and funding liquidity. Review
of Financial Studies, Vol. 22 No. 6, pp. 2201–2238.
Chakrabarty, K.C.,;1; 2012. Systemic risk assessment – the cornerstone for the pursuit of
financial stability. Inaugural at the International Seminar on Operationalizing tools for macro-
financial surveillance: country experiences, organized by the Financial Stability Unit (FSU).
Reserve Bank of India, Mumbai, 3 April 2012.
Commandeur,J. J. F. and Heiser,W. J.,;1; 1993. Mathematical derivations in the proximity
scaling (PROXSCAL) of symmetric data matrices. Tech, Rep, No RR-93-03, <PL>Leiden,
the Netherlands: department of Data Theory, Leiden</PL> University.
Danielsson, J., James, K.R., Valenzuela, M. and Zer, I.,;1; 2011. Model Risk of systemic risk
models'', working paper, first version of November 2011.
De Bandt, O., and Hartmann, P.,;1; 2000. Systemic Risk: A Survey. Working Paper 35,
European Central Bank.
Demirguc-Kunt, A., Kane, E. and Laeven, L,;1; 2009. Deposit Insurance Around the World:
issues of Design and Implementation. MIT Press, <PL>Cambridge Mass</PL>.
Forbes, J. and Rigobon, R.,;1; 2001. Measuring Contagion: conceptual and empirical issues.
in: Classens,S. and Forbes,K.(Eds.). International financial contagion, <PN>Kluwer
Academic Publishers</PN>, <PL>Boston</PL>, pp.43-66.
Girardi, G. and Tolga Ergün, A.,;1; 2013. Systemic risk measurement: Multivariate GARCH
estimation of CoVaR. Journal of Banking & Finance, Vol. 37 No 8, pp. 3169-3180.
Groenen, P. and Velden, M.,;1; 2004. Multi Dimensional Scaling. Econometric Institute
Report EI 2004-15.
Hakwa, B., Manfred, J.A. and Barbara, R.,;1; 2012. Measuring and Analysing Marginal
Systemic Risk Contribution using CoVaR: A Copula Approach. arXiv preprint
arXiv:1210.4713.
Hansen, L.P.,;1; 2013. Challenges in identifying and measuring systemic risk. University of
Chicago and the NBER.
International Monetary Fund and World Bank.;1; 2012. <PL>Tunisia Financial Sector
Assessment Program Report, Internal Mimeo, World Bank, Washington, DC</PL>.
Kaufman, G.,;1; 1995. Comment on Systemic Risk. Research in Financial Services: Banking,
Financial Markets, and Systemic Risk. Vol.7, pp. 47-52.
Kruskal, J.B. and Wish, M.,;1; 1978. Multidimensional scaling. <PL>Sage</PL> University
paper series on quantitative applications in the social sciences, 07-011, <PL>Beverly Hills,
California, Sage</PL>.
Le petit, J.F.,;1; 2010. Rapport sur le risque systémique. French Ministry of Economy,
Industry and Employment.

14
Mainik,G. and Shaanning, E.,;1; 2012. On dependence consistency of CoVaR and some other
systemic risk measures. arXiv: 1207.3464v3 [q-fin.RM] 29 Aug 2012.

Moussa, A.,;1; 2011. Contagion and Systemic Risk in Financial Networks. Ph.D. thesis,
<PL>Columbia </PL> <PN>University</PN>.

Pritsker, M.,;1; 2000. The channels for financial contagion. in:Classens,S. and
Forbes,K.(Eds.), International financial contagion, <PN>Kluwer Academic Publishers</PN>,
<PL>Boston</PL>.

Procopio, S., Krause,D., Hofmann,T and Becker,T.,;1; 2013. Significant amino acids in aroma
compound profiling during yeast fermentation analyzed by PLS regression. LWT-Food
Science and Technology, Vol. 51, No. 2, pp. 423–432.

Schwarcz, S.L.,;1; 2008. Systemic risk. Georgetown Law Journal, Vol. 97 No. 1, pp.193-249.

Torgerson, W.S.,;1; 1958. Theory and methods of scaling. <PL>New York: </PL>
<PN>Wiley</PN>.

Wold, S., Sjorstrom, M. and Eriksson, L.,;1; 2011. PLS-regression: a basic tool of
chemometrics. Chemometrics and Intelligent Laboratory Systems, Vol. 58 No. 2, pp.109–130.

</BIBL>
Figures Legend
<Figure>Figure 1: Subjects space
<Figure>Figure 2: Variables Space

Tables
<Table>Table 1: Returns Statistics Summary
Adjusted Std.
Observation Mean Skewness Kurtosis
mean Dev
BT 300 -0,027 0,7 4,091 -4,446 37,920
BIAT 300 0,164 0,14 2,935 0,160 3,632
UBCI 300 -0,102 -0,003 4,963 -4,981 59,819
ATTIJARI 300 0,246 0,133 3,126 -0,080 9,016
BH 300 -0,240 -0,3 3,228 0,382 3,302
UIB 300 -0,028 0,072 2,690 0,359 3,629
AB 300 0,039 0,17 5,069 -8,984 127,695
STB 300 -0,219 -0,23 3,929 -0,436 5,825
ATB 300 -0,092 -0,07 3,145 -1,099 12,269
BNA 300 -0,014 -0,055 3,497 -0,199 7,242
BTE 300 -0,075 -0,05 1,686 0,089 9,051

<Table>Table 2: Systemic risk interdependency table

15
 <remove
picture
𝑆𝑦𝑠
pageno BT BIAT UBCI ATTIJARI BH UIB AB STB ATB BNA BTE ∆𝐶𝑜𝐸𝑆𝑞,𝑡
21> j
i
BT 5,57 6,04 5,70 5,90 5,86 4,50 6,10 6,46 5,53 5,21 5,35
BIAT 5,87 5,82 5,10 6,14 3,09 6,69 5,96 6,83 5,80 6,00 5,33
UBCI 4,05 7,52 6,64 6,67 6,43 5,26 7,17 6,16 7,30 7,30 8,52
TIJARI 4,82 4,37 3,80 4,47 4,51 4,62 4,57 4,45 4,87 4,50 5,64
P= BH 7,28 5,17 5,96 5,46 5,47 6,53 6,52 5,50 6,42 5,52 6,17
UIB 5,07 4,60 4,47 5,10 5,75 4,25 5,25 4,68 5,18 3,22 4,69
AB 4,94 5,51 6,11 5,13 5,46 5,06 5,13 5,83 4,93 4,99 4,97
STB 6,97 7,18 6,64 6,06 7,23 5,78 6,38 6,71 7,28 6,51 7,09
ATB 5,48 5,89 4,76 5,29 5,36 4,95 5,34 5,20 5,49 4,54 5,25
BNA 6,95 6,09 6,33 6,36 5,95 5,98 5,61 7,44 6,89 5,43 6,91
BTE 3,42 3,37 3,63 2,91 3,41 3,24 2,92 3,45 3,22 3,51 3,40

∆𝐶𝑜𝐸𝑆𝑞𝑖 4,04 3,96 3,69 3,91 4,17 3,45 3,34 4,51 4,16 4,42 2,93

<Table>Table 3: Goodness-of-fit indices

Normalized Raw Stress 0,03671
Tucker’s congruence coefficient 0,98147

<Table>Table 4: Banks' Ranking

Banks STB BNA UBCI BH BIAT UIB BT ATB AB TIJARI BTE
MDS Implication 1 2 3 4 5 6 7 8 9 10 11
∆𝐶𝑜𝐸𝑆𝑞𝑖 1 2 8 3 6 9 5 4 10 7 11

𝑠𝑦𝑠
∆𝐶𝑜𝐸𝑆𝑞 2 3 1 4 7 10 6 8 9 5 11

<Table>Table 5: Spearman's rank correlation

𝑠𝑦𝑠
Ranking according to MDS Implication ∆𝐶𝑜𝐸𝑆𝑞𝑖 ∆𝐶𝑜𝐸𝑆𝑞

MDS Implication 1 0,836** 0,745**

∆𝐶𝑜𝐸𝑆𝑞𝑖 0,836** 1 0,655*

𝑠𝑦𝑠
∆𝐶𝑜𝐸𝑆𝑞 0,745** 0,655* 1

*. The correlation is significant at the 0.05 level (bilateral).

**. The correlation is significant at the 0.01 level (bilateral).

<Table>Table 6: SRICI' Descriptive Statistics

statistics Standard Error

Mean 0,2522 0,13004
lowerbound -0,0375
SRICI Confidence interval of 95% for the average
upperbound 0,5419
Trimmedmeanat 5% 0,2542

16
 Median 0,1592
Variance 0,186
standard deviation 0,43128
Minimum -0,53
Maximum 1,00
Interval 1,53
interquartile Interval 0,69
Skewness 0,071 0,661
Kurtosis 0,020 1,279

<Table>Table 7: correlation matrix

𝑠𝑦𝑠
SRICI ∆𝐶𝑜𝐸𝑆𝑞𝑖 ∆𝐶𝑜𝐸𝑆𝑞

SRICI 1 0,754** 0,899**

∆𝐶𝑜𝐸𝑆𝑞𝑖 0,754** 1

𝑠𝑦𝑠
∆𝐶𝑜𝐸𝑆𝑞 0,899** 0,597 1

**. The correlation is significant at the 0.01 level (bilateral).

<Table>Table 8: Factorial structure associated with the three measures

Dimension1
Loading Extraction
SRICI 0,972 0,945
∆𝐶𝑜𝐸𝑆𝑞𝑖 0,917 0,840
𝑠𝑦𝑠
∆𝐶𝑜𝐸𝑆𝑞 0,850 0,722
Information Restored 83,563%
Eigen value 2,507
α de Cronbach 0,727

<Table>Table 9: PLS results

F3
R2 = 0 ,969 F1 F2 F4
Interbank Beta
AdjustedR2 =0,939 Size Liquidity Technical efficiency
commitments
VIP 1,117 0,817 0,774 0,726
number of Branches Weight 0,322 0,029 -0,123 0,057 0,109
Loadings 0,356 -0,024 -0,215 -0,147
VIP 1,438 1,198 1,121 1,061
total assets Weight 0,415 0,245 0,116 -0,139 0,075
loadings 0,381 0,092 0,121 -0,247
VIP 1,56 1,472 1,478 1,399
share of credit Weight 0,45 0,394 0,437 0,183 0,454
loadings 0,366 0,108 0,29 -0,305
VIP 0,524 1,002 0,746 0,822
number of
Weight 0,151 -0,288 0,137 0,351 0,156
securitiestreated
loadings 0,279 0,478 0,116 -0,345
VIP 0,204 1,258 1,179 1,119
numbre of exchange 0,224
Weight 0,059 0,527 -0,133 -0,162

17
 loadings 0,263 0,608 0,172 -0,461
VIP 0,835 1,105 0,637 0,857
0,227
marketcapitalization Weight 0,241 -0,133 -0,042 0,497
loadings 0,324 0,636 -0,288 -0,233
VIP 0,473 0,544 1,126 1,079
interbankclaims Weight 0,136 -0,177 -0,775 -0,198 0,361
loadings 0,221 0,107 0,845 0,42
VIP 0,963 1,087 1,292 1,129
interbankdeposits
Weight 0,278 0,35 0,393 0,603 0,612
and assets
loadings 0,186 0,138 0,298 0,003
VIP 1,056 1,185 1,118 1,435
bank revenue Weight 0,414 0,234 0,162 0,126 0,289
loadings -0,131 0,057 0,03 0,384
VIP 1,05 0,952 0,883 1,248
operating expenses Weight 0,36 0,119 -0,012 -0,524 0,285
loadings 0,366 0,019 0,238 0,493
VIP 0,623 0,916 0,993 1,15
net earnings Weight 0,18 -0,335 -0,397 -0,548 0,643
loadings 0,329 -0,288 -0,048 0,382

TDENDOFDOCTD

<en><xps:span class="xps_label">1</xps:span>The institute of international finance

<en><xps:span class="xps_label">2</xps:span>According to IMF’s report on the

Assessment of the Tunisia Financial System (2012).

<en><xps:span class="xps_label">3</xps:span>In 2011, the Central Bank of Tunisia issued a circular allowing
banks to reschedule loans due from firmsaffected by the political turmoil of 2011.

𝒔𝒚𝒔
<en><xps:span class="xps_label">4</xps:span>All the details of the ∆𝑪𝒐𝑬𝑺𝒒,𝒕 and the
∆𝑪𝒐𝑬𝑺𝒊𝒒,𝒕 calculations are presented in the Appendix in page 16

18