Bank Stability and Systemic Risk Measurement: Application of Z-Score Variations to the Turkish Banking Sector

Abstract

The banking system’s role as an intermediary between depositors and borrowers makes its stability a crucial element of the economy. Individual bank stability impacts the stability of the financial sector and, consequently, the real sector and economic growth. This study is a detailed investigation of bank stability scores for the Turkish banking sector. Turkey experienced a major banking crisis from 2000 to 2001. However, with an improved regulatory environment and structural reforms, its resilience to shocks increased significantly in subsequent years. The present study contributes to the literature in three areas. First, it applies eight variations of the Z-score to the Turkish banking sector to test the importance of time and risk variation. Second, it investigates the impact on stability scores of several factors–the implementation of new capital regulations and Basel III liquidity rules, the Turkish lira crisis and the effects of COVID-19). Finally, it measures the contribution to systemic risk of each Turkish bank using the leave-one-out (LOO) Z-score method; this is the first study to use this method to analyse the contribution of different bank types to the overall stability of the sector.

Full text

INTERNATIONAL JOURNAL OF FINANCE & BANKING STUDIES 12 (1) (2023) 63-73
* Corresponding author.
© 2023 by the authors. Hosting by SSBFNET. Peer review under responsibility of Center for Strategic Studies in Business and Finance.
https://doi.org/10.20525/ijfbs.v12i1.2319
Bank Stability and Systemic Risk Measurement: Application of
Z-Score Variations to the Turkish Banking Sector
Coskun Tarkocin (a),(b)* , Murat Donduran (c)
(a) PhD Candidate in Economics, Graduate School of Social Sciences, Yildiz Technical University, Istanbul, Turkey.
(b) Head of Liquidity Optimisation, HSBC Group, London, United Kingdom.
(c) Director of Graduate School of Social Sciences, Yildiz Technical University, Istanbul, Turkey.
A R T I C L E I N F O
Article history:
Received 25 Jan. 2023
Received in rev. form 9 May 2023
Accepted 10 May 2023
Keywords:
Bank stability, financial stability, Z-
score, systemic risk, The Leave-one-
out score, Covid-19
JEL Classification:
C10, G01, G21, G33
A B S T R A C T
The banking system’s role as an intermediary between depositors and borrowers makes its stability a
crucial element of the economy. Individual bank stability impacts the stability of the financial sector
and, consequently, the real sector and economic growth. This study is a detailed investigation of bank
stability scores for the Turkish banking sector. Turkey experienced a major banking crisis from 2000
to 2001. However, with an improved regulatory environment and structural reforms, its resilience to
shocks increased significantly in subsequent years. The present study contributes to the literature in
three areas. First, it applies eight variations of the Z-score to the Turkish banking sector to test the
importance of time and risk variation. Second, it investigates the impact on stability scores of several
factors–the implementation of new capital regulations and Basel III liquidity rules, the Turkish lira
crisis and the effects of COVID-19). Finally, it measures the contribution to systemic risk of each
Turkish bank using the leave-one-out (LOO) Z-score method; this is the first study to use this method
to analyse the contribution of different bank types to the overall stability of the sector.
© 2023 by the authors. Licensee SSBFNET, Istanbul, Turkey. This article is an open access article
distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license
(http://creativecommons.org/licenses/by/4.0/).
Introduction
The banking sector has been a crucial component of the economy throughout its history. Banks are intermediaries between agents
with a surplus of funds and borrowers who require these funds for their consumption or investment plans. Banks provide their
customers with risk-management solutions, from so-called ‘vanilla’ products to complex derivatives, and they trade in financial
products and function as market-makers. A bank’s role is thus fundamental to the economic system. This unique position has
necessitated a highly regulated environment to ensure that banks are resilient to a variety of shocks. Such regulation aims to promote
financial stability and limit the cost to the economy of bank failure and, specifically, the cost to taxpayers. Within this context, bank
stability is an important concept for individual banks and stakeholders, such as regulators, investors and customers.
However, the existence of regulation and the unique position of banks in the economy does not prevent all bank failures. This is
evident in emerging markets like Turkey and in developed markets like the United States. At its peak in 1999, the Turkish banking
sector comprised 81 banks; by 2003, following the 2000–2001 banking crisis, the number of banks had fallen to 50, and the cost of
state bail-outs totalled $11.9 billion USD, excluding government support to state-owned banks (Banking Regulation and Supervision
Agency [BDDK], 2010). The sector has since experienced very few bank failures. By contrast, according to the United States Federal
Deposit Insurance Corporation, since 2001, 564 deposit institutions in the United States have failed. Of these, 414 of these failures
occurred between 2008 and 2011 in the aftermath of the 2007–2008 global financial crisis. In December 2010, to address the lessons
of these crises and to improve the resilience of the banking sector to shocks, the Basel Committee on Banking Supervision (BCBS)
issued the Basel III rules, which included improvements to capital standards and new liquidity measures (BCBS, 2010).
It became evident in the 2007–2008 global financial crisis that many institutions with a sufficient level of capital still faced liquidity
issues. When conditions rapidly deteriorated, central banks intervened to support the markets (BCBS, 2010). One important lesson
Finance & Banking Studies
IJFBS VOL 12 NO 1 ISSN: 2147-4486
Available online at www.ssbfnet.com
Journal homepage: https://www.ssbfnet.com/ojs/index.php/ijfbs

Tarkocin and Donduran, International Journal of Finance & Banking Studies 12(1) (2023), 63-73
64
from this crisis was that, in addition to strong capital requirements, there is a need for harmonised global liquidity standards and
robust regulatory supervision.
Understanding bank stability is paramount when identifying and managing sector vulnerabilities. And given the importance of banks
for the wider economy, ongoing research into bank stability and systemic risk is of the utmost importance. This study focuses on
individual bank stability and its relationship to systemic risk by comparing the stability of different banks within the same
environment.
This study has three main aims. The first is to measure variations of the Z-score and analyse its impact on the stability score of the
Turkish banking sector. The second is to investigate whether this score was impacted by important recent events, namely the
implementation of the Basel III liquidity coverage ratio, the Turkish lira crisis in August 2018 and the COVID-19 pandemic. The
third aim is to measure the contribution to systemic risk of each bank in terms of its impact on sector stability. The LOO Z-scores
are calculated following the approach in Li et al. (2020). To the best of our best knowledge, this study is the first to address the three
questions embedded in these aims with regard to the Turkish banking sector. This study will be the first to measure Z-score variations,
test the statistical significance of those and measure the LOO Z-score for the sector to analyse each bank’s contribution to systemic
risk.
This study is organised as follows. Section 2 contains a review of existing literature. The study data are outlined, and the data
transformation process is summarised in Section 3. The study’s methodology is outlined in Section 4. Section 5 sets out the empirical
results, and in Section 6, the conclusions and their policy implications are discussed.
Literature Review
The literature concerning bank stability and bank failure prediction is extensive due to banking’s direct linkage with and impact on
the wider economy, including the potential costs to taxpayers. Literature on bank stability can be grouped into two main categories.
The first category investigates factors impacting bank stability and the causes of bank insolvency. The second category aims to
measure bank stability and how it can be predicted. This study’s main aim relates to the intersection of these categories; therefore,
the most relevant selected literature will be discussed.
Theme 1: Factors Impacting Bank Stability or Causes of Bank Insolvency
One of the most frequently cited papers on the financial crisis is Demirguc-Kunt and Detragiache’s (1998) examination of the
determinants of the banking crisis. They investigated the likelihood of banking crises using a multivariate logit specification with
data from 1980–1994. They found that low GDP growth, a high rate of inflation and explicit deposit insurance schemes tended to
increase the probability of banking sector crises due to the moral hazard of banks taking excessive risk. Castro (2013) investigated
macroeconomic determinants of bank credit risk for GIPSI countries (Greece, Ireland, Portugal, Spain and Italy) over the period of
1997–2011 by employing dynamic panel data. Castro found that a decrease in GDP growth, share price indices and house prices, as
well as increases in the unemployment rate, interest rate and credit growth, resulted in higher credit risk in the banking system.
Imbierowicz and Rauch (2014) investigated the two major sources of bank default risk using a sample of all US commercial banks
during the period from 1998–2010. They found that both liquidity risk and credit risk had an impact on the probability of bank default.
Out of 254 bank defaults during the 2007–2010 period, 118 were due either exclusively to liquidity loss or to a combination of loan
and liquidity loss. An important conclusion from the study was the need for joint management of liquidity risk and credit risk in
banks.
Petropoulos et al. (2020) predicted bank insolvencies using machine learning techniques. They reported that earnings and capital
metrics have the highest marginal contribution to the prediction of bank failure, and they also demonstrated that the random forests
(RF) method has superior out-of-sample prediction performance. Gogas et al. (2018) proposed a forecasting model for bank failures
using US banks that failed during the 2007–2013 period. They employed the support vector machine (SVM) model and reported high
prediction power using four of the most relevant variables that they filtered in two steps from 144 original variables in the dataset.
Theme 2: Z-score Applications as a Bank Stability Measure
There has been extensive research into the relationship between the Z-score and other commonly used measures. For instance,
Demirguc-Kunt et al. (2006) found that Z-scores were positively correlated with Moody’s ratings, and this correlation was statistically
significant. Beck et al. (2009) included the Z-score as an indicator of banking stability and found a weak correlation between
profitability and stability. The study also reported lower Z-scores in the years leading up to 2007, which could be explained by lower
capital or the higher volatility of returns.
Chiaramonte et al. (2015) examined whether the Z-score was a reliable metric for predicting bank distress in 12 European countries.
They found that using the natural logarithm of the Z-score identified banks in distress with strong predictive power. It is significant
that, in this study, the Z-score performed as well as CAMELS variables but with less data.
Li et al. (2017) compared different approaches to measuring the Z-Score and empirically applied their methods to New Zealand
banks. They also provided detailed information regarding Z-Score calculations. The study proposed a risk-adjusted Z-score in which

Tarkocin and Donduran, International Journal of Finance & Banking Studies 12(1) (2023), 63-73
65
total assets are replaced by risk-weighted assets (RWA), and equity is replaced with the tier 1 capital ratio. Li et al. (2020) proposed
leave-one-out (LOO) as a new systemic risk measure and used it to identify the largest New Zealand banks as systemically important.
Lepetit and Strobel (2013) examined an alternative time-varying Z-score by calculating the mean and standard deviation of the return
on assets (ROA) over full samples with the current value of the capital–asset ratio. This alternative Z-score measure provided
satisfactory results overall and displayed a low level of intertemporal volatility at the bank level.
Ekinci and Erdal (2017) compared machine learning models’ forecasting performance for the Turkish banks’ failure during 1997–
2000. Using financial ratios as part of the CAMELS framework, they reported that hybrid ensemble learning models outperformed
other models.
Yuksel (2017) analysed the determinants of credit risk using Turkish banking sector data from 2004 to 2014 using the non-performing
loans (NPL) ratio as a dependent variable. Using a binary probit model with several internal and external variables, the industrial
production index was found to impact the NPL ratio with statistical significance. Danisman (2018) explored the determinants of bank
stability using Turkish bank data from 2007–2015. The study employed GMM estimation techniques and found that return on assets,
loans to asset ratio, inefficiency index, non-interest income share and loans loss provision share can all explain NPLs.
Aksoy and Donduran (2019) found a positive relationship between competition and bank stability using data for the Turkish banking
sector from 2006–2016. Alihodzic et al. (2020) investigated the determinants of bank stability for Turkey and six other Balkan
countries using the Z-score and NPL as dependent variables. This study reported that GDP, cost to income ratio, the net interest
margin and the Lerner index had the highest correlation with the Z-score. Kasman and Kasman (2016) reported that bank size is
positively related to the Z-score, implying that larger banks are less risky.
Data
The data that support the findings of this study are available in the Turkish Banking Association website’s statistical reports section
at https://www.tbb.org.tr/en/banks-and-banking-sector-information/statistics-and-data-query/statistical-reports/20 (Turkish Banking
Association [TBB], 2021).
Publicly available data from March 2003 to September 2020 (71 quarters in total) was used in this study. This time period was chosen
because it provides homogenous, long-term, stable data following the 2000–2001 banking crisis.
The largest number of banks since 1959 was 81 banks, in 1999. Following the 2000–2001 banking crisis, the number of banks
decreased to 50 by 2003 due to several bankruptcies. As of December 2020, there were 48 banks. Of these, three were state-owned
banks, nine were private local banks, 21 were banks owned by foreign investors, 14 were investment and development banks and
one was a bank under the control of the Savings Deposit Insurance Fund (SDIF).
Historical data on the banking sector’s asset composition can be seen in Figure 1. In the last two decades, the proportion of loans
increased from 30% to 62%, while the proportion of financial assets decreased from around 40% to below 20%. This data supports
the idea that the banking sector transformed to fund the real sector instead of holding a significant amount of government securities
with a high yield. This figure depicts the long-term transformation of the Turkish banking sector when it moved to a more traditional
intermediary role, with loans constituting a large portion of the total balance sheet assets.

Tarkocin and Donduran, International Journal of Finance & Banking Studies 12(1) (2023), 63-73
66
FIGURE 1: HISTORICAL COMPOSITION OF THE BALANCE SHEET COMPONENTS FOR THE TURKISH BANKING SECTOR
Source: Turkish Banking Association (TBB, 2021)
The Turkish banking sector is dominated by few large banks. Concentration of the top 10 banks is 87% while the top 20 represents
96% of total assets. The three state banks’ assets comprise 40% of the banking sector.
In this study, only banks with a licence to receive deposits were used; therefore, investment and development banks were excluded.
Investment and development banks accounted for 6.9% of total assets as of September 2020, 3.7% of which is owned by Turkish
Eximbank (Export Credit Bank of Turkey, owned by the Ministry of Treasury). Therefore, the exclusion of investment and
development banks did not eliminate a significant percentage of the banking sector from the analysis.
Methodology
The Z-score is a common measure of stability. It can be used for individual institutions, or it can be aggregated to measure the stability
of the whole banking system. It measures a bank’s distance to default by comparing its capitalisation and return to the volatility of
returns. The Z-score has a negative correlation with bank insolvency risk. A higher Z-score indicates a more stable bank and thus a
lower probability of default (TheWorldBank, 2021). The Z-score is used by the World Bank to monitor banking sector stability across
countries. Detailed statistics are then shared as part of the Global Financial Development Report (TheWorldBank, 2019).
The Z-score has several limitations. These include being completely reliant on accounting data, which may not be available frequently
and may have data quality issues; failing to capture the linkage with other market participants, such as contagion risk; and failing to
capture other risk types that may exist within an institution’s portfolio, such as large FX risk and concentration risk. Conversely, the
main advantages of the Z-score are that it is simple to use; it can give a reliable indication of bank stability when banks from the
same environment are compared; it can be used to measure the stability of institutions not publicly listed or not rated by external
credit ratings; and lastly, it allows for comparison of different groups of institutions based on characteristics such as size and
ownership characteristics (Cihak et al., 2012).
In this study, different variations of the Z-score will be applied to the Turkish banking sector. First, the risk based and the standard
Z-score will be calculated, then time variation will be applied by using different numbers of quarters to calculate the volatility of
return on assets and return on risk-weighted assets (RORWA). Once all measurements are performed, the statistical significance of
the variations will be tested. Following standard practice from the literature, the natural logarithm of the calculated Z-scores will be
used throughout this analysis to reduce the impact of outliers.
Li et al. (2017) extended the traditional Z-score by replacing balance sheet assets with RWAs to measure risk-adjusted Z-score and
empirically testing this measure using data from New Zealand and Australian banks. Li et al. (2020) defined the LOO Z-score by
calculating the aggregate Z-score for all the banks under consideration and then measuring the contribution of a single bank by re-
calculating the aggregate Z-score after omitting one bank from the aggregation. The present study applies the same approach to the
Turkish banking sector and analyses its impact at the bank ownership and size level. Li (2018) reported a long list of Z-score variations
used in the literature.

Tarkocin and Donduran, International Journal of Finance & Banking Studies 12(1) (2023), 63-73
67
Z-score Measure
The Z-score can be calculated as Z=(k+µ)/σ. In this formula k is equity, µ represents return as a percentage of total assets and σ is
standard deviation of ROA. A lower Z-score implies a higher probability of insolvency. Z-score simply measures how many standard
deviations are required to exhaust the equity (Cihak et al., 2012).
The standard Z-score formula for a bank is calculated as follows:
      󰇛
 󰇜
󰇛󰇜
TABLE 1: LIST OF VARIABLES IN Z-SCORE
Acronym
Name
Explanation
ROA
Return on Assets
Return divided by Total Assets
σ(ROA)t
Standard deviation of ROA
Last t quarters’ ROA used to calculated standard deviation
Risk-Based Z-Score Measure
Standard Z-score is risk sensitive, but it may fail to capture risks that have not yet resulted in any volatility impact on returns. Using
RWAs will add operational risk and market risk components into the calculation while also capturing risk due to off-balance-sheet
assets.
      󰇛
 󰇜
󰇛󰇜
TABLE 2: ADDITIONAL VARIABLES IN RISK-BASED Z-SCORE
Acronym
Name
Explanation
RORWA
Return on Risk-Weighted Assets
Net profit divided by Total Risk-Weighted Assets
σ(RORWA)t
Standard deviation of ROA
Last t quarters’ RORWA used to calculate standard deviation
Leave-One-Out (LOO) Z-Score
Two steps approach is followed to calculate the LOO score. The first step involves measuring the aggregated Z-score for all banks
included in the study. The second step involves calculating the aggregated Z-score for all banks while omitting one bank each time
the calculation is performed; the impact of removing that specific bank will indicate its contribution to the systemic risk of the banking
sector. When applied over time, this measure can provide important insights about changing dynamics within the sector.
    󰇛󰇜  󰇛
 󰇜
󰇛󰇜
TABLE 3: LIST OF RAW DATA AND TRANSFORMATION
Source Data From TBA
Name in the Z-score Calculation
Net Profit and Loss After Tax
Return
Total Assets
Assets
Total Equity
Equity
Capital Adequacy Ratio
Equity/RWA
Risk-Weighted Assets
RWA
Statistical Tests to Compare Results
Statistical tests in MATLAB (2021a version) will be used to compare two or more groups and test whether different variations of the
Z-score are statistically significant. These tests will also be used to check whether significant movement occurred between different
periods. For instance, they will be used to check variation before and after the implementation of new regulations, as well as before
and after the Turkish lira crisis and the Covid-19 crisis. However, it should be noted that this will only demonstrate the statistical
significance of any change, not necessarily whether the event in question caused the change. Understanding the determinants of any
movement that may occur during a given period will be an aspect of future studies. The tests performed, their description and the
MATLAB function that was used are summarized in Table 4.

Tarkocin and Donduran, International Journal of Finance & Banking Studies 12(1) (2023), 63-73
68
TABLE 4: STATISTICAL TESTS AND DESCRIPTIONS
Test Name
Description
Null Hypothesis
MATLAB
Function
Two-Sample Kolmogorov–
Smirnov Test
Testing whether two groups come
from the same continuous
distribution
The data in vectors x1 and x2 are
from the same continuous distribution
kstest2
Wilcoxon Rank Sum Test
Testing whether two independent
groups with equal medians come
from the same continuous
distribution
The data in vectors are from
continuous distributions with equal
medians.
ranksum
Two-Sample t-Test
Testing whether two independent
groups with equal variances have
the same mean
The data used in the vectors is
randomly selected with normal
distributions, with equal means, equal
but unknown variances.
ttest2
Source: MATLAB Documentation (MathWorks, 2021)
Empirical Results
Z-score Variations Results and Analysis
Standard Z-Score Time Variations Comparison
FIGURE 2: COMPARISON OF STANDARD LN_Z-SCORES ACROSS DIFFERENT TIME VARIATIONS
The statistical results for all combinations are summarized in Table 5. The P-values indicate that, for all combinations except 16Q vs
12Q, the results are different, with a statistical significance of 5%. The student’s t-tests generated a P-value of less than 5%, on the
basis of which we reject the null hypothesis of two samples from distributions with equal means. Furthermore, 16Q and 12Q can be
used interchangeably since, at the 1% level of significance, the results of all tests allow us to reject the null hypothesis. This shows
that time variation in the Z-score calculation can result in statistically different stability levels. Therefore, for comparability with the
results of other studies, the same time variation needs to be applied.

Tarkocin and Donduran, International Journal of Finance & Banking Studies 12(1) (2023), 63-73
69
TABLE 5: STATISTICAL TEST RESULTS COMPARING Z-SCORE TIME VARIATIONS
P-values
Ttest2
KStest2
Ranksum
28Q vs 16Q
0.01%
0.08%
0.01%
28Q vs 12Q
0.00%
0.03%
0.00%
28Q vs 8Q
0.00%
0.00%
0.00%
16Q vs 12Q
6.62%
3.38%
7.08%
16Q vs 8Q
0.01%
0.00%
0.03%
12Q vs 8Q
1.17%
1.76%
1.19%
Standard Z-score vs Risk-Based Z-Score Comparison
The standard and risk-based Z-score results are compared using the results for the 16Q variation. One reason for this is that the
difference between standard and risk-based Z-scores is not significant. Assets in Turkish banks attract a higher risk weight than assets
in developed markets. Specifically, when an asset attracts a 100% risk weight, total assets and total RWAs become highly correlated.
Nevertheless, the gap between the two Z-score measurements appeared to widen after June 2019. Similar tests could be performed
in the future to determine whether the deviation is temporary or a sign of structural changes.
FIGURE 3: COMPARISON OF STANDARD AND RISK-BASED LN_Z-SCORES
None of the statistical tests applied allows us to reject the null hypothesis at the 5% significance level. These results indicate that,
statistically, the two samples have the same mean, median and distribution. However, the KStest2 refutes the null hypothesis at the
10% significance level, suggesting a slight difference in the distribution. Although this is the case for the banks in Turkey, there may
be a greater deviation in countries using advanced models for the RWA calculations.
TABLE 6: STATISTICAL TEST RESULTS BETWEEN RISK-BASED Z-SCORE TIME VARIATIONS
P-values
Ttest2
KStest2
Ranksum
Standard Ln_Z 16Q vs risk-based Ln_Z 16Q
83.00%
6.19%
80.55%
The LOO Z-score Calculation Results and Analysis
The LOO Z-score calculation measures individual bank contribution to the aggregated bank stability measure. Impact analysis was
completed based on a standard Z-score with 16Q time variation for the LOO Z-score measurement. The LOO Z-score results based
on 2020 September data are summarized in Table 8.
The findings largely match the expected results; that is, as a bank’s size decreases, its impact on the aggregated sector stability score
also decreases. However, this study also provides an interesting insight that may be specific to the Turkish banking industry: when
individual state banks are excluded, the total system stability score increases. One potential reason for this is that, for state banks, the
average standard deviation of return on assets (ROA) was significantly higher than that of other banks for the period 2019 Q1 to

Tarkocin and Donduran, International Journal of Finance & Banking Studies 12(1) (2023), 63-73
70
2020 Q3 compared to the period from 2017 Q1 to 2018 Q4. Historical movements analysis indicates that this directional change is
evident from the beginning of 2019.
TABLE 7: THE LOO Z-SCORE AND INDIVIDUAL BANK CONTRIBUTION (2020 Q3)
Size
Ownership
Bank Name
The LOO
Z Score
Contribution to
Aggregated Ln_Z
score
% Contribution to
Aggregated Ln_Z
score
Top5
State
State Bank 1
3,70
0,12
3,4%
Top5
Private_Local
Private Local 1
3,54
-0,04
-1,0%
Top5
State
Sate Bank 2
3,73
0,15
4,3%
Top5
State
State Bank 3
3,66
0,08
2,2%
Top5
Private_Local
Private Local 2
3,51
-0,06
-1,8%
Top5-10
Private_Local
Private Local 3
3,51
-0,07
-1,9%
Top5-10
Private_Local
Private Local 4
3,53
-0,05
-1,3%
Top5-10
Private_Foreign
Private Foreign 1
3,54
-0,03
-0,9%
Top5-10
Private_Foreign
Private Foreign 2
3,58
0,00
0,0%
Top5-10
Private_Local
Private Local 5
3,56
-0,01
-0,4%
Top10-20
Private_Foreign
Private Foreign 3
3,54
-0,04
-1,1%
Top10-20
Private_Foreign
Private Foreign 4
3,56
-0,02
-0,4%
Top10-20
Private_Local
Private Local 6
3,58
0,01
0,2%
Top10-20
Private_Foreign
Private Foreign 5
3,57
0,00
-0,1%
Top10-20
Private_Local
Private Local 7
3,57
-0,01
-0,2%
Top10-20
Private_Foreign
Private Foreign 6
3,57
0,00
-0,1%
Top10-20
Private_Foreign
Private Foreign 7
3,58
0,00
0,0%
Top10-20
Private_Local
Private Local 8
3,57
-0,01
-0,3%
Aggreagated Ln_Z Score (pre LOO)
3,58
An important finding of this study is that the impact of state-owned banks is currently the opposite of other banks; excluding these
results in a higher level of stability in the banking sector, based on Z-scores. Further studies could focus on understanding the
underlying reasons for this effect and how it can be mitigated to improve sector stability. In absolute terms, state banks have the
highest impact on the aggregated stability measure. In terms of ownership, the next major contributor is privately owned local banks,
which dominate the market in terms of market share.
TABLE 8: THE LOO Z-SCORE IMPACT BY BANK OWNERSHIP TYPE (2020 Q3)
Ownership
% Contribution to Aggregated Z-score
State
9.9%
Private_Local
-6.8%
Private_Foreign
-2.7%
A Closer Look at Bank Stability based on Z-score
Closer Look 1: Major Banking Regulation Changes
The period from March 2016 to September 2020 was analyzed to investigate why the difference between standard and risk-based Z-
score measurements changed with statistical significance after June 2019. On 23 October 2015, the regulatory authority in Turkey,
the Banking Regulation and Supervision Agency (BRSA), published new regulations regarding the calculation of RWA that came
into effect on 31 March 2016. The Liquidity Coverage Ratio became effective on 1 January 2017 (BDDK, 2021). In addition, Turkey
Financial Reporting Standards 9 – financial instruments (TFRS 9) came into effect on 1 January 2018.
Since this change in the regulatory environment may have impacted the RWA standard Z-score vs the risk-based Z-score, tests were
performed again for March 2016 to September 2020. In contrast to the results for the full time frame used in Section 5.1, for this
specific period, all statistical tests reject the null hypothesis at the 5% significance level. These results indicate that the difference
between the standard Z-score and risk-based Z-score does not originate from the same continuous distribution, and the mean and
median are unequal.
TABLE 9: STATISTICAL TEST RESULTS BETWEEN STANDARD AND RISK-BASED Z-SCORE (16Q) MARCH 2016 TO SEPTEMBER 2020
P-values
Ttest2
KStest2
Ranksum
Standard ln_Z 16Q vs risk-based ln_Z 16Q
0.50%
1.81%
1.20%
Closer Look 2: Turkish Lira Crisis in 2018 and Covid-19 Crisis 2020
An initial look at the results Z-score results since 2015 indicate that the aggregated Z-score was at the highest level in 2017. Since
then, it has been decreasing, a trend that accelerated in 2020. Another insight is that state bank stability scores decreased during this
period. In 2020, state banks’ stability scores were at the lowest levels since 2015.

Tarkocin and Donduran, International Journal of Finance & Banking Studies 12(1) (2023), 63-73
71
TABLE 10: THE STANDARD Z-SCORE (16Q) 2015–2020
Bank Name
2020
2019
2018
2017
2016
2015
State Bank 1
2,98
3,12
3,77
4,04
4,18
4,42
Private Local 1
3,74
4,05
4,01
4,16
3,83
3,66
Sate Bank 2
2,56
2,62
3,23
3,80
3,17
3,13
State Bank 3
2,83
2,94
3,37
3,35
3,68
3,50
Private Local 2
3,94
4,25
3,82
3,99
3,82
3,62
Private Local 3
4,16
4,26
3,88
3,91
3,13
3,07
Private Local 4
3,98
4,40
3,90
3,93
3,78
3,76
Private Foreign 1
4,08
3,92
3,78
3,82
3,51
3,34
Private Foreign 2
3,52
3,57
3,58
3,49
3,46
2,39
Private Local 5
3,98
3,88
4,05
3,98
3,95
4,23
Private Foreign 3
3,23
3,10
3,02
3,03
3,45
3,02
Private Foreign 4
2,16
2,25
2,37
2,72
2,65
2,51
Private Local 6
1,92
1,58
4,26
3,71
3,20
3,02
Private Foreign 5
3,34
2,61
2,49
2,61
2,36
3,19
Private Local 7
4,04
4,15
4,41
4,27
4,15
3,73
Private Foreign 6
3,22
3,75
3,72
3,33
2,92
2,53
Private Foreign 7
2,55
2,85
2,41
2,58
2,28
2,37
Private Local 8
3,06
3,44
3,63
3,59
3,24
2,95
Aggregated_sector Z-score
3,58
3,81
3,95
3,97
3,82
3,63
The Turkish lira experienced a major foreign exchange shock in 2018. On 13 August, the lira depreciated 23% against the USD in
one day, and cumulative depreciation over the previous 12-month period reached 97% on 14 August
1
. According to the results
presented in Table 11, all major banks’ stability scores either decreased or remained constant in 2018.
The impact of the Covid-19 pandemic is also visible. In 2020, 18 out of 14 banks’ stability scores decreased, and the aggregated
stability scores decreased by the most significant amount in the decade. Although since 2017 there has been a general downwards
trend, it must be noted that following the 2001–2002 financial crisis, the Turkish banking sector was well capitalised and regulated.
Structural reforms and a developed regulatory environment increased bank resilience to a variety of shocks. In addition, most of the
poorly managed banks failed in 2001–2002, so it can be inferred that the remaining banks were better diversified and employed
effective risk management practices. Therefore, it is likely that the drop in stability scores in 2020 can be ascribed to the impact of
the Covid-19 pandemic. Figure 4 depicts a decreasing trend from 2017 and an increasing standard deviation between individual banks
since its peak level in 2017.
FIGURE 4: COMPARISON OF STANDARD AND RISK-BASED LN_Z-SCORES
1
Yahoo Finance USDTRY=X historical data.

Tarkocin and Donduran, International Journal of Finance & Banking Studies 12(1) (2023), 63-73
72
Conclusion
This paper empirically tested the impact of different Z-score measurements on stability scores for the Turkish banking industry and
tested the statistical significance of several Z-score variations. The risk-based and standard Z-scores were the main variations, with
four different time variations applied for each. The statistical tests show the Z-score results were distinct for all time variations, with
the exception of 12Q vs 16Q. This result suggests that, when comparing studies, the same time variation needs to be used. Standard
deviation, skewness, and range statistics were lowest for the 16Q variation of the standard Z-score; therefore, this variation was
preferred for the detailed analysis and the calculation of the LOO Z-score.
We found, using a full sample period, that the standard or risk-based Z-scores did not produce statistically different results for banks
in Turkey. However, for the period from March 2016 to September 2020, the standard and risk-based Z-scores were statistically
different. The gap between the two also widened after June 2019. This suggests that as more data points are collected in future,
similar tests can be performed to confirm whether this is a temporary or structural deviation.
This study also applied the LOO Z-score measurement to the Turkish banking sector for the first time. The LOO Z-score was used
to measure each bank’s systemic risk contribution, yielding insight into sector dynamics and how these have evolved. As expected,
the results suggest that as bank size decreases, systemic risk contribution decreases. However, there is one critical finding worth
noting: when state banks are excluded, the aggregated Z-score for the banking sector improves. Analysing Z-score components
indicates that from 2019, ROA volatility was significantly higher for state banks than for other banks. This is an important insight
for the central banker, and further investigating and understanding the underlying reasons for this could improve overall sector
stability. Another finding is that, in absolute terms, after state banks, private local (rather than foreign-owned) banks have the second
highest impact on systemic risk. This is in line with the fact that private local banks currently dominate the market.
Banks and policymakers can benefit from using the LOO Z-score metric to monitor banks’ contribution to systemic risk. A sudden
deterioration in LOO Z-scores can provide an early warning of sector risk since interlinkages between financial institutions can
amplify systemic risk.
A closer look at the bank stability results indicates that the aggregate Z-score was highest in 2017, but there has since been a
decreasing trend, for which the main contributors are the state-owned banks. The score continued to decrease following the 2018
Turkish lira crisis. The impact of the Covid-19 pandemic was visible in the 2020 stability scores of 14 banks. As banks report more
data, the full impact of the pandemic on the financial stability of the sector can be measured.
Acknowledgement
The views and opinions expressed in this paper are those of the authors and they do not necessarily reflect the views of the HSBC
Group or the Yildiz Technical University.
Author Contributions: Conceptualization, C.T., M.D.; Methodology, C.T., M.D.; Data Collection, C.T.; Formal Analysis, C.T.; Writing—Original
Draft Preparation, C.T.; Writing—Review And Editing, C.T., M.D.; All authors have read and agreed to the published the final version of the
manuscript.
Institutional Review Board Statement: Ethical review and approval were waived for this study, due to that the research does not deal with
vulnerable groups or sensitive issues.
Data Availability Statement: The data presented in this study are available on request from the corresponding author. The data are not publicly
available due to privacy.
Conflicts of Interest: The authors declare no conflict of interest.
References
Aksoy, T., & Donduran, M. (2019). Competition and stability in Turkish banking. Finans Politik & Ekonomik Yorumlar, 57(651),
133–147.
Alihodzic, A., Eksi, I. H., & Dogan, B. (2020). The determinants of bank stability: Evidence from selected Balkan countries and
Turkey. International Review, (1-2), 94–105.
Basel Committee on Banking Supervision [BCBS]. (2010, December). Basel III: A global regulatory framework for more resilient
banks and banking systems. https://www.bis.org/publ/bcbs189_dec2010.pdf
Banking Regulation and Supervision Agency [BDDK]. (2010). Krizden Istikrara Turkiye Tecrubesi (Calisma Tebligi).
BDDK. (2021, April 22). Regulations. https://www.bddk.org.tr/Legislation-Category/Communiqu%C3%A9s/33
Beck, T., Demirguc-Kunt, A., & Levine, R. (2009). Financial institutions and markets across countries and over time – data and
analysis (Policy research working paper no. WPS 4943). Washington, DC, World Bank Group.
Castro, V. (2013). Macroeconomic determinants of the credit risk in the banking system: The case of the GIPSI. Economic Modelling,
31, 672–683. https://doi.org/10.1016/j.econmod.2013.01.027
Chiaramonte, L., Croci, E., & Poli, F. (2015). Should we trust the Z-score? Evidence from the European banking industry. Global
Finance Journal, 28, 111–131. https://doi.org/10.1016/j.gfj.2015.02.002
Cihak, M., Demirguc-Kunt, A., Feyen, E., & Levine, R. (2012). Benchmarking financial systems around the world (English, Policy
research working paper no. WPS 6175). Washington, DC, World Bank Group.
Danisman, G. O. (2018). Determinants of bank stability: A financial statement analysis. Sosyoekonomi 26(38), 87–103.
https://doi.org/10.17233/sosyoekonomi.2018.04.06

Tarkocin and Donduran, International Journal of Finance & Banking Studies 12(1) (2023), 63-73
73
Demirguc-Kunt, A., & Detragiache, E. (1998). The determinants of banking crises in developing and developed countries. IMF Staff
papers, 45(1), 81-109. https://doi.org/10.2307/3867330
Demirgüç-Kunt, A., Detragiache, E., & Tressel, T. (2006). Banking on principles: Compliance with Basel Core Principles and bank
soundness (Policy research working paper no. 3954). World Bank.
Ekinci, A., & Erdal, H. I. (2017). Forecasting bank failure: Base learners, ensembles and hybrid ensembles. Computational
Economics, 49(104), 677–686. https://doi.org/10.1007/s10614-016-9623-y
Gogas, P., Papadimitriou, T., & Agrapetidou, A. (2018). Forecasting bank failures and stress testing: A machine learning approach.
International Journal of Forecasting, 34(3), 440–455. https://doi.org/10.1016/j.ijforecast.2018.01.009
Imbierowicz, B., & Rauch, C. (2014). The relationship between liquidity risk and credit risk in banks. Journal of Banking & Finance,
40, 242–256. https://doi.org/10.1016/j.jbankfin.2013.11.030
Kasman, A., & Kasman, S. (2016). Bank size, competition and risk in the Turkish banking industry. Empirica, 43, 607–631.
https://doi.org/10.1007/s10663-015-9307-1
Lepetit, L., & Strobel, F. (2013). Bank insolvency risk and time-varying Z-score measures. Journal of International Financial Markets,
Institutions & Money, 25, 73–87. https://doi.org/10.1016/j.intfin.2013.01.004
Li, X. (2018). An examination of bank risk measures and their relationship to systemic risk measurement [Unpublished doctoral
dissertation]. Massey University, Manawatu (Turitea).
Li, X., Tripe, D. W., & Malone, C. B. (2017, January 20). Measuring bank risk: An exploration of Z-score.
http://dx.doi.org/10.2139/ssrn.2823946
Li, X., Tripe, D., Malone, C., & Smith, D. (2020). Measuring systemic risk contribution: The leave-one-out z-score method. Finance
Research Letters, 36, 101306. https://doi.org/10.1016/j.frl.2019.101316
MathWorks. (2021, April 18). MATLAB documentation statistics and machine learning toolbox.
https://uk.mathworks.com/help/stats/available-hypothesis-tests.html
Petropoulos, A., Siakoulis, V., Stabroulakis, E., & Vlachogiannakis, N. E. (2020). Predicting bank insolvencies using machine
learning. International Journal of Forecasting, 36(3), 1092–1113. https://doi.org/10.1016/j.ijforecast.2019.11.005
Turkish Banking Association [TBB]. (2021). https://www.tbb.org.tr/
TheWorldBank. (2019). Global financial development report. Washington DC: World Bank Group.
TheWorldBank. (2021, April 17). Global financial development report 2016 backround: Financial stability.
https://www.worldbank.org/en/publication/gfdr/gfdr-2016/background/financial-stability
Yuksel, S. (2017). Determinants of the credit risk in developing countries after economic crisis: A case of Turkish banking sector. In
Ü. Hacioğlu, & H. Dinçer (Eds.), Global financial crisis and its ramifications on capital markets (pp. 401–415). Springer.
Publisher’s Note: SSBFNET stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.
© 2023 by the authors. Licensee SSBFNET, Istanbul, Turkey. This article is an open access article distributed under the terms and conditions of the
Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
International Journal of Finance & Banking Studies (2147-4486) by SSBFNET is licensed under a Creative Commons Attribution 4.0 International
License.