Does higher competition lead to lower probability of license withdrawal in Russian banking sector?

Размещено на http://www.allbest.ru/

Abstract

The purpose of the study is to identify whether there is a connection between competition in banking sector and banks' license withdrawal, and to discover type of this connection if it exists in the sample (whether it is positive, negative or non - linear). The sample includes data about Russian commercial banks for the period from July 2013 to January 2020. The research shows that default probability depends on competition on the loan market non - linearly, inversely U-shaped, and that applying concentration ratio CR5 of loans gives a better model than applying other competition measures considered. The result implies, firstly, that when banking system is close to monopoly, no more banks' licenses are withdrawn, secondly, that high competition in banking provides banks with low default probability. Also, the analysis shows that Russian banking sector has become more concentrated, therefore, less competitive, and that government's involvement has increased in Russian banking system.

Introduction

Advantages of competition are well known: higher quality of products and services, lower prices, more efficient processes, lower probability of collusion, etc. Whereas monopoly may lead to lack of differentiation and high prices. And there is a high risk of collusion in case of oligopoly. However, high competition in banking may have a drawback. In condition of high competition banks start to follow risk - taking behavior that results in financial instability.

On the one hand, monopolies might abuse their powers and economic agents will suffer. On the other hand, economic agents probably will suffer in case of excessive freedom in banking. Thus, free - banking laws that were introduced in 1836 to reduce power of large banks resulted in wildcat banking (issuing notes without adequate security) and unsound banking practices. During this period, also known as Free Banking Era, banks were free of federal regulation and it was easy to start a bank. A lot of banks' failures with large losses followed this period.

The main research question of this work is whether higher competition leads to lower probability of license withdrawal in Russian banking sector. Here, this question relates to the situation in Russian Federation since 2013. This period is under consideration as Russian banking industry since then has been experiencing new tighter regulation under new administration. Therefore, object of the research is Russian banking sector.

The purpose of this paper is to identify whether there is a connection between banks' license withdrawal and competition in this sector, and if there is a relation, the purpose includes discovering type of this connection (whether it is positive, negative or non - linear).

To achieve the goal stated above, it is necessary to set the following objectives:

To prepare a literature review on the topic;

To consider different types of competition and types of banks;

To state hypotheses of the study;

To collect data on banks' parameters and macroeconomic indexes since 2013;

To build several measurements of competition;

To construct appropriate models to test stated hypotheses;

To compare model results, to analyze which competition indicator shows better results; license bank competition

To make a conclusion.

The topic of the study is relevant as Russian banking sector is changing, specifically, quantity of commercial banks is reducing nowadays. Thus, there were 894 http://www.cbr.ru/statistics/bank_system_new/inform_13/ commercial banks in Russian Federation by 01.07.2013 and by 01.01.2020 there were 402 http://www.cbr.ru/statistics/bank_sector/lic/ (that is 55% reduction), 6 of them are new banks that were registered in 2014, 2015 and 2019 years. Banking regulation has toughened, and it is important to understand whether it has led to lower competition and whether its influence on soundness and stability of financial system in Russia is positive. Knowing these makes clearer what should be done in the future and what should not, what corrections should be made in current legislation.

Novelty of the paper is that the analysis is conducted for a recent period - from 2013 to 2020. Also, competition is explored in various groups of banks. Thus, “Small” cluster and “Medium” cluster of banks are characterized as highly competitive.

The sample for modeling includes information about 378 Russian commercial banks from July 2013 to January 2020. To construct time series of competition measures, values of assets, loans and deposits of every Russian bank from July 2013 to January 2020 are collected. Then, CR5, CR10, CR20, HHIs and HT indices of assets, loans and deposits are calculated. According to these parameters concentration has increased, Russian banking sector has become less competitive. Afterwards, binary choice models are built for the sample. The chosen model shows that there is a non-linear dependence between competition and probability of license withdrawal and this dependence is inversely U-shaped.

Chapter 1. Literature review and theoretical part

Literature review

Stability of a banking sector is an essential issue as development of economy and society depends on continuous and proper functioning of banking system. And stability of a banking sector is based on soundness of each separate bank. That is why, a lot of studies and researches are devoted to the problem of bankruptcy of financial organizations, and Russian banking system is also of interest to many authors. It is found that lots of factors are significant in determining default probability of a bank, for example, such factors as capital adequacy, liquidity, macroeconomic situation, etc. And level of competition might be one of the determining factors. Thus, a number of papers are focused on competition or concentration in banking as an aspect of financial stability.

Impact of banking sector on economy

In many studies growing importance of banking system and its interrelation with soundness of economy are highlighted. In a document of Basel Committee on Banking Supervision “Basel III: A global regulatory framework for more resilient banks and banking systems” [2] the resilience of banking is considered as a foundation for sustainable economic growth, and off-balanced financial system may cause economic and financial crises. Authors Monnin P. and Jokipii T. (2010) in their article studied whether banking sector stability is related with the real output growth or inflation. They found that when banking sector is stable, generally, an increase in real output growth follows, while instability is followed by subsequent periods of reduced growth. Moreover, they found that stable banking system reduces uncertainty in the economy. The results are supported by other researches. Hence, Gulzar A. (2018) in the article emphasizes that the better banking sector is, the more available capital is, and it leads to more investments that play a crucial role in the economy. Also, it is noted in the article that recently financial sector has become even more important. Lanets S. (2015) highlights that banking sector plays an important role in the socioeconomic development in any country and for Russia it is also true, and nowadays Russian banks face difficulties because of sanctions from other influential countries.

Existing competition measures and their classification

There are several approaches in understanding what competition is and how to measure it, specifically in banking. Leon F. (2014) considers two views of competition: competition as a static state and competition as a process. According to the first view, competition is a static state in which companies cannot charge more to earn extra profit. The second vision of competition defines competition as complex process of rivalry between businesses that forces them to behave in accordance with the purpose to cope with this rivalry. Also, the author highlights two paradigms to measure competition: structural and non - structural. The structural approach measures competition through characteristics of the market: number of banks, their absolute and relative size, entry and exit conditions, etc. This paradigm includes concentration ratios (CRk), the Herfindahl - Hirschman Index (HHI), the Hall - Tideman index, the U - index, the CCI, etc. Whereas non - structural method focuses on the conduct of banks. Thus, the Lerner index, the conjectural variation model, the Boone indicator, the Panzar - Rosse model were developed to observe conduct directly.

This classification of measurement approaches partly corresponds with the categorization in the thesis by Mamonov M. [Мамонов М.Е.] (2014). He distinguishes two methods: direct and indirect. Direct approach assesses the market power using the Lerner index. Indirect measurement is divided to structural and non - structural. Structural estimation considers concentration, while non - structural evaluation reflects banks' income sensitivity. The author in his thesis applies the Lerner index, the Boone index and the Panzar - Rosse model to measure competition.

Fungacova Z. et al. (2010; 2014) in their articles and Clark et al. (2018) apply the Lerner index for the analyses.

Fu X. et al. (2014) in addition to the Lerner index apply the efficiency - adjusted Lerner index and CR3.

Phan H.T. et al. (2019) in their article consider different modification of the Lerner index: the conventional Lerner index, the efficiency - adjusted Lerner index, the funding - adjusted Lerner index. The conventional Lerner index, the efficiency - adjusted Lerner index are also studied by Phan H.T.M. and Kevin Daly (2020).

Saif-Alyousfi A.Y.H. et al. (2018) in the research investigate CR5, HHI, the Lerner index and Boone indicator as a measure for competition. The Boone indicator is also considered by Glass A.J. et al. (2020) in their paper.

Stenbacka R., Takalo T. (2016) in their paper measure deposit market competition as switching costs for a depositor. The higher these switching costs are, the higher market power is, and the lower competition is.

Impact of competition

Competition in banking sector influences access to financial services and products, quality of these services and products and financial stability. And competition has an ambiguous effect. On the one hand, as many authors state (Fungacova Z. et al. (2010); Fu X. et al. (2014); Clark E. et al. (2018); Hasan I. and Marinи M. (2013)) there is a positive impact of competition in banking on economic growth, stability and efficiency; that the higher competition is, the lower loan rates are, the more available capital is. On the other hand, there is a trade - off between quantity and quality. Some researchers (Fu X. et al. (2014); Phan H.T.M., Daly K. (2020); Saif-Alyousfi A.Y.H. et al. (2018); Phan H.T. et al. (2019)) indicate that lower loan rates lead to higher risks for banks; competition induces inefficiencies and risk - taking behaviour and it stimulates destabilization; tougher restrictions might benefit financial stability.

Mamonov M. [Мамонов М.Е.] (2014) in his thesis says about nonlinear dependence between market power and stability of a bank. Thus, we cannot claim that competition in bank sector has a positive or negative influence on fragility because the relation is rather U - shaped. Some other authors also come to conclusion that the effect is controversial, depending on indicator for competition applied or on measure for stability.

License withdrawal modeling

Many authors use binary choice specifications to model probability that a bank will go bankrupt or its license will be withdrawn. Thus, authors Peresetsky A. et al. (2011) in their article use logit specification separately for each cluster. Clusters are defined as ``small'', ``medium'' and ``large'' for the chosen parameters: total assets, government bonds ratio, credits-to-non-financial-firms ratio, equity ratio. Logistic regressions are also applied by Tong X. (2015) to predict defaults over yearly horizons from one to five years. There is a model for each horizon of forecasting.

Grammatikos T. and Papanikolaou N.I. (2018) estimate failure probability using linear probability OLS regression and dividing their sample into three clusters. Clusters are formed according to banks' size reflected by total assets. Anginer D. et al. (2018) also apply linear probability OLS regression to identify what influences distance to default. As well, linear equation is implemented by Nguyen D.T. and Nguyen H. (2018) in their paper. Also, they apply two - step generalized method of moments (GMM) technique to understand what Z - score is sensitive to, where Z - score reflects insolvency probability. The authors find out that GMM outperforms OLS approach. GMM and OLS are also considered by Mamonov M. [Мамонов М.Е.] (2014). In his thesis Z - score and overdue loans ratio reflect banks' stability.

Calabrese R. and Giudici P. (2015) in their article compare logistic regression with GEV (generalised extreme values) regression models. The authors conclude that GEV regressions are better at predicting bank defaults because these models assign more importance to default cases as they are rarer.

Gуmez-Fernбndez-Aguado P. et al. (2018) in their article calculate the individual default probabilities based on SYstemic Model of Bank Originated Losses (SYMBOL model). The authors generate individual banks' losses via Monte Carlo simulation. Monte Carlo simulations are also implemented by Montesi G. and Papiro G. (2018). They consider a bank to be in default if value of its capital falls beneath a pre-set threshold or if its equity value is less than zero. Discounted cash flow (DCF) model is applied for each simulated scenario generated to estimate possible future values of equity. There, probability of default is the frequency of scenarios where the value of equity is less than zero.

Default determinants

Besides competition there are factors that also might affect probability of bank's default. There are researches that show that some factors as capital adequacy or liquidity have a significant impact on default probability of a bank. These factors can be divided into several groups: financial, macroeconomic, institutional, political and time. This distribution is presented in Table 1.

Table 1. Factors that might influence default probability of a bank

Financial	Macroeconomic	Institutional	Political
Total assets Liquidity Capital Equity Capital assets and non-working assets Earning assets Deposits and accounts Reserves Profit Loans Non-performing loans Non - current loans Credits from other banks Non-government securities Government bonds ROA, ROE Liability Turnover on correspondent accounts Sensitivity to market risk	GDP growth rate Inflation Export to import ratio Currency rate	Competition Existence of branches Region Management quality Ownership Bank participation in a Deposit insurance system Level of involvement in cross-country transactions Business model complexity Political connections	Periods of election

Financial factors according to the authors Peresetsky A. et al. (2011); Karminsky A. et al. (2012); Tong X. (2015); Grammatikos T. and Papanikolaou N.I. (2018); Gуmez-Fernбndez-Aguado P. et al. (2018) include logarithm of bank's total assets as a measure of its size, capitalization as a proportion of bank's capital to its total assets, balance profit or profit before tax to total assets ratio, non-government securities to total assets ratio, reserves for possible losses, government bonds, equity, liquid assets and liquidity, private customers' deposits and accounts, capital assets and other non-working assets, non-performing loans to total loans, loans to non-financial institutions, credits from other banks, non - current loans to total loans, net loans to bank capital equity, non-working assets, ROA (return on assets) and ROE (return on equity), liability to assets, earning assets to total assets, risk-weighted assets to total assets ratio, leverage ratio, net stable funding ratio, sensitivity to market risk.

As many studies show macroeconomic situation plays a crucial role in soundness of banks. GDP growth rate, inflation, export to import ratio, currency rate are considered as macroeconomic measurements by Gуmez-Fernбndez-Aguado P. et al. (2018) and by Karminsky A.M. [Карминский А.М.] (2015) in the book “Credit ratings and their modeling”.

Additionally to financial and macroeconomic factors, institutional parameters are implemented in the literature. According to Karminsky A.M. et al. (2012), Karminsky [Карминский А.М.] A.M. (2015), Grammatikos T. and Papanikolaou N.I. (2018), these include region where a bank is headquartered or operates; foreign or state ownership; bank participation in a Deposit insurance system; management quality; level of bank's involvement in cross-country transactions; business model complexity, political connections, etc.

Moreover, as it is noted by Jackowicz K. et al. (2013), performance of state - owned banks differs during elections, therefore, it is fair to state that political factors also take place in banking activity.

To detect trends time is included in many analyses.

A lot of different parameters in various combinations are considered in the existing literature on this topic. It shows that financial sector stability is of interest to many researchers and that understanding of this issue varies.

Theoretical part

Banks operate on various markets. For example, it includes credit market and deposit market. Therefore, competition should be measured for these spheres separately as it is made in some researches. In this work different competition indicators are applied with respect to loan portfolio, total deposits and banks' total assets. Level of competition probably varies dramatically across these financial parameters. CRk, HHI and Hall - Tideman index (HT) are implemented here. These indicators reflect structural concentration.

Also, competition may take place in different groups of banks. These groups are formed according to the existing types of banks or their size. Type of ownership is one of parameters for bank classification. In Russia there are state - owned banks, foreign - owned ones, private and public banks. Another classification parameter - branch network diversity (Belousova V. et al. (2018)). Banks can be grouped according to their number of branches. Bank's location also has an impact. Hence, organizations can be classified by territory they are headquartered in. Moreover, clusters might be formed by banks' size (Peresetsky A. et al. (2011)). And competition should be measured separately for a group or a cluster.

For purposes of this study and to answer the main research question the hypotheses are stated below.

Hypothesis 1: Russian banking sector has become more concentered since 2013. According to Vernikov A.V. [Верников А.В.] (2013), Russian banking sector had become less competitive by 2012 year as state - owned banks had strengthened their positions on the markets of financial services. Here, it is considered if this trend to less competitiveness described by Vernikov A.V. [Верников А.В.] (2013) continues to exist. The analysis is conducted for another period - from 2013 to 2020 year. Criterion applied to identify if concentration has increased in banking sector is the CRs, the HHIs and HT indices. The higher values of these ratios and indices imply higher concentration and, therefore, lower competition.

Hypothesis 2: There is a non-linear dependence between competition and probability of license withdrawal. As mentioned in the literature there is an ambiguous effect from competition on company's stability. This controversial impact can be explained by non-linear dependence. Mamonov M. [Мамонов М.Е.] (2014) tests for a non-linear dependence between competition and bank's stability and between competition and banks' overdue loans. The aim here is to conduct a similar test for a different period. To test if hypothesis 2 is true for the sample of this study, coefficients before competition parameters and squared competition parameters must be statistically significant.

Hypothesis 3: Dependence between competition and probability of license withdrawal is U-shaped. Mamonov M. [Мамонов М.Е.] (2014) in his thesis concludes that banking system is the most unstable while level of competition is the highest and lowest. His analysis indicates inversely U-shaped dependence between competition and bank's stability. Here, if coefficients before squared competition parameters are positive and statistically significant, hypothesis 3 is confirmed.

Bank failure is defined differently in the existing literature. In this research a bank is counted as default if its license was withdrawn or if this bank was merged or acquired or if it was sanitated.

Chapter 2. Methodology and data

Methodology

Competition measures

Firstly, time series of competition measurements are constructed. The results are presented in the appendix, Table A.1. Quarterly data from 01.07.2013 to 01.01.2020 on total assets, total loans and total deposits of every operating bank are collected. Afterwards, share of every operating commercial bank for each of the tree parameters is calculated as value of individual value divided to the overall one. For concentration ratios CR5, CR10, CR20 shares are ordered from the highest to the lowest and sum of respectively top 5, 10 and 20 shares are taken (formula 1). For the Herfindahl - Hirschman indices (HHI) each bank's share is squared and sum of these squared shares is the HHI (formula 2). Shares must be ranked to calculate the Hall - Tideman index (HT). Share and its rank are multiplied, then sum of these products is doubled and 1 is subtracted, this product in power of -1 is the HT index (formula 3).

Formula 1. Concentration ratios CRk:

(1),

where k is a number of organizations that occupy largest k positions on a market, and here, k = 5, 10, 20; is n bank's share on market i; i = a (assets), l (loans), d (deposits); n is an index for banks' shares that are sorted from the highest share to the lowest one.

Formula 2. The Herfindahl - Hirschman index (HHI):

(2),

where is n bank's share on market i; i = a (assets), l (loans), d (deposits); N is quantity of functioning commercial banks in Russia.

Formula 3. The Hall - Tideman index (HT):

(3),

where is n bank's share on market i; i = a (assets), l (loans), d (deposits); n is a rank for banks' shares that are ordered from the highest to the lowest share; N is a number of functioning commercial banks in Russia.

Concentration ratios show what share several largest firms on a market hold. If this share is lower than 0.5, it implies highly competitive conditions. CR value between 0.5 and 0.6 indicates moderate concentration. CR higher than 0.6 might suggest oligopoly. However, this measurement does not take into account share distribution between banks and number and influence of smaller institutions.

According to the literature if HHI is lower than 0.15, it indicates highly competitive unconcentrated market. If HHI is higher than 0.15 but lower than 0.25, it shows moderate concentration. HHI above 0.25 refers to highly concentrated market, where HHI of value 1 defines monopoly.

The Hall - Tideman index, in contrast to HHI, focuses more on number of firms and smaller ones. It is more sensitive to banks that hold the lowest shares. Russian banking sector is characterized by several big commercial banks and many small ones. That is why HT index should be considered here and compered with HHI.

Groups of banks

Secondly, different groups of banks and competition in these groups are considered.

Commercial banks are divided into groups according to their ownership structure, there are state-owned banks and banks that are not controlled by the government. In this research a bank is considered to be a state-owned one if state institutions directly own more than 50% of this bank or if the government owns more than 50% of this bank through companies that are controlled by the state. State-owned banks differ from the other ones by the fact that the government can realize its plans and interests through state-owned banks, and it leads to cheaper resources for these banks and protection from the government. Afterwards, competition measurements are calculated for these two groups separately: CR5, CR10, CR20, HHI, HT.

Also, banks are divided here into clusters according to their size. Logarithm of assets is taken as a measure for size. Tree categories are created: “big” banks, “medium” and “small” ones. K-means approach with respect to logarithm of assets is implemented in this work to build clusters for every year from 2014 to 2020. Concentration in these groups is calculated as CR5, CR10, CR20, HHI, HT. Competition may exist in groups by size because depending on its size a bank focuses on different purposes and competes with similar financial institutions.

Default probability modeling

Thirdly, to analyze questions of this paper default probability is modeled with binary choice models: probit model and logit model. Here, a bank is default if its license was withdrawn or if this bank was merged or acquired or if it was sanitated.

Dependent variable is probability of bank's license withdrawal. Factors that are considered as regressors are presented in Table 2. In the left column of Table 2 there are factors that are of interest here, and in the right column there are regressors that reflect corresponding factors and can be used for modeling.



Table 2. Research variables

Factor	Variables
Time	Dummy variable for each period
Size of a bank	Logarithm of total assets
Credit portfolio	Total loans to total assets ratio
Deposits	Total deposits to total assets ratio
Bank's performance	Net profit to total assets ratio
Risks accepted by a bank	Reserves for possible losses to total assets ratio
Quality of loan portfolio	Non-performing loans to total loans
Level of bank's access to financial resources	Net interbank credits (received loans minus granted credit)
Bank's capital adequacy	Capital to total assets ratio (N1 or N1.0)
Bank's liquidity	Liquid assets to current liabilities (N2 and N3)
Bank's efficiency	ROA, ROE
Macroeconomic factors	GDP growth rate, inflation growth, currency rate
Bank's ownership type and location	State or non-state-owned bank; region where a bank is headquartered
Institutional factors	Region, where a bank is headquartered; ownership; systematically important or not; bank's participation in deposit insurance system
Competition	Concentration ratios, HHI, HT indices

Logarithm of total assets is in analysis as a measure of banks' size. To apply logarithm, values must be positive, therefore, if bank's assets are negative or zero, this bank is be counted as default. It corresponds with current legislation and common sense. As banks are different in terms of size and activities, ratios are considered in the research instead of absolute values. Some parameters are rationed by either total assets or total loans.

As indicators of performance and efficiency net profit to total assets, ROA and ROE are implemented. Reserves appear as an approximate measure of taken risks; non-performing loans to total loans indicate quality of loan portfolio. Such factors as high taken risks and poor loan portfolio quality impact banks' activity negatively and threaten their stability. As a result, these indicators may affect default probability.

Five dummy variables are included in the research: state or non-state-owned bank; whether a bank is headquartered in Moscow and Moscow region or not; whether a bank is headquartered in Saint Petersburg and its region or not; whether a bank participates in the deposit insurance system and whether a bank is systematically important in Russia. State - owned banks usually do not go bankrupt as they are likely to have government support in case of difficulties. Region where a bank is headquartered might play a role in its ability to operate and be efficient. Thus, business activity in Moscow and in Saint Petersburg is higher than in other regions of Russia and it may create more favorable conditions for banks. Bank's participation in the deposit insurance system may positively influence bank's resilience as individual clients' trust depends on this factor. Systematically important organizations are unlikely to be liquidated because even in case of distress the government will not let such companies go bankrupt.

Here, a commercial bank is considered as a state - owned one as mentioned before if more than 50% of this bank is owned by state institutions or by companies that are controlled by the government. Information about ownership status, status in the deposit insurance system and systematically important status is collected from different sources as banki.ru, the Central Bank's database and Deposit Insurance Agency.

Macroeconomic situation in the country also influences banks' conditions. For example, if economy is expanding, bank services are in more demand and there is more trust in banks, and vice versa. The GDP growth in Russia is moderate (less than 2,5% per year) since 2013 and even negative in 2015 https://www.gks.ru/accounts according to The Federal State Statistic Service. Therefore, macroeconomic conditions are not very favorable for banks recent years. Quarterly GDP growth is implemented in this work. Moreover, some international relationships may have impact on banks' operations. Thus, Russian international relationships have worsened and as a result some banks suffer from foreign restrictions and sanctions.

In this paper probit binary and logit binary models are applied with different combinations of factors to identify a better model and a better measurement for competition for collected data.

Logit model is constructed as following:

. (4)

where zit is a sum of multiplications of regressors to coefficients, plus constant; i is a bank's number in the sample, t is an index for the period.

Probit model equation is:

. (5)

where zit is the same as in Logit model.

In this work it is checked what lags of what regressors are improving the model because generally banks do not go bankrupt instantly, usually their situation is worthening gradually before license withdrawal.

First, a logit model with first lags of parameters is build. Then, logit models with first lags with fixed and random effects are constructed. Logit and probit specifications with and without robust estimator of variance are compered. Different competition measures are compered in models to identify which one performs better. Afterwards, quadratic effect of competition is estimated to understand if there is a U-shaped dependence.

Models are compared with such parameters as AIC, BIC, LL, pseudo R2, Chi2, significance of the parameters, sensitivity, specificity.

Graphs and models are build using a spreadsheet Microsoft Excel and statistical software package Stata.

Data overview

To calculate competition parameters CR5, CR10, CR20, HHI, HT of assets, loans and deposits, the information about banks' total assets, loan portfolio and total deposits is collected from the database of the Central Bank of the Russian Federation interpreted by the service banki.ru.

The data for modeling consist of quarterly information about 378 commercial banks from the Russian Federation from July 2013 to January 2020. Financial data, norm values and regions where banks are headquartered are taken from the site of the Central Bank of the Russian Federation and from the database of the Central Bank of the Russian Federation interpreted by the service banki.ru. Information about banks that participate in deposit insurance system and about banks that have been sanitated is provided by the Agency of deposit insurance. Dollar currency rate is also taken from the database of the Central Bank. Statistics about quarterly GDP growth and inflation change is found on RF Federal State Statistics Service.

After financial parameters are normalized to assets and non-performing loans to total loans variables for modeling are formed and listed in Table 3.

Table 3. Description of the research variables

Variable name	Variable description
id	Bank's license number
time	This variable identifies periods
lnA	Logarithm of total assets
Lo	Total loans to assets
De	Total deposits to assets
P	Net profit to total assets
R	Bank's total reserves for probable losses to total assets
B	Non-performing loans to total loans
I	Net interbank loans to total assets
N1	Norm N1 or N1.0 that shows capital adequacy, %
N2	Norm N2 that shows bank's instant liquidity, %
N3	Norm N3 that shows bank's current liquidity, %
ROA	Return on assets, %
ROE	Return on equity, %
GDPg	Quarterly GDP growth in the RF, %
INFg	Quarterly inflation growth in the RF, %
Cur	US Dollar exchange rate, in rubles
regM	Region where a bank is headquartered, if it is headquartered in Moscow or in Moscow region, regM=1, otherwise, regM=0
regSP	Region where a bank is headquartered, if it is headquartered in Saint Petersburg or in its region, regSP=1, otherwise, regSP=0
State	If it is state-owned bank state=1, otherwise, state=0
SI	If a bank is included in the list of systematically important banks of Russia, SI=1, otherwise, SI=0
DIS	If a bank participates in the deposit insurance system, DIS=1, otherwise, DIS=0
CR5_a	The concentration ratio of assets for top 5 banks, share
CR10_a	The concentration ratio of assets for top 10 banks, share
CR20_a	The concentration ratio of assets for top 20 banks, share
HHI_a	The Herfindahl - Hirschman index of assets, share
HT_a	The Hall - Tideman index of assets, share
CR5_l	The concentration ratio of loans for top 5 banks, share
CR10_l	The concentration ratio of loans for top 10 banks, share
CR20_l	The concentration ratio of loans for top 20 banks, share
HHI_l	The Herfindahl - Hirschman index of loans, share
HT_l	The Hall - Tideman index of loans, share
CR5_d	The concentration ratio of deposits for top 5 banks, share
CR10_d	The concentration ratio of deposits for top 10 banks, share
CR20_d	The concentration ratio of deposits for top 20 banks, share
HHI_d	The Herfindahl - Hirschman index of deposits, share
HT_d	The Hall - Tideman index of deposits, share
D	This variable identifies if a bank is default in the period

Here, Dit is a dependent variable. It equals 1 if a bank i is default at period t, otherwise it equals 0.

Norm N1.0 used to be named N1 till 2014. The regulator changed its name in April 2014 because other norms of capital adequacy were introduced.

Banks with abnormally high or low values of ROA, ROE, N1, N2, N3 are removed from the sample. Also, banks with missing data are excluded from the sample.

Descriptive statistics of banks' parameters is presented in Table 4.



Table 4. Descriptive statistics of banks parameters

Variable	Obs	Mean	Std. Dev.	Min	Max
D	10,206	.2933568	.4553229	0	1
lnA	7 184	16.396	1.813	12.382	23.401
Lo	7 184	.498	.216	.0001	.971
De	7 184	.498	.216	.0001	.971
P	7 184	.007	.025	-.3465	.362
R	7 184	.084	.081	5.25e-07	.738
B	7 185	.065	.100	0	1
I	7 184	-.051	.221	-.935	.853
N1	10 206	17.889	19.802	0	192.952
N2	10 206	84.958	138.556	0	3 402.06
N3	10 206	118.772	169.355	0	5 444.87
ROA	10 206	.834	3.899	-96.16	74.56
ROE	10 206	4.279	17.062	-257.76	245.17
regM	10 206	.540	.498	0	1
regSP	10 206	.050	.219	0	1
state	10 206	.049	.215	0	1
SI	10 206	.0169	.129	0	1
DIS	10 206	.793	.405	0	1

At period 01.01.2020 there are 50% of default banks and 50% of operating banks in the sample.

Chapter 3

Results

Competition measurements in banking sector

Concentration ratios CR5 of assets, loans and deposits are presented on the figure below (Figure 1). From the graph and Table A.1 it can be seen that five largest commercial banks in Russia have owned more than a half of each mentioned markets (assets, credits and deposits) since 2013 and their share has been growing since then. According to the concentration ratio CR5 there is a strong trend of more concentrated financial markets, therefore, banking sector is becoming less competitive. It includes loan and deposit markets. An abrupt increase of top five banks' share took place in the first quarter of 2018. It happened because 2 organizations that were ones of the largest banks in Russia were merged in January 2018 (VTB 24 and VTB). Their assets, loans and deposits were joined. It led to a rise of the concentration ratios. Also, it can be noted that loans and deposits are more concentrated than banks' assets.

Figure 1. It represents CR5 of assets (green line), loans (orange line) and deposits (blue line).

CR10 and CR20 show similar results and trends as CR5. They are presented on Figure A.1 and Figure A.2. The largest banks in Russia have strengthened their positions on the market during the last five years. Top twenty banks own over 80% of the total assets in banking sector since 01.04.2018.

Dynamics of the Herfindahl - Hirschman indices from 01.07.2013 to 01.01.2020 is illustrated on Figure 2. HHI of assets identifies highly competitive market during the whole period observed. However, the graph also shows that loan and deposit markets turned from unconcentrated to moderately concentrated after 01.04.2018 as HHIs are higher than 0.15 for those periods. The results correspond with the results of the concentration ratios. Russian banking system is becoming less competitive and credit and deposit markets are more concentrated than the market of assets.

Figure 2. Time series of HHI are shown on Figure 2. HHI of assets is a green line, of loans is an orange line and of deposits is a blue line.

The Hall - Tideman indices are presented on Figure 3. In general, HT indices results coincide with the results of the previous competition measurements. HT indices are gradually increasing over the chosen period. As HT is sensitive to the number of operating banks, it shows less concentration on the markets than HHI and CRk, but also, it shows a greater change in competition for Russian banking sector. It is so because quantity of commercial banks has reduced by 492 organizations from 894.

Figure 3. Time series of HT indices (HT index of assets is a green line, HT index of loans is an orange line and HT index of deposits is a blue line) are illustrated on Figure 3.

Results of all considered competition measures correspond with each other. Several conclusions can be made from the graphs:

There is stable trend of more concentration on the mentioned markets. Number of operating banks is reducing while largest banks' share is increasing.

Deposit and loan markets are more concentrated than banks' assets. It can be explained by the fact that some commercial banks do not have license to hold deposits and some banks are not interested in giving loans.

These results coincide with conclusions made by Vernikov A.V. [Верников А.В.] (2013). Therefore, banking sector has continued becoming less competitive.

Competition in different groups and clusters

The first group considered is state-owned banks. Share of state-owned banks' assets, credits and deposits in total Russian banking sector has been increasing since 01.07.2013. Share of state-owned banks' assets has increased from 59% to 72% over the period (Figure 4). It indicates that government's control over the Russian banking sector has widened.

Figure 4. It presents share of banks in Russian banking sector that are controlled by the government or by the companies that are owned by the government (share in total assets, loans and deposits).

Competition between state-owned banks is quite low. It is a highly concentrated group, HHIs are higher than 0.25 (Table A.2). Dynamics of HHIs is presented on Figure 5. It shows that competition here has slightly improved over the period.

Figure 5. Here, HHIs of state-owned group are illustrated (HHI of assets is a green line, HHI of loans is an orange line and HHI of deposits is a blue line).

CR5 of state-owned group of assets, loans and deposits indicates oligopoly markets as these concentration ratios are higher than 0.6 (Table A.2).

Competition in the group of banks that are not controlled by the state is high. This group is characterized as highly competitive according to the competition measures (Table A.3). However, the competition measurements dynamics shows that the markets have gradually become more concentrated.

Figure 6. CR5 of non-state-owned group of assets, loans and deposits are presented here (CR5 of assets is a green line, CR5 of loans is an orange line and CR5 of deposits is a blue line).

Figure 7. It illustrates HHIs of the group of banks that are not controlled by the state (HHI of assets is a green line, HHI of loans is an orange line and HHI of deposits is a blue line).

These graphs (Figure 4; Figure 7) confirm a theory expressed by Vernikov A.V. [Верников А.В.] (2013) about “national champions” in Russian financial system. Positions of these `national champions” have only strengthened since 2013, government involvement in this sector has increased.

Another banks grouping is by their size that is expressed by assets. K-means approach with 3 clusters gives here 3 groups for every year from 2014 to 2020 with from 69 to 439 banks in a cluster. The first cluster “Big” consists of the largest banks in Russia (from 69 to 166 banks in a group depending on a year), the second cluster “Medium” contains midsize financial organizations (from 158 to 439 banks) and the third cluster “Small” includes the smallest banks (from 191 to 299 companies). Values of competition measures in these three categories are presented in Table A.4, Table A.5, Table A.6. Category “Big” is the most concentrated group over the whole period, it is characterized as moderate concentrated on loan and deposit markets, while clusters “Medium” and “Small” are highly competitive. And as it can be seen from Figure A.3, Figure A.4 and Figure A.5 in all three clusters markets have become less competitive over the period.

To sum up, it can be concluded that hypothesis 1 is true. Russian banking sector has become more concentered since 2013.

Results of default probability modeling

To measure competition influence on default probability several models are constructed. First, different panel logit models with first lags are build. Results are presented in Table 5.

Table 5. Different logit specifications

	Dependent variable - Default probability
lag=1quarter	Pooled logit	Logit with fixed effects	Logit with random effects
Logarithm of assets	-.569***	2.292***	-.569***
Profit/ assets	-5.802	21.136***	-5.802
Reserves/ assets	2.449***	44.740***	2.449***
Non-performing loans/ loans	-1.206	-4.396	-1.206
Net interbank loans/ assets	.875**	.416	.875**
DIS	-.654***		-.654***
State-owned	-1.599		-1.599
Headquartered in Moscow	-.267*		-.267*
GDP growth	-.057	-.858***	-.057
ROA	-.021	-.031	-.021
N1 (capital adequacy)	-.027***	.025	-.027***
N3 (liquidity)	.0006*	.001	.0006*
HHI asset	-21.722	468.163***	-21.723
HHI loan	7.785	534.638***	7.786
HHI deposit	24.983*	90.857**	24.984*
constant	4.392***		4.392***
# of observation	6 996	2 100	6 996



Table 5 continued

	Pooled logit	Logit with fixed effects	Logit with random effects
Pseudo_R2	.112	.715
Chi2	194.741	582.417	159.917
P-value	.000	.000	.000
AIC	1 575.195	255.887	1 577.196
BIC	1 684.845	323.683	1 693.698
ll	-771.598	-115.943	-771.598
ll0	-868.968	-407.152
	legend: * p<.1; ** p<.05; *** p<.01

According to the information criteria, specification with fixed effects is better than the other two models.

Afterwards, adding second and third lags to the model makes it better according to the information criteria AIC and BIC and to the pseudo-. Also, because some parameters are highly correlated, they are excluded from the models to avoid multicollinearity. Results are in Table 6.

Table 6. Logit models with fixed effects with different lags

Dependent variable - Default probability
	First lags	Second lags	Third lags
Logarithm of assets (lag=1 quarter)	2.292***
Profit/ assets (lag=1 quarter)	21.136***	23.470***	30.496***
Reserves/ assets (lag=1 quarter)	44.740***	37.181***	42.168***
lag=2 quarters		16.319**	18.561
Non-performing loans/ loans (lag=1 quarter)	-4.396
Net interbank loans/ assets (lag=1 quarter)	.416
GDP growth (lag=1 quarter)	-.858***	.449	-.201
lag=2 quarters		-1.774***	-.352
lag=3 quarters			-1.678***
ROA (lag=1 quarter)	-.0307
N1 (lag=1 quarter)	.025	-.0274	-.041
lag=2 quarters		.080**	.118***
N3 (lag=1 quarter)	.001
HHI asset (lag=1 quarter)	468.163***
HHI loan (lag=1 quarter)	534.638***	701.557***	685.174***
lag=2 quarters		674.256***	702.530***
lag=3 quarters			696.543***
HHI deposit (lag=1 quarter)	90.857**
# of observation	2 100	1 909	1 731

Table 6 continued

	First lags	Second lags	Third lags
Pseudo_R2	.715	.802	.873
Chi2	582.417	607.267	614.963
P-value	.000	.000	.000
AIC	255.887	168.203	111.547
BIC	323.683	218.192	171.568
ll	-115.943	-75.102	-44.774
ll0	-407.152	-378.735	-352.255
		legend: * p<.1; ** p<.05; *** p<.01

The information criteria AIC and BIC show that adding third lags of some variables better the model.

Then, other specifications are tested. Pooled probit model, pooled probit model with robust estimator of variance, probit model with random effects are compared with chosen previously logit specification. Parameters show that logit specification performs better (Table A.7).

Subsequently, different competition measures with their quadratic effects are compared. Here, HHIs are multiplied by 100 000. Models are presented in Table A.8. U-shaped dependence is confirmed in models with CR5 of assets, CR5 of loans and HHI of loans as competition measures. Corresponding coefficients are significant at 95% level of confidence. According to the information criteria AIC and BIC, model where CR5 of loans is a measure for competition is better than the other models in Table A.8. In this model coefficients before profit to assets ratio and reserves for probable losses to assets ratio of the previous quarter and capital adequacy two quarters before are significant at 90% level of confidence and positive. Also, this model shows that concentration influences default probability non-linearly, its quadratic effect is statistically significant at 95% level of confidence. Results of the model are presented below in Table 7. CR5 of loans is multiplied by 100 000 for Table 7.

Table 7. Results of the model (logit with fixed effects specification) with CR5 of loans as a measure for competition

Dependent variable - Default probability
Variable	Odd ratios	Std.	P>z
Profit/Assets (lag=1quarter)	5.30e+23**	1.29e+25	0.025
Reserves/ assets (lag=1quarter)	3.76e+30***	9.83e+31	0.007
lag=2 quarters	3.170	74.623	0.961
N1 (lag=1quarter)	.871	.099	0.228
lag=2 quarters	1.302*	.178	0.055
GDP growth (lag=1quarter)	1.043	.630	0.944
lag=2 quarters	.489	.301	0.245
lag=3 quarters	.732	.438	0.602

Table 7 continued

Variable	Odd ratios	Std.	P>z
CR5 loan (lag=1quarter)	1.037**	.017	0.024
lag=2 quarters	1.036**	.017	0.028
Squared CR5 loan (lag=1quarter)	.9999997**	1.23e-07	0.037
lag=2 quarters	.9999998**	1.21e-07	0.044
# of observations	1 731
Pseudo-R2	.950
legend: * p<.1; ** p<.05; *** p<.01

Odd ratios measure strength of correlation between a variable and default probability. Thus, for example, if capital adequacy (N1) increases by 1 (1 percentage point) two quarters previously, the odds of being in default increases by 3.02% while all other things are being equal. Also, an increase in profit to assets ratio or in reserves to assets ratio leads to higher default probability. And if CR5 of loans increases by one from 55 000 (by 0.001 percentage point from 55%) in the previous quarter, the odds of being in default increases by 0.86% while all other things are being equal. The turning point in the model for default probability is at 72% level of CR5. Before 72% level of CR5 default probability is increasing with this concentration ratio, if CR5 is higher than 72%, default probability is going down with growing level of CR5.

According to this model dependence between competition and probability of license withdrawal is inversely U-shaped. It is opposite of hypothesis 3. The lowest probability of default appears with the highest and lowest level of competition in the sample. Probably, it happens because, on the one hand, when banking sector is close to monopoly the Central Bank cannot withdraw more licenses as it will lead to too much risk and distress for society (or even to collapse of the banking sector as no more banks operate), on the other hand, there is no reason to eliminate licenses when banking system is highly competitive as high competition forces organizations to follow high standards to own a share on financial markets.

Therefore, after a certain level of concentration on the loan market higher competition indeed leads to lower probability of license withdrawal.

Limitations

Some limitations of this research can be highlighted. First, unfortunately, because of missing data, in the sample for default probability modeling there is no banks that have been sanitated. Therefore, it cannot be concluded from the research how close sanitated banks were to default right before help.

Second, total loans are considered as one market as well as market of deposits. However, dividing loan market into a market of mortgages, a market of consumer loans, a market of big corporations crediting, etc. and deposit market into a market of corporate deposits, a market of individual deposits, etc. could give clearer understanding of competition in Russian banking sector.

Third, as rules for banks are often changed, some banks' parameters are calculated slightly different over the period. It might cause heterogeneity of the data.

Robustness check

For robustness check models with different competition measurements on the loan market can be compered. As well as the chosen model where CR5 of loans is an explanatory variable model with HHI of loans as an independent variable shows the same result regarding significance of coefficient before competition parameters and their signs. These two models imply non-liner inversely U-shaped dependence between competition and probability of license withdrawal. Moreover, the turning point for default probability in the model with HHI of loans as an explanatory variable is at 0.18 level of HHI of loans. Both values: HHI = 0.18 and CR5 = 72%, indicate lack of competitiveness in the sector.

Therefore, it can be concluded that hypothesis 2 is confirmed. There is a non-linear dependence between competition and probability of license withdrawal. In the model coefficients before competition parameters and before squared competition parameters are statistically significant. It coincides with conclusions made by Mamonov M. [Мамонов М.Е.] (2014) about non-linear dependence. However, type of the dependence is not U-shaped as it is suggested in hypothesis 3. Dependence between competition and probability of license withdrawal is inversely U-shaped according to the chosen model. Coefficients before squared competition parameters are negative and statistically significant. That is the opposite of hypothesis 3. If default probability is considered to be approximately opposite to Z-score of stability (distance to default) applied by Mamonov M. [Мамонов М.Е.] (2014), then, results of this study contradict findings made by Mamonov M. [Мамонов М.Е.] (2014). Though, Z-score of stability is not exactly the opposite of default probability implemented in this work as methodologies are different.