Размещено на http://www.allbest.ru/

Федеральное государственное автономное образовательное учреждение

Высшего профессионального образования

«Национальный исследовательский университет

«Высшая школа экономики»

Международный институт экономики и финансов

Выпускная квалификационная работа - Бакалаврская работа

по направлению подготовки 38.03.01 «Экономика»

образовательная программа «Программа двух дипломов по экономике НИУ ВШЭ и Лондонского университета»

Оценка вероятности банкротства российских банков

Measuring probability of default in Russian banking system

Кузнецова Ольга Андреевна

Рецензент PhD, доцент А.А. Сирченко

Научный руководитель

К.э.н, доцент В.К. Шпрингель

Москва 2016

Contents

• Introduction
• Basic concepts about banks and risks in banking system
• Macroeconomic concepts and their basic overview for Russia
• Previous studies
• Methodology
• Data description
• Factors influencing bank probability of default
• Profitability measures
• Credit risk
• Liquidity
• Operating efficiency
• Determine factors that will be included in the model
• Model construction
• Testing the models
• Prediction on possible bankruptcies
• Conclusion
• Bibliography
• Appendix 1
• Appendix 2
• Appendix 3

Introduction

Banking system is one of the core elements of the whole financial system. The stability of banks and their constant development bring positive results to the economy of particular country. Russian banking system is considered to be quite young. It does not have much experience and demands elaboration to rich high level of development. For well-functioning the favourable conditions should be present and a sufficient level of control should be made. Problems systematically appearing to the banks suggest hurt the whole system and show that some changes should be done.

In resent years an increasing quantity of Russian banks are becoming bankrupt. By the term bankruptcy the license deprivation by Russian Central Bank is meant. Such situation hurts the economy and results in huge costs associated with bank defaults. Social marginal costs associated with defaults of banks are much higher than private ones, since a failure hurts bank customers, other commercial banks and the whole economy (Gwilym, 2011). Increasing number of defaults makes the economy more fragile and more sensitive to shocks. Effectiveness of the economy depends on many factors, and banking system is one of the most influential. This means that lower stability in banking sector places sufficient risk for economy as a whole.

Measuring probability of default (PD) is helpful for looking at bank exposure to risk, namely - to credit risk. This in turn may signal that a particular bank requires stronger regulation. Moreover, probability of default plays a crucial role in risk management process and helps to analyse how particular actions and characteristics of the bank influence its stability. Earlier determinants of bankruptcy risks enable to react rapidly and to take steps for improvement. Knowledge about such parameter as PD is useful not only for the bank itself. Creditors of the bank observe this characteristic in order to decide whether to put money in that particular bank or not and what risk premium should they accept if they decide to put funds in that bank. Higher risks make individuals less willing to put money in a bank, as a result financial organisation experiences lack of funds. This means that smaller amount of loans will be accepted. As a result, economy slows down. It is important to provide sufficient regulations and to determine weak banking organisations before they become insolvent in order to put more severe restrictions on them and to lower probability of default. However, it is problematic to carry out high control for numerous financial organizations that exist in Russian federation. That is why it is important to develop models that will be used to determine what bank needs to be paid a higher attention and which one is absolutely well functioning, so that there is no since to lose much time for its control.

Many researches have been conducted with the aim to measure probability of bank bankruptcy. However, most of them were constructing models based on information on foreign banks. Russian economy has several specific features that are not included in those works. This fact makes these models inapplicable for Russian banks or at least significantly lowers their power. Moreover, being very volatile the economy needs to adjust models constantly, so that particular characteristics that appear within some period of time are included and the influence of other characteristics is adjusted.

This work is aimed to determine main factors that influence probability of bank default in current economic and geopolitical situation and to include factors that are specific to Russian economy today. The research will determine what indicators should be used while analysing the stability of Russian bank and an attempt to look at whether the introduction of anti - Russian sanctions influenced banking system. In addition, at the end of the study a sorting of banks will be provided according to their PD and the most risky organisations will be listed.

Basic concepts about banks and risks in banking system

The first thing to do is to understand more deeply how the bank works, what kinds of banks exist in Russian banking system, what are the main sources of income and main sources of expenses of banks. It is also important to look at core problems that banks may face and to understand how different risks are related to banking activities. Since this work is concentrated on Russian banking system, the main focus will be on it.

The crucial role of the bank as a financial organisation is identical in each country. Banks are seen as financial intermediaries on the market. The main function of the bank is to meet needs of those people who have money with those who demand funds. This financial organisation offers clients the ability to obtain needed sum of money at some interest financing that by attracting funds from others. People who give their funds to the bank receive interest income on their money. So, in simple terms banks borrow money in order to lend them and receive profits equal to the difference of interest rates that is equal to the net interest margin (NIM).

NIM = Interest income on loans - Interest expense on deposits

(Van Greuning, Brajovic Bratanovic, 2009). Higher NIM means that bank is earning higher income. This in turn leads to the expansion of the bank, that means that more lending becomes possible and thus more projects will be realised and economy will grow. So, it seems that banks play a huge role in the economy and should be rewarded a great attention.

In Russian Federation there is a two-tier banking system, that means that the first level of the system refers to the Russian Central Bank and the second level is presented by commercial banks (Beloglasova, Krolivetskaya, 2003). Commercial banks provide financial services to their customers, while Central Bank is aimed to control commercial banks' activities and to ensure development of banking system as a whole (Federal Law «On the Central Bank of the Russian Federation»). Russian Central Bank is given the right to issue money and is responsible for attaining domestic currency (rubble) stability (Constitution of Russian Federation, 75).

Commercial banks activities are mainly referred to borrowing short term and lending long term. This makes banks suffer from several risks. Increased lending increases profitability of the bank at the same time making it more risky (increasing probability of default). This brings to the conclusion that banks should make their decisions on how much loans to make taking into account associated risks. To control banks' activities some rules and requirements were made that force banks to hold the required amount of capital and reserves. They will be used as a cushion against losses that bank may face and as a result will lower probability of default.

The main risks banks face are:

· credit risk - that is associated with probability that the counterparty will not meet its' obligations (Gwilym, 2011);

· liquidity risk - associated with lack of liquidity to meet demands of customers (Gwilym, 2011);

· interest rate risk - related to possible losses in case market interest rates change (Gwilym, 2011);

· market risk - associated with «adverse deviations in the value of trading portfolio» (Gwilym, 2011, p. 36);

· solvency risk - referred to situation when bank does not have enough capital to cover its losses (Gwilym, 2011);

· operational risk - related to losses associated with system failures, internal or external frauds, damages to physical assets (Gwilym, 2011);

The global economic crisis that hurt the whole world in 2007 (Russia was not an exception) was caused mainly by the actions of financial intermediaries. To be more precise, a sharp increase in issues of subprime mortgages in US and increased use of securitization without provision of sufficient regulations resulted in increased default rates on mortgages and led to enormous losses faced by financial organizations (Gwilym, 2011). Crisis expanded and caused serious problems and bank runs in other countries. This situation revealed the weaknesses of banking systems and lack of control. Bank runs are more likely to occur when depositors are not sure in their bank. The main problem is that one bank run usually results in a situation when clients in other banks loose confidence in their bank and also decide to bank run.

Deposit insurance may help to prevent such situation, but they give rise to moral hazard problem, that means that banks will be willing to take on more risk. Diamond and Dybvig developed a model that helped to provide an insurance against bank runs (Gwilym, 2011). The main idea of the model is that financial intermediary can diversify consumption shocks and should hold some amount of money that is equal to the estimated amount its clients are expected to withdraw at time t=1. Early withdrawal makes people loose money, since for each unit put in a bank they get only L*1, where

0 < L < 1.

Clients who decide to hold money in a bank until time t=2 receive premium; for each unit put in a bank they receive R*1, where R > 1 (Gwilym, 2011). According to the logic of Diamond and Dybvig this will prevent individuals from early withdrawals and will lower risks for the bank.

As it was discussed earlier, capital plays a crucial role in minimising banking risks by absorbing losses and providing confidence to clients of the bank. It has no exact fixed maturity and due to this fact is seen as the best means of long term financing. Basle committee decided on capital adequacy regulations that are aimed at forcing banks to hold sufficient amounts of capital. For this sake minimum capital requirements were used. The value of total risk-based capital ratio, that is the ratio of total capital to credit risk-adjusted assets was agreed to be not less than 8% (Basle, 1988).

Second Basle agreement changed capital standards by dividing capital requirements into three categories: those associated with credit, operational and market risk (International Convergence of Capital Measurement and Capital Standards A Revised Framework Comprehensive Version, 2006). Risks are estimated for each of the group and based on these estimations the amount of capital that is necessary to hold is found. The main aim of such requirements is to control bank exposure to credit risk (control PD).

Bank's probability of default is mainly referred to credit risk. The level of such risk could be measured either using standardised approach that implies the use of external credit ratings that are provided by credit rating agencies (the biggest agencies are: Standard & Poor's, Moody's and Fitch Group) or using internal rating based approach. Standardised approach is problematic to be widely used in Russia, since credit ratings are given only to a limited number of companies and thus not applicable to most Russian banks. Internal rating based approach models are widely used for internal management purposes. They are rarely revealed to the public. Such models could be constructed using statistical - based process, constrained expert judgement or pure expert judgement process (Gwilym, 2011). Being formed using different methods depending on the purpose of the rating, they cannot be compared among banks. As a result, such ratings are inappropriate to for comparing default probabilities among different financial organisations.

Another tools that could be used in order to analyse bank performance are early warning systems. The results obtained by using such methods are then further used in bank evaluation process and decision-making. One of the most popular early warning systems is CAMELS model. «The CAMELS stands for Capital adequacy, Asset quality, Management, Earning and Liquidity and Sensitivity» (Rostami, 2015, p.1). The underlying idea of this model is to assign rating to the bank that is calculated as average score of 6 characteristics. Each characteristic is given a mark out of 5, where 1 means the highest quality of particular characteristic and 5 means the lowest. First category is referred to capital adequacy, that is rated taking into account the general level and trend of capital, capital requirements, risks the bank faces and some other factors; asset quality mainly refers to evaluation of credit risk bank is exposed to; management quality evaluation is based on the ability of managers to quickly respond to any changes in risks and to adopt bank activities so that the bank could adjust to new conditions without incurring high losses; earnings and liquidity are usually given a mark after looking at their level and trends; last - sensitivity is given a rating based on sensitivity of the bank to market risk, nature of activities the bank participates in and some other factors (Gwilym, 2011; Rostami, 2015). The main disadvantage of CAMELS is that they are not refreshed very often and thus make it problematic to react to problem operatively. Such early warning systems as Statistical CAMELS Off-Site ratings, SEER and OCC model also exist. They were discussed in paper «Early warning models for bank supervision: Simpler could be better» by Julapa Jagtiani, James Kolari, Catharine Lemieux, and Hwan Shin (2003). Most of these models take into account CAMELS rating. So it seems that CAMELS method must work well, and measuring bank risk and performance by looking at the main groups of characteristics is a good idea.

Macroeconomic concepts and their basic overview for Russia

(This part includes a sufficient amount of data that was taken from 3^rd year course work)

Russian economy has several specific features that should be taken in account while conducting analysis. To be exact, Russian economy largely depends on exports of primary energy resources, mainly oil. Any demand or price shocks may cause serious changes in Russian economy. This is clearly seen after a little glance at current situation in the economy. Moreover, in resent years the geo-political situation, but not only economic one is not stable. And Russia is in the centre of many political issues and conflicts. It could be interesting to look at how macroeconomic features and political events may influence risks faced by banking system and by each bank particularly. To do this it is important to look at key macroeconomic indicators and their tendencies in Russia during last years. Then most important will be included in further analysis.

GDP (Gross Domestic Product) is often used for revealing «economic health» of a particular country. It measures the total value of goods and services produced in the economy. Another indicator that plays an important role in making conclusions about the economy is GDP growth rate, that shows its percentage changes and is calculated according to the formula:

GDP growth rate = ((GDP_t+1 -GDP_t)/GDP_t)*100% .

One more factor to be taken into account is the exchange rate. It describes the convertibility of domestic currency into foreign currencies. Exchange rates depend largely on the supply and demand of currency and are often influenced by Central Bank monetary policy decisions. In this work the exchange rate will be defined as the number of units of domestic currency needed for one unit of foreign currency.

Box 1 Appendix 1 shows dynamic of Russian GDP in billion rubbles during the period 2011 to 2015. Being equal to 15663,569 billion rubbles at the end of 2011, it increased up to 22016,107 billion rubbles by the end of 2015. However, it is more interesting to look at GDP growth rate that is presented in Box 2 Appendix 1. From the graph it is clearly seen that growth rate has fallen significantly from 2011 to the first half of 2013. After that in the beginning of 2014 a little improvement is observed, that is followed by a sharp drop starting from the fourth quarter of year 2014. The fall must be associated with significant drop in oil prices and rubble depreciation. A fall in GDP growth rate in second half of 2015 became not as large, as it was before. This shows that economy starts adapting to new conditions and GDP growth rates will probably soon go up. Such sectors as agriculture, food industry, and domestic tourism showed significant improvements after a negative economic shock, as newspaper ВЗГЛЯД (18 May, 2016) sais. Due to the development of these areas Russian fall in GDP is slowed down.

Monthly changes in exchange rate are presented in Box 3 Appendix 1; fluctuating around 30 Russian rubble per US dollar, it jumped up to 70 at the end of 2014. A little improvement was seen in the beginning of 2015, when rubble appreciated to 50, but then depreciation continued.

Since Russian economy strongly depends on exports of oil, it is reasonable to look at how oil prices changed during the last years. Box 4 Appendix 1 shows prices of oil mark Brent. Starting from 2014 a decline in oil prices is seen, and in year 2015 after a little increase the decline continued. This of course seriously hit the economy and led to fall in the value of exports. This resulted in rubble depreciation and increase in inflation. Inflation is associated with falling purchasing power of money caused by increasing price level. It is clear that a significant increase in general price level results in lower aggregate demand and makes firms produce less, causing a fall in GDP and making firms hire less workers, thus resulting in higher unemployment. Lower production in turn decreases demand for funds as well as to bank loans, since production process slows down. This hurts banks a lot.

Another indicator of macroeconomic situation is unemployment rate. Unemployed people are those who refer to economically active population and are willing and are able to work but can not find job. Box 5 Appendix 1 shows the dynamics of unemployment rate in Russia. Country experienced improvements in unemployment levels during last years, but a slight worsening is seen starting from the second half of 2014. This period corresponds to annexation of Crimea and introduction of sanctions.

Current macroeconomic situation definitely seriously hurts banking system. Banks face increasing difficulties in attracting funds. This fact is a straightforward result. Rubble is becoming weaker, Russia is becoming more risky for investors and foreigners prefer to extract money from Russia. Interest rates rise making loans too costly to people who are stressed by rising prices and increasing cost of life in general followed by unchanged salaries. Russian CB on official site publishes information that the value of outstanding debt is rising both among corporate and private borrowings and the proportion of liquid assets in total assets of banks is falling.

Previous studies

It is quite long since the very first attempts to measure risk exposure of the financial organisation were made. One of the early works that was aimed at analysing and detecting financial problems faced by economic organisation refers to work of Altman «Financial ratios, discriminant analysis and the prediction of corporate bankruptcy» (1968). He has shown that traditional view at financial ratios does not work well for analysing probabilities of default and from time to time leads to inappropriate conclusions. Instead, he offers a multiple discriminant analysis for these purposes. Some function:

Z = ,

where x is explanatory variable and v is discriminant coefficient of that variable is formed. The time period analysed is: 1958 - 1961. After results are received the cut-off points for the value of function Z are set, that are further used to indicate whether the organisation of interest is in the risk zone or not. This method implies grouping variables according to some characteristic of explanatory variable. Too many variables have shown to be significant during the analysis. For this sake Altman made a decision to choose variables based on their frequency in literature and taking into account the importance of each particular parameter for the study. The ratios that were decided to be included are as follows:

· Working capital to total assets

· Retained earnings to total assets

· EBIT to total assets

· Market value of equity to total debt book value

· Sales to total assets

Individual discriminating ability of each variable was tested using F-test. Results showed that ratio of profitability (EBIT to total assets) has appeared to be the most significant. For evaluation of model quality Type one and Type two errors were analysed. The model was said to be appropriate for determining bankruptcies within a two year period.

Binary choice model in estimating probability of bankruptcy was used in work of Daniel Martin «Early warning of bank failure. A logit regression approach.»(Martin, 1977). The sample used by author includes US banks from 1970 to 1976. Variables for the analysis were classified in four groups:

· Asset risk (loans / total assets)

· Liquidity (net liquid assets / total assets)

· Capital adequacy (basic capital / asset ratios)

· Earnings (operating expenses / operating income; net income / total assets; net interest margin / earning assets)

The main conclusions the author made after analysis are that gross capital to risk asset ratio is very significant in determination of probability of bank bankruptcy. Similarly, commercial loans to total loans ratio showed high level of significance. Martin mentions that he does not include liquidity variables in the model, since they do not appear to be significant after the regression is ran. It is explained by the fact that during the analysed period no bank runs occurred. Author concluded that logit model in general gave better results than linear discriminant or quadratic linear discriminant model.

In research paper «The determinants and early detection of inadequate capitalization of US commercial banks» conducted by Jagtiani, Kolari, Lemieux, Hwan Shin (2002) authors explain logistic regression method and trait recognition approach. The period from 1988 to 1990 is analysed. As explained, the time interval was chosen to have enough events of default. Mainly small banks are of interest as authors see them being more problematic. However, banks that are too small were excluded due to being very special and being focused on a very limited audience. The step equal to one year is taken for explanatory variables. Authors used 41 financial ratios in research. In general, after analysing type one and type two errors, it was seen that some outcomes are better predicted by trait recognition model rather than logit, but this method has several serious disadvantages. First of all, it is more complicated and takes much more time to be conducted. Secondly, and most important, it does not allow to analyse the significance of each explanatory variable and does not allow to determine what factors are important while analysing bank's fragility. So, trait recognition is not appropriate if the aim is to understand what characteristics should be paid more attention while measuring PD.

Since Russian banking sector is quite young, attempts to analyse it were made much later than in US or Europe. Modelling probability of bankruptcy based on analysis of three main issues like information on the bank's balance sheet, information of the balance of the banking system as a whole or information taken on a macro level was discussed in research «Models of probability of Russian banks bankruptcy. Preliminary division into clusters» by Golovan, Karminsky, Kopylov, Peresetsky (2003). Authors took data on Russian banks that corresponds to the period 1997-1998. The method used in analysis is logit model. The main criteria of clustering the data was the size of the bank. Within each cluster size showed to be insignificant. The ratio of investments in the government debt to loans to non-financial organizations was found to be of high importance. It is noticed that the ratio of the bank's own capital to its liabilities is very different for good and bad banks and so is believed to be important while constructing the model. Regardless of the size such ratios as liquid assets to the value of bank's liabilities, capital to total liabilities and reserves to total liabilities were found to be highly significant and this makes authors to believe that exactly these ratios play crucial role in estimating probability of default of the bank.

Further analysis «Probability of default models of Russian banks» (Golovan, Evdokimov, Karminsky, Peresetsky, 2004) was aimed to determine the importance of macroeconomic conditions in measuring probability of default of the bank in Russia. Probit model was decided to be used and the time period used is 1996 - 2002. Such explanatory variables as GDP, GDP growth rate, exchange rate, CPI, unemployment rate and number of unemployed people were analysed. The results suggest that introduction of macroeconomic indicators have a positive effect on the general quality of the model and on the significance of the coefficients particularly. Due to high correlation it is impossible to include many macroeconomic factors. Authors found out that the best macro indicators are exchange rate and the ratio of export to import.

In work «Econometric approach to remote analysis of the activity of Russian banks and banking supervision» (Peresetsky, 2009) the data on Russian banks during the period 1996 to 2002 is used. All the information is taken for each quarter during the chosen period. A lag used in the model is 2 years. Such a decision was made based on the idea that 2 years is enough to complete the bankruptcy procedure. Another thing that stands in favour of such time interval is that it was said to have the highest predicting ability as compared with one year time interval. Taking into account the fact that the number of non-bankrupt banks seriously exceeds the number of bankrupts the decision to reduce the number of good banks was made. This would help to increase the proportion of defaults in the sample and to make the data more balanced. Logit model was chosen to be used, and the data analysed is the one that is easily publicly available. The advantage of such data is explained by the fact that it is easy to obtain and thus does not require much time and resources. Author analyses whether clustering might be helpful and improving the model predictability power. The best results according to significance of coefficients were obtained while clustering banks according to the share of government securities in bank balances. However, the model without clustering by such criteria worked better for identifying banks in the risky zone. For weak banks a model with clustering by the level of capitalisation worked a bit better.

Research paper «Factors of stability of Russian banks in 2007-2009» (Drobyshevskiy, Zubarev, 2011) drives attention to one of the most significant factors in bank stability - the size of its capital. Authors conducted a study taking information about banks for the period 2006 - 2009 that is presented in a form of an unbalanced panel data. The results showed that the biggest and the smallest banks did not experience many cases of bankruptcies. A significant number of banks that are in the middle if sorted by the size became insolvent. This was concluded after obtaining different signs of coefficients before banks' logarithm of assets and the squared logarithm of assets. Loans to households and foreign debt have shown to be highly significant in the model. This was explained by the fact that mainly good banks have an ability to give loans to the public and to have foreign debt. It is argued that the total value of foreign debt is doubtfully important for measuring probability of default. The negative sign of coefficient before the outstanding debt is explained by the fact that only good banks are not afraid of showing it on their balance sheets. Smaller banks are said to camouflage outstanding debt by writing it as if the new credit was given. This fact raises the problem of wrong data presented in reports from weaker banks. The ratio of market debt to the total bank's liabilities and the ratio of reserves to loans to the non-banking sector showed to be the most significant and most influential while measuring probability of default. Authors obtained an insignificant coefficient before the ratio of liquid assets to the total assets of the bank. They explain this by the fact that the analysed period is 2006 to 2009, and since there was no bank run and it was quite easy to get the loan from Central Bank at that time contrary to 2004. Attention is paid to the significance of the ratio of export to import and exchange rate. As explained in the work, fluctuations in exchange rate tend to increase currency risk and so make bank more fragile.

One of the latest works for the described theme is «Modelling of the probability of default of the Russian banks: advanced abitilies» (Karminsky, Kostrov, 2013). Authors conducted a study forming a model based on the data about banks during 1998 to 2011 year. The decision on time interval that should be taken is accepted to be a quarter. Monthly data is said to have high level of misleading information. Logit model is used, and explanatory variables are taken with lags, equal to 8 quarters. A high attention is paid to the fact that the data is highly misbalanced that leads to biased estimations and does not allow the model to predict defaults well. This happens since the number of well functioning banks is much higher than the number of those became bankrupt. For the sake of improving the quality of the model, data is balanced by synthetically increasing the number of observations. Researches showed that increasing the value of lag the quality of the model falls. One of the disadvantages of logistic regression is said to be the absence of time indicator. Authors used dummy variables to include time. After conducting some analysis the conclusion was made, that only dummy that referred to year 2009 was significant. Similarly to other studies researches say that including macroeconomic factors improves the quality of the model. It was also mentioned that macroeconomic factors could be highly correlated with each other and thus only two factors were decided to be included: GDP growth rate and CPI.

Methodology

After analysing previous researches it seems reasonable to use a binary choice logit model in the analysis, as it was proved many times that such method is most appropriate for the purpose of measuring probability of default. It allows measuring the strength of influence of the explanatory variables on the dependent variable and is appropriate for determining core factors that influence probability of bank default. Results obtained using this method are easily interpreted.

Logit model is a parametric binary choice model. The dependent variable can only take values from 0 to 1. Since probability is by definition takes values in the interval

0<=p<=100,

linear regression can not be used. However, it should be understood that coefficients obtained in the logit model do show only whether the variable affects the dependent variable positively or negatively. If the strength of this influence is of interest, marginal effects should be obtained.

The function used in logit model looks as follows:

P_i = F (Z_i) = e^Zi / (1+e^Zi) (1),

where Z_i is a linear combination of explanatory variables (exogenous factors) (Peresetsky, 2009).

Z_i = b₀ + b₁ * X₁ + b₂ * X₂ + … + b_n * X_n (2)

Z_i = ln (P_i / (1-P_i))

(lecture of Prof. Sharyn O'Halloran Sustainable Development U9611 Econometrics II)

Multicoliniarity is a potential problem for parametric regressions, as it negatively influences model quality by increasing error terms. This hurts the power of regression. That is why it is important to control for it. This could be done by analysing correlation matrix. Variables that are highly correlated with each other should not be simultaneously included in the regression. Perfect multicoliniarity occurs if the correlation of variables is equal to one. In this case there is no sense at all to include both indicators in the model, as the quality of predictions will be damaged. For similar reason it is important to control for heteroskedasticity.

While analysing the quality of regression, its predicting power and type one and type two errors should be looked at. By predicting power the number of truly defined bankrupts is meant. Type one and two errors are discussed later on.

Data description

The data that is analysed in this work refers to banks that were functioning well and that became bankrupt during the period 2012 to 2015. Information on bankruptcies and exchange rate was taken from official Internet resource of Russian Central Bank (CB). Data on macroeconomic characteristics like GDP and unemployment was taken from the official Internet resource of Federal Service of State policy. Prices of oil mark Brent - from Independent statistics and analysis U.S. energy information administration.

By bankruptcy in this work is understood both compulsory and voluntary deprivation of license by Russian Central Bank. On official Internet resource of CB the following formulations of reasons for license deprivation are provided:

· Deprivation of license due to unreliability of the accounting data provided by organisation (Guided by Article 20 of the Federal Law «About Banks and Banking activity» and Article 74 of the Federal Law «On the Central Bank of the Russian Federation»)

· Deprivation of license due to counteraction to legalization of income received through criminal activities and financing terrorism and money laundering (Guided by Article 20 of the Federal Law «About Banks and Banking activity» and Article 74 of the Federal Law «On the Central Bank of the Russian Federation»)

· Deprivation of license due to the level of capital adequacy below 2 %

· Deprivation of license due to inability to meet creditors' claims

· Deprivation of license due to inability to meet requirements on charter capital

· Sanitation of the bank

· Merge of the bank

The reason for taking the time interval 2012 - 2015 is to analyse what factors influence banks probabilities of default today. Period around year 2008 is not included, since factors that are important in current economic condition are of the interest. Russian economy is very unstable and it is important to refresh models, since at different times different factors influence the economy. Furthermore, this time interval allows to determine whether introduction if sanctions and the whole situation of political instability that expanded in 2014 puts the effect on Russian banks, and if it does, the analysis will determine how serious it is. During this period a sharp increase in the number of banks that became bankrupt is observed. The Diagram 1 below presents the data on the number of bankruptcies from 2009 to 2016 provided by Russian Central Bank reports. It is clearly seen that starting from 2013 the number of banks that become insolvent was steadily increasing. Especially huge jump is seen in 2014, when the number of bankrupts rocketed from 45 to 93. In 2015 the rise continued reaching a number of 108.

Diagram 1. Number of bankruptcies (Source: http://www.cbr.ru/statistics/?PrtId=lic)

Model constructed further will mainly be based on analysing information on major accounting characteristics that was obtained from Mobile database. Information from year 2012 - 2014 is used for model construction, while year 2015 will be used for testing the model. The total number of initial observations is 277 bankruptcies (978 functioning banks) and 13859 observations for non-bankruptcies. In the analysis the indicator «1» (default = 1) will be used for banks that became bankrupt and «0» (default = 0) for those banks that did not do so. Each bank has a corresponding number - some index (ind). Explanatory variables are taken for each bank corresponding to each period analysed. If bank becomes bankrupt at time t_i, no information is presented about it at that time, bank is indicated as bankrupt at time t_i-1 and is set as non-bankrupt at time t_i-2. This logic was used by Peresetsky (2009). The step used in data is equal to one quarter. Earlier studies indicate a good performance of models constructed using such step. It allows to determine potential bankrupts quite rapidly and to react to warning signs operatively. The data used is unbalanced since banks are observed during different time periods. Once the bank becomes bankrupt it does not appear in the next period. More over, data that is disclosed by banks is often missed and gaps are seen.

Banks like Сбербанк, ВТБ, ВТБ24, Газпром банк, Альфа - банк, Банк Москвы and Россельхоз банк were decided to be excluded from the sample for the reason that they are «too big to fail», and if they face difficulties there is no doubt that government would provide support to them. Too small banks were excluded from the sample as well, following the logic of Jagtiani, Kolari, Lemieux and Hwan Shin (2003), they are too specific and should not be used in a model. The sorting was made using the value of logarithm of assets that is used as a proxy for bank size.

Factors influencing bank probability of default

One of the main parts of the analysis is to look at what factors should be included as explanatory variables in the regression. Based on previous studies analysis was conducted and the decision on what factors to include was made. Most previous works are based on analysis of accounting figures. They are said to be easily found and quite well determining potential weaknesses of the bank. Accounting characteristics that could be potentially included in analysis were chosen taking into account CAMELS classification. As said in work «The SCOR System of Off-Site Monitoring: Its Objectives, Functioning, and Performance» by Charles Collier, Sean Forbush, Daniel A. Nuxoll, and John O'Keefe (2003), CAMELS ratings are accepted by many experts as best indicators of bank position.

Table 1 below shows the abbreviations of parameters that will be further discussed forming explanatory variables for the model.

Table 1. Financial indicators (as used in Mobile database)

CA	Net assets of the bank
VB	Total assets of the bank
RES	Reserves of the bank
KE	Loans to non-financial organisations
GDO	Government securities
SK	Capital of the bank
LA	Liquid assets of the bank
PZS	Outstanding debt
CP	Net profit
SO	Total liabilities of the bank
VDFL	Deposits of individuals
PNA	Non-working assets
ODB	Operational earnings
ORB	Operational costs
NCB	Non-government securities
ORCB	Obligatory reserves in CB
TA	Total assets
NORM_AR	Bank risk-weighted assets
KE_F	Loans to individuals and individual businesses
RK	Interest expense on taken loans
PZS_F	Outstanding debt of individuals and individual businesses
PZSB	Outstanding debt of banks
MBK	Loans to other banks
PDFL	Interest income from loans to individuals
RPFL	Interest on individual deposits
PD	Interest income
PR	Interest expense
NORM_LAM	Highly liquid assets (according to norm H1)
BP	Balance profit

Based on these indicators the following explanatory variables are formed:

Size measure

1) Ln (CA) = ln__ca

As it was proposed in earlier studies, that logarithm of assets of organisation is a good proxy for its size. It is intuitively that banks that are bigger in size are more stable, and so increasing size of the bank should be associated with higher safety.

Expected sign of the probability influence	Negative «-»

Capital adequacy measure

2) SK / CA = sk_ca, SK / TA = sk_ta and SK / NORM_AR = sk_norm_ar

Capital absorbs losses. Higher amounts of capital make bank safer and lower risks of default. Since capital does not have fixed maturity it is a superior tool to protect the bank against losses.

Expected sign of the probability influence	Negative «-»

3) SO / TA = so_ta

Ratio shows how much total liabilities bank has per unit of total assets. Based on this ratio banks should understand how much capital it should hold for the case if not all liabilities would be repaid for the bank or if some will become outstanding debt and matching among assets and liabilities will be broken.

Expected sign of the probability influence	Positive «+»

Profitability measures

High profitability ratio usually indicates the well functioning banks that are strong enough and do not experience significant problems.

4) CP / CA = roa

Ratio shows how much net profits are generated per each unit of assets. Higher profitability usually refers to stable banks.

Expected sign of the probability influence	Negative «-»

5) CP / SK = roe

Ratio shows how much profits are generated on each unit of equity. This indicates how much equity holders benefit from banking performance. Higher ROE ratio is a good sign.

Expected sign of the probability influence	Negative «-»

6) NCB / CA = ncb_ca

Non-government securities tend to have higher risks. At the same time following the risk-return trade off they offer higher return and thus increase profitability. Due to higher associated risk weak banks prefer not to invest in such securities. This means that investments in non-government securities tend to indicate the confidence of the bank.

Expected sign of the probability influence	Negative «-»

7) BP / CA = bp_ca

Ratio depicts how much profit is generated per unit of asset. Higher profits are expected to be a sign of well functioning bank.

Expected sign of the probability influence	Negative «-»

8) RK / TA rk_ta

Interest expense to total assets indicates how much interest expenses are per unit of asset. Higher expenses hurt banks and may indicate that bank does not have enough assets.

Expected sign of the probability influence	Positive «+»

9) CP / SO = cp_so

Ratio presents the value of net profit per unit of liability. Higher ratio indicates higher profitability of the bank.

Expected sign of the probability influence	Negative «-»

10) PDFL / PRFL = pdfl_prfl and PD / PR = pd_pr

Ratios show how proportion of interest income to interest expense associated with individual loans and deposits and total interest income to interest expense correspondingly.

Expected sign of the probability influence	Negative «-»

Credit risk

11) RES / CA = res_ca and RES / TA = res_ta

The ratio shows whether bank holds a sufficient amount of reserves. The higher the value of the ratio the lower is expected probability of default, since acting as a cushion against potential losses capital will absorb them allowing the bank to well function further. If for example bank gave a loan and the holder of the loan did not pay for it, reserves would help the bank to cover these unexpected losses and will not create serious problems for the bank.

Expected sign of the probability influence	Negative «-»

12) KE_F / CA = ke_f_ca

Loans to individuals and individual business organisations are one of the sources of interest income of the bank. At the same time they are sources of risk. More confident banks are able to give more loans as they are sure that they will cope with losses if any occur.

Expected sign of the probability influence	Negative «-»

13) PRZDL = (PZS_F + PZSB) / (KE_F + MBK) = przdl

Calculated as the ratio of outstanding debt to loans given, this ratio the proportion of loans that were not repaid by individuals and banks. Outstanding debt creates risk, as bank does not receive expected amounts and thus may need to fire sell assets or use other methods to get liquidity.

Expected sign of the probability influence	Positive «+»

Liquidity

14) LA / CA = la_ca and LA / TA = la_ta

Liquid assets to total assets ratio depicts what proportion of bank assets can be easily transformed into cash. Higher liquidity positively affects bank strength and reduces probability of default.

Expected sign of the probability influence	Negative «-»

1) GDO / CA = gdo_ca

Government securities (sometimes viewed as substitutes for cash) can be easily sold back to the government and required liquidity will be soon received. The higher is the ratio the lower is expected probability of default. Big good banks tend to hold larger amounts of government debt obligations.

Expected sign of the probability influence	Negative «-»

2) Norm_lam / TA = norm_lam_ta

Proportion of highly liquid assets in total assets. Higher value of the ratio means better liquidity positions of the bank.

Expected sign of the probability influence	Negative «-»

Operating efficiency

3) ODB / ORB = odb_orb

The difference between operating profits and costs presents operation income. Ratio shows how much operating profit is generated per unit of operating costs. More efficient banks tend to have economies of scale and to lower operating costs per unit of operating income.

Expected sign of the probability influence	Negative «-»

4) ODB / CP = odb_cp

Ratio shows how high is the proportion of operational income in net income. The higher the ratio, the more profitable operations are.

Expected sign of the probability influence	Negative «-»

In this work macroeconomic factors will be included as well as indicator of political situation that is sanctions. Earlier main macroeconomic indicators were discussed. It is obvious that many of them are highly correlated with each other. For this reason the model will not include all of the macroeconomic factors. Since the period 2012 to 2015 is analysed, it seems reasonable to look at exchange rate (dollar exchange rate will be used, RUB / USD) (abbreviated as USD), oil prices (abbreviated as barrel, measured in USD), GDP (gdp, measured in bln. RUB) and GDP growth rate (gdp_gr, in %).

Expected sign of the probability influence
USD	Negative «-»
Barrel	Negative «-»
Gdp	Positive «+»
Gdp_gr	Positive «+»

Another indicator that will be included in the model is introduction of sanctions, which will appear in the model as a dummy. This will show whether political conflicts influence the banking sector. The introduction anti-Russian sanctions corresponds to the same time period as inclusion of Crimea back to Russia and to the beginning of an armed conflict in Ukraine. All these factors are included as a dummy variable that takes zero values before April 2014 (second quarter of 2014) and 1 for the following period.

Expected sign of the probability influence
Sanctions (dummy)	Negative «-»
Year 2014 (dummy)	Negative «-»

Determine factors that will be included in the model

One of the problems that are present in parametric models is multicollinearity. To avoid it the correlation table is constructed and factors with high correlation are not included. Expanded information on correlations is available in the Appendix 1 Box 6. Results suggest that the most highly correlated variables are: look at Table 2 below.

Table 2. Highly correlated variables

Sk_ca	Sk_ta
Bp_ca	Roa
La_ta	La_ca	Norm_lam_ta
Ln__ca	Odb_orb	So_ta
Res_ta	Res_ca

The result of high correlation between sk_ca and sk_ta is obvious, since ta and ca only differ by the amount of reserves required by Russian CB. Sk_ca has lower correlation with remaining variables in the model. Further decision on what factor to include will be done after looking at t-test results for the difference of mean (Box 7 and Box 8). The null hypothesis is that the difference between mean sk_ca for defaulters and non-defaulters is equal to zero.

H₀: mean (sk_ca non-defaulters) - mean (sk_ca defaulters) = 0

H₁: not H₀

Group 0 means non-defaulters.

Group 1 means defaulters.

Box 7 (t-tests for equality of means sk_ca)

t-test sk_ca, by (default)

Two-sample t test with equal variances ______________________

Group | Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] ____________

0 | 10527 .6868581 .0050266 .5157368 .677005 .6967113

1 | 134 .4523991 .0431521 .4995218 .3670459 .5377523 ______________

combined |10661 .6839112 .0049992 .5161748 .6741119 .6937105 ______

diff | .234459 .0448182 .146607 .3223111

diff = mean(0) - mean(1)

t = 5.2313

Ho: diff = 0

degrees of freedom = 10659

Ha: diff <> 0

Pr(|T| > |t|) = 0.0000

Box 8 (t-tests for equality of means sk_ta)

...