On the row with coefficients, the results of the model can be reported as odds ratios. Assuming that the probability of target event (bank failure) may potentially be equal to the probability of non-target event (survival), then in this case the odds will equal to 1. Proceeding from this, if the probability of target event is greater than the probability of non-target one, the odds are expected to be greater than 1. On the contrary, if the occurrence of non-target event is more likely comparing to the target one, these odds ratios will be less than 1. In this paper the two options of the regression outcomes reporting will be used and further presented in the Results section.

Finally, the rationale to build the model with random effects is explained by the aim of the regression results and the study in general. If using fixed effects method, then the model will not calculate the probability of the outcome (i.e. failure). Instead of this, the results will model the probability that, among all the observations of the particular panel unit (i.e. bank), namely these ones will have non-zero outcomes (Conway, 1990). Moreover, the changes in the variables across time are little (from approximately one thousandth to one hundredth). This makes the application of fixed effects regression rather complicating or even impossible. These points of view were also supported and used in prior studies in which logistic regression was applied.

The model building and all the supplementary data analysis will be conducted in STATA software package, as it is a quite convenient instrument when dealing with such kind of regression.

3.Results

3.1 General statistics of failed and survived banks

The study involves 896 Russian commercial banks. During the period of 8 years from 2012 to 2019, 511 of them has been deprived of a license. Such a high failure percentage (57 per cent of failed banks and 43 per cent of survived ones) indeed demonstrates the peculiarity of this period concerning radical changes of the Russian banking system, hence, generates the interest for the research (Table 3).

Table 3. Failure / Survival statistics

Figure 2 illustrates survival statistics graphically. Implying the time period to be 96 months, the survival curve is decreasing with every consequent month. It also can be seen that half of the credit institutions that were operating in January 2012 has been gone approximately after the 81^st month of observation. In addition, the numbers at risk are included that show how many banks were in danger of failure in a particular time period with the interval of 20 months.

Other plots estimating the hazard function and cumulative hazard function are presented in Appendix section and illustrate survival analysis more in details (Appendix 2).

Figure 2. Kaplan-Meier survival estimate plot

3.2 Descriptive statistics of the dependent and independent variables

Table 4 describes the “Bank Failure” binary dependent variable where “1” stands for the failed bank in the panel dataset and “0” means a survived credit institution. The slight difference in the percentage with the previous table (Table 3) is explained by the exact number of observations participating in the study, not the total number of failed and survived banks. Observing the higher frequency of failures and aligning it to the financial indicators that represent one of the six bank strategies, the regression outcomes are expected to reveal strategy's role in bank performance.

Table 4. Tabulation of Bank Failure

Both Table 5.1 and Table 5.2 characterize independent variables used in the regression. They were divided for the convenient illustration, as Table 5.1 includes the key regressors of the model (“Bank strategy” variables) and another one delineates the control regressors.

Analysis showed that most of the banks have distinct strategy (Table 5.1). For instance, the majority of the “Rusfinans Bank” assets are being formed from issuing individual loans. Their share in the eight years period was varying in the diapason of 88 to 97 per cent (the maximum value for Individual Loans). On the contrary, the engagement of the bank in issuing commercial loans is incomparable to the individual ones and has not exceeded even 1 per cent of the total assets (Appendix 3).

The summarize of the additional control variables presented in the second table (Table 5.2) indicates the relevant performance of a bank concerning its profitability and state of compliance with the Central Bank mandatory normatives. The extreme data outliers were excluded from the statistics, as the information retrieved from the special CB's files (containing the values of bank normatives) showed several unusually positive and negative numbers. However, for bank normatives such a vast deviation from the required percentage turned out to be frequently observed among banks in the data sample, hence, demonstrated their fluctuating and unstable performance.

For the ROA, the significant difference between the least and the most profitable credit institutions is visible. When the financial performance of the “Central Commercial Bank of Surgut” has experienced a sharp decrease in the end of 2014 - beginning of 2015 (ROA = -2.96 as of January 01, 2015), no wonder that in a couple of months this bank has failed. “Pochta Bank”, in turns, was demonstrating sufficient performance throughout the studied period with only positive ROA numbers, which in the beginning of 2013 even exceeded 50 per cent (Banki.ru [Electronic Source]).

Taking the Capital Adequacy Ratio (H1 normative) as an additional example, the minimum value of -2.76 per cent calculated as a ratio of bank's capital to its assets, illustrates the complete inability of “Rinvestbank” to satisfy the requirements set by the financial regulator, as the minimum established value is 10 per cent. Thus, demonstration of such a poor performance has inevitably led to “Rinvestbank” bank license revocation (Appendix 4).

These data demonstrating the fluctuating financial position of credit organizations is believed to make the analysis of strategy's efficiency more accurate and significative. The inclusion of all indicators in the regression model will help to reveal patterns of how particular bank strategies may impact overall financial performance of a bank, hence, have an effect on the probability of its failure or survival.

Table 5.1 Descriptive statistics of the independent variables

Table 5.2. Descriptive statistics of the independent variables

3.3 Tables of correlations

Tables of correlations are necessary for the study in order to check the independent variables for multicollinearity issues. In this case, all variables were divided into two tables according to the specifics of the data. Financial indicators in the Table 6.1 were checked separately from the Central Bank normatives in the Table 6.2, as these groups of variables have different nature, thus, insignificant correlation degrees.

Table 6.1 Pairwise correlations between financial indicators

Overall, the “Bank Strategy” variables, as well as the profitability indicator do not have high correlation between each other. None of them demonstrates correlation of more then 0.5. However, it is worth pointing out that there are pairwise correlations with negative sign. This is explained by the fact that each of the three strategies concerning bank's asset portfolio and three strategies in liabilities portfolio constitute their own shares in the respective portfolio. Therefore, if, for instance, the share of individual loans in asset portfolio constitutes more than a half, the share of commercial loans is obviously less. Thus, the majority of “Bank Strategy” variables are negatively correlated, meaning in most cases that the bigger share of a loan or deposit type is, the less share of other loan and deposit types becomes.

The pairwise correlations between bank normatives variables are not very high as well, except for correlation degree between H10.1 and H7 normatives (0.947). This is most likely to occur because of the similar ways these normatives are calculated. Both formulas include bank's capital as a denominator.

Table 6.2 Pairwise correlations between mandatory normatives

банкротство стратегия банковский

Considering all these, the VIF test is additionally conducted in order to reveal possible multicollinearity more accurately and, if necessary, to omit highly correlated variables. In the Table 7.1 it is clearly seen that variables H7 and H10.1 have quite a high level of multicollinearity, hence, the simultaneous usage of these indicators in regression model may lead to poor regression outcomes. In order to avoid this, the normative H10.1 is omitted which has led to favorable VIF values presented in Table 7.2.

Table 7.1 Variance inflation factor of H7 and H10.1

Table 7.2 VIF test of all independent variables

Finally, the independent group t-test has been conducted to determine whether the independent variables have a statistical difference between the means of two groups (failed and non-failed banks). The estimation is given based on the 95 per cent confidence interval, where all the variables' means statistically differ from 0 even on the 99.9 per cent confidence interval, except for two indicators (Table 8).

Table 8. Independent group t-test

3.4 Regression results

Table 9. Results of the panel logistic regression

Table 9 illustrates the outcomes of the model after the panel logistic regression with random effects was applied. As was stated in the methodology section, the table contains two models reporting the results as coefficients and odds ratios. This is made for clearer understanding of specifics of the binary logistic regression, as well as of the analysis outcomes. Although random effects are applied in regression, the months dummies were still included in order to specify all the financial indicators for the each of the 96 periods. The aim of this is to control the regression outcomes in terms of the gradual implementation of the new license revocation policy in relation to insufficient banks. Overall, 58 535 observations are included in the model that are clustered in 896 groups according to the number of banks taking part in the research.

Such a vast availability of the data enabled to conduct the quantitative analysis with 13 independent variables - six key variables representing bank strategies and 7 additional control regressors that help to reflect banks performance more expository. As a result, the crucial regression coefficients have been obtained that indicate their levels and directions of influence on the probability of occurrence of the event (bank failure).

Proceeding directly to the explanation of the coefficients values, the first key variable IL_to_TA, which stands for the share of individual loans in bank assets, has a negative coefficient of -0.839. That means that all else being equal, the strategy of increasing the share of individual loans in bank assets has a positive effect on bank's performance implying that the one-unit change in this variable leads to -0.839 change in the log-odds of the probability of bank failure. The odds ratio of 0.432 of the given regressor also indicates that if adopting this strategy, the failure of a credit organization is unlikely to happen (the odds ratio less than 1).

The second variable CL_to_TA indicating the strategic focus of increasing the share of commercial loans in bank's assets has a coefficient value of 0.340. Unlike the previous variable, this one has a positive sign and the odds ratio more than 1 (1.405). This means that based on the given dataset, banks that were focusing on crediting legal entities were experiencing more frequent license revocations with the one-unit change in the regressor leading to 0.340 increase in the log-odds of the failure probability.

The share of interbank loans in total assets (IB_to_TA) has even higher probability of survival if following the strategy in comparison to issuing individual loans. The value of -1.606 shows that those banks which assets constituted mainly of this type of loans had the lowest chances of default.

Summing up the regression results concerning bank strategies efficiency in terms of allocation of the assets in a portfolio, the focus on issuing interbank and individual loans positively affects bank performance, while concentrating on crediting legal entities is likely to lead a bank to failure.

The second group of key variables relates to a bank liabilities portfolio. Here the indicators CD_to_TL (share of commercial deposits in total liabilities) and IB_to_TL (share of raised interbank loans in total liabilities) both have negative coefficients of -0.767 and -1.424 respectively. Hence, the log-odds of the bank failure probability decreases when there is a one-unit increase of any of these strategies if all other things are equal. On the contrary, following the strategy of attracting deposits of individuals showed the higher chances of a bank to be deprived of a license, thus, to demonstrate poorer performance.

All the six coefficients reporting the levels of efficiency of various bank strategies are strongly significant on the 95 per cent confidence with the p-values > |z| equaling to 0.000.

Proceeding to additional control regressions used in the model, not all of them appeared to considerably affect the probability of bank failure. Among them, Return on Assets (ROA) showed a positive impact on bank performance with the 0.031 decrease in the log-odds of the event occurrence, considering a one-unit change in the variable. This makes sense, as the indicator shows the profitability of a credit organization in relation to its assets. Thus, the more money a bank generates, the more stable its activity is considered.

The values of bank normatives in majority did not have big influence on the model. It is assumed that the changes in these indicators were insignificant for the majority of the time periods (except for several shocks) in comparison to general financial reporting of banks.

The interpretation of additional indicators pointed out in the Table 7 is also required to demonstrate sufficiency and reliability of the model. Here the Prob>chi2 value confirms its statistical significance on the 95 per cent confidence interval with the value of 0.000. Based on the Wald chi-square outcome of 9 993.54, the Prob>chi2 indicator shows the probability of obtaining the given chi-square statistics if there was not any effect of all the independent variables on a bank failure. Taking into account this probability equals to zero (completely significant) and the statistical significance of all independent variables representing bank strategies, the null hypothesis can be rejected. This means that the way a credit organization structures its assets and liabilities portfolio does have an effect on the probability of default, hence, is capable to influence bank performance.

3.5 Postestimation of the model

3.5.1 Sensitivity analysis

For further check of the validity of the model, some postestimation techniques have been applied. Table 10 illustrates the results of sensitivity and specificity analysis that explain how well the model is able to indicate positive and negative outcomes of bank performance, i.e. to predict failure or survival correctly. Based on the general statistics of bank performance in 2012-2019 (511 failures and 385 non-failures), the accuracy of the model is presented below.

Table 10. Predictive abilities of the regression model

The final percentage of correctly classified outcomes (84.58 per cent) enables to admit that the majority of bank failures and survivals were accurately predicted. Thus, the forthcoming implications about viable bank strategies are believed to be trustworthy.

3.5.2 Analysis of marginal effects

The non-linear nature of relationships between independent and the dependent variable assumes different interpretation of initial regression results. In order to make such interpretation easier, as the change in the log-odds does not provide a clear understanding of actual influence on the probability of bank failure, the marginal effects of each independent regressor are estimated for the obtained model. As all the variables are continuous, the marginal effects will measure the instantaneous rate of change, i.e. how the probability of bank failure will be affected if an independent variable changes by one unit. These effects are more convenient to consider, as they enable to interpret the results as in usual linear regression model (Table 11).

The general direction of marginal effects of the independent variables is the same as when the log-odds values were analyzed. Concerning the asset portfolio, the variables IL_to_TA and IB_to_TA have a negative effect on the probability of bank failure, while CL_to_TA has positive interdependence with failure event.

The regressors CD_to_TL and IB_to_TL in the liabilities portfolio also have a negative effect on bank failure occurrence, whereas ID_to_TL may increase the chances of license revocation.

Table 11. Marginal effects

This analysis has eventually led to the formulation of main implications for the study about identifying efficient bank strategies.

1. Concentration on issuing loans to individuals decreases the probability of bank failure. With the one unit increase in share of individual loans in asset portfolio of a bank, its default probability decreases by 0.193.

2. Concentration on crediting legal entities increases the probability of bank failure. With the one unit increase in share of commercial loans in asset portfolio of a bank, its default probability also increases by 0.08.

3. Concentration on issuing loans to other banks decreases the probability of bank failure. With the one unit increase in share of interbank loans in asset portfolio of a bank, its default probability decreases by 0.373.

4. Concentration on attracting deposits of individuals increases the probability of bank failure. With the one unit increase in share of individual deposits in liabilities portfolio of a bank, its default probability increases by 0.174.

5. Concentration on attracting deposits of legal entities decreases the probability of bank failure. With the one unit increase in share of commercial loans in liabilities portfolio of a bank, its default probability decreases by 0.176.

6. Concentration on raising loans from other banks decreases the probability of bank failure. With the one unit increase in share of interbank loans in liabilities portfolio of a bank, its default probability decreases by 0.33.

Figure 3. Average shares of strategies followed by failed and survived banks

The visibility and veracity of the regression results is supported by Figure 3. For the given graph the average values of each share of loan and deposit types were taken. These shares were calculated separately among failed and survived Russian commercial banks 2012-2019 based on the dataset that was used in regression building.

Comparing each indicator of bank strategies between the defaulted credit institutions and those who are still on the market, the resilience of particular strategies can be observed. Taking into account the regression results illustrating the efficiency of structuring banks asset and liabilities portfolio in favor of individual loans, commercial deposits, issued and raised interbank loans, it is seen that all these strategies prevail in survived credit institutions.

On the contrary, strategies that likely to lead banks to failure (issuing commercial loans and attracting individual deposits) were higher adopted by failed banks (shares of 0,396 and 0,332 compared to 0,343 and 0,294 respectively).

The results are additionally supported in Appendix 3, where it is shown that “Rusfinance Bank” is mainly focusing on individual loans and raised interbank loans, which stand for its sufficient performance (Appendix 3). The performance of the “Yugra” bank, in turns, has experienced bad outcomes due to following inefficient strategies of crediting legal entities and attracting individual deposits (Appendix 1).

Summing up the results of the study, the diagram of analyzed bank strategies is presented (Figure 4).

Figure 4. Efficient and inefficient bank strategies

4.Discussion

4.1 Contribution of the research

The given study contributes to the relevant research field due to several reasons. Firstly, the hypothesis of bank strategies ability to affect its probability of failure was proved. This has led to an appearance of the new approach to critical evaluation of various bank strategies based on the identifying the degree of their efficiency. Such evaluation has helped to analyze the period of massive bank license revocations in Russia during 2013-2019.

Overall, such period in the history of Russian banking system can be considered a natural experiment that was aimed to test the viability of various bank strategies, which is impossible for the majority of countries of the globe. Therefore, the results obtained from the study may be treated as universal. They are interesting and beneficial for a wide range of researches and practitioners all over the world.

Among those there may be financial experts, investors and other economic professionals who would like to manage their personal assets more rationally or to widen general knowledge about commercial credit organizations. Also, the outcomes can be taken into account by bank themselves in order to form new or modify their current strategies as a measure of failure prevention and enhancement of financial position. In addition, the financial regulator may consider the outcomes of the study for making implications about those majority of banks who have been deprived of a license, which may serve a tool of proper identification of credit institutions that are exposed to risk of default in the future.

4.2 Comparison of the research outcomes with prior studies

The comparison of the study with prior ones will demonstrate the discovery of new insights regarding the causes of bank failure and enable to evaluate the degree of the research contribution.

As the intention of the research was comparatively new, because namely the evaluation of bank strategies regarding structuring asset and liabilities portfolio has not been deeply studied before, this comparison may be conducted only in the framework of some parts of other prior studies.

In the study of Karminsky and Kostrov (2017) that attempted to forecast failures of banks with negative capital, one of the regressors that was used as a control variable in the analysis was “Share of loans to individuals in assets”. Being considered a type of market strategy, the indicator showed the negative interdependence with the probability of default (-5.29***).

Wheelock and Wilson (2000) discovered that the higher the share of commercial and industrial loans in bank's assets, the more chances of failure it has. This finding was further supported by Mamonov (2017), which research was aimed to identify the reasons of bank capital holes appearance. According to the author, such capital holes tend to be bigger when a credit organization has high turnovers of commercial loans (coefficient of 0,026***) and big focus on attracting expansive client deposits that are further placed in cheap corporate loans (coefficient of 0,032***).

As for the additional control regressors, then the positive effect of ROA on bank performance was widely admitted (Cole & Lawrence, 2012; Lanine & Vennet, 2006; Zakirova et. all, 2018).

Implying that the aim of these studies was different, the coincidence of parts of the results in terms of the key regressors role in bank performance enhances the significance of the obtained model.

4.3 Limitations and future research

The research still has the limitation that to some extent is able to interfere with the outcomes to be highly reliable. As the dataset containing bank list is relevant for the January 01, 2012, the bank number dynamics from 2012 to 2019 will be reflected and studied only among these credit institutions without taking into account banks that appeared within this time period. The point of concern here lies in possible omission of bank failure cases potentially crucial for the study that could add some benefit to the research reliability.

However, the number of new incumbents on the bank market is not that big to be able to significantly affect the results of the regression if adding them to the data sample. Nonetheless, to deal with such kind of limitations future research is, undoubtedly, needed, as each country has its specific banking system where various strategies can be efficient in different ways. Analysis of more diverse data samples will result in greater accuracy and applicability of the approach.

Conclusion

This research aimed to discover efficient strategies of forming asset and liabilities portfolio of a bank. For the analysis, the eight-year period from 2012 to 2019 was considered, when there were numerous cases of commercial bank license revocations initiated by the Central Bank of Russia. The econometric model with an application of panel logistic regression showed the statistical significance of the research hypothesis and has proved that the way a bank distributes shares of different loan and deposit types affects the probability of default. As a result, the strategies of focusing on issuing individual and interbank loans, as well as attracting commercial deposits and raising interbank loans appeared to have a positive impact on bank performance. However, the concentration on commercial loans and individual deposits showed the increasing probability of failure of credit institutions that follow such strategies.

Several crucial implications aroused from the study conduction. Certainly, banking system of any country has to be healthy and favorably functioning. It is one of the main missions of all financial regulators to revoke licenses from organizations that demonstrate poor performance, thus, endanger the welfare of country's households and general condition of an economic system.

Nonetheless, such a radical and severe policy of the CB of Russia raises the question of what were the key factors of removing a great number of incumbents from the market? Namely intending to go into the matter of this, the identification of the most vulnerable strategies was as important as revealing the viable ones. There is a hope that the new findings will have a potential to be considered as a tool to follow performance of banks more informatively.

References

phenomenon banking credit

1.Alves, A. J., Dymski G.A. and Paula, L.-F. (2008). Banking Strategy and Credit Expansion: A Post-Keynesian Approach. Cambridge Journal of Economics, 32(3), 395-420. doi:10.1093/cje/bem035

2.Arena, M. (2008). Bank failures and bank fundamentals: A comparative analysis of latin America and east Asia during the nineties using bank level data. Journal of Banking and Finance, 32, 299-310.

3.Banki.ru [Electronic Source]. Retrieved from https://www.banki.ru/

4.Banki.ru [Electronic Source]. Kak nachinalis' kommercheskie banki v Rossii. Retrieved from https://www.banki.ru/news/bankpress/?id=5162110

5.Banki.ru [Electronic Source]. Obzor: bankovskiy sector v 2018 godu. Retrieved from https://www.banki.ru/news/research/?id=10890092

6.Berardi, S., Tedeschi, G. (2017). From Banks' Strategies to Financial (in)Stability. International Review of Economics & Finance, 47, 255-72. doi: 10.1016/j.iref.2016.11.001

7.Central Bank of the Russian Federation [Electronic Source]. Retrieved from https://cbr.ru/

8.Central Bank of the Russian Federation [Electronic Source]. Mery po ozdorovleniu bankov. Retrieved from http://www.cbr.ru/credit/PrBankrot/

9.Civil Code of the Russian Federation (part 1) (1994). № 51-FZ (ed. as of 16.12.2019). Sobranie zakonodatelstva RF, art. 65.

Размещено на Allbest.ru