ANALYSIS OF EMPIRICAL DETERMINANTS OF CREDIT RISK IN THE BANKING SECTOR OF THE REPUBLIC OF SERBIA

Cilj ovog rada je otkrivanje i analiza empirijskih determinanti kreditnog rizika u bankarskom sektoru Republike Srbije. Rad je baziran na analizi rezultata primene linearnog regresionog modela, u vremenskom periodu od trećeg kvartala 2008. do trećeg kvartala 2014. godine. Tri su osnovna nalaza istraživanja. Prvo, veća kreditna aktivnost banaka doprinosi povećanju udela visoko rizičnih kredita u ukupnim kreditima (zakasneli efekat od 3 godine). Drugo, rast kreditnih plasmana u odnosu na depozite doprinosi većoj izloženosti banaka kreditnom riziku. Treće, faktori koji smanjuju izloženost kreditnom riziku banaka su rast profitabilnosti, rast kamatnog spreda i realni rast bruto domaćeg proizvoda. Imajući u vidu sveukupne tržišne okolnosti i dinamiku privrednog oporavka zemlje, na osnovu rezultata se može izvesti generalni zaključak da će u narednom periodu pitanje visoko rizičnih kredita u Republici Srbiji biti aktuelan izazov i za kreditore i za korisnike kredita.


Introduction
In the last ten years, the banking sector of the Republic of Serbia has faced numerous challenges.Insufficiently rapid recovery of the economy (Ranisavljević, Vuković, 2015), partially successful privatization of state enterprises, the global financial crisis and political instability are just some of the factors that threatened to jeopardize the stability of the domestic banking sector.However, despite all of the above-mentioned factors, the banking sector has remained stable.The conservative policy of the National Bank of Serbia (NBS), the focus of banks on traditional deposit and lending activities, developed competition in the banking market and professional attitude of banking managers towards their business operations (Matić, 2015) resulted in stability, which further led to a high degree of liquidity and a high capital adequacy ratio.
The stability of banks, in operational terms, is most dependent on the quality of risk management (Saunders, 2011;Barjaktarović, 2013;Račić et al, 2014).The exposure to credit risk depends on many factors including the relation between business operations and the regulations in force, selection of suitable credit strategies and policies, level of development of the country and macroeconomic developments in the market in which banks operate.The empirical indicator which is commonly used for determining the level of banks' exposure to credit risk is the share of NPLs in total loans (asset quality).It is believed that the share of NPLs testifies that there is a growing exposure to credit risk and vice versa.
Chart 1 shows the movement of the share of NPLs in the total loans of the national banking sector.It is notable that under the impact of the global financial crisis the share of NPLs in the total loans showed an upward trend.The reaction of the banks to the crisis led to the reduced volume of investments, liquidity growth and reduction in active and passive interest rates.This trend could result in the fall in the share of NPLs, and credit expansion that could, by pumping fresh liquidity into the real economy, encourage the future economic growth (Report on the results of the survey on lending activities, 2015).
The aim of this paper is to identify and evaluate some of the significant empirical determinants of the exposure of banks to credit risk and thereby contribute to the monitoring and analysis of exposure to this type of risk.The subject of the research are NPLs as well as the performance of the banking sector of the Republic of Serbia in the area of credit risk management, published in the quarterly reports of the National Bank of Serbia on business operations of banks in the Republic of Serbia, in the period from 09/30/2008 to 06/30/2014 (The banking sector of Serbia -Quarterly Reports).The study is based on the general hypothesis that the macroeconomic and internal characteristics of banks determine the level of NPLs in the total credit portfolio of the banking sector of the Republic of Serbia.

References review
The identification and analysis of the empirical determinants of credit risk have been attracting the attention of researchers around the world since the second half of the twentieth century.The research on the influence of numerous factors have resulted in a general assessment that losses from the growth of exposure to credit risk can be caused by many microeconomic and macroeconomic factors.In the beginning, the majority of studies were carried out on a sample of banks operating in the territory of the United States, but later the research expanded to include banks in other parts of the world.The factors that are usually analyzed as empirical determinants of credit risk are gross domestic product (GDP), inflation, unemployment, number of branches and number of employees in banks, return on assets (ROA), return on equity (ROE), net interest income, size of banks, the level of capital, solvency, credit activity, the size of the deposit base and the like (Makri et al, 2014;Ganić, 2014, Das, Ghosh, 2007).
Most studies on banks in the United States led to the conclusions that reducing the operating business efficiency of borrowers leads to a rise in the share of NPLs and that less efficient and less capitalized banks tend to take higher risks (Kwan, Eisenbis, 1997;Berger, DeYoung, 1997).On the other hand, the research conducted in India (Rajaraman et al, 1999) resulted in the findings that microeconomic and macroeconomic characteristics have an equally important influence on the movement of the NPLs share.Thus, the increase in the number of NPLs positively correlated with the worsening of the macroeconomic conditions and with micro-economic factors such as the reduction in operational efficiency of the borrower and the expansionist conquest of the new geographical markets by banks.The research on banks operating in Spain (Salas, Saurina, 2002) showed that GDP growth affects the growth of the capacity to repay the loan of business entities and citizens, which leads to a drop in the share of NPLs in the total credit portfolio.The research concerning the Greek banks showed that there is a significant negative relationship between profitability indicators and movements of NPLs (Dimitrios et al, 2012).In the end, the conclusion that can be drawn from the research on macroeconomic determinants of credit risk is that NPLs show counter-cyclicality.This means that in times of economic growth, their share in the total loan portfolio decreases, and vice versa (Ganić, 2014).The growth of lending activities and bank interest rates are the characteristics of the upward stage of the economic cycle, which after its transition to the recession stage cause a growth in the share of NPLs (Jimenez, Saurina, 2005).The results of many empirical studies confirm this, testifying that a few years after the credit expansion there is an increase in the share of NPLs in the total credit portfolio of banks (Salas, Saurina, 2002;Alihodžić, 2015).

Methodology and data
The previous studies of the determinants of credit risk are primarily based on the use of dynamic panel models, which are used for research purposes when the dependent variable values from the previous periods stand as independent variables (Arellano, Bond, 1991;Bond, 2002;Baltagi, 1995;Baltagi, 2011).Due to the lack of data on the share of NPLs in the total credit portfolio at the level of individual banks in Serbia, in this research we applied the ordinary least squares model (OLS) rather than the dynamic panel models.This fact is also the greatest limitation of the research because the application of OLS models can question the quality of the obtained score.
The applied regression model can be illustrated by the following formula: with: β 0 -constant of the model, β 1… β 9 -the scores of the regression coefficients with independent variables that describe the nature of the intensity of their effects on the dependent variable, and ε i -residuals of the model.The model is applied on the basis of an analysis of quarterly data of the consolidated bankarskog sektora u Republici Srbiji, objavljenih u izveštajima Narodne banke Srbije, uz pomoć softverskog paketa STATA 11.Istraživanjem je obuhvaćen vremenski period između 30.09.2008.i 30.06.2014.godine.Opis korišćenih varijabli je prikazan u tabeli 1.
When interpreting the results, it is important to take into account the fact that the dependent variable model includes the variations of the share of the proper loans.Thus, the growth in the value of pl variable indicates a decrease in the share of NPLs in the total credit portfolio of banks, and vice versa.

Checking the fulfilment of the standard assumptions of the model
The application of OLS model results in reliable assessments only in case of the fulfilment of the standard assumptions of the model.Otherwise, the assessments of the regression coefficients are biased (Mladenović, Petrović, 2007).For this reason we examined the fulfilment of the basic standard assumptions of the multiple regression model.
The first assumption of the OLS model refers to the absence of multicollinearity.Multicollinearity implies a strong correlation between two or more independent variables of the model.The fulfilment of this assumption was tested by the VIF test (Regression with STATA Web book).The variables whose VIF values are greater than 10 should not be included in the model.Also the value of 1/VIF is taken into consideration, which should be higher than 0.1.The results of the VIF test are shown in Table 2.
According to the results shown in Table 2 it can be concluded that there is no multicollinearity in the model, i.e. all independent variables can be included in the model.
Another assumption that needs to be fulfilled refers to the normality of the distribution of residuals.The fulfilment of this assumption is tested using the Shapiro-Wilk W test, which starts from H 0 that the residuals have a normal distribution.According to P-values (P=0.99669) it can be seen that H 0 is confirmed, which means that the residuals have a normal distribution.The visual representation of the distribution of the residuals of the model is given in Figure 1.  4.
Based on the P-value of Breusch-Pagan/ Cook-Weisberg test that starts from H 0 and shows that the variance of the residuals is constant, we can conclude that there is no expressed heteroscedasticity in the model (Prob>chi2=0.4239).The visual representation of scattering of random errors around its mean value is shown in Figure 2.
The last step in the statistical processing of data relates to the elimination of autocorrelation of residuals, which is often present in the case of time series.Durbin-Watson test is used to detect autocorrelation, while for the purpose of its elimination we applied Cochrane-Orcutt model (Newbold, 2010;Mladenović, Petrović, 2007).The obtained results are shown in Table 5.

Analysis of the research results
The results support the assessment that the applied model is suitable for the analysis of the relationship between the observed variables (Prob>F=0.0000).Based on the value of the coefficient of determination it can be concluded that 93.6% of the variation of the share of NPLs in the total loans is explained by the variations of the observed regresses (Adj R-squared=0.9360).The variables that in the statistically significant level determine the proportion of NPLs are pllag, loans1, loans3, spread1, roe, lodep and gdp.
The variations of lending activities are the first characteristic that determines the exposure of banks to credit risk.The paper examines the impact of the banks' credit activity in the period of one, two and three years ago in relation to the moment of observing the level of NPLs.In other words, we considered the delayed effect of variations in credit activity on the observed exposure.The estimates support the conclusion that one year after the credit growth there is a reduction in the share of NPLs (Ganić 2014), while after three years there is an increase in their share in total loans (Salas, Saurina, 2002).We will now examine some of the significant factors contributing to this.The banking market in Serbia is showing signs of saturation.Despite the low overall level of concentration (Herfindahl-Hirschman Index), the market share of the top five banks in the balance sheet recorded a continuous growth, which significantly determines the level of NPLs in the total loan portfolio (Dimić, 2015).Also, it is important to draw attention to the fact that banks predominantly grant loans with a foreign currency clause, causing the problem of exposure to currency risk, which during the analyzed time period contributed significantly to the rise in the share of NPLs.The borrowers' primary sources of repayment are in dinars, while their obligations are indexed in FX (i.e.EUR or CHR exchange rate), which fluctuates and threatens the creditworthiness of the borrowers.
A contribution to the credit standards improvement can also be the reduced difference between the banks' interest-based revenues and expenditures.The profitability of domestic banks relies on interest income, due to the predominant orientation to the credit-deposit activity.This means that a reduction in the difference between interest-based revenues and expenditures threatens to jeopardize their profitability.A decrease in profitability may require an adjustment of the credit policy of banks to relax the lending standards, with the aim of maintaining profitability at the desired level with higher credit risk premiums (arising from risky investments and being included in the interest rate spread).Bearing in mind the overall market conditions, it can be concluded that the easing of credit standards increases the exposure of banks to credit risk.This is confirmed by the research results that support the assessment that the reduction in interest rate spread results in an increased share of NPLs in the total credit portfolio of banks.
Another feature of banks that determines their exposure to credit risk is their profitability (ROE).The previous studies estimate that profitable banks are less motivated to expose themselves to risky credit placements (Makri et al, 2014).The results of our research confirm the assessment that the reduction in profitability affects the growth of the share of NPLs in the total credit portfolio, which reduces the possibility to cover credit losses from operating income (covered from equity).This is supported by the negative correlation between the rate of capital adequacy ratio and the share of NPLs, which was, within the covered time period, ρ = -0.53(Pearson coefficient), noting that at the beginning of 2013 the value of this coefficient was as high as ρ=-0.82.
Variations in the loans-to-deposits ratio also represent a significant determinant of the observed exposure.The previous studies show that the trends of this ratio are positively correlated with the trends of the share of NPLs (Ganić, 2014;Makri et al, 2014).This is in accordance with the results of our study, which assessed that the growth of credit activity in relation to the deposit base affects the increased exposure of banks to credit risk.One of the important reasons for this is the banaka kreditnom riziku.Jedan od značajnih razloga za to jeste činjenica da su domaće banke prelikvidne i da nisu zainteresovane da prikupljaju depozite jer nemaju po njihovom mišljenju zdrave projekte u koje bi plasirali sredstva (kredite), dok vlasnici viška novčanih sredstava nisu motivisani potencijalnim prinosom (visinom kamatne stope) da investiraju u depozite poslovnih banaka.
The last significant variable of credit risk covered by this research is the trend of real GDP growth.The estimated negative relationship between the movement of the real GDP growth and the share of NPLs is also in line with the previous findings (Das, Ghosh, 2007;Jović, 2015;Makri et al, 2014).The GDP growth causes an increase in the funds for repayment of loans by retail and corporate borrowers, which results in a reduction of the share of NPLs in the total credit portfolio.When it comes to the economic growth in the Republic of Serbia, after the impact of the global economic crisis, the national economy was mainly faced with the periods of stagnation and recession, which is one of the main reasons for the higher banks' exposure to credit risk during this period.

Conclusion
The study of empirical determinants of exposure of the banking sector of the Republic of Serbia to credit risk led to the conclusions that confirm the general hypothesis.This means that macroeconomic and internal characteristics of banks in statistical terms significantly determine the level of NPLs in the total credit portfolio of the banking sector of the Republic of Serbia.The results support the assessment that the growth of credit activity of the banking sector after a period of three years resulted in the increase in the share of NPLs in the total credit portfolio.The same effect on the share of NPLs is realized through the reduction in the difference between interest-based revenues and expenditures, and an increase in credit activity in relation to deposits.Also, it can be concluded that profitable banks have a smaller share of NPLs, which means that they are less exposed to the credit risk.
Bearing in mind the overall market conditions and the dynamics of economic recovery, it can be concluded that in the future the issue of NPLs in the Republic of Serbia will present a challenge both for lenders and borrowers.The focus of managers should be on addressing the following questions.Firstly, how to prepare the appropriate "Early Warning System", identify the potential high-risk loans and address them preventively.Secondly, what mechanisms should be used to provide the borrowers with the suitable restructuring and thus enable them to continue to operate successfully if prevention does not help?Finally, how to identify at what point the turn-around managers should be involved, and how to sell NPLs to the specialized investment funds that will help the borrower to successfully overcome the crisis and continue to operate profitably.
The next assumption relates to the checking of homoscedasticity of random errors.Random errors should show the same degree of scattering around their mean value.If the variance of random errors differs significantly, the random errors are called heteroscedastic.Heteroscedasticity of residuals was checked by applying Cameron & Trivedi's decomposition of IM test and Breusch-Pagan/Cook-Weisberg test.The results of these tests are shown in Table

Figure 1 .
Figure 1.The representation of the distribution of the residuals of the model -Kernel density estimate

Table 1 .
Description of variables of the OLS model ROEReturn on Equity Ganić, 2014; Makri and assistants, 2014 Lodep Loans to Deposit Ratio Ganić, 2014; Makri and assistants, 2014 GDP Growth Rate of Real GDP Das, Ghosh, 2007; Makri and assistants, 2014 Public Finance Bulletin (www.mfin.gov.rs)Source: The overview prepared by the authors, NBS, Ministry of Finance of the Republic of Serbia

Table 2 .
Checking the existence of multicollinearity in the model -VIF test

Table 3 .
Checking the regularity of the distribution of residuals -Shapiro-Wilk W test Source: The calculation made by the authors

Table 4 .
Tests of heteroscedasticity of residuals Source: The calculation made by the authors