Resilience of the Polish Banking System to Credit Risk: Macroeconomic Perspective

Last Updated: 26 Jan 2021
Pages: 27 Views: 134
Table of contents


1. Introduction

The most significant type of risk in every Polish bank’s loan portfolio is the credit risk. Credit risk is a type of performance risk through a financial contract (such as a loan or bond). In case of a loan, the risk arises from the potential failure of a borrower to honour its obligations from the contract: repaying borrowed amount, the principal and relevant interest payments. For banks whose primary focus is on lending and credit activities over capital market activity, credit risk is more important other types of risk. This is the case with commercial banks, which dominate the Polish banking sector.

Order custom essay Resilience of the Polish Banking System to Credit Risk: Macroeconomic Perspective with free plagiarism report

feat icon 450+ experts on 30 subjects feat icon Starting from 3 hours delivery
Get Essay Help

Credit risk is found on the asset side of the banking book (loans and bonds held to maturity) Polish banking sector pirmarily deals with the credit risk from loans. This is essentially due to the fact that corporate bonds comprise only a small portion of the Polish commercial banking sector, and are therefore only marginally represented in the commercial banks’ balance sheets. Credit risk primarily arises in the claims on non-financial sector customers (households and corporates) which, leaving out the general sector of government, constitute almost half of banks’ assets. This type of risk taken on by the commercial banking sector is important to the central bank, considering that significant credit losses lower banks’ capital and negatively influence financial sector performance. Such scenarios can futher negatively influence corporates and households.

Economic literature to date highlights the significance of the connection between the loan losses of banks and business cycle. Such relationship is crucial in analysing financial stability. Thus, the objective of this dissertation is to examine the relationship between the loan losses of the Polish commercial banks’and the Polish economy’s key macroeconomic trends. ADD in the years which represent WHAT. In investigating this relationship, the bank-specific traits will be considered, as they can indicate the differences among individual banks’ loan portfolio losses.

Rephrase: The thesis is comprised of 5 parts. In the first chapter, the relevant literature review is laid out in relation to this specific body of work. Secondly, the model used and appropriate variables are explained. The third chapter presents the empirical estimation and results. The forth segment undertakes and discusses stress test.

The paper consists of 5 sections. The first one presents a short review of the relevant literature. Next, we discuss the variables used in the model. The third section contains the empirical results. The fourth part presents the results of a stress-testing exercise. The last chapter lays out the conclusions reached and propositions for additional research in the future.

2. Literature review

The period of 1990s embodies works attempting to develop models which could correctly capture the risk arising from banks’ lending practices. To date, literature on empirical studies of banks’ loan losses broadly offers two distinct streams of academic work. The first stream represents those models intended to explain probability distribution of potential losses on a bank’s credit portfolio. The typical illustrations of such credit risk modelling method originated in the second part of the ‘90s, and were presented to the public by the eminent financial institutions: Credit Suisse’s CreditRisk+ model, CreditMetrics/Credit Manager offered by J.P. Morgan, KMV model from the KMV coporatation, and CreditPortfolioView by McKinsey & Co (as explained in Crouhy et al. [15], CSFP [38] and Wilson [44]).

Although these models are highly diverse in their apporach, all of them, except CreditPortfolioView, are unconditional to macroecnonmic variables. They explain loan loss probablility distrubution without examining the relationship between macroeconomic indicators and individual borrowers’ credit quality. As a result, only CreditPortfolioView fully illustrates the relationship between the extent of loan losses and the busines cycle (to some extent this is also the case with KMV model). Usually, all aforementioned types of models are used to calculate credit Value-at-Risk (VaR) and are a helpful instument for capital allocation and relevant management of risk.

The second stream of literature represents those models that aim to present a practical connection among credit risk measures (loan losses and probability of default), macroeconomic conditions, and other components, e.g. the individual banks’ characteristic traits and strategies. As such, these models can provide point forcasts which, when based on a specific macroeconomic scenario, can be a useful tool to policy makers and supervisors for macro stress testing. Macro stress tests were particularly promoted by the joint initiative of the IMF and World bank in 1999 through Financial System Assessment Program (FSAP) and represent a conventional quantitative assessment of financial system resilience. This second stream of models is also applied in discovering which macroeconomic elements have impact on loan losses, and as such these models can be used to support monitoring of financial stability.

As the purpose of this dissertation is to discover the set of micro and macroeconomic variables that EXPLAIN the development of loan losses within the banking system of Poland, the methodological approach used in this dissertation comes from the second stream of literature.

This stream of literature is usually devided into three subgroups:

A. Econometric modelling of loan losses with the data on individual bank-level and macroeconomic data in order to model the percent of nonperforming loans (classified loans) at individual bank level (for instance Quagliariello [39] REPLACE w/reference they used Gizycki [22]).

B. Economectric modelling of loan losses or percent of nonperforming (classified) loans by combining data on aggregate banking sector-level into one country (e.g. Hoggarth, Logan and Zicchino [27] OR Whitley, Windram and Cox [42],) or multiple countries (e.g., Pesola [35] and [36], Bikker and Hu [10]).

C. Economectric models which illustrate the bankruptcy probability of a firm, on the industy average-level or individual firm-level. Collectively with the data on banking sector exposure to the examined industries/firms and data on prices of collateral (usually estimated by real estate price indexes) this subgroup of models is useful for forecasting and explaining loan losses. (Sorge and Virolainen [40], Boss [12], Andreeva [4], Bernhardsen [8]].

Studying the results that the aforementioned examples of second stream of literature offer, it is possible to recognise the set of variables that can estimate those economic processes which are responsible for the development of loan losses and quality of loans.

relationship between a set of macroeconomic indicators (e.g. GDP growth, interest rates) and either loan loss provisions or non-performing loans.

2.1 Econometric models using data on individual-bank level

Among earlier research examining the factors that influence loan quality, Robert Clair’s paper based on Texas banks data, indicates there is a connection between “extreme” loan growth (relative to the business cycle stage) and the quality of loans. His paper illustrated that fast loan growth reduces quality of loans, whereby this effect is most prominent in the case of banks which have low capital adequacy ratio.

In examining which components have an impact on charges to loan loss provisions among major banks in the UK (observing the period pning 22 years), Pain [34] differenciated between mortgage and commercial banks. In case of commercial banks, the key factors having an impact were: world and the UK GDP growth (negative impact), average rate of loan growth for the entire banking sector and real interest rates level. Concerning bank-level variables, three were found as significant (all three had positive influence): cost-to-income ratio, share of loans which extended to the sector of commercial real estate (considered to be a high-risk sector), and finally the concentration of loan portfolio (based on Herfindahl-Hirschmann index, calculations used industry shares). In case of mortgage banks, significant macroeconomic variables were real interest rates and UK GDP growth, while significant bank-level factors were cost to income ratio, share of portfolio indicating mortgage-secured loans and finally net interest margine (the increase in this variable was thought to represent expansion of banks’ business to riskier clients).

In order to explain the flow of new nonperforming loans and new loan loss provisions (prepresented as a percentage of loan portfolio), Quagliariello developed both static and dynamic models using panel data on Italian banks in the period 1985-2002. In his paper, the increase in nonperforming loans is described by the following factors: long-term interest rates level (positive influence), GDP growth (negative influence), spread between deposit and lending rates (average of the sector; positive), cost-to-income ratio (negative), and finally loan growth at individual bank-level (negative influence, where growth in loans is indicative of solid business conditions). Quagliariello’s paper explains the flow of new loan loss provisions by the following factors: flow of new nonperforming loans, long term interest rates, spread between the rates of deposit and lending, GDP growth, stock market index (serving as proxy for business cycle), and profitability of a bank (determined by ROA).

With the purpose of describing the share of nonperforming loans in the loan portfolios of Spanish banks between 1984-2002, Saurina and Jimenez [28] applied a panel data model. Concurring with other relevant research, the authors found real interest rates had a positive influence, while the lagged GDP growth had a negative impact. Loan portfolio concentration and share of secured loans were both found to have an important influence. The concentration of loan portfolio was measured based on Herfindahl-Hirschmann index, whereby higher loan portfolio concentration increases the share of nonperforming loans. This paper also examines the influence loan growth rate has on nonperforming loans. The authors established that increase in the rate of loan growth negatively impacts the quality of loans (with 4-year lag). They further found that loan quality is also influenced by the higher growth in loans at the bank-level, when compared to the banking sector average. This model was estimated based on Arellano-Bond approach, considering an autoregressive component was used in the model.

Based on the work of Bikker and Metzemakers [11], Kearns [30] explains the flow and share of loan loss provisions as percentage of loans on the case of 14 Irish banks for a 21-year period. In his panel model, the significant risk factors are: unemplyment rate, GDP growth, loan growth, share of income before tax, share of loans in assets and loan loss provisions in net income from banking operation (which is an indicator of banks’ income smoothing practices) .

In the original model based on twenty-nine OECD countries’ bank-level data, Bikker and Metzemakers [11] additionally examined the impact of banks’ capital position (represented by the share of capital in balance sheet total), and found this variable to have negative influence in their simulation. The authors explained this result by existing regulatory framework which allows for certain provisions to be incorporated in bank capital for regulatory purposes, which enables the banks to increase their capital adequacy rations by the means of higher loan loss provision charges.

The potential connection between bank capital resources, loan quality, and cost efficiency is investigated by Berger and DeYoung [7], who suggested 4 hypotheses:

1) Deterioration in loan quality is caused by low cost efficiency. This hypothesis is called “bad management” as weak control of costs represents a sign of inadequate management skills. Such lack of adequate skills is equally illustrated by poor credit risk management, causing lower loan quality.

2) Lower cost efficiency is caused by a deterioration of loan quality. According to this hypothesis, dubbed as “bad luck”, a weakening in loan quality due to an exogenous shock (e.g. bankruptcy of major borrower) increases bank’s expenses i.a. on collection, causing lower cost efficiency.

3) Deterioration in loan quality is caused by low capital adequacy (“moral hazard”). With low capital adequacy there is a likely supervisory action and/or higher probability of bank’s bankruptcy, which entices management to gamble and take on more risk in order to boost up returns, which in turn, leads to lower loan quality (higher borrower risk = lower quality loan in portfolio).

4) Deterioration in loan quality is caused by the high cost efficiency. According to this hypothesis, known as “skimping”, if high cost efficiency is actually the result of diminishing resources allocated for risk management, then such policy leads to inferior loan quality.

In order to verify the four aforementioned hypotheses, authors used Granger causalty tests using panel dataset of US banks for a nine-year period. Berger and DeYoung found instances of “moral hazard” for banks with inadequate capitalisation, but couldn’t establish sufficiently convincing evidence to choose among one of their hypothesis linking loan quality and cost efficiency.

Another comparable assessment was conducted by Williams [43] on the basis of savings banks from six Western European countries for the period of 1990-98, where the results advocated that “bad management” is the most common occurence in European commercial banking.

In addition to variables used in previously mentioned research, a model offered by Gizycki [22] found that another statistically significant variable influences loan quality: lagged share of interest payments in the income of firms and households. Based on Australian bank data in the period 1990-99, Gizycki examined the share of classified loans in banks’ assets. This model also included a commercial real estate price index as a proxy for the influence of real estate boom on the quality of loans. Another factor was the share of construction in GDP, as it demonstrated that higher share has a negative influence on loan quality.

Similarly to the majority of other research, Gerlach, Peng and Shu [21] in their examination of Hong Kong banks, identify factors that influence the share of NPL in the retail banking portfolios. The authors found significant influence of nominal interest rates, inflation, GDP growth, and changes in the price of real estate. Observing the last variable, the results indicate that in banks which hold portfolios with high share of real estate loans, the quality of loans is less influenced by the real estate prices. In their conclusion, real estate prices are understood as a business cycle indicator, because of the observed effect they produce under hypothesis that the quality of real estate loans is less influenced by business cycle than other types of loans.

2.2 Econometric models using aggregate-level data

In order to explain the share of classified loans in portfolios of Spanish savings and commercial banks (1981-2001), J. Saurina and Delgado (2004) used cointegration technique and ECM (error correction model. The results indicated that interest rates level and GDP growth are significant in the short-term as well as long-term equations. The authors didn’t find statistically significant loan growth or debt burden of household and firms .

Contrary to these authors, Windram, Cox and Whitely found that loan quality is significantly influenced by the interest burden of households. They developed ECM models to explain the loan quality of credit card loans and mortgage loans in observing UK commercial banks (period of 1985 – 2002). Similar results were obtained by Logan, Hoggarth and Zicchino [27] regarding the corporate bankruptcy rates and loan quality.

Gambera [20] applied VAR models to examine loan quality on the sample of banks from the the Chicago Federal Reserve District. Significant variables found to be explenatory of the loan quality were: unemployment rate, business cycle indicators, borrower income levels, and the number of bankruptcy filings.

Comparable VAR model was applied on the case of Czech Republic economy, to examine the relationship between loan quality and macroeconomic indicators. The results indicated that worsening in loan quality leads to increase in inflationary pressure, but a decrease in the unemployment rate. These results can be clarified if the worsening in loan quality is interpreted as the sign of looser credit policy implemented by banks.

Observing Polish foreign exchange loan quality, research by Zochowski et. al. applied VAR models separately for household loans and corporate loans. The latter included nominal f/x rate (PLN/EUR), GDP growth and export growth rates, while the model on households loan quality replaced export growth rate with the yearly change in the unemployment rate. These two models similarly pointed to the negative influence of local currency depreciation and economic slowdown on the quality of foreign exchnage loans.

Finally, in explaining the relationship between loan losses or loan quality and macroeconomic indicators, several research papers combined banking sector data from multiple Western European countries. Specifically, Pesola constructed a modelling approach to explain the elevated level of loan losses initially in Scandinavian[35], and later in Western European countries. To account for unexpected shocks, models included ‘surprise’ variables (which are obtained as the difference between realised and expected values), confirming the macroeconomic influence on the amount of loan losses.

2.3 Probability of Default (Bankruptcy probability) Models

This part of academic research examines factors influencing loan losses by the means of default probability models, which are able to forecast loan losses of a bank based on its exposure to specific classes of borrowers, and its expectiations concerning loss given dafault.

An illustration of this approach can be found in the work of Marco Sorgea and Kimmo Virolainenb, who constructed probability of default models for six different sectors of economy of Finland and found significant influence of interest rates, GDP growth and sector-level leverage. [40] The authors further applied the approach to model loan loss probability distribution of a given loan portfolio, assuming explenatory variables under AR process. This type of modelling was also applied by Michael Boss from the Austrian National Bank.

The influence of macroeconomic variables on the probability of default can also be examined based on the data from individual borrowers. Sampling significant Polish bank exposures, Glogowski found that liquidity and profitability (on the level of industry), interest rates and GDP growth were all significant.

2 .4 Conclusions from the literature review

Based on the variables extensively applied in the aforementioned papers, several concepts can be identified as fundamental for modelling bank loan losses:

Debt service cost, as the cost of borrower’s (firms or households) payments on outstanding debt (consisting of interest payments and repayment of principal). It is commonly proxied by the change in real or nominal interest rates.

Debt burden: debt as a percent of income; represents borrowers’ sensitivity to changes in debt service cost and their income. The tools most frequently used for approximation are: leverage, debt servicing ratio (ratio of interest payments to income), loans to GDP ratio, debt to income ratio (for firms) and household debt to disposable household income ratio.

In particular, debt service cost-to-income ratio is used extensively by the banking industry in determining prospective borrower’s risk level. Nevertheless, the application of other comparable tools which use aggregate income data, whereby data on borrowers income is not separated from income of other agents can lead to incorrect conclusions. This is particularly the situation in Poland and other emerging economies with significantly expanding parts of population using banking services. Under this scenario, increasing aggregate debt burden is rather the result of progressively more borrowing customers, which is not necessarily the increase in credit risk.

Borrower income, relating to the changes in free cash flows available to borrowers for the repayment of interest and principal on debt. This concept further incorporates unemployment rate, earnings, GDP growth (or associated alternatives, such as output gap), and other measurements of firm liquidity and profitability.

Collateral, in particular, the changes in the value of underlying asset serving as security. The share of secured (collateralised) loans and changes in the value of collateral impact loan losses by the means of their influence on loss given default. The most widely used proxy for the value of collateral is the price of real estate.

Bank behaviour, or predominantly, bank’s lending policy. Restrictive lending policy of banks is usually proxied by loan growth. However, comparable to debt burden, such proxy is limited in the case of economies, whose commercial banking sectors experience rapid expansion.

From a qualitative perspective, increased loan growth can be the consequence of increase in demand for loans, caused by optimistic expectations of borrowers concerning their future income.

Another illustrated bank behaviour – “income smoothing” practices – can be examined by the means of past profitability. Positive impact of this explanatory variable on loan losses would sustain the presence of “income smoothing” behaviour. On the other hand, decreasing capital adequacy increases the probability of moral hazard bank behaviour.

Although some research found statistically significant impact of bank portfolio diversification on loan losses, in practice, portfolio diversification impacts the variability of loan losses over time rather than their average level.

Additionally, bank efficiency (for instance cost efficiency) is also practical in explaining loan losses, provided that the bank efficiency is rational estimate of management quality.

3. Data

3 .1 Dependent variable

The dependent variable will stand for loan losses of Polish banks. When analysing banks’ balance sheets, it can be observed that the stock of loan loss provisions corresponds to accumulated credit risk losses. Moreover, stock of loan loss provisions decreases the net book value of impaired loans.

Thus, the dependent variable net flow of loan loss provisions represents the approximation of credit risk losses that each bank incurs over a certain period of time. Net flow of loan loss provisions (N_LLP_f_ln_nf) was chosen as the most suitable dependent variable based on the following grounds:

Prior to 2005, Polish banking regulatory framework was not compliant with IFRS, but rather implied the locally specified rules on loan classification. According to the previous loan regulations, bank loans needed to be sorted into one of the five categories as “satisfactory, special mention, substandard, doubtful or loss”, whereby the last three categories were considered irregular (impaired/advesly classified) loans. Previous regulations also applied to the minimum coverage of loans by loan loss provisioning for each of the five categories, wherby the last three categories were subject to particular provisioning requirements.

Moreover, throughout the observed period for the purpose of research, changes had been made to the definition of adversly classified loans. The first change in regulation took place in 2004: the minimum arrears period of thirty days was extended to ninty days and for the purposes of loan classification collateral was accepted. The second set of changes came in 2005 with gradual adoption of IFRS by those banks obliged to follow international regulations. (see [3], [1] and [2]) The aforementioned regulatory changes had significant influence on the adjustments of adversly classified loans share of portfolio (for instance certain smaller insitutions experienced ratio of fifty percent plummet to close to zero). On the other hand, the change affecting the flows of loan loss provisions was insignificant in scope, leaving little room for error. Another factor considered was the fact that data on nonperforming loans is available only from 2003 onwards.

Until 2004, Polish accounting and tax rules didn’t support banks with direct option to remove mature, fully provisioned non-performing loans (NPL) off their balance sheets (see NBP [32]), which is why these loans were accumulated in the balance sheet. This means that the share of adversely classified and/or NPL of a portfolio reveals bank’s historically accumulated loan losses, and not simply the current loan portfolio performance.

Other than a potential issue of dealing with nonstationarity in estimation, it is also­­­­ a challenge to interpret the results from the financial stability standpoint, without incorporating further qualitative analysis. The model applied for the purpose of this work can be useful in stress testing major balance sheet positions and profit and loss items of the banks, which is why the dependent variable is conveniently linked to the profit and loss account.

It can be anticipated that the flows of loan loss provisions will be different among different types of loans. This can lead to having different optimal sets of explanatory variables for each loan type. Although the supervisory data format doesn’t permit for the flows of loan loss provisions to be separated by the type of loan, type of borrower or by the currency of loan, this kind of separation exists for the data on the level of loan loss provisions.

When using the share of loan loss provisions as a dependent variable, whereby the common assumption is that it takes the values between 0 and 1, it is suggested to apply logit transformation of the variable. However, since the definition provided for the dependent variable does not restrict the range of values it can take, it is not necessary to apply any transformation of the dependent variable.

3 .2 Explanatory variables

The potential explanatory variables which were assessed for incorporating into the model are based on the broadly identified areas from the literature review, and are further specified in continuation.

Borrower income: the fundamental variable in this segment is the real GDP growth, as an aggreagte measure. Additionally taken into account for assessment were labour market trends (yearly change in unemplyment rate D4_unemp) and firms’ financial standing (liquidity ratio and pre-tax profitability ratio). The change in unemployment rate is intended to give a better explenatory power than just the level of unemployment, considering that it can be expected that increase in unemployment rate may be correlated with the number of borrowers who in the observed timeframe became unemplyed. Alternatively, labour market trends can be explained by quarterly changes (seasonally adjusted) in the number of people employed (D1_n_empl_ds). To calculate in the potential interbank differences in the sensitivity to the developments in the labour market, in certain measurements these variables were weighted by the share of loans -to-households of each bank’s portfolio. The two weighted variables are marked as D4_unempl_wgt and D1_n_empl_ds_wgt. In some instances, variables explaining labour market can as well approximate firms’ financial standing, which may influence their decidions regarding the optimal employment level.

Costs of servicing debt: these costs were estimated combining two different interest rate types. The first type was computed as the weighted average of real lending rate (b_ir_r_wgt denominates the average rate of lending to firms and households, weighted for every bank by the share of firms and households in a bank’s loan portfolio. The weighted rate of lending was computed based on the average rates of lending, as published by the NBP. Such measure was selected, taken into consideration that lending rates on a bank-level are not available for every banks. Real lending rate on corporate loans was computed as the interest rate on 1-year loans in domestic currency (PLN), deflated using producer price index (PPI). Real lending rate on household loans was computed as the interest rate on consumer loans (prior to 2002 known as cash loans), deflated using consumer price index (CPI). The decision was made by the necessity to minimise unexpected shifts in time series (a structural break) as the consequence of change in the interest rate reporting method (in 2002).

Alternative interest rate measurement can be the 3-month WIBOR (the interbank rate), deflated by CPI. The changes in such interest rate could be a useful approximation of changes in lending rates, under the premise of constant margins (risk premiums) on loans. However, the stability of margins does not hold true for the entire sample, as risk premiums on loans only became stable after 2001. The advantage of this measurements is the easy application in strss test scenarios, since it doesn’t call for forecasting margins.

Polish banks grant a significant portion of loans denominated in a foreign currency. Many of these foreign exchange loans are lended out to households that don’t hold matching income in a foreign currency. This is why the change in exchnage rate can impact the cost of repayment and the corresponding burden laid on the household income. Also, depreciation of the Polish zloty can increase loan losses of those banks which hold a large proportion of foreign currency loans in the profolios.

Nevertheless, under the observed period, the quality of loans denominated in foreign currency was better than the quality of loans in PLN, which points to the idea that the impact of changes in exchange rates could be only minor.

Additionally, mortgage loans given out in recent years represent a large part of total foreign currency loans. Academic research on f/x loans suggests the relationship between the age of morgage loan and the probability of default, whereby the greatest probability of default takes place around 5-8 years into the loan release. Considering that most of mortgage loans in foreign currency were released in Poland after 2003, the probability of default is still low for the majority of those loans. This is why the relationship between the changes in exchnage rate and loan loss provisions may result as statistically insignificant. This relationship was examined on the basis of yearly changes in nominal eur/pln rate. The change in nominal f/x rate is supposed to be the closest estimation of the changes in actual costs of debt servicing of a borrower without income in foreign currency.

Alternative measurements applied were effective real and nominal interest rates (deflated by CPI and PPI, respectively). In addition, the impact of foreign currecy interest rates were also examined (using average 3-month interbank rates for Swiss franc, EUR, and USD).

Chart 1: Progression of loan losses with key macroeconomic factors

Increase in loan losses follows after decrease in economic activity and increase in the rate of unemployment. The change in real interest rate levels is consistent with the change in loan loss levels, particularly notable after the year 2000.

Chart 2: Volume of adversely classified loans

Debt burden was computed individually for households and firms, as per two types of borrowers. Firms’ debt burden was computed on the basis of the ratio of firm loans to firms’ quartery sales (kred_cor_ przychody), while for households it is computed as the ratio of household loans to quarterly salary total (kred_gd_place). Total salaries are computed by multiplying the average gross salary to the number of people employed (apart from agricultural sector).

To capture bank behaviour, it is first necessary to look at banks’ lending policy, by examining loan growth rates. Specifically, the change in the rate of loan growth indicates certain change in the lending policies. For this purpose, the paper looks at loan growth on bank-level (b_ln_nf_gr_qq) and loan growth rate’s deviation from sector median. To take into consideration the possibility of “moral hazard”, the paper examines the capital resourses of each bank, based on the capital adequacy ratio (CAR), or precicely, its deviation from the sector median (dm_car). Here, using deviation from median is justified by the change in circumstances over the observed period, whereby the regulatory definition of CAR changed a couple of times, causing sudden chnages in the banks’ values of CAR. Additionally, two substitute are dummy variables (low_car10) and (low_car9), having the value of 1 if CAR falls below 10 percent or 9 percent, respectively. The efficient bank behaviour is captured by the cost-to-income ratio, frequently used in the literature and also known as efficiency ratio. It is defined as the “non-interest costs, excluding bad and doubtful expense, devided by the total of net interest income and non-interest income”6. In order to take into consideration the possiblity of “income smoothing” practices, the paper looks at the past performance, as captured by return-on-assets and return-on-equity ratios. [The income smoothing incentive is captured by a variable

that measures the difference between the historical return on assets and the pre-gain return on assets. ] If these two explanatory variables have a positive impact on loan loss provisions, such result would indicate the presence of income smoothing practice.

The extent of credit risk varies among different categories of borrowers and loan types. This is exemplified in Chart 2, which depicts the variation in the quality of loan portfolios. Loan losses of each bank are impacted by the structure of their corresponding loan portfolios. To account for such impact, banks are intially categories into “strategic groups”, as suggested in the methodological approach of Porter [37], [26]. Companies within the same industry that have similar business models and implement similar decisions in the major segments of their business are defined as strategic groups.SG For banking industry, those decisions are predominantly manifested in the composition of banks’ assets and liabilities. Polish group of authors in their paper “Strategic groups in Polish banking sector and ?nancial stability” performs cluster analysis to determined five existing groups within banking instury: “universal banks, corporate banks, car ?nance and mortgage banks, retail banks, regional banks”PSG. To describe the participation of banks in one of the groups, the approach will be to use dummy variables group1 – group5 (1 – retail, 2 – corporate, 3 – regional, 4 – specialised, 5 – universal). Taking into considerating that literature on strategic groups advocates the difference in profitability between different groups of firms, the approach taken here will be to examine if the differences among bank group profiles (as classified above) have the impact on the loan loss levels and on the loan loss sensitivity to macroeconomic conditions. In addition to the aforementioned approach, another variable can be taken to capture bank’s business profile, as represented by the volume of household loans in non-financial sector loans.

To account for potential impact of collateral, measure will be the portion of real estate loans in total household loans (b_hous_ln_share). This method is selected, due to the limited data on the categories of assets recognised as collateral. The higher proportion of real estate loans is expected to decrease loan losses. This is because real estate loans are predominantly secured by a motgage, which decreases loss given default, but also because gobally they represent the type of loan with the lowest default rates among different types of loans, since the penalty on default is especially significant for borrower whose mortgage is their primary residence (owner-occupant).

Lastly, in order to measure the influence of relevant regulatory changes, two dummy variable will be used: dummy_reg04 – value of 1 for all quarters of 2004 for capturing changes in the official classification of loans, and dummy_ifrs – value of 1 for all quarters of 2005 for changes resulting from International Fiancnail Reporting Standards.

The short average “age” of mortgages may overstate the influence of this category of loans (relative to a long-run average impact, which may be possible to estimate after an economic slowdown). The lack of long enough and representative enough time series of real estate prices made it impossible to use such data.

4. Empirical Estimation

4. 1 Data description and sample selection

The quantitative analysis is conducted using the data collected from quarterly bank reports, which are subject to auditing and submitted to the National Bank of Poland, as required by Polish law on banking (site WHICH here).

The dataset is taken for the ten-year period (with quarterly observations, totalling 2661), based on 108 banks operating on the Polish market in the years pning 1997 to 2006. This period us chosen in order to observe the phase of the rapid integration of Polish banking system with European markets, starting with the 1997 Reform of the Banking Law and ending with 2006, two years into the official integration of Poland into the EU, and one year before the onset of global financial market turbulence. 4

The sample consists of 2661 observations and in order to prevent the skewness caused by outliers, several measures were initially taken:

uBanks that were in bankruptcy proceedings or in the process of liquidation were not taken in the sample.

uFor new players in the market and those banks under significant restructuring (where low base effects could distort the data) and for the banks in poor financial standing (which have to take actions that are not reporesentative of the sample), the observations taken into the sample were those with the capital adequacy ratio between 5th and 95th percentile of the distribution.

uObservations with the values of dependent variable and explanatory variables falling outside the given range above were not included in the sample, apart from the dummy variables and those explanatory variables whose values were limited by their definitions.

The descriptive statistics of variables measured in the estimation phase are given in Table 1.

In order to test for robustness of estimates, the procedure is to re-estimate the models with an extended sample, whereby the values of the dependent variable are now between the 1st and 99th percentiles. This procedure expands the range of values in comparison to the baseline estimate. The results achieved with the extended sample correspond to the initially acquired results based on the baseline. As presented in the table 2 below, the results are comparably similar regarding both the statistical significance of coefficients, as well as the interval scale and each explanatory variable’s direction of influence.

4. 2. Estimation results

Every explanatory variable is lagged one quarter, or more, depending on the relevant regulatory framework. The incentives behind this transformation are the following:

uAccounting rules pertaining to loan losses create a gap between the time when borrower’s financial standing worsens (non-payment) and the time when the loan loss provision is created (after the loan has in actuality been classified as impaired). According to national regulations, such gap in the sample was taken as a lag of 30 days, prior to 2004, and 90 days afterwards. From 2005, banks which implemented IFRS were allowed to a certain degree to set their own threshold level of arrears, but the majority was in compliance to Basel II regulatory framework, which proposes 90 days arrears as a threshold of default on payment.

uAdditionally, applying lag to explanatory variables is useful in resolving possible endogeneity problems.

Choosing the best possible set of macroeconomic independent (explanatory) variables was challenging due to few instances of autocorrelation and correlation among them. This could cause approximate collinearity, resulting in imprecise parameter estimates. This is why in selecting independent (explanatory) variables, first choice were those variables that can be forecasted from the widely used models in literature, in particular ECMOD model of the Polish economy*, serving as a point of reference in forcasts. Also taken into consideration was the analysis conducted on correlation between prospective independent variables, with obtained conclusions as follows:

ucorporate sector indicators of liquidity and profitability were not included, since these are highly correlated with the change in the rate of unemployment (correlation coefficients’ absolute value ranges from 0.6 to 0.8)

uthe use of debt burden indicators may pose a challenge, since these are highly correlated with the change in the rate of unemployment and with GDP growth (value around 0.5)

uusing foreign interest rates is challenged by the correlation between Warsaw Interbank Offered Rate (WIBOR) and foreign interest rates of over 0.8

usince y.o.y GDP growth rates show significant autocorrelation at lags of 3 quarters or less, this can be solved by applying lags separated by 4 quarters or more

Table 1 Descriptive statistics of independent variables – under restrictions on dependent variable and capital adequacy

The initial procedure involves fixed effects model, which was applied in order to test for the statistical significance of independent (explanatory) variables. Estimating fixed effects model doesn’t require hypothesis of no correlation between independent variables and individual effects. Taking into consideration that independent variables include bank-specific features, such correlation couldn’t be ruled out presumptively.

Lag distribution of independent variables was selected based on exclusion of the lags with statistically significant coefficients[1]. The next step was to examine the presence of autocorrelation in the residuals, which was conducted using Durbin-Watson type test, of Sargan and Bhargava.6 The procedure indicated existance of autocorrelation[2], which was taken into account in further model estimations. In addition to fixed effects regression, random effects approach was also used, particularly when including in the specification strategic banking group membership dummy variables (as these predominantly change over cross-section and only slightly over time). Also, in order to select between random and fixed effects, Hausman test was applied and the relevant detailed statistics of the estimated models are offered in the tables 2-4.

The final specification included the following variables:

uGDP – year on year GDP growth

ureal_wib3m – 3-month interbank (WIBOR) rate, deflated by CPI

uD4_unempl – y.o.y. change in the rate of unemployment

ud1_n_empl_ds – quarterly changein the number of people emplyed, deseasoned beforehand

udm_car – capital adequacy ratio (deviation from sector median),

ub_hous_ln_share – housing loans share in loans to households (bank-level data),

ub_hh_ln_share – loans to households share in non-financial sector loans (bank-level data)

u dm_ln_nf_gr_qq_all – loan growth (deviation from sector mean) on bank-level

uDummy variables (group1 to group5) classifying banks into five strategic groups – relationship between strategic groups and macroeconomic indicators was utilised to examine potential differences among groups in their sensitivity to macroeconomic conditions.

uQuarterly dummy variables (q1-q3, etc.) – applied to account for seasonality

uDummy variables tracking regulatory changes

Coefficients estimated for macroeconomic indicators have usual signs: banks loan losses are increased by: incease in the real interest rates, increase in the rate of unemplotment, decrease in employment and decrease in gdp growth. The change in the foreign exchnage rate does not seem to impact banks’ loan losses. This is so even in the case when change in f/x rate is weighted by the share of f/x loans in a bank portfolio.

Debt burden indicators have not resulted in a significant impact. Similarly, combined effects of changes in interest rate and debt burden have no significant influence.

Table 2

Estimation results: changes in employment as explanatory variable, fixed effects model with AR(1) disturbance

With regard to the individual bank business profile, the strongest influence is noted in the structure of lending portfolio. The results indicate that the higher share of household loans in a bank’s portfolio increases loan losses. In developed economies, the relevant research on banking systems indicates that household lending usually carries less risk than corporate lending. This is probably due to the fact developed banking systems have a greater share of housing loans in the portfolio of loans to households than it is the case in emerging economies. Just by comparison within the EU in 2005, this share of housing loans in portfolio of loans to households in original EU countries comprised 69%, whereas in Poland this share was 37%.

The factor of categorising banks into 5 strategic groups has minor impact. The results indicate that specialised banks on average have lower loan losses, whereas universal banks demonstrate above average sensitivity to changes in unemployment. The strategic group of corporate banks indicate higher sensitivity to GDP growth.

Table 3 Estimation results: GDP growth and change in unemployment rate as explanatory

variables, random effects model with AR(1) disturbance

Also, greater loan losses are observed among the banks having low level of capital resources. This impact of capital adequacy is only significant in the short run. This outcome shows that moral hazard hypothesis doesn’t stand, since changing lending strategy in case of a “gamble” for survival would also affect the level of loan losses in the long run, considering the time required to extended loans on the aggregate portfolio level.

Table 4 Estimation results : GDP growth and change in unemployment rate as explanatory

variables, group-specific sensitivity, random effects model with AR(1) disturbance

The level of effect brought forward by regulatory changes in the period 2004-2005 is not definite, since the significance of the two dummy variables varies among regressions (specifications).


In analysing which macroeconomic variables exert influence on the loan losses in the Polish banking system, several key inficators were identified. In general, the results obtained are similar to the outcomes in some countries, but unlike elsewhere, in Poland debt burden variables did not result in having impact on loan losses.

The loan portfolio structure has emerged as an influential factor that determines the differences in the loans loss levels among banks.

It is useful to underline that under the observed period, the Polish banking industry was going through a dynamic phase of development, during which different varieties of lending products on offer changed in their importance and volume. For instance, at the beginning of observed period mortgage lending was not as influential to the bank performance as it was towards the end of the observed period, particularly with the rise in the price of property in 2005 and 2006.

Cite this Page

Resilience of the Polish Banking System to Credit Risk: Macroeconomic Perspective. (2019, Apr 15). Retrieved from

Don't let plagiarism ruin your grade

Run a free check or have your essay done for you

plagiarism ruin image

We use cookies to give you the best experience possible. By continuing we’ll assume you’re on board with our cookie policy

Save time and let our verified experts help you.

Hire writer