Determinants of Nonperforming Loans in Central

Credit risk assessment and, in particular, nonperforming loan (NPL) projection are a crucial part of macro-stress tests. These tests are based on macroeconomic assumptions in order to provide common scenarios for all financial institutions participating in such an exercise.

However, due to limited data availability, short time series and structural breaks, especially in emerging markets, it is challenging to estimate a robust model for any single country. It is therefore useful to provide an estimate based on panel data from emerging Europe to be used as a benchmark for countries in this region.

Credit risk is one key risk for financial stability in Central, Eastern and Southeastern Europe (CESEE), where banks apply the traditional business model based on accepting deposits and granting loans. Although the recent financial crisis had its origin in advanced economies, average bank asset quality in CESEE deteriorated sharply when the subsequent economic crisis hit the global economy.

Many research studies empirically confirm that credit quality is tightly linked to the economic cycle. However, there is some disagreement as to the importance of other factors driving credit quality. In this study, we focus on some specifics of the CESEE region that could determine the key drivers of NPL development.

This study follows research conducted by Beck, Jakubík and Piloiu (2013), who empirically investigated the key drivers of NPL development for a global panel covering 75 countries by using annual NPL data. Unlike the above study, we employ quarterly data and a more detailed dataset covering some aspects that were not available to Beck et al. (2013).

This paper is structured as follows: Section 1 provides a review of related studies on determinants of bank asset quality and credit risk. Section 2 presents a comprehensive description of the employed dataset and discusses NPL development in CESEE.

Section 3 focuses on the econometric methodology which is applied for quantifying the relationship between nonperforming loans (NPLs) and macroeconomic and financial indicators.

On this basis, we present the estimation results of an econometric model which quantifies this relationship. Section 4 summarizes the obtained results by drawing some conclusions with regard to policy implications. The last section concludes.

1 Related Studies

The literature on determinants of bank asset quality or credit risk comprises various approaches. Here, we focus on literature directly relevant for the present paper, following three main criteria. First, we cover papers that have the same regional focus, namely CESEE.

Second, our coverage extends to papers that follow a macro-approach by using macroeconomic variables as determinants for economy-wide aggregate NPLs. Third, our focus is on papers that apply a similar econometric framework as our paper, at least in as far as they apply panel techniques as well.

To our best knowledge, no study has been published so far that meets all three criteria. So, we see the present study as the first to use a macro-based approach to estimate economy-wide aggregate NPL ratios for a set of CESEE countries by applying a panel technique.

There are, however, many CESEE-related studies on bank asset quality and credit risk. For a large set of CESEE countries, Barisitz (2011, 2013) compares the national definitions and concepts of NPLs in detail and provides suggestions for aligning the statistical methodologies that measure NPLs.

Available CESEE-related studies on the macro-determinants of changes in the economy-wide aggregate NPL ratio of the banking sector are country-specific.

For Croatia, Erjavec, Cota and Jakšic´ (2012) set up a vector-autoregressive (VAR) model with macro-variables (real economic activity in Croatia and in the EU, inflation and short-term interest rates) and variables of the aggregate banking sector (return on equity (RoE), NPL ratio) and employ Uhlig’s sign restriction approach.

The VAR model is based on quarterly data for the period from Q2 2000 to Q2 2010. The above-mentioned authors find a strong sensitivity of the Croatian banking sector to contractionary monetary policy shocks and to negative demand shocks.

For the Baltic countries, Fainstein and Novikov (2011) published a comparative analysis of credit risk determinants in the banking sector, applying a separate vector-error-correction model (VECM) for each of these three countries, based on quarterly data for the period from (depending on the country) Q3 1997/ Q1 2002/Q1 2004 to Q4 2009.

In addition to the unemployment rate, real GDP growth and banks’ aggregated loan growth, the authors introduce the growth rate of the real estate market as explanatory variable.

Their results show real GDP growth as the most significant determinant of NPL growth in all three countries and that real estate market growth plays an important role in two of these countries (Latvia and Lithuania).

For Albania, Mancka (2012) estimates the impact of the exchange rate (in relation to the euro and to the U.S. dollar) and of a dummy variable for the world financial crisis on aggregate credit risk (measured by the NPL ratio) for the period from 2002 to 2010 on the basis of quarterly data.

Both the exchange rates of the national currency and the dummy variable proved to have significant influence on NPLs.

Beyond the CESEE region, several studies estimate aggregate NPL ratios under a macro-based approach for a large set of countries (that may or may not include CESEE countries) by application of panel techniques.

In their paper on macroprudential stress testing of credit risk, Buncic and Melecky (2012) incorporate estimates of NPL elasticities by dynamic panel data regression (unbalanced panel, Arellano-Bond GMM estimator) on the basis of annual data for 54 high- and middle-income countries in the period from 1994 to 2004.

Explanatory variables are the NPL ratio, real GDP growth, CPI inflation, the (ex post) real interest rate and the change in the nominal U.S. dollar exchange rate for each country, while a vector of control variables comprises the log of GDP per capita, the credit-to-GDP ratio and the share of foreign currency loans in total loans.

Buncic and Melecky find the exchange rate changes and the control variables to be not statistically significant.

For 26 advanced economies in the period from 1998 to 2009, Nkusu (2011) investigates the macroeconomic determinants of the NPL ratio and of the first difference of the NPL ratio in various panel regressions on the basis of annual data that include the lagged dependent variable.

The results confirm that adverse macroeconomic developments, in particular a contraction of real GDP, a higher unemployment rate, higher interest rates, a fall in house prices and a fall in equity prices, are associated with rising NPLs.

In a second step, the feedback between NPLs and macroeconomic variables is estimated in a panel vector autoregressive (PVAR) model. For 25 emerging market economies in the period from 1996 to 2010, De Bock and Demyanets (2012) estimate various panel regressions on the basis of annual data that include the lagged dependent variable and unobserved country effects.

Real GDP contraction, currency depreciation against the U.S dollar, weaker terms of trade and outflows of debt-creating capital (portfolio debt and bank loans) lead to a higher aggregate NPL ratio of the banking sector.

The sharp deterioration of loan quality following a reversal of portfolio inflows is particularly noteworthy. The (first lagged) increase in the private credit-to-GDP ratio has no significant impact in the whole sample but is significant with a negative sign in the 2004 to 2010 subsample.

In a second step, feedback effects from the financial sector on the wider economy are found to be significant according to a PVAR model with fixed effects, in which GDP growth falls in the wake of shocks that drive NPLs higher or generate a contraction in credit.

For 75 advanced and emerging economies in the period from 2000 to 2010, Beck, Jakubík and Piloiu (2013) estimate fixed-effects and dynamic panel regressions on the basis of annual data for the change in the aggregate NPL ratio.

Real GDP growth, share prices, the nominal effective exchange rate of the local currency and the bank lending interest rate are found to significantly affect changes in the NPL ratio.

In the case of exchange rates, the direction of the effect depends on the extent of foreign exchange lending to unhedged borrowers. In the case of share prices, the impact is found to be larger in countries which have a large stock market relative to GDP.

2 Data, Stylized Facts and Hypotheses

In contrast to the study by Beck et al. (2013), we focus only on CESEE and have a richer data sample with quarterly frequency. Hence, we are able to better capture some specific effects for emerging Europe that cannot be fully revealed with a global data sample at annual frequency.

Our study covers the following nine CESEE countries: Bulgaria, Croatia, the Czech Republic, Hungary, Poland, Romania, Russia, Slovakia and Ukraine. We include these countries for two main reasons: First, for these countries we can rely upon studies that investigate the applicability of the corresponding national data on NPLs and credit classifications (Barisitz, 2011 and 2013, see below).

However, we could not include all the ten countries covered by Barisitz, as we had to exclude Serbia due to problems with data availability related to the explanatory variables. Second, these nine countries together account for a very large part of Austrian banks’ credit exposure to the CESEE region.

For NPLs, i.e. the dependent variable to be explained, there is, so far, no internationally harmonized definition that has been applied in all or most countries of the world for a considerable period of time.

Rather, efforts toward harmonizing NPL definitions have been getting steam only in recent years in the wake of the financial and economic crisis. For the time being, one can resort only to national NPL data established by the national supervisory authorities on the basis of their respective national definitions.

However, we use the results of the detailed investigations performed by Barisitz (2011 and 2013) in an effort to minimize the methodological differences reflected in the data.

Barisitz looked at both primary elements (“90days+” and/or “well-defined weaknesses”) and secondary elements (treatment of replacement loans, impact of collateral and/or guarantees, share of loans classified as NPLs, downgrade requirement) in the CESEE countries’ NPL definitions in order to derive NPL ratios that are based on roughly internationally comparable definitions in a transparent and replicable manner.

Hence, we use the time series of NPL ratios as suggested by Barisitz (2011 and 2013), built as the country-specific sums of various components of the credit volume classified according to the respective national definitions.

From our point of view, this enhances the added value of this empirical study. Moreover, these level data that were derived by Barisitz on a best-effort basis, form the starting point of the ensuing transformation, as we aim at explaining the relative change in the share of NPLs in total loans, i.e. the percentage change (as opposed to the change in percentage points).

Therefore, any remaining methodological differences inherent in the level data are of lesser importance, as we follow the development of each national NPL time series in its own right.

Turning to possible explanatory variables, we take into account both the stylized facts of NPL developments in CESEE sketched out above and the body of literature in which econometrical models typically explain NPL ratios by including variables for economic activity, aggregate credit and some additional variables.

As regards real economic activity, we look at real GDP as well as at the two main components of final demand, namely real exports and real domestic demand. Obviously, we hypothesize a negative relation between real GDP growth and a change in the NPL ratio. Moreover, we expect this to be true also for both main demand components.

As real economic activity in each CESEE country is heavily influenced by the international environment, we include the Chicago Board Options Exchange (CBOE) Market Volatility Index (VIX), a popular measure of the implied volatility of Standard and Poor’s (S&P) 500 index options, the emerging market bond index global (EMBIG) and the national stock indices as proxies for the risk attitude among international financial investors and for the international environment and thus as leading indicators for the financial and economic developments in the CESEE countries.

Regarding the credit aggregate, we look at domestic bank credit to the private sector, including both households and nonfinancial corporations. We highlight, however, that this credit aggregate includes loans denominated in foreign currency or indexed to the exchange rate.

Therefore, as we would like to avoid having the development of this credit aggregate blurred by purely statistical effects of exchange rate changes, we used this credit aggregate after adjustment for valuation changes resulting from exchange rate changes.

The drawback of this approach was the fact that we had to shorten our sample to the period from 2004 to 2012. At the same time, however, this had the advantages that we avoided the statistical breaks in the NPL series at the end of the 1990s and early 2000s and that we ended up with a nearly balanced panel.

We use the credit aggregate relative to GDP as explanatory variable in the model. An increase in the credit-to-GDP ratio via higher credit growth than GDP growth may indicate a sound, sustainable process of financial deepening on the one hand, but it may also result from excessive loan growth as part of a boom-bust cycle on the other hand.

We hypothesize a two-fold relation between developments in the creditto-GDP ratio and the NPL ratio. First, in the short run, we expect a denominator effect that has a negative sign. Second, in the medium to long run, we expect a credit cycle effect with the opposite sign.

We expect episodes of high credit growth, which are often coupled with low lending standards, to raise the share of NPLs only with a considerable time lag, as the distinction between borrowers evolves over time and the probability of default in the course of a credit’s entire life cycle is higher than the probability of default just in the first payment period.

In several CESEE countries the share of foreign currency loans (loans denominated in foreign currency or linked to the exchange rate) in total loans is sizeable; there is an indirect credit risk as borrowers are exposed to higher debt servicing costs if their national currency depreciates against the loan currency and only a part of these borrowers may be adequately hedged.

We use the exchange rate against the euro for most CESEE countries and the one against the U.S. dollar for Ukraine and Russia, where foreign currency loans are mostly U.S. dollar denominated.

As we assume that a (substantial) part of the foreign currency-denominated loans is extended to unhedged borrowers, our hypothesis is that a depreciation of the national currency leads to a higher NPL ratio, depending on the size of the share of foreign currency-denominated loans in total loans.

Therefore, we use the multiplicative term exchange rate change multiplied by the share of foreign currency loans (weighted exchange rate change) as an additional explanatory variable.

Clearly, if one of the two components is very small or even zero – like in the case of the Czech Republic, where foreign currency loans play a very limited role, or in the case of Bulgaria, where the currency does not change against the euro given its currency board arrangement – this variable will hardly have any impact on the development of the NPL ratio.

However, obviously, this explanatory variable does not cover all types of risks potentially attached to foreign currency loans. Apart from possible risks related to the corresponding foreign currency funding of banks, “the interest rate risk profile of foreign currency loans differs from the risk profile of domestic currency loans.

This can be detrimental to the quality of foreign currency loans if the interest rate cycles of the foreign currency diverge from that of the domestic economy.” (ESRB, 2011). In addition, also the size of the amplitude of the cycles may be quite different.

As foreign interest rates are generally less linked to domestic inflation than domestic interest rates, the volatility of implicit real interest rates will be higher.

In as far as the share of foreign currency loans is positive, this interest rate risk is relevant also for countries where the respective exchange rate has not changed in the period of observation.

Both the exchange rate-related credit risk of foreign currency loans and the interest rate-related one are two-sided risks.

As these risks are not necessarily synchronized, the materialization of foreign interest rate risk may mitigate or exacerbate the impact of exchange rate changes, depending on the specific periods under observation.

Moreover, the extent of foreign interest rate risk depends on (i) the dominant type of interest rate-setting regime for foreign currency loans in an individual country2 and (ii) the country-risk premium (thus, in particular, on the anticipated fiscal position of the public sector) and how this premium is taken into account in the interest rate-setting regime of foreign currency loans to the private sector.

However, investigating which type of pricing regime is followed to which extent in which country and for which type of foreign currency loan and inserting this information into an NPL model would be a task of its own that we leave for further research.

For the moment, we would like to stress that the materialization of foreign interest rate risk, which depends inter alia on the interest rate-setting regime, may blur the measured impact of exchange rate changes on the NPL ratio.

It is conceivable that banks’ profitability is somehow related to the NPL ratio in later periods. Thus, we look at the return on assets (RoA) as a measure for banks’ profitability that we consider superior to the RoE, as the latter is heavily influenced by the degree of capital adequacy and leverage.

One may follow the traditional management quality hypothesis, stipulating that more profitable banking sectors are better managed and more prudent in their granting of credit so that higher profitability in the past leads to a lower NPL ratio.

For the sake of clarity, we would like to mention that, in addition, there may well be an impact of the NPL ratio on banks’ profitability in later periods, in particular via net creation of loan loss provisions, with the calculation of impairment charges usually taking more time.

However, we do not investigate this feedback loop in the present study, but limit our focus on the factors determining the NPL ratio. Taking all these explanatory variables together, we aim at explaining the development of NPL ratios to a large extent.

For each of the explanatory variables, we tried to get time series ranging from Q1 1993 to Q4 2012. However, due to the aforementioned structural breaks and the limited availability of data that conform to the quality requirements sketched out above for the years up to the first years of the new millennium, we finally ended up with a nearly balanced panel that we could estimate for the time span ranging from 2004 to 2012.

Location: Indonesia