The determinants of credit risk in the Greek banking system

(1)

1

The Determinants of Credit Risk in the Greek

Banking System

University of Amsterdam

MSc Economics

Monetary Policy, Banking and Regulation

Author:

Paraskevas A. Tsouroukidis 10827412

Supervisor:

Prof. dr. A.C.F.J. (Aerdt) Houben

Co-reader:

(2)

2

Statement of Originality

This document is written by Student Paraskevas A. Tsouroukidis who declares to take full responsibility for the contents of this document. I de-clare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its refer-ences have been used in creating it. The Faculty of Economics and Busi-ness is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

Abstract

This empirical study explains the events that caused an unprecedented banking crisis for Greece by utilizing theoretical reasoning and econometric testing. Using a single-equation time series approach, I examine the macro-economic determinants of non-performing loans in the Greek banking sys-tem for a time span of fifteen years (2001 - 2016) for total non-performing loans and ten years (2005 -2014) for three sub-categories of non-performing loans. The analysis includes indicators measuring the general state of the economy, private and public indebtedness and price stability. I find that variables such as GDP and unemployment have a significant impact in all categories of NPLs. Moreover, the set of variables accounting for public finances exert significant influence on credit risk, highlighting the feedback loop effect of the sovereign and banking crisis.

Keywords: Non-performing loans, Macroeconomic determinants, Greek

(4)

4

1 Introduction

Both in developing and advanced countries, the deterioration of banks’ asset portfolios has been at the center of the financial crisis. Adverse eco-nomic conditions have increased households’ and business’ defaults, leading to losses for banks. Increasing levels of non-performing loans (henceforth NPLs) is a warning indicator of the beginning of a crisis (Reinhart & Rogoff, 2009) and are associated with negative solvency and profitability effects in banks’ portfolios (Buncic & Melecky, 2012; Marcelo, Rodríguez, & Trucharte, 2008). Furthermore, there is a broad consensus that NPLs have a negative impact on banks’ lending to the economy (European Central Bank, 2017), highlighting banks’ responsibility to maintain suffi-cient capital buffers. Thus, understanding the determinants of NPLs is of vital importance for the stability of financial institutions and the whole economy.

The adoption of Basel III illustrated that factors such as business cycle effects and the macroeconomic environment should be considered in the credit risk modeling. This has triggered studies that investigate the deter-minants of credit risk at both the country and regional level. In this study, I analyze the macroeconomic determinants of credit risk for the Greek banking system, initially by estimating a model with a restricted set of variables that will serve as a baseline, and then expanding the model by incorporating with other variables that contribute explanatory power. I select a set of macroeconomic variables drawing on hypotheses put forward in the literature on credit risk modeling, as well as in the literature on the Greek crisis. This study contributes to the existing literature in three ways. First, it focuses on the aggregate NPLs (aggregate banking data), instead of individual banking data. Second, this study has a time span of 15 years for total NPLs and 10 years for the three sub-categories of NPLs (business, consumer, mortgages), thus covering both the boom as well as the crisis period. Finally, this study includes variables such as long-term interest rate, private indebtedness, housing price index, nominal effective exchange rate, stock price index, investment and fiscal balance which have not been used in this combination before in the case of Greece.

The study proceeds as follows: In section 2, I provide a survey of the re-cent literature on the Greek crisis and credit risk. Section 3, presents the

(6)

6

data analysis, statistical diagnostic tests, variable descriptions and the hy-pothesis I test. Section 4, displays the econometric model. Section 5, pre-sents the empirical results. Section 6, is a discussion of the results and a comparison with those yielded from the existing literature. Finally, Section 7 concludes.

(7)

7

2 Literature on the Greek Crisis and Credit Risk

In this section, I discuss the existing studies on the Greek crisis as well as the literature on the credit risk determinants to emphasize on the moti-vation for my research and to explain why the variables, that are selected and are included in my empirical analysis, are relevant.

2.1 Greek Crisis - The feedback loop between the sovereign

and the banking system

From 2001 to 2008 the Greek economy exhibited high rates of economic growth (average 4,0% of GDP) due to consumer spending, business confi-dence, profits, and investment that resulted in a reduction of the official unemployment from 10,8% to 7,8% (Bank of Greece, 2014). Despite these favorable conditions, the Greek governments demonstrated an inability to exercise fiscal discipline. Government expenditure rose from 46,7% to 50,4% of GDP, (60% of which was allocated to social transfers and com-pensation of public employees) while at the same time revenues declined from 41% to 36,9% of GDP (E.U. Directorate for Economic and Financial Affairs, 2010). This resulted in an increasing net General Government Bor-rowing, causing a rise of Public Debt to GDP by 9,2%. According to Pagoulatos and Triantopoulos (2009) this poor budgetary performance was related to a chronically inefficient public administrative and budgetary structure, tax evasion, an inadequate collection of revenues, a tradition of clientelistic appointments in the public sector and high military expendi-ture.

Following the adoption of the Euro in 2001, the current account balance changed, as the government no longer had the ability to use the exchange rate to restore imbalances in the external accounts. The Greek economy systematically lost competitiveness against its economic partners, especial-ly compared to the advanced economies of the EMU (Gagales, Rossi, & Badia, 2007). This is reflected in the appreciation of the real effective ex-change rate both in terms of unit labor cost and consumer prices, which increased annually by 2,3% and 1,9% respectively until 2008 (Bank of Greece, 2014). As a result, the current account deficit rose from 5,2% to 14,6% of GDP and external debt accelerated from 74,8% to 146,8% of

(8)

8

GDP during the period 2001-2008 (Petrakis, 2012). The persistent current account deficits are mainly due to deficits in the trade balance (Figure 1a).

In order to identify the determinants of the unsustainable current ac-count imbalances, we have to take a closer look at the saving and invest-ment sectoral imbalances of the Greek economy. As illustrated in Figure 2, both the Government and the Households contributed to the deterioration of the savings balance in the years preceding the crisis. While those sectors had negative savings during the whole period, on the contrary, the non-financial and non-financial corporations were net lenders.

-22.00 -17.00 -12.00 -7.00 -2.00 3.00 8.00 13.00 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 Figure 1: Current account balance (% GDP)

Goods Services Primary Income

Secondary Income Current Account

Data source: Bankof Greece

-20.00 -15.00 -10.00 -5.00 0.00 5.00 10.00 15.00 20.00 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

General Government Corporations

Financial Corporations Households Total Economy

Figure 2: Borowing Lending Private Sector - Goverment (% GDP)

(9)

9

By 2008, Greece was running a vulnerable twin deficit economy. High deficits and rising external debt raised the vulnerability of the economy to a shock. (Mitsopoulos & Pelagidis, 2011). An external funding was essen-tial for the balance of external accounts of the private and public sector.

A small proportion of the current account deficit was financed by the surplus of the capital account balance, and the rest was financed by the financial account (Bank of Greece, 2014). The negative sign of the finan-cial account from 2001 to 2009 means that net capital inflows were funding the corresponding excess of domestic investment instead of domestic sav-ings of government and households. Capital inflows consisted mainly of debt financing and portfolio investments, while foreign direct investment remained weak between the period of 2001-2008 (Pagoulatos & Triantopoulos, The Return of the Greek Patient: Greece and the 2008 Global Financial Crisis, 2009). Portfolio investment, mainly consisting of bond financing, financed almost 3/5 of the current account deficit (Bank of Greece, 2014). This is mainly attributed to the fact that the Greek gov-ernment, encouraged by low-interest rates, was increasingly relying on fi-nancial markets to finance its deficits. Finally, other investments consisted of bank lending was mostly allocated to the private sector (Bank of Greece, 2014).

The developments mentioned above, especially the imbalances of the public sector and households, would not have been possible without the financial intermediation of banks. During 1990-1998, a period when the liberation and modernization process took place, the banking sector oper-ated as the primary driver of the economy and bank credit had an annual average growth rate of 4,3% (Athanasoglou, Georgiou, & Staikouras, 2009). The following years, and especially after the introduction of the euro, cross-border bank activity increased dramatically, and the exposure of the non-stressed countries’ banks to the Greek banking system rose to approx-imately 42% of the Greek GDP in 2008 (Constâncio, 2014). Bank credit growth more than doubled (8,9%), with financial intermediation being the most critical category of activities (Athanasoglou, Georgiou, & Staikouras, 2009).

Bank lending increased sharply during this period, reflecting the increase in the deposit base and the reduction of barriers to cross-border banking integration (Petrakis, 2012). Funds coming from abroad were allocated through financial intermediaries mainly to the private sector of the

(10)

econo-10

my, where credit tripled, while lending allocated to the Government re-mained in high levels but relatively constant (Figure 3). Foreign banks were financing the excessive current account deficit and public debt.

Credit expansion stimulated economic performance but also financial fragility. First, a boost in demand of households and businesses fueled by credit expansion led to inflationary pressures and a rise in the real ex-change rate (Petrakis, 2012; Constâncio, 2014). Second, high domestic de-mand prompted a cross-sector redistribution of resources in favor of the non-tradable sectors (Pagoulatos & Triantopoulos, The Return of the Greek Patient: Greece and the 2008 Global Financial Crisis, 2009; Ioannou, 2013). The non-competitive sectoral structure can be illustrated by the re-allocation of 17,8% of the total credit of the private sector from business loans to consumer loans and mortgages. In the first quarter of 2001, the share of corporate credit, mortgages and consumer loans in total credit was 71,5%, 18,9%, and 9,4% respectively while at the end of 2008 it was 53,7%, 31,4%, 14,7% (Figure 4). Finally, asset prices escalated rapidly, within a period of 8 years, housing prices increased by 73% while stock prices more than tripled. As early studies have indicated, the joint effect of rising asset prices and increasing indebtedness is a warning indicator of financial fragil-ity (Borio & Lowe, 2002).

0 50,000 100,000 150,000 200,000 250,000 300,000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 General Government

Provate Sector ( Business and Households)

Figure 3: Bank lending (millions of euros)

(11)

11

In the period preceding the Greek crisis, the 10-year bond interest rate spread between Greek and German sovereigns had fallen from 600 basis points to 10-50 basis points (Graph 4) despite the growing imbalances1_{. In}

the context of the euro area’s debt crisis, the level of government debt combined with the low competitiveness of the economy should have raised concerns about the country’s ability to meet its obligations (Mitsopoulos & Pelagidis, 2011). But this only became clear in 2009 when the newly elect-ed Greek government stunnelect-ed the markets announcing that the deficit was higher than projected (Provopoulos, 2014). The following years, a self-reinforcing feedback loop between ratings’ downgrades and spreads esca-lated (Featherstone, 2011).

Those developments caused a downward spiral in the banking system. First, uncertainty rose, leading to liquidity problems following a massive withdrawal of deposits2_{(Figure 5). This deteriorated situation caused the}

downgrading of banks and their exclusion from international financial markets, including the European interbank market. This resulted in an increasing reliance on the Emergency Liquidity Assistance through the Greek Central Bank (Provopoulos, 2014). Second, the voluntary public debt restructuring that took place in 2012 implied a 70% haircut on the

1_{As Gibson, Hall and Tavlas (2014) showed, markets were mispricing sovereign risk in the euro}

area.

2_{A total amount of €63 billion deposit withdrawals occurred between 2010 to 2012. This can be}

explained by the general uncertainty of the Greek citizens caused by the increasing speculation on the probability of Greece leaving the Eurozone after the bailout agreement.

0.00 20.00 40.00 60.00 80.00 100.00 2001-Q1 2002-Q3 2004-Q1 2005-Q3 2007-Q1 2008-Q3 2010-Q1 2011-Q3 2013-Q1 2014-Q3 2016-Q1

Corporations Housing Consumer

Figure 4: Loans to the private sector (% of the total)

(12)

12

value of government bond holdings by the domestic banks. This caused banks to collectively lose €37 billion (Provopoulos, 2014). Finally, unprec-edented adverse economic conditions affected the banks’ asset portfolio, prompting a surge in NPLs.

The public sector provided the majority of the financial aid for banks’ stabilization through the Hellenic Financial Stability Fund (HFSF). In eight years, there were three recapitalization processes and the amount effectively disbursed from HFSF to the banks was €34.09 billion, whereas the investment of the private sector amounted to €13.51 billion (Muguruza, Efstathios Efstathiou, Savin, & Diana, 2017). Despite the massive injection of funds in the country’s banking system, credit conditions did not recover. NPLs are a main reason why credit growth and the Greek economy have not recovered. Before the shock of 2008, the average ratio of non-performing loans to total gross loans was 6%3_{in Greece, with NPLs offset}

by higher provisions4_{. This is above the average of European Monetary}

Union’s 2,2% and European Union’s 2,4%. From 2009 to 2016, the ration of NPLs to total gross loans increased sharply from 6,9% to 36,9% with an

3_{To the best of my knowledge there is no study explaining the high magnitude of NPLs in the}

pre-crisis period. However, a possible explanation could be that the culture of sticking to a contract is not very strong in Greece. This combined with the fact of unequal distribution of taxes (Petrakis, 2012) and the tax evasion of the richer (Georgakopoulos, 2016) may cause an inconsistent attitude of borrowers towards their obligations.

4_{According to financial stability assessment of the IMF (Gagales, Rossi, & Badia, 2007) NLPs}

(13)

13

average 21,9%. At the same time, the EMU countries had an average NPL ratio of 6,4%, and the average of EU countries was 5,8%. Specifically, post-crisis NPLs for consumer loans had an average of 33%, followed by busi-ness loans which amounted to 19,7% and for mortgages which rose to 18%. With that burden, banks were not able to operate and provide funding to the real economy. As Constâncio (2017) estimates, the replacement of NPLs with performing assets may improve profitability with an increase of up to 5% of ROE and additionally the resolution of NPLs can bring a tan-gible relief to bank capital and increase loan supply up to 10% for the case of Greece.

Overall the development of the Greek crisis depicts the feedback loop be-tween the government and the banking sector. In contrast with other Eu-ropean countries, the Greek banking system was not affected by the initial phase of the Global financial crisis, due to the limited exposure of domestic banks to (the) “toxic” financial instruments (Pagoulatos, 2014). A moder-ate recession of the Greek economy that started in 2008 caused an upward trend of NPLs5_{. However, the banks’ balance sheets were severely affected}

by the sovereign crisis and the austerity measures through the channels of their exposure to the Greek government bonds, which amounted to 54 bil-lion in 2010 (Bank of Greece, 2014), and the conditions of the real econo-my. To support the financial system, the Greek state proceeded to a recap-italization of the systemic banks (Muguruza, Efstathios Efstathiou, Savin, & Diana, 2017) that was funded from the bailout programs and burdened the public debt. This way the transfer of the burden from the banks to the public sector generated and further deteriorated the environment and the banking crisis turned back into a sovereign crisis.

2.2 Key Studies of the determinants of Non-performing

loans

Empirical studies have highlighted the cyclical nature of bank credit and NPLs (Bofondi & Ropele, 2011). During the upturn phase of the business cycle, competition between banks drives interest rates down, lending standards become looser, and credit expands. Credit risk and NPLs tend to

5_{As a precautionary measure the Greek government provided a financial aid of €28 billion}

bailout plan, of which €3.5 used for recapitalization proposes and remainder served as guarantees (Petrakis, 2012).

(14)

14

be at low levels as income increases and optimism prevails about the fu-ture outlook of the economy. The quality of the banks’ asset portfolios generated during this expansionary phase of the business cycle depends on the regulatory authorities’ standards and defines to a substantial degree the banks’ resilience to future shocks of the economy. In downturns, the situation reverses as credit expansion shrinks, the value of collateral de-clines and credit risk rises.

In this framework, the determinants of NPLs are both bank-specific and macroeconomic. Regarding bank-specific indicators, they refer to the risk attitude of a bank’s management and, to an extent, they reflect the results of financial regulation and supervision. There is strong evidence that accel-eration of credit growth, driven by increasing competition between banks, eventually leads to loans losses. Macroeconomic determinants, on the oth-er hand, reflect the conditions that influence the debt soth-ervicing capacity of borrowers, including the changes in their wealth. There is a broad consen-sus among studies on the countercyclical behavior of NPLs.

Several studies indicate that bank-specific variables are important de-terminants of credit risk. One of the most widely used variables is credit growth, assuming a positive relationship over time between credit growth and impaired loans. In early studies, (Keeton & Morris, 1987) they investi-gated the variations in loan losses using data on commercial banks in dif-ferent Federal Reserve Districts. They argued that banks with relatively high lending take on more risk, leading to higher losses. Furthermore, Salas and Saurina (2002) confirmed the same result for the Spanish bank-ing system. Khemraj and Pasha (2009) analyzed the determinants of NPLs in the Guyanese banking system and concluded that banks that lend more than other banks acquire higher levels of low-quality loans. In addition, profitability variables are also included in several studies; Godlewski (2004) used a sample of commercial banks for emerging market economies and showed that return on assets (ROA) is negatively correlated with NPLs. In adverse, Garcia-Marco and Robles-Fernandez (2008), using data for 129 Spanish banks, showed that initially high levels of return on equity (ROE) are positively correlated with credit risk. The different sign of the correla-tion of NPLs with profitability variables can be explained given that ROE may be higher in the short-term, but lower in the long-term.

Surveying studies on the macroeconomic determinants of problem loans at a country level, Salas and Saurina (2002) used panel data for the period

(15)

15

1985-1997. The authors compared two institutional regimes in Spain, for commercial and savings banks and their results indicated that macroeco-nomic fluctuations are quickly transmitted to problem loans. In particular, the GDP growth rate negatively affects problematic loans contemporane-ously and with a one-year lag. Furthermore, they found that the general level of firms’ indebtedness positively affects the level of problem loans in commercial banks.

In addition, Brookes, Dicks and Pradhan (1994) estimated a model to explain building society mortgage arrears and repossessions in the UK. Their results provide evidence of the positive relationship between macroe-conomic data on mortgage defaults, unemployment, inflation, and interest rates.

Furthermore, Macit (2012) investigated the determinants of non-performing loan ratios for commercial banks in Turkey running a general-ized least square and a panel data estimation. Regarding his results for macroeconomic variables, he found that the lagged value of GDP growth has an impact on the NLPs ratio. He also took into account the foreign exchange rate, as some of the loans were indexed to a foreign currency, and found that a depreciation in domestic currency increases NPLs. In contrast with these findings regarding Turkish banks, is the research of Vatansever & Hepşen (2015) who used linear regression models, with data from January 2007 to March 2013. They found that neither the GDP growth nor the exchange rate with Euro and USD and the interest rate have a significant effect. However, industrial production index, the Istan-bul Stock Exchange 100 Index, negatively affect NPL ratio, while the un-employment rate has a positive impact.

Turning to studies that analyze credit risk at a regional level, Sinkey and Greenawlat (1991) studied a sample of large commercial banks in the United States from 1984 to 1987. Their findings suggest that both macroe-conomic factors such as regional emacroe-conomic activity and bank-specific fac-tors such as excessive lending and interest rates affect the loan loss rate in the banking system. In a more recent study regarding the US, Ghosh (2015) investigates the determinants of NPLs in both commercial banks and sav-ings institutions in 50 states for the period 1984-2013. Using both fixed effects and dynamic-GMM estimations he finds that NPLs are negatively related with higher state real GDP, real personal income growth rates and

(16)

16

changes in the housing price index, while NPLs are positively correlated with inflation, unemployment and US public debt.

Ali and Daly (2010) compared the cases of Australia and the U.S. for the period 1995-2009. They found that the same set of macroeconomic variables display different default rates in the two countries. Moreover, their results demonstrate that in addition to the rate of GDP growth, which negatively affects defaults rates, interest rates and debt to GDP ratio have the opposite effect.

Moreover, Rinaldi and Sanchis‐Arellano (2006) analyzed household fi-nancial fragility using an empirical model for the ratio of non-performing loans, in a sample of EMU member countries, namely Belgium, France, Finland, Ireland, Italy, Portugal and Spain. They found that an increase in the ratio of indebtedness, inflation and lending rates is associated with a higher level of non-performing loans. On the contrary, the housing price index is negatively related to NPLs, supporting the idea that housing wealth increases households’ debt servicing capacity.

Skarica (2014) studied a sample of seven Central and Eastern European countries using aggregate, country-level data on problem loans. Her results suggest that higher levels of NPLs are associated with economic slowdown and increased unemployment and inflation.

Nkusu (Nkusu, 2011) analyzed factors influencing asset quality in a sample of 26 advanced economies from 1998 to 2009, concluding that ad-verse economic conditions such as lower GDP, higher unemployment and decreasing asset prices (stock prices and housing price index) tend to in-crease credit risk. In addition, he examined the causal relationship between non-performing loans and macro indicators for Sub-Saharian Africa. He indicated that the real effective exchange rate has a positive relationship on NPLs. Similarly, Beck, Jakubik, and Piloiu (2013) examined the role of key macroeconomic indicators in NPLs for a sample of 75 advanced and emerging economies and found that the nominal effective exchange rate depreciation leads to an increase of non-performing loans especially in loans that are denominated in foreign currencies.

In terms of methodology and dataset, the existing literature can be clas-sified into three strands. The first strand explains credit risk using general-ized methods of moments with individual banking data for a single country (Louzis, Vouldis, & Metaxas, 2010; Ghosh, 2015). The second strand ex-amines cross county analysis with vector autoregressive models (Espinoza

(17)

17

& Prasad, 2010). The third branch, which closely relates to this thesis, captures the dynamics of NPLs using single equation with time series ag-gregate data (Arpa, Giulini, Ittner, & Pauer, 2001; Shu, 2002; Kalirai & Scheicher, 2002; Bofondi & Ropele, 2011).

Surveying the relevant literature of the latter strand, Shu (2002) exam-ines the determinants of asset quality in the banking sector of Hong Kong using a single equation time-series approach. The author used time series for 1995-2002 and his findings suggest that loans’ quality is affected by macroeconomic variables such as GDP growth, lending rates, inflation and by variables measuring real wealth such as property prices. Likewise, Arpa, Giulini, Ittner, and Pauer (2001) employ a single equation to assess the role of key macroeconomic indicators in Austrian banks’ risk provisions using time series data for the period 1990-1999. They find that GDP growth negatively; real estate prices and real interest rate positively affect credit risk. Moving towards a more recent study, Bofondi and Ropele (2011) use aggregate data for two different classes of borrowers in the Ital-ian banking system for a period of twenty years to examine the macroeco-nomic determinants of loans’ quality. The authors use a single-equation time series method to explain impaired loans for firms and households. They found that NPLs are mainly influenced by variables measuring the general state of the economy, the burden of debt and disposable income. Furthermore, bad loans for households have an adverse relationship with money growth, consumption of durable goods and the housing price index. This points to a negative feedback loop from banking system health to the broader economic environment. In addition to the above variables, bad loans to firms are also negatively affected by gross fixed investments and the values of the stock prices, while exhibiting a positive relation with the lag autoregressive coefficient, the degree of leverage and the slope of the yield curve.

2.3 Contribution to the existing literature

Although studies have examined the determinants of NLPs for many countries, the studies that focus on Greece are scarce. Louzis, Vouldis, & Metaxas (2010) focus on the macroeconomic and bank-specific determi-nants of the NPLs, with individual banking data of nine institutions, from 2003Q1 to 2009Q3. They found that NPLs were affected by GDP growth, lending rates and unemployment. Furthermore, they found a negative

(18)

rela-18

tionship between NPLs and Return on Equity (ROE) and Return on As-sets (ROA). Furthermore, Makri and Papadatos (2014) analyzed the credit risk in the Greek banking system using loss loan provisions (LLPs) as a proxy from 2001 to 2012. Their findings indicate that LLPs are positively affected by unemployment, public debt, loans loss provisions of the previ-ous quarter and negatively by the capital adequacy ratio.

In this study, I try to define the macroeconomic determinants of Non-performing loans to Total loans ratio from 2001Q1 to 2016Q4 for total loans and from 2005q1 to 2014q3 for different categories of loans (consum-er, mortgages, business). The contribution of this study to the existing literature is the following: First, I focus on the aggregate Non-Performing Loans (aggregate banking data), instead of individual banking data (Louzis, Vouldis, & Metaxas, 2010). This method is widely used, for exam-ple at Salas and Saurina (2002), Vogiazas and Nikolaidou (2014), Fainstein and Novikov, (2011), Jakubík and Reininger (2013). Second, this study explains the determinants of NPLs’ for the last 15 years for total loans, and 10 years for the three categories of NPLs, including both the booming period as well as the crisis. Finally, this study incorporates variables such as the long-term interest rate, the private indebtedness, the housing price index, the nominal exchange rate, the stock price index, investments, and the fiscal balance which have not been used in this combination before in the case of Greece.

(19)

19

3 Data and Variables description

In this section, I will present the main features of the data, I will de-scribe the variables that are selected based on the analysis of the Greek crisis and the findings of previous studies, as well as the formed hypothesis and the expected results.

3.1 Data Description

My dataset consists of nineteen time-series that are retrieved from sev-eral sources. Table 1 presents the name of the data, the description, the units of account, their sources, the expected sign according to the hypothe-sis that I will explain later, and the supporting literature.

The data can be divided into two subsets. The first subset is that of the dependent variables. It comprises the aggregate NPLs to total loans ratio (2001Q1 – 2016Q4) and the ratio of NPLs of each borrower type to total loans (2005Q1 2014Q3). Although NPLs exhibit similar dynamics, the ex-isting guidelines indicate that a distinction between the different types of risk exposure enables the final impact to be determined adequately (Marcelo, Rodríguez, & Trucharte, 2008). Furthermore, different catego-ries of borrowers (business, consumer, mortgages) are affected by different macroeconomic indicators (Louzis, Vouldis, & Metaxas, 2010; Bofondi & Ropele, 2011). The second subset of variables consists of macroeconomic indicators measuring the general state of the economy, private and public indebtedness and price stability.

The description, relevance and the expected signs of the selected varia-bles are as follows:

3.1.1 Dependent Variables.

NPLs are considered the primary transmission channel of macroeconom-ic shocks to the banks’ balance sheets and are used in stress test modeling (Buncic & Melecky, 2012). Credit Risk is measured as the ratio of Non-Performing Loans to Total loans. European supervisors consider a loan non-performing when more than 90 days have passed without the borrower paying the agreed installments (European Central Bank, 2017)

(20)

20

3.1.2 Independent Variables

• There is substantial evidence from empirical studies that concludes favorable economic conditions (positive GDP growth, a rise in employment and disposable income) imply that both businesses and consumers have sufficient income to repay their obligations. Thus, in the case of increasing GDP and disposable income, I expect relatively low credit risk. Conversely, unemployment is expected to have a positive relationship with NPLs.

• Long-term government bond yields contain information about mar-ket participants’ expectations of the future economic activity. Furthermore, what we have seen in the Greek case is that financial intermediaries were borrowing from abroad (repo market), using government bonds as collat-eral. Therefore, we can assume that increasing yields of long-term govern-ment bonds6_{caused a deterioration of the lending conditions of the}

bank-ing system. I expect a positive relation between long-term bond yields and credit risk.

• There are two variables included in the analysis to account for pub-lic finances: pubpub-lic debt and fiscal deficit/surplus. One channel through which a sovereign crisis is transmitted to the balance sheets of banks is the following: An increase in government debt may give incentives to the gov-ernment to force banks to buy its securities. Then, banks’ balance sheets are affected directly if the government defaults, which in turn imply a squeeze on lending and rise in credit risk (Reinhart & Rogoff, 2010). Fur-thermore, the sign of fiscal deficit/surplus is ambiguous7_{. On one hand, a}

rise in government expenditure has a positive impact on individual’s in-come. Similarly, austerity measures (cuts in social expenditure and the wage component of government consumption) imposed by the government to consolidate public finances affects negatively the loan servicing capacity of households and businesses. On the other hand, a rise in government ex-penditure leads to a lower present value of future after-tax income, gener-ating a negative wealth effect on private income and consumption (Baxter & King, 1993). Ultimately, this leads to lower output of the economy, worsening the servicing capacity of borrowers.

6_{Debt contracts with floating rate.}

7_{Previous studies (Makri and Bellas 2014; Ghosh,2014) that have tested this hypothesis have}

mixed results, although the coefficients of those studies were not statistically significant different from zero.

(21)

21

• A hike in real interest rates causes a decline in the servicing capaci-ty of borrowers, as it increases the installments’ payment and the value of debt. This way, it leads to an increase in defaults and higher credit risk. The effect is expected to be stronger mainly in categories of loans that are denominated with variable rates.

• A rise in inflation implicitly amortizes debt and makes debt servic-ing easier by reducservic-ing the real value of outstandservic-ing loans. However, when nominal wages are sticky, a higher rate of inflation reduces real income, which in turn may lead to higher NPLs. The sign of inflation is ambiguous.

• The lack of competitiveness of the Greek economy is a critical issue of the ongoing crisis. An increase in the exchange rate means that export-companies are less competitive, business activity shrinks, and it is more difficult to service debt. A depreciation of the nominal effective exchange rate encourages export-companies to increase their turnover and, thus, their ability to repay their loans (Fofack, 2005). In contrast, it increases the debt servicing cost in local currency terms and the vulnerability of un-hedged borrowers with loans dominated in foreign currency8_(Beck,

Jakubik, & Piloiu, 2013). Thus, the sign between credit risk and the nomi-nal effective exchange rate is indeterminate.

• A rise in private indebtedness can have mixed implications. Pesola (2007) indicates that high private indebtedness makes borrowers more vul-nerable to adverse shocks affecting their income, which increases the prob-ability of running into debt servicing problems. Furthermore, in economic upturns, an increase in private indebtedness may reflect softening in credit standards and thus indicate an inadequate risk management. In this case, I expect a positive correlation with NPLs.

• To compensate for changes in the real and financial wealth, we ac-count for the housing price index and the Greek stock price index. An in-crease in the prices of houses boosts the value of the underlying asset used as collateral, thus eases the access of the owners to credit. Furthermore, as the housing price index increases responding to a solid property market, in case of difficulty in meeting debt obligations, a household can easily sell the house and repay the existing loan. Similarly, a buoyant stock market means additional income to business and households for servicing their debt. Consequently, I expect a negative correlation with credit risk.

(22)

22

• A rise in gross fixed capital formation boosts aggregate consumption and reflects a buoyant business outlook. I anticipate a negative sign with credit risk.

3.2 Statistical analysis

Tables 2a and 2b depict the main features of the statistical distributions of the time-series variables described. Furthermore, figures 6a, 6b (Appen-dix) show the original series, many of them found to be stationary accord-ing to ADF test, KPSS and Philips – Perron test (Table 3a, Figure 3b). Hence, following the relevant literature, their logarithmic9_{differences have}

been taken (Figure 7a, 7b - Appendix) to remove the unit root and they are standardized to allow for comparisons (Lopez, Colli, & Coporale, 2013; Ghosh, 2015).

4 Methodology

The current study uses time series data to capture the dynamics of cred-it risk. Based on the empirical lcred-iterature, a single equation time series is employed (Arpa, Giulini, Ittner, & Pauer, 2001; Kalirai & Scheicher, 2002; Bofondi & Ropele, 2011; Lopez, Colli, & Coporale, 2013).

𝛥𝑙𝑜𝑔𝐶𝑅_𝑡 = 𝑎 + 𝑝 𝛽_𝑗𝛥𝑙𝑜𝑔𝐶𝑅_{𝑡− 𝑗}

𝑗= 1

+ q 𝛾_𝑠,𝑗𝛥𝑙𝑜𝑔𝛸_{𝑠,𝑡− 𝑗}

𝑗= 1

+ 𝜀_𝑡

Thus, Credit Risk (CR), which is the ratio of the non-performing loans to total loans, is regressed on its own past values and on current and past values of a set of macroeconomic explanatory variables. The lag structure of the above equation is selected by checking the statistical significance of estimated coefficients and the Akaike and Schwartz information criteria. The latter applies also for the decision of the inclusion or not of a variable into the model. Furthermore, is an intercept, 𝜀𝑡 is the error term, p is

the lag order autoregressive component and, similarly, q is the lag order of the other independent variables.

(23)

23

5 Empirical Results

5.1 Macroeconomic determinants of Credit Risk (Total

Loans)

To begin with, I regress the Credit risk of total loans in a restricted set of variables, namely, GDP, Unemployment, Long-Term interest rate. Spe-cifically, regression [a] considers GDP and Long-Term interest rate, regres-sion [b] replaces GDP with Unemployment and [c] includes all three varia-bles together. As Table 4 indicates, estimated coefficients are highly statis-tically significant and have the expected sign. An increase in GDP affects credit risk negatively with one-quarter lag, while Unemployment and Long-Term interest rate are positively associated with a deterioration in loan quality with two and ten-quarter lags respectively. Furthermore, the autoregressive term is statistically significant in all models indicating the persistent nature of NPLs. On the basis of the goodness of fit, as well as of the Akaike and Schwarz criteria, model [c] is the preferred one to serve as a baseline model for the analysis.

Based on this conclusion, models [d], [e], [f], [g], [h], [i], [j], [k] expand the baseline model with other macroeconomic variables, namely Fiscal Deficit, Debt to GDP, Disposable Income, Household Debt, Gross Capital For-mation, Inflation, Housing Price Index, Share Price adding them one by one. With the exception of Unemployment in model [d], the variables of the baseline equation remain highly statistically significant. It is worth noting the highly significant coefficient of Long-Term interest rate in all specifications, which highlights the influence of a country’s risk on the quality of its bank’s assets through the channel of bonds used as collateral on the balance sheet of banks. I use government Debt to GDP and Fiscal to measure the impact of public finances on credit risk. The negative rela-tion of credit risk with Fiscal in model [d] matches the fact that a great deal of government expenditure was boosting private income, thus reduc-ing credit risk and enablreduc-ing households to service their loans. Moreover, an increase of public debt augments credit risk (Table 4, model [e]). This is in line with the fact that the Greek financial intermediaries were exposed to government bonds and faced losses in their balance sheets, which were

(24)

24

then transmitted to borrowers through a financing squeeze. Furthermore, as expected, the relation of credit risk with disposable income is negative, instantaneous and with 3-quarter lags respectively. In addition, an increase in private investments and share price (both with 5-quarter lags) reduces credit risk.

Furthermore, the coefficients of inflation and housing are not statistically different from zero (Models [i] and [j]).

Finally, I tested the joint inclusion of all variables. As reported in model [l], in addition to the baseline model, the variables of public finances, household debt and private investments are statistically significant and with the expected signs. The next step is the estimation of a model includ-ing all variables found to be statistically significant in model [m]. Com-pared with the previous models, the value of adjusted R2_{increases further,}

and Akaike and Schwarz criteria also improve.

5.2 Macroeconomic Determinants of Credit Risk (Business

loans, Mortgages, Consumer loans)

Following the same strategy as with the credit risk of total loans, we an-alyze the determinants of credit risk of different categories of loans (busi-ness, mortgages, consumer) on just three variables (GDP, unemployment rate, and lending rates) and then extend the model with the inclusion of a list of variables. Tables 5-7 report the results.

In the case of business loans (Table 5), the estimated coefficients have the expected sign in models [a], [b], and [c]. An increase in unemployment raises the credit risk with 2-quarter lags, while GDP is inversely related to the NPLs with 1-quarter lag. Lending rates’ coefficients are statistically significant at 1% significance level in specifications [b] and [c]. Further-more, the second lag of credit risk is statistically significant in these three models. Model [a], which includes unemployment and GDP, has a satisfy-ing goodness of fit and accordsatisfy-ing to Akaike and Schwarz criteria this speci-fication is the appropriate baseline.

Later on, as in the case of total loans, fiscal deficit/surplus and Debt to GDP ratio variables are statistically significant at 10% and 1% level re-spectively and have the expected signs (models [d] and [e]). As model [f] presents, a rise in disposable income of households reduces credit risk in the same quarter. Furthermore, a rise in Private Investments (model [g])

(25)

25

of the private sector reduces credit risk with 2-quarter lag. The interpreta-tion of this result reflects that fixed investments is a leading indicator of the next business cycle and usually concerns long-term decisions of corpo-rations. An increase in investment spending of firms indicates a good out-look which induces financial intermediaries to refinance their loans. In model [h], a decline in financial wealth deteriorates the quality of business loans with 3-quarter lags, highlighting the fact that deflation of asset pric-es generatpric-es a negative balance sheet effect for firms and, with a time lag, for their lenders. Furthermore, the nominal effective exchange rate has no explanatory power (Model [i]). Finally, in specification [j] we include all variables together, and in model [k] we test only those that have explana-tory power in model [j]. Model [k] explain more than 77% of the observed variation of credit risk for business loans and is the preferred one according to Akaike and Schwarz criteria.

Next, in the case of credit risk related to households, we focus on mort-gages and consumer loans. The results (Table 6 and 7) highlight the strong relation of households with public finance variables, the persistence of the autoregressive coefficient and the significance of the regressors measuring real and financial wealth.

In particular, models [a] and [c] of the credit risk for mortgages (Table 6) have about the same goodness of fit and according to Akaike and Schwarz criteria model [a] is a better specification to serve as the baseline model. As expected, an increase in lending rate has a positive relation with credit risk. Furthermore, public finance variables (Debt to GDP, Fiscal deficit) explain credit risk also in the case of mortgages but the goodness of fit slightly increases. A noteworthy finding in specification [d] is that the magnitude of the coefficient of fiscal deficit, shows a strong negative rela-tionship with the credit risk of mortgages and is larger in comparison to the other categories of loans. This result highlights, (all else equal) the effect of the heavy property taxation imposed from 201010_{onwards which}

resulted in a reduction of the fiscal deficit and a surge of NPLs. Both coef-ficients of share value and asset prices have high explanatory power (mod-els [f] and [g]) giving support to the aforementioned argument for the busi-ness loans. The nominal effective exchange rate has a positive effect on

10_{According to (DG Taxation and Customs Union Taxation , 2017) taxes on property increased}

from 1% to 2,7% of GDP. In detail, taxes on land, buildings or other structures raised from 446 to 3317 million euro (Commission, 2017).

(26)

26

credit risk for mortgages which can be explained by the fact that an im-portant amount11_{of mortgages was denominated in foreign currency}

(mainly in Swiss francs), which appreciated significantly from 2008 on-wards, increasing default rates. In addition, the autoregressive coefficient is statistically significant in all models. According to the goodness of fit and the Akaike and Schwarz criteria, specification [l] is the most appropriate to model the credit risk of mortgages.

In the case of the credit risk of consumer loans (Table 7), specification [b] appears to have the best goodness of fit among the first three models and includes output and unemployment (Akaike and Schwarz criteria also con-firms that conclusion). Subsequently, the baseline model for consumer loans is augmented with the inclusion of five supplementary variables ac-counting for public finances, household indebtedness, housing price index and the value of shares. All coefficients, except for the private indebted-ness variable, have the expected signs. In model [h], stock value has a neg-ative relationship with credit risk. Disposable income and nominal effective exchange rate have no explanatory power in specifications [i] and [j]. In addition, as in the cases of business and mortgages, specification [k] jointly includes all the variables. Finally, regression [l], which takes into account the variables of public finances and wealth, has the best goodness of fit compared with the other models and AIC and BIC criteria improve.

11_{According to (Avlogiari, 2017) 75000 unhedged borrowers faced difficulties to repay their loans}

(27)

27

6 Discussion

Comparing my results with those of earlier studies, the positive sign of the autoregressive coefficient of the credit risk is in line with the findings of previous studies for euro area countries such as Italy (Bofondi & Ropele, 2011) as well as for studies focused on other regions (Ghosh, 2015; Beck, Jakubik, & Piloiu, 2013). This result indicates, that all else equal, a posi-tive movement in NPLs is statistically more likely to be related with an increase of credit risk in the next period. It is worth noting that this result is in contrast with previous results of Louzis, Vouldis, & Metaxas (2010) for the Greek banking sector. Specifically, they found a significant negative coefficient for Business and Consumer loans implying that NPLs are likely to decrease when they have increased in the previous quarter due to the loan write-offs. However, this difference can be explained by the different time span of our datasets, as I also include the crisis period.

Next, comparing the findings of the country’s economic conditions vari-ables, the results for GDP growth and unemployment match the results of a broad body of the existing literature (Salas & Saurina, 2002; Nkusu, 2011; Fainstein & Novikov, 2011; Skarica, 2014) confirming the countercy-clical behavior of credit risk. Furthermore, the long-term interest rate that has a positive sign and is statistically significant, is in contrast with other studies that didn’t find a significant relationship (Bofondi & Ropele, 2011).

In all models, the impact of Government Debt to GDP ratio in credit risk is found to be statistically significant and exerts the expected sign, and follows suit with the existing literature for other European countries (Makri, Tsagkanos, & Bellas, 2014), United States (Ghosh, 2015) and sin-gle country analysis for Greece (Louzis, Vouldis, & Metaxas, 2010; Makri & Papadatos, 2014). Turning to Fiscal Deficit, this is a variable tested for the first time in the Greek case, the sign of the coefficient is in line with previous studies for Europe (Makri, Tsagkanos, & Bellas, 2014) and in contrast with the regional study of Ghosh (2015). Unlike the other studies (Ghosh, 2015; Makri, Tsagkanos, & Bellas, 2014) this coefficient is statisti-cally significant different from zero in all categories of loans indicating the effect of fiscal policy on banks’ asset quality. Finally, in all models the role of financial and housing wealth (proxied by the share price and housing

(28)

28

price index) appears to have high explanatory power giving support to the hypothesis that wealth can be used as a buffer against negative shocks and that deflationary pressures on asset prices affect asset quality (Kazuo, 2003).

Notably, different macroeconomic variables have different effect depend-ing on the type of borrowers. Nominal effective exchange rate is negatively related with credit risk of mortgage loans, providing further evidence to the existing literature about the vulnerability of unhedged borrowers with loans dominated in foreign currency (Beck, Jakubik, & Piloiu, 2013). Addi-tionally, in line with a previous study results for corporate loans in the Italian banking system (Bofondi & Ropele, 2011), investment is found to exert significant explanatory power also in the Greek case. An interesting observation is the strong dependence of business loans on the disposable income of households. This can be explained by the fact that the highest concentration of business NPLs is noted in the case of small and medium size enterprises (Bank of Greece, 2017) that are most active in the service sector (50% of the total) and a large proportion of them in the retail trade. In general, the positive sign of real interest rates in credit risk is consistent with the studies of Shu (2002), Beck, Jakubik, and Piloiu (2013), Ali and Daly (2010) and Rinaldi & Sanchis‐Arellano (2006). Although, their ex-planatory power is limited, this is in line with the findings of previous studies for the Greek banking system (Louzis, Vouldis, & Metaxas, 2010) conducted in the pre-crisis period, that indicated a relative insensitivity of mortgages and business loans to lending rates. In addition, the insensitivi-ty coefficient of real interest rate for consumer loans is in contrast with previous studies results (Louzis, Vouldis, & Metaxas, 2010).

(29)

29

7 Conclusion

Understanding the determinants of asset quality deterioration is of great importance for financial and economic stability. In this empirical study, I explain the events that caused an unprecedented banking crisis for Greece utilizing theoretical reasoning and econometric testing.

I use time series data to investigate the macroeconomic determinants of credit risk in the Greek banking sector. I find that variables such as GDP and unemployment have significant impact in all categories of NPLs. Fur-thermore, particular categories of NPLs are affected by different variables and in diverse ways than in the pre-crisis period. Finally, a set of other variables measuring the impact of fiscal policy, namely Debt to GDP ratio and fiscal deficit/surplus, are incorporated in this analysis and exert signif-icant influence on credit risk, highlighting the destructive feedback loop effect of the sovereign and banking crisis.

Further understanding the nature of credit risk raises an open question that has been covered for other countries (Lopez, Colli, & Coporale, 2013) and needs to be addressed in future research for the Greek case; whether excessive loans granted during expansionary phases can explain the more than proportional increase in non-performing loans during the crisis period.

(30)

30

8 Bibliography

Ali, A., & Daly, K. (2010, March 17). Macroeconomic determinants of credit risk: Recent evidence from a cross country study. International Review of Financial Analysis, σσ. 165-171.

Arpa, M., Giulini, I., Ittner, A., & Pauer, F. (2001). The influence of macroeconomic developments on Austrian banks: implications for banking supervision. BIS.

Athanasoglou, P., Georgiou, E., & Staikouras, C. (2009). Assessing output and productivity growth in the banking industry. The Quarterly Review of Economics and Finance, Volume 49(Issue 4), Pages 1317-1340.

Avlogiari, E. (2017). FX Loans. Hellenic Review of European law, International Edition, 55-66.

Bank of Greece. (2014). The Chronicle of the Great Crisis, Public interventions and institutional actions to safeguard financial stability and overcome the crisis. Athens: Bank of Greece.

Bank of Greece. (2017). Report on Operational Targets for Non-Performing Exposures. Athens: Bank of Greece.

Baxter, M., & King, R. G. (1993). Fiscal Policy in General Equilibrium. 83, 315–334.

Beck, R., Jakubik, P., & Piloiu, A. (2013). Non-Performing Loans: What Matters in Addition to the Economic Cycle? ECB Working Paper Series 1515.

Bofondi, M., & Ropele, T. (2011). Macroeconomic determinants of bad loans: evidence from Italian banks. Questioni di Economia e Finanza (Occasional Papers) della Banca d’Italia, n. 89.

Borio, C., & Lowe, P. (2002, July). Asset prices, financial and monetary stability: exploring the nexus. Monetary and Economic Department. Bank for International Settlements.

Bortz, P. G. (2015, November 19). The Greek 'Rescue': Where Did the Money Go? An Analysis. Institute for New Economic Thinking Working Paper Series(29), σ. 40.

Brookes, M., Dicks, M., & Pradhan, M. (1994). An Empirical Model of Mortgage Arrears and Repossessions. Economic Modelling, 11, 134-144.

(31)

31

Buncic, D., & Melecky, M. .. (2012). Macroprudential Stress Testing of Credit Risk – A Practical Approach for Policy Makers. World Bank Policy Research Working Paper 5936.

Constâncio, V. (2014). The European Crisis and the Role of the Financial System. Journal of Macroeconomics, Volume 39(Part B), Pages 250-259.

Constancio, V. (2017). Resolving Europe’s NPL burden: challenges and benefits . Tackling Europe's non-performing loans crisis: restructuring debt, reviving growth. Brussels: Bruegel.

Darvas, Z. (2012, March 15). Real effective exchange rates for 178 countries: A new database. Working Paper , 06.

DG Taxation and Customs Union Taxation . (2017). Trends in the European Union Data for the EU Member States, Iceland and Norway. Luxembourg: European Union.

E.U. Directorate for Economic and Financial Affairs. (2010). The Economic Adjustment Programme for Greece. European Commission. Brussels: European Union.

Espinoza, R., & Prasad, A. (2010). Nonperforming Loans in the GCC Banking Systems and their Macroeconomic Effects. International Monetary Fund. Washington: IMF Working Paper 10/224.

European Central Bank. (2017). Guidance to banks on non-performing loans. Frankfurt am Main: European Central Bank.

European Commission. (2017). TAXATION AND CUSTOMS UNION. Ανάκτηση από Taxation and Custom Union: https://ec.europa.eu/taxation_customs/business/economic-analysis-taxation/data-taxation_en

Fainstein, G., & Novikov, I. 2. (2011, March). The Comparative Analysis of Credit Risk Determinants in the Banking Sector of the Baltic States. Review of Economics and Finance.

Featherstone, K. (2011). The Greek Sovereign Debt Crisis and EMU: A Failing State in a Skewed Regime. Journal of Common Market Studies, 193–217.

Fofack, H. (2005). Non-performing loans in sub-Saharan Africa: Causal Analysis and Macroeconomic Implications. World Bank Policy Research Working Paper n° 3769.

Gagales, A., Rossi, M., & Badia, M. (2007). Greece: Selected Issues. Washington: International Monetary Fund.

(32)

32

Garcia-Marco, T., & Robles-Fernandez, M. D. (2008). isk-taking Behaviour and Ownership in the Banking Industry: The Spanish Evidence. Journal of Economics and Business, 60(4), 332-354.

Georgakopoulos, T. (2016). Tax Evasion in Greece – A Study. Athens: Dianeosis.

Ghosh, A. (2015, August 19). Banking-industry specific and regional economic determinants of non-performing loans: Evidence from US states. Journal of Financial Stability, σσ. 91-103.

Gibson, H., Hall, S., & Tavlas, G. (2014). Fundamentally wrong: market pricing of sovereigns and the Greek financial crisis. Journal of Macroeconomics, 39, 405–419.

Godlewski, C. (2004). Bank capital and credit risk taking in emerging market economies. Journal of Banking Regulation, 6(2), 128-145. IMF. (2006). Financial Soundness Indicators Compilation Guide.

Washington: International Monetary Fund.

Ioannou, D. (2013). The structural collapse of the Greek economy. Foreign Affairs.

Jakubík, P., & Reininger, T. (2013). Determinants of Nonperforming Loans in Central, Eastern and Southeastern Europe. Focus on European Economic integration. Oesterreichische Nationalbank. Kalirai, H., & Scheicher, M. (2002). Macroeconomic Stress Testing:

Preliminary Evidence for Austria. Austrian National Bank Financial Stability Report , May, no. 3.

Kazuo, U. (2003). Japan's Deflation and Policy Response. Meeting on Economic and Financial Matters in Nara City. Nara Prefecture: Bank of Japan.

Keeton, W., & Morris, C. (1987, May). Why do banks’ loan losses differ? Economic Review, 3-21.

Khemraj, T., & Pasha, S. (2009). The Determinants of Non-Performing Loans: An Econometric Case Study of Guyana. Munich Personal RePEc, 53128.

Lopez, J., Colli, S., & Coporale, M. (2013). Bank Lending Procyclicality And Credit Quality During Financial Crises. Economics and Finance Working Paper Series, Department of Economics and Finance. Brunel University London.

Louzis, D., Vouldis, A., & Metaxas, V. (2010). Macroeconomic and bank-specific determinants of non-performing loans in Greece: A

(33)

33

comparative study of mortgage, business and consumer loan portfolios. Journal of Banking & Finance, 1012–1027.

Macit, F. (2012). What determines the non-performing loans ratio: evidence from Turkish commercial banks. Cent. Econ. Anal. J. Econ, 13, 33–39.

Makri, V., & Papadatos, K. (2014). How accounting information and macroeconomic environment determine credit risk? Evidence from Greece. International Journal of Economic Sciences and Applied Research , 7(1), 129-143.

Makri, V., Tsagkanos, A., & Bellas, A. (2014). Determinants of Non-Performing Loans: The case of Eurozone. PANOECONOMICUS 2, 193-206.

Marcelo, A., Rodríguez, A., & Trucharte, C. (2008, April 1). Stress tests and their contribution to financial stability. Journal of Banking Regulation, σσ. 65–81.

Mitsopoulos, M., & Pelagidis, T. (2011). Understanding the crisis in Greece: from boom to bust. Basel: Springer.

Muguruza, B. T., Efstathios Efstathiou, A. K., Savin, A., & Diana, G. (2017). The Commission's intervention in the Greek financial crisis. Brussels: European Commission.

Nikolaidou, E., & Vogiazas, S. (2014). Credit Risk Determinants for the Bulgarian Banking System. International Advances in Economic Research, 87–102.

Nkusu, M. ‘. (2011). Nonperforming Loans and Macrofinancial Vulnerabilities in Advanced Economies’. IMF Working Paper , 11(161).

Pagoulatos, G. (2014, July 22). State-driven in Boom and in Bust: Structural Limitations of Financial Power in Greece. Government and Opposition, σσ. 452-482.

Pagoulatos, G., & Triantopoulos, C. (2009). The Return of the Greek Patient: Greece and the 2008 Global Financial Crisis. South European Society and Politics, 35-54.

Pesola, J. (. (2007). Financial fragility, macroeconomic shocks and banks' loan losses: evidence from Europe. Bank of Finland Research Discussion Papers n°15.

Petrakis, P. E. (2012). The Greek Economy and the Crisis: Challenges and Responses. Basel: Springer International Publishing AG.

(34)

34

Provopoulos, G. (2014). The Greek Economy and Banking System: Recent Developments and the Way Forward. In Journal of Macroeconomics, Volume 39(Part B), 240-249.

Reinhart, C. M., & Rogoff, K. S. (2009). This Time is Different: Eight centuries of financial folly. Princeton: Princeton University Press. Reinhart, C., & Rogoff, K. (2010, March). From financial crash to debt

crisis. Working Paper 15795, σ. 46.

Rinaldi, L., & Sanchis‐Arellano, A. (2006). Household debt sustainability: What explains household non‐performing loans: An empirical analysis. ECB Working Paper n° 570.

Salas, V., & Saurina, J. (2002). Credit risk in two institutional regimes: Spanish commercial and savings banks. J. Financ. Serv. Res.(22), 203–224.

Shu, C. (2002). The Impact of Macroeconomic Environment on the Asset Quality of Hong Kong's Banking Sector. Hong Kong Monetary Authority Research Memorandums.

Sinkey, J. F., & Greenawlat, M. B. (1991). Loan-Loss Experience and Risk- Taking Behaviour at Large Commercial Banks. Journal of Financial Services Research, 5(1), 43-59.

Skarica, B. (2014). Determinants of non-performing loans in Central and Eastern European countries. Financ. Theory Prac(38), 37–59.

Vatansever, M., & Hepşen, A. (2015). etermining Impacts on Non-Performing Loan Ratio in Turkey. Journal of Applied Finance & Banking, 5(1), 1-11.

(35)

35

(36)

36

Definition Units Source Expected Sign Literature

Credit Risk Total loans Credit Risk Business loans Credit Risk Mortgage loans Credit Risk Consumer loans

Gross Domestic Product (GDP) Gross Domestic Product (seasonally adjusted) mln. Euros Eurostat, Elstat

- Sales and Saurina, 2002; Kalirai and Scheicher, 2002; Shu, 2002; Rajan and Dhal, 2003; Quagliariello, 2003; Jimenez and Saurina, 2005; Fofack, 2005; Babouèek and Janèar, 2005; Louzis et al., 2010; Festi et al., 2011, Macit, 2012; Ghosh, 2014;Skarica, 2014;

Unemployment (UN) Quarterly percentage of Unemployment rate

% Eurostat, Elstat + Brookes et al., 1994; Babouèek and Janèar, 2005; Jakubík, 2007; Louzis et al., 2010, Nkusu, 2011; Skarica, 2014;

Long term Interest rate (LIR) Long-term interest rate for convergence purposes - 10 years maturity,

percent %

Eurostat, Elstat + Keeton and Morries, 1987; Sinkey and Greenwalt, 1991; Shu, 2002; Fofack, 2005; Gerlach et al., 2005; Jakubík, 2007; Louzis et al., 2010; Beck et al. 2013; RIR Business loans

RIR Mortgage loans RIR Consumer loans

Disposable income Disposable income, gross

mln. Euros Eurostat (Author's Calculation)

Bofondi and Ropele, 2011; Fiscal Deficit Ratio of Net borrowing Lending of General Government to GDP

% Elstat (Author's Calculation)

+ / - Ghosh, 2012; Makri, 2014; Government Debt Ratio of Government consolidated gross debt to GDP % Eurostat + Ali and Daly, 2010; Louzis, 2012; Household Debt Ratio of Morgages, Consumer loans and other liabilities to GDP, in

percentage (Definition of Househod Indebtness according to IMF (2006) %

Bank of Greece (Author's Calculation)

+ / - Sales and Saurina, 2002; Rinaldi et al., 2006; Bofondi and Ropele, 2011;

Inflation (In) Consumer Price index

%

Eurostat + / - Brookes et al., 1994; Shu, 2002; Kalirai and Scheicher, 2002; Babihuga, 2007; Babouèek and Janèar, 2005; Dash and Kabra, 2010, Ghosh, 2014; Rinaldi et al., 2006;

Housing Price index (HPI) Quarterly change of Housing Price Index

%

Bank of Greece (Author's Calculation)

- Arpa et al., 2001; Shu, 2002; Rinaldi et al., 2006; Quagliarello 2007; Nkusu, 2011; Bofondi and Ropele, 2011; Vatansever & Hepsen, 2013; Ghosh, 2014; Skarica, 2014;

Share Value Share price indices are calculated from the prices of common shares of companies traded on national or foreign stock exchange

OECD - Nkusu, 2011; Bofondi and Ropele, 2011; Vatansever & Hepsen, 2013;

Real Effective exchange rate Real effective exchange rate calculated against 30 trading partners Brugel, Darvas (2012) +

Nominal Effective Exchange Rate Calculated as geometric weighted averages of bilateral exchange rates BIS + / - Nkusu, 2011; Macit, 2012; Beck et al., 2013; Private Investments Gross Fixed Capital Formation of the Non Fin. Sector mln. Euros Elstat - Bofondi and Ropele, 2011;

Disposable income Adjusted gross disposable income of households thous. Euros Eurostat - Pesola, 2007; Bofondi and Ropele, 2011; Table 1. Data description , Sources and Suported Literature

Credit Risk is measured as the ratio between the Non-performing Loans and the gross loans of each category

Dependent Variables

Independent Variables

Sinkey and Greenawlat, 1991; Brookes et al., 1994; Shu, 2002; Rinaldi et al., 2006; Ali and Daly, 2010; Louzis et al., 2010; Vatansever & Hepsen, 2013; Bank of Greece +

Weighted averages of interest rate (adjusted for inflation) of the outstanding amounts of euro-dominated loans vis-à-vis euro area residents

Gross Loans - Bank of Greece , NPLs - Greek Parliament, World Bank, IMF %

(37)

37 mean sd min max skewness kurtosis N Total Loans 176512,1 58156,04 63066,65 248527,1 -0,5711 1,9942 64 Total NPLs 27390,86 24689,74 4194,29 69335,76 0,7735 1,8556 64 CR Total Loans 0,15 0,12 0,04 0,37 0,9500 2,1436 64 GDP 53458,47 6070,999 45474,5 63334,63 0,1292 1,5534 64 Unemployment rate 15,23 7,32 7,30 27,80 0,6325 1,6481 64 Long-Term interest rate 7,71 5,14 3,41 25,40 1,9827 6,5671 64 Fiscal Deficit -7,32 5,40 -29,99 4,93 -1,1157 6,4182 64 Debt to GDP 131,97 31,14 99,90 181,20 0,4464 1,4563 64 Inflation 0,021 0,022 -0,024 0,055 -0,6523 2,2133 64 HPI 0,001 0,023 -0,042 0,051 0,1997 2,3009 64 Disposable Income 34207,3 5652,184 23678 45620 0,3219 2,1227 64 Household Debt 49,28 15,29 18,93 65,48 -0,7268 1,9842 64 Gross Capital Formation 3037,69 875,57 1108,10 5490,37 0,4883 3,2920 64

Table 2a. Descriptive Statistics - Total Loans 2001q1-2016q4

mean sd min max skewness kurtosis N Business Loans 109809,7 17383,9 73243,6 132923,1 -0,4850 2,2413 39 Mortgage Loans 68726,0 12992,0 35764,2 81088,0 -1,2296 3,3268 39 Consumer Loans 30350,3 5177,3 17969,6 36412,3 -0,6946 2,6191 39 NPL Business 13962,0 10486,7 5170,8 34939,1 0,9797 2,3838 39 NPL Mortgage 8152,5 6373,2 1619,1 19534,6 0,6042 1,8487 39 NPL Consumer 6274,3 4532,2 1432,7 13902,2 0,4661 1,6330 39 CR Business 4,08 0,78 2,84 5,10 -0,0531 1,4584 39 CR Mortgage 13,58 0,79 12,18 14,73 -0,1374 1,6632 39 CR Consumer 6,63 0,64 5,51 7,63 -0,2909 1,7691 39 RIR Business 3,54 2,00 -0,55 7,87 0,1229 2,7403 39 RIR Mortgages 1,81 1,63 -1,91 5,23 -0,2552 3,0240 39 RIR Consumer 8,44 1,72 5,32 12,13 0,2312 2,4708 39 GDP 55308,9 6402,8 45940,3 63334,6 -0,3413 1,5191 39 Unemployment rate 15,10 7,63 7,30 27,80 0,6530 1,7275 39 Fiscal Deficit -9,06 5,58 -29,99 -0,66 -1,2866 6,1572 39 Debt to GDP 133,40 28,50 102,20 181,20 0,3449 1,5561 39 HPI -0,005 0,024 -0,042 0,041 0,3468 2,1786 39 Disposable Income 37103,4 4970,5 27244,0 45620,0 -0,0348 1,9727 39 Household Debt 55,99 7,83 36,17 65,48 -0,7793 2,8259 39

Gross Capital Formation 3168,34 980,87 1108,10 5490,37 0,3738 2,8335 39 Real Effective Exchange Rate 74,24 13,11 49,03 100,00 -0,2896 2,3246 39

neer 100,32 1,64 97,77 103,65 0,4602 2,2691 39

Share Value 138,52 83,25 35,81 300,24 0,4840 1,8361 39

The determinants of credit risk in the Greek banking system