BANK OF GREECE EUROSYSTEM Working Paper Macroeconomic and bank-specific determinants of non-performing loans in Greece: a comparative study of mortgage, business and consumer loan portfolios Dimitrios P. Louzis Angelos T. Vouldis Vasilios L. Metaxas 1 8 SEPTEMBER WORKINKPAPERWORKINKPAPERWORKINKPAPERWORKINKPAPERWO 2010
BANK OF GREECE Economic Researc Department Special Studies Division 21, Ε. Venizelos Avenue GR-102 50 Atens Τel: +30210-320 3610 Fax: +30210-320 2432 www.bankofgreece.gr Printed in Atens, Greece at te Bank of Greece Printing Works. All rigts reserved. Reproduction for educational and non-commercial purposes is permitted provided tat te source is acknowledged. ISSN 1109-6691
MACROECONOMIC AND BANK-SPECIFIC DETERMINANTS OF NON- PERFORMING LOANS IN GREECE: A COMPARATIVE STUDY OF MORTGAGE, BUSINESS AND CONSUMER LOAN PORTFOLIOS Dimitrios P. Louzis Bank of Greece and Atens University of Economics and Business Angelos T. Vouldis Bank of Greece and University of Atens Vasilios L. Metaxas Bank of Greece ABSTRACT Tis paper uses dynamic panel data metods to examine te determinants of nonperforming loans (NPLs) in te Greek banking sector, separately for eac type of loan (consumer, business and mortgage loans). Te study is motivated by te ypotesis tat bot macroeconomic and bank-specific variables ave an effect on loan quality and tat tese effects vary between different categories of loans. Te results sow tat NPLs in te Greek banking system can be explained mainly by macrofundamentals (GDP, unemployment, interest rates) and management quality. Differences in te quantitative impact of macroeconomic factors among types of loans are evident wit non-performing mortgages being te least responsive towards canges in te macroeconomic conditions. Keywords: Non-perfoming loans; Greek banking system; Macroeconomic determinants; Bank specific determinants; Dynamic panel data JEL classifications codes: G21; C23 Aknowledgments: We are grateful to H. Gibson, S. Hall, N. Tsaveas, M. Mavridou, K. Zavandis, N. Stavrianou, V. Nydrioti, I. Zourka, S. Liakopoulou and all colleagues from te Financial Stability Department of te Bank of Greece for teir valuable comments and useful discussions. We would also like to acknowledge te discussants comments on tis paper from M. Melecky at te World Bank Seminar Advances in Stress Testing in Central and Sout Eastern Europe on 19-20 May 2010 in Tessaloniki. Te views of te paper do not necessarily reflect tose of te Bank of Greece. Correspondence: Vouldis Angelos, Financial Stability Department, Bank of Greece, 3 Amerikis Str., 105 64, Atens, Greece, Tel: +30 2103205130, e-mail: avouldis@bankofgreece.gr
4
1. Introduction Exploring te determinant factors of ex post credit risk is an issue of substantial importance for regulatory autorities concerned about financial stability and banks management. Te ex post credit risk takes te form of non-performing loans (NPLs). Despite te fact tat banks ave developed sopisticated tecniques for quantifying ex ante credit risk by focusing on te borrower s idiosyncratic features, te ex post credit risk as reflected in te number of NPLs seems to be primarily driven by macroeconomic developments as te business cycle literature as sown. Anoter strand of te literature as focused on te effect of bank-specific caracteristics suc as te quality of management, policy coices, size and market power on problem loans. Tis paper aims to provide a syntesis of tese two approaces on te issue of NPLs determinant factors. In tis direction, te explanatory power of bot macroeconomic and bank-specific variables will be investigated. Anoter issue, wic according to our knowledge as been neglected so far, is a comparative study of NPLs on different types of loans and teir corresponding determinants. It may be conjectured tat macroeconomic and bank-specific variables ave a differential impact on NPLs depending on te type of loan. Tis could be attributed to institutional settings creating different incentive structures for eac type of loan wit regards to te costs of bankruptcy. Moreover, differences in te sensitivity of various types of NPLs to macroeconomic developments may be linked to differential effects of te business cycle on agents cas flows and collateralized assets values. In tis study, NPLs on consumer, business and mortgage loans are examined separately and teir corresponding determinants are compared. Te paper focuses on te Greek financial sector wic as been affected by te recent macroeconomic developments in te Greek economy. It utilizes a panel data set comprising te 9 largest Greek banks (wic account for approximately 90% of Greece s banking sector) and spans te period 2003Q1 to 2009Q3. Te time period 2003-9 includes bot a period of growt (wic began in te mid-1990s) as well as te downturn. following te financial crisis and te ensuing manifestation of Greece s own structural weaknesses. 5
Te rest of te paper is structured as follows. Section 2 briefly presents te evolution of te Greek banking system and links it wit empirical observations regarding non-performing loans. Section 3 overviews te teoretical and empirical literature on te determinants of non-performing loans and formulates a number of ypoteses relating bank-specific variables to te number of non-performing loans. Section 4 presents preliminary econometric results wile Section 5 investigates te use of a dynamic panel data approac and provides a discussion of empirical results. Finally, Section 6 concludes. 2. Greek banking system and non-performing loans evolution 2.1 Development of te Greek banking sector Until te end of te 1980s te Greek banking system was igly regulated. Regulations pertained bot to te quantity (and direction) of credit supply (e.g. requirements for financing specific sector and in particular te public sector) and te interest rates carged. 1 Voridis (1993) as examined te supply of credit for fixed capital investment in te context of te a financial repressed regime 2 wic caracterized te Greek banking sector trougout te period and found significant distortions as regards te operation of te credit market 3. Te early 1990s saw te gradual lifting of regulative restrictions on credit markets. Te ensuing transformation of te financial system sould be attributed to te aspiration of Greek policy makers to armonize te institutional environment of Greece wit tat of te European Union (following Greece s accession to te EC in 1981). During te 1990s 1 According to te Economic Survey of Greece, publised by te OECD (1986), te regulations on te Greek financial sector featured "extensive controls of volume, direction, and price of credit flows wile direct government intervention was also mentioned (OECD 1986, p. 52). 2 Defined as a financial system were te scope of regulation is pervasive. 3 Specifically, in is estimation of te determinants of investment,,e finds a positive sign for all te significant lags of te user cost of capital (wic approximates te real cost of capital). Subsequently, a number of arguments from te literature on financial repression are invoked to justify tis sign wic contradicts te predictions of standard neoclassical teory. 6
liberalization proceeded more decisively as te prospect of joining te EMU became a primary goal for Greece. 4,5 Overall, te 1990s constituted a landmark for te development of te Greek financial system. From te late-1980s to te early 2000s, a total of sixteen commercial banks were incorporated (Kamberoglou et al 2004). A wave of mergers and acquisitions was observed during tat period, caused by financial institutions need to capture market sare, exploit economies of scale and become more efficient 6. Tecnological improvement was reflected in increased investment in IT. Tecnological progress accordingly led to te introduction of innovative products (Panopoulou 2005) 7. Furtermore, a surge in branc expansion took place as te Greek credit market was underdeveloped compared to te average of te EU and banks sougt to expand and gain market sare (Eicengreen and Gibson 2001, p. 563). Competition intensified as te relatively small banks exibited ig growt rates and gained market sare 8. Te decade starting in 2000 saw a continuation of tis trend and accordingly credit expansion proceeded apace, following Greece s accession to te euro area in 2001 and te impetus for growt provided by te Olympic Games in 2004. Tere were also significant canges in te direction of furter improving te institutional settings in te financial sector (suc as te setting up of a national credit bureau in te form of an interbank company, te imperative to comply wit te standards set fort by Basel II (2007) and te new bankruptcy law enacted in 2007). Te outbreak of te financial crisis in 2007 and te deterioration of public finance in Greece, wic became evident in 2009, put te banking sector in a stressed situation. Te transformation of te economic environment witin wic te banks in Greece operated dictated a cange of attitude wit regards to risk management and as a 4 Gibson and Tsakalotos argue tat te general trend towards less government since te late 1970s was anoter contributing factor conducive to financial liberalization (Gibson and Tsakalotos 1992, p. 11). 5 Specifically, te Second Banking Directive enacted by te EU in 1992 was critical in directing efforts towards armonization of national regulations. 6 For a more detailed account of te mergers wave in te Greek financial system during te 1990s, see Eicengreen and Gibson (2001, p. 584-7) 7 Cristopoulos and Tsionas (2001) found tat tecnical inefficiency dropped drastically during te postliberalization period. 8 Te downward trend of te Herfindal-Hirscman index during te 1990s testifies to tis empirical observation. 7
consequence is expected to ave affected decisively te factors determining nonperforming loans (NPLs). 2.2 Non-performing loans in Greece: institutional settings and evolution Before te liberalization of te financial sector, te regulatory restrictions determined to a large extent, te risk attitude of te banking institutions. According to Tsakalotos (1991, quoted in Gibson and Tsakalotos 1992, p. 61) credits granted by Greek banks were frequently made on te basis of personal contacts and social pressure leading, as a consequence, to inefficiency wit regards to risk management and subsequently to problems wit non-performing loans. By contrast, te canging economic environment witin wic te banks operated, led banks in Greece to adopt a different mode of operation wit regards to te ways tey andled risk. In order to acieve satisfactory levels of profitability and survive in te face of more intense competition, te banks were forced to improve teir risk management efficiency and to adopt sopisticated related tecnology. Taking into account te transformation of te banking sector, as analyzed above, it is logical to infer tat te determinants of NPLs sould ave canged over time. In te place of determinants related to public policy directions, market forces are expected to ave taken over as te major drivers of NPLs. Tus, it makes sense to restrict our investigation to te post-liberalization time period. Tis study focuses on te period from 2003 until today and uses a panel data set of te 9 largest banks in Greece wic, in total, account for almost 80% of te market sare. During tis period, te Greek banking system can safely be caracterized as a relatively mature financial sector were market forces govern its functioning. In addition tis period encompasses a part of te booming period (wic started since te mid-1990s) and te current financial crisis, originating from te US subprime mortgages market. Tus, in te time period examined various pases of te business cycle are represented. Te evolution of te aggregate NPLs ratio for all types of loans is depicted in Figure 1. A common feature for all tree types of loans is tat te NPL ratio exibited a downward trend from 2003 onwards wic was reversed abruptly after te outbreak of 8
te financial crisis (te trend reversal is evident in te last two quarters of 2008). It can also be observed tat te problem loans ratio in te business sector was noticeably lower (relative to teir average value) compared to te consumer and mortgage loans. [Insert Figure 1 about ere] In addition to te concerns raised by te current financial crisis for a furter NPL ratio deterioration, te steep credit expansion wic occurred during tis decade (see Figure 2) also poses te question weter te quality of loans granted during tis period was accurately evaluated by te banking system. Generally, te ig rates of credit growt during te 2000s can be attributed to rigtward sifts bot in te demand and te supply curves. On te one and, te deregulation of te financial system wic took place in te 1990s and te ensuing competition between banks for market sare, fuelled credit growt. On te demand side, te increase in debt ceilings, brougt about by bank competition, induced ouseolds to attempt to smoot teir consumption troug borrowing 9. In addition, te ig rates of growt tat prevailed in Greece since te mid- 1990s 10, motivated firms to undertake investments, leading to increased debt obligations for te business sector as well. [Insert Figure 2 about ere] Te fact tat te Greek financial sector as crystallized in its current form for little more tan a decade implies tat tere are many unresolved issues wit regards to te determinants of NPLs. More generally, tere is a deart of studies focusing on credit risk in te Greek economy. In te single related study, Kalfaoglou (2006) presents a stresstesting exercise conducted as a collaboration of te Bank of Greece and te IMF, aiming to quantify te resilience of Greek banks to internal and external socks related to credit and market risk. Te credit risk impact turned out to be te most significant risk component (as would be expected). 9 In a teoretical contribution, Antzoulatos (1994) argues (using a stocastic optimization framework) tat increases in te debt ceiling may lead to increases in optimal consumption. Debt ceiling is assumed to be exogenous in is model, so tat one can interpret it as a coice variable for bank policy. Antzoulatos links tis teoretical result wit te observed decrease in savings, presumably related to improved consumer access to credit (caused by financial deregulation), across a diverse set of countries. Furtermore, te proposed model implies tat improved access to credit primarily affects middle-income groups. 10 For a periodization of te growt pases for te Greek economy, see Boswort and Kollintzas (2001). 9
Te present study aims to fill tis gap by investigating te macroeconomic and bank-specific determinants of problem loans, per type of loan (business, consumer and mortgage loans) for te Greek banking system, taking into account te empirical information provided by te current financial crisis. 3. Determinant factors of NPLs 3.1 Macroeconomic factors Te relation between te macroeconomic environment and loan quality as been investigated in te literature linking te pase of te business cycle wit banking stability. In tis line of researc te ypotesis is formulated tat te expansion pase of te economy is caracterized by a relatively low number of NPLs, as bot consumers and firms face a sufficient stream of income and revenues to service teir debts. However as te booming period continues, credit is extended to lower-quality debtors and subsequently, wen te recession pase sets in, NPLs increase 11. Empirical studies tend to confirm te aforementioned link between te pase of te cycle and credit defaults. Quagliarello (2007) find tat te business cycle affects te NPL ratio for a large panel of Italian banks over te period 1985 to 2002. Furtermore, Cifter et al (2009), using neural network based wavelet decomposition, find a lagged impact of industrial production on te number of non-performing loans in te Turkis financial system over te period January 2001 to November 2007. Finally, Salas and Saurina (2002) estimate a significant negative contemporaneous effect of GDP growt on te NPL ratio and infer a quick transmission of macroeconomic developments to te ability of economic agents to service teir loans. It seems plausible to include oter macroeconomic variables, aside from GDP growt, suc as unemployment and interest rates as tese may provide additional information regarding te impact of macroeconomic conditions on ouseold and firms. 11 Te inability of lower-quality debtors (eiter ouseolds or firms) to service teir loans during a recession is also caused by te decrease in asset values wic serve as collateral and te subsequent contraction of credit as banks become more risk-averse (See e.g. Fiser 1933, Minsky 1986, Kiyotaki and Moore 1997, Geanakoplos 2009). 10
More specifically, an increase in te unemployment rate sould influence negatively te cas flow streams of ouseolds and increase te debt burden. Wit regards to firms, increases in unemployment may signal a decrease production as a consequence of a drop in effective demand. Tis may lead to a decrease in revenues and a fragile debt condition. Te interest rate affects te difficulty in servicing debt, in te case of floating rate loans. Tis implies tat te effect of te interest rate sould be positive, and as a result te increasing debt burden caused from rising interest rate payments sould lead to a iger number of NPLs. Te coice of GDP, unemployment and interest rate as te primary determinants of NPLs may also be justified from te teoretical literature of life-cycle consumption models. Lawrence (1995) examines suc a model and introduces explicitly te probability of default. Te model implies tat borrowers wit low incomes ave iger rates of default. Tis is explained by teir increased risk of facing unemployment and being unable to pay. Additionally, in equilibrium, banks carge iger interest rates to riskier clients. Rinaldi and Sancis-Arellano (2006) extend Lawrence s model by including te possibility tat agents can also borrow in order to invest in real or financial assets. After solving te optimization problem of an agent, tey derive te probability of default wic depends on current income, te unemployment rate (wic is linked to uncertainty regarding future income) and te lending rate 12. 3.2 Bank specific factors Te determinants of NPLs sould not be sougt exclusively in macroeconomic factors wic are viewed as exogenous forces influencing te banking industry. On te contrary, te distinctive features of te banking sector and te policy coices of eac particular bank wit respect to teir efforts for maximum efficiency and improvements in teir risk management are expected to exert a decisive influence on te evolution of NPLs. A strand in te literature as examined te connection between bank-specific factors and NPLs. 12 Te probability of default, in tis model, also depends on te amount of loan taken, te volume of investment and te time preference rate. 11
In teir seminal paper, Berger and DeYoung (1997) investigate te existence of causality among loan quality, cost efficiency and bank capital using a sample of U.S. commercial banks for te period 1985-1994. Tey codify and test four ypoteses concerning te flow of causality between tese variables: 1) Bad luck ypotesis: exogenous increases in nonperforming loans cause decreases in measured cost efficiency. Te underlying argument is tat a ig number of problem loans leads to extra operating costs associated wit dealing wit tem. 2) Bad management ypotesis: low cost efficiency is positively associated wit increases in future nonperforming loans. 13 Te proposed justification links bad management wit poor skills in credit scoring, appraisal of pledged collaterals and monitoring borrowers. 3) Skimping ypotesis: ig measured efficiency causes increasing numbers of nonperforming loans. According to tis view, tere exists a trade-off between allocating resources for underwriting and monitoring loans and measured cost efficiency. In oter words, banks wic devote less effort to ensure iger loan quality will seem to be more cost-efficient; owever, tere will be a burgeoning number of NPLs in te long run. 4) Moral azard ypotesis: low capitalization of banks leads to an increase in nonperforming loans. Te link is supposed to be found in te moral azard incentives on te part of banks managers wo increase te riskiness of teir loan portfolio wen teir banks are tinly capitalized.14 Berger and DeYoung find support bot for te bad management ypotesis and te bad luck ypotesis implying bidirectional causation between cost efficiency and NPLs (negative association). In addition, tey find evidence for te moral azard ypotesis. 13 i.e. tis ypotesis concerns te same variables as te bad luck ypotesis but implies te opposite temporal ordering (cost inefficiency causes NPLs rater tan te opposite). 14 See Berger and DeYoung (1997, pp. 852-854) for a more detailed formulation of tese ypoteses. 12
Podpiera and Weill (2008) examine empirically te relation between cost efficiency and non-performing loans in te context of te Czec banking industry for te period 1994 to 2005. Tey conclude tat tere is strong evidence in favor of te bad management ypotesis and propose tat regulatory autorities in emerging economies sould focus on managerial performance in order to enance te stability of te financial system (by reducing nonperforming loans). Salas and Saurina (2002) combine macroeconomic and microeconomic variables as explanatory regressors to explain NPLs in a study wic is concerned wit Spanis Commercial and Savings Banks (for te period 1985-1997). Tey estimate a statistically insignificant effect of lagged efficiency on problem loans (probably as a consequence of te counteraction of te bad management and skimping effects) and a negative influence of lagged solvency ratio to NPLs wic is consistent wit te moral azard ypotesis. In addition, tey find a size effect i.e. large banks seem to ave fewer NPLs. Tus te following ypotesis may also be formulated: 5) Size effect ypotesis: Te size of te bank is negatively related to nonperforming loans.15 Te link between lagged measures of performance and problem loans is ambiguous in its direction. One ypotesis is tat worse performance may be a proxy of lower quality of skills wit respect to lending activities (similar to te bad management ypotesis). Tis implies a negative relationsip between past earnings and problem loans. Te inverse direction of effect is also possible as in te model of Rajan (1994). Te argument is tat a bank may attempt to convince te market about te profitability of its lending by adopting liberal credit policies and tus inflating current earnings at te expense of future problem loans. A bank may also use loan loss provisions in order to boost its current earnings. 16 As a consequence, past earnings may be positively linked to future NPLs. 15 Salas and Saurina explain tis by noting tat size allows for more diversification opportunities (Salas and Saurina 2002, p. 219). 16 Amed et al (1999), owever, do not find evidence of earnings management via loan loss provisions for a sample of U.S. banks over 1986-1995. 13
6) Bad management II ypotesis: worse performance is positively associated wit increases in future nonperforming loans. Tis may be justified in a way analogous to te bad management ypotesis by regarding past performance as a proxy for te quality of management. Furtermore, we could conjecture credit growt as a cause of future NPLs, in accordance wit te business cycle literature: 7) Procyclical credit policy ypotesis: Banks adopt a liberal credit policy (defined by Rajan (1994) as a negative NPV extension of credit ) during te boom of te cycle, and a tigt policy in te contraction pase (defined in an inverse manner). Table 1 presents te bank specific variables used in te econometric analysis and teir mapping to specific ypoteses. [Insert Table 1 about ere] Finally, it is noteworty tat (according to our knowledge) tere is no comparative study for te determinants of NPLs between different types of loans. It is to be expected tat te weigt of te various determinants (bot macroeconomic and bank-specific) will vary across different types of loans. For example, macroeconomic developments suc as te rate of unemployment may ave different quantitative implications for te future number of business and consumer NPLs as te strategic beaviour and incentives of economic agents differ across economic units. Tis could be attributed to differences in te strictness of bankruptcy laws regarding firm defaults and ouseold defaults 17,18. 4. Preliminary econometric analysis As a starting point, we investigate te existence of a cointegrating relation between NPLs and macroeconomic variables. Subsequently, we estimate a fixed effects model of te first differences intended to capture sort-run movements in te NPL ratio. 17 For an international comparison of bankruptcy laws see Kolecek (2008). 18 Te regulatory framework e.g. te obligation of te banks to write off NPLs quickly or not is anoter factor tat affects te observed value of NPLs. 14
Te data set consists of a balanced panel of nine (9) Greek commercial banks spanning te period from 2003 q1 to 2009 q3 on a quarterly basis. Te analysis is conducted in a disaggregated manner by classifying te banks total loan portfolio into tree main categories i.e. mortgages, business and consumer loans. Eac category of problem loans is examined separately so as to detect possible similarities or differences in te beavior (i.e. te determinants) of eac type of portfolio. Te dependent variable is te NPL ratio wic is defined as te ratio of te NPLs to te value of total loans. Our analysis is built on te baseline model were te relation between NPLs and a set of fundamental macroeconomic indicators, specifically, te real GDP growt rate, te unemployment rate and te real lending rates (corresponding to eac type of loan) is investigated. Subsequently, te baseline model is furter extended by te inclusion of additional microeconomic variables (see section 5). However, as empirical evidence suggests tat te NPL ratio may follow a unit root process inting at a possible cointegrating relation wit macroeconomic variables (see e.g. Rinaldi and Sancis-Arellano (2006)), we perform a preliminary panel unit root and cointegration analysis. Te panel unit root results for te NPLs and te real lending rates are presented in Table 2. [Insert Table 2 about ere] Te null ypotesis of te presence of a unit root cannot be rejected for all tree categories of NPLs and across bot tests. Te presence of unit root can also not be rejected for te lending rates wit te exception of te Im, Pesaran and Sin (1997 (IPS) test for te business rates. Tese results confirm te empirical findings of Rinaldi and Sancis-Arellano (2006) wo report non-stationarity of NPLs for ouseold loans and real lending rates in a European cross country study. Table 3 summarizes te unit root results for GDP and te unemployment rate. [Insert Table 3 about ere] Te ypotesis of a unit root process is not rejected for bot macroeconomic variables, as expected. Te natural extension of te analysis is to test for a sustainable long-run equilibrium relationsip between NPLs and te macroeconomic indicators as sown in Rinaldi and Sancis-Arellano (2006). Noneteless, our data sample covers a 15
period, were te downward trend in te NPLs is abruptly interrupted by an upward sift approximately in te first alf of 2008 (see Graps 1,2 and 3) due to te ongoing crisis. Tis may signal an interruption of te common stocastic trend between NPLs and te macroeconomic factors. In oter areas of financial researc, suc as stock markets, it as been sown tat establised long-run relations in a pre-crisis period are (temporarily) interrupted during te crisis (e.g. see Jocum et.al (1999) and Yang et.al (2006)). In order to detect possible sifts in te long-run beavior of te variables, we proceed as follows. First, we test for cointegration utilizing te full data sample. For eac type of NPL, a precrisis period is defined by observing in te data, te date wen te jump in te NPLs occurred. Te new data set is used in order to perform panel cointegration test. Te Pedroni panel cointegration test (Pedroni, 1999) for te full sample is sown in Table 4. [Insert Table 4 about ere] For all panel cointegration tests and for eac type of loans, te null ypotesis of no cointegration is not rejected implying te absence of a cointegrating relation between te NPLs and te macroeconomic variables wen te full data sample is used. Table 5 presents te Pedroni panel cointegration results for te pre-crisis sample. [Insert Table 5 about ere] Te rejection of te null ypotesis for four out of seven statistics for almost all types of NPLs generates evidence in favor of a long-run equilibrium in te pre-crisis period. In addition, as te Panel and Group t-statistics are te most powerful amongst its counterparts for N<T and for relative small T (Pedroni, 2004), non-rejection of te cointegration ypotesis is a reasonable conclusion. Tese empirical findings are indicative of a long-run equilibrium between NPLs and macroeconomic factors wic was interrupted by te unfolding crisis. However, we would like to utilize te wole data sample in order to enance te consistency of our estimates. As a result, we coose a model in first differences rater tan a specification involving a cointegration relation. Te baseline model for te tree categories of loan portfolios is initially estimated using te Fixed Effects (FE) metod. Te regression results are presented in Table 6. [Insert Table 6 about ere] 16
From Table 6 we see tat te signs of all coefficients are as expected. Te real growt is negatively related to canges in te NPL ratio, wile te unemployment rate and te real lending rates ave a positive impact on te dependent variable. Noneteless, for te NPLs of mortgage loans, te first lag of te growt rate is te only statistical by significant variable. For te oter two categories of loans, a statistical by significant relation is establised wit te second lag in te growt rate and te first lag of unemployment and real lending rates for business and consumer loans, respectively. Across all tree categories of loans, te GDP growt seems to govern te relationsip between te NPLs and te macroeconomic factors. Anoter interesting conclusion is tat te NPLs of consumer loans are te most sensitive 19 towards a cange in bot growt rate and lending rates, wile te NPLs of business loans are te most sensitive towards a cange in te unemployment rate. Te NPLs of mortgage loans is te least sensitive to canges of macroeconomic factors. Tese results are indicative of te differences in te driving forces beind NPLs for various types of portfolio loans. However, as it is expected tat some degree of persistence may exist in te evolution of NPLs, te analysis is subsequently extended to a dynamic setting. In addition, te dynamic specification is suitable for testing te ypoteses formulated in Section 3 regarding te microdeterminants of NPLs. 5. Dynamic panel data analysis 5.1 Dynamic panel data estimators Following te recent literature in panel data studies (e.g. see Salas and Saurina (2002), Atanasoglou et al. (2009) and Merkl and Stolz (2009) on banking related studies, Calderon and Cong (2001), Ceng and Kwan (2000), Beck and Levine (2004), Santos-Paulino and Tirlwall (2004) and Carstensen and Toubal (2004) on macroeconomic studies), a dynamic approac is adopted in order to account for te time persistence in te NPLs structure. 19 Te sensitivity is measured by te long-run coefficients wic are computed as te sum of te individual parameters (see section 6). 17
Te main feature of a dynamic panel data specification is te inclusion of a lagged dependent variable in te set of regressors i.e.: y it it ( L) X it + ηi ε it = α y 1 + β +, α < 1, i = 1,..., N, t = 1,..., T (1) were te subscripts i and t denote te cross sectional and time dimension of te panel sample respectively, y is te first difference of te NPLs, β ( L) is te 1 k lag it polynomial vector, X is te k 1 vector of explanatory variables oter tan y, it i η i are te unobserved individual (bank specific) effects and ε it are te error terms. As te lagged dependent variable, y i is inerently correlated wit te bankspecific effects, η i, OLS estimation metods will produce biased and inconsistent parameters estimates. Equation (1) is consistently estimated utilizing te Generalized Metod of Moments (GMM) as proposed by Arellano and Bond (1991) and generalized by Arellano and Bover (1995) and Blundell and Bond (1998). Te GMM estimation of Arellano and Bond (1991) is based on te first difference transformation of equation (1) and te subsequent elimination of bank-specific effects: it it ( L) X it + ε it y = α y 1 + β (2) were is te first difference operator. In equation (2), te lagged dependent variable, y 1 is, by construction, correlated wit te error term,, imposing a bias in te εit estimation of te model. Noneteless, y i, wic is expected to be correlated wit y i and not correlated wit for t = 3,..., T, can be used as an instrument in te εit estimation of (2), given tat ε it are not serially correlated. Tis suggests tat lags of order two, and more, of te dependent variable satisfy te following moment conditions: E [ y ] = 0 for t = 3,..., T and s 2 it s ε it (3) A second source of bias stems from te possible endogeneity of te explanatory variables and te resultant correlation wit te error term. In te case of strictly exogenous variables, all past and future values of te explanatory variable are uncorrelated wit te error term, implying te following moment conditions: 18
E [ X ] = 0 t 3,..., T and for all s. it s it = ε (4) Te assumption of strict exogeneity is restrictive and invalid in te presence of reverse causality i.e. wen [ ] 0 E ε for t < s. For a set of weakly exogenous or X is it predetermined explanatory variables, only current and lagged values of X it are valid instruments and te following moment conditions can be used: E [ X ] = 0 t = 3,..., T and for s 2. it s ε it (5) Te ortogonality restrictions described in (3) (5) form te underpinnings of te one-step GMM estimation wic produces, under te assumption of independent and omoscedastic residuals (bot cross-sectionally and over time), consistent parameter estimates. Arellano and Bond (1991) propose anoter variant of te GMM estimator, namely te two-step estimator, wic utilizes te estimated residuals in order to construct a consistent variance covariance matrix of te moment conditions. Altoug te two-step estimator is asymptotically more efficient tan te one-step estimator and relaxes te assumption of omoscedasticity, te efficiency gains are not tat important even in te case of eteroscedastic errors (e.g. see Arellano and Bond (1991), Blundel and Bond (1998) and Blundell et al. (2000)). Tis result is furter supported by te empirical findings of Judson and Owen (1999), wo performed Monte Carlo experiments for a variety of cross sectional and time series dimensions and sowed tat te one-step estimator outperforms te two-step estimator. Moreover, te two-step estimator imposes a downward (upward) bias in standard errors (t-statistics) due to its dependence to estimated values (as it uses te estimated residuals from te one-step estimator) (Windmeijer, 2005), wic may lead to unreliable asymptotic statistical inference (Bond, 2002, Bond and Windmeijeir, 2002). Tis issue sould be taken into account, especially in te case of data samples wit relatively small cross section dimension (see Arellano and Bond, 1991 and Blundell and Bond, 1998). As noted above, te validity of te instruments used in te moment conditions as well as te assumption of serial independence of te residuals is crucial for te consistency of te GMM estimates. Ιn line wit te dynamic panel data literature, we test te overall validity of te instruments using te Sargan specification test proposed by 19
Arellano and Bond (1991), Arellano and Bover (1995) and Blundel and Bond (1998). Te Sargan test for over-identifying restrictions is based on te sample analog of te moment conditions used in te estimation process so as to determine te suitability of te instruments. Under te null ypotesis of valid moment conditions, te Sargan test statistic is asymptotically distributed as ci-square. Furtermore, te fundamental assumption tat te errors, ε it, are serially uncorrelated can be assessed by testing for te ypotesis tat te differenced errors, εit are not second order autocorrelated. Rejection of te null ypotesis of no second order autocorrelation of te differenced errors implies serial correlation for te level of te error term and tus inconsistency of te GMM estimates. 5.2 Econometric specification Equation (1) takes te following form in te baseline model: it it 1 + β j= j NPL = a NPL 2 1 1 GDP t j = 2 2 + j j UNt j + β i j RLR 1 2 = 1 3 it j β + η + ε i it (6) wit a < 1, i = 1,...,9 and t = 1,..., 27. In equation (6) te superscript denotes te type of loan (i.e. mortgages, business and consumer loans), NPL it is te first difference of te non-performing loans ratio, GDP t RLR it is te real GDP growt rate, UNt is te cange in te unemployment rate and is te first differences of te te real lending rates. For eac type of loan, te baseline model, i.e. equation (6), is estimated in its dynamic form. Ten, eac of te microeconomic indicators of Table 1 is added to te baseline model in order to examine its additive explanatory power. Moreover, from an econometric point, te limited number of cross-sectional units in te sample (tere are nine (9) banks) poses additional limitations on te number of instruments tat can be used in te estimation and subsequently te number of exogenous variables tat can be added to equation (6). Specifically, wen te number of instruments is greater tan or equal to 20
te number of cross-sectional units, ten bot te standard errors and te Sargan test are downwards biased and as a consequence te asymptotic inference may be misleading. To cope wit tis problem, we implement a restricted GMM procedure ( see Judson and Owen, (1999) 20, i.e. we use only a limited number of lagged regressors as instruments and, secondly, as already mentioned, we add just one bank-specific variable at a time reducing te need of extra instruments. Te number of instruments is cautiously determined so tat teir total number does not exceed te number of cross-sectional units in te sample. Te baseline model in (6) is extended to account for te extra microeconomic variable: it it 1 + β j= j NPL = a NPL 2 1 1 GDP t j 2 2 4 + j j UNt j + β i j RLRit j + β i j X = = 1 3 = 1 4 it j β + η + ε 1 2 (7) i it were X it denotes a bank-specific variable from tose presented in Table 6. Te set of tese bank-specific variables was cosen in order to enable te testing of te ypoteses formulated in Section 3 21. In bot equations (6) and (7), te macroeconomic variables are considered as strictly exogenous wile te bank-specific regressors are treated as weakly exogenous in te sense tat NPLs can reversely cause te microeconomic factors used in (7). Tus, te macroeconomic variables are instrumented following condition (4) and microeconomic variables following condition (5) were only current and lagged values of te regressors are valid instruments. In addition to te individual lags estimations, we also calculated te long-run coefficient for eac of te regressions, wic is defined as: LR m n β j 1 mj ( a) β = = 1 (8) 20 Judson and Owen (1999) sow tat te use of te restricted procedure does not materially worsen te performance of te GMM estimation. 21 Te Bad luck ypotesis could not be tested using te dynamic panel data framework represented by Eq. 7. We test te ypotesis tat NPLs do not Granger-cause inefficiency and found mixed results. Te p- values were found as follows: Business NPLs (0.124), Consumer NPLs (0.010) and mortgage NPLs (0.404). 21
were m = 1,...,4. Its variance is calculated from te following formula: n ( βmj ) n ( β ) j= 1 mj n 2 ( βmj ) n ( βmj ), ( 1 a) ) 2 ( ) ( ) ( ) ( )( ) ( ) Var Cov LR j= 1 j= 1 Var a Var β m = 2 + (9) 2 n 2 1 a 1 β 1 a a = 1 j= 1 mj j n n were Var( β ) = Var( β ) + 2 ( β, β ) 5.3 Estimation results j= mj j= mj Cov 1 1 j l One-step GMM estimation 22 results for all categories of NPLs are presented in Tables 7, 8 and 9. Eac table contains bot te baseline specification estimates and tose obtained wen an extra bank-specific variable is included as an explanatory variable. For eac model, te Sargan test of overidentifying restrictions and te m 2 test of second order serial correlation are reported. [Insert Table 7 about ere] [Insert Table 8 about ere] [Insert Table 9 about ere] It may be observed tat most estimated coefficients ave signs wic are compatible wit economic intuition and te teoretical arguments surveyed in Section 3. Te coefficient of te lagged dependent variable is negative and statistically significant for business and consumer loans. Te implication is tat te NPL ratio is likely to decrease wen it as increased in te previous quarter, due to te write-offs 23. On te oter and, tis coefficient is statistically insignificant for mortgages. Specifically, for tis category of loans, macrofundamentals and bank-specific variables seem to be te main drivers of te NPL ratio. Te NPL ratio is negatively affected by a slowdown in economic growt for all types of loans. Te effect of te GDP growt rate is found to be stronger for business mj ml. 22 Te Blundell-Bond system GMM estimator (Blundell and Bond 1998) as also been proposed in te literature. However using te Sargan - difference test we rejected its underlying assumptions. 23 Sorge and Virolainen (2006) report a negative coefficient for te lagged dependent variable in teir estimated equation of loan loss provisions for te Finnis banking system. Te economic interpretation for te negative coefficient in bot cases is similar. 22
loans. In tat case, all lags are significant at a 1% significance level. Tis result points to a strong dependence of te business sector s ability to repay its loans on te pase of te cycle. Specfically, an increase of one percentage point of GDP eads to a decrease of 0.28 in te te NPL ratio during te first quarter and a furter decrease of 0.43 during te second quarter. Tus, te ypotesis tat a recession pase as an adverse impact on NPLs is confirmed. Tis effect is exacerbated by te small average size of Greek firms wic results in tem being most vulnerable to adverse macroeconomic socks. Te NPL ratio for consumer loans is also a negative function of te GDP growt rate. Te GDP growt rate affects te NPL ratio of consumer loans wit a lag of two quarters. Finally, in te case of mortgage loans, te coefficient of te lagged GDP growt rate is also statistically significant, owever its quantitative impact is attenuated compared to te oter two types of loans. Lagged unemployment affects, in particular, business loans, wit one lag. Tus, it seems tat firms cut teir labor cost before tey face credit repayment problems. Additionally, unemployment wit one-period lag is a leading indicator of NPLs in te consumer loan portfolio. It may be inferred tat a rise in unemployment affects ouseolds ability to service teir debts wit a tree-mont time delay. Mortgages are again te least sensitive type of loans. Tis could be explained by te fact tat mortgage loans are mostly extended to public servants and ig-skilled workers of te private sector and consequently unemployment does not ave any noticeable effect on te corresponding NPL ratio (Mitrakos et al, 2005). Te coefficients on te lagged real lending rates are positive as expected. Te onequarter lagged coefficient is statistically significant in te case of consumer and business loans. Despite te fact tat, in te case of mortgage loans, te statistical significance of eac individual lag is low, tis could be attributed to te ig level of covariance between te individual coefficients (as te estimation of te long-run coefficients sows see below). It sould be noticed tat te great majority of bot consumer and business loans are floating rate loans, compared to mortgages, were tere is a significant portion of fixed rate loans. Tis is compatible wit te aforementioned finding tat te one quarter lag for bot consumer and business loans is statistical significant, implying a direct effect of interest rates on te NPL ratio, in contrast to mortgages, were only te long-run 23
coefficient turns out to be statistically significant. Consumer loans are not easily refinanced as banks tend to resort to tigt credit granting policies wit regards to consumer loans during recessions. In contrast, firms facing difficulties in servicing teir debt possess greater bargaining power to renegotiate interest payments on teir loans. For eac type of loans, te coefficients for te macroeconomic variables are rater stable across alternative models (wic include bank-specific explanatory variables), tus providing a robustness ceck of our estimated results. In order to assess in a more transparent manner te differential impact of macroeconomic and bank specific variables one sould look at te long-run coefficients (see Tables 10 and 11 respectively). For all macroeconomic variables, te estimated longrun coefficients estimates are statistically significant wit te expected sign. Focusing on specific macrofundamentals, te real GDP growt ave te strongest long-run effect on te NPL ratio of business loans, as does unemployment. On te oter and, lending rates impact noticeably, in te long- run, on te NPL ratio of te consumer loans portfolio. [Insert Table 10 about ere] [Insert Table 11 about ere] Te most striking implication of te estimated long-run coefficients is te relative insensitivity of mortgage NPLs, to macroeconomic conditions, compared to consumer and business NPLs. One possible explanation is tat a borrower faces greater incentives to avoid defaulting on a mortgage loan due to te pledged collateral. In addition, ome ownersip is igly valued in Greece, a feature tat may be considered as a social specificity. Te estimated coefficients for te bank-specific explanatory variables imply te existence of certain regularities as regards te relation between banks features and te quality of loans. Te empirical evidence for te ypoteses presented in Table 1 is summarized in Table 12. Performance indicators (suc as ROE and ROA) are found to be significant and negatively related to te NPLs for mortgages and consumer loans wile tey are not significant for business loans. In te case of mortgages and consumer loans, tis provides evidence in favor of te bad management II ypotesis. Te fact tat 24
empirical evidence in favor of te bad management II ypotesis is restriced to mortgages and consumer loans, but is absent for business loans, may signify tat te effect of management quality is mainly reflected on te efficiency of cross grating procedures to ouseolds wic are primarily based on te development of quantitative modeling tecniques wile te quality of case-by-case assignment procedures wic caracterize te granting of business loans does not differ substantially among banks. [Insert Table 12 about ere] Banks risk attitude, as reflected in te solvency and loans-to-deposit ratio, does not seem to ave explanatory power over NPLs for all types of loans. Tus, te moral azard ypotesis does not find any support in te Greek banking system. A possible explanation is tat te small size of te market for bank managers in Greece creates disincentives for reckless risk-taking and sort-termism for reputation reasons. In addition, due to te small number of banks, te regulatory autorities tend to ave an accurate onsite overview of te riskiness of eac individual bank s loan portfolio and tus tey can intervene accordingly to ensure financial stability. As a result, te potential of bank managers to generate causing ig levels of NPLs because of moral azard incentives is minimized. On te oter and, inefficiency index as a positive and statistically significant tus lending support to te bad management ypotesis. Moreover, its impact is quantitatively similar for NPLs of all types. Empirical support for te bad management ypotesis is also consistent wit te aforementioned finding tat lagged performance is negatively related to te problem loans troug te bad management II ypotesis. Tus, bot performance indicators and te inefficiency index may serve as proxies for te quality of management and bot ave explanatory power over te NPL ratio. Past credit growt does not go a long way to explain NPLs implying tat sort sigtedness on te part of managers was not present in te present sample and tat aggressive lending does not necessarily coincide wit reckless risk taking. Consequently, te procyclical credit growt ypotesis is rejected for te Greek banking system. Finally, market power indicators ave a significant impact only for business loans NPLs 25
(specifically te market sare). It can be deduced tat diversification opportunities are more ample in te business loan portfolio. 6. Concluding remarks and discussion In tis study we used dynamic panel data metods to examine te determinants of non-performing loans (NPLs) in te Greek financial sector. It was found tat macroeconomic variables, specifically te real GDP growt rate, te unemployment rate and te lending rates ave a strong effect on te level of NPLs. Furtermore, bankspecific variables suc as performance and efficiency indicators were found to possess additional explanatory power wen added to te baseline model, tus lending support to te bad management ypotesis linking tese indicators to te quality of management. Empirical results also indicate significant differences wit regards to te quantitative effects of te various NPLs determinants depending on te category of loans. Our findings ave several implications in terms of regulation and policy. Specifically, tere is evidence tat performance and inefficiency measures may serve as leading indicators of future problem loans. Tis suggests tat te regulatory autorities could use tese measures to detect banks wit potential NPLs increases. Moreover, regulators sould place greater empasis on risk management systems and procedures followed by banks in order to avert future financial instability. In addition, te econometric relations establised in te paper can be used for forecasting and stress testing purposes by bot regulators and banks. In a macro-stress testing exercise, alternative scenarios for te evolution of te macrovariables can be used in order to assess if NPLs are likely to exceed a tresold indicative of financial instability and to evaluate te adequacy of loan-loss provisions in te banking system. On te oter and, similar exercises could be performed on a bank-specific level in order to assess future problems tat may ensue in particular banks caracterized by relatively low performance and efficiency. Given tat te analysis as been conducted on a disaggregated basis, stress testing exercises may focus on different types of loan portfolios, enancing te reliability of te results. 26
Te study could be extended in various ways. In te first place, a vintage loan analysis could be used to pinpoint any differences in te quality of loans granted during te cycle. Suc a type of analysis would be directly linked to te ypotesis wic posits a cange in te risk attitude of banks between te pases of te cycle. Additionally, furter investigation of te crisis effects would be worty of study. It may be conjectured tat te financial crisis represents a structural break affecting te interrelations between non-performing loans and teir determinant factors, but additional data is required to test ypotesis. 27
REFERENCES Amed, A. S., Takeda, C. and Tomas, S., 1999. Bank Loan Loss Provisions: A Reexamination of Capital Management, Earnings Management and Signaling Effects. Journal of Accounting and Economics 28, 1-25. Antzoulatos, A. A., 1994, Credit Rationing and Rational Beavior. Journal of Money, Credit and Banking 26 (2), 182-202. Arellano, M. and Bond S., 1991. Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations. Review of Economic Studies 58, 277 297. Arellano, M. and Bover, O., 1995. Anoter Look at te Instrumental Variable Estimation of Error-Component Models. Journal of Econometrics 68, 29 51. Atanasoglou, P. P., Brissimis, S. N. and Delis M. D., 2008. Bank-Specific, Industry- Specific and Macroeconomic Determinants of Bank Profitability. Journal of International Financial Markets, Institutions and Money 18 (2), 121 136. Beck, T. and Levin R., 2004. Stock Markets, Banks, and Growt: Panel Evidence. Journal of Banking and Finance 28 (3), 423 442. Berger, A. N. and DeYoung R., 1997. Problem Loans and Cost Efficiency in Commercial Banks. Journal of Banking and Finance 21, 849-870 Blundell, R., Bond, S., 1998. Initial Conditions and Moment Conditions in Dynamic Panel Data Models. Journal of Econometrics 87, 115 143. Blundell, R., Bond, S. and Windmeijer F., 2000. Estimation in Dynamic Panel Data Models: Improving on Te Performance of te Standard GMM Estimator. in B. Baltagi (ed). Advances in Econometrics, Volume 15: Non-stationary Panels, Panel Cointegration and Dynamic Panels, JAI Elsevier Science. Bond, S., 2002. Dynamic Panel Data Models: A Guide to Micro Data Metods and Practice, Portuguese Economic Review 1 (2). Bond, S., Windmeijer F., 2002. Finite Sample Inference for GMM Estimators in linear Panel Data Models: A Comparison of Alternative Tests. mimeo, Institute for fiscal studies, London. Boswort, B., Kollintzas, T., 2001. Economic Growt in Greece: Past Performance and Future Prospects in Bryant R. C., Garganas N. C. and Tavlas G. S. (Eds.), Greece s Economic Performance and Prospects, Bank of Greece and te Brookings Institution Calderon, C., Cong, A., 2001. External Sector and Income Inequality in Interdependent Economies Using a Dynamic Panel Data Approac. Economics Letters 71, 225 231. Carstensen, K., Toubal, F., 2004. Foreign Direct Investment in Central and Eastern European countries: a Dynamic Panel Analysis. Journal of Comparative Economics 32, 3 22. 28
Ceng, L.K., Kwan, Y.K., 2000. Wat are te Determinants of te Location of Foreign Direct Investment? Te Cinese experience. Journal of International Economics 51, 379 400. Cristopoulos, D. K., Tsionas, E., 2001. Banking Economic Efficiency in te Deregulation Period: Results from Heteroscedastic Stocastic Frontier Models. Te Mancester Scool 69 (6), 656-676. Cifter, A., Yilmazer, S., Cifter E., 2009. Analysis of Sectoral Credit Default Cycle Dependency wit Wavelet Networks: Evidence from Turkey. Economic Modelling 26, 1382-1388 Eicengreen B. and Gibson H. D., 2001. Greek Banking at te Dawn of te New Millenium in Bryant R. C., Garganas N. C. and Tavlas G. S. (Eds.), Greece s Economic Performance and Prospects, Bank of Greece and te Brookings Institution Fiser, I., 1933. Te Debt Deflation Teory of Great Depressions. Econometrica 1, 337-57. Geanakoplos, J. 2009. Te Leverage Cycle. Cowles Foundation Discussion Paper No. 1715. Gibson, H., Tsakalotos, E., 1992. Economic Integration and Financial Liberalization: Prospects for Soutern Europe. St. Antony/MacMillan Press. Im, K., Pesaran H., Sin Y., 1997. Testing for Unit Roots in Heterogeneous Panels. Discussion Paper, University of Cambridge. Jocum, C., Kircgassner, G. and Platek, M., 1999. A Long-Run Relationsip Between Eastern Stock Markets? Cointegration and Te 1997 1998 Crisis In Emerging Markets. Weltwirtscaftlices Arciv 135, 455 479. Judson, R. A. and Owen, L. A., 1999. Estimating Dynamic Panel Data Models: A Guide for Macroeconomists. Economics Letters 65, 9 15. Kalfaoglou, F. 2006. Stress Testing of te Greek Banking System. Bank of Greece: Economic Bulletin 27. Kamberoglou, N., Liapis, E., Simigiannis, G. and Tzamourani, P., 2004. Cost Efficiency in Greek Banking. Working Paper Bank of Greece 9. Kiyotaki, N., Moore, J., 1997. Credit Cycles. Journal of Political Economy 105 (2) 211-248. Kolecek, L. 2008. Bankruptcy Laws and Debt Renegotiation. Journal of Financial Stability 4, 40-61. Lawrence, E. C., 1995. Consumer Default and te Life Cycle Model. Journal of Money Credit and Banking. 27, 939-954. Merkl, C., Stolz S., 2009. Banks Regulatory Buffers, Liquidity Networks and Monetary Policy Transmission. Applied Economics 41, 2013 2024. Minsky, H. 1986. Stabilizing an Unstable Economy, McGraw-Hill. 29
Mitrakos, T., Simigiannis, G. and Tzamourani P. 2005. Indebtedness of Greek Houseolds: Evidence from a Survey. Economic Bulletin, Bank of Greece. Organization for Economic Cooperation and Development (OECD), Economic Surveys: Greece Paris, 1986 Panopoulou, M., 2005. Tecnological Cange and Corporate Strategy in te Greek Banking Industry, Atens: KEPE Pedroni, P., 1999. Critical Values for Cointegration Tests in Heterogeneous Panels wit Multiple Regressors. Oxford Bulletin of Economics and Statistics 61, 653 70. Pedroni, P., 2004. Panel Cointegration, Asymptotic and Finite Sample Properties of Pooled Time Series Tests wit Application to PPP Hypotesis. Econometric Teory 20, 597 625. Podpiera, J., Weill L., 2008. Bad Luck or Bad Management? Emerging Banking Market Experience. Journal of Financial Stability 4, 135-148. Quagliarello M. 2007. Banks Riskiness Over te Business Cycle: a Panel Analysis on Italian Intermediaries. Applied Financial Economics 17, 119-138. Rajan R., 1994. Wy Bank Credit Policies Fluctuate: A Teory and Some Evidence. Te Quarterly Journal of Economics 109, 399-441 Rinaldi L. and Sancis-Arellano A., 2006. Houseold Debt Sustainability: Wat explains Houseold Non-performing Loans? An Empirical Analysis. ECB Working Paper Salas, V. and Saurina, J., 2002. Credit Risk in Two Institutional Regimes: Spanis Commercial and Savings Banks. Journal of Financial Services Researc 22 (3), 203-224. Santos-Paulino, A. and Tirlwall A.P. 2004. Te Impact of Trade Liberalization on Exports, Imports and Balance of Payments of Developing Countries. Te Economic Journal 114. Sorge, M. and Virolainen K., 2006. A Comparative Analysis of Macro Stress-Testing Metodologies wit Application to Finland. Journal of Financial Stability 2, 113-151. Voridis, H., 1993. Ceilings on Interest Rates and Investment: Te Example of Greece. Review of Economics and Statistics 75 (2), 276-283. Walen, J. M., 1994. Te Nature of Information in Commercial Bank Loan Loss Disclosure. Te Accounting Review 69 (3), 455-478. Windmeijer, F., 2005. A Finite Sample Correction for te Variance of Linear Efficient Two-Step GMM Estimators. Journal of Econometrics 126 (1), 25 51. Yang, J., Hsiao C., Q. Li and Z. Wang 2006. Te Emerging Market Crisis and Stock Market Linkages: Furter Evidence. Journal of Applied Econometrics 21, 727 744. 30
Appendix Table 1. Definition of bank specific variables Variable Definition Hypotesis tested Return on Assets Return on Equity Solvency Ratio Loans to Deposit Ratio Inefficiency Credit growt Market power Size Profitsit ROA = Bad management II it Total Assetsit (-) Profitsit ROE = Bad management II it Total Equityit (-) Owned Capitalit SOLR = Moral azard it Total Assetsit (-) Loansit LtD = Moral azard it Deposit it (+) Operating Expensesit Bad Management INEF it = Operating Income (+) it Skimping (-) t Loansit Loansi GLOANSit = Procyclical credit policy Loans (+) i Loansit MPOW = Size it 9 Loans (-) i = 1 it SIZE it = Total Assets 9 i = 1 it Total Assets it Size (-) Note: All ratios are expressed in percentage points. Te expected coefficient signs are sown in parentesis. 31
Table 2. Panel unit root tests Mortgages Business Consumer LLC IPS LLC IPS LLC IPS NPL [0.948] [0.964] [0.973] [0.946] [0.911] [0.967] NPL [0.000] [0.000] [0.008] [0.000] [0.003] [0.000] RLR [0.677] [0.179] [0.153] [0.016] [0.516] [0.433] RLR [0.000] [0.000] [0.001] [0.001] [0.001] [0.000] Notes: NPL and RLR denote te non-performing loans ratio and real lending rates respectively, wile te operator is te first difference operator. LLC stands for te Levin, Lin and Cu (2002) test were te null ypotesis tat eac individual time series is a unit root is tested against te alternative tat all of tem are stationary. IPS stands for te Im, Pesaran and Sin (1997) test were te assumption of omogeneity is relaxed under te alternative ypotesis tat some of te individual time series are stationary. Te p-values of te corresponding t-statistics are sown in brackets. Table 3. Unit root tests ADF PP GDP [0.984] [0.994] GDP [0.013] [0.012] UN [0.583] [0.463] UN [0.002] [0.001] Notes: GDP and UN denote te gross domestic product and te unemployment rate respectively wile te operator is te first difference operator. ADF stands for te Augmented Dickey Fuller and PP stands for te Pillips Perron test. For bot tests te null ypotesis of unit root is tested against te alternative of stationarity. Te p-values of te corresponding t-statistics are sown in brackets. Table 4. Pedroni panel cointegration test for te full sample Mortgages Business Consumer Panel v-statistic [0.497] [0.906] [0.671] Panel ρ-statistic [0.869] [0.974] [0.797] Panel t-statistic (PP) [0.174] [0.902] [0.127] Panel t-statistic (ADF) [0.769] [0.810] [0.743] Group ρ-statistic [0.955] [0.998] [0.967] Group t-statistic (PP) [0.147] [0.975] [0.154] Group t-statistic (ADF) [0.685] [0.949] [0.874] Notes: Te Pedroni test is an Enlge-Granger type test were te residuals are tested for te presence of unit root. Te null ypotesis is tat of no cointegration and te decision is based on seven statistics. Te main difference between te panel and group statistics is tat te latter allows for potential eterogeneity in te individual units troug its less restrictive alternative ypotesis. Te p-values for te corresponding statistics are sown in brackets. 32
Table 5. Pedroni panel cointegration test for te pre-crisis period Mortgages (Period: 2003 q1 2008 q2) Business (Period: 2003 q1 2008 q4) Panel v-statistic [0.830] [0.507] [0.856] Panel ρ-statistic [0.297] [0.892] [0.409] Panel t-statistic (PP) [0.001] [0.000] [0.004] Panel t-statistic (ADF) [0.000] [0.000] [0.000] Consumer (Period: 2003 q1 2008 q3) Group ρ-statistic [0.772] [0.951] [0.836] Group t-statistic (PP) [0.000] [0.000] [0.010] Group t-statistic (ADF) [0.000] [0.000] [0.138] Notes: Te Pedroni test is an Enlge-Granger type test were te residuals are tested for te presence of unit root. Te null ypotesis is tat of no cointegration and te decision is based on seven statistics. Te main difference between te panel and group statistics is tat te latter allows for potential eterogeneity in te individual units troug its less restrictive alternative ypotesis. Te p-values for te corresponding statistics are sown in brackets. Table 6. Fixed Effects (FE) regressions Mortgages Business Consumer 0.570*** (5.252) 0.392*** (3.164) 0.788*** (4.312) GDP GDP -0.336*** (-3.682) -0.077 (-0.871) -0.153 (-1.475) -0.305*** (-3.051) -0.205 (-1.408) -0.511*** (-3.946) UN UN 0.099 (1.245) 0.047 (0.572) 0.181** (1.917) 0.095 (0.980) 0.108 (1.132) 0.002 (0.023) RLR 1 RLR 2 0.128 (1.377) 0.078 (0.799) 0.122 (1.124) 0.044 (0.377) 0.345*** (2.985) 0.004 (0.032) Wald test (p-value) [0.000] [0.000] [0.000] Adjusted R-square 0.185 0.427 0.464 Notes: GDP is real growt rate, UN is te deseasonalized unemployment rate and RLR is te differenced real lending rate. *,** and *** indicate significance at a 10%, 5% and 1% significance level. Te t-statistics are depicted in parentesis under te parameter estimations. 33
Table 7. GMM estimation results for mortgages Baseline Model 1 Model 2 Model 3 Constant 0.031** 0.038** Constant 0.033** 0.032** (2.342) (2.452) (2.485) (2.387) NPL -0.003 0.006 0.000-0.019 1 NPL (-0.068) 1 NPL (0.113) 1 NPL (0.006) 1 (-0.392) GDP -0.238** (-2.48) GDP -0.041 (-0.670) GDP -0.170 (-1.527) GDP -0.008 (-0.113) GDP -0.198* (-1.748) GDP -0.030 (-0.452) GDP -0.219** (-2.060) GDP -0.027 (-0.346) UN 0.135* (1.723) UN -0.000 (-0.006) UN 0.126 (1.416) UN -0.059 (-0.852) UN 0.135 (1.618) UN -0.036 (-0.522) UN 0.124 (1.545) UN 0.011 (0.175) 0.096 (1.582) 0.070 (0.918) 0.067 (1.05) 0.086 (1.263) 0.086 (1.412) 0.092 (1.201) 0.069 (1.08) 0.050 (0.596) ROA 0.077 1 (0.990) ROA -0.253*** 2 (-3.973) ROA -0.054 3 (-1.343) ROA -0.150* 4 (-1.881) ROE 0.005 1 (1.282) ROE -0.009** 2 (-2.103) ROE -0.000 3 (-0.024) ROE -0.008** 4 (-2.281) SOLR -0.109** 1 (-2.04) SOLR -0.025 2 (-0.397) SOLR 0.071 3 (1.60) SOLR -0.029 4 (-0.557) Sargan test m 2 119.0 [0.820] -1.795 [0.073] 131.6 [0.871] -1.844 [0.065] 135.8 [0.807] -1.881 [0.060] 128.4 [0.909] -2.122 [0.034] Notes: ***,** and * denote significance at 1%, 5% and 10% respectively. t-statistics are reported in parentesis. Te p-values for te Sargan and te m 2 test are reported in brackets. 34
Table 7. (continued) GMM estimation results for mortgages NPL 1 Model 4 Model 5 Model 6 Model 7 Model 8 0.045* (1.885) -0.031 (-0.709) GDP -0.236** (-2.493) GDP -0.019 (-0.321) UN 0.128* (1.789) UN 0.008 (0.184) 0.130** (2.491) 0.043 (0.553) LtD -0.016 1 (-1.232) LtD 0.028 2 (1.365) LtD -0.002 3 (-0.183) LtD -0.017** 4 (-2.338) 131.8 Sargan test m 2 [0.868] -2.039 [0.041] NPL 1 0.036** (2.301) -0.011 (-0.220) GDP -0.193* (-1.760) GDP -0.014 (-0.199) UN 0.127 (1.457) UN -0.031 (-0.444) 0.074 (1.178) 0.086 (1.276) INEF 0.000 1 (0.036) INEF 0.002 i (0.436) INEF 0.001 3 (1.064) INEF 0.005** it 4 (2.211) 134.5 [0.829] -2.006 [0.045] NPL 1 0.030** (2.101) -0.000 (-0.015) GDP -0.234** (-2.382) GDP -0.033 (-0.587) UN 0.119 (1.569) UN -0.002 (-0.033) GLOANS 1 GLOANS 2 GLOANS it 3 GLOANS it 4 0.092 (1.285) 0.084 (1.078) -0.003 (-0.616) -0.000 (-0.035) -0.009 (-1.332) -0.001 (-0.199) 131.1 [0.878] -1.849 [0.064] NPL 1 0.031** (2.281) -0.013 (-0.277) GDP -0.230** (-2.412) GDP -0.054 (-0.874) UN 0.131* (1.753) UN 0.004 (0.0737) 0.106 (1.507) 0.074 (0.988) SIZE 0.005 it (0.056) 132.0 [0.900] -1.976 [0.048] NPL 1 0.031** (2.210) -0.008 (-0.187) GDP -0.229** (-2.422) GDP -0.047 (-0.740) UN 0.130* (1.731) UN 0.003 (0.057) MPOW it 0.109* (1.701) 0.079 (0.969) 0.018 (0.148) 130.1 [0.920] -1.986 [0.047] Notes: ***,** and * denote significance at 1%, 5% and 10% respectively. t-statistics are reported in parentesis. Te p-values for te Sargan and te m 2 test are reported in brackets. 35
Table 8. GMM estimation results for business loans NPL 1 Baseline Model 1 Model 2 Model 3-0.037** (-1.983) -0.102** (-2.081) NPL 1-0.038** (-1.973) -0.087* (-1.824) NPL 1-0.042** (-2.033) -0.091** (-2.074) NPL 1-0.038* (-1.768) -0.084 (-1.604) GDP -0.280*** (-4.562) GDP -0.436*** (-2.685) GDP -0.245*** (-4.743) GDP -0.418*** (-2.502) GDP -0.256*** (-4.332) GDP -0.428*** (-2.513) GDP -0.254*** (-5.543) GDP -0.446*** (-2.373) UN 0.156** (2.036) UN 0.107 1.084 UN 0.122* (1.698) UN 0.067 (0.683) UN 0.121* (1.645) UN 0.075 (0.791) UN 0.137* (1.784) UN 0.093 (1.313) 0.175* (1.803) 0.043 0.505 0.137 (1.504) 0.048 (0.630) 0.156* (1.734) 0.045 (0.589) 0.157* (1.684) 0.066 (0.887) ROA -0.193*** 1 (-2.501) ROA 0.076 2 (0.813) ROA -0.221* 3 (-1.818) ROA 0.125 4 (1.245) ROE -0.009*** 1 (-3.724) ROE 0.006* 2 (1.688) ROE -0.011 3 (-1.483) ROE 0.007 4 (1.232) SOLR -0.012 1 (-0.156) SOLR -0.119 2 (-0.902) SOLR 0.214*** 3 (2.741) SOLR -0.033 4 (-0.680) Sargan test m 2 181.7 [0.388] -0.462 [0.644] 184.2 [0.682] 0.225 [0.821] 184.9 [0.669] 0.2336 [0.815] 182.8 [0.707] -0.331 [0.741] Notes: ***,** and * denote significance at 1%, 5% and 10% respectively. t-statistics are reported in parentesis. Te p-values for te Sargan and te m 2 test are reported in brackets. 36
Table 8. (continued) GMM estimation results for business loans NPL 1 Model 4 Model 5 Model 6 Model 7 Model 8-0.066** -0.036* -0.039* -0.040* -0.040** (-1.984) (-1.873) (1.785) (-1.893) (-2.048) -0.138*** -0.106* -0.077-0.109** -0.119** (-2.603) NPL 1 (-1.877) NPL 1-1.094 NPL 1 (-2.162) NPL 1 (-2.569) GDP GDP -0.328*** (-4.643) GDP -0.461*** (-2.762) GDP -0.249*** (-3.968) GDP -0.405*** (-2.432) GDP -0.284*** (4.823) GDP -0.423*** (2.674) GDP -0.282*** (-4.628) GDP -0.438*** (-2.714) GDP -0.288*** (-4.633) -0.434*** (-2.703) UN UN 0.145** (1.984) UN 0.095 (1.074) UN 0.119* (1.668) UN 0.083 (0.809) UN 0.152* (1.864) UN 0.089 (0.998) UN 0.157** (2.151) UN 0.103 (1.067) UN 0.163** (2.133) 0.107 (1.098) 0.177*** (2.521) 0.031 (0.330) 0.145 (1.598) 0.023 (0.280) 0.158* (1.817) 0.045 (0.487) 0.175* (1.914) 0.043 (0.522) 0.142* (1.757) 0.047 (0.586) LtD 1 LtD 2 LtD 3 LtD 4 Sargan test m 2-0.022** INEF (-1.971) 1-0.004 INEF (-0.208) i 0.029*** INEF (2.725) 3 0.009 INEF (0.523) it 4 180.2 [0.753] -1.021 [0.307] 0.006*** (2.758) -0.0008 (-0.335) 0.003 (0.974) 0.002 (0.787) 185.1 [0.664] -0.215 [0.829] GLOANS 1-0.010 GLOANSit 2 (-0.674) 0.006 GLOANS it 3 (0.865) 0.004 GLOANS it 4 (0.283) 182.7 [0.709] -0.778 [0.436] 0.002 SIZE (0.179) it 0.096 1.432 187.9 [0.667] -0.562 [0.574] MPOW it 0.112** (2.451) 186.9 [0.686] -0.564 [0.573] Notes: ***,** and * denote significance at 1%, 5% and 10% respectively. t-statistics are reported in parentesis. Te p-values for te Sargan and te m 2 test are reported in brackets. 37
Table 9. GMM estimation results for consumer loans NPL 1 Baseline Model 1 Model 2 Model 3 0.034*** (3.232) -0.175*** (-3.103) NPL 1 0.044*** (3.153) -0.191*** (-3.235) NPL 1 0.033*** (2.624) -0.185*** (-3.209) NPL 1 0.030*** (2.948) -0.198*** (-3.489) GDP -0.150 (-1.433) GDP -0.398*** (-3.204) GDP -0.110* (-1.648) GDP -0.343*** (-2.635) GDP -0.132* (-1.905) GDP -0.365*** (-2.928) GDP -0.203** (-2.224) GDP -0.436*** (-3.835) UN 0.160** (1.893) UN 0.053 (0.525) UN 0.143 (0.909) UN -0.012 (-0.143) UN 0.141 (0.956) UN -0.005 (-0.066) UN 0.176 (1.216) UN 0.057 (0.580) 0.438*** (4.438) 0.081 (1.081) 0.371*** (3.733) 0.043 (0.818) 0.416*** (4.314) 0.072 (1.597) 0.400*** (5.114) 0.019 (0.226) ROA -0.217 1 (-1.108) ROA -0.150 2 (-1.224) ROA -0.059 3 (-0.281) ROA -0.226** 4 (-1.908) ROE -0.002 1 (-0.439) ROE -0.011 2 (-1.615) ROE 0.001 3 (0.149) ROE -0.011 4 (-1.448) SOLR -0.087 1 (-1.150) SOLR -0.058 2 (-0.949) SOLR 0.157 3 (1.264) SOLR -0.056 4 (-0.425) Sargan test m 2 137.1 [0.363] -1.027 [0.304] 154.8 [0.355] -0.844 [0.399] 151.2 [0.435] -0.448 [0.654] 156.5 [0.321] -1.641 [0.101] Notes: ***,** and * denote significance at 1%, 5% and 10% respectively. t-statistics are reported in parentesis. Te p-values for te Sargan and te m 2 test are reported in brackets. 38
Table 9. (continued) GMM estimation results for consumer loans NPL 1 Model 4 Model 5 Model 6 Model 7 Model 8 0.014 0.037*** 0.045*** 0.033* 0.033*** (0.716) 2.784 (3.374) (2.043) (3.204) -0.170*** -0.187*** -0.208*** -0.185*** -0.191*** (-3.060) NPL 1 (-3.265) NPL 1 (-3.765) NPL 1 (-3.735) NPL 1 (-3.779) GDP GDP -0.190** (-1.963) GDP -0.467*** (-3.604) GDP -0.152** (-2.327) GDP -0.393*** (-3.179) GDP -0.162 (-1.564) GDP -0.397*** (-3.272) GDP -0.186** (-2.414) GDP -0.433*** (-3.360) GDP -0.186** (-2.299) -0.432*** (-3.708) UN UN 0.204 (1.459) UN 0.053 (0.595) UN 0.152 (0.957) UN -0.005 (-0.062) UN 0.183 (1.309) UN 0.053 (0.529) UN 0.186 (1.401) UN 0.052 (0.563) UN 0.191 1.443 0.052 0.568 0.415*** (4.894) 0.005 (0.117) 0.376*** (3.724) 0.019 (0.406) 0.399*** (5.564) 0.075 (0.644) 0.395*** (4.764) 0.042 (0.556) 0.405*** 4.757 0.034 0.436 LtD 1 LtD 2 LtD 3 LtD 4 Sargan test m 2 0.004 INEF (0.269) 1 0.005 INEF (0.257) i -0.025* INEF (-1.879) 3 0.025 INEF (1.492) it 4 156.4 [0.322] -1.541 [0.123] 0.003 0.827 0.004 (0.952) 0.0007 (0.155) 0.006* (1.844) 156.0 [0.330] -0.860 [0.390] GLOANS 1 0.005 GLOANSit 2 (0.279) 0.023* GLOANS it 3 (1.765) 0.018 GLOANS it 4 (0.961) 152.0 [0.416] -0.778 [0.214] -0.015 SIZE (-1.015) it -0.167 (-1.312) 159.1 [0.330] -1.255 [0.210] MPOW it Notes: ***,** and * denote significance at 1%, 5% and 10% respectively. t-statistics are reported in parentesis. Te p-values for te Sargan and te m 2 test are reported in brackets. 0.016 0.251 164.5 [0.230] -1.268 [0.205] 39
Table 10. Long-run coefficients for te macroeconomic variables GDP -0.278** (-2.501) UN 0.134*** Mortgages Business Consumer (3.065) RLR 0.166** (2.031) -0.650*** (-3.534) 0.239** (2.208) 0.199** (1.972) -0.466*** (-2.850) 0.181* (1.893) 0.442** (6.331) Notes: ***,** and * denote significance at 1%, 5% and 10% respectively. t-statistics for te long run regression coefficients are reported in parentesis. Table 11. Long run coefficients for te bank specific variables ROA -0.383*** (-3.044) ROE -0.012** (-2.491) SOLR -0.092 (-0.994) LtD -0.008 (-1.036) INEF 0.009*** (3.061) GLOANS -0.014 (-0.727) Mortgages Business Consumer -0.195 (-1.172) -0.006 (-0.801) 0.045 (1.418) 0.011 (1.333) 0.010** (2.323) 0.002 (0.194) -0.549* (-1.897) -0.020*** (-3.759) -0.037 (-0.375) 0.008 (0.904) 0.012* (1.770) 0.026 (1.328) Notes: ***,** and * denote significance at 1%, 5% and 10% respectively. t-statistics for te long run regression coefficients are reported in parentesis. Table 12. Empirical evidence for tested ypoteses Hypotesis Tested Mortgages Business Consumer Bad luck ypotesis No No Yes Bad management ypotesis Yes Yes Yes Skimping ypotesis No No No Moral azard ypotesis No No No Size ypotesis No No No Bad management ypotesis Yes II No Yes Procyclical Credit Policy ypotesis No No No Note: Te empirical evidence is based on te sign and te statistical significance of te long run coefficients 40
Figure 1. Aggregate NPL ratio of loan portfolios (demeaned series) Percentage NPL ratio (Business loans) 6 NPL ratio (Mortgage loans) 5 NPL ratio (Consumer loans) 4 3 2 1 0-1 2003Q1-2 2003Q3 2004Q1 2004Q3 2005Q1 2005Q3 2006Q1 2006Q3 2007Q1 2007Q3 2008Q1 2008Q3 2009Q1 2009Q3-3 Source: Bank of Greece Figure 2. Credit expansion by type of loan Loans by type (Total loans in 2003Q1=1) 3 2,5 Consumer loans Mortgages Business loans 2 1,5 1 0,5 0 2003Q1 2003Q3 2004Q1 2004Q3 2005Q1 2005Q3 2006Q1 2006Q3 2007Q1 2007Q3 2008Q1 2008Q3 2009Q1 2009Q3 Source: Bank of Greece 41
42
BANK OF GREECE WORKING PAPERS 90. Tavlas, G., H. Dellas and A. Stockman, Te Classification and Performance of Alternative Excange-Rate Systems, September 2008. 91. Milionis, A. E. and E. Papanagiotou, A Note on te Use of Moving Average Trading Rules to Test for Weak Form Efficiency in Capital Markets, October 2008. 92. Atanasoglou, P.P. E. A. Georgiou and C. C. Staikouras, Assessing Output and Productivity Growt in te Banking Industry, November 2008. 93. Brissimis, S. N. and M. D. Delis, Bank-Level Estimates of Market Power, January 2009. 94. Members of te SEEMHN Data Collection Task Force wit a Foreword by Micael Bordo and an introduction by Mattias Morys, Monetary Time Series of Souteastern Europe from 1870s to 1914, February 2009. 95. Cronis, P., Modeling Distortionary Taxation, Marc 2009. 96. Hondroyiannis, G., Fertility Determinants and Economic Uncertainty: An Assessment using European Panel Data, April 2009 97. Papageorgiou, D., Macroeconomic Implications of Alternative Tax Regimes: Te Case of Greece, May 2009. 98. Zombanakis, G. A., C. Stylianou and A. S. Andreou, Te Greek Current Account Deficit: Is It Sustainable After All?, June 2009. 99. Sideris, D., Optimum Currency Areas, Structural Canges and te Endogeneity of te OCA Criteria: Evidence from Six New EU Member States, July 2009. 100. Asimakopoulos, I. and P. Atanasoglou, Revisiting te Merger and Acquisition performance of European Banks, August 2009. 101. Brissimis, N. S. and D. M. Delis, Bank Heterogeneity and Monetary Policy Transmission, August 2009. 102. Dellas, H. and G. S. Tavlas, An Optimum-Currency-Area Odyssey, September 2009. 103. Georgoutsos, A. D. and P. M. Migiakis, Bencmark Bonds Interactions under Regime Sifts, September 2009. 104. Tagkalakis, A., Fiscal Adjustments and Asset Price Movements, October 2009. 105. Lazaretou, S., Money Supply and Greek Historical Monetary Statistics: Definition, Construction, Sources and Data, November 2009. 106. Tagkalakis, A., Te Effect of Asset Price Volatility on Fiscal Policy Outcomes, November 2009. 107. Milionis, E. A and E. Papanagiotou, An Alternative Metodological Approac to Assess te Predictive Performance of te Moving Average Trading Rule in 43
Financial Markets: Application to te London Stock Excange, December 2009. 108. Babecký, J., P. D. Caju, T. Kosma, M. Lawless, J. Messina and T. Rõõm, Te Margins of Labour Cost Adjustment: Survey Evidence from European Firms, December 2009. 109. Droucopoulos, V. and P. Cronis, Assessing Market Dominance : A Comment and an Extension, January 2010. 110. Babecký, J., P. D. Caju, T. Kosma, M. Lawless, J. Messina and T. Rõõm, Downward Nominal and Real Wage Rigidity: Survey Evidence from European Firms, February 2010. 111. Mitrakos, T., P. Tsakloglou, Analyzing and Comparing te Impact of Alternative Concepts of Resources in Distributional Studies: Greece, 2004/5, Marc 2010. 112. Hall, G. S, G. Hondroyiannis, P.A.V.B. Swamy and G. Tavlas, Bretton-Woods Systems, Old and New, and te Rotation of Excange-Rates Regimes, April 2010. 113. Konstantinou, P. and A. Tagkalakis, Boosting Confidence: Is Tere a Role for Fiscal Policy?, April 2010. 114. Atanasoglou, P. P, C. Backinezos and E. A. Georgiou, Export Performance, Competitiveness and Commodity Composition, May 2010. 115. Georgoutsos, A. D and P. M Migiakis, European Sovereign Bond Spreads: Monetary Unification, Market Conditions and Financial Integration, June 2010. 116. Tagkalakis, A., Fiscal Policy and Financial Market Movements, July 2010. 117. Brissimis, N. S., G. Hondroyiannis, C. Papazoglou, N. T Tsaveas., and M. A. Vasardani, Current Account Determinants and External Sustainability in Periods of Structural Cange, September 2010. 44