Bank-specific factors and credit risk: evidence from Italian banks in different local markets

Abstract

Purpose

Using bank-level data over the 1994–2015 period, the authors aim to investigate the role of bank-specific factors on credit risk in Italy by considering two different groups of banks, namely, cooperative and non-cooperative (commercial and popular), in different local markets.

Design/methodology/approach

Relying on highly territorially disaggregated data at labour market areas’ level, the authors estimate the impact of the role of bank-specific factors on credit risk in Italy from the estimation of a fixed-effect estimator. Non-performing loans to total loans has been used as a proxy of credit risk; the bank-specific factors are as follows: growth of loans, reflecting credit policy; log of total assets, controlling for banks’ size; loans to total assets, reflecting the volume of credit market; equity to total assets, capturing the solvency of banks and reflecting their capital strength; return on assets, reflecting the profitability of banks; deposits to loans, reflecting the intermediation cost; cost of total assets, reflecting the banks’ efficiency or volume of intermediation cost.

Findings

The empirical findings suggest that regulatory credit policy, capitalisation, volume of credit and volume of intermediation costs are the main bank-specific factors affecting non-performing loans. Nevertheless, the present analysis suggests that the behaviour of cooperative banks’ behaviour seems to be in line with that of commercial rather than popular banks, casting doubts about the feasibility of their credit policies. It turns out that recent reforms involving popular and cooperative banks represent the first step toward the enhancement of the stability and efficiency of the Italian banking system. While the present study’s benchmark results are not particularly affected by the degree of competition in the banking sector and by banks’ size, it shows that both cooperative and non-cooperative banks have undertaken more prudent credit policies after the advent of the financial crisis and the introduction of the Basel regulation.

Originality/value

The relationship between bank-specific factors and credit risk has been analysed using a rich sample of cooperative, commercial and popular banks in Italy over the 1994–2015 period. The authors rely on labour market areas being sub-regional geographical areas where the bulk of the labour force lives and works. The contribution is motivated by the financial distress experienced after the 2008 financial crisis, which has significantly hit the Italian banking system and cooperative banks in particular.

Citation

Barra, C. and Ruggiero, N. (2023), "Bank-specific factors and credit risk: evidence from Italian banks in different local markets", Journal of Financial Regulation and Compliance, Vol. 31 No. 3, pp. 316-350. https://doi.org/10.1108/JFRC-04-2022-0051

1. Motivation and literature

The link between bank-specific factors and bank risk has been the subject of much attention in the literature, with several contributions, both at national and cross-country levels, been devoted to the identification of the relevant bank-specific drivers of credit risk. Salas and Saurina (2002), using data for Spain over the 1985–1997 period, empirically assess the impact of both bank-specific and macroeconomic determinants on credit risk, focusing on two different institutional regimes, namely, commercial and saving banks, and show that factors like branch expansion, inefficiency, portfolio composition, size and market power play a key role in explaining credit quality and that saving banks are largely more affected by bank-specific factors.

Haq and Heaney (2012) relied on a set of 117 banking institutions for 15 European Union (EU) countries over the 1996–2010 period to investigate the impact of various bank-specific factors, like bank capital, charter value, off-balance sheet activities, dividend pay-out ratio, operating leverage and size, on both bank equity risk and credit risk. Their evidence points out to a U-shaped relationship between bank capital and both equity and credit risk. In contrast, while off-balance sheet activities determine an increase in bank risk, evidence of an inverse relationship between dividend pay-out ratio and all the measures of risk that were employed was found. Further, their analysis revealed that both equity and credit risk depend upon banks’ size, with large banks contemporaneously exhibiting higher systematic risk but lower credit risk.

Ghosh (2015) relied on a generalised method of moments (GMM) approach to deal with the relationship between bank-industry-specific factors and non-performing loans, considering the whole set of commercial and savings banks in the 50 US States and the District of Columbia over the 1984–2013 period. While higher profitability was found to be inversely correlated with credit risk, capitalisation, liquidity risks, cost inefficiency and the size of the banking industry were found to have a detrimental effect on the variable of interest.

Chaibi and Ftiti (2015), within a dynamic panel data framework, studied the effects of both macroeconomic and bank-specific factors on non-performing loans for a set of commercial banks for France and Germany over the 2005–2011 period, to assess whether the effects of the relevant explanatory variables depended upon the type of banking system, as the two countries are, respectively, classified as market-based and bank-based. The set of bank-specific variables included in the econometric analysis were loan loss provision, inefficiency, leverage, solvency ratio, non-interest income, size and bank profitability, and their evidence revealed that bank-specific factors affect credit quality mostly in the case of the French market-based system.

Other papers, like those of Agoraki et al. (2011) and Tabak et al. (2012), dealt instead with the relationship between market structure and risk-taking propensity. Agoraki et al. (2011) indeed analysed the nexus between capital requirements, market structure and risk-taking propensity, considering a sample of 546 banks for a set of 13 countries from Central and Eastern Europe over the 1998–2005 period, finding evidence of an inverse relationship between market structure and risk-taking and that banks with market power tend to display lower probabilities of default. This contribution further revealed that capital requirements reduce credit risk, although this effect vanishes or can be even reversed once market power is considered. Tabak et al. (2012), using data for ten Latin American countries over the 2003–2008 period, focused on the role of competition, size and capital requirements on bank risk. Competition and bank risk exhibited a non-monotonic relationship, with the former further shown to strengthen financial stability. Capital ratios were instead found to play a key role in the case of collusive markets, with beneficial effects on the stability of large banks, if they operate in moderately or highly competitive markets.

Another strand of the literature, see, for instance, Louzis et al. (2012) and Abid et al. (2014), has focused on the relationship between management quality and risk. Louzis et al. (2012), using quarterly data for the nine largest banks in Greece over the 2003–2009 period, proposed the application of dynamic panel methods to identify both the aggregate and bank-specific variables affecting non-performing loans. They showed that most of the variation in the non-performing loans ratio is driven by macroeconomic factors, while management quality plays a preeminent role among the set of bank-specific factors assumed to influence bank risk. Using a similar econometric approach, based on the application of the system GMM, Abid et al. (2014) identified the main bank-specific and macroeconomic drivers of households’ non-performing loans in a sample of 16 Tunisian banks over the 2003–2012 period. In line with Louzis et al. (2012), they showed that although most of the variation in credit risk is explained by variations in the underlying macroeconomic environment, higher quality of management is associated with lower credit risk.

Finally, other contributions have instead focused on of the relationships between ownership structure on both risk-taking behaviour and banks’ efficiency.

Zribi and Boujelbène (2011), using data for a set of commercial banks for Tunisia over the 1995–2008 period, showed that beyond customary macro factors, ownership structure, prudential regulation of capital and profitability play a key role in explaining credit risk dynamics.

Forssbæck (2011) investigated the effects of market discipline and ownership on credit risk for a set of 331 banks worldwide over the ten-year period from 1995 to 2005, finding evidence of a negative relationship between market discipline and credit risk, and favourable evidence concerning the existence of a U-shaped relationship between ownership structure and the variable of interest. The paper further showed that the joint effect of market discipline and ownership has a limited impact on credit quality.

Haque and Brown (2017), relying on data for a set of 132 commercial banks of 12 Middle East and North Africa (MENA) region, dealt with the relationship between ownership structure and efficiency, through the application of a data envelopment analysis (DEA) approach, and found that ownership concentration and governmental ownership increase cost efficiency, while there is positive and significant impact on profits only during the pre-crisis period. The impact of ownership structure on bank productivity has instead been studied by Sanyal and Shankar (2011), who relied on data for 107 Indian banks over the 1992–2004 period, showing that private banks are 38% more productive compared to their publicly owned counterparts.

This paper attempts to contribute to the recent literature concerning the effects of bank-specific factors upon the ratio of non-performing loans, trying to assess whether the impact of the relevant bank-specific factors on credit risk varies according to the institutional regime, as our analysis incorporates different types of banks. More specifically, in our analysis, we assess how bank-specific factors influence credit risk by alternatively considering cooperative and non-cooperative banks (popular and commercial), a partition which is relevant given the structure of the Italian banking system. Indeed, as highlighted by various contributions in the literature, see, for instance, Gutiérrez (2008), Manetti and Bagnoli (2013), Destefanis et al. (2014), Bruno et al. (2018), cooperative banks play a key role in the Italian banking system, as these financial institutions, which are smaller than commercial banks, mostly operate in peripheral, less-developed areas; have consolidated and deep roots with the territory; and represent the main source of funding for small and medium enterprises (SMEs), which would face stricter credit constraints from commercial banks. Further, as emphasised by Berger and Mester (1997) and Pestana Barros et al. (2010), the peculiar governance structure of cooperative banks, in which customers are fully involved in the decision-making process of the intermediary, should, in principle, allow them to better monitor their borrowers, as they are better informed about the quality of debtors. Nevertheless, though, various contributions in the literature, see, for instance, Barra and Zotti (2019a, 2019b), have proven that cooperative banks are more efficient than other types of financial intermediaries, and the performance of cooperative banks has significantly shrunk with the advent of the 2008 financial crisis, a result that, according to Destefanis et al. (2014), beyond unfavourable economic conditions, is essentially ascribed to some regulatory features faced by cooperative banks, with a key role played by localisation constraints. It turns out that a genuine analysis aimed at assessing whether the effects of bank-specific factors on credit risk are similar between the different types of banks that compose the Italian banking system would shed light about its functioning and the way the different financial institutions manage credit risk. In particular, the aim of this paper is to assess whether cooperative banks’ credit policies are aligned to the ones adopted by commercial or popular banks. Indeed, if the behaviour of cooperative banks mimics that of commercial rather than popular banks, it turns out that the recent reforms of popular and cooperative banks represent a step forward to strengthen financial stability and improve the efficiency of the banking sector. Moreover, we bring additional insights by testing whether the influence of the main bank-specific determinants of non-performing loans varies according to the degree of competition in the banking sector, variations in the degree of economic activity (summarised here by the 2008 financial crisis) and changes in the regulatory environment, such as the introduction of Basel Accords, i.e. Basel II and Basel III.

Empirical evidence presented in this paper reveals that the ratio of non-performing loans for both cooperative and non-cooperative banks is influenced by the same bank-specific factors, with a significant role found for regulatory credit policy, capitalisation, volume of credit and volume of intermediation costs. As the estimated parameters are quantitatively similar among the three types of banks (i.e. cooperative, commercial and popular), and especially between cooperatives and commercial, it seems reasonable to assert that cooperative banks operate like their commercial counterparts, a result that is further reinforced by the fact that the impact of bank-specific factors on non-performing loans is not particularly affected by the degree of competition in the banking system and by banks’ size. It turns out that the 2015 implementation of the reform of popular banks, which introduces size limits by requiring popular banks to be transformed into joint stock companies provided that their amount of assets exceeds €8bn, and the 2016 reform of cooperative banks, aimed at reducing the fragmentation of the sector and its structural weaknesses, allowing these intermediaries to obtain funds outside the cooperative sector in case of financial distress, probably represent a step forward to enhance financial stability and to allow for a more efficient allocation of resources within the financial system. Further, empirical findings presented here further suggest that, after the advent of the 2008 financial crisis and the introduction of the Basel II and III accords, both cooperative and non-cooperative banks adopted more prudent credit policies, implying that the introduction of new capital and deposits requirements was effective in reducing credit risk.

The paper is structured as follows. Section 2 discusses the data and the empirical strategy employed in our econometric analysis, while in Section 3, the main empirical findings are introduced, emphasising the behaviour of the two different types of banks considered and the relevance of market structure, to test whether the benchmark results are affected by the degree of competition in the banking system. Section 4 discusses the robustness checks. Specifically, to test the robustness of our estimates, we initially propose to split the sample of banks according to their size, as large and small banks might behave differently in terms of risk-taking propensity, and then test whether the advent of the financial crisis and the introduction of the Basel accords have caused structural breaks in the Italian banking system, therefore affecting the parameters of interest. As an additional robustness check, our baseline regressions are extended by employing an alternative dependent variable, represented by the z-score on the return on asset (ROA). We further assess the robustness of our findings by properly considering the size of the local market areas (LMAs), as the latter might affect the degree of competition in the banking market and, in turn, provide bias estimates of the parameters of interest. Section 5 concludes and discusses the main policy implications drawn from the econometric analysis proposed in this paper.

2. Empirical framework

2.1 Data and variables

To empirically identify the main bank-specific drivers of the ratio of non-performing loans and to test whether these effects vary unevenly across the relevant types of banks scrutinised, our analysis is developed using bank-level information. The main source of data is represented by the BilBank 2000 database distributed by ABI (Associazione Bancaria Italiana), which provides a large amount of information concerning both bank balance sheets and bank-specific factors, covering the 1994–2015 period (see Table 1 for more details on the definition of the variables) [1].

The sample of banks consists of cooperative, commercial and popular banks. Branches of banks located abroad, however, are excluded from our sample, as they follow a different regulation, based on both the national and the local rules of the country in which they operate.

Banks in our sample are classified as per the Bank of Italy as follows: major (average funds intermediated more than €65bn), large (average funds intermediated between €27bn and €65bn), medium (average funds intermediated between €9bn and €27bn), small (average funds intermediated between €1.3bn and €9bn) and minor (average funds intermediated less than €1.3bn). Our data set contains, on average, information for 694 banks, categorically and territorially portioned as follows: 437 for cooperative banks (63% of our sample), 206 for commercial banks (30% of our sample) and 51 for popular banks (7% of our sample); 238 for North-West (34% of our sample), 147 for North-East (21% of our sample), 130 for Centre (19% of our sample) and 179 for South (26% of our sample). Due to the territorially highly disaggregated data availability, the financial reforms (privatisation and second banking directive) occurred after 1990 and the integration of markets, Italy is a promising field of analysis (Table 2).

All monetary aggregates are in thousands and deflated at 2005 Euros. The sample begins in 1994, since some variables are not available before that year.

Table 3 reports descriptive statistics for the main variables employed in the paper, partitioned both at bank and territorial levels. According to Table 3, the proportion of non-performing loans is higher for popular banks, followed by cooperatives and commercial banks. A result that suggests that, at least in the Italian case, popular and cooperative banks are not necessarily less risky than their commercial counterparts. While the ROA is larger for cooperative and popular banks, the degree of capitalisation, volume of credit and volume of cost are similar between the three different types of banks considered. A different picture, however, emerges with respect to both intermediation costs and size, which are higher for commercial banks. Once statistics are partitioned at the territorial level, a clear dualism emerges in the Italian banking system. Descriptive statistics reported in Table 3 indicate that southern regions display a higher proportion of non-performing loans and that banks operating in these areas are smaller than those in other macro areas, but better capitalised. Evidence presented in Table 3 further highlights that, over the period considered, less-developed southern regions exhibit a negative growth of loans, a result that can be ascribed to the advent of the crisis, which has been truly severe for these regions, consistently harming their economic performances. Finally, according to Table 3, banks operating in southern regions have lower returns than those in the most developed regions of the country and exhibit lower volumes of credits.

Table 4 reports the pairwise correlations among the main variables used in our econometric analysis. We document a negative and significant correlation between the growth in total loans and the quality of credit. ROA is negatively and significantly correlated with the main variables, but growth of loans. Capitalisation, on the other hand, is found to be inversely and significantly correlated with the ratio of non-performing loans, ROAs and growth in total loans. The log of total assets, which proxies the bank dimension, is shown to be directly correlated with non-performing loans’ ratio and growth in total loans and inversely correlated with capitalisation, volume of credit and volume of intermediation cost. The volume of credit is shown to be inversely correlated with non-performing loans’ ratio and capitalisation and positively with growth in total loans and ROAs, while intermediation cost correlates negatively with non-performing loans’ ratio and ROAs and positively with capitalisation. Finally, the volume of intermediation cost exhibits a positive and statistically significant relationship with all the main variables employed in the paper.

2.2 Empirical specification

To empirically assess the role of bank-specific factors on non-performing loans, we follow the approach proposed by Hesse and Cihák (2007) and rely on the application of a fixed-effects estimator [2] with cluster standard errors, to deal with the presence of group-wise heteroscedasticity. As highlighted by Wooldridge (2002), the application of the fixed-effects estimator allows unobserved specific factors to be correlated with the drivers of asset quality, while Beck et al. (2013) emphasised that the presence of these unobserved specific factors allows the fixed-effect estimator to solve the problem of omitted variable bias and to capture the existing heterogeneity within cross-sectional units. It must be further highlighted that, although the main literature surveyed here relies on the application of dynamic panel methods, most notably the GMM, this econometric procedure has little ground in this environment, given the structure of our data set. Indeed, it has been recognised that the GMM is suitable for panels with few time periods and a higher number of individuals (small T, large N) [3]. As our data set is characterised by both a large time span and a large number of cross-sectional units, the application of the GMM is inappropriate in the current framework. As suggested by some preliminary estimates and by Akaike and Schwartz information criteria, we propose a dynamic specification in which all the regressors enter with their first lag in order to control for possible endogeneity problems, while an AR(1) in credit risk captures the degree of persistence. Formally, the model is specified as follows:

(1) CRi,t=α1CRi,t−1+βjBANK−SPECIFICi,t−1+REGIONk+TIMEt+εi,t   

where CR is the credit risk (i.e. non-performing loans to total loans) (Espinoza and Prasad, 2010; Louzis et al., 2012; Beck et al., 2013); BANK − SPECIFIC is a vector of bank-specific factors, which includes: GRLO: growth of loans, reflecting credit policy (Salas and Saurina, 2002; Quagliariello, 2007); TA: log of total assets, controlling for banks’ size (Salas and Saurina, 2002; Hesse and Cihák, 2007; Quagliariello, 2007; Louzis et al., 2012); LTA: loans to total assets, reflecting the volume of credit market; ETA: equity to total assets, capturing the solvency of banks and reflecting their capital strength (Quagliariello, 2007; Hesse and Cihák, 2007; ROA: return on assets, reflecting the profitability of banks (Quagliariello, 2007; Louzis et al., 2012); DL: deposits to loans, reflecting the intermediation cost; CTA: cost of total assets, reflecting the banks’ efficiency or volume of intermediation cost (similar to Hesse and Čihák, 2007; Quagliariello, 2007); REGION is the set of region dummies included in the model in order to capture geographical or spillover effects; TIME is the set of time dummies included in the model to capture changes in macroeconomic dynamics (e.g. the lowering of interest rates) and rules (e.g. the processes of financial deregulation and privatization in Italy). Finally, ε is the error term. Subscripts i and t refer, respectively, to financial intermediaries and time periods (years), while k refers to Italian regions. We include the first lag of CR as an explanatory variable, since the ratio of non-performing loans displays a significant degree of persistence [4]. This further serves to capture capital reserves built in the previous periods. All the relevant variables included in our empirical analysis are deflated (at 2005 prices) to avoid that the relationship might be affected by a price effect. Finally, all standard errors are clustered at the bank level.

3. Empirical evidence

3.1 Regression results: How do cooperative and non-cooperative banks behave?

Table 5 reports the results of our baseline regressions, for the whole sample of banks and for the different types of banking institutions composing the Italian financial system, namely, cooperative, commercial, and popular. Evidence indicates that regulatory credit policy and volume of intermediation costs significantly affect the quality of credit, regardless of the type of banking institutions. Specifically, we show, consistently with the previous findings of Vithessonthi (2016), that regulatory credit policy reduces credit risk. Accordingly, if banks expect a future increase in the amount of non-performing loans, caused by a deterioration in living standards, the growth of credits will reduce credit risk. Higher intermediation costs, on the other hand, have a detrimental effect on the proportion of non-performing loans. This result may suggest, in line with Poghosyan (2013), that as credits become riskier, the higher is the cost of intermediation required by financial institutions to finance the project. Among the other bank-specific factors included in our econometric analysis, we find that volume of credit and intermediation costs reduce the proportion of non-performing loans, although this effect is limited to the whole set of banks and to cooperatives. On the other hand, higher capitalisation is shown to have a negative and significant effect on the left-hand-side variable for all the groups considered in our analysis, but popular banks. The empirical findings presented here suggest an important role for regulatory credit policy, with the loan restriction of banks that reduces credit risk. In other words, intermediation cost has a negative influence on bank risk, supporting the idea that a strict credit policy can be beneficial in reducing credit risks. The same applies to the level of bank capitalisation, which is found to reduce credit risk, in line with the predictions of Agoraki et al. (2011), Tabak et al. (2012) and Ghosh (2015). This result, in line with the predictions of Berger and DeYoung (1997), seems to suggest that managerial incentives to behave opportunistically shrink in the presence of improved capital standards, with beneficial effects on the riskiness of banks’ loan portfolios.

Nevertheless, as the main goal of the paper is to assess whether the impact of bank-specific factors on the ratio of non-performing loans is similar between the different types of financial institutions composing the Italian banking system, our results seem to suggest that cooperative banks’ behaviour is more aligned to the one of the commercial rather than popular banks. Indeed, the estimated coefficients and their significance indicate that cooperative banks’ credit risk policies tend to mimic commercial banks’ credit risk policies and diverge from those adopted by popular banks. However, the structural and market differences between cooperative and non-cooperative banks should not be underestimated. As pointed out by Fonteyne (2007) and Ayadi et al. (2010) and emphasised by Barra et al. (2016), cooperative banks are an important feature of the European banking landscape. These banks are characterised by small size, self-governance and the principle of mutuality (internal: their activity is biased in favour of members; external: there must be activities aimed at supporting the moral, cultural and economic development of the local community). They also face localisation constraints, providing loans only within a restricted area, the so-called area of territorial competence (area di competenza territoriale), which includes the municipality in which the bank has its head office and the contiguous areas. These fundamental characteristics could explain some different results between cooperatives and non-cooperatives, especially in terms of monitoring of borrowers and reducing of bad loans.

3.2 Does the market structure affect the estimation?

Recent theoretical and empirical literature has emphasised that the degree of competition is crucial in explaining the dynamics of credit risk, showing mixed evidence. In line with the “competition-fragility” hypothesis, one strand of the literature, summarised here by the contributions of Keeley (1990), Carletti and Hartmann (2002), Berger et al. (2009), Agoraki et al. (2011) and Jiménez et al. (2013), highlights that higher competition in the banking sector decreases the profit margin and reduces the franchise value, therefore increasing credit risk and reducing financial stability. A second strand of the literature, see, for instance, Salas and Saurina (2002), Boyd et al. (2006), De Nicolò and Lakouianova (2007), Schaeck et al. (2009), Tabak et al. (2012), Fiordelisi and Mare (2014) and Kasman and Carvallo (2017), supports instead the “competition-stability” view, according to which increased competition in the financial sector reduces credit risk and the probability of banks’ failures. In this section, we test whether the impact of bank-specific factors on the ratio of non-performing loans depends upon the market structure in which banks operate. We therefore construct a set of four dummy variables aimed at identifying the degree of competition at LMA level. According to the official definition provided by the Italian Statistical Office (ISTAT):

Labour market areas are sub-regional geographical areas where the bulk of the labour force lives and works, and where establishments can find the largest amount of the labour force necessary to occupy the offered jobs. They respond to the need for meaningfully comparable sub-regional labour market areas for the reporting and analysis of statistics. LMAs are defined on a functional basis, the key criterion being the proportion of commuters who cross the LMA boundary on their way to work.

LMA is therefore represented by a set of two or more contiguous municipalities, which constitute a single labour market and that are comparable in terms of bilateral flows of population. The ISTAT has identified 686 LMAs, which are partitioned into twenty macro regions [5], respectively, represented by North-West, North-East, Centre and South. Over the 686 LMAs, 114 are in North-West, 119 in North-East, 128 in Centre and the remaining 325 in Southern. Our data set contains information for 321 LMAs, territorially partitioned as follows: 89 for North-Western, 55 for North-Eastern, 65 for Centre and 112 for South. As we rely on highly disaggregated data and given the special stratification embedded in our information, our analysis allows us to not only obtain a rich and detailed picture of the characteristics of the Italian banking system but also capture the existing heterogeneity across geographical areas, the contribution of local credit institutions and of different market structures on credit risk. It turns out that, differently from the existing literature, we measure the degree of competition in the banking sector by building indices of the market structure at the LMA level. Specifically, following Destefanis et al. (2014), (see Table 6 for more information on statistics of variables based on different market structure). Formally, we construct a variable taking the value of 1 when the LMA level, there is one bank (MON: MONOPOLY); 2 when at the LMA level there are two banks (DUOP: DUOPOLY); 3 when at LMA level there are three banks (OLIG: OLIGOPOLY) and finally 4 when at LMA level there are more than three banks (COMP: COMPETITION) the model described in equation (1) becomes:

(2) CRi,t=α0+α1CRi,t−1+βjBANK−SPECIFICi,t−1+TIMEt+εi,t  if market structure=[MON,DUOP,OLIG,COMP] 

As the empirical assessment of the role of the market structure would cause a consistent loss of observations which might bias our estimates, we proceed by excluding, one-by-one, all the relevant market structures. Table 6 reports descriptive statistics of the main variables based on the four market structures identified. According to the descriptive statistics presented in Table 6, the ratio of non-performing loans is higher in the case of the monopolistic market and lower in the case of the fully competitive environment, while the opposite holds for growth of loans, which is found to be higher in the case of perfect competition and lower in the presence of market power. ROAs are instead shown to be quantitatively similar in the case of imperfect competition and largely higher compared to the fully competitive market. While capitalisation is similar among the different market structures, oligopolistic markets display higher levels of volume of credit compared to monopoly and perfect competition. The volume of intermediation cost and bank dimension are instead found to be consistently higher for competitive markets compared to their market power counterparts.

Table 7 assesses the impact of bank-specific factors on the ratio of non-performing loans for the whole sample and for the three main different types of banks, considering the degree of competition in which banks operate. Empirical evidence, in line with the estimates obtained so far, indicates that regulatory credit policy, capitalisation and volume of credit market significantly reduce credit risk and are not significantly influenced by either the market structure or the type of banks scrutinised. Finally, while volume of intermediation cost has a detrimental effect upon the variable of interest, regardless of the degree of competition and the type of bank, the effects of intermediation cost are significant and depend on both the market structure and the financial institution considered. Specifically, we show that intermediation cost is not significant in the case of cooperative banks, but negatively and significantly affects the proportion of bad loans held by commercial banks, once we exclude monopolistic and competitive markets. Finally, we find evidence of a weak and significant effect in the case of popular banks in all the market structures considered but the case in which we exclude the competitive market. These results reinforce our main findings according to which cooperative banks substantially behave like their commercial counterparts. Differently from the main literature considered in this paper, but in line with the findings of Martín-Oliver et al. (2020), the empirical evidence presented in Table 7 indicates that, at least in the Italian case, the ratio of non-performing loans is not particularly affected by both the market structure in which banks operate and by the type of financial institution considered.

4. Robustness checks

4.1 Does the bank size affect the estimation?

To assess the robustness of our estimates, we initially partition banks according to their size, as the latter, as shown by various contributions in the literature, see for instance Salas and Saurina (2002), Ghosh (2015) and Chaibi and Ftiti (2015), is a relevant bank-specific factor of credit risk. To test whether banks’ size affects the main results, we follow the classification proposed by the Bank of Italy, already discussed in Section 2.1. Since the empirical assessment of the role of the bank’s size would cause a consistent loss of observations, which might bias our estimates, we will alternatively exclude major, large, medium, small and minor banks. Evidence presented in Table 8 suggests that, once banks are partitioned according to their size, the main findings are almost unaffected, as regulatory policies, capitalisation, volume of credit, intermediation costs and volume of intermediation costs are shown to statistically affect the proportion of non-performing loans. Once again, our estimates indicate that more stringent credit policies, capitalisation, volume of credit and intermediation costs reduce credit risk, while the volume of costs has a positive impact upon the variable of interest. Differently from the main literature surveyed here, our evidence indicates that, at least in the Italian case, banks’ size has no sizable impact on the share of non-performing loans.

4.2 Does financial crisis affect the estimation?

The advent of 2008 financial crisis has stimulated a large amount of literature aimed at addressing the impact of the recession on several dimensions of the financial system.

Altunbas et al. (2011) and Claessens and van Horen (2015) focus their attention on how banks’ business models were affected by the crisis, with great emphasis on their location decisions. While Altunbas et al. (2011) show that financial institutions with a high level of risk were less capitalised, large in terms of size and with an aggressive credit policy, Claessens and van Horen (2015) show that, following the financial crisis, several financial institutions decided to reduce their presence abroad.

A second strand of the literature (Beltratti and Stultz, 2012; Demirguc-Kunt et al., 2013 and Berger and Bouwnan, 2013) has instead emphasised the role of capital requirements during the various recessionary episodes. While Beltratti and Stultz (2012) show that banks with less leverage and lower stock returns displayed better performances during the recessionary periods, the opposite is found by Demirguc-Kunt et al. (2013) whose evidence suggests that better-capitalised banks exhibited better performance in terms of stock returns, a result that is further strengthened by the application of other measures of capital, like TIER 1 capital.

On the other hand, the contribution of Berger and Bouwnan (2013) focuses on the relationship between capital and size during the crisis, finding that small banks take advantage of their capital levels during both expansions and recessions.

As the advent of the financial crisis might have caused a structural break in the functioning of the Italian banking system which, in turn, might affect the parameters of interest, we exploit the nature of our data and propose a set of econometric estimates, limitedly to the 1994–2007 period. Formally, the model described in equation (1) becomes:

(3) CRi,t=α0+α1CRi,t−1+βjBANK−SPECIFICi,t−1+TIMEt+εi,t  if pre− crisis 

Evidence reported in Table 9 indicates that once we allow for the effects of the financial crisis, the benchmark results are only partially affected. Specifically, compared with our benchmark specifications, our results confirm a negative and significant impact of regulatory credit policy. While capitalisation loses its significance, both the volume of credit market and intermediation cost are now significant only in the case of cooperative banks. Volume of intermediation cost is instead significant only once we consider the whole set of banks and cooperatives.

4.3 Do Basel agreements affect the estimation?

The effects of Basel regulation on credit risk have been widely debated in the economic literature, both theoretically and empirically. Hakenes and Schnabel (2011) analyse the impact of Basel II accord on the nexus between bank size and risk-taking within a theoretical framework. They show that one of the effects of the Basel II accord was to provide large banks with a competitive advantage in the market and that increased competition led small banks to undertake risky projects, with a detrimental effect on aggregate risk-taking. Gavalas (2015) investigated the effects of the Basel III accord on bank performance using data for some EU countries, showing that capital requirements increased banks’ marginal costs which, in turn, determined an increase in lending rates. Moreover, the paper finds evidence of an inverse relationship between the Basel agreements and the total volume of loans. Buch et al. (2015) investigated the effects of uncertainty on bank lending, showing that better-capitalised banks, where capitalisation is measured using TIER1 regulatory quality, are less affected by higher uncertainty. Kim and Sohn (2017), using data from US-insured commercial banks, analyse the effects of liquidity on the relationship between bank capital requirements and bank lending. They show that the relationship between capital requirements and credit growth crucially depends on the level of liquidity, with these requirements representing a beneficial policy only for large banks, if they have enough liquid assets. Roulet (2018), using data from a set of commercial banks of 22 countries over the 2008–2015 period, studied the effects of capital requirements introduced by the Basel III accord after the 2008 recession. Their paper shows that capital requirements had a negative effect on lending growth, especially for large banks, while liquidity indicators are found to have a positive but perverse effect on the variable of interest. Naceur and Routlet (2018) address the effects of capital ratios and liquidity requirements for the United States and Europe after the financial crisis, showing that the regulation determined an increase in the ability of US banks to absorb risks when they expand their credit capacities, while the effects of capital requirements and liquidity are like the ones found by Roulet (2018).

In the context of the current contribution, it seems therefore reasonable to address whether the introduction of the capital requirements embedded in the Basel regulations changed banks’ incentives which, in turn, might affect the estimated bank-specific parameters. To empirically assess whether our benchmark results are altered by the introduction of the Basel regulation, the model is re-estimated considering both Basel II (1995–2005 period) and Basel III (1995–2013). Formally, the model described in equation (1) becomes:

(4) CRi,t=α0+α1CRi,t−1+βjBANK−SPECIFICi,t−1+TIMEt+εi,t  if pre−Basel agreements 

Results reported in Table 10 indicate that the introduction of both Basel II and Basel III accords did not significantly change the benchmark results. Specifically, we still find evidence of a significant effect of growth of loans, volume of the credit market and volume of intermediation cost. While capitalisation is found to have a weak impact for the whole set of banks and for commercial, once Basel III is considered, the intermediation cost is significant in the case of cooperative banks and in the case of popular banks, although this effect is limited to the Basel III accord.

4.4 Does a different dependent variable affect the estimation?

To further test the robustness of our estimates, in this section, we propose a set of econometric specifications in which an alternative measure of bank insolvency risk is employed, summarised here by the z-score on the ROAs. Although the ratio of non-performing loans represents the most widely and commonly used measure of credit risk, the z-score of the returns of asset, often associated with the concept of volatility, is used to capture both bank soundness and financial instability [6] (see, among others, De Nicolò and Lakouianova, 2007; Laeven and Levine, 2009; Hesse and Cihák, 2007; Uhde and Heimeshoff, 2009; Fiordelisi and Mare, 2014; Chiaramonte et al., 2015; Barra and Zotti, 2019a, 2019b). Specifically, for each bank i and time t, the z-score indicator is defined as follows:

Z−SCOREROA,i,t=ETAi,t+ROAi,tσROAi,t

where ETA is the level of capitalisation of the bank (i.e. equity to total assets), ROA denotes the ratio between profit and total assets (i.e. return on assets) and σ_ROA is the standard deviation of the ROA. Following Agoraki et al. (2011) and Soedarmono et al. (2013), σROA at time t is calculated based on observations of ROA from time t to t − 2 (a three-period rolling window). The z-score therefore represents a good proxy of a bank’s distance to default and does not require strong assumptions about the distribution of the ROAs (Strobel, 2011). It combines banks’ buffers (capital and profits) with the risks they face (measured by the standard deviation of returns), reflecting the number of standard deviations by which returns would have to fall from the mean to wipe out equity. A higher value of z-score implies a lower probability of insolvency risk (Uhde and Heimeshoff, 2009) and greater stability (e.g. inverse of the probability of defaults), providing a direct measure of the banking system stability. Further, as emphasised by Kasman and Kasman (2015), the application of the z-score allows to obtain a more direct measure of bank soundness compared to other measures of risk. The z-score is a positive function of both banks’ profitability and capital ratio and a negative function of the conditional volatility.

Formally, the model described in equation (1) becomes:

(5) Z−SCOREROA,i,t=α0+α1Z−SCOREROA,i,t−1+βjBANK−SPECIFICi,t−1+TIMEt+εi,t   

Evidence reported in Table 11 suggests that growth of loans is significant among the different types of banks examined, but the whole sample. Capitalisation has a positive, as expected, and significant impact on the variable of interest, though it is not significant in the case of commercial banks. The volume of the credit market exhibits a positive and statistically significant relationship with the endogenous variable limitedly to the whole set of banks and cooperatives, while intermediation cost is significant only in the case of cooperative banks. Finally, while the volume of intermediation cost is significant only in the case of the whole set of banks, higher bank size is found to reduce the z-score on ROA limitedly to commercial banks. All in all, evidence suggests that the benchmark results are only weakly affected by the application of an alternative dependent variable.

4.5 Do potential differences inside LMAs affect the estimates?

As the proxies of the market structure employed might be biased by the size of the LMAs, as larger cities might be more attractive for financial institutions and, hence, more competitive, we propose a set of econometric estimates where we exclude LMAs belonging to the first and tenth decile. This econometric exercise rules out the possibility that the size of the LMAs might bias our measures of the market structure and, in turn, to bias our evidence.

The evidence reported in Table 12 indicates that the growth of loans significantly reduces credit risk, although we do find evidence of a statistically insignificant effect for both commercial and popular banks once we remove the tenth decile.

In line with the evidence reported so far, the ROA play no significant role in affecting the ratio of non-performing loans, while both capitalisation and volume of the credit market are found to reduce credit risk. A relatively limited impact is instead found for intermediation costs, while the volume of intermediation cost is found to have an adverse impact on credit risk.

The log of total assets, in line with the estimates reported so far, is instead found to have no significant impact on the variable of interest.

All in all, the evidence reported in Table 12 seem to confirm the robustness of our findings and the results reported in our benchmark analysis, hence implying that our evidence is confirmed even after we exclude smaller and larger LMAs, which might affect the degree of competition and, in turn, our findings.

Table 13 reports a set of econometric estimates in which we employ the z-score instead of the ratio of non-performing loans as our proxy of credit risk and remove the smallest and the largest LMAs. Accordingly, there is some evidence of an inverse and significant relationship between the growth of loans and the variable under scrutiny. While the effects of the ROA on the variable of interest are weak, both ETA and LTA are found to have a positive and significant effect on the z-score. Moreover, while there is some evidence of a positive effect of intermediation costs on the variable of interest, both the volume of intermediation costs and total assets play no role in affecting the z-score. The results reported in

3 provide additional evidence concerning the robustness of our findings and confirm that our benchmark findings are not significantly affected by the size of the LMAs.

5. Conclusions and policy implication

Using bank-level data over the 1994–2015 period, this paper investigates the impact of bank-specific factors upon the ratio of non-performing loans in Italy, by addressing whether the effects of bank-specific factors upon credit risk depend on the institutional system in which banks operate. Specifically, the goal of the paper is to test whether the impact of the relevant bank-specific factors included in the econometric analysis depends upon the type of financial intermediary considered, by partitioning the population of banks according to cooperative and non-cooperative banks, i.e. popular and commercial, dichotomy. The advent of the financial crisis has indeed caused various and prolonged situations of financial distress that have involved cooperative banks as well, casting doubts about the ability of these financial institutions to efficiently monitor their borrowers and about the feasibility of their credit strategies. It turns out that a genuine analysis aimed at assessing whether the effects of bank-specific factors on credit risk are similar between cooperative and non-cooperative banks would shed light about the functioning of the Italian banking system. Empirical evidence presented in the paper reveals that the ratio of non-performing loans for both cooperative and non-cooperative banks is influenced by the same bank-specific factors, with a significant role found for regulatory credit policy, capitalisation, volume of credit and volume of intermediation costs. Specifically, while regulatory credit policy, capitalisation and volume of credit are found to reduce the ratio of non-performing loans, higher volume of intermediation costs has a positive effect on the latter. The empirical findings presented here suggest an important role of regulatory credit policy, with the loan restriction of banks that reduces the risk. In other words, intermediation cost has a negative influence on bank risk, supporting the idea that a strict credit policy can be beneficial in reducing the risks related to an increase in the proportion of non-performing loans. The same applies to the level of bank capitalisation, which is found to reduce credit risk, as higher levels of capitalisation allow banks to cover any risks that could harm their balance sheets, with beneficial effects for the stability of the financial system and, in turn, for the real economy. As the estimated parameters are quantitatively similar, especially once we compare cooperative and commercial banks, it seems reasonable to assert that the cooperative banks’ behaviour is closest to commercial rather than popular banks. To provide additional insights, our analysis further considers the role of the market structure and of banks’ size. Results presented here highlight that the ratio of non-performing loans is not particularly affected, either by the degree of competition or by banks’ size, a result that reinforces the idea that, although different in nature and size, cooperative banks tend to behave like their commercial counterparts. To assess the robustness of our estimates, various alternative strategies have been proposed. Specifically, we initially assess whether the estimated parameters remain stable following the financial crisis and the introduction of Basel II and III, as both variations in the macroeconomic and regulatory environment might have changed banks’ behaviour. We then rely on the application of an alternative measure of credit risk, summarised here by the z-score on the ROA. Further, as the size of the LMAs might affect the degree of competition in the banking market and, through that, bias our estimates, as an additional robustness check, we propose a set of econometric estimates where we alternatively remove the smallest and the largest LMAs. This battery of robustness checks mostly confirms the main findings, although the advent of the financial crisis and the introduction of the capital requirements embedded in the Basel regulation have forced both cooperative and non-cooperative banks to undertake more prudent credit policies, based on credit restrictions. Evidence presented here has various policy implications, especially with respect to the stability of the Italian banking sector. Our results indicate that policy measures aimed at improving banks’ monitoring abilities would be beneficial, as the higher the ability of banks in monitoring their borrowers, the lower is the amount of non-performing loans, thus determining an efficient allocation of resources. A similar picture concerns the role of capitalisation, where regulators should impose capital standards, rendering banks more and better capitalised, reducing credit risk.

At the same time, policy measures which reduce the costs of intermediation would be highly beneficial in reducing credit risk. These policies should mostly be tailored towards commercial banks, as the latter face significantly larger costs of intermediation compared to other types of financial intermediaries. Measures based on the expansion of the cooperative banking sector, on the other hand, should be taken with caution. Indeed, although our evidence points to an inverse relationship between the volume of credits provided by these banking institutions and their ratios of non-performing loans, at the same time, it has been proven that the expiation of their branches caused a worsening in the levels of credit risk (Bernini and Brighi, 2018).

From this perspective, however, the recent reforms of cooperative banks, aimed at reducing its fragmentation and to increase the availability of external sources in case of financial distress, represent a significant step forward to enhance the stability and the resilience of the Italian financial system. The same applies to the recent reforms of banche popolari, which requires these banks to be converted into joint stock companies provided that their assets exceed €8bn.

Table 1.

Description of the variables

Variables	Symbol	Description
Bank risk
Credit risk	CR	Non-performing loans to total loans.a
Financial stability (1)	ZSCOREROA	Capitalisation plus return on assets over standard deviation of return on assets.a
Financial stability (2)	ZSCOREROE	One plus return on equity over standard deviation of return on equity.a
Bank-specific factors
Credit policy	GRLO	Loans at time t minus loans at time t − 1.a
Return on assets	ROA	Profit to total assets.a
Capitalisation	ETA	Equity to total assets.a
Dimension of bank	TA	Log of total assets. a
Volume of credit market	LTA	Loans to total assets. a
Intermediation cost	DL	Deposits to loans. a
Volume of intermediation cost	CTA	Cost to total assets. a
Controls
Dimension of banks	DYM	Large, medium, small and minor dummies; major used as benchmark group
Timing	TIME	Set of time dummies
Macro regions	MR	Set of region dummies (i.e. 20) included in the model to capture the geographical or spillover effects
Market structure
Monopolistic dummy	MON	Monopolistic market (assuming 1 if at LMAs’ level there is one bank [cooperative or non-cooperative banks], 0 otherwise)
Duopolistic dummy	DUO	Duopolistic market (assuming 2 if at LMAs’ level there are two banks [cooperative or non-cooperative banks or both], 0 otherwise)
Oligopolistic dummy	OLIG	Oligopolistic market (assuming 3 if at LMAs’ level there are three banks [cooperative or non-cooperative banks or both), 0 otherwise)
Competitive dummy	COMP	Competitive market (assuming 4 if at LMAs’ level there are more than three banks [cooperative or non-cooperative banks or both], 0 otherwise)

Note:

Deposits and loans are those granted to customers

Source: ^aOwn calculations upon BilBank 2000 database from ABI

Table 2.

Description of sample used in the analysis: the market structure

Market structure	North-West	North-East	Centre	South	Cooperative banks	Commercial banks	Popular banks	All banks
MON	582	507	669	1,359	2,059	375	233	3,117
DUO	747	509	549	961	1,852	604	310	2,766
OLIG	939	198	479	579	1,515	555	118	2,188
COMP	2,967	2,024	1,170	1,035	3,735	3,003	458	7,196
ITALY	5,235	3,238	2,860	3,934	9,611	4,537	1,119	15,267

Source: Own calculations upon BilBank 2000 database from ABI (Associazione Bancaria Italiana)

Table 3.

Summary of statistics

Banks/Variables	CR	GRLO	ROA	ETA	LTA	DL	CTA	TA
Cooperative banks
(N = 9,611)	0.0163	9938.54	0.0053	0.1337	0.6213	1.1752	0.0746	274437.2
	0.0194	41383.03	0.0270	0.0415	0.1928	16.1989	0.052	465499.4
Commercial banks
(N = 4,537)	0.0137	239758.6	−0.0000	0.1275	0.6005	24.8065	0.0802	1.24e+07
	0.0226	5193563	0.03288	0.0847	0.2827	349.4974	0.0704	4.22e+07
Popular banks
(N = 1,119)	0.0192	285044	0.0049	0.1332	0.6453	0.9080	0.0715	9993896
	0.0286	2513528	0.0159	0.0629	0.1905	0.6550	0.0498	2.10e+07
North-West
(N = 5,235)	0.0120	95452,23	0.0061	0.1376	0.6921	2.3599	0.0743	2894735
	0.0179	2077605	0.0327	0.0512	0.2068	76.9778	0.0510	17400000
North-East
(N = 3,238)	0.0134	292389.8	0.0005	0.1286	0.5940	17.6598	0.0798	8832653
	0.0202	4944393	0.0326	0.0720	0.2637	218.0379	0.0709	35400000
Centre
(N = 2,860)	0.0165	16129.28	0.0039	0.1229	0.6135	8.095	0.0758	5646839
	0.0219	2199006	0.0253	0.0535	0.2122	258.9389	0.0562	21200000
South
(N = 3,934)	0.0222	−8276.78	0.0029	0.1334	0.5382	8.1774	0.0755	2581162
	0.0241	1575316	0.0186	0.0610	0.1816	214.7064	0.0563	22800000
All sample
(N = 15,267)	0.0158	95258,62	0.0037	0.1318	0.6169	8.1783	0.0760	4588873
	0.0213	2855534	0.0284	0.0593	0.2236	191.2491	0.0581	24400000

Notes:

See Table 1 for more details about the description and construction of variables included in the regression. All variables averaged between 1994 and 2015. All monetary aggregates are in thousands of euros (at 2005 prices). Standard deviation in parentheses

Source: Own calculations upon BilBank 2000 database from ABI (Associazione Bancaria Italiana)

Table 4.

Pairwise matric correlations between variables

Variables	CR	GRLO	ROA	ETA	LTA	DL	CTA	TA
CR	1.0000
ΔLO	−0.0226***	1.0000
ROA	0.1276***	−0.0046	1.0000
ETA	−0.1273***	−0.0144	−0.1601***	1.0000
LTA	−0.1574***	0.0408***	0.1154***	−0.0929***	1.0000
DL	−0.0275***	−0.0012	−0.1423***	0.1569***	−0.1055***	1.0000
CTA	−0.3455***	0.0135	−0.4079***	0.1681***	0.2210***	0.0737***	1.0000
TA	0.0282***	0.1558***	−0.0053	−0.0837***	−0.0191**	−0.0071	−0.0693***	1.0000

Notes: *p < 0.10; **p < 0.05; ***p < 0.01Source: Own calculations upon BilBank 2000 database from ABI (Associazione Bancaria Italiana)

Table 5.

Bank-Specific Factors and credit risk

Regressors	All banks	Cooperative banks	Commercial banks	Popular banks
CRt−1	0.7906*** (0.0282)	0.7841*** (0.0252)	0.7902*** (0.0768)	0.6554*** (0.1559)
Δln(GRLO)t−1	−0.0025*** (0.0008)	−0.0049*** (0.0015)	−0.0018** (0.0008)	−0.0093* (0.0052)
ROAt−1	0.0028 (0.0032)	0.0077 (0.0053)	−0.0040 (0.0063)	0.0191 (0.0279)
ln(ETA)t−1	−0.0036*** (0.0008)	−0.0040*** (0.0012)	−0.0036*** (0.0010)	−0.0046 (0.0031)
ln(LTA)t−1	−0.0034*** (0.0010)	−0.0121*** (0.0019)	−0.0017 (0.0014)	−0.0073 (0.0051)
ln(DL)t−1	−0.0016* (0.0010)	−0.0084*** (0.0011)	−0.0001 (0.0013)	−0.0007 (0.0012)
ln(CTA)t−1	0.0033*** (0.0010)	0.0061*** (0.0013)	0.0027** (0.0013)	0.0076* (0.0038)
ln(TA)t−1	0.0004 (0.0004)	0.0010 (0.0007)	−0.0006 (0.0006)	−0.0041 (0.0032)
Constant	0.0080 (0.0062)	−0.0077 (0.0074)	0.0191* (0.0097)	0.0933* (0.0535)
Observations	13,581	8,732	3,870	979
Period	1994–2015	1994–2015	1994–2015	1994–2015
R2	0.7022	0.7369	0.6426	0.7822
Region fixed effects	Yes	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes	Yes

Notes:

CR is the measure of bank credit risk calculated through non-performing loans to total loans; BANK-SPECIFIC is a vector of bank-specific factors, such as: GRLO: growth of loans, reflecting credit policy; ROA: return on assets, reflecting the profitability of bank; ETA: equity to total assets, capturing the solvency bank and reflecting capital strength of bank; LTA: loans to total assets, capturing the volume of credit market; DL: deposits to loans, reflecting the intermediation cost; CTA: cost of total assets, reflecting a common indicator of bank’s efficiency or volume of intermediation cost; TA: total assets, controlling for size of banks. See Table 1 for more details about the description of the variables; own calculations upon BilBank 2000 database from ABI (Associazione Bancaria Italiana); standard errors in brackets; *p < 0.10, **p < 0.05, ***p < 0.01

Table 6.

Descriptive statistics of variables based on different market structure

Market structures/Variables	CR	GRLO	ROA	ETA	LTA	DL	CTA	TA
MON (N = 3,117)
	0.0193	10676.52	0.0053	0.1340	0.5938	1.0924	0.0712	499086
	0.0212	286223.90	0.0203	0.0506	0.1951	0.7855	0.0507	1604318
DUOP (N = 2,766)
	0.0185	34272.60	0.0059	0.1299	0.6302	3.5151	0.0709	1273917
	0.0207	288778.10	0.0143	0.0502	0.2007	80.4731	0.0501	3823786
OLIG (N = 2,188)
	0.0157	12441.17	0.0052	0.1258	0.6458	1.1759	0.0757	1350776
	0.0197	397681.60	0.0138	0.0451	0.1991	8.5640	0.0530	4314434
COMP (N = 7,196)
	0.0132	183224.40	0.0017	0.1335	0.6130	15.1692	0.0802	8619165
	0.0217	4178454.00	0.0373	0.0690	0.2479	273.8622	0.0647	34900000

Source: Own calculations upon BilBank 2000 database from ABI (Associazione Bancaria Italiana)

Table 7.

Bank-specific factors and credit risk: the role of market structure

	All banks				Cooperative banks				Commercial banks				Popular banks
Regressors	NO MON	NO DUOP	NO OLIG	NO COMP	NO MON	NO DUOP	NO OLIG	NO COMP	NO MON	NO DUOP	NO OLIG	NO COMP	NO MON	NO DUOP	NO OLIG	NO COMP
CRt−1	0.7992*** (0.0375)	0.7706*** (0.0359)	0.7905*** (0.0297)	0.7494*** (0.0297)	0.8016*** (0.0348)	0.7735*** (0.0335)	0.7788*** (0.0305)	0.7765*** (0.0334)	0.7889*** (0.0291)	0.7947*** (0.0292)	0.7919*** (0.0291)	0.7778*** (0.0252)	0.7900*** (0.0290)	0.7845*** (0.0289)	0.8005*** (0.0266)	0.7884*** (0.0286)
Δln(GRLO)t−1	−0.0023*** (0.0008)	−0.0025*** (0.0008)	−0.0027*** (0.0008)	−0.0026** (0.0012)	−0.0022*** (0.0008)	−0.0025*** (0.0008)	−0.0024*** (0.0008)	−0.0023*** (0.0008)	−0.0026*** (0.0008)	−0.0025*** (0.0008)	−0.0029*** (0.0008)	−0.0029** (0.0012)	−0.0024*** (0.0007)	−0.0024*** (0.0007)	−0.0024*** (0.0007)	−0.0025*** (0.0008)
ROAt−1	0.0011 (0.0024)	0.0020 (0.0027)	0.0014 (0.0023)	0.0463*** (0.0142)	0.0016 (0.0025)	0.0022 (0.0028)	0.0026 (0.0028)	0.0069 (0.0080)	0.0028 (0.0029)	0.0027 (0.0028)	0.0019 (0.0026)	0.0067 (0.0056)	0.0028 (0.0029)	0.0031 (0.0030)	0.0030 (0.0029)	0.0028 (0.0029)
ln(ETA)t−1	−0.0038*** (0.0009)	−0.0042*** (0.0009)	−0.0043*** (0.0008)	−0.0014 (0.0011)	−0.0038*** (0.0008)	−0.0037*** (0.0008)	−0.0039*** (0.0008)	−0.0031*** (0.0008)	−0.0037*** (0.0008)	−0.0038*** (0.0008)	−0.0040*** (0.0008)	−0.0031*** (0.0010)	−0.0036*** (0.0008)	−0.0039*** (0.0008)	−0.0035*** (0.0008)	−0.0035*** (0.0008)
ln(LTA)t−1	−0.0032*** (0.0011)	−0.0031*** (0.0011)	−0.0029*** (0.0011)	−0.0071*** (0.0019)	−0.0031*** (0.0011)	−0.0031*** (0.0010)	−0.0030*** (0.0010)	−0.0027** (0.0011)	−0.0035*** (0.0010)	−0.0033*** (0.0010)	−0.0033*** (0.0010)	−0.0073*** (0.0016)	−0.0034*** (0.0010)	−0.0034*** (0.0010)	−0.0033*** (0.0010)	−0.0034*** (0.0011)
ln(DL)t−1	−0.0015 (0.0011)	−0.0012 (0.0010)	−0.0012 (0.0010)	−0.0043*** (0.0013)	−0.0015 (0.0010)	−0.0012 (0.0010)	−0.0014 (0.0010)	−0.0009 (0.0010)	−0.0017* (0.0010)	−0.0016 (0.0010)	−0.0014 (0.0010)	−0.0051*** (0.0011)	−0.0017* (0.0010)	−0.0017* (0.0010)	−0.0017* (0.0010)	−0.0017 (0.0010)
ln(CTA)t−1	0.0034*** (0.0011)	0.0032*** (0.0011)	0.0031*** (0.0011)	0.0042*** (0.0013)	0.0033*** (0.0011)	0.0030*** (0.0010)	0.0032*** (0.0010)	0.0031*** (0.0011)	0.0034*** (0.0010)	0.0035*** (0.0010)	0.0033*** (0.0010)	0.0043*** (0.0010)	0.0033*** (0.0010)	0.0033*** (0.0010)	0.0032*** (0.0010)	0.0033*** (0.0010)
ln(TA)t−1	0.0002 (0.0005)	−0.0000 (0.0005)	0.0003 (0.0005)	0.0012 (0.0008)	0.0002 (0.0005)	−0.0000 (0.0005)	0.0002 (0.0005)	0.0003 (0.0005)	0.0003 (0.0005)	0.0004 (0.0005)	0.0004 (0.0005)	0.0010 (0.0006)	0.0004 (0.0004)	0.0004 (0.0004)	0.0005 (0.0004)	0.0004 (0.0005)
Constant	0.0166** (0.0069)	0.0164** (0.0070)	0.0061 (0.0067)	0.0047 (0.0102)	0.0077 (0.0068)	0.0127* (0.0067)	0.0092 (0.0066)	−0.0037 (0.0057)	0.0109* (0.0062)	0.0078 (0.0064)	0.0075 (0.0067)	0.0024 (0.0086)	0.0120* (0.0064)	0.0110* (0.0063)	0.0058 (0.0059)	0.0120* (0.0065)
Observations Period	10,771 1994–2015	11,079 1994–2015	11,607 1994–2015	7,286 1994–2015	11,311 1994–2015	11,893 1994–2015	12,2 1994–2015	10,197 1994–2015	13,249 1994–2015	13,044 1994–2015	13,09 1994–2015	11,071 1994–2015	13,373 1994–2015	13,306 1994–2015	13,485 1994–2015	13,181 1994–2015
R2	0.6865	0.6811	0.6965	0.7277	0.6977	0.6928	0.6897	0.6923	0.6979	0.6990	0.7035	0.7361	0.6980	0.6969	0.7078	0.6971
Region fixed effect	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes

Notes:

CR is the measure of bank credit risk calculated through non-performing loans to total loans; BANK-SPECIFIC is a vector of bank-specific factors, such as: GRLO: growth of loans, reflecting credit policy; ROA: return on assets, reflecting the profitability of bank; ETA: equity to total assets, capturing the solvency bank and reflecting capital strength of bank; LTA: loans to total assets, capturing the volume of credit market; DL: deposits to loans, reflecting the intermediation cost; CTA: cost of total assets, reflecting a common indicator of bank’s efficiency or volume of intermediation cost; TA: total assets, controlling for size of banks. See Table 1 for more details about the description of the variables; own calculations upon BilBank 2000 database from ABI (Associazione Bancaria Italiana); standard errors in brackets; *p < 0.10, **p < 0.05, ***p < 0.01

Table 8.

Bank-specific factors and credit risk: the role of bank size

Regressors	Excluding minor banks	Excluding small banks	Excluding medium banks	Excluding large banks	Excluding major banks	Minor banks	Small banks
CRt−1	0.7700*** (0.1050)	0.7958*** (0.0295)	0.7880*** (0.0279)	0.7788*** (0.0253)	0.7913*** (0.0284)	0.7820*** (0.0256)	0.6535*** (0.0834)
Δln(GRLO)t−1	−0.0034 (0.0025)	−0.0021*** (0.0007)	−0.0023*** (0.0006)	−0.0024*** (0.0007)	−0.0025*** (0.0008)	−0.0021*** (0.0007)	−0.0026 (0.0021)
ROAt−1	0.0033 (0.0174)	0.0031 (0.0031)	0.0046 (0.0034)	0.0019 (0.0031)	0.0026 (0.0032)	0.0057* (0.0034)	0.0103 (0.0162)
ln(ETA)t−1	−0.0041*** (0.0014)	−0.0036*** (0.0008)	−0.0040*** (0.0008)	−0.0036*** (0.0008)	−0.0035*** (0.0008)	−0.0041*** (0.0010)	−0.0053** (0.0022)
ln(LTA)t−1	−0.0026 (0.0017)	−0.0026* (0.0015)	−0.0036*** (0.0011)	−0.0043*** (0.0008)	−0.0034*** (0.0010)	−0.0069*** (0.0011)	−0.0034** (0.0014)
ln(DL)t−1	0.0003 (0.0015)	−0.0014 (0.0015)	−0.0018 (0.0011)	−0.0027*** (0.0006)	−0.0017* (0.0010)	−0.0053*** (0.0009)	−0.0008 (0.0007)
ln(CTA)t−1	0.0035** (0.0017)	0.0023* (0.0014)	0.0039*** (0.0008)	0.0034*** (0.0010)	0.0034*** (0.0011)	0.0051*** (0.0010)	0.0038** (0.0018)
ln(TA)t−1	−0.0005 (0.0010)	0.0009* (0.0005)	0.0004 (0.0004)	0.0004 (0.0005)	0.0005 (0.0005)	0.0011* (0.0006)	−0.0012 (0.0013)
Constant	0.0162 (0.0166)	0.0027 (0.0080)	0.0093 (0.0061)	0.0122* (0.0069)	0.0072 (0.0064)	0.0045 (0.0090)	0.0134 (0.0164)
Observations Period	3,437 1994–2015	11,379 1994–2015	12,892 1994–2015	13,256 1994–2015	13,360 1994–2015	10,144 1994–2015	2,202 1994–2015
R2	0.6649	0.6937	0.6983	0.7239	0.7004	0.7210	0.7230
Region fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes

Notes:

CR is the measure of bank credit risk calculated through non-performing loans to total loans; BANK-SPECIFIC is a vector of bank-specific factors, such as: GRLO: growth of loans, reflecting credit policy; ROA: return on assets, reflecting the profitability of bank; ETA: equity to total assets, capturing the solvency bank and reflecting capital strength of bank; LTA: loans to total assets, capturing the volume of credit market; DL: deposits to loans, reflecting the intermediation cost; CTA: cost of total assets, reflecting a common indicator of bank’s efficiency or volume of intermediation cost; TA: total assets, controlling for size of banks. See Table 1 for more details about the description of the variables; own calculations upon BilBank 2000 database from ABI (Associazione Bancaria Italiana); standard errors in brackets; *p < 0.10, **p < 0.05, ***p < 0.01

Table 9.

Bank-specific factors and credit risk: the effects of the financial crisis

Regressors	All banks	Cooperative banks	Commercial banks	Popular banks
CRt−1	0.5727*** (0.0442)	0.5903*** (0.0253)	0.6098*** (0.1455)	0.2100 (0.1365)
Δln(GRLO)t−1	−0.0029*** (0.0009)	−0.0059*** (0.0013)	−0.0017* (0.0010)	−0.0094** (0.0047)
ROAt−1	0.0082 (0.0061)	0.0107 (0.0083)	0.0127 (0.0079)	0.0140 (0.0324)
ln(ETA)t−1	−0.0005 (0.0007)	−0.0010 (0.0012)	−0.0001 (0.0009)	−0.0041 (0.0031)
ln(LTA)t−1	−0.0019 (0.0015)	−0.0087*** (0.0021)	−0.0003 (0.0020)	−0.0039 (0.0035)
ln(DL)t−1	0.0003 (0.0014)	−0.0029** (0.0013)	0.0010 (0.0020)	0.0024 (0.0015)
ln(CTA)t−1	0.0024* (0.0013)	0.0055*** (0.0014)	0.0016 (0.0019)	0.0068 (0.0041)
ln(TA)t−1	0.0006 (0.0005)	0.0000 (0.0008)	0.0001 (0.0006)	−0.0027 (0.0022)
Constant	0.0047 (0.0069)	0.0048 (0.0079)	0.0019 (0.0086)	0.0412 (0.0291)
Observations Period	9,755 1994–2008	6,442 1994–2008	2,643 1994–2008	670 1994–2008
R2	0.4759	0.5792	0.3565	0.4210
Region fixed effects	Yes	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes	Yes

Notes:

CR is the measure of bank credit risk calculated through non-performing loans to total loans; BANK-SPECIFIC is a vector of bank-specific factors, such as: GRLO: growth of loans, reflecting credit policy; ROA: return on assets, reflecting the profitability of bank; ETA: equity to total assets, capturing the solvency bank and reflecting capital strength of bank; LTA: loans to total assets, capturing the volume of credit market; DL: deposits to loans, reflecting the intermediation cost; CTA: cost of total assets, reflecting a common indicator of bank’s efficiency or volume of intermediation cost; TA: total assets, controlling for size of banks. See Table 1 for more details about the description of the variables; own calculations upon BilBank 2000 database from ABI (Associazione Bancaria Italiana); standard errors in brackets; *p < 0.10, **p < 0.05, ***p < 0.01

Table 10.

Bank-specific factors and credit risk: the effects of Basel agreements

Regressors	All banks	All banks	Cooperative banks	Cooperative banks	Commercial banks	Commercial banks	Popular banks	Popular banks
	Basel II	Basel III	Basel II	Basel III	Basel II	Basel III	Basel II	Basel III
CRt−1	0.4693*** (0.0335)	0.6357*** (0.0360)	0.5151*** (0.0257)	0.6731*** (0.0235)	0.3791*** (0.0518)	0.6080*** (0.1103)	0.1469 (0.1221)	0.2990** (0.1456)
Δln(GRLO)t−1	−0.0028** (0.0011)	−0.0028*** (0.0008)	−0.0064*** (0.0017)	−0.0059*** (0.0013)	−0.0010 (0.0012)	−0.0018** (0.0009)	−0.0088* (0.0050)	−0.0101** (0.0047)
ROAt−1	−0.0055 (0.0045)	0.0063 (0.0052)	−0.0003 (0.0085)	0.0137 (0.0084)	0.0022 (0.0043)	0.0062 (0.0063)	0.0001 (0.0316)	0.0273 (0.0351)
ln(ETA)t−1	−0.0005 (0.0008)	−0.0014** (0.0007)	−0.0020 (0.0013)	−0.0013 (0.0011)	0.0010 (0.0009)	−0.0018* (0.0010)	−0.0051 (0.0040)	−0.0018 (0.0027)
ln(LTA)t−1	−0.0027** (0.0012)	−0.0027** (0.0012)	−0.0098*** (0.0028)	−0.0091*** (0.0019)	0.0005 (0.0012)	−0.0013 (0.0017)	−0.0014 (0.0042)	−0.0097*** (0.0036)
ln(DL)t−1	−0.0006 (0.0006)	−0.0005 (0.0011)	−0.0034** (0.0016)	−0.0044*** (0.0011)	0.0007 (0.0007)	0.0004 (0.0016)	0.0034 (0.0029)	−0.0018** (0.0009)
ln(CTA)t−1	0.0014* (0.0008)	0.0025** (0.0010)	0.0033*** (0.0013)	0.0052*** (0.0013)	0.0001 (0.0009)	0.0024* (0.0013)	0.0055 (0.0044)	0.0102*** (0.0038)
ln(TA)t−1	0.0003 (0.0006)	0.0006 (0.0004)	−0.0018* (0.0010)	0.0005 (0.0007)	0.0005 (0.0007)	−0.0001 (0.0006)	−0.0031 (0.0024)	−0.0024 (0.0023)
Constant	0.0045 (0.0070)	−0.0068 (0.0047)	0.0180* (0.0100)	0.0159* (0.0086)	−0.0048 (0.0089)	0.0187** (0.0087)	0.0600 (0.0372)	0.0431 (0.0320)
Observations Period	7,835 1994–2005	11,692 1995–2011	5,189 1994–2005	7,642 1995–2011	2,091 1994–2005	3,251 1995–2011	555 1994–2005	799 1995–2011
R2	0.5416	0.5377	0.6070	0.6253	0.4678	0.4224	0.4082	0.5089
Region fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes

Notes: CR is the measure of bank credit risk calculated through non-performing loans to total loans; BANK-SPECIFIC is a the vector of bank-specific factors, such as: GRLO: growth of loans, reflecting credit policy; ROA: return on assets, reflecting the profitability of bank; ETA: equity to total assets, capturing the solvency bank and reflecting capital strength of bank; LTA: loans to total assets, capturing the volume of credit market; DL: deposits to loans, reflecting the intermediation cost; CTA: cost of total assets, reflecting a common indicator of bank’s efficiency or volume of intermediation cost; TA: total assets, controlling for size of banks. See Table 1 for more details about the description of the variables; own calculations upon BilBank 2000 database from ABI (Associazione Bancaria Italiana); standard errors in brackets; *p < 0.10, **p < 0.05, ***p < 0.01

Table 11.

Bank-specific factors and credit risk: changing dependent variable (Z-SCORE)

Regressors	All banks	Cooperative banks	Commercial banks	Popular banks
ln(Z.SCORE-ROA)t−1	0.1291*** (0.0107)	0.3639*** (0.0887)	0.5010*** (0.0686)	0.5246*** (0.0723)
Δln(GRLO)t−1	−0.0443 (0.0330)	−0.1321*** (0.0335)	−0.0728** (0.0321)	−0.2339*** (0.0644)
ROAt−1	0.6131 (0.4467)	−0.3136 (0.5202)	0.0365 (0.7505)	−1.3444 (0.8905)
ln(ETA)t−1	0.6413*** (0.0331)	0.3488*** (0.0929)	0.1133 (0.0779)	0.1153* (0.0632)
ln(LTA)t−1	0.0954** (0.0381)	0.0694** (0.0282)	0.0141 (0.0388)	0.1361 (0.0876)
ln(DL)t−1	0.0230 (0.0232)	0.0775*** (0.0161)	0.0076 (0.0252)	−0.0140 (0.0325)
ln(CTA)t−1	−0.0748** (0.0355)	−0.0189 (0.0278)	0.0036 (0.0375)	−0.1128 (0.0895)
ln(TA)t−1	0.0239 (0.0228)	−0.0076 (0.0149)	−0.0524** (0.0223)	−0.0648 (0.0430)
Constant	3.4162*** (0.3149)	2.9679*** (0.5428)	1.7716*** (0.4605)	2.5444*** (0.9035)
Observations Period	13,528 1994–2015	8,726 1994–2015	3,824 1994–2015	978 1994–2015
R2	0.7022	0.9755	0.7864	0.9188
Region fixed effects	Yes	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes	Yes

Notes: Z-SCORE is the measure of bank stability (see equation in Section 4.2); BANK-SPECIFIC is a vector of bank-specific factors, such as: GRLO: growth of loans, reflecting credit policy; ROA: return on assets, reflecting the profitability of bank; ETA: equity to total assets, capturing the solvency bank and reflecting capital strength of bank; LTA: loans to total assets, capturing the volume of credit market; DL: deposits to loans, reflecting the intermediation cost; CTA: cost of total assets, reflecting a common indicator of bank’s efficiency or volume of intermediation cost; TA: total assets, controlling for size of banks. See Table 1 for more details about the description of the variables; own calculations upon BilBank 2000 database from ABI (Associazione Bancaria Italiana); standard errors in brackets; *p < 0.10, **p < 0.05, ***p < 0.01

Table 12.

Bank-specific factors and credit risk: the potential differences inside LMAs

Regressors	All banks- Excluding 1st decile	All banks- Excluding 10th decile	Cooperative banks- Excluding 1st decile	Cooperative banks- Excluding 10th decile	Commercial banks-Excluding 1st decile	Commercial banks-Excluding 10th decile	Popular banks-Excluding 1st decile	Popular banks-Excluding 10th decile
CRt−1	0.7711*** (0.0265)	0.8110*** (0.0291)	0.7771*** (0.0269)	0.7802*** (0.0259)	0.7338*** (0.0596)	0.9029*** (0.0930)	0.6535*** (0.1593)	0.6338*** (0.1578)
Δln(GRLO)t−1	−0.0025*** (0.0008)	−0.0019** (0.0007)	−0.0061*** (0.0017)	−0.0039*** (0.0012)	−0.0015* (0.0009)	−0.0007 (0.0009)	−0.0099* (0.0057)	−0.0080 (0.0052)
ROAt−1	0.0009 (0.0024)	0.0054 (0.0044)	0.0040 (0.0032)	0.0047 (0.0034)	−0.0118 (0.0089)	0.0175 (0.0149)	0.0330 (0.0476)	0.0229 (0.0474)
ln(ETA)t−1	−0.0033*** (0.0008)	−0.0030*** (0.0007)	−0.0041*** (0.0013)	−0.0039*** (0.0011)	−0.0030*** (0.0011)	−0.0022** (0.0008)	−0.0049 (0.0033)	−0.0035 (0.0030)
ln(LTA)t−1	−0.0037*** (0.0008)	−0.0031*** (0.0012)	−0.0131*** (0.0022)	−0.0117*** (0.0019)	−0.0019** (0.0009)	−0.0007 (0.0018)	−0.0075 (0.0057)	−0.0059 (0.0050)
ln(DL)t−1	−0.0023*** (0.0006)	−0.0013 (0.0012)	−0.0087*** (0.0013)	−0.0082*** (0.0011)	−0.0009 (0.0006)	0.0010 (0.0019)	−0.0007 (0.0013)	−0.0003 (0.0012)
ln(CTA)t−1	0.0028** (0.0011)	0.0035*** (0.0009)	0.0055*** (0.0014)	0.0062*** (0.0013)	0.0021 (0.0014)	0.0025** (0.0013)	0.0081** (0.0040)	0.0070* (0.0040)
ln(TA)t−1	0.0002 (0.0005)	0.0009* (0.0005)	0.0004 (0.0009)	0.0014* (0.0007)	−0.0005 (0.0007)	−0.0000 (0.0007)	−0.0049 (0.0036)	−0.0042 (0.0032)
Constant	0.0162** (0.0072)	0.0061 (0.0071)	0.0151 (0.0104)	0.0045 (0.0091)	0.0212* (0.0110)	0.0109 (0.0107)	0.1187* (0.0623)	0.1106* (0.0593)
Observations	12,163	12,440	7,625	8,442	3,612	3,085	926	913
Period	1994–2015	1994–2015	1994–2015	1994–2015	1994–2015	1994–2015	1994–2015	1994–2015
R2	0.7339	0.7163	0.7454	0.7331	0.7231	0.7008	0.7796	0.7753
Region fixed effects	YES	YES	YES	YES	YES	YES	YES	YES
Time fixed effects	YES	YES	YES	YES	YES	YES	YES	YES

Notes: CR is the measure of bank credit risk calculated through non-performing loans to total loans; BANK-SPECIFIC is a the vector of bank-specific factors, such as: GRLO: growth of loans, reflecting credit policy; ROA: return on assets, reflecting the profitability of bank; ETA: equity to total assets, capturing the solvency bank and reflecting capital strength of bank; LTA: loans to total assets, capturing the volume of credit market; DL: deposits to loans, reflecting the intermediation cost; CTA: cost of total assets, reflecting a common indicator of bank’s efficiency or volume of intermediation cost; TA: total assets, controlling for size of banks. See Table 1 for more details about the description of the variables; own calculations upon BilBank 2000 database from ABI (Associazione Bancaria Italiana); standard errors in brackets; *p < 0.10, **p < 0.05, ***p < 0.01

Table 13.

Bank-specific factors and credit risk: the potential differences inside LMAs

Regressors	All banks-Excluding 1st decile	All banks-Excluding 10th decile	Cooperative banks-Excluding 1st decile	Cooperative banks-Excluding 10th decile	Commercial banks-Excluding 1st decile	Commercial banks-Excluding 10th decile	Popular banks-Excluding 1st decile	Popular banks-Excluding 10th decile
ln(Z.SCORE-ROA)t−1	0.1226*** (0.0114)	0.1379*** (0.0118)	0.3658*** (0.0732)	0.3914*** (0.0707)	0.4727*** (0.0727)	0.5322*** (0.0819)	0.5326*** (0.0716)	0.5252*** (0.0669)
Δln(GRLO)t−1	−0.0314 (0.0342)	−0.0497 (0.0379)	−0.1212*** (0.0350)	−0.1627*** (0.0262)	−0.0661** (0.0326)	−0.0485 (0.0395)	−0.2433*** (0.0678)	−0.2005*** (0.0599)
ROAt−1	0.3423 (0.3982)	0.4055 (0.4748)	−0.4405* (0.2665)	−0.4764* (0.2696)	0.0908 (0.9750)	1.3351 (0.8898)	−1.1280 (1.2063)	−2.0820 (1.3222)
ln(ETA)t−1	0.6447*** (0.0359)	0.6535*** (0.0354)	0.3513*** (0.0793)	0.3227*** (0.0751)	0.1217 (0.0819)	0.1250 (0.0883)	0.1097* (0.0622)	0.1356** (0.0626)
ln(LTA)t−1	0.1349*** (0.0440)	0.0719 (0.0464)	0.0930*** (0.0294)	0.0514** (0.0256)	0.0077 (0.0380)	0.0342 (0.0468)	0.1090 (0.0949)	0.1540 (0.0946)
ln(DL)t−1	0.0440* (0.0247)	0.0177 (0.0255)	0.0922*** (0.0181)	0.0755*** (0.0158)	0.0060 (0.0247)	0.0334 (0.0265)	−0.0035 (0.0255)	−0.0073 (0.0301)
ln(CTA)t−1	−0.0788** (0.0378)	−0.0143 (0.0411)	−0.0160 (0.0297)	−0.0169 (0.0270)	0.0056 (0.0344)	−0.0094 (0.0474)	−0.0743 (0.0773)	−0.1086 (0.0935)
ln(TA)t−1	0.0304 (0.0250)	0.0386* (0.0231)	−0.0053 (0.0176)	−0.0162 (0.0148)	−0.0552** (0.0235)	−0.0427* (0.0231)	−0.0569 (0.0483)	−0.0694 (0.0443)
Constant	3.3918*** (0.3395)	3.4439*** (0.3338)	3.4052*** (0.4325)	3.2814*** (0.4090)	1.8867*** (0.4801)	1.4792*** (0.5427)	1.9740** (0.9466)	2.1520** (0.9183)
Observations Period	12,121 1994–2015	12,415 1994–2015	7,619 1994–2015	8,436 1994–2015	3,577 1994–2015	3,067 1994–2015	925 1994–2015	912 1994–2015
R2	0.6905	0.7394	0.9752	0.9755	0.7947	0.8307	0.9264	0.9259
Region fixed effects	YES	YES	YES	YES	YES	YES	YES	YES
Time fixed effects	YES	YES	YES	YES	YES	YES	YES	YES

Notes: Z-SCORE is the measure of bank stability (see equation in Section 4.2); BANK-SPECIFIC is a vector of bank-specific factors, such as: GRLO: growth of loans, reflecting credit policy; ROA: return on assets, reflecting the profitability of bank; ETA: equity to total assets, capturing the solvency bank and reflecting capital strength of bank; LTA: loans to total assets, capturing the volume of credit market; DL: deposits to loans, reflecting the intermediation cost; CTA: cost of total assets, reflecting a common indicator of bank’s efficiency or volume of intermediation cost; TA: total assets, controlling for size of banks. See Table 1 for more details about the description of the variables; own calculations upon BilBank 2000 database from ABI (Associazione Bancaria Italiana); standard errors in brackets; *p < 0.10, **p < 0.05, ***p < 0.01

Notes

1. Unfortunately, we do not have information on some of the variables used in the analysis for years before 1994 and after 2015. For this reason, our analysis only covers the 1994–2015 period. Furthermore, the ABI-data set is compared with Bankscope-data set. The debate is in favour of the first because it has some valuable information.

2. For the purposes of the paper and given the structure of our data, we have opted for the application of a pooled regression, i.e., fixed or random effect. The first step is to determine whether individual effects are relevant or not. If the individual effects are not relevant, the pooled regression model is the best option. The Breusch–Pagan test (Breusch and Pagan, 1980) was performed. Contrast results suggest that the constant coefficient model was not suitable for the data. The second step is to choose between the fixed-effects and the random-effects models. The Hausman test provides an answer for the best choice (Hausman 1978; Hausman and McFadden 1984). The null hypothesis assumes that the most appropriate model is the random-effects, with the fixed-effect model being an alternative hypothesis. Test results rejected the null hypothesis, thus concluding that the best estimation technique for the data was provided by the model of fixed coefficients.

3. For a more detailed discussion concerning the properties of the GMM estimator, please refer to Roodman (2009).

4. The choice of the lag structure is suggested by some preliminary estimates.

5. More specifically, the Italian regions are divided as follows: South (Abruzzo, Molise, Campania, Puglia, Basilicata, Calabria, Sicilia and Sardegna), North‐West (Valle d'Aosta, Lombardia, Piemonte and Liguria), North‐East (Trentino Alto Adige, Friuli Venezia Giulia, Veneto and Emilia‐Romagna) and finally Centre (c—Toscana, Marche, Umbria and Lazio).

6. As suggested by Roy (1952), the indicator of financial stability corresponds to the inverse of the probability of default, and it’s considered in literature as one of the main indicators to quantify the financial stability in the banking sector (see, among others, Boyd and Graham, 1986, 1988; Boyd and Runkle, 1993; Maechler et al., 2005; Laeven and Levine, 2009; Uhde and Heimeshoff, 2009; Fink et al., 2009; Demirguc-Kunt and Huizinga, 2010; Houston et al., 2010; Beck et al., 2012; Fiordelisi and Mare, 2014).