The Home Bias and the Credit Crunch: A Regional Perspective

Transcription

1 The Home Bias and the Credit Crunch: A Regional Perspective Andrea F. Presbitero Gregory F. Udell Alberto Zazzaro February 24, 2012 Abstract A major policy issue is whether troubles in the banking system reflected in the bankruptcy of Lehman Brothers in September 2008 have spurred a credit crunch and, if so, how and why its severity has been different across markets and firms. In this paper, we tackle this issue by looking at the Italian case. We take advantage of a dataset on a large sample of manufacturing firms, observed quarterly between January 2008 and September Thanks to detailed information about loan applications and lending decisions, we are able to identify the occurrence of a credit crunch in Italy which has been found to be harsher in provinces with a large share of branches owned by distantly-managed banks. Inconsistent with the flight to quality hypothesis, however, we do not find evidence that economically weaker and smaller firms suffered more during the crisis period than during tranquil periods. By contrast, we find that large and healthy firms, the segment of borrowers which, according to theoretical predictions, are cream-skimmed by distantly-headquartered banks, were more intensely hit by the credit tightening in functionally distant credit markets than in the ones populated by less distant banks. This last result is consistent with the hypothesis of a home bias on the part of nationwide banks. JEL Classification: F33, F34, F35, O11 Key words: Banking; Credit crunch; Distance; Home bias; Flight to quality. Andrea F. Presbitero (corresponding author), Department of Economics Università Politecnica delle Marche (Italy), Money and Finance Research group (MoFiR) and Centre for Macroeconomic and Finance Research (Ce- MaFiR). a.presbitero@univpm.it; personal web page: Gregory F. Udell, Indiana University, Kelley School of Business and Money and Finance Research group (MoFiR). E- mail: gudell@indiana.edu Alberto Zazzaro, Department of Economics Università Politecnica delle Marche (Italy) and Money and Finance Research group (MoFiR). a.zazzaro@univpm.it; personal web page: We thank Raoul Minetti and the participants at the MoFiR workshop on banking (Ancona, 2012) and at seminars held at the Università di Milano Bicocca and Università Politecnica delle Marche for valuable suggestions. 1

2 1 Introduction The financial crisis that began in the third quarter of 2007 originated with the bursting of a real estate bubble in the US and hit its peak in the quarters immediately after the collapse of Lehman Brothers in September This led to massive capital shocks to the US banking system that quickly propagated to Europe as global interbank loan markets seized up. The contagion from the US shock was subsequently exacerbated by Europe s own problems in the real estate sector in countries like Ireland and Spain compounded by sovereign debt problems particularly in the southern Euro zone. One of the most feared and debated consequences of the crisis in both Europe and the US has been the possible credit crunch caused by the contraction of banks capital and the adverse liquidity shocks in interbank markets. However, identifying the existence of a credit crunch during a global crisis, disentangling the shrinking of credit supply from the parallel reduction in credit demand, and distinguishing the factors that may have driven differences in the severity of the crunch across firms and markets are major concerns to policymakers and one of the biggest challenges facing empirical work. In the absence of unusual natural experiments that create an easily identifiable supply shock (e.g. Khwaja and Mian; 2008; Peek and Rosengren; 1997) several identification strategies have been employed in the literature. One strategy is to exploit credit registry data on firms that have multiple lenders in order to control for demand effects (e.g. Albertazzi and Marchetti; 2010; Iyer et al.; 2010; Jiménez et al.; 2011; Gobbi and Sette; 2012). Another approach is to apply a disequilibrium model to identify credit constrained firms (e.g. Carbò-Valverde, Rodriguez-Fernandez and Udell; 2011; Kremp and Sevestre; 2011). An alternative approach to identify constrained firms, that we will follow in this paper, is to use survey data that contain information on loan applications and bank decisions (e.g. Popov and Udell; 2012; Winston Smith and Robb; 2011; Ferrando and Mulier; 2011; Puri et al.; 2011). A number of studies have analyzed the effects on the credit markets in the country where the crisis began (i.e., the US). These studies have found evidence of significant shocks to the supply of credit by large and small banks (e.g. Contessi and Francis; 2010; Gozzi and Goetz; 2010; Ivashina and Scharfstein; 2010; Santos; 2010). However, missing from the research on the impact of the credit crunch in the US is an analysis of the impact across different categories of borrowers and regions. Virtually all of this research on credit in the US during the crisis either focuses on large firms the least likely to be affected by the crunch or on indirect evidence such as the Federal Reserve s Senior Loan Officer Survey (e.g. Udell; 2009) 1. As a consequence of data limitations in the US 2, firm level analysis of the effect of the current crisis on small and medium enterprises (SMEs) has been substantially limited to Europe. In general these studies have confirmed a credit crunch in the European credit markets (e.g. Albertazzi and Marchetti; 2010; Carbò-Valverde, Degryse and Rodriguez-Fernandez; 2011; Carbò-Valverde, Rodriguez- Fernandez and Udell; 2011; Ferrando and Mulier; 2011; Iyer et al.; 2010; Jiménez et al.; 2011; Puri et al.; 2011). The evidence also suggests that younger, smaller and informationally more opaque firms may have been more severely affected (e.g. Artola and Genre; 2011; Canton et al.; 1 One exception is a study of how start-up firms faired during the crisis (Winston Smith and Robb; 2011). This study used the Kaufman Firm Survey and was confined to very young and very small firms. Examples of indirect evidence include a study of how large firms supplied trade credit during the crisis, some of which likely went to small firms (Garcia-Appendini and Montoriol-Garriga; 2011), and a study by Gozzi and Goetz (2010) who show that metropolitan areas where banks relied less on retail deposits experienced a more severe economic downturn during the crisis. 2 Unlike many European countries the US does not have a public credit registry. In addition, the best available firm level data on SME finance in the US, the Federal Reserve s Survey of Small Business Finance (SSBF), was discontinued just before the crisis began. While the SSBF data were not panel data, they did contain extensive data on firm characteristics, financial statements and loan terms. Moreover, the next survey would have been conducted in the middle of the crisis, had it not been discontinued. 2

3 2011; Holton et al.; 2011; Popov and Udell; 2012). Our paper adds to this growing empirical literature on the determinants of the credit crunch in two ways. First, we explore whether and how the hierarchical structure of banks in the local market affects the severity of the credit crunch in that market. Second, we look deeper into the question of which type of firms are more exposed to credit tightening by investigating the common conjecture that small and risky firms suffer most if operating in credit markets largely populated by nationwide, distantly-managed banks rather than by local banks. Our specific focus on the hierarchical structure of the local banking market centers on the issue of whether borrowers whose banks are less local are more vulnerable. The theoretical and empirical literature on commercial lending suggests that hierarchical banks are less able to provide relationship lending to SMEs because of difficulties associated with producing and transmitting soft information (Stein; 2002; Berger et al.; 2005; Liberti and Mian; 2009). This implies that as the functional distance between the loan officer and the headquarters where final lending decisions are made increases, banks are less able to make relationship-based loans and access to credit to local firms becomes tighter (Alessandrini et al.; 2009). In this paper, we explore this issue by conjecturing that, in times of crisis, banks retract disproportionally from markets which are distant from their headquarters. If this actually occurrs, then the adverse effect of functional distance on firms access to credit should be observed to be more pronounced in the months following the collapse of Lehman Brothers. In addition, we investigate whether the withdrawal of banks from local markets is the result of a flight to quality or a home bias effect. To establish which of the two effects prevails, we test whether more small and risky enterprises in more functionally distant banking systems are more (flight to quality) or less (home bias) likely to suffer from a contraction of credit after Lehman s collapse. Our study is closely related to studies that have examined the foreign ownership of banks and whether shocks to parent banks are propagated across borders affecting the lending activities of their foreign operations (e.g. Cetorelli and Goldberg; 2011; Popov and Udell; 2012). A few recent contributions have considered the existence of a home bias in banks lending reactions to adverse shocks to their own financial conditions at times of global crisis, by looking at the behavior of international banks in syndicated loan market (Galindo et al.; 2010; de Haas and van Horen; 2011; Giannetti and Laeven; 2011). In this paper, we take a national perspective by studying the credit crunch in Italian provinces during the present financial crisis. To this end, we exploit detailed survey information on loan applications and their outcome concerning a large sample of manufacturing firms. This allows us to separate demand and supply effects, and to identify the existence and severity of credit crunch across firms and markets. In particular, we merge firm-level data with information on the spatial distribution of bank branches in order to assess the effect of the organizational structure of the local banking systems on access to credit to local firms. Data availability apart, Italy, like many other countries in Europe and elsewhere, is characterized by a large number of small firms which are strongly dependent on loans from local banks to finance their investments and business activity, and by a number of nationwide banks which spread their subsidiaries and branches to provinces at a great distance from the home province where they are headquartered. This makes Italy a representative case study, having broader implications for the analysis of the current credit crunch and the relevance of the home bias effect in shaping the supply of loans. By way of preview we find that the shock to global liquidity surrounding the Lehman collapse was transmitted to the real sector in Italy in terms of a significant contraction in both demand for and supply of credit. Hence, we find evidence that a credit crunch occurred in Italy. However, inconsistent with the flight to quality hypothesis that the credit crunch has been significantly 3

4 more severe for small and economically weaker firms than for large and good-quality ones, we show that the likelihood of the former being credit-rationed during the crisis period was not significantly higher than in tranquil periods. By contrast, we find robust evidence that the penetration of functionally distant banks in local credit markets exacerbated the credit crunch. However, our results reject the hypothesis of a flight to quality in functionally distant credit markets, while they are consistent with the hypothesis of a home bias on the part of nationwide banks. In fact, we find that the contraction of credit supply has targeted large and healthy firms the ones that, according to theory, are likely to borrow from distantly-headquartered banks relatively more in functionally distant credit markets than in the ones populated by banks functionally close to the local economy. The remainder of the paper is structured as follows. In the next section we offer a brief review of the related literature. In sections 3 and 4 we present our data and variables and discuss our identification strategy. In section 5 we provide a descriptive analysis of the credit crunch, while sections 6 and 7 present the results of our model estimations and the robustness exercises. Section 8 offers a discussion of our findings and a conclusion. 2 Functionally distant banks, home bias and access to credit Our paper builds on the literature that has analyzed the link between banks operating in functionally distant local credit markets and firms access to credit. There are several reasons why the presence of subsidiaries and branches of (foreign or domestic) banks headquartered at a geographical distance may adversely influence the availability of credit to local firms, resulting in a home bias. These reasons have to do with: (i) asymmetric information and agency costs; (ii) internal capital markets and corporate politics. 2.1 Asymmetric information and agency costs The existence of information asymmetries between the bank and the firm makes lending a very local activity. 3 A crucial part of information about the firm s creditworthiness is soft and socially embedded. As a result, it can be conveniently recovered and processed only by loan officers working and living in the same neighborhoods where the borrowers operate, but they can also be only imperfectly (and at some cost) transmitted to the senior managers at the upper layers of the parent bank. Accordingly, loan officers benefit from informational rents and banks bear agency costs in order to align the interests of the former with those of bank shareholders and mitigate moral hazard in communication (Agarwal and Wang; 2009; Agarwal and Hauswald; 2010; Hertzberg et al.; 2010; Uchida et al.; 2012). The fact is that the more hierarchically organized and (physically and culturally) distant from the local economy is the parent bank, the greater the shortfalls in communication channels (Stein; 2002). For example, costs and uncertainty of loan reviews increase with physical distance from the bank s headquarters where loan reviewers report (Udell; 1989), as well as trust between bank s managers and local loan officers tends to be lower when the cultural distance between the geographical areas where the staff of the bank s decisional centres and local offices work and live is great (Cremer et al.; 2007). For such reasons, distant banks have an incentive to constrain local branches from lending to soft-information-intensive borrowers, such as small and innovative enterprises (Dell Ariccia and Marquez; 2004). Similarly, career-concerned loan officers can be induced to assume a 3 With regard to lines of credit, in the US, the median distance between the firm and the lending bank branch was 4 miles in 1993 and 3 miles in 2003 (Brevoort and Wolken; 2009), while it was still lower in Europe: 1.58 miles in Italy (Bellucci et al.; 2010), 1.4 in Belgium (Degryse and Ongena; 2005) and 0.62 in Sweden (Carling and Lundberg; 2005). 4

5 conservative attitude towards loans to small and new borrowers based on soft, uncommunicable information and a too liberal approach towards large and well-established borrowers who are evaluated with hard, easily transferable information (Hirshleifer and Thakor; 1992; Berger and Udell; 2002). To the extent that the cost of funds is lower for branches of large and functionally distant banks, local credit markets tend to segment, with distant banks cream-skimming highreturn, informationally transparent borrowers, and local banks specializing in lending to softinformation borrowers. Market segmentation may result in higher overall lending to the economy or in extensive credit rationing to small firms, according to whether local banks are more or less efficient at screening small, opaque firms and the average quality of such firms is high or low (Dell Ariccia and Marquez; 2004; Sengupta; 2007; Detragiache et al.; 2008; Gormley; 2011). Consistent with theoretical predictions, empirical evidence shows that branches and subsidiaries of functionally distant banks tend to be less engaged in loans to small businesses and other soft-information-based investment projects, have a comparative disadvantage in relationship lending, ask for lower collateral, have a lower share of bad loans, are less prone to assist firms facing financial distress and are less efficient (Berger et al.; 2001; Berger and DeYoung; 2001; Mian; 2006; Alessandrini et al.; 2008; DeYoung et al.; 2008; Jiménez et al.; 2009; Micucci and Rossi; 2010). In addition, a number of studies provide evidence in support of the hypothesis that firms in markets populated by functionally distant banks have, on average, lower access to credit. Detragiache et al. (2008) look at poor countries and find that the total amount of loans to the private sector (normalized to GDP) and the rate of credit growth are negatively correlated with the foreign bank penetration (measured by the share of bank assets owned by foreign banks). The negative impact of foreign banks is confirmed by Gormley (2010), who documents that in India firms in districts with a foreign bank, especially if they are small sized and endowed with low tangible assets, have a significantly lower probability of obtaining long-term loans. Using Italian data, Alessandrini et al. (2009, 2010) show that firms are more likely to be financially constrained and less inclined to introduce innovations if they are located in provinces where a large share of branches belong to banks headquartered in physically distant provinces and in provinces with different social and economic environments. Similar findings with regard to France are documented by Djedidi (2010), while Presbitero and Zazzaro (2011), again with Italian data, find that in highly competitive markets the presence of functionally distant banks is detrimental to relationship lending. Finally, Özyildirim and Önder (2008) show that in Turkish provinces whose bank branches are distant from their headquarter the credit-to-gdp ratio has a low or even negative impact on local growth, suggesting that local branches of distant banks tend on average to fund less profitable projects. 2.2 Internal capital markets and corporate politics The existence of an internal capital market has contrasting effects on lending to local firms by branches of banks headquartered at a distance (Morgan et al.; 2004). On the one hand, capital inflows from parent banks allow branches and affiliate banks to promptly increase lending in response to a boom in the local economy (de Haas and van Lelyveld; 2010) and to be partly insulated from idiosyncratic liquidity shocks (Houston et al.; 1997; Dahl et al.; 2002). On the other hand, by having the opportunity to move funds across regions, multi-market banks (whether foreign or nationwide) may transmit financial shocks from one economy to another (Peek and Rosengren; 1997, 2000; Berrospide et al.; 2011; Imai and Takarabe; 2011; Schnabl; 2011; Cetorelli and Goldberg; 2011) and may be more inclined to reduce local lending when the local economy and deposit growth slow down (Campello; 2002; Cremers et al.; 2010). 5

6 However, capital allocation and the internal liquidity flows across bank branches and affiliates are only partially driven by lending opportunities, while they are affected by corporate politics, by the power and influence of local managers inside the organization and by the economic, social and political importance of the local economy for the CEOs and other top managers at the bank s headquarter (Meyer et al.; 1992; Scharfstein and Stein; 2000). Once again, the headquarter-tobranch distance tends to affect both the local managers ability to attract internal resources and the bank s favoritism attitude towards local economy needs (the home bias). For example, it is reasonable to assume that the managers of branches located at a great distance from the center have little power to attract internal resources and influence capital budgeting decisions (Carlin et al.; 2006). In the same vein, the more physically and culturally close the bank s headquarters and top management are to a region, the greater is the incentive to favor the local economy and local firms (Landier et al.; 2009). The feeble links of functionally distant banks with the local economic community and the weight of the home bias and corporate politics for internal capital allocation can be thought to become more pronounced in times of global crisis when the amount of loanable funds is lower. Consistently, Giannetti and Laeven (2011) find that the portfolio share of syndicated loans issued by a bank in the home country is larger than that issued in foreign countries, and that the home bias tends to significantly increase in periods when the home country undergoes a banking crisis. At the same time, when it is the host banking system to experience a crisis, foreign banks contract loans to local borrowers less than domestic banks, but to a much smaller extent than when they face negative shocks at home. Similarly, de Haas and van Horen (2011) find that during the financial crisis international banks participating in cross-border syndicated loans reduced their lending exposure to countries far away from their headquarter more than their exposure to geographically close countries. Galindo et al. (2010) find instead some evidence to suggest that also cultural distance matters. They show that, while during the global crisis foreign banks generally amplified external financial shocks in Latin American countries, contracting credit more than domestic banks, Spanish banks behaved similarly to domestic banks. In the same vein, and with reference to the same period, Aiyar (2011) shows that foreign subsidiaries and branches in the United Kingdom decreased their lending to local businesses by a larger amount than domestically owned banks. Likewise, using data on emerging eastern European countries, Popov and Udell (2012) find that small firms located in cities where the majority of lending banks are headquartered abroad are more likely to be creditrationed, especially if banks in the area are financially distressed, while de Haas et al. (2011) find that foreign banks reduced their loan supply to local firms earlier and faster than domestic banks. Finally, Gambacorta and Mistrulli (2011) show that during the global crisis Italian firms borrowing from distant banks experienced a larger increase in interest rates and decrease in loan supply, while Barboni and Rossi (2012) find that firms mainly borrowing from local banks were less credit rationed. 3 Data and variables 3.1 Data sources and the construction of the dataset We draw on data concerning firms financial conditions and the geographical distribution of bank branches from two sources: 1) a monthly survey of about 3,800 Italian manufacturing firms, interviewed from March 2008 to February 2010 by the ISAE (Institute of Studies and Economic Analysis), recently becoming part of the ISTAT (Italian Institute of Statistics); and 2) the monthly data on bank branch openings and closures compiled by the Bank of Italy. The ISAE-ISTAT survey data cover Italian firms with at least five employees (the average 6

7 size is 74 employees, while the median is 18) and was recently updated and re-engineered with the main aim of increasing the comparability of the Italian data with those released by the other European institutions, such as the Ifo Business Climate Survey, while still maintaining a focus on the traditional sectors of Italian specialization (Malgarini et al.; 2005). A specific section of the survey was added in March 2008 to cover bank-firm relationships, providing useful information to distinguish firms demand for credit and the banks rationing decisions. Additional data include information on the firm s industrial sector, labor costs, status of liquidity and the level of (domestic and foreign) orders, production and unsold goods. Since data on the openings and closures of branches are at the bank-province level, it is possible to calculate the stock of branches per bank per province in any quarter. These data are complemented with information on bank governance (i.e. mutual, cooperatives, listed) and asset size (large, medium, small), on the location of their headquarters and holding company structure (when applicable). Considering that we only have the ISAE survey data for the months of March, June, September and December and excluding observations with missing values and outliers, we end up with an unbalanced panel made by 3,623 firms and 23,140 observations, observed quarterly between 2008:1 to 2009:3. Within this dataset, we distinguish two main periods: the pre-lehman period (P RE LEHMAN), from 2008:1 to 2008:3, and the post-lehman (P OST LEHMAN), from 2008:4 to 2009:3. In the following subsections we will describe in detail the data and the construction of the variables (see also Appendix A). 3.2 Firm-specific variables Access to credit The two main dependent variables regarding firms access to bank credit distinguish the demand and supply of credit. DEMAND is an indicator variable which assumes value one for firms which report direct contacts with one or more banks in the previous quarter in order to seek credit (i.e., we exclude firms stating that they just went to the bank to ask for information). RAT IONED, a variable observed only for firms which applied for credit in the given quarter, is a dummy variable which is equal to one for firms which stated they did not obtain the desired amount of bank credit. Using information on the demand for and supply of credit we build a variable measuring the financial health of firms in the pre-lehman period. P RE LEHMAN RAT IONED is a time-invariant dummy variable observed only in the pre-lehman quarters which is equal to one for firms which were quantity-rationed at least in one quarter in the pre-lehman period (RAT IONED = 1) and zero for firms which were not rationed (RAT IONED = 0) or nonapplicants (DEMAND = 0). In other words, this variable identifies firms which had problems in accessing bank credit in the tranquil period, compared to firms which either obtained the desired amount of credit or did not apply for a bank loan between 2008:1 and 2008: Other firms characteristics The survey provides other useful information about firms characteristics, their economic activity and financial condition that allows us to take into account several factors that might influence the demand for credit by firms and the availability of banks to satisfy such demand. First, we consider the firms size (SIZE), the capacity to operate abroad (EXP ORT ) and the labor costs they incur (LABOR COST ). SIZE is measured, for each period, by the logarithm of the average number of employees in each quarter. Enterprises with no more than 7

8 20 employees are identified as small firms (SM ALL). The export status is measured, in each quarter, by a dummy variable equal to one for firms which sold part of the production abroad (EXP ORT ). As a measure of the cost of production, we can take advantage of a specific question asking for the percentage change in labor cost per employee in the previous 12 months (LABOR COST ). Second, we build variables measuring the trend of the economic activity of the surveyed firms and their financial condition. In particular, the level of production and potential demand (LOW DEMAND) is measured by a dummy variable that assumes the value 1 for firms that answer low to the question How are the level of orders and the general demand for products? and zero otherwise (i.e., for firms answering normal or high ). Similarly, firms financial health is captured by three dummy variables which are constructed on the basis of a question about the level of liquidity with respect to operational needs, which respondents can evaluate as good, neither good nor bad, or bad (LIQUIDIT Y ). Finally, the industry a firm belongs to and its location in the South of Italy 4 are taken into account for the possible effect that a specific industry or geographic location could have upon access to credit. 3.3 Credit market variables We match firm-level data with aggregate indicators of the structure of local credit markets, defined at provincial level. 5 First, we consider the organizational structure of the local banking systems as proxied by the headquarter-to-branch functional distance (DIST ANCE). Following Alessandrini et al. (2009), we measure functional distance at the province level as the ratio of the number of branches in the province weighted by the logarithm of 1 plus the kilometric distance between the province of the branch and the province where the parent bank is headquartered, over total branches in the province. Second, we include the degree of credit market concentration in the province by building the Herfindhal-Hirschman index computed on the share of branches held by banks operating in the province (HHI). Finally, in order to take into account the financial crisis and how it has hurt the local banking system, we follow by considering the share of branches belonging to the five largest Italian banking groups (LARGE BAN KS) which were most seriously affected by the crisis, slowing down their lending activity (Albertazzi and Marchetti; 2010; Gobbi and Sette; 2012). 6 All the banking-system variables are calculated at the end of each quarter from September 2007 to December The credit market variables shows a great heterogeneity across provinces, as visually confirmed by the maps reported in Figure 1. This variability is not exclusively driven by the traditional Italian dichotomy between more financially developed Northern regions and a less financially developed South, as shown by the between component of the standard deviation of the credit market structure variables within Northern and Southern provinces (Table A.2). 4 As is well documented in the banking literature, Italy s southern regions are economically and financially less developed, and local firms have greater difficulties in accessing bank credit (Lucchetti et al.; 2001; Guiso et al.; 2004a; Alessandrini et al.; 2009). 5 In Italy there are currently 110 administrative provinces, with some being recently established. For reasons of data availability, we refer to the standard classification into 103 provinces. 6 Alternative measures of the presence of banks severely affected by the liquidity crisis are discussed in the robustness, see Section 7. 7 For robustness, we have also taken the values of banking variables at the beginning of the two periods, finding very similar results. 8

9 By contrast, the quarterly frequency and the short duration of the data set seriously limit the variability of credit market structure variables over time. Figure 1: The spatial heterogeneity of credit market structure variables Functional distance Quintiles [.773,2.5] (2.5,3.02] (3.02,3.76] (3.76,4.45] (4.45,5.87] Herfindhal-Hirschman index Quintiles [.0355,.0855] (.0855,.105] (.105,.117] (.117,.141] (.141,.563] Share of branches held by top-5 banking groups Quintiles [.14,.412] (.412,.528] (.528,.579] (.579,.639] (.639,.878] The maps report the provincial distribution of the time-average values of DIST AN CE, HHI and LARGE BAN KS over the period 2008:1 2009:3. 4 The identification strategy The critical problem in order to correctly identify a credit crunch effect on the likelihood of firms being credit-rationed is the selection bias arising from the fact that not all firms in the sample had a positive demand for credit and that those that are observed to be rationed might not be randomly drawn from the population of Italian firms. Such a bias may be especially strong in times of financial crisis, when many firms can decide not to apply for bank loans either because they have limited financing needs or because they are discouraged from applying by the worsening lending conditions in the local credit market and the high probability of seeing their application rejected (Popov and Udell; 2012). To address the left-truncation of the sample, our identification strategy is based on a sample selection model a la Heckman, in which the selection mechanism results from sampled firms not responding to the survey questions about access to bank credit.8 Since also the dependent variable in the outcome equation is dichotomous, the presence of a credit crunch is tested estimating a binary response model with sample selection with maximum likelihood (Wooldridge; 2011). To estimate the impact of the functional distance of the local banking system from the local economy and of the flight to quality and home bias in banks lending decisions on the intensity of the crunch after Lehman s collapse, we proceed in two steps. 8 Specifically, only firms which stated, in a previous question, they had had direct contact with banks (DEM AN D = 1) were asked the question Did you get from the bank the requested amount of credit?. 9

10 4.1 Credit crunch, banks functional distance and flight to quality In the first step, we test whether the crisis has actually produced a credit crunch in Italy. Then, we investigate whether the credit crunch was harsher for firms in provinces whose banking systems are more functionally distant, and whether it was the result of a generalized flight to quality on the part banks. We measure firms quality in terms of economic prospects, productivity and informational transparency. The former is proxied by the expected level of product demand (LOW DEMAND), while EXP ORT and SIZE are taken as proxies for firms productivity and informational transparency as usual in the literature (Berger and Udell; 2002; Wagner; 2007). Hence, we estimate the following bivariate model: RAT IONED ijt = 1[αP OST LEHMAN t + β 1 DIST ANCE jt + β 2 P OST LEHMAN DIST ANCE jt + (1) 3 3 m + γ 1k Q kijt + γ 2k P OST LEHMAN Q kijt + δ h X hijt + ɛ ijt > 0] k=1 h=1 DEMAND ijt = 1[aP OST LEHMAN it + bdist ANCE jt + h=1 3 m c k Q kijt + d h X hijt + k=1 h=1 2 g rir it + η ijt > 0] where i, j and t indicate firms, provinces and quarters respectively, P OST -LEHM AN is a dummy variable which is equal to 1 for the quarters 2008:4 2009:3 and 0 otherwise, Q k are the firm quality variables conditioning the severity of the credit crunch with k = {SIZE, EXP ORT, LOW DEMAND} and X is the set of bank-market-structure and firmlevel control variables. The second equation is the selection equation, where the firms liquidity needs and the variation in labor costs are included as identifying restrictions (IR), while the dependent variable in the rationing equation is observed only for firms which applied for bank credit (DEMAND = 1). The error terms in the two equations, ɛ i,j and η i,j, are assumed to be independent of the explanatory variables, with a zero-mean normal distribution, but possibly reciprocally correlated. 9 The sign and significance of coefficients for P OST -LEHMAN and its interaction terms in the rationing equation capture the impact of the crisis on the supply of loans and its heterogeneity across markets and firms. Namely, a value of α significantly greater than zero would indicate that Italian firms experienced a credit crunch after Lehman, β 2 > 0 suggests that the crunch was harsher in credit markets mostly populated by functionally distant banks, while γ 2k > (<)0 provides evidence of a flight to quality by banks which contracted loans disproportionally more (less) to small and risky firms. 4.2 Who is hurt by functional distance? Home bias vs flight to quality Since distantly-managed banks are usually found to be at a disadvantage in soft-information production and relationship lending, a common conjecture is that during crisis periods small, risky and informationally opaque firms would be the most hurt by banking systems with a large presence of branches belonging to functionally distant banks. However, to the extent that the penetration of distant banks produces a segmentation of local credit markets into safe/transparent borrowers served by distantly-managed banks and risky/opaque borrowers served by local lenders, and if nationwide banks have actually rebalanced their loan portfolio 9 Given the limited variability of the credit market structure variable over time (see Table A.2), we can not include provincial fixed effects. However we include a dummy for firms located in Southern provinces and, in the robustness section, we show that our results are confirmed considering exclusively the sub-sample of Centre-North provinces. r=1 10

11 away from distant provinces, credit retrenchment in functionally distant banking systems should prove to be relatively more pronounced for the former type of borrower than for the latter. Therefore, in this second model we focus on the post-lehman period to test whether in provinces populated by functionally distant banks the flight to quality effect was severer or whether it was the good-quality firms, the market segment typically served by nationwide banks, which suffered relatively more than in provinces with a close banking system. To the best of our knowledge, there is no direct evidence about these contrasting effects. A partial exception supporting the flight-to-quality view is Albertazzi and Marchetti (2010), who document that after Lehman s collapse, banks belonging to the five largest banking groups reduced outstanding loans to riskier borrowers significantly more than other banks. This evidence, however, does not consider credit rationing, is limited to banks with a low risk-weighted capital ratio (lower than 10%), does not control for the distance between the lending branch and the parent bank headquarters and does not explicitly model the demand of credit. As in model (1), we proxy firms quality by LOW DEMAND, EXP ORT and SIZE. In addition we consider the firm s financial risk by using the status of being credit-rationed in the pre-crisis period (P RE LEHM AN RAT ION ED). Hence, we estimate the following bivariate model: RAT IONED ijt = 1[αDIST ANCE jt + + k=1 n δ h X hijt + ɛ ijt > 0] h=1 DEMAND ijt = 1[aDIST ANCE jt + 4 β k Q hijt + 4 γ k DIST ANCE Q kijt + (2) k=1 4 m b k Q hijt + d h X hijt + k=1 h=1 2 g rir it + η ijt > 0] where i, j and t indicate firms, provinces and the post-lehman quarters, Q includes the four k-characteristics for firms quality mentioned above and X is the set of bank-market-structure, firm-level control variables. and four time dummies. As in the previous model, the demand equation includes the firms liquidity needs and labor costs as identifying restrictions (IR) and error terms in the two equations are assumed independent of the explanatory variables, but possibly correlated. In the rationing equation, a coefficient α > 0 indicates that the credit crunch is more severe in provinces dominated by nationwide banks, while coefficients on the variables included in Q identify the flight-to-quality effect. The interaction terms DIST ANCE Q allow us to test which type of borrower was hurt by functionally distant banking systems. A coefficient γ k > 0 indicates that a large presence of distantly-managed bank branches in a province spurs flight from risky, less productive and opaque firms. Conversely, a γ k not significantly greater than zero would suggest that the strength of flight to quality is not affected by the structure of the local banking system and possibly (if γ k < 0) that safe and transparent firms are relatively more harmed in areas with functionally distant banking systems. This would imply that the credit crunch was the result of the home bias by nationwide banks. 5 Descriptive analysis From firm-level data available in the ISAE/ISTAT survey it is possible to have clear descriptive evidence of the intensity of the credit crunch. In the last quarter of 2008, the share of firms judging as restrictive the condition to access bank credit increased to 41%, while during the first nine months of 2008 one firm out of every four had the same perception (Figure 2, left r=1 11

12 panel). 10 The firms perception of banks lending behavior is strongly correlated with their product demand and liquidity levels. In the right-hand panel in figure 2, we consider the difference between the share of firms assessing their demand as high less the share assessing it as low, and the difference between the share of firms assessing their liquidity as good less the share assessing it as bad. Both indicators follow a similar trend, with a decline in the business demand and liquidity from the second half of 2008 and the bottom reached in the first quarter of However, the climate of the product market is judged to evolve worse than liquidity conditions by a larger share of firms: on average, during the sample period 47% of firms faced a low level of demand, while only 6% stated that the level of demand was high. Figure 2: Access to bank credit and business climate: Access to credit (share of firms) Business climate q1 2008q2 2008q3 2008q4 2009q1 2009q2 2009q q1 2008q2 2008q3 2008q4 2009q1 2009q2 2009q3 Accomodating Restrictive Product demand Liquidity needs (a) Conditions of access to credit (b) Business climate Notes: Elaborations based on the ISAE/ISTAT Survey of Manufacturing Firms. In any quarter, the business climate is defined, alternatively: as the share of firms assessing their demand as high less the share assessing it as low (product demand), and as the share of firms assessing their liquidity as good less the share assessing it as bad (liquidity). In Figure 3 we look directly at the evolution between the first quarter of 2008 and the third quarter of 2009 of the two dependent variables used in the bivariate probit model. In the left-hand panel we plot the share of firms which applied for bank credit (DEMAND), while in the right-hand panel we focus on the share of firms which have been credit-rationed (RAT ION ED). To take into account possible differences in the severity of the credit crunch according to the structure of local credit markets, we calculate these shares separately for firms located in provinces where the banking system is functionally close (DIST ANCE below the 75 percentile of its 2008:3 provincial distribution) and functionally distant (DIST AN CE below the 75 percentile of its 2008:3 provincial distribution). Two main patterns emerge from the diagrams. The first concerns timing and shows that while the demand for credit remained quite stable before and after Lehman s collapse (apart from a temporary and sharp increase in the second quarter of 2009), the restraining response of the banking system to the reduction in global liquidity was evident and immediately transmitted to the real sector. The share of rationed firms increased from 11.6 percent in the third quarter of 2008 to 21.6 percent in the last quarter of the year and further to 25.5 and 27.5 percent respectively in the first and third quarters of The second pattern is related to the geographical differences in the access to bank credit. On average, over the sample period, firms located in provinces densely populated by distant banks are less likely to seek credit. The share of firms asking for fresh flows of bank credit in each quarter is 30.1% in provinces where the functional distance of the banking system is 10 Similar findings are reported by Costa and Margani (2009). 12

13 particularly high, while the same share increases to 32.7% in provinces where the banking system is functionally closer; this difference is statistically significant at the usual level of confidence. The opposite trend is observable with the share of rationed firms. Just before the onset of the crisis (2008:3), the share of credit-rationed firms is 11.6%, irrespective of the functional distance of local banking systems. In the first quarter after the Lehman collapse, the tightening of credit conditions is found everywhere, but the increase in credit rationing is statistically higher in provinces dominated by distant banks. 11 Figure 3: Demand and supply of bank credit: 2008:1 2009:3.4.4 Share of firms demanding bank credit Share of rationed firms q1 2008q2 2008q3 2008q4 2009q1 2009q2 2009q q1 2008q2 2008q3 2008q4 2009q1 2009q2 2009q3 Functionally close provinces Functionally distant provinces Functionally close provinces Functionally distant provinces (a) Demand for bank credit (b) Credit rationing Notes: Elaborations based on the sample of 3,631 firms (24,651 observations); source: ISAE/ISTAT Survey on Manufacturing Firms. Provinces are classified as functionally close (distant) whether DIST ANCE is below (above) the 75 percentile of its 2008:3 provincial distribution. Table A.1 reports the descriptive statistics for the sample of 3,623 firms (23,140 observations) used in the first empirical exercise (Table B.1). From the descriptive analysis of quarterly data, it emerges that firms asking for bank credit do not differ significantly from the non-applicants in terms of size and product demand, while they are more likely to export. Moreover, the two groups of firms are not located in provinces with different degrees of functional distance. By contrast, significant differences emerge in the sub-sample of applicants between the ones which were credit-rationed and the others which were untouched by credit restrictions. The former are smaller, less internationalized, with a lower product demand and predominantly located in provinces where the banking system is functionally distant (Table A.1). This preliminary evidence on the heterogeneity of the credit crunch is formally tested in the first empirical exercise in the next Section. 6 Econometric results In line with the recent empirical evidence about the credit crunch in Italy (Costa and Margani; 2009; Gambacorta and Mistrulli; 2011; Gobbi and Sette; 2012), our regression results show that Italian banks significantly reduced credit supply after the collapse of Lehman Brothers. In addition, they clearly show that the organizational structure of local credit markets has been 11 The share of rationed firms is 26.4% (20.9%) in provinces where DIST ANCE is above (below) the 75 percentile of its 2008:3 provincial distribution, and this difference is statistically significant at the 10 percent level of confidence. Over the whole 2008:1 2009:3 period, 21.6% of firms located in provinces where the banking system is functionally distant are credit-rationed, while this share drops to 16.7% in provinces with a closer banking system. Also in this case the difference is statistically significant. 13

14 a (statistically and economically) major determinant of the severity of the credit crunch and the spatial heterogeneity across Italian provinces. The greater the share of branches held by out-of-market, distantly-managed banks, the higher the probability of local enterprises being credit rationed in the crisis period. In particular, in provinces where the banking system is functionally distant, even firms which before the onset of the financial crisis had full access to bank credit increased their likelihood of being rationed after Lehman s collapse. In those provinces, financing constraints proved relatively more binding for larger and exporting firms, and for enterprises with a higher expected product demand and financially healthier than the average. This suggests that the harsher credit crunch in functionally distant credit markets has not been the result of a stronger, generalized flight from risk on the part of bank branches lending locally, but of a contraction of the engagement by nationwide banks in provinces far away from their headquarter due to a home bias effect. 6.1 Firms financing constraints pre- and post-lehman Table B.1 reports the coefficients for the model (1) estimated over the pooled sample using preand post-lehman quarters. The negative and significant ρ confirms the presence of a negative correlation between the equations modeling the demand and supply of credit, supporting the discouraged borrower hypothesis according to which, in anticipation of a high probability of credit rationing, a self-selection mechanism is at work leading riskier firms to stay out of the credit market Demand equation Looking at the demand equation, there is clear evidence that, with the onset of the global crisis, there has been a significant contraction of the demand for credit by firms, whose loan applications are 8.7% less frequent on average. The coefficients for the firm-level variables are generally significant and with signs consistent with the hypothesis that applicants tend to self-select. Namely, we find that large firms, with part of their production being exported, increasing labor costs, a stronger demand for their products and with lower levels of liquidity, are significantly more likely to apply for a bank loan. By contrast, the credit market structure variables (DIST ANCE, HHI and LARGE BANKS) are not significantly correlated with the firm-level demand for bank credit, indicating that the firms credit demand depends on their own characteristics more than that of banks operating in the local market. Finally, the demand for credit in has not been affected by the geographical location of firms in the less-developed Southern regions Rationing equation Consistent with previous literature, the estimation results of the rationing equation show that firms financing constraints decrease with their size, export attitude and the level of demand for their products (Alessandrini et al.; 2009; Minetti and Zhu; 2011). As for the demand side, as with the supply of credit there is no significant difference between the North and the South of Italy, once firm-specific characteristics and the structure of provincial credit markets are taken into account. The significant and positive coefficient for the dummy P OST LEHMAN reported in column (1) suggests that there has been a tightening in the supply of bank credit since the collapse of Lehman Brothers. The calculation of the average partial effects shows that, other things being equal, in the post-lehman period Italian firms have had a 7.8% higher probability of being credit rationed. 14