Credit rationing in small business bank relationships

Transcription

1 Credit rationing in small business bank relationships Karolin Kirschenmann* November 2010 Job market paper Abstract This paper studies how credit rationing develops over bank-borrower relationships. I analyze a unique dataset of matched loan application and loan contract information which allows me to establish the actual degree of credit rationing by relating a borrower s requested loan amount to the bank s granted loan amount. In line with theoretical predictions I find that, at the outset of bank relationships, credit rationing is significantly higher for opaque borrowers but that it subsequently decreases over bank relationships for both opaque and transparent borrowers. The analysis indicates that the decrease in credit rationing stems from a more pronounced increase in granted than in requested loan amounts. Furthermore, I find that borrowers who were credit rationed in their previous loan increase their demand more moderately than previously non-rationed borrowers. When taken together with the finding that the increase in granted loan amounts does not differ between previously rationed and non-rationed borrowers and the result that rationed borrowers perform worse ex-post, this suggests that credit rationing has real effects on firms economic prosperity. Keywords: Credit rationing, relationship lending, small business lending, asymmetric information, loan applications JEL classification: D82, G20, G21, G30 * Karolin Kirschenmann, University of Mannheim, Department of Banking and Finance, L 9, 1-2, D Mannheim/Germany, kirschenmann@bwl.uni-mannheim.de Acknowledgements: I am grateful to Franklin Allen, Mitchell Berlin, Martin Brown, Geraldo Cerqueiro, Hans Degryse, Daniel Foos, Todd Gormley, Mark Jenkins, Leonard Nakamura, Lars Norden, Steven Ongena, Christian Opp, Maria Fabiana Penas, Michael Roberts, Sascha Steffen, Luke Taylor, Eva Terberger and Wolf Wagner, seminar participants at the University of Mannheim, the Wharton School of the University of Pennsylvania and the Federal Reserve Bank of Philadelphia as well as participants at the 6 th Annual Conference of the Research Committee Development Economics of the German Economic Association, the 2010 Münster Banking Workshop and the 37 th Annual Meeting of the European Finance Association for helpful comments and suggestions. I am especially grateful to the management and employees of the bank which provided me with the data. The paper was partially completed while I was visiting the Finance Department of Tilburg University and the Financial Institutions Center of the Wharton School of the University of Pennsylvania whose hospitality I greatly appreciated.

2 1 Introduction In the presence of asymmetric information both adverse selection and moral hazard may lead to credit rationing (e.g. Jaffee and Russell (1976), Stiglitz and Weiss (1981)). Over the course of bank-borrower relationships, the ability of banks to produce information (e.g. Diamond (1984), Ramakrishan and Thakor (1984), Boyd and Prescott (1986)) decreases the risk of adverse selection and moral hazard, and therefore the need to ration credit. The empirical evidence on credit rationing and its development over bank relationships, however, is scarce due to a lack of demand data. This paper uses information from loan applications to establish the degree of credit rationing as well as its evolution when informational asymmetries resolve themselves over the course of bank-borrower relationships. Understanding the extent and the effects of credit rationing is important to design adequate policies to overcome the obstacles that impede opaque firms in their access to credit. I analyze a unique panel dataset of matched loan applications and loan contracts that includes both requested and granted loan terms as well as borrower and relationship characteristics at the time of loan origination. The dataset consists of nearly 99,000 loans to small enterprises extended by one bank in Bulgaria over the period April 2003 to September It allows me to provide evidence on the degree of credit rationing by relating borrowers demand to the bank s supply. Since the wedge between demand and supply is informative about the resolution of informational asymmetries over time, I investigate not only how it relates to firm characteristics but also how it evolves over sequential loan contracts. This analysis of chains of short-term investment loans complements studies that focus on credit lines to assess how banks use the information they gather from multiple interactions with their borrowers (e.g. Berger and Udell (1995) and Norden and Weber (2010)). Furthermore, by establishing the dynamic patterns of requested and granted loan amounts that arise when bank and borrowers interact repeatedly, I am able to connect the degree of observed credit rationing to the evolution of its components over time. Finally, I 2

3 complement the analysis by studying differences in the ex-post loan performance of rationed vs. non-rationed borrowers to assess how credit rationing may affect borrowers and the bank. The results show that credit rationing due to informational asymmetries is considerable in lending to small businesses. In line with predictions from the credit rationing theory, I find that opaque firms (i.e. firms that are comparatively young or small when starting to borrow from the bank) are significantly more rationed than more transparent firms and that the degree of credit rationing decreases significantly over loan sequences. Furthermore, loan officer changes lead to higher credit rationing, which seems to be driven by the loss of private information. Studying the dynamics behind the observed reduction of credit rationing over bankborrower relationships I find that granted loan amounts increase significantly over time. Requested loan amounts also increase over relationships but on a lower level. The resolution of information problems over bank relationships therefore leads to lower credit rationing because the bank is willing to increase its stakes to meet the borrowers (growing) demand. 1 Furthermore, I find that borrowers who were credit rationed at their previous loan increase their demand more moderately than previously non-rationed borrowers. Together with the finding that the increase in granted loan amounts does not differ between previously rationed and non-rationed borrowers and the result that rationed borrowers perform worse ex-post this suggests that credit rationing may have real effects on firms economic prosperity. In the absence of real effects, in contrast, the bank should be expected to make up for the previous rationing because borrowers who repay their loans duly have revealed themselves as good borrowers. 1 The concept of starting small is also established in the corporate finance literature (e.g. Tirole (2006)) to model so-called staged financing, in the industrial organization literature to explain the development of business partnerships in states of uncertainty (e.g. Rauch and Watson (2003)) and in the venture capital literature when venture projects are financed under uncertainty and the threat of moral hazard (e.g. Bergemann and Hege (1998) and Wang and Zhou (2004)). 3

4 The remainder of the paper is organized as follows. Section 2 derives testable hypotheses from the banking theories on credit rationing and then reviews the related empirical literature. Section 3 provides institutional details on the loan granting process and describes the data while section 4 presents the findings from the empirical analyses. Section 5 concludes. 2 Related literature 2.1 Theory Modern theories of credit rationing rationalize why banks may rather set an interest rate below the market-clearing rate and ration credit than increase interest rates when facing an increased demand. Stiglitz and Weiss (1981) show that credit rationing occurs if borrowers are heterogeneous with respect to their default risks and banks are unable to distinguish between borrowers with different characteristics. A rise in interest rates then may affect the quality of demand as the least risky borrowers drop out of the market and the average risk of the borrower pool increases (adverse selection) and its behavior because some borrowers may opt for the riskier project after having received funding (moral hazard). Both effects, in turn, impact negatively on banks expected profits and therefore induce banks to ration borrowers. 2 While banks do accumulate private information about their borrowers, there nevertheless remain informational asymmetries, e.g. concerning borrower heterogeneity in entrepreneurial ability, especially in the case of very young and small firms. 2 Rationing can take the two forms of (i) loan size rationing, which means that, at the current interest rate, all borrowers are served but demand a larger loan amount than they finally receive from the bank and (ii) borrower rationing, which means that some borrowers get no loan at all although they may have profitable investment projects and are indistinguishable from those borrowers who receive loans (Keeton (1979)). Jaffee and Russell (1976) establish loan size rationing in a model in which borrowers differ in their probability of default in the sense that some borrowers are honest and repay whenever they are able to while other borrowers are dishonest experiencing a utility increase from defaulting and do so whenever the costs of default are lower than the contracted repayment. In the model of Parker (2003) adverse selection and thus borrower rationing occurs because borrowers differ in their ability. In a moral hazard model, Watson (1984) shows that borrower rationing may occur because borrowers reduce their effort as a response to an increased interest rate. Ghosh, Mookherjee and Ray (2000) derive both loan size and borrower rationing in the presence of moral hazard. 4

5 In a multi-period setting (see Sobel (1985), Ghosh and Ray (2001) and the experimental evidence by Brown and Serra-Garcia (2010)), credit rationing is expected to decrease over bank-borrower relationships. In the case of adverse selection, on the one hand, the need for credit rationing is reduced over multiple interactions between the same borrower and lender because informational asymmetries are resolved. On the other hand, bad borrowers reveal themselves over time and are not granted a further loan. Adverse selection thus impacts on the degree of credit rationing over loan sequences along the intensive and extensive margins. In the case of moral hazard, credit rationing should also decrease over bank relationships as borrowers become more transparent and firm owners stakes in their companies increase over time so that they can bear a larger share of the risk decreasing the risk of moral hazard. In sum, this leads to the following testable hypotheses: H1: Credit rationing is tighter for opaque borrowers (e.g. young and small firms) than for more transparent borrowers in the beginning of bank relationships. H2: Credit rationing decreases over bank-borrower relationships. 2.2 Related empirical studies An extent set of papers studies the influence of bank relationships on credit availability and generally finds a positive relation between various measures of relationship strength and credit availability (Petersen and Rajan (1994, 1995), Harhoff and Körting (1998), Machauer and Weber (1998)). Cole (1998) and Angelini, Di Salvo and Ferri (1998) establish that the valuable private information seems to be gathered very early in the relationship, while Elsas and Krahnen (1998) find that especially risky borrowers benefit from bank relationships. Scott (2006) shows that loan officer turnover, which is connected with a loss of soft information, is positively related to the probability that banks deny credit. Finally, several studies document increasing contracted loan amounts empirically (e.g. Armendariz and Murdoch (2000), Ioannidou and Ongena (2010)). 5

6 While these papers have established the value of close bank relationships on the availability of credit for small firms, they have not been able to directly observe borrowers requests, relate them to the actual loan terms granted and establish the extent of credit rationing. Furthermore, many of these studies are confined to using cross-sectional data. This paper uses a more comprehensive and direct measure of credit rationing which incorporates the demand side: the wedge between requested and granted loan amounts for those borrowers receiving credit. The panel structure of the employed dataset allows to explicitly follow borrowers over their bank relationships and to analyze the evolution of credit rationing over individual loan sequences. 3 Several studies examine borrower (type I) rationing providing evidence on the firm characteristics that determine the probability of being denied credit (e.g. Brown, Ongena, Popov and Yesin (2010), Cole (2010)). Their results show that younger, smaller and observationally riskier firms are more likely to be denied credit. This study, on the contrary, is concerned with loan size (type II) rationing for those borrowers who receive credit and thus complements these findings. My results show that the comparatively young and small firms are also more loan size rationed than the older and larger firms once they receive credit, with this effect being especially pronounced in the beginning of bank relationships. Finally, this study is related to the literature that aims at disentangling demand and supply effects on credit availability by using information from both loan applications and loan contracts. Cheng and Degryse (2010) observe requested and granted loan amounts and study the impact of the introduction of a public credit registry on credit rationing in the Chinese credit card market. Furthermore, Jimenez, Ongena, Peydro and Saurina (2010) examine the impact of macroeconomic and financial shocks on the probability that a loan request results in a loan granted and Puri, Rocholl and Steffen (2010a) examine how the US financial crisis 3 One caveat to this approach is that it assumes requested and granted loan amounts to mirror real demand and supply although both may be driven by strategic considerations. Whereas the dataset at hand does not allow me to fully resolve this issue, it takes the analysis of credit availability one step further by incorporating loan applications and shedding some first light on the degree and the dynamic evolution of credit rationing. 6

7 affected retail bank lending at German savings banks. This study extends the existing evidence on demand and supply effects in bank lending by analyzing requested and granted loan amounts, i.e. the dynamic patterns that arise on both the demand and the supply side, in a sequence of interactions between borrowers and a bank. 3 Data and methodology 3.1 The data and the Bank s loan granting process The dataset used in this study comprises all small annuity loans, credit lines and overdrafts (loans with amounts up to 50,000 EUR) to firms extended by one Bulgarian bank (henceforth called the Bank ) between April 2003 and September For each loan the dataset includes information from the borrowers loan applications on the loan terms that were requested. I match this information with data on the actually granted loan terms as stated in the loan contracts as well as with borrower characteristics and relationship indicators at the time of loan origination. Definitions of all variables are provided in Table 1. [Insert Table 1 here] All observations with missing loan or firm-level data are excluded. Since the following empirical analysis focuses on the evolution of requested and granted loan sizes and their relation over a loan sequence, all loans after the ninth are excluded due to very few observations in these categories. Based on the fact that interest rate and collateral requirements are fixed for small loans whereas they are individually negotiated in the loan granting process for medium loans (loans with amounts of more than 50,000 EUR), eventually all medium loans are excluded from the main analysis. This leads to the final sample of 98,888 loans to 53,305 firms among which are 20,355 repeat clients with loan sequences of up to nine loans. The subsample of loans to repeat clients will be important 7

8 throughout my empirical exercise as it allows me to control for unobserved (time-invariant) firm heterogeneity. As of September 2007, 30 commercial banks and branches of foreign banks operate in the Bulgarian banking system. Total sector assets amount to 26.5 billion EUR (an increase of 34% over the preceding twelve months mainly funded by an increase in deposits and equity) with the five largest banks accounting for 56% of total banking assets (Bulgarian National Bank (2007)). The Bank that provided me with the data is a commercial, full-service bank with a nationwide branch network. As with the majority of banks in Bulgaria and the region, it is owned by foreign strategic and commercial investors 4 who implemented the Bank s structure, its lending technology, risk management, management information system and staff training and accomplish regular supervision. The Bank provides a wide range of financial services to private and business clients but has a special focus on and expertise in lending to small enterprises. Compared to the aggregate banking system, where only 41% of assets are loans to enterprises, 70% of the assets of the Bank are enterprise loans. At the heart of the Bank s lending technology is a personnel-intensive analysis of the borrower s debt capacity. Approaching the Bank, a prospective borrower first of all meets a client advisor who assesses whether the borrower meets the Bank s basic requirements. If this is the case, the client fills in a loan application form. On this form the client indicates her preferred loan amount, maturity and currency as well as the purpose of the loan. The client also has to provide information about the firm ownership, other bank relations and the free cash flow available for the repayment of the loan. In a next step, the Bank s credit administration prepares information on the borrower s credit history with this Bank and other banks. 5 At the same time, the loan officer conducts the financial analysis which includes a 4 In 2007, 82% of bank assets in Bulgaria were in the hands of institutions with majority foreign ownership. In Central and Eastern Europe the average share of foreign bank assets in 2007 was 80%. 5 Enterprise loans in Bulgaria are covered both by the public credit registry and, since 2005, a private credit bureau. At the same time, contract enforcement remains problematic as indicated by the 2010 Doing Business Indicators which rank Bulgaria 87 th among 183 countries in this category (see 8

9 personal visit to the borrower s site. Eventually, the loan officer presents the client s demand and the suggested loan terms together with the information gathered during the financial analysis to the Bank s credit committee which makes the final decision on the granted loan terms. Collateral requirements and interest rates are largely standardized and play a minor role in the individual loan contracting process for my sample of small loans. Therefore, I will not explicitly consider these loan terms throughout the empirical analysis. Concentrating the analysis on small loans from one bank in an emerging market provides an ideal ground for studying credit rationing because informational asymmetries are presumably severe, especially among young and small firms and in the beginning of bank relationships. The Bank s largely standardized loan contracts for small loans leave mainly loan amount (and maturity 6 ) as means for the Bank to deal with borrowers credit risks and the risks arising from informational asymmetries. Finally, since the loan granting process is the same for all observed loans possible heterogeneity is reduced at this level. 3.2 Credit rationing Since I observe requested and granted loan amounts I am able to establish credit rationing by the extent to which borrowers receive a smaller loan amount than they requested. I denote this as the Share granted and measure it as the ratio of granted to requested loan amounts. Note that this is an inverse measure of credit rationing with smaller values indicating tighter rationing. Figure 1 contains a scatterplot of borrowers requested loan amounts against their respective granted loan amounts (both measured in log EUR) and provides an indication of the importance of credit rationing in the sample. All observations above the 45 degree line are loans that are granted with a lower than requested loan amount, i.e. are credit rationed (Share granted < 1). In total, these loans make up 26.0% of the sample. 6 Since I find amount and maturity to be complementary loan contract terms, the analysis focuses on requested and granted loan amounts. 9

10 For loans on the 45 degree line requested and granted loan amounts are identical, while 4.6% of the loans are granted with larger than requested loan amounts (Share granted > 1). [Insert Figure 1 here] Figure 2 displays the distribution of the degree of credit rationing (Share granted) for all loans and the subsample of first loans. In both samples, the median (indicated by the triangle) is one and coincides with the 75 th percentile. This matches the above illustration in Figure 1 that the majority of loans are granted at their requested amount and only very few loans exhibit larger requested than granted amounts. Interestingly, the dispersion of values is considerably larger for first loans than for the full sample which is a first indication that over multiple interactions between borrowers and the Bank the Share granted increases towards one. The following analysis explores this in more detail. [Insert Figure 2 here] Theory predicts that the degree of credit rationing depends on the extent of asymmetric information. Table 2 therefore displays the degree of credit rationing (Share granted) for subsamples of opaque vs. more transparent borrowers. To capture the effect of different levels of asymmetric information between borrowers and to separate it from the effect of repeated interactions over time the table provides evidence for the subsample of first loans as well as the subsample of all later loans. Two proxies for firm opaqueness widely used in the banking literature are firm age (e.g. Berger, Klapper and Udell (2001)) and firm size (e.g. Berger and Udell (1995) and Petersen and Rajan (1995)). I define Initially young firms as those with firm age of up to two years at their first loan because such firms have not had the time to establish a public track record (see Petersen and Rajan (1994)). To define Initially small firms, I follow 10

11 Petersen and Rajan (1995) and split the sample at the median value of firm size at the first loan. [Insert Table 2 here] Table 2 shows that credit rationing is significantly larger for the Initially young than for the initially old firms and that this result holds for the subsamples of first and later loans. The economic difference, however, is very small in the subsample of later loans while it is large for first loans. In line with this, the last column of Table 2 shows that the decrease in credit rationing between first and later loans is more pronounced on average for the Initially young firms than for the initially old firms. The difference-in-difference estimate (in bold) indicates that this difference in the decrease of rationing is indeed significant. Findings for the Initially small vs. initially large firms are very similar with differences between the two groups being somewhat more pronounced. Thus, Table 2 clearly indicates that these measures of asymmetric information play an important role in the Bank s decision to grant a lower than requested amount. Confirming hypotheses H1 and H2, the group of opaque borrowers is found to be more credit rationed than the group of transparent borrowers and first loans are more rationed than later loans in a loan sequence. To measure relationship strength I use the Loan number which indicates how many interactions between the borrower and the Bank have taken place providing the Bank with the opportunity to monitor borrowers and to observe their repayment behavior. Table 3 provides summary statistics on the evolution of the Share granted over loan sequences. Panel A shows that the observed degree of credit rationing decreases considerably over an average loan sequence with this effect being most pronounced in the beginning of bank relationships. The Share granted increases significantly in the beginning of loan sequences from 0.90 to

12 between the first and the fifth loan. 7 While borrowers start, on average, with receiving 90% of their requested loan amount, they almost receive their requested amount after several interactions with the Bank. At the same time, Requested amount and Granted amount increase over a bank-borrower relationship nearly doubling on average between the first and the ninth loan. This indicates that the observed reduction in credit constraints over bank relationships is due to a disproportionate increase in granted loan amounts and not driven by decreased demand because, for instance, borrowers establish lending relationships with other banks. [Insert Table 3 here] Panel B of Table 3 reports the evolution of the degree of credit rationing over loan sequences of different lengths. It shows that the Share granted increases similarly for all loan sequences no matter how many loans they comprise. Furthermore, the initial degree of credit rationing hardly varies for sequences of different lengths. These results provide further evidence for hypothesis H2: they show in particular that the decrease in credit rationing over bank relationships does not only occur because risky borrowers drop out of the sample but most notably because repeat borrowers receive increasing loan amounts over multiple interactions with the Bank. The descriptive statistics provided in Tables 2 and 3 confirm findings from previous studies that employ indirect or equilibrium outcome measures for credit availability (e.g. Petersen and Rajan (1994) and Ioannidou and Ongena (2010)). Yet, this paper is, to the best of my knowledge, the first being able to provide evidence on the actual degree of rationing and its evolution over bank-borrower relationships. A crucial part of the following analysis 7 To rule out that the observed pattern is driven by changes in the Bank policy over years, I also investigate loan sequences that start in different years and find similar patterns no matter in which year bank relationships begin. I also study a measure of credit rationing which is the ratio of the difference between requested and granted loan amounts over total firm assets to account for firm size and find qualitatively the same results. 12

13 will be concerned with the determinants of credit rationing and the underlying dynamics on the borrower and Bank side over the course of bank relationships. 3.3 Determinants of the degree of credit rationing I study the factors that influence the degree of credit rationing in the sample in two steps. First, I estimate an OLS model for the full sample with Share granted i,k,t (the ratio of the granted loan amount to the requested loan amount for loan k taken out by firm i in month t) as the dependent variable: Share granted i,k,t = a + β 1 A i,t + β 2 F i,t + β 3 L k + β 4 B t + β 5 T t + e i,k,t (1) A i,t is a vector of indicators measuring the level of asymmetric information, while F i,t and L k are vectors of firm and loan characteristics. B t and T t are vectors of branch and year-month dummies accounting for the branch-specific (such as local competition) and general (such as macroeconomic and monetary conditions, the Bank s refinancing situation and the Bank s prevailing interest rate and collateral requirements for small loans) environment at the time of loan disbursement. In a second step, I estimate outcome equation (1) as a panel model with firm fixed effects to control for any unobserved time-invariant borrower heterogeneity that may influence the Share granted. The fixed effects estimator concentrates the analysis on the factors that determine differences in the degree of credit rationing over the course of individual bankborrower relationships which include at least two loans. 13

14 Indicators of asymmetric information The variable Loan number indicates the number of the current loan and measures the length of the bank-borrower relationship. 8 Most importantly, it captures the dynamic patterns that arise along a chain of interactions between borrowers and the Bank. To allow for nonlinear effects I include the dummy variables Loan number_2,, Loan number_5 (which pools interactions number five to nine because of the fewer observations in these categories and because the descriptive analysis has displayed that most of the action happens in the beginning of the relationship) and use Loan number_1 as the reference category. As outlined in section 3.2, I use the dummy variables Initially young and Initially small to identify the groups of opaque borrowers that might face tighter credit rationing. To study whether credit rationing evolves differently over bank relationships for opaque vs. transparent firms I assess the interaction effects Loan number_2*initially young,, Loan number_5*initially young and Loan number_2*initially small,, Loan number_5*initially small. When a borrower applies for a loan, it is the loan officer who collects all the borrowerspecific data necessary for the subsequent decision on whether to grant a loan and under which conditions. 9 If the information gathered by the loan officer cannot fully be transmitted within the bank, which is likely for qualitative soft information, part of it is lost in case a loan officer change takes place. The variable Loan officer change captures whether the loan officer has changed during the duration of the previous loan. Firm and loan characteristics I include a vector of firm characteristics to control for (observable) borrower risk, but which may also serve as indicators of asymmetric information. Sole proprietorships are more 8 I do not include the duration of a Bank relationship to measure the level of asymmetric information because it is highly correlated with Loan number. However, rerunning all regressions with Bank relationship instead of Loan number yields qualitatively and quantitatively very similar results. 9 See Berger and Udell (2002), Stein (2002) and for empirical papers using loan officer information e.g. Liberti (2005), Scott (2004, 2006), Uchida, Udell and Yamori (2006), Beck, Behr and Güttler (2009) and Liberti and Mian (2009). 14

15 opaque than incorporated firms because they do not have to provide certified annual reports according to Bulgarian law, hence the dummy variable Sole proprietorship equals one if the firm is a sole proprietorship and zero otherwise. Borrowers that are highly indebted face a higher risk of default in case of external shocks to their income so that I introduce Leverage, the firm s total debt as share of its total assets at the disbursement date of the loan. A firm with little financial scope (Ln(Disposable income)) to react to unforeseen cuts to its income is more vulnerable to external shocks and thus more risky because the repayment of the loan may be endangered more easily. I also include an indicator whether the firm has loans from Other banks. 10 Having access to other funding sources may reduce the firm s need for funds at this Bank reducing observed credit rationing due to lower demand. At the same time, firms borrowing from multiple sources may be the more productive and successful firms and therefore less credit rationed. Finally, whether a borrower faced repayment problems at her previous loan may influence the degree of rationing at the next loan so that I include the dummy variable Previous arrears which is one if the borrower was past-due on interest or principal payments more than 30 days at her previous loan, and zero otherwise. To account for all remaining differences in firm characteristics the regressions contain seven Industry dummies. 11 The variable Fixed capital loan indicates whether a loan is for fixed capital financing or working capital otherwise. If a loan is intended for fixed capital financing, the underlying asset may be sold in case of default lowering the risk associated with such loans. Similarly, an Annuity loan (dummy variable which is one if the loan is an annuity loan and zero if it is a credit line or overdraft) may be considered less risky because of its regular repayment 10 Note that Other banks is only a crude proxy for other bank relationships. I retrieve it by calculating a firm s outstanding debt at this Bank and comparing it to its total liabilities at the time of each loan disbursement. Other banks takes on the value of one if total liabilities are larger than the outstanding debt at this Bank, and zero otherwise. 11 These are agriculture (the baseline category), construction, manufacturing, trade, transport, tourism and other services. 15

16 schedule. To account for the fact that new loans in a sequence may be granted before the previous loan is fully repaid, I include the dummy variable Add-on loan. I do not include loan maturity as this is possibly endogenous to the determination of loan amount. Studying requested and granted loan amounts and maturities reveals that both loan terms are complements because for 67% of all loans they are adjusted into the same direction, i.e. requests for both loan terms are either higher, lower or equal to both granted loan terms. The Spearman rank correlation between the Share granted and the ratio of granted to requested maturity is 0.44 and significant (p-value <0.01) which means that the two variables are not independent and I therefore concentrate the analysis on requested and granted loan amounts. 3.4 Summary statistics Table 4 compares the observable loan and firm characteristics for the two subsamples of non-rationed vs. rationed loans. The credit rationed loans in column (2) exhibit significantly larger requested and smaller granted loan amounts than the non-rationed loans (loans granted with an amount at least as high as the requested loan amount) in column (1). Interestingly, credit rationed loans are twice as likely to be in Arrears for over 30 days as non-rationed loans. Furthermore, credit rationed firms are more likely to be young and small at their first loan, are younger in general and have shorter bank relationships than the non-rationed firms. They are also smaller in terms of total assets and disposable income so that they are clearly the less transparent firms, which is in line with predictions from the credit rationing theory. A t-test confirms that these differences in firm characteristics are statistically significant at the 0.01-level when comparing the two groups. [Insert Table 4 here] 16

17 4 Multivariate results 4.1 Determinants of the degree of credit rationing Table 5 displays the regression results on the determinants of Share granted presenting estimations for both the full sample and the panel of repeat clients. Regressions for the full sample include industry, branch and year-month dummies, but they do not include the variables Loan officer change, Previous arrears and Add-on loan because they are not defined for all first loans. The regressions for the subsample of repeat clients include firm fixed effects to account for unobserved time-invariant firm heterogeneity and year-month dummies. The branch and industry dummies as well as the variables Initially young, Initially small and Sole proprietorship are excluded from the panel regressions due to (almost) no withinvariation. Standard errors are reported in parentheses and are adjusted for clustering at the firm level. Effects of asymmetric information indicators and firm and loan variables Column (1) of Table 5 presents OLS estimates for the full sample. The results confirm that firms with more intense bank relationships as well as more transparent and less risky firms are less credit rationed. The variables Loan number_2,, Loan number_5 capture the effect which the intensity of the bank-borrower relationship has on observed loan size rationing for the initially older and larger firms. The more often such firms borrow from the Bank, the less credit rationed they are with the degree of credit rationing decreasing significantly between the first two interactions by 2.7 percentage points. A Wald test for differences in coefficients of the further adjacent loan numbers reveals that they do not differ significantly. This is in line with empirical studies concluding that most of the valuable private information is gathered in the beginning of a bank-borrower relationship (e.g. Cole (1998)). Those firms which are Initially young or Initially small experience credit constraints that are significantly higher than those 17

18 for the initially older (4.2 percentage points on average) or initially larger (6.3 percentage points on average) firms. The significantly positive coefficients for the interaction effects of Loan number_2,, Loan number_5 and Initially young and Initially small respectively indicate that the reduction of credit rationing over a loan sequence is more pronounced for initially younger and smaller firms. For instance, between the first two interactions Initially young firms experience on average an additional 2.7 percentage points decrease in loan size rationing compared to initially older firms. For Initially small firms this additional decrease is 3.3 percentage points. The further firm and loan characteristics show that credit rationing also depends on the observable credit risk of the firm. Larger firms in terms of Disposable income and firms taking out a Fixed capital loan are less credit constrained. Since firms with more disposable income are less vulnerable in case of external shocks to their business and since fixed capital assets may be sold in case of default these loans may be considered as less risky. Besides, investments in fixed assets may be more difficult to be split which leaves less scope for loan size constraints. At the same time, firms that show a higher Leverage are more rationed. Surprisingly, Sole proprietorships which are considered to be less transparent than incorporated firms face lower credit constraints. Nevertheless, the Bank may assess them to be less risky because of their owners unlimited liability and because the firm management does not easily change. Having lending relationships with Other banks leads to a decrease in the degree of credit rationing. This may either stem from lower demand due to other available funds or from increased supply because firms with several bank relationships are the more profitable and vital ones. I will explore supply and demand issues in section 4.2. These results confirm hypotheses H1 and H2: credit rationing is higher for the group of opaque firms, and later loans are less rationed than first loans. To test whether H2 is also confirmed along individual borrower relationships, I introduce firm fixed effects in a next step and analyze the sample of repeat clients. 18

19 [Insert Table 5 here] Repeat clients The results from the repeat client analysis presented in column (2) show that hypothesis H2 is also confirmed for individual borrower relationships. Interestingly, the economic impact of the Loan number dummies is stronger than in the pooled analysis. Thus, when focusing on the variation of variables over borrowers loan sequences and controlling for unobserved borrower heterogeneity I can confirm that repeated interactions with the Bank and thus decreased informational asymmetries are an important determinant of the reduction of credit rationing. This implies that the decrease in the degree of credit rationing observed over time is not only due to a sample effect because bad borrowers drop out (the extensive margin effect) but also because individual borrowers experience less credit rationing in later stages of their bank relationships (the intensive margin effect). Besides, I find confirmation for the earlier result that borrowers who are observably less risky are less credit rationed (Leverage, Ln(Disposable income), Fixed capital loan). Column (3) adds those explanatory variables which are not defined for all first loans and therefore uses the second loan as the reference category. The results show that a previous Loan officer change leads to higher loan size rationing, but the economic effect is quite small (1.4 percentage points). 12 This confirms the conjecture of Berger and Udell (2002) that not all of the soft information gathered by loan officers can be transformed into common knowledge within the Bank The loan officer changes observed in the dataset mostly occur because loan officers are promoted within the Bank or because they leave the Bank. The Bank does not follow a policy to regularly rotate its loan officers internally to avoid too close relationships between clients and loan officers that might lead to decisions rather based upon personal considerations than objective judgements (see Hertzberg, Liberti and Paravisini (2010) for positive effects of loan officer rotation). 13 An alternative explanation for the impact of Loan officer change on credit constraints is that the borrower and the loan officer were colluding leading to better loan terms than the borrower risk would justify. Although there are a few relationships between borrowers and loan officers for which collusion might be a possible explanation 19

20 When a borrower was in arrears for more than 30 days at her previous loan (Previous arrears) the degree of rationing at the next loan is significantly higher again confirming that the Bank is also concerned with the observable credit risk of a firm when rationing credit. Finally, add-on loans are less rationed although they increase the overall exposure of the client at the Bank. This implies that the Bank only grants new loans before the full repayment of the previous loan if it has collected sufficient positive information about the borrower. Not surprisingly, the interaction effects between the Loan number dummies and Initially young or Initially small do not play any significant role in this specification since most of the withinfirm variation for these variables is between the first and the second loan. 4.2 Requested and granted loan amounts over loan sequences The analysis so far has provided evidence for the magnitude of credit rationing, its determinants as well as its evolution over bank-borrower relationships. The observed decrease in credit rationing over bank relationships may be interpreted as confirmation for the positive relation between close bank relationships and credit availability shown by previous studies that use indirect measures of credit constraints (e.g. Petersen and Rajan (1994, 1995)). But how are credit constraints relieved over time in bank-borrower relationships? Do banks increase the granted loan amount to meet borrowers demands, or do borrowers revise their loan requests downwards to the level which banks are willing to provide? The structure of my dataset enables me to disentangle whether the observed reduction in credit rationing is due to increasing stakes in bank lending or whether it is driven by processes on the demand side. 14 because they last up to nine interactions, the average number of interactions with a loan officer is 1.7 for repeat clients leaving little room for collusion. Therefore, it seems as if the partial loss of the acquired information during a loan officer change is the main driver of the observed increase in credit rationing after a loan officer change. 14 The analysis cannot incorporate whether borrowers are credit rationed at other banks nor whether borrowers would not prefer to realize a larger loan amount if it was possible. However, the structure of the dataset allows me to observe the evolution of both bank s granted loan contract terms as well as borrowers requests over multiple interactions and to draw conclusions on their relation from the results. 20

21 Table 6 reports results from a panel model with firm fixed effects for the determinants of granted and requested loan amounts, respectively. 15 All specifications include year-month effects to account for the macroeconomic environment as well as the Bank s prevailing loan contract terms for small loans at the time of loan disbursement. Standard errors are reported in parentheses and are adjusted for clustering at the firm level. [Insert Table 6 here] The regression results in column (1) show that granted loan amounts indeed increase over loan sequences. A Wald test for differences in adjacent loan numbers indicates that this increase is significant up to loan number four. For instance, granted loan amounts for the second loan are, on average, by 23.0% higher than for the first loan. The other firm and loan level variables show that Ln(Granted amount) is furthermore determined by the firm s observable credit risk. If firms are more indebted (Leverage) they are granted smaller amounts, while loans that finance a fixed asset (Fixed capital loan) which may be sold in case of default and Annuity loans with regular repayment schedules and thus lower risk show higher granted amounts. Finally, the existence of lending relationships with Other banks leads to higher granted loan amounts at this Bank indicating that firms with several lending relationships seem indeed to be the more successful ones. In column (2) the explanatory variables that are not defined for first loans are added and thus loan sequences with the second loan as the reference category are examined. It turns out that the fact whether a borrower s previous loan was credit rationed (Previously rationed) I repeat the same analysis with relative requested and granted loan amounts, i.e. requested and granted loan amounts scaled by total firm assets, and find qualitatively the same results. 16 This adds a dynamic component to the model because the extent of the previous rationing includes the granted amount of the previous loan (i.e. the lagged dependent variable). However, I have to refrain from estimating a dynamic panel model since in my setting the time intervals (between adjacent loan numbers) differ between sequences of observations. Using the dummy variable Previously rationed reduces the bias that would arise from the direct introduction of the lagged Share granted in a fixed effects regression. 21

22 does not influence the Bank s decision about the granted loan amount at the current interaction. The further firm level variables reveal that granted loan amounts are smaller after a Loan officer change (18.7% on average). As shown in section 4.1 this may be explained by the loss of proprietary information when loan officers are assigned new portfolios or leave the Bank. Finally, a borrower s previous repayment performance has a considerable impact on the amount granted for the current loan. A borrower who was in arrears for more than 30 days during the maturity of her previous loan (Previous arrears) receives a 31.5% lower loan amount at the current interaction. Turning to Ln(Requested amount), column (3) reveals that requested amounts also increase significantly over multiple interactions with the Bank (as indicated by a Wald test for differences in adjacent loan numbers) but on a smaller level than the granted amounts. Together with the significantly positive but again smaller coefficient for Other banks, this provides clear evidence that the observed decrease in credit rationing over loan sequences does not stem from borrowers reduced demand at this Bank due to their access to debt capital from other sources. Interestingly, column (4) shows that, in contrast to the Bank side, the fact of being Previously rationed seems to influence borrowers requested amounts. The significantly negative interaction terms between the Loan number dummies and Previously rationed indicate that those borrowers who experienced credit rationing at their previous loan increase their demand more moderately than the non-rationed borrowers. While models of credit rationing do not incorporate dynamic patterns on the demand side, the results above hint at borrower learning. Agarwal, Driscoll, Gabaix and Laibson (2008) study borrower learning in the credit card market and find that borrowers seem to learn to avoid paying future fees through negative feedback, i.e. the experience of past fees. Similarly, borrowers may learn from the negative feedback they receive from previous credit constraints how much they may reasonably request from this Bank and adapt their requested loan amounts accordingly. An alternative explanation is that credit rationing has real effects 22

23 on borrowers economic activities so that rationed borrowers prosper less than non-rationed borrowers and therefore increase their demand more moderately. This would also explain why the Bank does not increase granted loan amounts relatively more for previously credit rationed borrowers a result one would expect to find in the absence of real effects if the group of rationed borrowers was a pool of good and bad firms due to adverse selection. In this case, the Bank should be expected to make up for the previous rationing of the then revealed good firms. To shed some more light on the effects of credit rationing, the next section will examine the factors that influence the ex-post performance of loans. 4.3 Ex-post loan performance Table 7 reports marginal effects from Probit regressions on the determinants of Arrears. All specifications include industry-, branch- and year-month fixed effects to account for the industry- and location-specific as well as the macroeconomic conditions at the time of loan disbursement to capture external economic shocks that might impact on borrowers loan repayment performance. The main interest lies in examining whether the group of credit rationed borrowers performs differently from the group of non-rationed borrowers. [Insert Table 7 here] The results in column (1) show that the group of credit rationed borrowers (Share granted < 1) is significantly more likely to be past-due more than 30 days on interest or principal payments during the maturity of their loans than the group of non-rationed borrowers (borrowers receiving the amount they requested or a larger than requested amount). The difference of 1.7 percentage points is economically relevant given that the mean of Arrears amounts to only 4%. This supports the above reasoning that credit rationing may 23