Endogenous Product Differentiation in Credit Markets: What Do Borrowers Pay For?

Kim, Moshe; Kristiansen, Eirik Gaard; Vale, Bent Working Paper Endogenous Product Differentiation in Credit Markets: What Do Borrowers Pay For? Working Paper, No. 2001/8 Provided in Cooperation with: Norges Bank, Oslo Suggested Citation: Kim, Moshe; Kristiansen, Eirik Gaard; Vale, Bent (2004) : Endogenous Product Differentiation in Credit Markets: What Do Borrowers Pay For?, Working Paper, No. 2001/8, ISBN 82-7553-184-5, Norges Bank, Oslo, https://hdl.handle.net/11250/2498715 This Version is available at: https://hdl.handle.net/10419/209797 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/4.0/deed.no ENDOGENOUS PRODUCT DIFFERENTIATION IN CREDIT MARKETS: WHAT DO BORROWERS PAY FOR? Moshe KimEirik Gaard KristiansenyBent Valez February 5, 2004 Abstract This paper studies strategies pursued by banks in order to di¤erentiate their services and soften competition. More speci…cally we analyze whether bank’s ability to avoid losses, its capital ratio, or bank size can be used as strategic variables to make banks di¤erent and increase the interest rates banks can charge their borrowers in equilibrium. Using a panel of data covering Norwegian banks between 1993 and 1998 we …nd empirical support that the ability to avoid losses, measured by the ratio of loss provisions, may act as such a strategic variable. A likely interpretation is that borrowers use highquality low-loss banks to signal their creditworthiness to other stakeholders. This supports the hypothesis that high-quality banks serve as certi…ers for their borrowers. Furthermore, this suggests that not only lenders and supervisors but also borrowers may discipline banks to avoid losses. JEL code: G21, L15 Keywords: Banking, product di¤erentiation, certi…cation, market discipline. Forthcoming, Journal of Banking and Finance University of Haifa, Department of Economics, yNorwegian School of Economics and Business Administration and Norges Bank (The central bank of Norway). zNorges Bank (The central bank of Norway), Corresponding author address: Norges Bank, C51, Box 1179, Sentrum, N-0107 Oslo Norway. Fax: +47 22 42 40 62, e-mail: bent.v[email protected] 1. Introduction What do borrowers pay for? Are borrowers willing to pay higher rates to banks exhibiting higher reputation? If this is the case, some banks would invest in reputation for quality and di¤erentiate their services from their rivals, thereby softening competition. In this paper we focus on such endogenous di¤erentiation among banks. More precisely, which “quality” characteristics (equity ratios, loss avoidance, size etc.) do banks choose in order to di¤erentiate themselves from competing banks. There are two major reasons for borrowers to be concerned with bank quality. First, banks provide certi…cation which can be used to alleviate consequences of asymmetric information and to contribute to borrowers’value. By borrowing from a bank known to have a high-quality loan portfolio (i.e. low loan-loss provisions) a …rm can signal its creditworthiness to its other stakeholders. In this manner a high quality bank certi…es its borrowers.1Thus, banks can segment the markets according to borrowers’ willingness to pay for borrowing from banks with highquality loan portfolios and extract higher rents from those valuing certi…cation. Second, borrowers may be concerned with re…nancing. Re…nancing is of crucial interest for locked-in customers. Some borrowers may face large lock-in e¤ects due to the fact that their current bank has an informational advantage vis a vis competing banks (see Sharpe (1990)). These borrowers are inclined to choose banks that they anticipate are able to extend credit lines or provide new loans in future periods (switching to another bank is costly, see Kim, Kliger, and Vale (2003)). This suggests that bank characteristics that are informative about a bank’s ability to provide loans in the future, as re‡ected in bank solvency and diversi…cation (size), is important 0We are grateful to two anonymous referees who greatly helped in improving the paper. We appreciate comments from Sonja Daltung, Øyvind Eitrheim, Alois Geyer, David F. Hendry, David B. Humphrey, Tor Jacobsson, Kjersti-Gro Lindquist, Øivind Anti Nilsen, Henri Pagés. Thanks also to seminar participants at The Hebrew University of Jerusalem, the E.A.R.I.E 2000 conference in Lausanne, Sveriges Riksbank, the 2nd workshop of the Basel Committee on applied banking research, the 8th meeting of the German Finance Association, the EEA 2002 congress in Venezia, and at the University of Heksinki.Views and conclusions expressed are the responsibility of the authors alone and cannot be attributed to Norges Bank nor any of the persons and institutions mentioned above. 1See e.g. Cook, Schellorn, and Spellman (2003), James (1987), Lummer and McConnell (1989), and Billett, Flannery, and Gar…nkel (1995). 2 for borrowers.2Well diversi…ed and well capitalized banks will less likely face large losses and are more able to withstand potential losses. Locked-in borrowers may prefer such banks (see Chemmanur and Fulghieri (1994)).3 The major interest of the empirical part of this study is to distinguish between the certi…cation and re…nancing motives. If borrowers pay a premium for borrowing from banks providing certi…cation (low loan-loss provisions) or from solvent banks with few problems in meeting future re…nancing needs, banks face market discipline induced by borrowers. This asset side market discipline e¤ect is di¤erent from the conventional one on the liability side (uninsured deposit and money market funding), which has been extensively studied in the banking literature.4A possible disciplinary e¤ect from borrowers may reinforce the market disciplinary e¤ect stemming from the liability side and make banks less …nancially fragile. The issue of product di¤erentiation in banking has been of interest for some time. Generally, banks can pursue two kinds of di¤erentiation strategies. A bank can di¤er from other banks in a way that all customers consider as better than its competitors (e.g., better services). When customers agree about the quality ranking of di¤erent banks at equal prices, we call it vertical product di¤erentiation. In contrast, horizontal product di¤erentiation does not imply that all borrowers agree about such a ranking. For example, a bank may move a branch from one city to another, to the bene…t of customers in the latter city. The empirical literature on product di¤erentiation in banking has mainly been concerned with horizontal di¤erentiation. See Matutes and Vives (1996), Berg and Kim (1998), Barros (1999), and Kim and Vale (2001). Degryse (1996) theoretically analyzes the interaction of horizontal and vertical di¤erentiation. See also Anderson, De Palma, and Thisse 2See Detragiache, Garella, and Guiso (2000). 3Peek and Rosengren (1997) provide empirical evidence for a negative relation between loan losses at banks and their concurrent supply of loans. 4See for instance Calomiris and Kahn (1991) for a theoretical model explaining how depositors can discipline bank managers. Rochet and Tirole (1996) provide a theory of peer monitoring among banks in the interbank market. Martinez Peria and Schmukler (2001) and Gunther, Hooks, and Robinson (2000) provide empirical evidence of depositors disciplining banks’risk taking. 3 (1992).5The present paper, however, focuses on vertical product di¤erentiation since we are interested in the e¤ect of reputation for quality which is intrinsically a vertical di¤erentiation phenomenon.6 In the present paper we restrict our attention to debt taken from the banking sector only. This is because most European countries have relatively thin markets for arm’s length debt (bonds and certi…cates). OECD statistics show that bond and certi…cates as of 1995 comprised only around 4.0%–6.0% of total funding for the private non-…nancial …rms in Europe (see OECD (1996)).7 Before conducting the empirical analysis, we provide a stylized, two-stage, theoretical model which can shed some light on ways banks can utilize borrowerheterogeneity in order to di¤erentiate themselves. In the empirical part, we use data from the Norwegian banking industry to illustrate along which dimensions banks may …nd it most pro…table to di¤erentiate and soften competition. The paper is organized in the following way: section 2 presents our stylized theoretical model which illustrates some of the main forces behind product di¤erentiation; section 3 describes the data used, variables calculations, and the empirical model. Empirical results and discussion are presented in section 4. Section 5 concludes the paper. 2. A theoretical model In this section we introduce a stylized two-stage model which illustrates the product di¤erentiation e¤ect discussed above. In this theoretical model we are deliberately vague about exactly which strategic variables banks use in their vertical product di¤erentiation strategy. In the empirical part we analyze di¤erent potential “quality” variables that banks can use to 5For literature about relationship lending and/or competition in credit markets see for instance Boot and Thakor (2000), Petersen and Rajan (1994), Petersen and Rajan (1995), or Winton (1997). These papers, however, are silent regarding vertical di¤erentiation issues. 6In the empirical model we do however control for some horizontally di¤erentiated elements like geographic location. 7These particular OECD statistics are not published for the years after 1995. Note, however, that 1995 is in the middle of our data sample extending from 1993 to 1998. 4 di¤erentiate themselves. For simplicity, we study the case with two banks, bank Aand bank B. At stage 1, banks choose their quality variables, qi,i=A; B and, at stage 2, banks choose interest rates, ri; i =A; B (price competition). This two-stage structure captures the notion that some characteristics are used as strategic variables, i.e. variables more costly or di¢ cult to alter than interest rates. Figure 1 presents a schematic diagram of the two-stage game: Stage 1 Banks choose quality variables, qAand qB;simultaneously. Stage 2 Banks choose interest rates, rAand rB;simultaneously. Borrowers accept an o¤er from one of the banks. Figure 1: Competition in a two-stage game There are numerous potential ways a bank can distinguish itself from its competitors. If bank relationships are important, borrowers may be concerned about the capabilities or characteristics of their main bank. Let us here point out some potential quality variables in banking. Certi…cation ( signalling): Bank loans may signal the …nancial quality of the borrowing …rm to other creditors and shareholders. A loan from a bank known to have a low level of loan loss provisions provides a more favorable signal than a similar loan from a high-loss bank. Low loan loss provisions can result from high skills in screening and monitoring or from the bank being very risk averse. An outsider cannot directly observe from which of these two low losses originate. However, in both cases obtaining a loan from such a bank would serve as a certi…cation of high credit worthiness. In this way, a bank loan can be used to alleviate the asymmetric information problems a …rm may face in negotiations with, for example, suppliers, buyers, and other stakeholders. In their theoretical model Chemmanur 5 and Fulghieri (1994) also show how loans from a more reputable bank provide more information than loans from a less reputable bank. Furthermore the empirical study of Billett, Flannery, and Gar…nkel (1995) shows that loans from high-quality lenders are associated with larger positive stock price reactions than loans from low-quality lenders. Re…nancing (solvency): Empirical literature has shown that borrowers may su¤er if their main bank is forced to restrict its lending capacity (see Slovin, Sushka, and Polonchek (1993)). Consequently, a borrower may be concerned about their main bank’s solvency or, more precisely, how likely it is that their bank may face di¢ culties in providing loans in the future. Both a high capital ratio and low loss provisions are variables that contribute to a bank’s solvency. All else equal, a bank that is more diversi…ed would be less likely to su¤er losses that may reduce its lending capacity. As larger banks tend to be more diversi…ed than smaller ones, borrowers concerned about re…nancing would prefer borrowing from larger banks. Furthermore, borrowers may believe a larger bank is also more likely to be considered as “too big to fail”by the government. Borrowers are assumed to have access to an investment project with present value, V(not including …nancing costs). There is a continuum of borrowers indexed by Ron the unit interval [0;1] with unit density according to borrowers’increasing appreciation for banks’quality. By Rjwe denote borrower j’s quality appreciation. The scale parameter, 0, is introduced in order to study how more heterogeneity among borrowers (i.e. increase in ) may a¤ect competition and product di¤erentiation. A borrower of type Rjgains Rjqutility from borrowing from a bank with quality q. As an example, a borrower who does not need re…nancing in the future has a low R(possibly 0). In contrast, a borrower who is locked into a relationship with a particular bank (high switching cost) and needs re…nancing in the future, would have a high R.8Banks cannot observe the Rs but they are well aware of the distribution 8See for example Sharpe (1990) and von Thadden (1998) for a discussion of switching costs due to information asymmetries between lenders. 6 of Rs in the economy. Furthermore, we assume for simplicity that a bank’s cost, e(q), associated with choosing a quality level, q, di¤erent from the cost minimizing level, q0, is quadratic, e(qi) = (qiqo)2i=A; B where is a positive parameter. Note that the cost minimizing quality level, qo, can be interpreted as the quality level that a bank would have chosen if borrowers did not appreciate the quality level in question. Note also that if both banks choose the cost minimizing level of q(i.e. q0=qA=qB), banks would o¤er identical services and competition would be …erce. Hence, banks have incentives to deviate from q0 and thereby soften competition. To …nd the sub-game perfect pure-strategy equilibrium in the two-stage game we start with stage 2. 2.1. Competition at stage 2 First, let us examine the demand for loans given qA,qB,rA, and rB:Without loss of generality assume qAqB, which implies that rArB(otherwise bank B’s o¤er dominates bank A’s o¤er). Borrower jcompares the net bene…ts from using bank Aand bank B:9 Bank A:VrA+qARj Bank B:VrB+qBRj A borrower of type b R, is indi¤erent between using bank Aand bank B, VrA+qAb R=VrB+qBb R b R=rArB (qAqB). 9For simplicity we have assumed that the project has a certain outcome. However, we could have assumed that there is a probability p < 1for success. In case of failure the project is worthless. Then, the expected value of the project would have been: p[Vri+qiRi]. The choice between the two banks would, however, not have changed. 7 Consequently, bank Aand bank Bface demand DA(rA; rB)and DB(rA; rB), respectively, DA(rA; rB) = 8 > < > : 0if b R1 1b Rif 0b R1 1if b R0 DB(rA; rB) = 8 > < > : 0if b R0 b Rif 0b R1 1if b R1, and the banks’stage-2 pro…t levels are A(rA; rB)=(rAr0)DA(rA; rB) B(rA; rB)=(rBr0)DB(rA; rB), (2.1) where r0is the banks’cost of funding. From the two banks’pro…t maximizing choice of interest rates, we get the Nash equilibrium at stage 2: rA=2 3(qAqB) + r0 rB=1 3(qAqB) + r0. (2.2) From, equation (2.1) and (2.2) we have A(qA; qB) = 4 9(qAqB) B(qA; qB) = 1 9(qAqB). (2.3) Notice that there are two e¤ects stemming from a change in a bank’s quality variable on the equilibrium interest rate charged. First, there is a direct e¤ect on the demand for its loans. If bank quality improves, borrowers are willing to pay higher interest rates. Second, there is an indirect competition e¤ect on the equilibrium interest rate charged. If bank A(the high quality bank) improves its quality, the two competing banks will become more di¤erentiated and competition is softened. Hence, both 8 and partially by lagged LHS variables. Furthermore, the macroeconomic part of the loan portfolio risk is controlled for by the time dummies and the regional speci…c part by the regional dummies.16 In what follows we present the de…nition of the RHS variables used to estimate (3.2): Variable Description si;t1,si;t2Lags of the spread of interest rate on credit lines Bank quality v(q)i;c;th: v(assets)i;c;t1Total assets of bank iend of year t1 v(cap88)i;c;t1Capital ratio(Basel 88) of bank iend of year t1 v(loss)i;c;t1Ratio of accumulated loss provisions to loans outstanding for bank iend of year t1 Gini coe¢ cients of quality g(q)c;th: g(assets)c;t1 g(cap88)c;t1 g(loss)c;t1 Controls (xi;t;fc;t;dummies): costrati;t Ratio of materialsand wage cost to loans outstanding for bank iin year t herfinc;t Her…ndahl index of the bank to business credit market in county cin year t i,c,tBank, county and year dummies v(q)i;c;t1is a vector representing the di¤erence between the value of bank i’s quality variables and the cross-sectional median of the corresponding bank quality variables in county cin period t1.g(q)c;t1is a vector containing for each bank quality variable a measure of the inequality in that variable across banks in county cin period t1. All lagged stock variables are aggregated backwards, i.e. the bank structure of year tis forced upon the variable in year t1. The variables listed under the heading ‘bank quality variables’v(q)i;c;t are vari16 In a previous version of the model we used another way of controlling for borrower risk. See discussion in section 4. 15 ables that borrowers are likely to take into account as signals by banks when choosing a bank. The operator vrepresents the cross-sectional di¤erence of a quality variable qfrom its relevant market’s median in the following way17: v(q)i;c;t =qi;t median i2c(qi;t). i2cstates that the median is calculated only over banks operating in county c. Note that even if qi;t varies only across banks and years, v(q)i;c;t will also vary across counties. As mentioned earlier, the dispersion of quality variables is of great importance since it softens competition. For dispersion we use the Gini measure of inequality calculated as follows: g(q)c;t = 1 + 1 nc 2 n2 cqc;t X i2c jqi;t where qc;t =1 ncX i2c qi;t ,j= 1;2;3;   ; nc, where ncis the number of banks operating in county c, and jis a rank number assigned to each qi;t in decreasing order of size. assets represents the size of a bank. The larger the bank the more diversi…ed its portfolio is likely to be, and all else equal, the less likely it is that the bank will su¤er huge losses and be forced to reduce its lending activity. Furthermore, borrowers may believe that a larger bank is also more likely to be considered as “too big to fail”by the government. cap88 (capital ratio (Basel 88)) represents the solvency of a bank in terms of its ability to withstand large loan losses without being forced to cut its lending in order to satisfy the capital requirements. This variable can have a positive impact on the spread, if borrowers are willing to pay for this sign of quality. This may be so if they need future re…nancing, and are locked in, as described earlier. loss (ratio of accumulated loss provisions to loans outstanding) represents the results of the bank’s ability to screen and monitor, as well as a bank’s willingness to 17 When taking the ln of v(q)i;c;t;then this variable will be the ln di¤erence from the median. 16 take on risk in its loan portfolio. Thus borrowers who need to signal their quality to their other stakeholders can do so by borrowing from a bank that has su¤ered few loan losses (see the discussion of the certi…cation role of banks in section 2). Low loss provisions will also increase the probability that the bank can maintain its solvency and hence its capacity to re…nance borrowers in the future. To the extent borrowers are willing to pay for this quality variable, the expected impact of v(loss)i;c;t1on the spread is negative. The accumulated loss provisions is a good indicator of the credit quality of a bank’s loan portfolio. According to the accounting standard for Norwegian banks in force during our sample period, banks are required to increase their loss provisions when and only when they get information indicating that the credit quality of their portfolio has deteriorated. This can be information on speci…c loans, for instance a borrower defaulting on his payments of interest and installments, or speci…c information regarding the overall loan portfolio. Thus, accumulated loss provisions in this data set are not general reserves that banks are allowed to set aside in good times independently of the level of risk in their loan portfolios. The expected sign of the estimated parameters for all the Gini coe¢ cients are positive. More dispersion among banks in terms of variables borrowers care about, serves to soften competition and hence increase the interest rate banks can charge their borrowers. Recall that a key assumption –and a fairly realistic one –in this paper is the heterogeneity of borrower preferences. If one of the Gini coe¢ cients of an underlying variable turns out to be insigni…cantly di¤erent from zero, this indicates that more dispersion among competing banks along this variable does not soften competition. Under our assumption about heterogeneous borrowers this also implies that the underlying variable is not a quality variable as de…ned in section 2. We use a lag of one year for all the quality variables. Borrowers have to base their evaluation of the bank on the values published in the bank’s annual report and …nancial statements for the last year. These are usually more comprehensive and more scrupulously audited statements than the quarterly statements made during the year. 17 Among the control variables, costrati;t (ratio of materials and wage cost to loans outstanding) represents the banks’ability under imperfect competition to pass their operating costs on to their credit line borrowers. The regional Her…ndahl index herfinc;t (Her…ndahl index of the bank to business credit market in county cin year t) controls for the competitive environment, as measured by market concentration, in which a bank operates. The more concentrated the market is the higher is the value of the Her…ndahl index. A more concentrated market is usually considered a less competitive market, and banks should be able to charge a higher interest rate. Hence the expected sign of this variable should be positive. However, it could also have a negative sign due to the ‘winner’s curse’ problem discussed in auction theory.18 The dummies control for bank, regional, and time speci…c e¤ects. Our model will support the re…nancing hypothesis if the variables relating to banks’future lending capacity (the capital ratio, assets (size/diversi…cation), and loan loss provisions) are found to be signi…cant quality variables. However, if only low loan loss provisions are found to be a signi…cant quality variable our model would support the certi…cation hypothesis and not the re…nancing hypothesis. 4. Results The model presented in (3.2) is estimated using two-stage least square. costrati;t is endogenous, it may be partially determined by the LHS variable si;t. It is therefore instrumented using its own one year lag, not aggregated backwards.19 The correlation between ln costrati;t and its lag is 0.90. We start by estimating the general model including all the RHS variables listed in section 3.2, The results are presented in Table 4.1 column (a), and indicate a model that satis…es certain misspeci…cation tests regarding lack of serial correlation in the residuals and no functional form misspeci…cation. 18 See for instance Bulow and Klemperer (2002) who construct a theory model of auctions where a reduction in the number of bidders actually raises the price when bidders are asymmetric. 19 Backward aggregation of a variable means that the bank structure of year tis forced upon the variable in year t1. 18 Table 4.1: Empirical results LHS variable ln si;t (credit line interest rate spread over money market interest rate) Variable (a) (b) ln si;t10:0296 (0:89) 0:0252 (0:76) ln si;t20:0230 (1:11) 0:0174 (0:85) Bank quality v(q)i;c;t1: ln v(assets)i;c;t10:0278 (1:31) – ln v(cap88)i;c;t10:1389 (3:19) 0:1562 (4:37) ln v(loss)i;c;t10:1458 (5:68) 0:1193 (6:23) Gini coe¢ cients of quality g(q)c;t1: ln g(assets)c;t10:0266 (0:09) – ln g(cap88)c;t10:061 (0:97) – ln g(loss)c;t10:1552 (3:16) 0:1528 (3:91) Controls (xi;t;fc;t;dummies): ln costrati;t 0:8217 (3:52) 0:7880 (3:43) ln herfinc;t 0:0396 (0:54) – iin in cin – tin in F-test, (a) –(b) – 0:81 AR(1,2) 0:19 0:62 RESET 0:91 0:94 R2adj. 0:4804 0:4869 Number of observations is 1241. v(q)i;c;t1is a vector representing the di¤erence between the value of bank i’s quality variables and the cross-sectional median of the corresponding bank quality variables in county cin period t1.g(q)c;t1is a vector containing for each bank quality variable a measure of the Gini coe¢ cient of that variable across banks in county cin period t1. Numbers in parentheses are White heteroscedasticity consistent t-values. The F-test is a test of the joint signi…cance of the variables excluded from model (a), the p-value is reported. AR(1,2) is a joint Breusch-Pagan test for …rst and second order serial correlation in the residuals. Pvalues for the F-test are reported (see Greene (1993) p. 428). RESET is the test for functional form using the square of the predicted value as RHS. P-values of the t-test are reported. 19 We …nd the following variables insigni…cant; ln v(assets)i;c;t1,ln g(assets)c;t1, ln g(cap88)c;t1,ln herfinc;t, and the county dummies. Exclusion of these insigni…- cant variables is statistically valid, as is shown by the reported F-test. Thus, we get the parsimonious model (b) which also passes the tests for functional form and for no serial correlation in the residuals. Note that due to the log-linear speci…cation all coe¢ cients of the model can be interpreted as elasticities. The re…nancing hypothesis implies that borrowers care about bank characteristics which indicate to what extent a bank will be able to stay behind its borrowers and extend loans in the future. In line with the previous discussion this hypothesis will gain support if size, capital ratio, and loan losses turn out to be signi…cant quality variables. In contrast, if borrowers only care about certi…cation (signalling), only the quality of a bank’s loan portfolio (loan losses) would be signi…cant. Our results lend support to the certi…cation hypothesis since only loan loss provisions (quality of the loan portfolio) turns out to be a signi…cant quality variable. Size and capital ratio turn out to be insigni…cant as quality variables. Consequently the re…nancing hypothesis is not supported by our empirical results. Banks can charge a premium to borrowers that want to signal their credit worthiness by borrowing from a low loss bank. Furthermore banks can segment the markets according to borrowers’willingness to pay for using low-loss banks. Borrowers’appreciation of banks with low loss provisions serves as an important disciplinary device, inducing banks to avoid losses. To illustrate the strength of this e¤ect, consider a bank at sample mean with an interest rate spread on its credit line loans of 4:74 pct. It will according to our results be ‘punished’by a reduction of the interest rate spread in the range of 0:38 to 0:74 pct. points, if its loss provisions relative to its competitors double.20 This suggests that there is a market discipline e¤ect at work not only in the money market, but also in the market for credit line loans. Both banks’lenders and borrowers punish banks with high loan losses. The negative and signi…cant sign for ln v(cap88)i;c;t1may be explained by different degrees of risk aversion among banks: Banks with high degree of risk aversion 20 This range is calculated as a 95 pct. con…dence interval. 20 choose to operate with both a high capital ratio – to minimize the possibility of moving below the minimum requirement –and at the same time lend to safe borrowers, borrowers from which they only can charge a low interest rate. This may explain why well-capitalized banks charge low interest rates.21 However, this result is not robust to shortening the length of the sample, as ln v(cap88)i;c;t1becomes insigni…cant when the two …rst years are taken out (see the discussion of robustness below). Among the control variables the coe¢ cient of the costrati;t is positive and significant. This may indicate that banks operating under imperfect competition in the market for credit line loans are able to pass some of their operating costs over to these borrowers. However, neither this is a robust result, since its coe¢ cient turns insigni…cant when the time length of the sample is shortened. As the Her…ndahl index does not obtain a signi…cant coe¢ cient we cannot say which is the more important theory; the traditional view of more concentrated credit markets leading to higher interest rates or the theories of ‘winner’s curse’. Our results that borrowers facing high switching costs do not seem to care about the future lending capacity of their bank, may stem from the way the banking crisis in the early nineties was handled by the Norwegian government. All banks –with one minor exception –were recapitalized or merged into other larger banks, such that lending activities could continue.22 This may explain why borrowers are not concerned with bank solvency. In fact in most industrialized countries facing a banking crisis, the crises have been resolved in similar ways by capital injection or even government takeover of the failed banks, see (Lumpkin, 2002, p. 123). As some of the government induced mergers during the banking crisis were not fully implemented until 1994, borrowers may not have rationally anticipated the 21 Similarly, a bank very close to or even below the minimum capital requirement may behave like a risk lover by lending to high-risk borrowers from which it charges a high interest rate. 22 In fact Ongena, Smith, and Michalsen (2003) …nd that …rms listed on Oslo Stock Exchange that maintained a banking relationship with any of the problem banks during the announcements of the banks’distress events, on average only had small and temporary negative excess returns around the distress announcement dates. Furthermore, Vale (2002) …nds evidence that small …rms borrowing from problem banks were not negatively a¤ected due to their bank relations. 21 outcome of this process as early as of 1993 or 1994. To account for this we reestimated our model, …rst taking out the year 1993 and then also leaving out 1994. The negative and signi…cant coe¢ cient for the capital ratio remained when just 1993 was taken out but it became insigni…cant when both 1993 and 1994 were taken out. The e¤ect of operating costs, however, was insigni…cant in both subsamples. However, our main result of low loan loss provisions as a signi…cant quality variable was not changed in any of these overlapping subsamples. This further strengthens the hypothesis of the certi…cation role of banks when it comes to market discipline from borrowers. In a previous speci…cation of the model, instead of using dummies to control for borrower risk, we used the loan loss ratio on credit line loans and the real money market interest rate. Our main qualitative results were similar to those of the model presented here. This indicates that our main results are relatively robust to di¤erent ways of controlling for borrower risk. The current speci…cation, however, performs better in terms of misspeci…cation tests than the previous one.23 This paper lends further support to the hypothesis of banks as certi…ers already found in existing litterature. James (1987) …nds a positive stock price response to the announcements of bank loans, whereas Lummer and McConnell (1989) …nd evidence that favourable loan renewals in particular give excess return to the stock issued by the borrowing …rm. Billett, Flannery, and Gar…nkel (1995) demonstrate that the equity response increases with the credit rating of the lender. In a recent paper Cook, Schellorn, and Spellman (2003) examine a sample of syndicated loans and show that lenders can extract a certi…cation premium from borrowers, particularly so when collateral is missing. In the present paper we also …nd evidence of certi…cation premiums for uncollateralized loans –lines of credit. This, however, relates to all loan sizes and not just to loans granted to publicly quoted …rms, as in the litterature mentioned above. Furthermore we demonstrate how banks’ability to act as certi…ers can be used strategically when banks compete in credit markets. 23 To further check the robustness of our results we reestimated the model for the whole sample including other potential quality variables like the size of a bank’s branch network and the liquidity of a bank. Neither of these two quality variables came out signi…cant however, nor did their inclusion have any impact on the other estimated coe¢ cients. 22 5. Concluding remarks In this paper we have studied strategies pursued by banks to di¤erentiate their services from those of their rivals and thereby soften competition. More speci…cally we have analyzed if the bank size, a bank’s ability to avoid losses, and its capital ratio can be used as such strategic variables. We also study to what extent borrowers are willing to pay for high quality along these dimensions. Using a panel of data covering Norwegian banks between 1993 and 1998 we did not …nd evidence for the use of high capital ratio as a strategic variable that borrowers are willing to pay for. This …nding may be explained by the way the banking crisis in the early nineties was handled. We do, however, …nd empirical support for the banks’ability to avoid losses, as a strategic variable, indicating that the quality of a bank’s loan portfolio is used to certify the credit worthiness of borrowers. This implies that borrowers in the market for credit line loans can discipline banks to avoid future losses. Hence, banks may face market discipline not only from the liability side (extensively discussed in the litterature), but also from the asset side. This further strengthens the arguments for putting more emphasis on Pillar 3 of the Basel II which promotes transparency and market discipline. References Anderson, S., A. De Palma, and J.-F. Thisse (1992): Discrete choice theory of product di¤erentiation. MIT Press, Cambridge and London. Barros, P. (1999): “Multimarket competition in banking with an example from the case of Portuguese market,”International Journal of Industrial Organization, 117, 335–352. Berg, S. A., and M. Kim (1998): “Banks as multioutput oligopolies: An empirical evaluation of the retail and corporate banking markets,”Journal of Money, Credit, and Banking, 30, 135–153. Berger, A. N., and G. F. Udell (1995): “Relationship lending and lines of credit in small …rm …nance,”Journal of Business, 68, 351–381. 23 Billett, M. T., M. J. Flannery, and J. A. Gar…nkel (1995): “The e¤ect of lender identity on a borrowing …rm’s equity return,”The Journal of Finance, 50, 699– 718. Boot, A. W., and A. V. Thakor (2000): “Can Relationship banking Survive Competition?,”Journal of Finance, 55, 679–713. Bulow, J., and P. Klemperer (2002): “Prices and winner’s curse,”RAND Journal of Economics, 33, 1–21. Calomiris, C. W., and C. M. Kahn (1991): “The role of demandable debt in structuring optimal banking arrangaments,”American Economic Review, 81, 497–513. Chemmanur, T. J., and P. Fulghieri (1994): “Reputation, renegotiation and the choice between bank loans and publicly traded debt,”Review of Financial Studies, 7, 475–506. Cook, D. O., C. D. Schellorn, and L. J. Spellman (2003): “Lender certi…cation premiums,”Journal of Banking and Finance, 27, 1561–1579. Degryse, H. (1996): “On the interaction between vertical and horizontal product di¤erentiation: an application to banking,”Journal of Industrial Economics, 44, 169–186. Degryse, H., and P. van Cayseele (2000): “Relationship lending within a bank based system: evidence from European small business data,”Journal of Financial Intermediation, 9, 90–109. Detragiache, E., P. Garella, and L. Guiso (2000): “Multiple versus Single Banking Relationships: Theory and Evidence,”Journal of Finance, 55, 1133–1161. Greene, W. H. (1993): Econometric Analysis. Macmillan, New York. Gunther, J. W., L. M. Hooks, and K. J. Robinson (2000): “Adverse Selection and Competing Deposit Insurance Systems in Pre-depression Texas,”Journal of Financial Services Research, 17, 237–258. James, C. (1987): “Some evidence on the uniqueness of bank loans,” Journal of Financial Economics, 19, 217–235. Kim, M., D. Kliger, and B. Vale (2003): “Estimating Switching Costs: The Case of Banking,”Journal of Financial Intermediation, 12, 25–56. 24