Firm and owner level characteristics of SME lending relationships : evidence from the Survey of Small Business Finance

(1)

Firm and owner level

characteristics of SME

lending relationships

Evidence from the Survey of Small Business

Finance

Fabian Fourné -‐ 10225579

B. Sc. in Economics - University of Amsterdam

Thesis supervisor: Ieva Sakalauskaite

Final version - 22.02.2015

Abstract: This paper empirically measures the effect of firm and owner level characteristics on the strength of the relationship between small and medium sized enterprises and their

financial intermediaries. Using data from the Survey of Small Business Finance and employing a multiple imputation model, we find that firm level characteristics tend to be more pronounced than owner level characteristics for both the length of the relationship and the degree of exclusivity. In particular we confirm existing theories that traditional measures of risk such as credit scores play an important role determining the strength of a relationship between borrowers and lenders. We also test existing theories towards the effects of non-‐conventional determinants of strong relationships in the context of relationship banking. More specifically the size of the firm, the distance to the lender and interestingly the use of computer technology is related to the strength of the relationship. These findings could for instance be used to identify SME subgroups that require special support in terms of subsidized loans, because the subgroup’s relationships tend to be less stable.

(2)

Acknowledgements

I would like to take this time to thank my thesis supervisor, Ieva

Sakalauskaite, for her support and guidance. I am extremely grateful for the advice and suggestions provided by Ieva throughout the entire process. Additionally, I would like to thank the numerous, anonymous Stata Forum members that without this study would not have been possible.

(3)

In the current research paradigm in small business lending academics and policy makers much focused attention on the needs of banks and other financial institutions to enhance credit issuance to small and medium sized enterprises. It is well known that in order to decrease information asymmetry between financial institutions and borrowers, the parties engage in a relationship – a practice commonly referred to as relationship banking. Despite the numerous findings on the origin, benefits and costs of relationship banking, researchers know surprisingly little about the determinants of the strength of the relationship from a firm and owner level perspective.

This lack of knowledge of what constitutes a strong relationship from the firm and owner perspective is the more surprising considering the importance of small and medium sized enterprises in the US economy. According to U.S. Census Bureau data firms with fewer than 500 workers accounted for 99.7% of all businesses in the US. Further, the U.S. Small Business Administration office found that in 2008 roughly half of the US nonfarm GDP was generated by SMEs. Next to large share of GDP, SMEs are a major driver of innovation. Data showss for instance that employees of small enterprises are up to 16 times as likely as employees of other firms to file a patent. Lastly and often considered most important by policy makers, employees of SMEs account for half of total employment in the US and small firms accounted for more than 63% of the jobs created between the nineties and today.

Therefore, after addressing the most pressing problems in the banking market in the post 2008 recession, policy makers turned their attention to the large group of SMEs to address problems directly. SMEs are generally more prone to swings in the credit market, because SMES heavily rely on traditional forms of lending such as loans and cannot easily obtain other forms of funding on for instance the public debt market. In addition, lenders typically have problems to determine the creditworthiness of applicants for small business loans, as the heterogeneity across small businesses in combination with widely varying use of the obtained funds prevented the development of standardized small business loan applications. Similarly, on the borrower side, small firms often cannot provide accurate financial statement data that would typically be used to assess credit worthiness in the case of transaction banking.

(5)

Consequently both borrowers and lenders built a relationship that helps the lender to access proprietary information about the firm based on repeated interaction.

Thus, when banks largely stopped lending to SMEs in the crisis, arguably stronger relationships could have helped to soften the credit crunch. Considering the the lack of information on the determinants of strong relationships on the demand side, my research on firm and owner level characteristics of strong relationships helps to explore unknown factors on the firm’s side and may thus help to implement better policies that help to nurture the growth of SMEs. In practice, my research may for instance unravel a statistical relationship between the firm size or location of a firm on the strength of the relationship. For example, if rural firms would have stronger relationships with banks or female owner would have weaker relationships with banks, policy makers could implement loan-subsidizing programs for the most vulnerable sub-groups of firms or owners.

The purpose of this thesis therefore is to explore the firm and owner level characteristics that determine the strength of this relationship between borrowers and lenders. The next section will shortly outline the foundations of relationship banking and also stress the added value of my research. I continue with a data description and methodology before I outline the results. Eventually I will present a conclusion and show an opportunity for further research as well as highlight the limitations of my findings.

2.1 Literature Review

The literature on relationship banking suggests that financial intermediaries exist because they enjoy economies of scale and have a comparative advantage in the production of information about borrowers (Berger & Udell, 1994). In contrast to large corporations that can raise funds on the public debt market small and medium sized enterprises often do not have access to the public debt market and thus rely on commercial loans. Diamond argues, that the borrowing decision changes over time. Less mature and smaller firms may choose to borrow funds from bank, because in this case they can take advantage of building a relationship with the bank loan officer and thereby reduce their cost of capital.

In the case of SMEs a strong relationship may often be needed to access soft information because information on financial measures may not be available due to

(6)

the high costs that are associated with collecting them. Berger & Udell (1994) for instance find that many SMEs struggle to comply with accounting standards that are widely used by the financial industry. As a result, the SME’s cost of financing is higher and they turn towards commercial loans, which often do not require the same standards as more complex financial instruments such as bonds.

Previous research on SME Finance mostly focused on the availability and cost of financing. Although these measures constitute important figures, this paper focuses on the strength of the relationship and thus will provide a short overview of previous research on the relationship strength between SMEs and lenders. Further, this review will outline possible advantages and disadvantages of relationship lending.

2.2 Benefits associated with relationship banking

Over the past two decades the literature on relationship banking mostly exhibits benefits to borrowers from strong relationships. According to Berger the research often finds that small businesses benefit from strong relationship by having better credit availability, as measured by a higher loan application rate, less dependency on costly alternatives such as trade credits, or by receiving more loans with less strict collateral requirements (e.g. Petersen and Rajan 1994, Berger and Udell 1994).

Further, U.S studies not only find a favorable effect of strong relationships on credit availability, but also a positive effect on the cost of credit (e.g. Petersen and Rajan 1994, Berger and Udell 1995). This effect may not hold true in other markets, as research in Europe did not confirm the existence of a positive effect of strong relationships on the rate of credit (Elsas & Krahnen, 1998, Machauer & Weber, 2000).

However, for instance a study of Italian manufactures by Herrera & Minetti (2007) established a link between the relationship strength and innovation by borrowing firms. Similarly, a paper about the role of the strength of the relationship in the onset of distress (Hoshi, Kashyap, & Scharfstein, 1990) finds that firms in Japan that have close ties to their banks are less likely to face liquidity constraints when investing than firms that do not have such ties. According to a study from James & Wier (1990) even stock markets perceive shares from firms with strong ties to financial intermediaries as less risky and see strong relationships with house banks as a positive sign for financial prudence.

(7)

2.3 Costs associated with relationship banking

Despite the numerous positive effects of strong relationships between borrowers and lenders, the literature also cites downside that can emerge from strong relationship – in particular when the relationship is exclusive. Firstly, borrowers that have an exclusive lending relationship with a lender expose their private information to single lender, possibly giving the bank a considerable market power of the firm. In a situation where the bank has market power over the firm (i.e. the bank is the single lender) Sharpe (1990) and Rajan (1992) for instance observed above average extraction of rents of the bank from the firm. To mitigate the risk that arises from dealing with a single bank, firms may thus establish relationships with multiple banks, resulting in duplicative costs. Secondly, firms may also choose to bear the cost of multiple banking relationships to protect themselves for instance from the risk stemming from financial institutions that become financially distressed in crisis periods (Berger, 2014). The literature on this issue is inconclusive and only finds ambiguous results, meaning that no consistent effect of bank fragility on the probability of multiple banking relationships is clear (e.g. Detragiache, Garella and Guiso 2000, Berger, Klapper, Martinez Oerua and Zaidi 2008). Thirdly, firms may have to bear the cost of multiple banking relationships if their main bank cannot offer the range of financial services needed (Berger 2013). This could be the case for local firms that outgrow their home markets and consequently require international services or specialized investment products and indeed the literature shows that larger firms are associated with multiple banking (e.g. Machauer and Weber 2000, Ongena and Smith 2000).

2.4 The dimensions of the strength of a relationship

Berger (2014) suggests that the relationship strength can be measured along three dimensions: Relationship length, exclusivity and depth. The relationship length allows banks to monitor firms over time and accumulate soft information through repeated interaction. Further, Boot (2000) cites the potentially increased trust of loan application officers towards firms they have previously interacted with. In economic terms we may refer to effect of the length of the relationship as a reduction of costs associated with information asymmetry.

(8)

Secondly, the literature cites exclusivity as an important indictor for the strength of the relationship. From the bank’s perspective an exclusive relationship could translate to a more efficient access to proprietary business information, which in turn would lead to a better relationship. If for instance a small firm interacts with multiple banks its resources may not suffice to provide adequate information to each of the banks. As a result, the strength of relationship would decline. On the other hand the exclusivity indicator has an ambiguous effect, which is outlined in more detail in the next section.

Lastly, authors like Boot (2000) also cite the depth of the relationship as an important measure of the strength of relationship. Typically, the depth of the relationship is measured as engagement between the lender and borrower in terms of for instance the number of deposit or saving accounts, outstanding credit, total number of deals in the history of the relationship or other metrics. As the third dimension of the strength of the relationship –depth- is difficult to objectively measure and the data set excluded most propriety information on such data, the empirical analyses of this paper will focus on the two other dimensions of the strength of the relationship: The length of the relationship and the exclusivity.

In the next section, I present a short description of the variables I use to gauge the strength of the relationship along with a short explanation of the expected effects on the respective dimensions of the endogenous variables. This paper adds to current literature a new perspective on the determinants of the strength of the relationship between borrowers and lenders. Previous research has much focused either on the credit availability or credit terms on one side and the banking structure on the other side. Contrary, my research highlights the borrower’s perspective and may guide the interested reader to understand the constraints policy makers face when they attempt to adapt legislation to favor small and medium sized enterprises in the market for credits.

(9)

3.1 Data description

3.2 Data

The data used in this analysis was collected by the 2003 Survey of Small Business Finance (SSBF). The 2003 SSBF is the fourth iteration of a survey on SMEs initiated by the Federal Reserve, following the surveys of 1987, 1993 and 1998. The set contains information on 4,240 small businesses that were in operation in December 2003 and at the time of the interview (between June and December 2004). After data cleaning that was required due to missing observations, the final sample size is 4035. This analysis uses the public data set of the survey, meaning that all information that could identify particular businesses, owners, or the financial resources used by business are excluded. Authors like Berger had exclusive access to the proprietary version of the survey and thus were able to merge the data with a parallel survey on banks that engaged in relationship banking Berger, Goulding, & Rice (2014). As a result, Berger’s analyses focused on the bank specific variables that determine the strength of the relationship. In contrast, my analyses will focus on the firm and owner specific characteristics that determine the strength of the relationship.

According to the survey administrators at the University of Chicago the target population of the SSBF provides information about a nationally representative sample of small businesses in the US. The target population is defined as follows: “All for-profit, nonfinancial, nonfarm, nonsubsidiary business enterprises that had fewer than 500 employees and were in operation as of year-end 2003 and on the date of the interview” (SSBF Codebook, 2003).

3.3 Method

The survey includes variables on a firm specific level, such as firm size, age, risk rating, type of organization and variables on owner specific level such as the ethnicity, gender and education of the owner. The sample design was random sample, stratified by census region, urban or rural location, and by employment size (less than 50 employees, 50 to 100 employees, more than 100 employees and less than 500 employees). The following regression analysis takes account of the survey design by adjusting for respective survey weights and the stratification; hence the data represents approximately 5.9 million small businesses.

(10)

As is true for virtually all survey data, the SSBF has a considerable amount of missing data that results from respondents who either choose to skip the respective question or who were not able to answer the question appropriately. To account for missing values and counter concerns regarding missing value bias, the authors of the survey build a multiple imputation (mi) model, based on a randomized regression model. The mi regression models every variable as a function of the other survey variables and estimates a value for missing observations. As a result the survey includes five implicates, yielding a total of 21,100 (5*4220) observations. That means that imputed values may differ across imputations, but values reported by respondents will not differ across imputations. As my regression model takes account of the imputation model employed by the authors of the survey, the standard errors of the coefficients of my model will likely be larger, but a potential risk for biasedness is reduced (Rubin, 1976). Reported t-statistics and p-values are already adjusted for the multiple imputation model. Unfortunately, also the summary statistics are affected by the multiple imputation model, hence the respective estimated means in will be displayed with 95% confidence intervals. Although multiple imputation models are widely used in social sciences and in particular in settings with survey data, one must interpret the results with caution. One crucial assumption of the multiple imputation model is that observations are missing at a random. If for instance a certain subgroup of respondents is less likely to respond to certain questions the estimated coefficients of any regression that is based on the imputation model are biased (Rubin, 1976). My analysis from the observation missing tables yielded no clear pattern of missing values; hence the regression model will assume a random pattern of missing values.

3.4 Sample

Table 3-1 shows the summary statistics of all variables of interest. In the following paragraph, I will give a short overview on how to interpret the variables and outline where applicable their predicted effects based on the literature. As outlined previously, the two main determinants of relationship strength are the length of the relationship between the firm and its financial institution and the degree of exclusivity, thus the following theoretical framework for the predicted effects of variables will analyze the effects with respect to the length of the relationship and with respect to the degree of exclusivity. The dependent variable of the Eq. 4-1 is the

(11)

length of the relationship between the firm and its primary financial institution. The mean length of the relationship in the sample is 125 month (approx. 10.5 years). The second equation aims to determine an empirical relationship between the firm and owner level characteristics and the degree of exclusivity. For Exclusive = 1, the binary variable indicates an exclusive relationship with a bank, meaning that the firm has only one relationship with a financial institution. Exclusive = 0 represents all firms that have more than one lending relationship with a financial institution. Although some information on the number of institutions a firm is dealing with is lost due to the binary specification of the variable, the binary specification is preferred in this case, because it mitigates potential problems resulting from large outliers. Additionally, the distribution of the number of relationships a firm has with banks is skewed to the left, meaning that most firms only have one or two lending relationships. The summary statistics show that 30% of all SMEs in the sample have an exclusive lending relationship.

3.5 Predicted effects of firm level characteristics

On the firm specific level I choose variables in line with the literature on the size, the type of organization, liabilities, equity, leverage, credit score and whether the firm is located in an urban or rural area. For the size variable, I created the dummy variables Small, Medium and Big and choose Medium as a base value for the regression. In contrast to a linear size variable, the categorical classification of the size variable may help to unravel differing coefficients with respect to class size. Further, I expect that the categorical specification yields more precise results than a linear specification, because the standard errors for the coefficients are based on the respective size sub group only. The firm size variables are categorized according to the firm’s total value of book assets, with cut off values of 10,0000$ in total assets for small firms, between 100,000$ - 1,000,000$ for medium-sized firms and more than 1,000,000$ for big firms. Please note that the indicator for big firms is still within in the target group of SMEs, meaning that this size variable is not an indicator for large corporations within the US, but only for relatively large firms within the target group as defined earlier. The sample includes 56% of small firms, 34% of medium sized firms and 10% of big firms. Based on the literature (e.g. Berger, Goulding & Rice, 2014; Cole, 2013), I expect significant differences between the respective size categories on both length of relationship and the degree of exclusivity. In particular, I

(12)

expect smaller firms to have longer and more exclusive relationships with financial institutions, because smaller firms tend to have a more opaque information structure. Further, bigger firms may require more financial services that can no longer be provided by a single bank. The length of the relationship of bigger firms may also be shorter, because bigger firms tend to engage more in transaction banking (Berger et all, 2014). Contrary one could argue that within in the definition of small and medium sized enterprises, the size of the firm may have a positive effect on the length of the relationship, because the bigger and more financially successful a firm is, the more stable is the relationship with its main bank.

The variables proprietorship, partnership and corporation indicate the type of ownership. In proprietorships there is a single business owner with full liability, whereas a partnerships can have multiple owners, although also with full private liability. The variable Corporation summarizes S and C corporations, thus neglecting some differences in taxation, but both share the characteristic of limited liability protection, meaning that shareholders are not personally liable for business debts or liabilities. The type of ownership may influence the strength of relationship due to differences in perception of default risk by banks, because shareholder of corporations are not personally liable for losses. The summary statistics show that corporations are the most frequent form of ownership with 53% of the sample, proprietorships are second most frequent form of ownership with 42% and only 5% of businesses are partnerships. I choose corporations as base value for my regression.

Next, I added as part of the standard set of financial indicators variables for equity and liabilities. Another financial indicator on the firm level is leverage. Leverage is a good measure for the firm’s indebtedness, because the ratio of debt over equity provides an indication of the relative level of debt of the firm. As high levels of indebtedness are often seen as risky, we may expect a negative effect of the leverage variable. Alternatively one could argue that if the leverage is high, banks have already assessed and trusted the firm and may thus continue to do so in the future. The latter effect of leverage may also be cyclical, meaning that during booms high leverage is seen as positive, whilst in recessions high leverage is perceived as additional risk.

The credit score variable serves as an indicator for the firm’s credit risk and it useful, because the opaque nature of information of SMEs often inhibits reports from rating agencies or financial data that is in line with accounting guidelines that are applicable for public firms. Thus, I expect to see higher credit scores corresponding to

(13)

longer relationships. The credit score was obtained from the Dun and Bradstreet Rank Credit Score and the score ranges between 1 and 6, where 1 represents the most risky and the 6 represents the least risky.

I included the distance (in miles) from the borrowing firm to its primary financial institution and a variable representing the main form of contact with personal contact as a base value for my regression and phone or mail contact as dummy variables. The average distance between borrower and lender is 32 miles, hence it is not surprising that for 82% of SMEs in the sample the main form of contact is in person. Elyasiani & Goldberg, 2004 ascertain that due to the lack of hard information that can be assessed by credit scoring software, distance plays an important role in SME lending, because geographic proximity reduces the costs associated with gathering information about the borrower. Thus, I expect the distance variable to have a negative impact on both endogenous variables. Considering that the data is from 2004, the variable indicating how many computers the firm uses in its operations may not be fully applicable to today’s SMEs, but will still provide an indication of openness to technology for the firms of the time. On the other hand, if a business is at ease of using computer technology, it may also be able to provide more sophisticated information for lending technology that is based on hard financial data, because businesses that use computer may be more likely to have automated business reporting processes that often come with easy to use software packages. The automated collection of data and increased reporting on processes may also reduce the switching costs associated with changing a bank or searching for a new lender. Thus, the computer variable may have a negative effect on the exclusivity indicator. On the other hand, (Zhang & Park, 2004) for instance find a strong positive effect of the use of computers on the sales of the firm with data from the 1998 SSBF iteration, hence one may infer that the increased sales also improve the lending relationship and increase the length of the relationship, because of better financial fundamentals. Although the 1998 SSBF data on sales proceeds may be inflated due to the dotcom bubble, I still gauge Zhang & Park’s findings as valid, because the dotcom bubble was mainly a result of inflated stock prices rather than inflated sales.

The variable on firm age could provide an indication on a relationship between the firm’s age and the strength of the relationship. Berger and Udell (1995) for instance argue in line with the conventional wisdom that with increased firm maturity the credit terms for firms improve, hence potentially increasing the length of

(14)

relationship. On the other hand, an increase in the firm maturity could require more sophisticated financial services that can no longer be offered by only one bank and may thus result in a decrease of the probability that a firm only deals with one firm. The average age of firms in the sample is 14.3 years, showing that the average firm in the sample is relatively mature. Lastly, the dummy variable rural indicates whether the firm is located in an urban or rural area. Due to more traditional community and business structures, rural businesses may engage more in relationship banking and thus have a longer, more exclusive relationship with their bank. In our sample about 21% of all SMEs are classified as rural.

3.6 Predicted effects of owner level characteristics

On an owner specific level the survey provides an array of data. The dummy variable Established, - indicating whether the firm was established by the current manager or not, captures effects that could be associated with trust from banks toward the manager of the firm. If the current manager has also founded the company, he may already have a relationship with the bank, because the manager potentially had to obtain initial funding by the bank. Thus, I expect a positive effect of the variable on both length of relationship and exclusivity. Next, if the owner of the firm is also the principal manager, banks could associate the owner’s personal finances with the assessment of creditworthiness. In this case, the bank may be less reluctant to rely on relationship banking, because hard information on private finances of owners is often more easily obtained, because consumer credit data in the US is widely available over extended period of times through credit scores. However, in the regression model it may be difficult to specify a significant coefficient for the difference between businesses that are owner managed and those that are not, as 94% of all businesses in the sample are owner managed. The dummy variable on owner bankruptcy reports if a business owner was bankrupt within the past three years. Less than 1% of business owners were bankrupt within the past three years. The variable could serve as another proxy of creditworthiness that may also be used by bank loan officers.

As the age of the firm, the age of the owner may play a role determining the strength of relationship. The variable on the age of the owner is specified as the natural logarithm of the age of the principal owner. The sample mean of the age is 51 year. The correlation between age of principle owner and the length of the lending relationship is 0.38, which is comparatively high. Although this mechanical

(15)

relationship between the variables may pose some risk regarding collinearity, I will keep the variables in the regression specification as the literature suggests a relationship between the maturity of a firm or of the owner of a firm and the strength of the relationship. Next, the analysis includes variables on ethnicity, gender and owner background. In this sample, 14% of SMEs are owned by minorities and 38% are owned by woman. As previously outlined, minority owners may face less favorable terms credit terms and as result have a less stable relationship with their financial institution. This effect of segregation on relationship strength would be captured by the minority dummy variable. Effects of potential bias against woman are captured in the dummy variable Female. To test the validity of studies from Europe that have found difference in credit terms for family owned businesses, the regression also includes a dummy variable on family businesses. The summary statistics show that 90% of all SMEs in the sample are family businesses.

Lastly, the business experience and the education of the owner may play a role in determining the strength of the relationship. The business experience is defined as the natural logarithm of the years the principal owner of the firm has been operating a business. Similar to the firm age and owner, the business experience variable shows that the survey data set includes many mature businesses and owners, as the average business experience of the owners is 19.7 years. The variable on owner education captures potential effects that the level of education has on the strength of the relationship. The variable ranges from 1 to 7, where one represents a person without a high school degree and 7 represents a person with a postgraduate degree like a PHD, MBA, MS or JD. The value for the average education of the firm owners in the sample corresponds to a vocational degree, meaning that the average firm owner has no college degree, but finished high school and continued education. I expect a positive effect of education on the length of relationship, because of potential greater financial literacy. In contrast, this better understanding of business or banking practices may lead educated business owners to search for more sophisticated financial services; hence I expect a negative effect on the exclusivity variable.

(16)

(17)

4.1 Methodology

The following regression model aims to establish an empirical relationship between the strength of the relationship and firm and owner specific characteristics. The strength of relationship will be investigated along its two dimensions: (1) The length of relationship in months and (2) the number of financial institutions a firm deals with. In this paper I use a cross-sectional data regression model to analyze a given sample of individual firms, as outlined in the data description. The regression model takes account of the special properties of multiple imputation data and also adjusts the coefficients and standards error to properties of survey data.

The following model is used to estimate the relationship between the length of the relationship and firm and owner specific characteristics:

Eq. 4-1

"𝐿𝑒𝑛𝑔𝑡ℎ 𝑜𝑓 𝑟𝑒𝑙𝑎𝑡𝑖𝑜𝑛𝑠ℎ𝑖𝑝 = 𝛼 + 𝛽_!∗ 𝐹𝑖𝑟𝑚 𝑙𝑒𝑣𝑒𝑙 𝑐ℎ𝑎𝑟𝑎𝑐𝑡𝑒𝑟𝑖𝑠𝑡𝑖𝑐𝑠 + 𝛽_!∗ 𝑂𝑤𝑛𝑒𝑟 𝑙𝑒𝑣𝑒𝑙 𝑐ℎ𝑎𝑟𝑎𝑐𝑡𝑒𝑟𝑖𝑠𝑡𝑖𝑐𝑠 + ū"

Where the left-hand side of the equation is the length of the relationship in month and the right hand side represents the previously specified firm level and owner level characteristics. The error term ū represents the effects of all omitted variables, is assumed to be uncorrelated with all independent variables and independent and

identically distributed (i,i,d) random variable with mean 0 and variance σ2.

The second model estimates the relationship between the probability of an exclusive relationship between the firm and the financial institution and is presented as follows: Eq. 4-2 Pr 𝐸𝑥𝑐𝑙𝑢𝑠𝑖𝑣𝑒 𝑅𝑒𝑙𝑎𝑡𝑖𝑜𝑛𝑠ℎ𝑖𝑝 = 1| 𝐹𝑖𝑟𝑚 𝑙𝑒𝑣𝑒𝑙 𝑐ℎ𝑎𝑟𝑎𝑐𝑡𝑒𝑟𝑖𝑠𝑡𝑖𝑐𝑠, 𝑂𝑤𝑛𝑒𝑟 𝑙𝑒𝑣𝑒𝑙 𝑐ℎ𝑎𝑟𝑎𝑐𝑡𝑒𝑟𝑖𝑠𝑡𝑖𝑐𝑠) = 1 1 + 𝑒!(!!∗!"#$ !"#"! !!!"!#$%"&'$&#'! !!∗!"#$% !"#"! !!!"!#$%"&'$&#' + ū

(18)

Where the left-hand side of the equation is a binary variable and can be interpreted as the probability that a firm has an exclusive lending relationship. The right hand side is a logit regression model and includes the same firm and owner level specific characteristics as the linear regression model from Eq. 4-2, but as a function of the cumulative standard logistic distribution function.

(19)

5.1 Empirical results

The results of regression 4-1 and 4-2 are displayed in Table 4A. Column (1) reports the results of Eq. 4-1 estimated as an multiply imputed model, adjusted for survey weights and stratification with a linear regression using the length of the relationship in months as the endogenous variable. Column (2) reports the results of Eq. 4-2 estimated as an multiply imputed model, adjusted for survey weights and stratification with a logit model using the exclusive relationship indicator as the endogenous variable.

5.2 Estimated effects of firm level characteristics

Firstly, I find that the coefficients of the size variables are mostly not statistically significant. According to the linear regression on the length of the relationship, the firm size seems to have no statistically significant effect. Although this result is in line with Berger’s finding on the significance of the size on the length of the relationship, the result is not conform with the conventional wisdom found in the literature. Petersen & Rajan (1994) for instance report above average benefits of relationship banking for small firms. As a result I would expect the length of the relationship to increase. Hence, my findings may support Berger’s conclusion that the difference that determines the use of hard lending versus soft lending is not strictly a function of the size of the borrower (Berger & Black, 2011). On the other hand, I can confirm that firm size has a positive effect on the exclusivity of the relationship, as small firms tend to have more exclusive relationships. The odds ratio of regression 4-2 suggests that small firms are 4-2.7 times as likely as other firms to have an exclusive lending relationship. This may confirm the initial hypothesis that small firms tend to require less diversified financial services and thus tend to rely on a single bank. On the other hand the result may indicate that establishing multiple lending relationships is too costly for small firms and they consequently rely on a single lender. The importance of the size variable in determining the exclusivity of the relationship is reinforced by the fact that the odds ratio is one the highest in the regression, with a strong p-value significant at a 1% level.

Surprisingly, the type of ownership of the organization has no statistically significant effect on the length of relationship or the probability of an exclusive

(20)

relationship. Next, the variables on equity and liabilities show no significance for either of the regressions. One reason for this finding could be the relatively high correlation between -0.18 and -0.58 of the equity and size variables, resulting in an underestimation of the equity coefficient. I included a variable on leverage to include another proxy for credit risk, because banks tend to perceive firms with relatively high debt to equity ratio as risky (Berger & Frame, 2005). The regression results report no significant effect of leverage on the length of the relationship or the degree of exclusivity. Although including variables on equity and liabilities and leverage (calculated as debt over equity) may have posed some risk with respect to introducing collinearity, this specification reduced the risk of omitted variable bias. Even with non-significant coefficients, the introduced variables reduced the sum of squared residuals.

The credit score seems to be a better proxy for the perceived credit risk of banks. As expected, the credit score has positive effect on the length of the relationship. The estimated coefficient translates to a 5.3 month increase in the length of relationship for a 1 point increase in the credit score; significant at the 1% level. Frame, Padhi, Woosley, & Padhi (2001) examined the effects of credit-scores on credit availability for SMEs and found a positive impact; hence I may infer that the results of my regression confirm Frame’s result by finding that an increase in the credit-score increases the length of the relationship. Frame et all (2001) conclude that the extensive use of credit scoring lowers the information costs associated with assessing credit risks, thereby reducing the value of relationship lending. Based on my findings, I conclude that with increased use credit scoring, the length of the relationship may be improved due to better pre-screening of borrowers, but the exclusivity of the relationship may be negatively effected, because as a result of more hard data based on credit scores the switching costs associated with using another bank may be reduced.

The distance variable is indeed significant in both regressions. The coefficient of -0.04, significant at the 1% level, confirms the initial hypothesis that geographic proximity increases the length of the relationship. This may be due to decreased costs associated with obtaining first hand information through direct means of communication. Another potential reason for the finding may be the existence of a ‘home-bias’ of banks for local lenders. Presbitero, Udell, & Zazzaro (2014) for instance found in a study of Italian manufactures during the 2008/2009 credit crunch

(21)

evidence that firms with primary lending relationships to banks that were distantly-managed had a significant disadvantage in obtaining credit.

The regression reports large and significant coefficients of the computer variable. Holding all other variables constant, the length of relationship increases by 18 months if the firm uses an additional computer. Considering the strong positive relationship, we may thus infer that the computer variable can serve as proxy for the openness to technology for SMEs in 2004. In line with the findings for the length of the relationship, the odds-ratio for the computer variable is also significantly different from 1. The reported ratio is 2.8, meaning that for each additional computer a firm uses, it is 2.8 times as likely to have an exclusive lending relationship. As a result we may refuse the hypothesis that the use of computers shortens the length of relationship. Conventional wisdom in the literature usually argues oppositely and cites reduced switching costs as a factor that facilitates multiple banking relationships in the context of increased use of computer technology (e.g. Petersen & Rajan, 2002).

The coefficient of the firm age variable has the greatest magnitude among all variables and is significant at the 1% level. However, the coefficient may partly be inflated due to a mechanical relationship between the length of the relationship and the firm age, - older firms can simply have longer relationships than firms that just got established. One reason for the relative importance of firm age variable is that banks may perceive older firms as more mature and less risky investments, as the information structure of mature firms tends to be more structured (Berger & Udell, 1994). Additionally, the literature cites increased trust and reduced costs due to the benefits of monitoring borrowers over time (Petersen & Rajan, 1994). It seems that in particular in the setting of relationship banking, which is characterized by repeated interaction, firms are better off providing full-disclosure of potential information on credit risk, because of the improved credit terms and credit availability that are associated with strong relationships to lenders. The odds-ratio ratio of the exclusivity indictor is smaller than 1 and almost significant at the 5% level, hinting at the interpretation that mature firms require a broader set of financial services and thus establish multiple banking relationships.

Lastly, the dummy variable on rural indicates a significant difference on the length of relationship. Holding all other factors constant, rural businesses have a 16.7 months longer relationship. This result broadly confirms previous findings in the literature and show that lending practices differ between urban and rural areas. The

(22)

odds ratio on exclusivity is not significant. One reason could be the previously cited banking consolidation in the US. Over the past three decades many rural and community banks have merged and transformed to corporations similar to large, urban banking corporation (Park, 2008). This trend may be reflected in the exclusivity indicator, as rural firms may have been forced to differentiate their banking relationships. On the other hand, the coefficient on the length of the relationship shows that despite this change towards less exclusive banking relationship, rural firms still engage in longer relationships.

5.3 Estimated effects of owner level characteristics

In summary, the estimated coefficient for the effect of owner level characteristics are less pronounced than expected, but the regression results still report some significant coefficients and odd ratios. The variable Established, which indicates whether the current owner of the firm established the firm, has a negative effect on the length of the relationship. This may confirm my initial hypothesis that firms that are still in the ownership of their initial founders may in some cases be managed less professionally, as the literature suggests that from a certain size firms may managed more efficiently by outsiders (Cole, 2013). Next, I find that whether the owner manages the firm or not has no significant impact on the length of the relationship or the exclusivity indicator. From the literature, I expected an ambiguous effect. On the one hand, much literature exists on the dichotomy between managers and shareholders and the resulting moral hazard problem (e.g. Cole, 2013). Banks could be aware of such problems and thus provide for instance less favorable credit terms, which in turn shorten the length of the relationship. On the other hand, in particular in the context of SMEs, banks may prefer owner managed firms, because they can rely on widely available private credit scoring information. I conclude that the specification of my regression is too limited to provide significant evidence on this hypothesis. Next, I find that a previous bankruptcy of the owner of the firm has no significant impact on the length of the relationship of the exclusivity indicator. This may be due to the small sample size of less than 1% of owners that were bankrupt within the past three years.

Among the owner characteristics that are associated with the owner’s age, ethnicity, gender or education, I find few significant coefficients. For the estimate of

(23)

the length of relationship, the only significant coefficients are the age and the years of business experience of the owner. Due to the mechanical relationship between the age, the years of business experience and the length of relationship, it is difficult to confirm the initial hypothesis that banks prefer more mature or more experienced business owners. The age variable also has a positive effect on the exclusivity indicator, meaning that according to the regression results older owners tend to have more exclusive banking relationships. Other characteristics like the gender or the ethnic background of the owner have no significant effect on either of the endogenous variables. Previous research found a gender gap in credit availability for consumer credits, but my regression results cannot confirm such a relationship for commercial lending relationships (Cole & Mehran, 2009). Additionally, the education of the owner has no statistically significant effect on either the length of the relationship or the exclusivity indicator.

(24)

(25)

6.1 Conclusion

This paper contributes to the existing literature on relationship banking by presenting one of the few empirical estimates of the firm and owner level characteristics that determine a strong relationship. The least square estimators are adjusted for multiply imputed data and represent approximately 5.9 million small businesses in the US. This thesis shows that an important determinant of the relationship strength between borrowers and lenders, here represented by the length of the relationship and an exclusivity indicator, are the firm and owner level characteristics of the respective small and medium sized enterprises. This finding confirms recent research by Berger (2014), which already suggested significance of firm and owner level characteristics on the strength of the relationship. Corresponding to existing theories, I found evidence for the significance of credit scores, as a measure of credit risk, to have a strong positive effect on the length of relationship. Furthermore, other in relationship banking theory historically deemed important indicators like the distance between the lender and borrower, the firm age, whether the firm was established by the current manager and the owner age were shown to have a significant effect on the strength of the relationship. What is even more interesting in the context of this analyses, are the variables that have no statistically significant coefficients. Theory on relationship banking predicts strong effects of the size and type of ownership variables on the strength of the relationship. While we found that small businesses tend to have significantly more exclusive relationships, probably reflecting the high costs of maintaining multiple banking relationships, none of the other coefficients on size or type of ownership were significant. This finding shows that the existing theory may overstate the effect of size and type of ownership on the strength of the relationship. On the other hand, this conclusion is limited within our definition of SMEs, hence in the context of small and medium sized relationship banking.

The most interesting of the paper is the relationship between the computer variable and both the length of the relationship and the exclusivity indicator. I found a strong positive effect of the use of computers on the length of the relationship. In addition, firms that use computers are 2.9 as likely as other firms to have an exclusive relationship with their lender. The finding violates most of the theory on relationship

(26)

banking, which typically predicts a negative effect of the use of computers on the strength of the relationship. My findings may thus be interpreted as an indicator that the transformative power of information technology does not disrupt long built relationships between borrowers and lender, but helps the firms to provide better and more accurate information. On the other hand, the strong estimate of the effect may also serve as measure for the openness of the technology for the respective firm. In this case, the observed effect may be due to superior firm performance that improved the lending relationship.

The advantage of this thesis, namely focusing on the owner and firm characteristics that determine the strength of the relationship between borrowers and lenders, is at the same time a limitation to the findings. Later research can improve on this topic by adding bank and loan officer specific information to the regression specification. In conclusion, my findings show an exciting perspective on the interplay between borrowers and lenders and can, besides confirming theoretical arguments, be very interesting for policy makers that want to target their legislative aid to small and medium sized enterprises.

(27)

Bibliography

Berger, A. N., & Black, L. K. (2011). Bank size, lending technologies, and small business finance.

Journal of Banking & Finance, 35(3), 724–735.

Berger, A. N., Goulding, W., & Rice, T. (2014). Do small businesses still prefer community banks?

Journal of Banking & Finance, 44(0), 264–278.

Berger, A. N., & Udell, G. F. (1994). Lines of credit and relationship lending in small firm finance.

Jerome Levy Economics Institute Working Paper, (113).

Boot, A. W. A. (2000). Relationship Banking: What Do We Know? Journal of Financial

Intermediation, 9(1), 7–25.

Cole, R. A. (2013). What Do We Know about the Capital Structure of Privately Held US Firms? Evidence from the Surveys of Small Business Finance. Financial Management, 42(4), 777–813. Cole, R. A., & Mehran, H. (2009). Gender and the availability of credit to privately held firms:

evidence from the surveys of small business finances.

Elsas, R., & Krahnen, J. P. (1998). Is relationship lending special? Evidence from credit-file data in Germany. Journal of Banking & Finance, 22(10), 1283–1316.

Elyasiani, E., & Goldberg, L. G. (2004). Relationship lending: a survey of the literature. Journal of

Economics and Business, 56(4), 315–330.

Frame, W. S., Padhi, M., Woosley, L., & Padhi, M. (2001). The Effect of Credit Scoring on Small Business Lending in Low- and Moderate-Income Areas. Federal Reserve Bank of Atlanta, 26. Hoshi, T., Kashyap, A., & Scharfstein, D. (1990). The role of banks in reducing the costs of financial

distress in Japan. Journal of Financial Economics.

James, C., & Wier, P. (1990). Borrowing relationships, intermediation, and the cost of issuing public securities. Journal of Financial Economics, 28(1), 149–171.

Machauer, A., & Weber, M. (2000). Number of bank relationships: An indicator of competition,

borrower quality, or just size?.

Maria Herrera, A., & Minetti, R. (2007). Informed finance and technological change: Evidence from credit relationships. Journal of Financial Economics, 83, 223–269.

Park, Y. (2008). Banking market concentration and credit availability to small businesses.

Petersen, M. A., & Rajan, R. G. (1994). The Benefits of Lending Relationships: Evidence from Small Business Data. The Journal of Finance, 49(1), 3–37.

Petersen, M. A., & Rajan, R. G. (2002). Does distance still matter? The information revolution in small business lending. The Journal of Finance, 57, 2533–2570.

Presbitero, A. F., Udell, G. F., & Zazzaro, A. (2014). The home bias and the credit crunch: A regional perspective. Journal of Money, Credit and Banking, 46(s1), 53–85.

Rubin, D. B. (1976). Inference and missing data. Biometrika, 63(3), 581–592. Zhang, F., & Park, T. A. (2004). Computer Adoption Patters of U.S Small Businesses

(28)

Firm and owner level characteristics of SME lending relationships : evidence from the Survey of Small Business Finance