March 2000 Does Distance Still Matter? The Information Revolution in Small Business Lending Mitchell A. Petersen Kellogg Graduate School of Management Northwestern University and Raghuram G. Rajan Graduate School of Business University of Chicago Abstract The distance between small firms and their lenders in the United States is increasing. Not only are firms choosing more distant lenders, they are also communicating with them in more impersonal ways. After documenting these systematic changes, we demonstrate that they do not stem from small firms locating differently, from consolidation in the banking industry, or from biases in the sample. Instead, they seem correlated with improvements in bank productivity. We conjecture that greater, and more timely, availability of borrower credit records, as well as the greater ease of processing these may explain the greater amount of lending at a distance. Consistent with such an explanation, unlike in the past, distant firms no longer have to be observably the highest quality credits, suggesting that a wider cross-section of firms can now obtain funding from a particular lender. These findings, we believe, are direct evidence that there has been substantial development of the financial sector in the past few years in the United States, even in areas such as small business lending that have not been directly influenced by the growth in public markets. From a policy perspective, that small firms now obtain wider access to financing suggests the consolidation of banking services may not raise as strong anti-trust concerns as in the past. Petersen thanks the Banking Research Center at Northwestern University’ Kellogg School while s Rajan thanks the Center for Research in Security Prices and the Center for the Study of the State and the Economy at the University of Chicago, as well as the National Science Foundation for support. Small business lending has historically been very costly, because of the paucity of information about small firms and the high costs of the personnel required to obtain even that information. There has, however, been a strong trend in small business lending; small businesses have been growing steadily more distant from their lenders. This demonstrates itself in two ways. First, borrowers are becoming physically more distant from lenders. The distance between small firms and their lenders has grown from an average of 51 miles for lending relationships that began in the seventies to 161 miles for relationships that began in the nineties. A similar pattern can be seen in the medians and seventy-fifth percentiles and across different kinds of lenders (see Table I). While banks are indeed closer to their clients than other lenders – reflecting the greater transactions and information intensive nature of bank lending – even banks have been moving away; from an average of 16 miles for lending relationships that began in the seventies to 68 miles for relationships that began in the nineties. A second trend is that the method of transacting business has moved increasingly from personal contact to using the phone or the mail. This trend is partly a result of the first one, but even correcting for physical distance, there seems to be a change in the manner of doing business. What accounts for this trend? We show that changes in the location of small businesses, consolidation in the banking industry, and sample selection biases, are unlikely explanations. One possible explanation of the trend, however, is that a variety of infomediaries such as credit bureaus now exist. They collect information on payment history and defaults that hitherto used to be available to a lender only after a long relationship. Advances in storage technology and computing allow these data to be easily retrieved, processed, and communicated quickly. As a result, even distant lenders have timely information that hitherto used to be soft and unavailable at arm’ length. s 1 Not only does this reduce information asymmetries between lender and borrower, but also timely information enables them to react quickly to unanticipated contingencies, minimizing the loss from borrower malfeasance or incompetence. Consistent with this conjecture, we show that the increase in distance is strongly positively correlated with measures of bank productivity, suggesting that reductions in the cost of gathering and analyzing information may indeed explain the results. If greater, and timely, availability of information about credit history is indeed part of the explanation for why loans can now be made at a distance, we should find that distance is no longer such a strong predictor of credit quality as in the past. Intuitively, in the past only unimpeachable borrowers would have obtained credit at a distance because the costs of detecting potential borrower distress or malfeasance, and resolving it, would have been prohibitively high for a distant lender. But if these costs have fallen, then riskier credits should now be obtaining finance at a distance. We do find that even though lender distance is a good indicator of creditworthiness, it is less important a measure now than in the past. This suggests weaker credits are obtaining loans at a distance. Consistent with our argument that lenders are better able to monitor and control these credits, we do not find a commensurate increase in lender loan losses over our sample period. We therefore conclude the evidence is consistent with greater information availability being responsible for the increasing distance between lender and borrower. Small business lending is indeed the final frontier in financial development. That even this frontier is being breached is suggestive of the dramatic effects of technology. But the implications of the breach can be large indeed. Small businesses are important, in part because most businesses start this way, in part because they have traditionally been associated with innovation, and in part because, increasingly, more of the employment in the United States is concentrated in this sector (see 2 Davis and Haltiwanger (1989)). In the long run, if the trend continues, the increased access of small business to finance may lead to more competition, greater efficiency, and a more egalitarian society. There has been a debate about whether institutional lending, especially bank lending, is in secular decline because of the greater availability of information to arm’ length financial markets s today (see Boyd and Gertler (1994) and Gorton and Rosen (1995) for two contrasting views). Our evidence suggests that institutional lending also seems to have benefitted from the greater availability of information. This has enabled institutions to lend to riskier clients whom they would have shunned in the past. Thus instead of driving financial institutions such as banks out of the business, technological change may also create new sources of comparative advantage for them (as conjectured by Merton (1995)). Our evidence may also better establish the sources of comparative advantage among different types of institutional lenders. The theoretical literature in banking (e.g., Diamond (1984)) has emphasized the role of banks as “close” lenders serving informationally opaque credits. Empirical work has typically focused on firm characteristics such as low firm age and small size as measures of opaqueness, and found that banks do indeed lend proportionately more to such firms (see, for example, Petersen and Rajan (1994) and Berger and Udell (1995)). But evidence that banks are close has been more anecdotal than systematic. Our metrics of closeness, physical distance and the method of interaction, provide confirmation that banks are close, even after we account for natural reasons for closeness such as the greater frequency of transactions implied by the existence of deposit accounts. Moreover, the firms that on average stay closest to lenders seem to be informationally opaque ones. Thus the new metrics we focus on are consistent with the theoretical conjecture of the role of banks, and could be useful in further empirical work. 3 Our findings also have policy implications. A significant focus of anti-trust policy in banking has been on the consequences of bank mergers on local competition, and hence on small business lending. While our evidence indicates that small businesses continue to use their local banks for deposit transactions, the effective size of the credit markets faced by small firms is continuously expanding. Anti-trust policy should optimally take into account the expansion in competition as a result of changes in technology while defining the size of the relevant market.1 Finally, there is an ongoing debate about whether the effects of the improvements in information technology result in improvements in productivity (see Gordon (1999), for example). Our evidence, while not directly addressing this issue, suggests that at least in the financial sector, the nature of transactions is changing in a direction consistent with greater information availability and reduced costs of processing it. When combined with evidence that these changes are correlated with improvements in bank productivity, this should strengthen our belief that information technology does indeed increase productivity. The rest of the paper is as follows. After a brief discussion of the issues, we document the growing distance between borrowers and lenders in section III. In section IV, we examine possible causes of the changes we document. In section V, we examine the consequences of these changes by asking if these changes had differential effects on the price of capital or access to capital. We conclude in section VI. 1 There may already be political changes resulting from technological development. Kroszner and Strahan (1999) argue that technological innovation reduced the value of geographical restrictions on banking to their traditional beneficiaries (the small banks, for example), and permitted the deregulation that started in the 1970s. 4 II. The Kinds of Information Available in Small Business Lending. A. The Market. Unlike for large firms, the information available about small and private firms has historically been limited and difficult to access. With the exception of some in high growth industries – which are a very small portion of our sample – analysts do not follow these firms. Since these firms do not raise capital in public markets, they are not required to disclose much information. The firm’s lenders clearly know about the firms, but these lenders are few in number and did not readily share information in the past. Since information about the firm was not compiled, stored, and distributed by a central bureau, but instead resided in the minds of the firm’ bankers, much of it tended to be s soft – whether the firm generally maintained adequate balances, for example, rather than hard information specifying when and to whom it had, or had not, made payments in the past. B. Information Technology and Small Business Lending. The use of information and communications technology, by which we mean everything from hardware like computers and phones to software like credit scoring and client profitability programs, has transformed the financial sector over the last three decades ( Mishkin and Strahan, 1999). Three aspects are particularly significant to us. First, the ability to collect, store, process, and communicate large amounts of information has expanded tremendously. Second, this has resulted in the expansion of the activities of infomediaries whose sole purpose is to collect, organize, and make available this information to paying customers. Third, the availability of hard, processed information lends itself to cost effective credit appraisal and monitoring techniques. Since the first aspect is fairly uncontroversial, let us examine the latter two in more detail. 1. Expansion of the Activities of Infomediaries. 5 Technological change has resulted in the expansion of the activities of infomediaries such as rating agencies and credit bureaus. Consider, for example, Dun and Bradstreet. It was founded as far back as1841, with the aim of establishing a network of correspondents that would function as a source of reliable, consistent and objective credit information.2 Among its early reporters were four U.S. presidents: Abraham Lincoln, Ulysses S. Grant, Grover Cleveland, and William McKinley. Big changes in its procedures came in the 1960s and the 1970s. In 1963, the introduction of the Data Universal Numbering System used to identify businesses numerically for data-processing purposes revolutionized the collation and distribution of business information. In the 1970s, a new "Advanced Office System" fully computerized D&B’ data-collection operations, giving it the ability s to link and analyze categories of information in entirely new ways, and to deliver information to customers faster and more economically. The pace of these changes have accelerated with the coming of the Internet (an era that post-dates our sample). D&B updates over 11 million business records on an annual basis. Over our sample period, the number of firms on which they have records has grown at 6.3 percent per year, a rate over two and a half times the real growth of the economy. Hundreds of millions of pieces of data, ranging from trade experiences to financial statements, are integrated every day into one file. D&B collects information from millions of on-site and telephone contacts with business owners and managers, as well as from all federal bankruptcy filing locations, from millions of trade and bank experiences, from public utilities, from over 2,500 state filing locations, and from daily newspapers, publications and electronic news services. The data that is entered is automatically checked, and also subject to random verification. Finally, D&B alerts its customers to increases in a business’ risk profile so s 2 The information on Dun and Bradstreet in this section is obtained from their web site. 6 as to prevent unnecessary losses. The point is that specialized infomediaries like D&B can save on duplication, and amortize the costs of information collection over a larger number of customers than could lenders in the past. As a result, they can distribute more information than ever available to lenders in the past. 2. More Efficient Appraisal and Monitoring. The increased availability of systematic reliable information has allowed loan officers to cut down on their own monitoring. Moreover, the information can now be automatically processed, eliminating many tedious and costly transactions. For example, Automatic Loan Machines (ALM) now offer loans on the spot to individuals who have a reasonable credit history, regardless of who they banked with in the past (see Rajan (1996)). In an ALM, the process of taking the client’s information, checking records, evaluating the expected profitability of the loan, and then making the actual loan has been completely automated. Credit scoring – a process by which a loan applicant’ credit history and characteristics are s summarized in a credit score which forms the basis for approval or rejection of most applications – is increasingly used by large banks such as Wells Fargo to make lending decisions even for small businesses (Mishkin and Strahan, 1999). By using financial histories, credit reports, and scoring methods, the banks can dramatically lower the time their loan officers spend on a given application and thus the cost (Padhi, Srinivasan, and Woosley, 1999, Mester, 1997). Small firms gain substantially by a lowering of the fixed costs in the lending process (Frame, Srinivasan, and Woosley, 1999). The median loan in our sample is for $18,000. Firms in our sample were asked the total fee (not including interest) that they paid to obtain their loan. The level of the fee is uncorrelated with the size of the loan across the sample (correlation = 0.001). Thus, fees as a 7 fraction of the loan size declined with the size of the loan. Every ten percent increase in the loan size, lowered the fees as a percent of the loan amount by 4.8 percent (t=13.4). Given that they are largely fixed, reductions in the cost of loan origination and information collection could therefore produce the largest gain for smallest firms. Additionally, if transactions costs drop sufficiently, the number of lenders that are willing to lend may expand. This has the possibility of not only expanding the supply of finance to small firms, but also reducing the cost of financing to the extent that geographically larger markets are more competitive. We explore this issue in Section V. If in addition to being available at low cost, information is timely, the lender reduces the potential loss from borrower moral hazard. If a borrower, either because of incompetence, or because of malevolence, takes improper actions, the lender can act quickly to stop further lending and can demand repayment. By contrast, if the lender acquires information after a long lag, he may have thrown good money after bad, the borrower’ assets may have deteriorated under poor management, s and other lenders may have seized anything of value. Thus timely information also reduces the costs of lending. There is a potential downside, however, to information technology. The literature on small firms has stressed the importance of relationship lending for small firms (see the survey by Berger et al. (1998) for references). Historically, part of the incentive for a lender to develop a relationship with a borrower, even if initial loans were not cost effective, was the knowledge that, if successful, the firm would be locked in to a long term relationship because the lender would have monopoly access to information about the borrower. The cost of the initial loan could be amortized over the longer relationship (see Greenbaum, Kanatas and Venezias (1989)). However, when information is widely shared, the credit market becomes more competitive, which could reduce the availability of 8 credit (see Petersen and Rajan (1995) for theory and evidence). Another potential problem is that soft information is difficult if not impossible to incorporate when credit decisions are made by computer models and credit reports. Thus a concern is that small firms who are truly good credit risks – but on paper look like bad credit risks – will find capital more difficult to obtain. As the relative costs of funding such firms rise with advances in information technology, lenders may simply ignore them, preferring to focus attention on the transparent.3 We will attempt to sort out these explanations in the rest of the paper. III. Changes in the Small Business Lending Market: Empirical Results. A. Description of the Data. Our data sample is drawn from the 1993 National Survey of Small Business Finance (NSSBF). This is a stratified random sample of small firms which was collected by the Board of Governors of the Federal Reserve System and the Small Business Administration. In addition to financial information about the firm (balance sheet and income statement information for 1992), the data contains a thorough documentation of the firm’ relationship with financial institutions. To be s in the sample, the firm must be a for-profit firm with fewer than 500 employees. Consequently the firms in our sample are small. The firms have a mean 1992 sales revenue of $3.6M (median $ 400,000) and a mean book value of asset of $1.7M (median $ 153,000). For a more complete description of the data see Cole and Wolken (1996). 3 “(W)e should not lose sight of the exceptional economic value of ...old-fashioned face-to- face interpersonal banking. The newer technologies may be awesome but human nature does not change – we still appreciate a face across the desk more than a computer screen.” Alan Greenspan, “Challenges Facing Community Banks,” speech before the Independent Community Bankers of America, March 8, 2000. 9 Firms in the survey were asked for an exhaustive list of the financial institutions with which they have a business relationship. The relationship can be a credit relationship (they borrow from the institution), a service relationship (they purchase financial services from the lender), or a deposit relationship (they have a checking or savings account with the institution). From this information, we can build a picture of how the lending relationship or environment for small firms has changed over time. For each institution with which the firm has a relationship, the firm is asked how long they have been doing business with the institution. From this we calculate the calendar year in which the business relationship between the firm and the lender began. To examine the change in the small business lending market, we focus on two measures of how close firms are to their lenders. The firms are asked how far the lending institution is from the firm. This is the distance from the main office of the firm to the office or branch of the lender that the firm uses most frequently. We also know the predominant way in which the firm and the lender conduct business (in person, by phone, or by mail). These variables will be the focus of our analysis. From the National Survey of Small Business Finance (1993), we generate a data set of lender borrower pairs. The firms in our data set may borrow from multiple lenders or they may borrow from a given lender in multiple ways (an uncollateralized line of credit and a collateralized mortgage). Since the nature of the loan and the lender may affect the nature of the relationship (physical distance and method of communication), we generate a data set where each observation represents a single lender-borrower pair. There are, on average, just over two lender-borrower pairs per firm. B. Distance from Lender. The distance to a firm’ lender is, not surprisingly, a skewed distribution (see Table I). Thus s we use the log of one plus the distance between a firm and its lending institution as our favored 10 measure. The average log distance to the lender is 2.6 (13.5 miles) with a standard deviation of 1.9 (6.7 miles).4 By taking logs, the distribution moves closer to a symmetric distribution. The median of the distribution is 2.3. Since distance is related to the cost of collecting and transmitting information about a potential borrower, we want to examine the factors which determine how far a firm is from its lender. In particular, firms that are informationally transparent will be cheaper and easier to evaluate from a distance and thus can borrow from further away. By examining the correlates of distance, we can explore the factors that might make small firms more or less informationally transparent. 1. Loan sources and lender characteristics. One of the primary functions of banks is their role as a monitor (see Diamond (1984), Fama (1985), James (1987)). This role includes an initial evaluation of the borrower’ type as well as s continuous monitoring of the actions of the borrower. Given the limited publicly available data on small/private firms, the role of monitor can be particularly valuable. Consistent with this intuition, small firms rely more on banks than on other kinds of lenders. The lenders in our sample can be classified into five basic categories – two internal and three external sources. Internal sources include loans from the owners (16 percent of debt) and loans from family and friends (6 percent of debt).5 The remaining 78 percent of debt comes from external sources. These sources can be divided into banks (69 percent of external debt), non-bank financial 4 The mean distance between firm and lender is 115 miles if we do not take logs first. The median distance is 9 miles (see Table I). 5 Government loans comprise a very small fraction of the firms’ borrowing (less than 1 percent). Although this is external debt, we do not include it the following analysis as we want to focus on the firm’ development of relationships with private, for-profit lenders. s 11 lenders (25 percent) and non-financial lenders such as suppliers (5 percent). The identity of the firm’ lender has a strong correlation with how far away the lender is. s Non-banks are significantly further from the firm than banks – 118% further, on average (see Table II, column I). This is consistent with banks being more active monitors and monitoring being difficult across greater distances. Thus firms that want or need monitoring choose close banks. The distance to non-financial firms in our sample lies in between that of the banks and non-banks. The frequency with which a firm has to transact with a lender can also determine how close that lender is. Checking accounts presumably imply more frequent transactions and possibly more frequent monitoring. Banks that also provide their borrowers with a checking account must be significantly closer. They are! Based on the results in Table II, the distance doubles with each step as a firm moves from a bank where it has a checking account, to a bank without a checking account, and finally to a non-bank.6 Interestingly, not all lenders that provide services are closer. Lenders that provide services other than checking or savings accounts are in fact further.7 This suggests transactions costs cannot be the entire story. A possible explanation is that checking accounts are special. Fluctuations in the firm’ balance in the account can provide the bank with additional s information (Nakamura (1989), Mester, Nakamura, and Renault, 1998)) and seems to increase the availability of credit (Petersen and Rajan (1994)). So the existence of a checking account may reflect, 6 In our sample, over ninety-nine percent of checking accounts are provided by banks. 7 We also examined the effect on distance of other services the firm may obtain from the lender. Firms which obtain additional financial services are further from their lenders on average. The magnitude of the effect depends upon the type of service: transaction services (such as making change, processing credit cards, and executing wire transfers) are 11% further (t=1.7), cash management services are 43% further away (t=5.3), and credit related services (such as bankers acceptance and sales financing) are 26% further away (t=3.1). 12 in part, the need for monitoring, which may also explain the closeness. This is also consistent with our finding that having a saving account (which may not reflect as much information as a checking account because it can be used less flexibly) moves the firm closer to its lender, but by only 16% (regression not reported). Lender type is not independent of other characteristics of the firm’ borrowing. The loans in s our sample are divided into six loan types and these are not distributed evenly across banks and non- banks.8 However, even when we include controls for loan type and whether the loan is collateralized, the coefficients in Table II do not change qualitatively (regression not reported). 2. Systematic changes in the lending market. Over time the environment in which our firms borrow has changed. To examine the systematic changes in the borrowing market, we included the year the firm began borrowing from a given lender. This variable measures systematic changes in the small business lending market independent of the characteristics of the firms. We find that the distance between the firm and their lenders is growing at 3.4% per year (t=7.4). Thus holding the firm characteristics constant, a firm that began borrowing from its bank in 1993 is 34% further away from the bank than an otherwise identical firm that first borrowed from its bank in 1983 (Table II, column I and Figure I). The conditional growth rate in the distance between firms reported in Table II, understates the unconditional growth in distance. If we do not control for lender type and the presence of a checking account, the estimated growth in distance rises from 3.4 percent per year to 6.9 percent per year (t=13.3 – regression not reported). This implies that over time, new lending relationships are 8 The firms in our sample borrow through lines of credit, leases, motor vehicle loans, mortgages, equipment loans, and other loans. Leases, motor vehicle loans, and equipment loans are collateralized by definition. The other three loan types may or may not be collateralized. 13 moving from banks with checking accounts (which are close) to non-banks without checking accounts (which are farther away). Although banks are closer to their borrowers on average, both banks and non-banks have been moving away from their lenders. The distance is growing slightly faster for non-banks than banks (4.4% versus 3.0% per year), but the difference is not statistically significant (regression not reported). Thus the factors that are driving this change are affecting all lenders. 3. Alternative Explanations For the Increase in Distance. a. Changing distribution of firms. The growing distance between firms and their lenders that we document could arise from several sources. One possibility is there has been a systematic change in the type of firms that are beginning a relationship with a lender over time. As a first test of this we included a set of control variables. Specifically, we include controls for the firm’ industry (two digit SIC) or the census s region in which the firm is located. This does increase the explanatory power of the model. The industry controls (F(58,5916)=3.0, p<0.01) and the census region control (F=8,5966)=3.1 p<0.01) are both statistically significant. However, our estimate of the rate at which firms are moving away from their lenders does not change when we add these controls (see Table II, columns II-III). We also exploit the panel structure of our data by including a control for each firm. This dramatically raises the explanatory power of the model (the R2 rises from 0.307 to 0.711). This implies that firm characteristics are an important determinant of distance which we explore below. Adding firm controls to the regression actually raises the coefficient on the year the relationship began slightly from 3.4 percent per year to 4.0 percent per year. Whether we estimate the growth in distance based on within firm variation (variables are defined as deviations from firm means) or based on between 14 firm variation (variables are defines as the mean for each firm), the estimated coefficient is essentially the same (see Table II, column IV and V). It is also possible that the location of firms is changing over time. For example, firms may now be moving from urban areas to predominantly rural areas where distances between businesses are larger. Such a change could in theory explain our findings. We know whether the firm is in a Standard Metropolitan Statistical Area (urban) or not (rural) and rural firms are 13 percent further from their lenders. We estimate the year coefficient separately for firms located in rural or urban areas. The coefficient is not statistically different across the two samples (p=0.23) suggesting that the possibly changing location of firms cannot explain the findings. b. Sample selection bias. The data set we use is a synthetic panel. Data on the year a firm began a relationship with a given lender helps us describe a firm’ borrowing patterns over time. However, the data set is s conditioned on the firm existing in 1993. Firms that do not survive will obviously not be included in our sample. If the selection mechanism is correlated with distance, then our estimated coefficient would be biased. In addition, to requiring that the firm survives, observations appear in our sample only if the firm-lender pairs also survive. If the type of pairs that die is correlated with distance, this will also bias our coefficient. We first describe each of the possible selection processes that must be at work for a survival bias to generate our results and then examine the empirical evidence to determine their validity. We examine firm survival first. The literature on relationship lending suggests that close lenders (often banks) are good at looking into informationally opaque firms and determining which are truly good firms. They may also be more able and more willing to bail out their good borrowers when they 15 experience temporary financial difficulties. As lenders migrate further away from firms, bailouts may become more costly and less common. If firms whose lenders are further away are less likely to be rescued, they will be less likely to survive. This would imply that surviving firms will be closer on average than the full sample of firms that begin relationships with lenders. This may be a problem for our results since the firms that start a relationship in 1991 need survive only 2 years to be in our sample, however, firms that begin a relationship in 1981 must survive 12. Thus, sample selection will induce a positive coefficient on the calendar time variable even if there is no change in the distribution of how far firms go to borrow from their lender.9 To test for the importance of these sample selection biases we supplemented our sample with data from the 1988 NSSBF. This allows us to directly examine how firm survival affects our estimates. In the expanded data set, we can examine two firms that both began borrowing from their first lender in 1985, for example, when they were both five years old. The one from the 1993 sample had to survive 8 years to be included in the sample while the one in the 1988 sample needed to survive only 3 years to be included in the sample. If the selection mechanism we described is working, we will see that controlling for the year the relationship started, observations in the 1993 sample are closer – as they have had to survive longer to remain in the sample. The results are displayed in column VI of Table II. Firms in the 1993 sample are 10 percent closer and the coefficient is statistically significant (t=2.0). This is consistent with the sample selection hypothesis, but also with any other difference between the two samples. Nevertheless, if sample selection is truly responsible for our results, once we control for it the coefficient on calendar year should fall. Instead, 9 If instead, firms that start closer to their lenders are riskier and are more likely to die, then the coefficient on the calendar year variable is understated and firms are moving away from their lenders faster than our estimates suggest. 16 we find the distance between lenders and borrowers is growing faster over time – 3.8 percent per year (Table II, column I). Our initial estimate was biased downward. The second potential selection bias arises if the firms in our sample drop their original lenders as they age. If firms become more well known as they age and grow, they might expand the geographic market in which they can borrow. We see this in Table II where subsequent lenders are further away than the firm’ first lender. A selection problem may arise, however, since we only s know which of the firm’ lenders is first among the ones they have retained. 10 If they drop the early s (and close lenders), this could bias our calendar year coefficient upwards. However, once we include the firm’ age in the regression, the firm’ age and not calendar year should measure this effect. That s s the coefficient on calendar year is still important after including firm age suggests this source of bias is not significant. We can also directly test the importance of this bias. First relationships are likely to start soon after a firm is founded. So if we include only firms whose first relationship begins soon after the firm was born, we will have a sample that is less subject to the alleged selection bias. When we cut the sample down to firms whose first observed lending relationship starts during their first five years of life, we lose a third of the observations. Truncating the sample in this way, however, raises, rather than lowers, the coefficient on calendar year. For the firms for whom we are less likely to miss the first lending relationship, the distance between firms and their lenders grows at 3.9% per year (Table II, column VII). 10 We define the firm’ first lender as the lender with whom the firm has been conducting s business the longest. If the firm began its relationship with multiple institutions in that year, we classify all such lenders as the ‘ first’ lender. Financial institutions with which the firm has no interaction at the time of the survey (credit, deposit, or service) would not appear in our sample although they may have been the firm’ first lender. s 17 4. Firm characteristics. In addition to systematic changes in the lending environment for small firms we are also interested in the cross sectional variation in the distance between a firm and its lender. Different firms have varying potential for moral hazard, and thus a different need to be monitored. They have different degrees of informational transparency. Both of these effects may help explain the cross sectional variation in how far firms are from their lenders. Additionally, by controlling for firm characteristics, we have an additional robustness test of our finding that there has been a systematic increase in distance between lender and borrower. Finally, we will also get a sense of what factors might reduce the problem of lending at a distance. Information about a firm accumulates over time. The longer it has operated, the greater the firm’ track record, and the more the market may know about the firm. To measure this effect we s include in the regression both the firm’ age as well as whether this is the firm’ first lender. Firm s s age is measured as of the date the relationship with the institution began. Thus if the firm began business in 1975 but began a relationship with the lender in 1980, we code firm age as five. Firms that are older when they start borrowing from an institution do borrow further from home, but the coefficient is effectively zero (0.001). We do find that as the firm expands its circle of lenders it also increases the distance to its lender. A firm’ first lender is about 16 percent closer than its subsequent s lenders (see Table II, column I). This coefficient should be interpreted with caution. If we estimate this coefficient using only within firm variation (comparing a firm’ first to later lenders), the s coefficient is zero (Table II, column IV). If we estimate the coefficient using between firm variation (comparing firms with only a first lender to firms with multiple lenders), we find that the coefficient is significantly larger (Table II, column V). Thus the firms that have expanded to multiple lenders 18 are further from their lenders (both first and later) on average than firms who have only a single lender. Informational transparency, or the ability to evaluate the firm’ credit quality at low cost, s should lower the cost of lending to a firm. To the extent that this has a differential effect on creditors that are physically further from the firm it will raise the average distance between firms and lenders. We measure information transparency four ways. First, we identify firms that have a business credit card. Since these are usually granted based on a credit report, this implies the external market knows a sufficient amount about the firm to grant it credit based on information in computer files. Thirty- two percent of the firms in the sample have a business credit card.11 The survey also asks whether the person answering the income statement and balance sheet questions had records such as tax forms and/or financial statements to help in answering these questions. The existence of such records suggests greater transparency to outside investors. The final measures of record keeping we use are from the governance structure of the firm. We use the fraction of the equity owned by the largest shareholder as an (inverse) measure of shareholder dispersion. The more dispersed the shareholders, the greater the need for the firm to systematize their reporting function and make information easily accessible to outside investors (alternatively, the firm attracts dispersed shareholders only when reporting is systematic). We also include an indicator of whether the firm is a franchise – a franchise is likely to have a more systematized reporting structure in order to measure franchise fees. The expanded regressions with the additional firm characteristics are reported in Table III. The presence of a credit card has no effect on the distance to the firm’ lenders. This may be because s 11 The survey also asks if the owner uses a personal credit card for business expenses. Fifty- four percent of the firms have either a business credit card or a personal credit card that the owner uses for business expenses. 19 credit cards are freely available, or because they depend on the personal history of the owner rather than that of the business. However, the remaining firm characteristics are important. Firms that have financial records detailing their financials do borrow further from their lenders. The difference is 9 percent and is statistically significant (see Table III, column II).12 The ownership structure has a large and statistically significant impact on distance. Expanding the largest shareholder’ stake from zero s (a completely diffuse ownership structure) to one hundred percent, lowers the distance to a firm’s lender by twenty-six percent (t=3.4). We also find firms that are franchises – and are thus expected to be more transparent – also borrow 29 percent further away (t=3.2). We examine the characteristics of the manager to determine whether older and more experienced managers borrow from institutions that are further from the firm. Both the age and years of business experience are measured in the year the relationship started. Neither variable has a significant effect on the distance between borrower and lender. Also, whether the firm is managed by a non-owner does not appear to matter. The final set of firm characteristics we examine is the sales region of the firm. The larger the firm’ sales region the more likely distant lenders will know about the firm. In the survey, the firms s are asked if their sales region is local, regional, or national. As the firm’ sales region expands, so s does the distance between the firm’ headquarters and their lender. Most of the increase, however, s occurs when the firm shifts from a regional sales region to a national sales region. The distance to the lender increases by 29% in this case (t=4.9). 12 If we allow the slope on calendar year to vary by whether the firm had records, we find that firms with records are not only further away on average, but this distance is growing faster. The coefficient on calendar year is 2.9 percent per year for firms without records and 3.9 percent per year for firms with records. The difference is not statistically significant (p=0.25). 20 C. Method of Communicating with Lender. Distance can be related to the costs of transmitting information. How this information is transmitted provides more evidence about the type of information being exchanged. In addition to knowing how far a lender is from a firm, we also know the most frequent method by which the firm and the lender conduct business. Firms conduct business with their lender in person (46% of the firm-lender pairs), by phone (19%), or by mail (35%). Relationship lending – which is based on the collection and processing of soft information – is more likely to be based on personal contact. Thus we would expect that the use of written communication (mail) instead of personal contact to be correlated with more informationally transparent firms. These firms do not need to rely on personal contact to provide the lender with the soft information they require. They are more likely to have their lender rely on the type of hard information that is easily transmitted using technology across greater distances and at lower cost. As with distance, we find that personal communication between lenders has declined from 68% for relationships started in the seventies to 34% in the nineties (Table I – Panel B). We examine the determinants of the communication method chosen by estimating an ordered logit model with “in person” being the lowest category and “by mail” being the highest category. Therefore, positive coefficients indicate variables that raise the probability of communicating by mail and lower the probability of communicating in person. We expect higher values to be correlated with more informationally transparent firms. The variables we use to explain the method of communication are the same as we used in the distance regression. 1. Firm Characteristics: Informational Transparency The role of banks as a monitor and the use of checking accounts to monitor the firm can be 21 seen in the estimates (see Table IV). Firms that borrow from a bank are more likely to communicate in person. If they have a checking account at the bank, this increases the probability of communicating in person even more. Comparing a firm that borrows from a bank where it has a checking account to one that borrows from a non-bank where it does not have a checking account, the probability of communicating in person drops from 74 to 14 percent.13 We also find that firms that we expect to be more informationally transparent, are more likely to use arm’ length communication (phone and mail) rather than communicating in person. The s probability of communicating in person drops for firms that used records to answer the financial statement questions in the survey, have diffuse ownership structures, and have a national sales region. The last finding is particularly interesting. It suggests that firms with a national reach do not borrow from lenders remote from the head office simply because the firm has local branches near the lender (see Table III). In fact, lenders do indeed appear far from such firms because the method of communication is by phone or mail (see Table IV). Rather, a national sales region seems to expand the pool of lenders the firm can borrow from because the firm has a national image and is therefore more transparent. Firms are more likely to use personal communication with their first lender and then move away from personal communication with later lenders. Thus as firms become better known, they rely less on personal communication. Such firms are more likely to have a documented track record – just the type of hard information which makes personal communication less necessary. Unlike our distance regressions, we find an independent role for firm age. Firms that are older when they begin a relationship with a lender are less likely to communicate in person, although the magnitude of the 13 All other variables are set equal to their sample means, when probabilities are calculated. 22 effect is small. A firm that is five years old when it begins borrowing from its lender, is one percent less likely to communicate in person than a start up. 2. Changes in the Lending Market. The distance between firms and their lenders have grown systematically over the last two decades. We would expect the form of communication to mirror these changes. Holding the firm characteristics constant, relationships that started more recently are less likely to communicate in person with their lender and more likely to communicate by mail (See Table IV, column I and Figure II). The probability that a firm will communicate with its lender in person drops from 59 percent for relationships that started in 1973 to 36 percent for relationships that started in 1993. This coefficient is statistically significant (t=7.3). Although the shift from communicating in person to communicating by the mail is related to the shift to more distant lenders, the two effects are not exactly the same. To test this hypothesis, we include the distance to a firm’ lender as an explanatory variable in our estimated multinomial s logit. As expected, firms whose lenders are further away are less likely to communicate in person (see Table IV, column II) because, not surprisingly, distance raises the cost of personal communication. Raising the log distance from the 25th percentile (1.1) to the 75th percentile (3.8) lowers the probability of communicating in person from 58 percent to 30 percent (t=23.3). Including distance in our model, reduces the coefficient on calendar year from the 0.047 to 0.032. The fact that the coefficient on the year the relationship started is still positive and statistically significant (t=4.7) implies that the systematic change in how firms communicate with their lenders is more than just the effect of increased physical distance. Holding the distance to a firm’ lender constant, there has s still been a change in the way firm’ communicate – away from face to face communication and s 23 toward communication through the phone or mail. The probability of communicating in person now drops from 52 percent for a relationship started in 1973 to 37 percent for a relationship started in 1993. The decline is 67 percent of the earlier estimates when distance is not included, and is still quite large. This last finding is interesting. Recall that all variables are measured in 1993. So a firm that started a relationship with a lender 25 miles away in 1993 tends to use the phone more often while a firm that started a relationship with a lender 25 miles away in 1983 tends to use personal contact more often in 1993. One explanation could simply be habit persistence. If the small business owner had to visit her lender personally in 1983, she may still pay visits instead of picking up the phone because she has built up a friendship. We are skeptical that this could account for such a large effect -- for a time-pressured small businesswoman, friendship is likely to rank after convenience. A more plausible explanation is selection; those who borrowed 25 miles away in 1983 were those who had a lower cost of maintaining personal contact with their lender. For example, they may have had other business in the town 25 miles away which reduced the cost of paying a visit to a lender in that town. As a result, in 1993 they are likely to still continue paying personal visits to the lender. An immediate implication is that the reduced necessity to do business in person has increased the availability of finance. No longer is finance available to only those that have low costs of traveling to meet the lender in business. “Distance” has effectively been reduced. To summarize, we find that the distance between firms and lenders is increasing over time. Firms and lenders appear to be using more impersonal ways of transacting over time, even accounting for the increase in distance. These changes does not seem to be because of any systematic change in the characteristics of firms. Could they be because the environment – and not necessarily 24 technology – has changed? This is what we examine in the next section. IV. Systematic Changes in the Lending Environment. By including a time trend we have been able to characterize the changing environment in which small firms and their lenders operate. However, we have not directly identified the source of this change – other than by showing it is not simply a change in the distribution of firm characteristics. In this section, we explore several changes in the small business lending environment that may explain our results. A. Bank Consolidation. One of the major trends in banking has been the consolidation of the industry. The number of banks in the U.S. has declined by about 30 percent in the last decade (Berger, Demsetz, and Strahan, 1999). Much of this decline has been in the form of mergers (Berger, Saunders, Scalise, and Udell, 1997). The existence of fewer banks could explain the trend we find toward greater distance between firms and their lenders. To control for differences in bank density across regions and across time we include a measure of bank density in our model. We don’ know the exact location of each firm. We only t know which of nine census regions the firm resides in. Thus our bank density variable (log of banks per 1000 square miles) varies across time (the year a relationship started) and area (the nine census regions). We also calculate branch density (log of branches per 1000 square miles) since it is the number of physical locations which most directly affect where and how firms build their lending relationships. Bank density is correlated with the average distance between firm and lender (see Table V, 25 column II). Every ten percent increase in the density of banks lowers the average distance by 0.8 percent (t=3.7). The coefficient on branch density is smaller economically and less statistically significant (t=1.9). Each ten percent increase in branch density lowers distance by 0.4% (regression not reported). Based on the magnitude of these coefficients, changes in bank or branch density do not explain the growing distance between firms and their lenders. The increase in distance attributable to changes in density over time is only 2 percent – a small fraction of the actual increase. Although there has been a significant loss of banks, the effect on the distance from lender and method of communication has been minimal. A simple explanation of this finding can be found in Figure III. The decline in banks has been dramatic, but only starts in the mid eighties. The effects we find begin much earlier (see Figure I and II). This is why the inclusion of bank density does not change the coefficient on calendar year (see Table V). More importantly, the number of branches – the physical locations that firms use to conduct their business and the primary determinant of distance – have not fallen (see Figure III). The number of bank branches has risen about 4 percent per year. Much of this growth in the last decade has come from banks acquiring savings and loans. Thus counting both sources, the total number of branches has remained static since the mid eighties (see Figure III and Berger, Demsetz, and Strahan, 1999). Finally, recall that we even find non-banks are moving away from firms. Since it is not clear that non-banks are consolidating to the same extent as banks, we have to find a better explanation. B. Growing Use of Information Technology. If computers and information technology allow lenders to do their job more efficiently, we should see greater use of the technology as well as a transformation in the way lending is conducted (Gorton, 1996). As discussed above, computers are very good at processing information using 26 predefined rules. Our finding that the physical distance between lender and borrower has expanded and that the interaction between the two has become less personal is consistent with the intuition that information technology is replacing the traditional role of the loan officer. In this section we try to connect our results to measures of information technology use. The use of credit scoring models is an example of the classic substitution of capital for labor. Previously, loan officers would read the application material from the borrower, talk with the borrower, maybe interview references and then make a decision – a very labor intensive process – but the nature of the information upon which the decision was based demanded such time (Mester, 1997). Credit decisions based on credit reports (computerized data) and analytic decision rules, however, require less of the loan officer’ time. It is not that personal intervention has been s eliminated – it has just been focused on the most marginal decisions. Loan originations involve fewer people and more computers. To see if this is what explains our findings, we collected data on bank employment. The ideal measure would be the number of loan originations per loan officer, as this is a measure of labor productivity and should rise systematically as information technology supplements the efforts of the loan officers. In the absence of this data, the empirical challenge is to find the correct normalization. Bank employment has grown over the last two decades. However, when normalized by either total lending or the size of the local economy, bank employment is shrinking (see Figures IV-A and IV-B). To test the hypothesis that changes we find in the lending market are due to the greater use of information technology, we include in our regressions the log of bank employees in the region standardized by total loan volume or by the size of the economy (total regional output). The results are reported in Table V. In both cases, fewer employees are associated with a greater distance 27 between lender and borrower. However, only when we standardize number of employees by the total output, is the coefficient statistically significant (t=3.5 – see Table V, column III). In this case, every ten percent decrease in the number of employees to gross regional product raises the distance to a firm’ lender by 5.4 percent.14 This explains a large fraction of the increase in distance over the last s two decades. The average ratio of bank employees to gross regional product (in employees per million dollars) falls from 0.77 to 0.26. Based on the coefficient estimate in Table V (column III), this implies that distance should have risen by 58 percent.When we include employees standardized by gross regional product, the coefficient on year the relationship originated drops significantly and is no longer statistically different from zero (t=-0.5, see Table V, column III). Thus increases in bank productivity correlate strongly with growing distance. The results for method of communication are qualitatively similar (see Table VI, column III). As the number of employees standardized by total loan volume or gross regional product has declined distance has increased. In this model, both variables are statistically significant (t=-4.8 and -2.8). As with distance, these variables explain a large fraction of shift away from personal communication. The coefficient on year the relationship originated is again not significant at standard levels (p = 0.21). The explanatory power of the time trend represented by this variable has been absorbed by the new variable. Given the regression results and Figures I and II there may be concern that any trending variable will explain the shift in distance and method of communication that we document. This is not correct, as the results in Table V and VI demonstrate. The data are able to distinguish between 14 Using total loans and total output in real dollars does not alter our results significantly. For example, the coefficient on the log of the number of employees to gross regional product in real dollars is -0.475 opposed to -0.537 when nominal dollars are used. The t-statistic is -3.2. 28 a simple time trend (the year variable) and the standardized bank employees variable, with only the latter being statistically significant. However, to demonstrate our point more strongly, we ran an additional test. We created a new variable which is the log of bank employees standardized by total loans and by total output, not for the region in which the firm is located, but for the United States as a whole. This variable has time series variation, but no cross regional variation, by construction. It is correlated with the region variable, but not perfectly (?=0.79). When we include both the region specific employee variables and the aggregate (US) employee variables, the national variables are statistically insignificant both singly and jointly (see Table V, column IV). The distance between a firm and its lender depends upon the use of more productive techniques by banks in the firm’s region, not in the country at large. 15 V. Does More Timely Information Explain the Growing Distance. Lenders may be willing to go further because they expect to get information for the purposes of monitoring and controlling firms more easily and quickly at a distance than in the past. Credit reports from centralized bureaus may be one source of this additional information. If new technologies permit better monitoring and control at a distance, we should see that the ability to borrow at a distance is no longer such a strong signal of the intrinsic credit quality of the firm. In other words, in the past only unimpeachable credits could borrow at a distance because there was no way for a distant lender to anticipate when a weaker borrower would get into trouble, and very costly to resolve distress at a distance when the borrower was in trouble. More recently, however, 15 This is consistent with Jayaratne and Strahan (1996), who argue that the liberalization of banking markets in various states over the 1980s did improve the quality and efficiency of lending. 29 lower quality borrowers should be able to get credit at the same distance because information costs are lower and timely intervention is now possible. We have two immediate problems in implementing this test. First, we do not have a measure of the credit quality of the borrower when the relationship started. However, we do know the rate on the last loan the firm obtained, as well as whether the firm's last loan application was approved or rejected. This gives us two measures of recent credit quality. A second problem is that lender distance is, by itself, not a perfect measure of a firm's intrinsic ability to tap into a wider pool of lenders. Lenders can be at a distance because information about the firm’ quality is widespread (i.e., it is informationally transparent). But a firm can also be s at a distance from its lenders simply because there are no nearby lenders (firms in rural areas are 13% further from their lenders – see Table III), or no nearby lender will lend. We can, however, determine the ability to borrow at a distance that is correlated with public information. This is the estimated distance based on the regression coefficients in Table III and is a measure of the transparency of the firm.16 If poorer credits can borrow at a distance now, while they could not in the past, predicted distance from lending relationships which were established in recent years should be a weaker signal of credit quality than distance from relationships set up in the distant past. In other words, if a firm could establish a relationship at a distance in the early 1980s, it must have been really high quality, 16 We use the coefficient estimates from Table III column II. However, when we predict distance, we use only those variables which we think measure the firm’ informational transparency. s These include whether the firm uses records, whether the firm has a business credit card, the ownership share of the largest shareholder, whether the firm is a corporation, whether the firm is a franchise, whether the firm is owner managed, the age and experience of the owner, and the sales region of the firm. 30 while if it could establish a relationship at a distance in the early 1990s this may be a positive but a weaker signal of quality. We examine the effect of predicted distance on the price and availability of credit in Tables VII and VIII, both by itself and when interacted with time. This allows us to see the effect on informational transparency on the cost and availability of finance and how this relationship has changed over our sample period. In Table VII, we regress the rate a firm obtains on its most recent loan against predicted distance as well as other controls from Petersen and Rajan (1995). Predicted distance has a negative and significant effect on the rate (see column I). Raising predicted distance by fifty percent lowers the cost of borrowing by 25 basis points. This is twice the effect of a fifty percent increase in the size of the firm. Thus the ability to borrow at a distance, predicted by measures of the firm’s informational transparency, seems to be a good signal about the credit quality of the firm. When coupled with our finding that banks are physically much closer to clients, this supports the theoretical intuition that banks focus on informationally opaque clients. When we include an interaction between the year the relationship started and distance, however, the coefficient for the interaction is not significantly different from zero (see column II). Although predicted distance has a large effect on the cost of capital, this effect does not appear to have changed over time. In column III we estimate standard errors by bootstrapping, and this does not significantly change the statistical power of our results.17 17 The reported standard errors in column II are OLS standard errors. This is a problem since predicted distance and predicted distance interacted with the year the relationship started are predicted regressors. We estimated standard errors by bootstrapping the model. A sample with replacement was drawn. We then estimated the first stage, created the predicted distance variables given those estimates and then estimated the second stage (the model in Table VII column II). This gave us a single estimate of the coefficients. This was done 1000 times and the standard error of this distribution of estimates is reported in column III (Table VII). 31 Instead of predicted distance as a measure of transparency, we can use the predicted method of dealing with the lender (in person or not). A lender who is willing to deal with a borrower by phone or mail, rather than in person, should be signaling the credit quality of a borrower. It turns out that predicted method has exactly the same qualitative effect as predicted distance. If a lender is willing to accept dealing with the firm by mail or phone, the firm gets a lower rate on its most recent loan suggesting the firm is lower risk. Again, the coefficient of the interaction between predicted method and the year the relationship started is insignificant. The absence of time effects may reflect the fact that the rate for small loans is often determined by standard boilerplates, based on standard information such as firm size and industry. This is especially likely when loans are made by branches of large banks (see Brickley, Linck, and Smith (2000), Petersen and Rajan (1994), and Stein (2000)). Petersen and Rajan find that while the rate charged on a loan is not sensitive to measures of the soft information that is generated about a borrower (such as the duration of the relationship with the lender) the availability of credit is. So if soft information is more widely available over time, the reduced effectiveness of distance as a signal of credit quality over time will not be seen in the rate but in the availability of credit. As we will now argue, this is indeed the case. A direct measure of whether credit is available for a firm is whether the firm’ last s application for a loan is approved. Of course, a loan can be approved only if it was applied for. The firm will apply for credit if it needs funds and it thinks approval is sufficiently likely, and not otherwise. However, those who need funds but do not apply because they think their application will be refused should also be thought of as rationed. Therefore, a firm is included in the regression below if it either applied for a loan, or needed financing but did not apply for a loan because it felt 32 it would be denied.18 Firms whose loan application was approved are coded as one, the rest of the sample is coded as zero. Those who needed funds but did not apply are considered equivalent to those who needed funds and were rejected. Together with controls for availability taken from Petersen and Rajan (1995), we include predicted distance as an explanatory variable in Table VIII. Predicted distance is indeed a measure of intrinsic credit quality of the firm. Higher predicted distance is strongly positively correlated with greater availability of funding (column I). Increasing predicted distance from the 25 to the 75 percentile raises the probability a loan will be approved by 21 percentage points. Unlike the results for the interest rate (Table VII), the inclusion of an interaction between the year the relationship was started and predicted distance has a significant negative coefficient (column II) which persists even after we compute boot strapped standard errors (column III). This suggests that the distance a firm is able to borrow at has become less significant in distinguishing the credit quality of a firm in recent times. The coefficient estimates imply that role of predicted distance on the firms’access to capital is approximately 25 percent less in 1993 as compared with 1973. As before, the results are similar when we use predicted method of transacting instead of predicted distance. To check the robustness of these results we estimated several alternative specifications. We do not have a firm’ characteristics at the time the relationship started, but only at the time of the s survey. These characteristics may be very noisy proxies of firm transparency for relationships that started very early – the longer the lag, the more characteristics have changed. To see how important this problem is, we include only relationships started after 1983. In the regression explaining whether 18 The survey asks firms if there was a time during the last three years that the firm needed funds but did not apply for a loan because the firm felt they would be turned down. 33 the loan is approved, the estimated coefficient of predicted distance is now larger in magnitude (ß=1.001 and t=3.2), as is the coefficient of the calender year times predicted distance interaction (ß=-0.025 and t=-2.6, regression not reported). Even though the standard errors go up, but the estimates are still statistically significant. So it does not appear that noise in firm characteristics biases coefficients in a way to produce our results. We also estimated separate slopes including predicted distance if the relationship started before 1983, and predicted distance if the relationship started after. Predicted distance for relationships that started after 1983 is less important in determining whether a loan is approved. In summary, firms that are informationally more transparent (have greater predicted distance and lower probability of personal communication) face less credit rationing and are charged lower interest rates. However, the relation between predicted distance and credit availability is weakening over time. If predicted distance is no longer such a strong signal of credit quality, the implication is that riskier credits are being financed at a greater distance. These are the ones that have a greater propensity to get into trouble and face reduced availability later. But this means that lenders are willing to make bolder credit decisions at a distance now. One explanation is simply that the credit standards of lenders have been steadily deteriorating over the last twenty years, and that they are less interested in monitoring. This does not square with the fact that loan losses have not steadily risen over this period, and have no strong correlation with distance (see Table V, column V). We are left to conclude that a more distant, and less creditworthy, set of borrowers have become viable to lenders because improvements in technology allow cheaper screening, monitoring, and control, at a distance – our results are consistent with the information and communications revolution making distance less important. 34 VI. Conclusions. We have documented a trend in the distance between small firms and their lenders in the United States. Firms are choosing further lenders and are also communicating in more impersonal ways with them. The trend correlates well with the increase in loans per bank employee or total regional output per bank employee over time, suggesting that the advances in computing and communications have reduced the human component in lending decisions, allowing more impersonal and distant lending decisions to be made. Moreover, distant firms are no longer only the highest credit qualities, suggesting that a wider, and more distant, cross-section of firms can now obtain funding from a particular lender. By implication, economy wide credit availability, and competition in credit markets, has increased. The paper makes a number of contributions. For one, it focuses on new metrics for informational closeness – physical distance and method of communication. Others have used distance as a proxy for informational asymmetry (for example, see Covals and Moskowitz (1999), Garmaise and Moskowitz (1999), and Grinblatt and Keloharju (1999)). To the best of our knowledge, however, we have not seen this correlated with the nature of the institution or with credit availability and price. Our findings that informationally opaque firms have closer lenders, and that banks are closer than other lenders, are consistent with the theoretical view that banks are informationally close lenders making loans to opaque borrowers. These findings also have policy implications. If information technology can increase the services provided to, and competition in, the sector that has historically been viewed as the most informationally sensitive and thus most local, i.e., small business lending, then the relevant size of the market for anti-trust policy will have to be revised upwards over time. While our paper does not 35 provide a ready metric for how this should be done, it highlights this as an issue that deserves more research. Perhaps most important, however, greater credit availability to small businesses can lead to greater innovation, more competition, and more opportunity to those without resources. Our work therefore highlights an effect of technological change that goes against the popular Luddite belief that technological change serves only to enhance the power of a few while destroying the jobs of many. Quantifying the effect of greater credit availability is an important task for future work. 36 References. Berger, Allen, Rebecca Demsetz, and Philip Strahan, 1999, “The Consolidation of the Financial Services Industry: Cause, Consequences, and Implications for the Future” Journal of Banking and Finance 23, 137-194. Berger, Allen, Anthony Saunders, Joseph Scalise, and Gregory Udell, 1998, “The Effect of Bank Mergers and Acquisition on Small Business Lending”, Journal of Financial Economics, 187-229. Berger, Allen, Anil Kashyap, and Joseph Scalise, 1995, The Transformation of the U.S. Banking Industry: What a Long Strange Trip It’ Been, “ Brookings Papers on Economic Activity 2, 55-217 s Berger, Allen, and Gregory Udell, 1995 "Small Firms, Commercial Lines of Credit, and Collateral" Journal of Business. Boyd, J. and M. Gertler, “Are Banks Dead? Or Are the Reports Greatly Exaggerated?”, Quarterly Review of the Federal Reserve Bank of Minneapolis (Summer 1994), 2-23. Brickley, J., J. Linck, and C. Smith, 2000, “Boundaries of the firm: Evidence from the Banking Industry”, working paper, University of Rochester. Cole, Rebel and John Wolken, 1996, “Bank and Nonbank Competition for Small Business Credit: Evidence from the 1987 and 1993 National Surveys of Small Business Finances”, Federal Reserve Bulletin, 983-995. Coval, Joshua, and Tobias Moskowitz, “Home Bias at Home: Local Equity Preference in Domestic Portfolios”, 1999, Journal of Finance, forthcoming. Davis, Steve, and John Haltiwanger, “The Distribution of Employees by Establishment Size: Patterns of Change and Comovement in the United States: 1962-1985", Mimeo, University of Chicago. Diamond, Douglas, 1984, "Financial Intermediation and Delegated Monitoring," Review of Economic Studies LI, 393-414. Fama, Eugene, 1985, What's different about banks?, Journal of Monetary Economics 15, 29-36. Frame, Scott, Aruna Srinivasan and Lynn Woosley, 1999, “The Effect of Credit Scoring on Small Business Lending” working paper. Garmaise, Mark and Tobias Moskowitz, 1999, “Confronting Information Asymmetries: Evidence from Real Estate Markets”, mimeo, University of Chicago. Gordon, Robert, J. “Has the New Economy Rendered the Productivity Slowdown Obsolete?” Northwestern University Economics Department working paper. 37 Gorton, Gary, 1996, “Reputation Formation in Early Bank Note Markets” Journal of Political Economy 104, 346-397. Gorton, Gary, and Richard Rosen, 1995, “ Corporate Control, Portfolio Choice, and the Decline of Banking”, Journal of Finance, 50: (5) 1377-1420 DEC 1995 Greenbaum, Stuart, George Kanatas, and Itzhak Venezia, 1989, Equilibrium loan pricing under the bank client relationship, Journal of Banking and Finance 13, 221-235. Grinblatt, Mark, and Matti Keloharju, 1999 “Distance Bias, Language Bias, and Investor Sophistication: Results from Finland” UCLA working paper. Jayaratne, Jith and Philip E. Strahan, 1996, ``The Finance-Growth Nexus: Evidence from Bank Branch Deregulation," Quarterly Journal of Economics, CXI, 639-671. James, Christopher, "Some Evidence on the Uniqueness of Bank Loans," Journal of Financial Economics, XIX (1987), 217-235. Kroszner, R. And P. Strahan, 1999, “What drives Deregulation? Economics and Politics of the Relaxation of Bank Branching Restrictions”, Quarterly Journal of Economics, 1437-1467, 114. Merton, R.,1995, “Financial Innovation and the Management and Regulation of Financial Institutions”, Journal of Banking and Finance 19, 461-482. Mester, Loretta, Leonard Nakamura, and Micheline Renault, 1998, “Checking Accounts and Bank Monitoring” Federal Reserve Bank of Philadelphia working paper Mester, Loretta, 1997, “What’ the Point of Credit Scoring” Federal Reserve Bank of Philadelphia s Business Review 3-16. Mishkin, Frederic S. and Philip E. Strahan, 1999, “What Will Technology Do to Financial Structure?” Brookings-Wharton Papers on Financial Services, 249-287. Nakamura, Leonard, 1988, “A Loan Workouts and Commercial Bank Information: Why banks Are Special,” Federal Reserve Bank of Philadelphia working paper. Nocera, Joseph, 1994, A Piece of the Action: How the Middle Class Joined the Moneyed Class, (Simon and Schuster: New York). Padhi, Michael, Aruna Srinivasan, and Lynn Woosley, 1999, “Credit Scoring and Small Business Lending in Low and Moderate Income Communities” Federal Reserve Bank of Atlanta working paper. Petersen, Mitchell and Raghuram G. Rajan, 1994, “The Benefits of Lending Relationships: “ Journal 38 of Finance , Petersen, Mitchell and Raghuram G. Rajan, 1995, “The Effect of Credit Market Competition on Firm-Creditor Relationships," Quarterly Journal of Economics 110, 407-443. Rajan, Raghuram G, 1996, “Why Banking Has a Future: Towards a New Theory of Banking,” Journal of Applied Corporate Finance Rhoades, Stephen, 1996, “Bank Mergers and Industrywide Structure 1980-1994", Federal Reserve Board, Staff Study 169. Stein, Jeremy, 2000, “Information Production and Capital Allocation: Decentralized vs Hierarchical Firms”, mimeo, Harvard University. 39 Table I: Firm’ Lending Relationship s Panel A: Distance to a Firm’ Lender by Lender Type and Time s Lender Type Year Lending Relationship Began 1973-79 1980-89 1990-93 Total Banks 15.8 34.0 67.8 42.5 2.0 4.0 5.0 4.0 6.0 12.0 20.0 14.0 Non-Banks 235.9 222.1 280.5 251.6 15.5 42.0 54.0 45.0 71.0 215.5 332.0 255.0 Non-Financial Firms 117.3 165.9 209.2 182.5 17.5 29.0 32.0 30.5 60.0 141.5 235.0 164.0 Total 51.2 92.6 161.3 114.7 3.0 7.0 15.0 9.0 10.0 33.0 91.5 42.0 Panel B: Method of Communication to a Firm’ Lender by Lender Type and Time s Lender Type Year Lending Relationship Began 1973-79 1980-89 1990-93 Total Banks 0.77 0.67 0.54 0.64 1.3 1.5 1.7 1.5 Non-Banks 0.27 0.12 0.09 0.11 2.2 2.5 2.6 2.6 Non-Financial Firms 0.35 0.20 0.18 0.20 2.2 2.4 2.4 2.4 Total 0.68 0.49 0.34 0.45 1.5 1.8 2.1 1.9 40 Note: Panel A contains the average distance between a firm and its lender classified by lender type and the year the relationship began. Non-banks include finance companies, insurance companies, brokerage firms, leasing companies, mortgage banks, and venture capitalists. The first entry in each cell is the mean distance, the second entry is the median distance, and the third entry is the 75 percentile distance. The far right column contains the sample average for lender types, and the bottom row contains the sample average for each decade. The sample contains 5981 firm lender pairs which were begun between 1973 and 1993. Panel B contains data on the predominant method of communication between the firm and its lender. The communication can be in person (1), by phone (2) or by mail (3). The first number is the fraction of firm-lender pairs that communicate predominantly in person. The second number is the average value of the method variable. Higher values are associated with less personal communication and more communication by mail. 41 Table II: Determinants of Distance to the Firm’ Lenders s Independent Variables Models I II III IV V VI VII Firm’ age s 0.001 0.001 0.001 0.000 0.00310 -0.01110 (0.002) (0.002) (0.002) (0.002) (0.001) (0.006) Year relation started 0.0341 0.0331 0.0331 0.0401 0.0351 0.0381 0.0391 (0.005) (0.005) (0.005) (0.011) (0.006) (0.004) (0.006) Lender is bank -1.1781 -1.1661 -1.1731 -1.1221 -1.1031 -1.1861 -1.1991 (0.062) (0.062) (0.062) (0.085) (0.088) (0.050) (0.076) Lender is a non-financial firm -0.3761 -0.3321 -0.3731 -0.1251 -0.4751 -0.3691 -0.4161 (0.096) (0.096) (0.096) (0.132) (0.139) (0.077) (0.113) Lender provides a -1.0691 -1.0811 -1.0641 -1.0641 -1.1191 -1.0021 -1.1151 checking account (0.059) (0.059) (0.059) (0.082) (0.080) (0.047) (0.074) Lender is firm’ first lender s -0.1641 -0.1431 -0.1611 -0.026 -0.2591 -0.2041 -0.1595 (0.049) (0.048) (0.048) (0.077) (0.085) (0.039) (0.072) Observation from 1993 sample -0.0985 (0.039) Industry controls (59) 3.02 F-statistics (p-value) (0.000) Region controls (9) 3.09 (0.002) Firm controls (2878) 1.50 (0.000) R2 0.307 0.327 0.310 0.711 0.297 0.296 0.316 Number of Observations 5981 5981 5981 5981 2878 9385 4094 42 Note: The table contains regressions of the log of one plus the distance to the firm’ lender. The sample includes only s relationships beginning since 1973. Dummy variables are included for whether the lender is a bank or a non-financial firm. The missing category is non-bank financial lenders. Each observation represents a firm-lender pair. In some cases, firms began borrowing from more than one lender at the same time. In these cases, there is more than one observation for each firm. Model Description: II: Industry controls. The regression includes dummy variables to control for differences across the 59 two digit industries in the sample. III: Census region controls. The regression includes dummy variables to control for differences across the 9 two census regions in which the firms are located. IV: Within estimates. A dummy variable is included for every firm. Thus the coefficients are estimated based on variation from firm specific means. The R2 includes the firm controls. Without including the explanatory power of the firm effects, the R2 would be 0.302. V: Between estimates. The coefficients are estimated based on variation between firm specific means. Each observation represents the mean value for a given firm. VI: Sample includes observations from the 1993 and the 1987 National Survey of Small Business Finance. VII: Sample includes observations where the first reported lending relationship is within five years of the firm’s founding. 43 Table III: Determinants of Distance to the Firm’ Lenders s Independent Variables Models I II Firm’ age s 0.001 0.000 (0.002) (0.002) Year relation started 0.0331 0.0371 (0.005) (0.005) Lender is bank -1.1781 -1.1561 (0.062) (0.062) Lender is a non-financial firm -0.3761 -0.3781 (0.096) (0.096) Lender provides a checking acct -1.0691 -1.1111 (0.059) (0.059) Lender is firm’ first lender s -0.1641 -0.1341 (0.049) (0.049) Firm had records for filling out survey 0.0915 (0.042) Firm has credit card -0.003 (0.043) Ownership share of largest owner (%) -0.2551 (0.076) Corporation (1=yes) 0.069 (0.052) Franchise (1=yes) 0.2921 (0.091) Owner managed -0.044 (0.050) Owner’ age s 0.000 when relation began (0.003) Owner’ experience s -0.003 when relation began (0.003) 44 Sales area regional (1=yes) 0.073 (0.047) Sales area national (1=yes) 0.3601 (0.060) Firm in MSA (1=yes) -0.1295 (0.055) Herfindahl Index > 1800 0.046 (0.045) R2 0.307 0.318 Number of Observations 5981 5974 Note: The table contains regressions of the log of one plus the distance to the firm’ lender. The sample includes only s relationships beginning since 1973. Dummy variables are included for whether the lender is a bank or a non-financial firm. The missing category is non-bank financial lenders. Each observation represents a firm-lender pair. In some cases, firms began borrowing from more than one lender at the same time. In these cases, there is more than one observation for each firm. Model Description: I. This is the regression from Table II, column I. It is reported here for comparison. II. This model includes additional controls for information about the firm. 45 Table IV: Determinants of the Method Used to Communicate with Lender In Person, By Phone, or By Mail Independent Variables Models I II III Firm’ age s 0.0055 0.0055 0.0065 (0.002) (0.002) (0.003) Year relation started 0.0471 0.0321 0.0441 (0.007) (0.007) (0.007) Lender is bank -1.2431 -0.8931 -1.6001 (0.079) (0.082) (0.098) Lender is a non-financial firm -0.4591 -0.3311 -0.6521 (0.121) (0.126) (0.160) Lender provides a checking acct -1.6601 -1.2651 -1.4261 (0.077) (0.080) (0.082) Lender is firm’ first lender s -0.2181 -0.1881 -0.2931 (0.064) (0.066) (0.075) Firm had records for filling out survey 0.087 0.051 0.11510 (0.057) (0.059) (0.066) Firm has credit card 0.048 0.033 0.12710 (0.057) (0.059) (0.066) Ownership share of largest owner (%) -0.2155 -0.108 -0.5071 (0.101) (0.105) (0.119) Corporation (1=yes) 0.109 0.094 0.1885 (0.071) (0.074) (0.081) Franchise (1=yes) -0.017 -0.178 0.088 (0.120) (0.127) (0.142) Owner managed -0.017 0.007 -0.13310 (0.067) (0.069) (0.079) Owner’ age s 0.004 0.005 0.005 when relation began (0.004) (0.004) (0.004) Owner’ experience s -0.003 -0.003 -0.005 when relation began (0.004) (0.004) (0.004) 46 Sales area regional (1=yes) -0.023 -0.061 0.090 (0.064) (0.066) (0.073) Sales area national (1=yes) 0.3851 0.2061 0.7431 (0.079) (0.082) (0.094) Firm in MSA (1=yes) 0.2961 0.4331 0.3101 (0.076) (0.080) (0.086) Herfindahl index > 1800 -0.077 -0.097 -0.1635 (0.060) (0.062) (0.070) Log (1 + Distance from lender) 0.4361 (0.019) a1 92.9441 63.163 85.6761 (12.956) (13.357) (14.448) a2 93.9451 64.479 (12.958) (13.358) Pseudo R2 0.215 0.263 0.283 Number of Observations 5945 5945 5945 Note: The dependent variable denotes the predominant method for communicating with lender (1 = in person, 2 = by phone, 3 = by mail). The coefficients are estimated as an ordered logit model. The sample includes only relationships beginning since 1973. Dummy variables are included for whether the lender is a bank or a non-financial firm. The missing category is non-bank financial lenders. Each observation represents a firm-lender pair. In some cases, firms began borrowing from more than one lender at the same time. In these cases, there is more than one observation for each firm. 47 Table V: Determinants of Distance to the Firm’ Lenders s Includes Changes in the Lending Environment Independent Variables Models I II III IV V Log(Banks/1000 Square Miles) -0.0831 (0.023) Log(Bank Employees/Loans) -0.157 -0.171 (0.115) (0.122) Log(Bank Employees/GDP) -0.5371 -0.5381 (0.153) (0.157) Log(Bank Employees) Log(Total Loans) Log(GDP) Log(Bank Employees/Total Loans) 0.408 (based on national numbers) (0.470) Log(Bank Employees/GDP) -1.383 (based on national numbers) (1.608) Log(Loan Losses/Total Loans) 0.011 (0.050) Firm’ age s 0.000 0.000 0.001 -0.001 0.000 (0.002) (0.002) (0.002) (0.002) (0.002) Year relation started 0.0371 0.0361 -0.007 -0.076 0.0361 (0.005) (0.005) (0.015) (0.092) (0.007) Lender is bank -1.1561 -1.1431 -1.1121 -1.1121 -1.1551 (0.062) (0.062) (0.063) (0.063) (0.062) Lender is a non-financial firm -0.3781 -0.3761 -0.3691 -0.3691 -0.3781 (0.096) (0.096) (0.098) (0.098) (0.096) Lender provides a checking acct -1.1111 -1.1111 -1.1601 -1.1571 -1.1111 (0.059) (0.059) (0.061) (0.061) (0.059) Lender is firm’ first lender s -0.1341 -0.1331 -0.1411 -0.1371 -0.1341 (0.049) (0.049) (0.050) (0.050) (0.049) 48 Firm had records for filling out 0.0915 0.0865 0.0975 0.0975 0.0915 survey (0.042) (0.042) (0.044) (0.044) (0.042) Firm has credit card -0.003 -0.015 -0.012 -0.011 -0.003 (0.043) (0.043) (0.044) (0.044) (0.043) Ownership share of largest owner -0.2551 -0.2551 -0.2831 -0.2841 -0.2551 (%) (0.076) (0.076) (0.078) (0.078) (0.076) Corporation (1=yes) 0.069 0.08810 0.076 0.077 0.069 (0.052) (0.052) (0.054) (0.054) (0.052) Franchise (1=yes) 0.2921 0.3051 0.3161 0.3181 0.2931 (0.091) (0.091) (0.093) (0.093) (0.091) Owner Managed -0.044 -0.046 -0.049 -0.051 -0.044 (0.050) (0.050) (0.052) (0.052) (0.050) Owner’ age s 0.000 -0.000 -0.000 0.000 0.000 when relation began (0.003) (0.003) (0.003) (0.003) (0.003) Owner’ experience s -0.003 -0.003 -0.003 -0.003 -0.003 when relation began (0.003) (0.003) (0.003) (0.003) (0.003) Sales Area Regional (1=yes) 0.073 0.072 0.051 0.051 0.074 (0.047) (0.047) (0.048) (0.049) (0.047) Sales Area National (1=yes) 0.3601 0.3671 0.3351 0.3371 0.3611 (0.060) (0.060) (0.061) (0.062) (0.060) Firm in MSA (1=yes) -0.1295 -0.1551 -0.1475 -0.1501 -0.1295 (0.055) (0.056) (0.058) (0.058) (0.056) Herfindahl Index > 1800 0.046 0.017 0.042 0.043 0.046 (0.045) (0.045) (0.046) (0.046) (0.045) R2 0.318 0.319 0.314 0.314 0.318 Number of Observations 5974 5974 5662 5662 5974 49 Note: The table contains regressions of the log of one plus the distance to the firm’ lender. The sample includes only s relationships beginning since 1973. Dummy variables are included for whether the lender is a bank or a non-financial firm. The missing category is non-bank financial lenders. Each observation represents a firm-lender pair. In some cases, firms began borrowing from more than one lender at the same time. In these cases, there is more than one observation for each firm. The estimates in column I are the estimates from Table III, column II, and are reported here only for comparison. The other columns contain controls for changes in the lending environment over the sample period. In most cases, the variables vary across both time and census region. Both total loan amounts and GDP are reported in real dollars. 50 Table VI: Determinants of the Method Used to Communicate with Lender Includes Changes in the Lending Environment Independent Variables Models I II III IV V Log(Banks/1000 Square Miles) -0.045 (0.031) Log(Bank Employees/Total Loans) -0.7441 -0.7791 (0.156) (0.165) Log(Bank Employees/GDP) -0.5631 -0.6011 (0.204) (0.209) Log(Bank Employees) Log(Total Loans) Log(GDP) Log(Bank Employees/Total Loans) -0.066 (based on national numbers) (0.654) Log(Bank Employees/GDP) 1.859 (based on national numbers) (2.198) Log(Loan Losses/Total Loans) 0.004 (0.068) Firm’ age s 0.0055 0.0055 0.0055 0.00510 0.0055 (0.002) (0.002) (0.002) (0.002) (0.002) Year relation started 0.0471 0.0471 -0.025 0.090 0.0471 (0.007) (0.007) (0.020) (0.125) (0.009) Lender is bank -1.2431 -1.2361 -1.2261 -1.2261 -1.2431 (0.079) (0.079) (0.080) (0.080) (0.079) Lender is a non-financial firm -0.4591 -0.4561 -0.4641 -0.4711 -0.4581 (0.121) (0.121) (0.124) (0.124) (0.121) Lender provides a checking acct -1.6601 -1.6611 -1.7041 -1.7061 -1.6601 (0.077) (0.077) (0.079) (0.079) (0.077) Lender is firm’ first lender s -0.2181 -0.2171 -0.2211 -0.2211 -0.2181 (0.064) (0.064) (0.065) (0.065) (0.064) 51 Firm had records for filling out 0.087 0.083 0.082 0.083 0.087 financial part of survey (0.057) (0.057) (0.059) (0.059) (0.057) Firm has a credit card 0.048 0.042 0.060 0.059 0.048 (0.057) (0.057) (0.059) (0.059) (0.057) Ownership Share of largest owner -0.2155 -0.2155 -0.2405 -0.2395 -0.2155 (%) (0.101) (0.101) 0.104 0.104 0.101 Corporation (1=yes) 0.109 0.12010 0.12910 0.12910 0.109 (0.071) (0.072) (0.074) 0.074 0.071 Franchise (1=yes) -0.017 -0.011 0.002 0.001 -0.017 (0.120) (0.120) (0.124) (0.124) (0.120) Owner Managed -0.017 -0.018 -0.024 -0.023 -0.017 (0.067) (0.067) (0.069) (0.069) (0.067) Owner’ age s 0.004 0.004 0.003 0.003 0.004 when relation began (0.004) (0.004) (0.004) (0.004) (0.004) Owner’ experience s -0.003 -0.003 -0.004 -0.004 -0.003 when relation began (0.004) (0.004) (0.004) (0.004) (0.004) Sales Area Regional (1=yes) -0.023 -0.023 -0.049 -0.048 -0.023 (0.064) (0.064) (0.065) (0.065) (0.064) Sales Area National (1=yes) 0.3851 0.3881 0.3261 0.3251 0.3851 (0.079) (0.079) (0.081) (0.081) (0.079) Firm in MSA (1=yes) 0.2961 0.2821 0.2171 0.2151 0.2961 (0.076) (0.077) (0.080) (0.080) (0.076) Herfindahl Index > 1800 -0.077 -0.093 -0.098 -0.103 -0.077 (0.060) (0.061) (0.062) (0.062) (0.060) a1 92.0611 91.3581 -46.467 181.12 91.968 (12.956) (12.973) (39.654) (248.46) (18.226) a2 93.2561 92.5531 -45.269 182.32 93.163 (12.958) (12.975) (39.654) (248.46) (18.227) R2 0.215 0.215 0.218 0.218 0.215 Number of Observations 5945 5945 5633 5633 5945 52 Note: The table contains regressions of the log of one plus the distance to the firm’ lender. The sample includes only s relationships beginning since 1973. Dummy variables are included for whether the lender is a bank or a non-financial firm. The missing category is non-bank financial lenders. Each observation represents a firm-lender pair. In some cases, firms began borrowing from more than one lender at the same time. In these cases, there is more than one observation for each firm. The estimates in column I are the estimates from Table IV, column I, and are reported here only for comparison. The other columns contain controls for changes in the lending environment over the sample period. In most cases, the variables vary across both time and census region. 53 Table VII: Determinants of Interest Rate on Firm’ Most Recent Loan s Independent Variables Models I II III IV V VI Maturity matched treasury rate 0.3671 0.3671 0.3671 0.3691 0.3701 0.3701 (0.029) (0.029) (0.029) (0.029) (0.029) (0.029) Log(Firm’ assets) s -0.2431 -0.2431 -0.2431 -0.2421 -0.2421 -0.2421 (0.021) (0.021) (0.021) (0.021) (0.021) (0.022) Outside debt/assets -0.107 -0.107 -0.107 -0.104 -0.103 -0.104 (0.067) (0.067) (0.080) (0.067) (0.067) (0.079) Corporation (1=yes) -0.3851 -0.3841 -0.3841 -0.3761 -0.3761 -0.3761 (0.086) (0.086) (0.099) (0.086) (0.086) (0.100) Log(1+firm age) -0.024 -0.032 -0.032 -0.015 -0.012 -0.012 (0.044) (0.048) (0.049) (0.044) (0.047) (0.050) Loan has floating rate (1=yes) 0.030 0.030 0.030 0.032 0.032 0.032 (0.074) (0.075) (0.071) (0.075) (0.075) (0.072) Loan is collateralized 0.091 0.091 0.091 0.088 0.087 0.087 (0.075) (0.074) (0.070) (0.074) (0.074) (0.071) Lender is a bank (1=yes) -0.5651 -0.5671 -0.5671 -0.5571 -0.5561 -0.5561 (0.099) (0.099) (0.148) (0.099) (0.099) (0.150) Lender is nonfinancial firm -0.4605 -0.4595 -0.459 -0.4435 -0.4445 -0.444 (0.230) (0.230) (0.485) (0.230) (0.230) (0.484) Owner has had delinquency 0.4571 0.4571 0.4571 0.4631 0.4631 0.4635 (0.114) (0.114) (0.146) (0.114) (0.114) (0.150) Firm has had delinquency 0.3481 0.3501 0.3501 0.3511 0.3511 0.3511 (0.078) (0.079) (0.079) (0.079) (0.079) (0.078) Number of financial lenders 0.0751 0.0761 0.0761 0.0731 0.0731 0.0731 (0.021) (0.021) (0.018) (0.021) (0.021) (0.018) Predicted distance from Lender -0.5461 -0.5281 -0.5285 (0.179) (0.184) (0.217) Predicted distance * -0.001 -0.001 Year relation started (0.004) (0.003) 54 Predicted method used to -1.3281 -1.3981 -1.3981 communicate with lender (0.418) (0.542) (0.546) Predicted method * 0.005 0.005 Year relation started (0.024) (0.021) R2 0.174 0.175 0.175 0.175 0.175 0.175 Number of observations 3523 3523 3523 3523 3523 3523 Note: The dependent variable is the rate on the most recent loan. The average loan rate is 8.4%. The estimates in columns III and VI are identical to those in columns II and V, except that the standard errors have been estimated by the bootstrapping technique (1000 repetitions). 55 Table VIII: Probability of Loan Approval on Most Recent Loan Independent Variables Models I II III IV V VI Log(Firm’ assets) s 0.3501 0.3491 0.3491 0.3521 0.3491 0.3491 (0.026) (0.026) (0.026) (0.026) (0.026) (0.025) Return on assets (profits/assets) 0.3751 0.3861 0.3861 0.3821 0.3811 0.3811 (0.080) (0.080) (0.080) (0.079) (0.080) (0.080) Outside debt/assets 0.005 0.005 0.005 0.004 0.004 0.004 (0.015) (0.015) (0.020) (0.015) (0.015) (0.021) Corporation (1=yes) -0.034 -0.028 -0.028 -0.037 -0.036 -0.036 (0.104) (0.104) (0.115) (0.105) (0.105) (0.111) Firm age 0.008 0.008 0.008 0.007 0.007 0.007 (0.006) (0.006) (0.005) (0.005) (0.005) (0.005) Length of longest relationship 0.012 0.004 0.004 0.012 0.004 0.004 (0.009) (0.010) (0.010) (0.009) (0.010) (0.010) Debt from financial service provider 1.5901 1.5631 1.5631 1.5841 1.5541 1.5541 (%) (0.107) (0.108) (0.115) (0.107) (0.108) (0.113) Number of financial lenders 0.021 0.028 0.028 0.020 0.029 0.029 (0.029) (0.029) (0.029) (0.029) (0.029) (0.030) Herfindahl index > 1800 0.1875 0.1845 0.1845 0.1915 0.1875 0.1875 (0.080) (0.080) (0.078) (0.080) (0.080) (0.079) Predicted distance from lender 0.6801 0.8241 0.8241 (0.257) (0.269) (0.297) Predicted distance * -0.01010 -0.01010 Year relation started (0.006) (0.006) Predicted method used to 1.6141 2.9271 2.9271 communicate with lender (0.638) (0.928) (0.967) Predicted method * -0.0845 -0.0845 Year relation started (0.024) (0.043) Pseudo R2 0.172 0.173 0.173 0.172 0.173 0.173 Number of observations 4548 4548 4548 4548 4548 4548 Note: 56 The independent variable is 1 if the firm was approved for a loan and zero otherwise. The sample include both those firms that applied for a loan as well as those firms that did not apply for a loan because they expected to be turned down. The last group is coded as being turned down for a loan. Only loans which were applied for (or considered) in the last three years are included. The estimates in columns III and VI are identical to those in columns II and V, but the standard errors have been estimated by the bootstrapping technique (1000 repetitions). 57 Figure I: Changes in the Distance to Lender over Time. 0.5 90 91 89 84 85 92 0 87 88 93 77 Log(distance) 81 83 86 75 73 82 80 78 79 -0.5 74 76 -1 73 78 83 88 93 Year Note: This is the estimated distance between a firm and its lender based on the estimates from Table III. All variables except the year in which the relationship began is set equal to the sample mean. In the regression, the year variable is replaced by a set of year dummy variables. This allows the functional form of the relationship between distance and year have any form. 58 Figure II: Changes in the Method of Communicating with Lender over Time 0.7 89 90 91 85 92 0.6 83 87 74 86 88 73 0.5 77 80 81 82 75 84 79 78 0.4 76 0.3 73 78 83 88 93 Year Note: The estimated probability of using mail or phone as the most frequent method of communicating with a lender as a function of the year the lending relationship started. The alternative method of communication is in person. The probability is calculated for a new firm which was borrowing and had a checking account from a bank. The probabilities are based on multinomial logit (as in Table IV) except the variable Year relationship was started was replaced with a series of year dummy variables. 59 Figure III: Number of Banks and Bank Branches 15000 90000 80000 Number of Branches 14000 Number of Banks 70000 13000 60000 12000 50000 11000 40000 10000 30000 73 78 83 88 93 Year Banks Bank Branches Total Branches Note: The number of banks and bank offices is from taken from the FDIC web site. The number of banks are graphed along the left axis as a line and are stable until 1984 when they begin to decline. The number of offices are graphed along the right axis. The graph contains both the number of bank offices and the number of bank plus S&L offices. The data on S&L offices is available only since 1984. 60 Figure IV: Total Bank Employment A: Relative to Total Bank Lending -5.5 -6.0 -6.5 -7.0 -7.5 1973 1978 1983 1988 1993 B: Relative to Gross Region Product 0.0 -0.5 -1.0 -1.5 -2.0 1973 1978 1983 1988 1993 1 2 3 4 5 6 7 8 9 61 Note: Figure IV-A is the graph by census region (1-9) of the log of total bank employment divided by total bank lending (in $1000s). Both numbers are from the Federal Deposit Insurance Corporation (FDIC) web site.19 Both number are reported by state. Figure IV-B is the graph by census region of the log of total bank employment divided by total gross region product. Gross state product data was collected from the Bureau of Economic Analysis (BEA), Commerce Department web.20 Gross state product is only available starting in 1977 site and is reported in millions of dollars. Regional numbers are the sum of the numbers for each state in the region. 19 http://www2.fdic.gov/hsob/ 20 http://www.bea.doc.gov/bea/regional/gsp/gsplist.htm 62