Business Commitments, Personal Commitments and Credit Risk: Evidence from China

Transcription

1 Business Commitments, Personal Commitments and Credit Risk: Evidence from China February 20, 2014 Abstract This paper studies the relationship between collateral/guarantees and credit risk for loans made to small and medium-size enterprises (SMEs) in China. We explicitly distinguish between personal commitments and business collateral/guarantees. Using a unique dataset for SME loans from a regional commercial bank in China, we find that, after controlling for characteristics of borrowers, other terms of loans, and other variables, SME loans with business collateral/guarantees are subject to lower credit risk. Our result is in sharp contrast to evidence from developed countries. Moreover, SME loans with personal commitments of managers/owners are associated with even lower credit risk, after controlling for the impact of business collateral/guarantees. Keywords: Personal commitments; Business collateral/guarantees; Credit risk; SME loans JEL Classification: G21; G32 0

2 Business Commitments, Personal Commitments and Credit Risk: Evidence from China Abstract This paper studies the relationship between collateral/guarantees and credit risk for loans made to small and medium-size enterprises (SMEs) in China. We explicitly distinguish between personal commitments and business collateral/guarantees. Using a unique dataset for SME loans from a regional commercial bank in China, we find that, after controlling for characteristics of borrowers, other terms of loans, and other variables, SME loans with business collateral/guarantees are subject to lower credit risk. Our result is in sharp contrast to evidence from developed countries. Moreover, SME loans with personal commitments of managers/owners are associated with even lower credit risk, after controlling for the impact of business collateral/guarantees. Keywords: Personal commitments; Business collateral/guarantees; Credit risk; SME loans JEL Classification: G21; G32 1

3 1 Introduction This paper analyzes the effects of collateral/guarantees on the probability of default (PD) of bank loans to small and medium-sized enterprises (SMEs). In particular, it examines both business collateral/guarantees and managerial personal commitments both pledges of personal assets and personal guarantees by business owners/managers. Using a unique dataset of loans to SMEs from a regional commercial bank in one province of China for , we are able to provide a direct test on the relationship between collateral/guarantees both business and manager/owner personal commitments and PD, after controlling for the other explanatory variables. Such variables include other terms of loans, characteristics and financial situations of the borrowers, and the macroeconomic environment. We find that loans with collateral and/or guarantees are associated with a lower PD. In particular, personal commitments play a significant role in reducing PD. Our study is the first to study the relationship of personal commitments and credit risk. We also contribute to the understanding of collateral by providing evidence from an emerging economy. Collateral is a prominent feature of debt contracts. It arises either as a solution to compensate for ex ante asymmetric information or as a method to reduce ex post incentive problems. The problems of adverse selection and moral hazard tend to be much more acute for SMEs than for larger firms and it is also generally more difficult for SMEs to obtain bank loans. There have been extensive theoretical contributions on the relationship between collateral and borrower quality. The ex ante asymmetric information models suggests that low quality (high risk) borrowers put up less collateral than high quality (low risk) borrowers (e.g., Stiglitz and Weiss, 1981; Bester, 1985 and 1987; Besanko and Thakor, 1987a, b; Chan and Thakor, 1987; Boot, Thakor and Udell, 1991). Collateral therefore serves as a signalling tool. The theories of ex post frictions suggest that riskier borrowers are more likely to post collateral because of moral hazard concerns (e.g., Boot, Thakor and Udell, 1991, Boot and Thakor, 1994; Aghion and Bolton, 1997; Holmstrom and Tirole, 1997), limited contract enforceability (e.g., Banerjee and Newman, 1993; Albuquerque and Hopenhayn, 2004; Cooley, Marimon, and Quadrini, 2004), and costly state verification (e.g., Townsend, 1979; Gale and Hellwig, 1985; Williamson, 1986; Boyd and Smith, 1994). However, few of these studies clearly distinguish between personal and business collateral. Mann (1997) is one exception. He argues that personal commitments are more effective than business collat- 2

4 eral since a loss of personal assets is taken more seriously than the loss of a business. This is particularly true for small businesses. The empirical evidence on the relationship between credit risk and the use of collateral is rather scant, partly due to data limitations. Most of such evidence measures credit risk using risk premium (Berger and Udell, 1990, 1995; Booth, 1992; Angbazo et al., 1998; Degryse and Van Cayseele, 2000). These studies show that collateralized loans tend to have a higher risk premium. Jiménez and Saurina (2004) is the only study we are aware of that evaluates the relationship between credit risk and the use of collateral using a measure of ex post credit risk. They find that collateralized loans have a higher PD. As Berger and Udell (1990) point out, using ex post data on credit risk has the advantage that ex post credit risk is not affected by the potential problems of collateral-related monitoring costs or other fees. In this study, we also use a measure of ex post credit risk to study the relationship between collateral/guarantees and credit risk. Our main contributions in this paper are based on loan data collected by a regional commercial bank in one province of China. The dataset contains information on loans to SMEs and ex post classifies these loans to five categories: normal, to be watched, substandard, doubtful, and loss, according to the Guidelines on Risk-Based Loan Classification promulgated by the People s Bank of China. Loans that are classified as substandard, doubtful and loss are non-performing loans. Much of the existing empirical literature on credit risk uses datasets that focus on big firms or large operations. Moreover, with few exceptions (such as Berger and Udell, 1990; Jiménez and Saurina, 2004), these datasets usually refer to only one date or, at best, to a short time period. On the contrary, our dataset covers just under 20,000 loans to SMEs with information about borrower characteristics in more than 11 industries for the period 1970 to Although our dataset is based at the transaction or loan level, not at the level of borrowers, none of the borrowers had multiple loans with the bank at the same time. It is possible for one firm to enter another loan after paying off an older loan. About 10 percent of the loans in our sample were entered by the same firms at different points in time. Firms characteristics had changed by the time they borrowed again. Therefore, our analysis of the probability of default is not only at the loan level, but also at the borrower level. We analyze the relationship between collateral/guarantees and credit risk using binomial logit models. We find that business collateral/guarantees is negatively related to PD, after control- 3

5 ling for industries, time fixed effects, and borrower s characteristics such as liquidity, leverage, profitability and coverage. This negative association is robust whether we use loss loans or nonperforming loans. Moreover, personal commitments are negatively associated with PD, in addition to the impact of business collateral/guarantees. Our binomial logit model estimates that the PD drops about one half if business collateral/guarantees (but no personal commitments) are provided when we use loss loans as defaulted loans. When we use non-performing loans, the PD decreases by nearly one third for loans with business collateral/guarantees. When personal commitments from managers/owners are present, the PD further decreases by approximately one half for both loss loans and non-performing loans. The economic significance of personal commitments suggests a potential solution to the ex ante asymmetric information problem in Chinese loan markets as business commitments are common but personal commitments from managers/owners are scarcely used. We also use an ordered logit model to assess the impact of business and personal commitments on PD, taking the gradations of default into account. The results of the ordered logit model are consistent with the binomial logit models. Loans with business collateral/guarantees are subject to lower credit risk, and personal commitments further reduce the PD. The empirical findings in our study are in contrast with existing evidence from developed economies. Studies have shown that in the developed economies (e.g., the US, Spain) collateralized loans are subject to greater credit risk. These studies suggest that the function of collateral as a solution to the ex post incentive problems dominate the role played by collateral as a signal for high quality borrowers. However, our study using loans to Chinese SMEs shows that in an emerging economy like China, the asymmetric information problem for SMEs is so acute that credit enhancement tools such as collateral or guarantees send a strong signal to lenders about the quality of borrowers. Moreover, our results emphasize the importance of the role of personal commitments as a borrower s risk signalling mechanism. Our paper makes a significant contribution to the literature in two respects. First, to our knowledge, we are the first to study the relationship between personal commitments and credit risk. Prior literature on credit risk does not specifically distinguish between business collateral/guarantees and personal commitments. Most of the articles that study personal commitments focus on the determinants of the use of personal commitments (e.g. Ang et al, 1995; Avery et al, 4

6 1998; Voordeckers and Steijvers, 2006) or the relationship between personal commitments and the allocation of small business credit (e.g. Avery et al, 1998). Our paper provides sound evidence on the negative relationship between personal commitments and credit risk using a unique dataset on loans to Chinese SMEs. We emphasize the significant role played by personal commitments as a signalling mechanism of borrowers quality. Our paper is also useful for bank managers and banking regulators. Personal commitments are not as commonly used in SME lending in China as they are in other countries. Only about 3.72% of loans in our sample were protected by personal commitments, while in the US and Belgium, 30-45% of SME loans were (Avery et al, 1998; Voordeckers and Steijvers, 2006). Our study suggests that banks may require personal commitments from managers/owners as an effective screening tool of potential borrowers to decrease credit risk of their loan portfolios. Second, this study deepens our understanding of (business) collateral/guarantees and credit risk. We present strong evidence that for loans to SMEs in China the pledging of collateral or guarantees is associated with lower credit risk. As mentioned above, this is in direct contrast with empirical findings from developed economies where collateralized loans are subject to greater credit risk. As Berger et al (2011) point out, both the ex ante asymmetric information theories and the ex post frictions theories are empirically valid. Our results complement their findings that ex ante theories are valid for customers that are relatively unknown to the lender, since loans in our dataset are for SMEs that are characterized by the lack of financial credibility because of their informal accounting, internal control and governance systems (Berger and Udell, 2006; Firth et al, 2009). The remainder of this paper is organized as follows. Section 2 discusses institutional background. Section 3 describes the dataset we use and econometric models. Section 4 presents and interprets the empirical results and Section 5 concludes. 5

7 2 Institutional Background 2.1 The Bank We first briefly discuss the evolvement of the Chinese banking sector and then describe the regional commercial bank we obtain our data from in the background of the reform of the banking sector. From 1949 to 1978, the People s Bank of China was the only bank in China; it had the dual role of a central bank and credit allocation. All of the bank s branches were part of one administrative hierarchy. Between 1978 and 1984, banking reform measures were taken to modify the structure and operations of the banking system. Four state-owned banks were resumed or established to become specialized banks to service different industries/areas. Before late 1990s, these state-owned banks were a conduit of government policies. In 1994, three specialized policy banks were established to reduce the policy lending burden of the state-owned banks which were given greater scope in raising and allocating capital. However, because of the history of policy lending, these state-owned banks were saddled with extensive portfolios of non-performing loans. The government was forced to inject public funds into these state-owned banks and took a series of reformative steps to transform them into business entities operating on commercial bases. During the reform process, banks other than the state-owned ones emerged with different ownership structure. Now the Chinese banking system consists mainly of state-owned banks, national-level domestic joint-equity banks, city-level commercial banks, urban and rural credit cooperatives, and postal savings banks. The system also includes policy banks, Chinese-foreign joint-equity banks, and foreign banks. We obtain our dataset on loans to SMEs from a regional commercial bank in one province of China. It evolved to an independent commercial bank during the reformative process of the Chinese banking sector. This bank (including branches at the municipal and county levels) was initially part of the administrative hierarchy as local branches of the People s Bank of China. Beginning from 1979, it became local branches of one of the stated-owned banks. In 1996, with the implementation of the Commercial Bank Law, it became independent from the state-owned bank and was managed by the People s Bank of China. In 2005, it registered with the provincial government and became a joint-equity regional commercial bank providing banking and financial 6

8 services. The focus of this bank and its predecessors has been taking deposits and issuing loans in rural areas in one province. By the end of 2010, it had total assets of billion RMB, total deposits of billion RMB, and a loan portfolio of billion RMB. 2.2 SMEs in China Since the launch of the economic reform in 1978, SMEs have played a significant role in employment creation and economic growth in China. Over 99 percent of Chinese enterprises are SMEs. They account for 60 percent of Chinese GDP; 70 percent of employment; 50 percent of tax revenues; 60 percent of exports; and 65 percent of patents filed each year. However, despite this large contribution to the Chinese economy, SMEs only use 20 percent of China s financial resources (Wang, 2007). The lack of appropriate financing channels is the main hurdle faced by SMEs (Fagan and Zhao, 2009). Furthermore, SMEs in China are characterized by the lack of financial credibility (Berger and Udell, 2006; Firth et al, 2009). SMEs generally keep informal accounting books that lack transparency, accuracy, and completeness. It is not uncommon for SMEs to prepare different books for the purposes of tax and borrowing than for themselves (Fagan and Zhang, 2009). As a result, the problem of information asymmetry between SMEs and lenders is particularly severe. Moreover, SMEs are often labor-intensive downstream firms with low profitability and thus are more prone to bankruptcy. Statistics show that the average life of the 40 million registered SMEs is about 2.9 years 1. Hence, it is crucial for lenders to find signals or indicators to infer the quality of the potential borrowers. This study provides sound evidence that owners/managers of SMEs who are willing to pledge their personal assets or provide personal guarantees can serve as a credible signal. 1 Source: (in Chinese) Last accessed on June 21,

9 3 Data and Empirical Specifications 3.1 Data Our data comes from retrospective verifications of classifications of loans done in 2007, 2008 and 2009 by a regional commercial bank in one province of China. The bank s headquarters administered these verifications in 11 prefectural branches on their loans. The verifications confirm the following: basic loan information, collateral, guarantees, classification of loans, loan-repayment record, organizational type and basic borrower information, and borrower financial information at the time applying for a loan. Note that the focus of this commercial bank has been on the rural areas so that its loans were made to SMEs in these rural areas. Each loan in our sample is classified into five categories: normal, to be watched, substandard, doubtful, and loss, according to the Guidelines on Risk-Based Loan Classification promulgated by the People s Bank of China. A loan is classified as normal if the obligor services the loan promptly and there is not enough reason to doubt the likelihood of repayment of either the interest or the principal. A loan is classified as to be watched if the obligor is able to service the loan now but circumstances have arisen that could negatively affect the repayment ability in the future. A loan is substandard if the obligor is not able to service the loan using income from its regular business operations. In this case, a loss is possible even with the enforcement of security or guarantees. A loan or part thereof is doubtful if it is obvious that the obligor will not be able to service the loan. A substantial loss is probable even with the enforcement of security or guarantees. A loan or portion thereof is loss if it is uncollectible or only a small portion is collectible after taking all possible measures to recover the loss. The difference among the verifications done in different years is that each loan is classified again in a later year using information available then and the years these loans were rolled over increase. For example, a loan that was classified as normal in the 2007 verifications became to be watched in the verifications done in 2008 and 2009, and the time period this loan had been rolled over increased by one year in the 2008 verifications and two years in the 2009 verifications. All the other variables, such as other terms of loans and borrower characteristics (except firm age), are the same for the verifications done in different years. Since the verifications done in 2009 are 8

10 more comprehensive and up to date, we use data from that verifications in our main analysis. We will use information on classifications of loans from the two earlier verifications as a robustness check. Our original dataset, the verifications of loans done in 2009, contains 21,496 loans issued between 1970 and Of these loans, we delete 1609 observations due to missing values in firm age (1,130 observations), loan age (440 observations), interest rate of the loans (6 observations), registered capital (59 observations), and information on collateral/guarantees (1 observation). Our sample with complete non-financial information consists of 19,887 observations. The total amount of loans in this sample is billion RMB. Among these, the amount of short-term loans is billion RMB, billion RMB is mid-term loans, and 2.72 billion RMB is longterm loans. For some of our analysis presented later in this article, we require financial information of borrowers. Therefore, we further clean the data by deleting observations with missing or erroneous values of financial ratio variables. We have 4,235 missing or erroneous values for the ratio of tangible assets to total assets. We also have 4,234 observations with missing or erroneous values in the ratios of current liabilities to total liabilities, current liabilities to total assets, net income to net assets, or other financial ratios. We end up with 14,244 observations with complete non-financial and financial information. 3.2 Sample Descriptions Table 1 provides the frequency of loan classifications and of other loan characteristics. From Table 1, we can see that approximately 38% of the loans are normal or to be watched, but loss loans account for just under 22% of the sample, non-performing loans account for 61% of the sample. This is in sharp contrast to default rates in developed countries. For example, the default rate (percentage of doubtful loans) in Spain is about 4.51% for all loans granted by Spanish credit institutions for a value of over 6,000 euros (Jiménez and Saurina, 2004). The default rate in our sample is much higher. It should be borne in mind that the non-performing loans are extensive for Chinese banks. Official statistics show that non-performing loans as a share of the state banks total loans was 20% to 35% during the period , while western observers 9

11 generally estimated the non-performing ratio to be 40% to 50% of outstanding loans (Firth et al, 2009). The non-performing ratio in our sample is even larger. Part of the reason is that the regional bank has kept information on non-performing loans from early years in record but deleted information on loans paid off promptly. Therefore the non-performing ratio is very high in our sample, especially in early years. We note that in our sample, the majority of loss loans are smaller loans. Of the loans with losses, 88% of them are small loans (with the size of loan smaller than median) and only 12% of them are of larger size. If we calculate the percentage of loss loans and non-performing loans using the RMB value of loans, then the ratios are 2.61% and 12.01%, respectively. If we consider the inflation during the years , these two ratios are 3.65% and 15.24%. These ratios based on the RMB value of loans suggest that bad loans in early years account for a large portion of the 61% non-performing loans since early loans were significantly smaller. We can also observe from Table 1 that business guarantees are widely used. About 77% of loans have business guarantees. The pledging of business collateral is not as common as guarantees with only approximately 13.3% of the loans being collateralized. Personal commitments from managers/owners are far less common. Only about 3.72% of the loans have personal commitments from managers/owners. Note that the profile of the use of credit enhancement in our sample is very different from other countries. For example, in Belgium in , 12.39% of loans to SMEs had no collateral protection, 57.26% had business collateral, and 30.34% had personal commitments (Voordeckers and Steijvers, 2006). In the US in 1993, just under 10% of the loans to SMEs were unguaranteed or unsecured, nearly 46% were collateralized, and approximately 45% were protected by personal commitments (Avery et al, 1998). In our sample, the use of business guarantees are far more prevalent while business collateral and personal commitments are much less common in loans to SMEs in China. SMEs have a lack of financial credibility that stems from limited transparency, as well as a lack of fixed asset collateral such as real estate and equipment. As such, more than 90% of the loans are short-term loans and nearly 80% of all the loans are for the financing of working capital. Ironically, these loans may be continually rolled over and thus become de facto long-term loans. In our sample, the average age of loans is 8.38 years, which indicates that rolling over of loans was a common practice. About 45% of borrowers in our sample are corporations, nearly 7% 10

12 are partnerships, and the rest are classified as Other. Our sample includes loans issued in the 1970s at which time there were only state-owned and collective enterprises. All of these older loans are classified as Other. Our sample loans are relatively concentrated on: manufacturing; wholesale and retailing; mining; and agriculture, forestry, animal husbandry and fishery. These four industries account for over 70% of the loans. Table 1 also reports the time periods in which loans were issued. From , there was high inflation. Beginning in 1996, there was a period of economic contraction during which workers were laid off from state-owned enterprises; this lasted until In the period of , the Chinese economy gradually recovered from said contraction. During this time, banks issued fewer loans due to increased accountability of loan officers was a transition period from recovery to the overheated economy in Loans in our sample are roughly equally distributed in the first five time periods with more loans (about 25% of the sample) being issued in the last period. Insert Table 1 here. Table 2 presents detailed definitions of the variables used in the regressions and Table 3 shows the summary statistics for these variables, excluding dummy variables. The average size of loans is about 2,958,200 RMB, the average interest rate charged on the loans is 11.79% with the minimum being 2.88% and maximum 32.4%, and the average time period these loans were rolled over is just over 8 years. Our sample SMEs have a longer average life (8.38 years) than the average life of all registered SMEs in China (2.9 years). This is reasonable since not all SMEs are able to obtain bank credit. More established SMEs are more likely to receive bank loans. Insert Table 2 here. Insert Table 3 here. Table 4 compares the percentages of loss loans for loans without any credit enhancement and collateralized loans. The percentage of loss loans for collateralized loans is 17.62%, which is about 7% lower than that for loans without credit enhancement. Further, for collateralized loans, 11

13 those with personal commitments from managers/owners have 6.94% loss loans while those without personal commitments have 18.21% loss loans. From these univariate analyses, we can see preliminarily that business collateral can be an effective way of reducing loss loans. Moreover, personal commitments from managers/owners provide a strong incentive for borrowers to repay their loans so that loss loans dramatically decrease. Insert Table 4 here. 3.3 Empirical Specifications Our central variables are loan classifications, business collateral/guarantees, and personal commitments. To test the impact of business collateral/guarantees and personal commitments on PD, we run regressions to examine the sign and significance of the coefficients of business commitments and personal commitments. As mentioned before, there is considerable debate on the impact of collateral on credit risk. The two alternative interpretations, ex ante asymmetric information theories and ex post frictions theories, empirically predict opposite relationship between collateral and credit risk. The ex ante asymmetric information theories (e.g., Stiglitz and Weiss, 1981; Bester, 1985 and 1987; Besanko and Thakor, 1987a, b; Chan and Thakor, 1987; Boot, Thakor and Udell, 1991) predict that unobservably safer borrowers are more likely to pledge collateral. Specifically, the use of collateral is an attempt to compensate for ex ante information gaps between borrowers and lenders that lead to adverse selection and credit rationing in the credit market. Collateral can serve as a signal for observationally equivalent borrowers with higher quality projects. Banks design loan contacts such that higher quality borrowers choose loans with collateral and lower risk premiums while lower quality borrower self-select into unsecured loans with higher risk premium. Therefore the use of collateral is associated with lower credit risk. The theories of ex post frictions predict that riskier borrowers are more likely to post collateral. The reasons for this positive relationship include: 1. moral hazard concerns that borrowers make less effort to ensure the success of the project for which finance was given (e.g., Boot, Thakor and Udell, 1991, Boot and Thakor, 1994; Aghion and Bolton, 1997; Holmstrom and Ti- 12

14 role, 1997); 2. limited contract enforceability as the borrower can choose to default (e.g., Banerjee and Newman, 1993; Albuquerque and Hopenhayn, 2004; Cooley, Marimon, and Quadrini, 2004); and 3. costly state verification that actual returns on any project can be observed by anyone other than the owner only at some cost (e.g., Townsend, 1979; Gale and Hellwig, 1985; Williamson, 1986; Boyd and Smith, 1994). These frictions lead to a positive relationship between collateral and credit risk. Berger et al (2011) point out that the ex ante and ex post theories are not mutually exclusive. We are striving to determine which of the theories empirically dominates in SMEs loans in China. In our study, significant and negative signs of the coefficients of business collateral/guarantees and personal commitments indicate that ex ante signalling theories dominate. By contrast, positive signs on these coefficients indicate the dominance of ex post frictions theories Logit Models Our empirical tests relate the incidence of default to the use of business collateral/guarantees, personal commitments, other characteristics of the loan, characteristics of borrowers, and industry and time fixed effects. This model can be summarized as P(Default it )=F(Business Collateral/Guarantees it, Personal Commitments it, Other Loan Characteristics it, Borrower Characteristics it, Industry i, γ t ) (1) where P( ) indicates the probability of default and i, t denote the loan and time period (year). The function F( ) is the logistic distribution function (as in Jiménez and Saurina, 2004) and the variable Default it is a dummy that equals one if default occurs. We use two measures of default. The first measure is to use loss loans, Default it = 1 if a loan is classified as a loss loan and zero otherwise. The second measure is to use non-performing loans, Default it = 1 if a loan is classified as loss, doubtful or substandard loans. The variables that we are most interested in are the use of business collateral/guarantees and personal commitments. As discussed, significant and negative signs of the coefficients of these variables indicate the dominance of ex ante theories while positive signs indicate the dominance of ex post theories. 13

15 To control for other potential factors that may also affect credit risk, we include other loan characteristics and borrower characteristics. Other characteristics of loans include the size of a loan (LNSIZE), interest rate charged (INT EREST ), maturity (SHORT), and rollover period (LNAGE). The size of a loan may affect a borrower s credit risk. Larger loans are usually issued to larger companies while smaller loans to smaller companies. Other things being equal, smaller companies have more operational risk than larger ones. Moreover, larger loans are subject to more stringent screening and monitoring. Hence we expect a negative sign for the loan size (LNSIZE). In a fully market-oriented banking system, lending rates reflects credit risk, with higher rates for greater credit risk and lower rates for lower credit risk. However, commercial banks in China did not have the freedom to choose lending rates until recent years. During the time period for our sample, the lending rates in China were set by the People s Bank of China. Hence, lending rates in our sample fluctuated with economic cycles, with higher rates in booms and lower rates in recession (or slower growth). We expect a negative sign for lending rates (INT EREST ) since credit risk is lower in booms with lending rates being higher. For the maturity of a loan, we do not have specific information on the length of loans. However, each loan is classified as shortterm, mid-term, or long-term. We include a dummy variable (SHORT) to indicate short-term loans. We expect short-term loans to have a lower PD. This is because longer-term loans take a longer time to materialize. Moreover, given the rapidly changing government policies and the severe asymmetric information problem between borrowers and lenders, longer-term loans are especially difficult for lenders to assess and thus are more risky than short-term loans. Another characteristics of loans in China is that loans can be continually rolled over and become de facto longer-term loans. The longer a loan is rolled over, the higher the credit risk. Therefore we expect a positive sign for LNAGE. We also include some characteristics of borrowers such as firm age and size, organization type, profitability, liquidity, turnover, capital structure, and debt structure. Firm age (FMAGE) may affect default as older firms are more established with their regular clientele and more experienced management team while younger firms may be relatively weaker. Hence the older a firm, the lower the credit risk. We use registered capital (CAPITAL) as a proxy for firm size, due to data availability. In general, larger firms are required to have more registered capital. Larger firms are less likely to default as they are more capable of weathering economic downturns. 14

16 Organization type (CORPORAT E) affects credit risk because owners/managers of corporations who have limited liability are more likely to default when in financial difficulty than partnerships or sole proprietorships with unlimited liability. Profitability is measured by EBITDA margin (EBITDAMG) and return on net assets (RONA). To measure a firm s ability to meet its debt obligations, we use quick ratio (QUICK) and interest coverage ratio (ICR). We also use turnover ratios such as average collection period of account receivables (ACPR) and inventory turnover ratio (V OI) as additional measures of liquidity. Leverage (DEBT ) is included to measure the overall indebtedness of the firm. Because financial difficulties are often caused by the inability of the borrower to repay short-term debt, we include the current to total liabilities ratio (CLT L) to measure the debt structure. We also include the ratio of current assets to total liabilities (CAT L) to see how current assets can cover liabilities. The ratio of tangible assets to total assets (TANG) is included as lenders face less credit risk for firms with more tangible assets. Moreover, we include the operating profit to net income ratio (OPNI) as a measure of how much of a firm s profit comes from its operations. In equation (1), Industry i and γ t represent industry and time (period) fixed effects, respectively. Our sample firms belong to 12 industries (see Table 1). Some of the industries are highly cyclical while others are not as cyclical, therefore we include industry fixed effects. And, because our sample spans a long time period, we include dummies to account for different time periods (see Table 1) in order to reflect time (period) fixed effects Ordered Logit Model We use the ordered logit regression model to further assess the impact of business and personal commitments on credit risk. The dependent variable CLASSIFICAT ION is treated as an ordered variable with five ordinal categories: normal, to be watched, substandard, doubtful, and loss. It takes values from 0 and 4 to account for gradations of default. For this type of data the ordered logit model is a suitable tool of analysis. The ordered logit model is based on the following specification: y i = β x i + ε i 15

17 where x i is the set of explanatory variables and ε i is the disturbance term that has a standard logistic distribution. As usual y i is unobserved. What we do observe is y i = 0 if y i µ 0 = 1 if µ 0 < y i µ 1 = 2 if µ 1 < y i µ 2 = 4 if y i > µ 3. The µ s are unknown parameters to be estimated together with β. Each loan has its own probability of default, which depends on certain measurable factors, x, and certain unobservable factors, ε. However, loan officers classify each loan into five categories that most closely represents each loan s PD. The conditional probability of each category is given by the following: Pr(y i = 0 x i,β,µ)=f(µ 0 β x i ); Pr(y i = 1 x i,β,µ)=f(µ 1 β x i ) F(µ 0 β x i ); Pr(y i = 4 x i,β,µ)=1 F(µ 3 β x i ). We use the same set of explanatory variables as in the binomial logit models. Again, our focus will be on the significance and sign of the coefficients of business collateral/guarantees and personal commitments. 4 Results Next we will discuss the estimation results for the binomial logit and ordered logit models. Table 5 shows the results of the maximum likelihood estimation of the binomial logit model using LOSS as the dependent variable for the verifications done in The sample size in model (2) is significantly smaller than in model (1) since about one quarter of our sample borrowers did not provide their financial information. The sample size in model (3) decreases further because we remove the observations that have financial ratios in the top 5% and bottom 5%. This is because 16

18 SMEs in China generally keep informal accounting books that are often inaccurate. As a result, the financial information they provide may not be reliable. We try to mitigate the impact of extreme values of financial ratios on our results by winsorizing the variables of financial ratios at 5% and 95%. In total we lose about 15% of the observations after winsorization. From Table 5, we can observe that most loan characteristics and borrower information are statistically significant at the 1% significance level. The inclusion of financial ratios decreases the significance of INT EREST, though only a few of the financial ratios are statistically significant. Moreover, winsorizing the variables of financial ratios at 5% and 95% does not qualitatively change our results. Insert Table 5 here. For collateral/guarantees, the coefficients of both business commitments and personal commitments are all negative and highly significant. The pledging of business collateral or guarantees decreases the probability of loss loans compared to loans without any business credit enhancement. Moreover, loans with personal commitments are associated with a lower probability of loss loans, after the impact of business commitments has been controlled. Our results show the empirical dominance of ex ante asymmetric information theories. In the loan market to SMEs in China, both ex ante asymmetric information problems and ex post frictions problems exist. Our results indicate that lenders are mainly concerned about the information that they do not know about, but which their borrowers do know. The asymmetric problem is particularly severe, given that SMEs lack financial creditability and transparency (Berger and Udell, 2006; Firth et al, 2009). Therefore, the signalling mechanism of collateral/guarantees is especially important to address the adverse selection problem that leads to credit rationing. Our results are in direct contrast to evidence from developed countries which indicates collateralized loans are subject to greater credit risk. In these countries the ex post frictions theories seem to empirically dominate the ex ante asymmetric information theories (e.g., Berger and Udell, 1990; Jiménez and Saurina, 2004). The impact of collateral/guarantees on PD is also significant in economic terms. We can calculate the marginal probability of loan loss using model (1) in Table 5. For instance, the prob- 17

19 ability of loan loss is 11.11% for a long-term loan issued to a median corporation in the industry of electronics and telecommunication in without any credit enhancement. When business collateral/guarantees are provided for this loan but no personal commitments from managers, the probability of loss decreases by 5.39% to 5.72%, which is an approximately one-half decrease. When personal commitments from managers are provided but no business commitments, the probability of loss decreases by 4.83% to 6.28%, which is also close to one-half decrease. Moreover, when both business commitments and personal commitments are present, the probability of loan loss drops by 7.96% to 3.15%. The economic significance of personal commitments suggests the importance of personal commitments from managers/owners as a solution to the severe asymmetric information problem because while business commitments are prevalent (87.49% of loans have at least one form of business commitments) personal commitments are far less common (only 3.72% of loans have personal commitments). We briefly examine the impact on PD of the remaining loan characteristics as well as borrower characteristics. All of the signs of the remaining loan characteristics are as expected and most of them are statistically significant at the 1% level. Larger loans are associated with a lower PD since larger loans are subject to more stringent screening and monitoring and are also applied by larger SMEs, which are often less risky. The coefficient of INT EREST is negative, as expected, but the statistical significance is reduced when financial ratios are included. In regards to maturity, short-term loans are associated with a lower PD. This is not surprising given the risk associated with the severe asymmetric information problem among SMEs in China and that the longer-term loans take a longer time to materialize and thus making them riskier. The PD is higher for loans with longer roll over periods, which is also expected. The longer a borrower rolls over a loan, the less likely the borrower s ability to pay it off becomes, which is also indicative of greater credit risk. For non-financial characteristics of borrowers, the coefficients of firm age and registered capital are as expected and statistically significant. Default rates for older SMEs are lower since older firms are more established and have less operational risk. Firms with more registered capital are larger and the credit risk is lower for larger firms. The coefficient for organization type CORPORAT E is negative and highly significant, which is opposite to our expectation. The reason could be that in our sample corporations on average are three times of the size of other organiza- 18

20 tion types. The variable CORPORAT E acts more as a proxy for firm size. Since larger firms are subject to lower credit risk, we observe that corporations are associated with a lower PD. We also use non-performing loans instead of loss loans as defaulted loans in the binomial logit models. Table 6 reports the maximum likelihood estimation results. Again three models are estimated in the same way as in Table 5. We can see from Table 6 that the results are similar to that using loss loans. The coefficients of both business commitments and personal commitments are negative and highly significant, suggesting the dominance of ex ante asymmetric information theories. The effects of business commitments and personal commitments on reducing non-performing loans are also economically significant. For instance, the probability of becoming non-performing for a long-term loan issued to a median corporation in the industry of textile and clothing during the period of with neither business commitments nor personal commitments decreases from 49.42% by about 13.78% to 35.63% when business collateral/guarantees but not personal commitments are provided. If personal commitments alone are present, the probability of non-performing for the above loan decreases by approximately 20.34% to 29.08%. When both business and personal commitments are present, the probability of non-performing decreases by 30.56% to 18.85%. Again, the economic significance of personal commitments highlight the importance of personal commitments from managers/owners as a potential solution to the ex ante information asymmetry problem in the Chinese loan market since personal commitments are scarce compared to business commitments. The estimation results for borrower characteristics and other terms of loans are qualitatively similar to that using loss loans. Insert Table 6 here. One would think that personal commitments of managers/owners only affect firms whose owners have limited liability, but have no impact on other types of firms. For example, owners of sole proprietorships and general partnerships are personally liable for their business debts, meaning that personal commitments would be redundant. By contrast, for a corporate shareholder who has limited liability, personal commitments clearly put her/his personal assets at risk, especially in financial distress. To check if the differential effects of personal commitments exist in our sample, we include an interaction term PERCOM CORPORAT E in the binomial logit model 19

21 using non-performing loans. Table 7 presents the estimation results. The coefficient of the interaction term is negative and statistically significant at the 5% level, indicating that loans issued to corporations that have personal commitments have a lower PD. However, the coefficient of personal commitments is not significant any more, though still negative, suggesting that personal commitments do not affect the PD of other types of firms. Insert Table 7 here. Table 8 presents the maximum likelihood estimation results for the ordered logit model with and without financial ratios of borrowers. The dependent variable is CLASSIFICAT ION which is an ordinal variable taking values of 0, 1 4 for normal, to be watched, substandard, doubtful, and loss loans. These results are consistent with those in the binomial logit models. The coefficient of business commitments is negative and statistically significant at the 1% level. The use of business commitments is associated with a lower probability of downgrading to a lower credit classification. Moreover, the coefficient of personal commitments is negative and significant at the 5% level when financial ratios are not included and at the 1% level when financial ratios are included. The estimation results for other variables are qualitatively similar to those in binomial logit models. Insert Table 8 here. To check if our results are biased by earlier observations in 1970s and 1980s, we also conduct the analysis for later years in our sample. The observations from earlier years are biased towards bad loans for a few reasons. In the older days, the bank tended to remove records of loans that were repaid in full or from firms it no longer had business with. As a result, the loans left on the record of the bank from earlier years tend to be ones that were not repaid in full on time or from firms that continued to borrow from the bank. Moreover, in earlier years, the bank issued more loans without any credit enhancement, but in later years, guarantees/collateral became much more common. Given these structural changes, we re-estimate the binomial logit model for the period of We choose to start the robustness check from 1998 because the People s Bank of China changed loan policies of commercial banks fundamentally by removing credit quotas. 20

22 At the same time, loan officers in commercial banks became much more accountable for their loan decisions. They may hold criminal liability for intentional wrongdoing while they were not subject to any sever penalty before Moreover, the period covers exactly one economic cycle with being a slow-growth cycle and being a rapid-growth cycle. Table 9 reports the results. We find that excluding earlier periods does not qualitatively affect our results. The coefficients of business commitments and personal commitments for models using non-performing loans as the dependent variable are negative and highly significant, indicating that both forms of credit enhancement decrease the probability of non-performing loans. For models using loss loans as the dependent variable, the coefficients of personal commitments are negative and highly significant, though the significance level of business commitments is lower (at 10% level) than that using the entire sample. Insert Table 9 here. Finally, we re-estimate the binomial logit models using the verifications of loan classifications done in 2007 and Tables 10 and 11 present the estimation results using loss loans and nonperforming loans, respectively. Financial ratios are included in Models (3) and (4) in both tables and we winsorize the variables of financial ratios at 5% and 95% in Models (3) and (4) in Table 10 and Model (4) in Table 11, but not in Model (3) in Table 11. The estimation results are very similar to that using verifications of loan classifications done in The coefficients of business collateral/guarantees are negative and statistically significant at the 1% level in both tables. The coefficients of personal commitments are also negative and statistically significant, though the significance level is lower in some models. Our main conclusions remain intact: loans with business collateral or guarantees are subject to lower credit risk, while personal commitments provide additional protection for creditors. Insert Table 10 here. Insert Table 11 here. 21

23 5 Conclusions Existing empirical evidence from developed countries suggests that collateralized loans are subject to greater credit risk, indicating that ex ante asymmetric information theories dominate the ex post contract frictions theories. In this paper we provide opposite evidence, using a unique dataset on about 20,000 loans to SMEs in from a regional commercial bank in China, that loans with collateral or guarantees are associated with a lower probability of default. In an emerging economy like China, severe asymmetric information problem exists between borrowers and lenders. This problem is even more acute for SMEs. As a result, borrowers with high quality projects use collateral/guarantees to signal their quality to banks. Our results complement the findings of Berger et al (2011) that ex ante theories hold for customers who are relatively unknown to the lender. More importantly, in our study, we explicitly distinguish between business collateral/guarantees and personal commitments. We present strong evidence that loans with personal commitments are subject to significantly lower credit risk in addition to the impact of business collateral/guarantees. Given the rare use of personal commitments in SME lending in China (personal commitments are provided to only 3.72% of loans in our sample), our results may be useful for bank managers and supervisors who seek to manage the risk of their loan portfolios. This is particularly relevant in the current environment in which the Chinese government encourages commercial banks to support the development of SMEs. References Aghion, P. and P. Bolton, (1997), A theory of trickle-down growth and development, Review of Economic Studies, 64, Ang, J.S., J. Lin and F. Tyler, (1995), Evidence on the lack of separation between business and personal risks among small businesses, Journal of Small Business Finance, 4, Angbazo, L.A., J. Mei, and A. Saunders, (1998), Credit spreads in the market for highly leveraged transaction loans, Journal of Banking and Finance, 22, Avery, R.B., R. Bostic and K. Samolyk, (1998), The role of personal wealth in small business finance, Journal of Banking and Finance, 22,