The effect of credit guarantees on credit availability and delinquency rates

Journal of Banking & Finance 59 (2015) 98–110

Contents lists available at ScienceDirect

Journal of Banking & Finance journal homepage: www.elsevier.com/locate/jbf

The effect of credit guarantees on credit availability and delinquency rates q Kevin Cowan a, Alejandro Drexler b,⇑, Álvaro Yañez c a

Inter-American Development Bank, 1300 New York Avenue NW, Washington, D.C., United States Federal Reserve Bank of Chicago, 230 South Lasalle Street, Chicago, IL 60604, United States c Superintendencia de Bancos e Instituciones Financieras Chile, Moneda 1123, Santiago, Chile b

a r t i c l e

i n f o

Article history: Received 8 April 2013 Accepted 27 April 2015 Available online 16 June 2015 JEL classification: G2 G21 G22 G28 G32 G38

a b s t r a c t We use new data to examine whether credit guarantees affect economic incentives and whether they affect the credit available to small- and medium-size enterprises (SMEs). We find that firms that have both guaranteed and non-guaranteed loans are 1.67% more likely to miss payments on their guaranteed loans, but are not more likely to default on these loans. These findings suggest that guarantees affect firms’ incentives to repay loans but not their long-term performance. We also find that firms selected into the guarantee programs are 1.17% more likely to default on their loans compared with similar firms that borrow without guarantees. Since we find evidence that long-term performance is not affected by guarantees, the higher default rates among firms selected into the guarantee programs must be the consequence of adverse selection. We also find that credit guarantees increase the aggregated amount of credit; in particular, one additional dollar of guarantees increases the total credit for SMEs by US$ 0.65. Ó 2015 Published by Elsevier B.V.

Keywords: Banking Financial intermediation Guarantees Credit constraints

1. Introduction A consistent finding in empirical economics is that small- and medium-sized enterprises (SMEs) experience stronger and more costly financial impediments in investing than large firms (Evans and Jovanovic, 1989; Beck et al., 2005, 2007). Substantial effort q This paper greatly benefited from the analysis of partial credit guarantees prepared by the authors for the Partial Credit Guarantee Conference organized by the World Bank in 2008, as well as the analysis of the effect of PCGs in the Chilean economy that the authors conducted in cooperation with the Chilean Central Bank, the Chilean Bank Regulation Office, and the PCG administration. We are grateful for the comments and suggestions of Antoinette Schoar, Roberto Rigobon, Daniel Paravisini, Andre Guettler, Alessandro Bozzo, Javier Torres, Ricardo Villarroel, Erik Feijen, Thorsten Beck, Patrick Honohan, Jiro Kondo, and Luis Opazo. We also thank the participants in the Partial Credit Guarantees Conference at the World Bank and the participants in the seminars at the Inter-American Development Bank and Banco Central de Chile. Furthermore we want to thank Kate Gordon, Niloufar Rohani, Mary Fischer, and Han Choi for precious help in writing the paper. All remaining errors are our own. The views expressed here do not reflect official positions of the Federal Reserve Bank of Chicago, the Inter-American Development Bank or the Superintendencia de Bancos e Instituciones Financieras de Chile. ⇑ Corresponding author. Tel.: +1 (312) 322 5343. E-mail address: [email protected] (A. Drexler).

http://dx.doi.org/10.1016/j.jbankfin.2015.04.024 0378-4266/Ó 2015 Published by Elsevier B.V.

has been exerted by governments and multilateral organizations to reduce these obstacles.1 However, the success of interventions to increase SME access to financing has been mixed at best (Jaramillo-Vallejo et al., 1993), and governmental interventions in particular have not been cost effective (Khwaja and Mian, 2005; Zia, 2008). In light of this evidence, many governments have adopted a more passive role and have delegated the administration of interventions to private institutions with more experience in the credit markets. A prominent example of this tendency is the partial credit guarantee (PCG), under which the government offers funds to guarantee the repayment of loans issued to SMEs but private institutions can freely choose which borrowers receive guaranteed loans.2 Lately, PCGs have become one of the most widely used strategies to improve SMEs’ access to credit. Green (2003) reports 1 For example, the Inter-American Development Bank estimates that their interventions to reduce financial market deficiencies in Latin America and the Caribbean between 1990 and 2004 account for US$ 22 billion. 2 In some regions guarantee funds are also provided by non-governmental institutions. In practice PCGs work like credit insurance. In the text we will use the terms insurance and guarantees indistinctly.

K. Cowan et al. / Journal of Banking & Finance 59 (2015) 98–110

that almost one hundred countries have some form of PCGs, and in the U.S. alone, PCGs support US$ 62.5 billion in loans to SMEs. The size of PCGs varies largely across other countries. For example, in Chile, PCGs represent 1% of the gross domestic product (GDP), whereas they represent 9% of the GDP in Korea (Beck et al., 2010). Despite the popularity of this type of intervention, there is still an intense debate among scholars and practitioners about its potential effects on economic incentives.3 Advocates argue that PCGs reduce collateral requirements, increasing the access to financing of some low-asset SMEs that have credit constraints despite having profitable investment opportunities. Furthermore, they argue that accessing the credit market helps these borrowers to build a credit score that might let them borrow without guarantees in the future. Detractors claim that profitable firms can afford competitive interest rates, and blame PCG for reducing market discipline, facilitating access to credit to low-quality firms and creating moral hazard and adverse selection problems (Kuniyoshi and Tsuruta, 2014; Gropp et al., 2013). There are other criticisms of PCGs that point out the high cost structure of credit guarantee programs; however, in this paper we center the analysis on the economic distortions associated with PCGs.4 Detractors base their views on the proliferation of risky loans made to poorly performing firms selected for guarantee programs in Europe, Asia, Africa and Latin America. Extreme cases have been reported in Nigeria, Malaysia and Indonesia, where the default rates on guaranteed loans are 12%, 34% and 50%, respectively (Gudger, 1998). While these rates are certainly above the average default rate on non-guaranteed loans, they do not necessarily prove guarantees are associated with economic distortions. Even if guarantees were proven to be the cause of increased default, it would be necessary to understand the mechanisms of these economic distortions and how these economic distortions affect repayment in order to re-design these interventions. To date, there are no studies that address these issues, largely because of data limitations. The main contribution of our paper is to shed light on this mostly unexplored dimension of credit guarantees. We study the operations of the PCG program in Chile between 2003 and 2006. What makes these PCGs special is that enterprises can borrow from multiple sources and can maintain insured and uninsured obligations with each of them. Moreover, we observe the repayment behavior for each of these obligations separately. These features allow us to study the effects of PCGs at the firm level by including in our specification a rich set of fixed effects that control for bank and borrower characteristics. Furthermore, we can control for time-varying characteristics of the relationship between the borrower and the bank. This is a major contribution to the empirical banking literature that has mainly focused on time invariant borrower–bank fixed effects. Our main analysis thus tests whether the same firm borrowing from the same bank shows a different repayment behavior on its insured loans compared with its repayment behavior on uninsured loans; in the paper, we refer to this approach as the ‘‘within bank-borrower’’ estimation. This specification requires information at the loan level, which we only have for credit outstanding and amount delinquent, but which we don’t have for the amount in default.5 Therefore, the analysis of default

3 For a description of theoretical implications of PCGs, see Innes (1991), Chaney and Thakor (1985), and Gale (1990); and for a description of the most important discussions among practitioners, see Gudger (1998) and Mhlanga et al. (2013). 4 In many countries PCG administration costs outweigh by several times the fees charged to obtain guarantees. This problem is particularly relevant in some European countries where PCG administration costs are close to 15% of the guaranteed funds, and fees range between 2% and 3% of the guaranteed funds (Bannock, 2005). 5 We observe credit outstanding and amount delinquent (defined as amount in arrears between 61 and 90 days) separately for each loan that borrower i maintains with bank b; however, we only observe the consolidated amount in default (defined as amount in arrears of more than 90 days) that borrower i maintains with bank b.

99

relies on a ‘‘within borrower’’ specification for which we include borrower-time fixed effects but exclude bank–borrower-time fixed effects. We find the delinquency rates of obligations with credit guarantees to be 1.02% higher over the first twelve months and 1.67% higher over the first twenty-four months than those similar uninsured obligations.6 While delinquency rates are higher when guarantees are present, outright default is not affected by them, suggesting that guarantees deteriorate the borrowers’ incentives to repay loans but not necessarily their managerial effort and their long-term performance.7 We also find interesting heterogeneous treatment effects; the repayment behavior of firms with relatively high assets is not affected by the presence of insurance, which is consistent with the finding in Berger et al. (2011b) in that collateral has an important role in disciplining borrowers after loans are issued. To understand how firms are selected into the guarantee programs, we next estimate a specification without firm fixed effects. This approach measures differences between firms selected into the guarantee programs and firms borrowing without guarantees; in the paper, we refer to this approach as the ‘‘between firms’’ estimation. A caveat of the between estimation is that findings can be explained by differences in firms’ characteristics (adverse selection) or by changes in firms’ behavior associated with the use of guarantees (moral hazard). However, by comparing the between estimation with the within estimation, we are able to disentangle these two alternative explanations. We find that firms selected into the guarantee programs are 1.44% more likely to become delinquent on their loan payments within 36 months compared with firms borrowing without guarantees. Furthermore, firms selected into the guarantee programs are also 1.17% more likely to enter into default within 36 months compared with firms borrowing without guarantees. Since we know from the within estimation that guarantees don’t affect firms’ long-term performance, the higher default rates among firms selected into the guarantee programs must be the consequence of adverse selection. We also find interesting heterogeneous treatment effects; among firms with high assets, the presence of guarantees is not associated with higher default rates. We think that the risk of losing assets deters high-asset firms from pursuing low-quality projects, even when they have access to guarantees. In contrast, low-asset firms, which have ‘‘nothing to lose,’’ are willing to pursue low-quality projects and try to get lucky. These inefficiencies can reduce the GDP by 0.04% and destroy 0.1% of jobs each year. Another interesting feature of the PCGs in Chile is that guarantees are allocated through an auction with sealed bids. Therefore, the amount of guarantees allocated to a bank depends not only on its demand for the guarantees, but also on the bids of other participants in the auction. This feature generates nonlinear variation in the amount of guarantees allocated to each financial institution. In the paper, this nonlinear variation is used to identify the effect of PCGs on the aggregate lending to SMEs. While similar approaches have been used by other researchers to test the effect of other types of government interventions, we are the first to use nonlinear variation to study the effect of partial credit guarantee programs.8 We find that PCGs are effective in increasing the aggregated amount of credit available to SMEs. In particular, an increase of one dollar in the guarantees available to a bank is 6 These are the point estimates in the within bank–borrower specification, but similar results are obtained in the within borrower specification. 7 An implicit assumption in our interpretation of this result is that, to some extent, projects are bank specific; i.e., it is costly or not possible for borrower i to pay bank b1 with the revenues of a project financed by bank b2 . We are also assuming that poor long-term performance on the loan is caused by poor long-term performance on the project. 8 For example, Paravisini (2008) uses nonlinear variation to study the effect of a direct subsidy to small businesses lending in Argentina.

100

K. Cowan et al. / Journal of Banking & Finance 59 (2015) 98–110

associated with an increase of US$ 0.65 in credit to SMEs; more specifically, the amount of credit for new borrowers increases by US$ 0.21, and the amount of credit for pre-existing borrowers increases by US$ 0.44. The remainder of the guarantees is allocated to loans that would have been issued anyway. Our work not only contributes to the understanding of economic distortions associated with PCGs but also provides policymakers with relevant knowledge to improve the design of this instrument. Our work complements the empirical literature on PCGs that so far has focused mostly on the additionality of PCG and its effect in production and job creation. For example, Craig et al. (2007a,b, 2008) study the effect of credit guarantee programs implemented by the U.S. Small Business Administration and find a weak relationship between guaranteed lending and future per capita income growth. Uesugi et al. (2010) study a government guarantee program in Japan; they find that though credit guarantees increase the availability of credit in the short term, most participants in the guarantee programs experience a deterioration in their long-term performance on the loans. Boocock and Shariff (2005) study a guarantee scheme in Malaysia but find no conclusive evidence about the effectiveness of the program. Their study also suggests important losses borne by lenders. Cowling (2010) studies a government credit guarantee program in the U.K. He finds that the program achieves the primary objective of increasing availability of credit; however, the effect of the program on economic incentives is unclear. Our paper also contributes to the literature on the role of collateral. For example, our findings are consistent with the results in Berger and Udell (1990) in that increasing the protection for the bank with collateral results in riskier loans. Our results also confirm the findings in Berger et al. (2011b) that collateral plays an important role in disciplining borrowers after loans are issued. Finally, our results support the predictions in Manove et al. (2001) in that relying excessively on collateral (or insurance in our setup) might reduce the quality of borrowers. The rest of the paper is organized as follows. In Section 2, we describe the institutional details of the PCG intervention in Chile and the details of the data. In Section 3, we describe the potential distortions that can be associated with PCG, and we identify a set of testable implications. In Section 4, we explain the methodology and present the main results. Lastly, in Section 5, we present the conclusion and policy implications.

2. Description of the intervention and data 2.1. Description of the intervention In a PCG program, a third party guarantees the partial repayment to the lender in the event that the borrower defaults. PCGs reduce the banks’ exposure to systematic risk and reduce their capital constraints when capital requirements are binding (Ayuso et al., 2004). Borrowers benefit from PCGs because the reduction in the banks’ capital constraints has a positive impact on the aggregate supply of credit (Paravisini, 2008). PCGs can also facilitate borrowing by early-stage entrepreneurs in capital-intensive industries until they accumulate enough assets and/or build the reputation to access the financial markets without external insurance (Berger et al., 2011a). On the flip side, PCGs can generate distortions in economic incentives, decreasing the quality of firms that receive credit and/or reducing managerial effort.9 9 To reduce this perverse incentive, the guarantees are made ‘‘partial’’ and the government or other guarantor only covers a fraction of the loss in the event of default; however, lending without (or with reduced) collateral might exacerbate asymmetric information problems and create important economic distortions even if the lender retains part of the risk (for a detailed explanation see Bester (1985), Jiménez et al. (2006), Rajan and Winton (1995)).

The appeal of PCGs compared with direct subsidies is that the screening, monitoring, and collection tasks can be performed by private financial institutions that might have more expertise in performing these tasks than the provider of the guarantees (Honohan, 2010). The delegation of management tasks in indirect interventions has a significant advantage over direct subsidies, given that most of the time direct subsidies managed by the government do not perform as expected (Khwaja and Mian, 2005; Zia, 2008). In this study, we focus on a PCG structure where the government provides the guarantees and the private sector provides the principal and performs the screening, monitoring, and collection of the loan installments. Though this structure is very common for PCGs in developing countries, it is not the only one. For example, in some countries, PCGs are privately funded. Also, in some countries, the government has a more active role in the screening and monitoring of loans. For a comprehensive description of the different PCG structures across countries, see Beck et al. (2010). The results for this study are based on the operations of the Chilean Credit Guarantee Program (FOGAPE) between January 2003 and September 2006. During that time frame, FOGAPE had a fund of US$ 60 million in liquid assets, and the law allowed the administration to guarantee credit for a total of up to ten times this amount (US$ 600 million). The administration of FOGAPE allocates guarantees among the financial institutions through an auction with sealed bids. Financial institutions can use the allocated guarantees to insure loans at their discretion, subject to satisfying the following conditions:  Only new loans are eligible for insurance (new loans to pre-existing clients are eligible, but pre-existing loans are not).  Insured loans cannot exceed US$ 200,000.  The maximum coverage ratio for loans below US$ 120,000 is 80%.  The maximum coverage ratio for loans above US$ 120,000 is 50%.  The client has to be up to date on all its financial obligations to be eligible for guaranteed loans.10  Only clients with yearly sales below US$ 1,000,000 can get insured loans. The management of the guarantees charges each financial institution a fee that depends on its historic default rate on insured loans. Because of legal restrictions, the fee cannot exceed 2%. Furthermore, the administration reserves the right to exclude institutions from the auction based on poor past performance. In each auction, the insurance administration offers to guarantee a fixed volume of loans. These guarantees are allocated among all participating financial institutions. Each financial institution requests guarantees for a certain volume of loans, at a certain coverage ratio. Institutions requesting the lowest coverage ratio have priority over other institutions to get guarantees for their lending. In the event of a tie in the requested coverage ratio, the tied institutions receive guarantees pro-rata at their requested volume. Therefore, if the insurance administration offers to guarantee a volume of loans larger than the aggregated volume requested by all financial institutions, then each institution receives guarantees for 100% of the loan volume requested. However, if the insurance administration offers to guarantee a volume of loans smaller than the aggregated volume requested by all financial institutions, then 10 This restriction is intended to avoid situations in which a financially distressed client pays off its uninsured debt with newly issued insured debt. In practice, it is easy for banks and clients to hide financial distress from the administration of the guarantees; therefore, this restriction is weakly enforceable.

101

K. Cowan et al. / Journal of Banking & Finance 59 (2015) 98–110 Table 1 Insurance bidding summary. (1)

(2)

Inst. 1

(3)

(4)

Inst. 2

(5)

(6)

Inst. 3

(7)

(8)

Inst. 4

(9)

(10)

Inst. 5

Date

Cover

Funds

Cover

Funds

Cover

Funds

Cover

Funds

Cover

Funds

03/31/03 06/19/03 09/22/03 12/19/03 03/31/04 06/30/04 09/30/04 12/30/04 04/01/05 07/01/05 09/01/05 11/01/05 01/02/06 03/16/06 05/01/06 07/01/06 09/01/06

80 80 80 80 70 70 70 70 70 70 69 67 60 60 60 63 70

98 61 47 67 100 100 100 100 100 100 100 100 100 100 100 100 100

80 80 80 80 80 80 80 80 80 80 70 60 60

98 61 47 67 92 81 77 61 58 12 8 100 100

80 80 80

100 100 100

80 80 80 80 80 80 80 80 80 80 70 65 65 65 65 80 75

98 61 47 67 92 81 77 61 58 12 8 100 100 100 100 100 100

80 80 80 80 80 80 80 80 80 80 70 70 60 65 70 80 80

98 61 47 67 92 81 77 61 58 12 8 7 100 100 100 100 100

80 80 80 80 80 80 80 80 80 80 70 67 65 65 65 70 70

98 61 47 67 92 81 77 61 58 12 8 100 100 100 100 100 100

This table presents the results of the auctions for guarantees for the five largest financial institutions in Chile. The odd columns present the coverage ratio requested by each institution, and the even columns present the volume of guarantees allocated to each institution as a fraction of the volume it requested. All values are expressed in percentages.

the guarantees are allocated in a pecking order from the institution with the lowest coverage ratio to the institution with the highest coverage ratio until guarantees are exhausted. In Table 1, we present the results of the auctions for the five largest financial institutions in Chile between 2003 and 2006. In the odd columns, we present the coverage ratio requested by each institution. In the even columns, we present the guarantees allocated to a financial institution as a fraction of its requested volume. For example, on 06/19/03 all institutions requested an 80% coverage ratio and the summation of the loan volume to be guaranteed requested by each bank was 64% higher than the available guarantees. As a consequence, all institutions were allocated guarantees for 61% of their requested loan volume. Additionally, on 09/01/05, institution 1 requested a 69% coverage ratio, the rest of the institutions requested 70% coverage ratios and the summation of the loan volume to be guaranteed requested by each bank was 89% higher than the available guarantees. Consequently, institution 1 received guarantees for 100% of its requested loan volume, while the other institutions received guarantees for 8% of their requested loan volume. 2.2. Relevance of the intervention in the economy As mentioned in Section 2.1, FOGAPE offers guarantees for a total of US$ 600 million at an average coverage ratio of 70% (see Table 2). Thus, credit issued with guarantees is US$70%600 ¼ US$ 857 million, which is roughly 0.85% of the Chilean GDP during the sample period. While 0.85% of the GDP might seem small, the scope of the intervention in the Chilean economy is not. During the sample period, FOGAPE provided funding for 38,000 firms: 34,000 small firms and 4,000 medium-sized firms. These firms represent 5% of all small firms and 27% of all medium-sized firms in the Chilean economy. In Chile 5% of small firms produce 0.8% of the GDP and hire 3.7% of the labor force, and 27% of medium-sized firms produce 2.9% of the GDP and hire 3.5% of the labor force. Therefore, the firms selected for the guarantee programs account for as much as 3.7% of the GDP and 7.2% of the labor force. 2.3. Data We use a pooled panel data set that merges data from three sources; The Partial Credit Guarantees Administration (FOGAPE),

the Chilean Bureau for Bank Regulation (SBIF), and the Chilean Tax Revenue Office (SII). FOGAPE provides us with the bids of each financial institution and the outcome of each auction. FOGAPE also provides us with the following information about insured loans: the identification number of the borrower, the identification number of the issuing institution, the issue date, the loan size, the coverage ratio, and the maturity.11 SBIF provides financial information at the bank– borrower-month level: For each firm i, bank b and time t, we observe the total credit amount, the fraction of the total that has been in arrears between 61 and 90 days, and the percentage of the total that has been in arrears for more than 90 days. The total credit amount and the amount in arrears between 61 and 90 days are presented by loan category. Therefore, for each borrower, we observe the delinquency on commercial loans, consumer loans, and mortgages independently. However, we only observe the amount in arrears for more than 90 days in the aggregate.12 This financial information is self-reported by all formal financial institutions; reporting is mandatory, and the data are audited by the SBIF.13 The SBIF uses the data to verify whether the banks satisfy the capital requirements recommended in the Basel treaty. The registry also provides financial institutions with information about the riskiness and leverage of potential borrowers. Finally, SII provides a measure of asset size. Because of legal limitations, SII cannot disclose asset amounts; instead, they provide the firms’ ranking according to asset size. This ranking has ten categories, with category one comprising the firms with the smallest asset amount and category 10 comprising the firms with the largest asset amount.

11 To comply with the Chilean law for bank information privacy, these institutions gave us fictitious identification numbers for the borrowers, but these numbers stay consistent across institutions and time. 12 For this study we are interested in loans that can be spent at the discretion of the borrower and that are not secured by a specific asset. Therefore, in most of the analysis, we exclude mortgages. In Table A2 in the Appendix, we show that the main results in the paper do not change if we include mortgages. 13 Similar registries exist in other countries. The information collected in these registries varies across countries. Some countries collect very detailed information about the borrowers that can include credit amount, late payments, demographic data, credit inquiries, ratings, and even utility payments. Other countries collect less detailed information, and in some countries the information is only collected for large borrowers (Djankov et al., 2007).

102

K. Cowan et al. / Journal of Banking & Finance 59 (2015) 98–110

Table 2 Number of new operations and average loan size.

2004 2005 2006

2004 2005 2006

New operations

Fraction of loans with insurance (%)

Loan size

Insured loan size

Coverage ratio (%)

435,205 436,421 461,671

7.97 7.57 5.56

26,238 25,867 25,669

12,433 14,173 17,466

69.9 68.9 66.8

Average N of banks

Delinquency rate (%)

Default rate (%)

Simultaneous loans (%)

2.38 2.46 2.57

4.5 5.7 6.9

2.6 4 5.5

15.4 18.1 18.3

The first panel presents the total number of new credit operations in the Chilean financial market, the fraction of those operations that are insured, the average loan size including all new operations, the average loan size of insured operations, and the average coverage ratio of insured operations. The second panel presents the average number of bank relationships for borrowers with insured loans, their average delinquency rate, their average default rate, and the fraction of those borrowers that have insured and uninsured loans coexisting in the same bank.

For the analysis in the paper, some variables are aggregated at the institution level. C bt is the total credit to SMEs by bank b at time t and equals the summation of all loans outstanding to SMEs P ( cibt ), N bt is the total number of new loans to SMEs by institution  bt is the average size of new loans to SMEs by instib at time t, and C tution b at time t.14 We also generate state variables at the borrower level. The variable nbanksit (or nit ) indicates the number of banks in which the borrower i has non-zero debt at time t, and the variable simit takes the value of one if the borrower i maintains both uninsured and insured loans in at least one financial institution at time t and zero otherwise. For our analysis, we use the fact that commercial loans are eligible for insurance, but consumer loans are not. The distinction between consumer and commercial loans is based on the borrower’s declared use of the money. However, since money is fungible, the borrower can indiscriminately use commercial or consumer loans for business or household expenses. In our fixed effect (or FE) approach, identification comes from entrepreneurs that borrow from more than one bank and simultaneously have consumer and commercial loans with these banks. During the period of analysis, approximately 100,000 operations were insured. In Table 2, we show that for the sample period, FOGAPE guaranteed between 5.5% and 8% of the total loans in the financial system. The average coverage ratio was roughly 68%, and the average size of an insured loan was $12,433 in 2004, $14,173 in 2005, and $17,466 in 2006. In Table 2, we also show that among clients with insured loans, the average number of banks from which they borrow is approximately 2.5, the delinquency rate ranges from 4.5% to 6.9% (depending on the year), and the average fraction of borrowers that maintain uninsured and insured loans at the same bank is approximately 17%. 3. Economic incentives and testable implications PCGs aim to reduce the credit constraints of SMEs, which are particularly vulnerable because of the lack of collateral (Menkhoff et al., 2006). However, PCGs can also affect the incentives of banks to screen or monitor loans, and the incentives of borrowers to exert managerial effort and to repay their obligations. These potential distortions are particularly important for SMEs, 14 We use loan size as a proxy for firm size. Any firm with credit equal to or less than US$ 200,000 is considered an SME.

which usually have more ‘‘opaque’’ information than larger businesses. Distortions can happen at the bank level and/or at the borrower level, and can affect the ex-ante selection of borrowers (adverse selection) or the ex-post behavior of borrowers (moral hazard). 3.1. Distortions at the borrower level 3.1.1. Adverse selection The availability of guarantees might affect the ex-ante willingness of firms to pursue available investment opportunities. Firms without assets (or with low assets) but with access to guaranteed funds might be tempted to invest in negative net present value (NPV) projects because these firms have little to lose when such projects fail; however, these firms might have profits if they get lucky. This opportunistic behavior is less likely to be exhibited by firms that have substantial assets that banks can seize in the event of default. 3.1.2. Moral hazard Guarantees might also affect borrowers’ ex-post behavior if the perceived cost of default decreases. The perceived cost of default might decrease if guarantees reduce the need to pledge collateral or if borrowers expect banks to be more lenient on firms that default on guaranteed loans compared with firms that default on non-guaranteed loans.15 The decrease in the perceived cost of default can cause at least two ex-post distortions: Firms might prioritize the repayment of non-guaranteed loans at the expense of guaranteed loans and/or they might reduce managerial effort. The former distortion would increase delinquency rates but not necessarily default rates; the latter distortion would certainly hurt firms’ performance and increase default rates. The aforementioned negative incentives might be smaller for firms with multiple bank relationships. Indeed, having more banks increases the firms’ outside financial options, reduces the probability that banks capture rents and increases the firms’ incentives to exert effort (Ongena and Smith, 2000; Doblas-Madrid and Minetti, 2013). 3.2. Distortions at the bank level 3.2.1. Adverse selection It is possible that banks reduce screening if the government bears the credit risk. It is also possible that, even if screening does not change, guarantees are allocated strategically by the banks after the screening is done. These two effects, as well as borrower adverse selection, might increase delinquency rates and/or default rates and therefore are observationally very similar. Even with our rich data set, these potential mechanisms are difficult to disentangle. 3.2.2. Moral hazard The banks can also change their ex-post behavior after granting guaranteed loans. For example, they can reduce monitoring of low-quality borrowers. However, while lax screening might take several months to show up in borrowers repayment behavior, changes in monitoring might affect repayment within one billing cycle see Behr et al. (2014). Therefore, by focusing on the timing of the changes in repayment behavior, we can disentangle bank moral hazard from bank adverse selection. 15 For a discussion of the effect of collateral on firms’ incentives, see Bester (1987), Besanko and Thakor (1987) and Chan and Thakor (1987). For a discussion of the effect of banks’ behavior on borrowers’ incentives, see Ambrose and Capone (1996) and Ambrose et al. (1997).

103

K. Cowan et al. / Journal of Banking & Finance 59 (2015) 98–110

Table 3 Within bank–borrower effect of guarantees on loan delinquency rate without interactions.

4. Methodology and results 4.1. Guarantees and economic incentives 4.1.1. Within bank–borrower estimation The most fundamental analysis of this paper tests whether firms exhibit a different behavior toward insured loans compared with their behavior toward uninsured obligations. This analysis relies on the study of two types of loans: loans that are eligible for insurance and loans that are not eligible for insurance. Before we proceed, we will clarify the analysis with an example. Borrower i borrows from two identical banks, b and c. He has two loans with each bank, of which one is eligible for insurance and the other is not. Let’s assume that borrower i agrees with bank b to insure its insurance-eligible loan and agrees with bank c to leave all loans uninsured. The difference between the delinquency rate of the loan eligible for insurance and the loan not eligible for insurance at bank b might be attributed to the effect of guarantees or to the fact that loans eligible for insurance are intrinsically different. However, the difference between the delinquency rates of the loans at bank c are only attributable to the loans eligible for insurance being different from the loans not eligible for insurance. By subtracting the difference in delinquency rates between loans at bank c from the difference in delinquency rates between loans at bank b, we get an unbiased estimation of the effect of guarantees on the behavior of the borrower. In our sample, loans labeled by the registry as ‘‘commercial loans’’ are legally eligible for insurance and loans labeled by the registry as ‘‘consumer loans’’ are not eligible for insurance. The category of a loan depends on the borrower’s declared use of the money; however, because money is fungible, the borrower can indiscriminately use commercial or consumer loans for business or household expenses. Therefore, the category of the loan is immaterial.16 We use the following specification to estimate the double difference described in the example:

dibl ¼ K þ aib þ ui þ cb þ vg ib þ jeibl þ bg ib  eibl þ ibl ;

ð1Þ

where dibl is the delinquency rate of firm i on loan l at bank b; aib ; ui and cb are fixed effects that capture how the specific relationship between bank b and borrower i affects delinquency, how the borrower’s characteristics affect delinquency, and how bank-wide policy changes affect delinquency, respectively; g ib is a variable that takes the value of one if borrower i holds an insured loan from bank b and zero otherwise; and eibl is a variable that takes the value of one if loan l of borrower i at bank b is eligible for insurance and zero otherwise.17 Therefore v captures differences between the delinquency rate of borrower i on loans at banks where the firm keeps insured loans and its delinquency rate on loans at banks where it does not keep insured loans, j captures differences between the borrower’s delinquency rate on insurance-eligible loans and delinquency rate on loans not eligible for insurance, and b captures the difference between the borrower’s delinquency rate on insured loans and delinquency rate on uninsured loans. Note that g ib  eibl takes the value of one if and only if loan l of borrower i at bank b is eligible for insurance and is insured.18 16 While not common, some clients might maintain several consumer loans at the same bank and/or several commercial loans at the same bank. Because of data limitations, we aggregate all consumer loans of client i at bank b and treat the collection as a single loan. Similarly, we aggregate all commercial loans of client i at bank b and treat the collection as a single loan. 17 In the paper we use both the specification with time-invariant fixed effects and the specification with time-varying fixed effects. 18 In this approach, the potential bias associated with borrower heterogeneity is addressed by including a borrower fixed effect and even a bank⁄borrower⁄time fixed effect. This approach has a strong advantage in that the findings are robust to changes in firms’ characteristics.

Delinquency rate after 12 months: (1) g e g⁄e Time f.e. Bank f.e. Firm f.e. Bank-Firm-time f.e. N Adj-r2

0.046 (0.084) 0.433⁄⁄⁄ (0.083) 1.028⁄⁄⁄ (0.128) Yes Yes Yes No 276,978 0.184

Delinquency rate after 24 months:

(2)

(3)

(4)

0.377⁄⁄⁄ (0.121) 1.016⁄⁄⁄ (0.173) No No No Yes 276,978 0.132

0.006 (0.095) 0.993⁄⁄⁄ (0.103) 1.604⁄⁄⁄ (0.167) Yes Yes Yes No 276,978 0.206

1.232⁄⁄⁄ (0.153) 1.668⁄⁄⁄ (0.227) No No No Yes 276,978 0.112

This table presents the results from a fixed effect estimation at the bank–borrowerloan level where the dependent variable is set to one if the loan has past due payments between 61 and 90 days. Each column represents one regression. The independent variables are g that takes the value of one if the borrower has at least one guaranteed loan with the bank and zero otherwise, e that takes the value of one if the loan is eligible for guarantees and zero otherwise, and the interactions between these variables. Note that the interaction between g and e takes the value of one if the loan is guaranteed and zero otherwise and therefore captures the effect of guarantees. All the regressions are controlled for loan size. Standard errors in parentheses are clustered at the borrower level. ⁄⁄⁄, ⁄⁄, and ⁄ indicate significance at the 1%, 5%, and 10% level, respectively.

We call Eq. (1) the within bank–borrower specification because we are estimating the effect of guarantees keeping the borrower– bank pair constant. In Tables 3–5, we estimate Eq. (1) excluding mortgages.19 In Table A2 in the Appendix, we show that the findings in the paper remain unchanged if we compare the performance of commercial loans with the performance of mortgages. In Table 3, we present the estimation of the within bank–borrower specification. Twelve months after a guaranteed loan is issued, its delinquency rate is on average 1.02% higher than the delinquency rate of similar loans issued without guarantees. The gap in the delinquency rate between guaranteed loans and regular loans widens over time. More specifically, twenty-four months after a guaranteed loan is issued, its delinquency rate is on average 1.67% higher than the delinquency rate on similar loans issued without guarantees. Table 4 shows heterogeneous treatment effects with the firm’s asset size, credit size, and number of bank relationships. The increase in the delinquency rate on guaranteed loans is 0.74% larger for large firms (firms with total credit outstanding above the firms’ median credit outstanding in the sample) twelve months after the loan is issued and 1.93% larger twenty-four months after the loan is issued. The change in the delinquency rate on guaranteed loans is 0.75% smaller for firms with high assets twelve months after the loan is issued and 0.43% smaller twenty-four months after the loan is issued, and indicates that firms with high assets have similar repayment behavior on their guaranteed loans and their non-guaranteed loans. The effect of guarantees on repayment behavior does not seem to be affected by the number of bank relationships. Table 4 also suggests that small firms have worse repayment behavior on guaranteed loans only in the short term; however, the repayment behavior of large firms on guaranteed loans deteriorates over time. Table 5 presents the effect of PCGs on different measures of delinquency. We observe that in the short term, guarantees affect delinquency rates but do not affect the number of times the firms are delinquent. This indicates that delinquency episodes start late 19 Mortgages are highly collateralized and are potentially very different from commercial and consumer loans.

104

K. Cowan et al. / Journal of Banking & Finance 59 (2015) 98–110

Table 4 Within bank–borrower effect of guarantees on loan delinquency rate.

g g⁄e g⁄n g⁄a g⁄s g⁄e⁄n g⁄e⁄a g⁄e⁄s Time f.e. Bank f.e. Firm f.e. Bank-Firm-time f.e. N Adj-r2

Delinquency rate after 12 months

Delinquency rate after 24 months

(1)

(3)

0.050 (0.221) 0.942⁄⁄⁄ (0.342) 0.171 (0.155) 0.044 (0.170) 0.081 (0.177) 0.079 (0.269) 0.752⁄⁄⁄ (0.288) 0.681⁄⁄ (0.280) Yes Yes Yes No 276,978 0.185

(2)

0.899⁄ (0.461)

0.086 (0.362) 0.748⁄ (0.389) 0.741⁄⁄ (0.378) No No No Yes 276,978 0.135

0.304 (0.227) 0.404 (0.383) 0.216 (0.175) 0.055 (0.176) 0.497⁄⁄⁄ (0.192) 0.122 (0.345) 0.423 (0.346) 1.943⁄⁄⁄ (0.326) Yes Yes Yes No 276,978 0.207

(4)

0.510 (0.517)

0.124 (0.465) 0.433 (0.467) 1.933⁄⁄⁄ (0.441) No No No Yes 276,978 0.114

This table presents the results from a fixed effect estimation at the bank–borrowerloan level where the dependent variable is set to one if the loan has past due payments between 61 and 90 days and is set to zero otherwise. Each column represents one regression. The independent variables are g that takes the value of one if the borrower has at least one guaranteed loan with the bank and zero otherwise, e that takes the value of one if the loan is eligible for guarantees and zero otherwise, n that takes the value of one if the firm has more than two bank lending relationships and zero otherwise, a that is the firm’s assets relative to its total credit amount and is described in Table A1 in the Appendix, s that is one if the firm’s credit size is above the average firm credit size and zero otherwise, and the interactions between these variables. Note that the interaction between g and e takes the value of one if the loan is guaranteed and zero otherwise and therefore captures the effect of guarantees. Standard errors in parentheses are clustered at the borrower level. ⁄⁄⁄, ⁄⁄ , and ⁄ indicate significance at the 1%, 5%, and 10% level, respectively. Some coefficients are omitted for the sake of brevity.

in the loan cycle (otherwise the number of times delinquent should already be larger after twelve months); we also observe that the amount delinquent (as a percentage of the loan size) does not change within the first twelve months. That said, all the measures of economic distortions – delinquency rate, times delinquent, and amount delinquent – are economically large and statistically significant after twenty-four months. In light of these results, we infer that PCGs are an important source of distortions to the borrowers’ incentives to repay loans. In particular, we think borrowers prioritize the repayment of non-guaranteed loans at the expense of guaranteed loans. Moreover, we observe that the presence of guarantees does not affect the repayment behavior of borrowers with high assets. This result supports the findings in Berger et al. (2011b) that collateral is important in disciplining the borrower after the loan is issued. The number of bank lending relationships does not seem to affect the magnitude of the distortions associated with guarantees. This suggests that different mechanisms that might link the number of bank relationships to the distortions associated with credit insurance seem to cancel out. For example, the potential benefits of sharing information between multiple banks described in Doblas-Madrid and Minetti (2013) might be of a similar magnitude than the negative effect associated with the free-riding problem described in Bolton and Scharfstein (1996). 4.1.2. Within borrower estimation A data limitation of specification 1 is that it needs information at the loan-borrower–bank level, which we have for total credit

outstanding and amount delinquent (amount in arrears between 61 and 90 days) but we don’t have for amount in default (amount in arrears of more than 90 days). In Tables 6 and 7, we estimate a less stringent specification with data aggregated at the bank–borrower level and under which we can only include borrower fixed effects but not borrower–bank fixed effects. We call this specification the within borrower estimation. In Table 6, we estimate the within borrower specification using the delinquency rate as the left-hand side variable. Our findings are consistent with the within bank–borrower specification: The delinquency rate of loans at the bank that provides insurance increases by 0.46% twelve months after the loan is issued (but this increase is not statistically significant), by 1.76% twenty-four months after the loan is issued (significant at the 1% level), and by 2% thirty-six months after the loan is issued (significant at the 1% level).20 These results suggest that borrowers perceive missing payments at banks where they have insured loans as less costly. An alternative explanation is that they exert less effort on projects financed by banks where they have insured loans. To test if guarantees generate a reduction in effort, we repeat in Table 7 the within borrower estimation using the default rate as the left-hand side variable (delays in payment of more than 90 days).21 We observe that borrowers do not seem to default more often on their credit with banks that provide guarantees. This result is important because it shows that while borrowers perceive missing payments on guaranteed loans as less costly than missing payments on non-guaranteed loans, the perception of the cost of default on guaranteed loans is similar to the perception of the cost of default on non-guaranteed loans. Furthermore, it also provides evidence that the overall managerial effort and long-term performance of the borrower is not affected by the guarantees. 4.1.3. Between borrowers estimation We next remove the firm fixed effects, and estimate the between firms effects of guarantees. By removing the firm fixed effects, we aim to compare the performance of firms selected for the guarantee programs with the performance of firms that borrow without guarantees.22 A problem of this approach is that findings can be explained by ex-ante differences in firms’ characteristics (adverse selection) or by ex-post changes in firms behavior associated with the use of guarantees (moral hazard). However, comparing the results of the between estimation with the results of the within estimation will help disentangle these two alternative explanations. In Table 8, we present the results of the between estimation. We find that firms selected into the guarantee programs do not show worse performance in the first year after the loan is issued compared with firms that borrow without guarantees. In the second year, the firms borrowing with guarantees show a higher delinquency rate, but not a higher default rate than those borrowing without guarantees. However, in the third year, the firms with loan guarantees show both a higher delinquency rate and a higher default rate compared with firms without loan guarantees. In particular, the delinquency rate of firms selected into the guarantee programs is 1.44% higher compared with the delinquency rate of firms not in the program, and the default rate of firms selected into the guarantee programs is 1.2% higher compared with the default rate of firms not in them. These findings are consistent with both the adverse selection problems and the moral hazard problems 20

The estimation after 36 months is included in Table A3 in the Appendix. Table 7 presents the effect of guarantees on the default rate twelve and twenty-four months after loans are issued. For the sake of completeness, the effect of guarantees on the default rate after thirty-six months is presented in Table A4 in the Appendix. 22 In this estimation, the variables that determine eligibility into the programs are used as controls; therefore, the results of the regression capture differences between firms that are eligible for the program and those that are not. 21

105

K. Cowan et al. / Journal of Banking & Finance 59 (2015) 98–110 Table 5 Within bank–borrower effect of guarantees on loan delinquency rate, number of months delinquent, and delinquent amount. Measures of delinquency after 12 months

g⁄e

(1) Rate

(2) Count

(3) Amount

(4) Rate

(5) Count

(6) Amount

(7) Rate

(8) Count

(9) Amount

(10) Rate

(11) Count

(12) Amount

1.016⁄⁄⁄ (0.173)

0.004 (0.011)

0.117 (0.119)

0.382⁄⁄ (0.167)

276,978 0.092

0.665⁄⁄ (0.317) 0.365 (0.248) 0.455⁄ (0.266) 0.560⁄⁄ (0.261) 276,978 0.094

0.178⁄⁄⁄ (0.027)

276,978 0.156

0.022 (0.028) 0.027 (0.022) 0.009 (0.024) 0.033 (0.023) 276,978 0.160

1.668⁄⁄⁄ (0.227)

276,978 0.132

0.899⁄ (0.461) 0.086 (0.362) 0.748⁄ (0.389) 0.741⁄⁄ (0.378) 276,978 0.135

276,978 0.112

276,978 0.146

276,978 0.080

0.510 (0.517) 0.124 (0.465) 0.433 (0.467) 1.933⁄⁄⁄ (0.441) 276,978 0.114

0.031 (0.065) 0.008 (0.057) 0.042 (0.059) 0.220⁄⁄⁄ (0.054) 276,978 0.151

0.615 (0.382) 0.094 (0.341) 0.503 (0.336) 1.078⁄⁄⁄ (0.337) 276,978 0.081

g⁄e⁄n g⁄e⁄a g⁄e⁄s N Adj-r2

Measures of delinquency after 24 months

This table presents the results from a fixed effect estimation at the bank–borrower-loan level where the dependent variables are the delinquency rate that takes the value of one if the loan has past due payments between 61 and 90 days and zero otherwise, the number of months delinquent, and delinquent amount (as a fraction of the loan size). Each column represents one regression. The independent variables are g that takes the value of one if the borrower has at least one guaranteed loan with the bank and zero otherwise, e that takes the value of one if the loan is eligible for guarantees and zero otherwise, n that takes the value of one if the firm has more than two bank lending relationships and zero otherwise, a that is the firm’s assets relative to its total credit amount and is described in Table A1 in the Appendix, s that is one if the firm’s credit size is above the average firm credit size and zero otherwise, and the interactions between these variables. Note that the interaction between g and e takes the value of one if the loan is guaranteed and zero otherwise and therefore captures the effect of guarantees. All the regressions include bank-firm-time fixed effects. Standard errors in parentheses are clustered at the borrower level. ⁄⁄⁄, ⁄⁄, and ⁄ indicate significance at the 1%, 5%, and 10% level, respectively. Some coefficients are omitted for the sake of brevity.

Table 6 Within borrower effect of guarantees on loan delinquency rate, number of months delinquent, and delinquent amount. Measures of delinquency after 12 months

g

(2) Count

(3) Amount

(4) Rate

(5) Count

(6) Amount

(7) Rate

(8) Count

(9) Amount

(10) Rate

(11) Count

(12) Amount

0.462 (0.306)

0.024 (0.019)

0.276⁄ (0.168)

0.632⁄⁄⁄ (0.241)

138,717 0.186

0.145 (0.441) 0.228 (0.290) 0.015 (0.297) 0.655⁄ (0.394) 138,717 0.186

0.112⁄⁄⁄ (0.040)

138,717 0.466

0.002 (0.050) 0.082⁄⁄ (0.033) 0.016 (0.034) 0.035 (0.045) 138,717 0.466

1.758⁄⁄⁄ (0.360)

138,717 0.267

0.460 (0.804) 0.409 (0.529) 0.207 (0.541) 0.400 (0.719) 138,717 0.266

138,717 0.209

138,717 0.465

138,717 0.148

0.808 (0.948) 0.244 (0.624) 0.250 (0.638) 1.475⁄ (0.847) 138,717 0.209

0.057 (0.106) 0.132⁄ (0.070) 0.062 (0.072) 0.192⁄⁄ (0.095) 138,717 0.465

0.208 (0.634) 0.137 (0.417) 0.627 (0.427) 0.631 (0.567) 138,717 0.148

g⁄n g⁄a g⁄s N Adj-r2

Measures of delinquency after 24 months

(1) Rate

This table presents the results from a fixed effect estimation at the bank–borrower level where the dependent variables are the delinquency rate that takes the value of one if the borrower has past due payments with the bank between 61 and 90 days and zero otherwise, the number of months delinquent, and the delinquent amount (as a fraction of total credit at the bank). Each column represents one regression. The independent variables are g that takes the value of one if the borrower has at least one guaranteed loan with the bank and zero otherwise, n that takes the value of one if the firm has more than two bank lending relationships and zero otherwise, a that is the firm’s assets relative to its total credit amount and is described in Table A1 in the Appendix, s that is one if the firm’s credit size is above the average firm credit size and zero otherwise, and the interactions between these variables. All the regressions include firm-time and bank fixed effects. Standard errors in parentheses are clustered at the borrower level. ⁄⁄⁄, ⁄⁄, and ⁄ indicate significance at the 1%, 5%, and 10% level, respectively. Some coefficients are omitted for the sake of brevity.

Table 7 Within borrower effect of guarantees on loan default rate, number of months in default, and default amount. Measures of default after 12 months

g

(1) Rate

(2) Count

(3) Amount

(4) Rate

(5) Count

(6) Amount

(7) Rate

(8) Count

(9) Amount

(10) Rate

(11) Count

(12) Amount

0.167 (0.163)

0.011 (0.010)

0.132 (0.097)

0.575⁄⁄⁄ (0.162)

138,717 0.473

0.124 (0.256) 0.137 (0.169) 0.499⁄⁄⁄ (0.172) 0.296 (0.229) 138,717 0.474

0.050⁄ (0.028)

138,717 0.664

0.014 (0.028) 0.002 (0.018) 0.012 (0.019) 0.003 (0.025) 138,717 0.664

0.313 (0.244)

138,717 0.390

0.074 (0.429) 0.022 (0.282) 0.336 (0.288) 0.525 (0.383) 138,717 0.390

138,717 0.281

138,717 0.479

138,717 0.277

0.638 (0.641) 0.596 (0.422) 0.078 (0.431) 0.729 (0.573) 138,717 0.281

0.018 (0.073) 0.006 (0.048) 0.078 (0.049) 0.093 (0.065) 138,717 0.479

0.900⁄⁄ (0.427) 0.541⁄ (0.281) 0.445 (0.287) 0.387 (0.382) 138,717 0.277

g⁄n g⁄a g⁄s N Adj-r2

Measures of default after 24 months

This table presents the results from a fixed effect estimation at the bank–borrower level where the dependent variables are the default rate that takes the value of one if the borrower has past due payments with the bank of more than 90 days and zero otherwise, the number of months in default, and the default amount (as a fraction of total credit at the bank). Each column represents one regression. The independent variables are g that takes the value of one if the borrower has at least one guaranteed loan with the bank and zero otherwise, n that takes the value of one if the firm has more than two bank lending relationships and zero otherwise, a that is the firm’s assets relative to its total credit amount and is described in Table A1 in the Appendix, s that is one if the firm’s credit size is above the average firm credit size and zero otherwise, and the interactions between these variables. All the regressions include firm-time and bank fixed effects. Standard errors in parentheses are clustered at the borrower level. ⁄⁄⁄, ⁄⁄, and ⁄ indicate significance at the 1%, 5%, and 10% level, respectively. Some coefficients are omitted for the sake of brevity.

106

K. Cowan et al. / Journal of Banking & Finance 59 (2015) 98–110

Table 8 Differences in loan delinquency rate, delinquent amount, default rate and default amount between borrowers with guarantees and borrowers without guarantees. Measures of delinquency 12 months

g g⁄n g⁄a g⁄s N Adj-r2

Measures of default 24 months

36 months

12 months

24 months

36 months

(1) Rate

(2) Amount

(3) Rate

(4) Amount

(5) Rate

(6) Amount

(7) Rate

(8) Amount

(9) Rate

(10) Amount

(11) Rate

(12) Amount

0.017 (0.226) 0.052 (0.245) 0.121 (0.212) 0.147 (0.225) 156,190 0.004

0.085 (0.117) 0.001 (0.099) 0.064 (0.103) 0.140 (0.107) 156,190 0.003

1.204⁄⁄⁄ (0.344) 0.220 (0.357) 0.493 (0.326) 0.010 (0.344) 156,190 0.013

0.654⁄⁄⁄ (0.188) 0.291 (0.185) 0.289 (0.185) 0.188 (0.192) 156,190 0.006

1.437⁄⁄⁄ (0.346) 0.343 (0.351) 0.139 (0.338) 0.226 (0.348) 156,190 0.029

0.933⁄⁄⁄ (0.218) 0.368⁄ (0.222) 0.759⁄⁄⁄ (0.227) 0.119 (0.234) 156,190 0.018

0.052 (0.077) 0.009 (0.084) 0.028 (0.070) 0.090 (0.070) 156,190 0.001

0.039 (0.063) 0.044 (0.059) 0.000 (0.053) 0.076 (0.057) 156,190 0.001

0.395 (0.246) 0.123 (0.246) 0.356⁄ (0.215) 0.148 (0.230) 156,190 0.012

0.126 (0.212) 0.056 (0.205) 0.108 (0.183) 0.033 (0.196) 156,190 0.010

1.173⁄⁄⁄ (0.332) 0.189 (0.303) 0.952⁄⁄⁄ (0.290) 0.006 (0.295) 156,190 0.034

0.738⁄⁄ (0.292) 0.300 (0.258) 0.492⁄ (0.251) 0.101 (0.260) 156,190 0.031

This table presents the results from an ordinary least squares (OLS) regression at the borrower level where the dependent variables are the delinquency rate that takes the value of one if the borrower has past due payments with the bank between 61 and 90 days and zero otherwise, the delinquent amount (as a fraction of total credit at the bank), the default rate that takes the value of one if the borrower has past due payments of more than 90 days and zero otherwise, and the default amount (as a fraction of total credit at the bank). Each column represents one regression. The independent variables are g that takes the value of one if the borrower was selected into the guarantee programs at any point during the sample period, n that takes the value of one if the firm has more than two bank lending relationships and zero otherwise, a that is the firm’s assets relative to its total credit amount and is described in Table A1 in the Appendix, s that is one if the firm’s credit size is above the average firm credit size and zero otherwise, and the interactions between these variables. All the regressions include time fixed effects and bank fixed effects (the bank where the borrower has the highest outstanding credit amount). Standard errors in parentheses are clustered at the borrower level. ⁄⁄⁄, ⁄⁄, and ⁄ indicate significance at the 1%, 5%, and 10% level, respectively. Some coefficients are omitted for the sake of brevity.

described in Section 3. However, from the analysis of Tables 7 and A4, we know that guarantees don’t affect the long-term performance of the firms, and therefore, we conclude the higher default rates among firms selected into the guarantee programs are the consequence of adverse selection. We also find interesting heterogeneous treatment effects. Even after three years, high-asset firms selected into the guarantee programs do not show higher default rates compared with similar firms borrowing without guarantees. This is consistent with the idea that high-asset firms have fewer incentives to pursue low-quality projects even if they have access to guarantees, which in turn reduces the adverse selection problem. 4.2. Guarantees and loan size at the borrower level In this section, we study how guarantees affect the size of loans to SMEs. The methodology resembles that in Section 4.1 with a small twist; we use the following specification:

log ib ¼ K þ aib þ ui þ cb þ vg ib þ ib ;

ð2Þ

where most variables are equivalent to those in Eq. (1) and log ib is the logarithm of the size of the loan of client i with bank b. In this case, we cannot identify the b of specification (1) because borrowers seldom get a loan eligible for insurance and a loan not eligible for insurance at the same time and from the same financial institution. We can, however, take the first time-difference to study how loan size growth is affected by the presence of guarantees. The ordinary least squares (OLS) estimation of the first time-difference of Eq. (2) is presented in Table 9. We observe that credit growth is 25% larger when a loan is issued with guarantees, compared with a similar loan issued without guarantees. The effect is economically large in all estimations, but is not statistically significant when including firm, firm-period, and firm-institution fixed effects. These results suggest that guarantees are effective in helping borrowers get larger loans if they need to. The caveat in this analysis is that we cannot control for time-varying changes in borrowing behavior. This might be problematic if borrower i has a different motivation for borrowing from bank b at time t than for ~ at time ~t. Thus, the results in Table 9 should borrowing from bank b be interpreted cautiously.

Table 9 Effect of guarantees on loan size growth at the borrower level. Loan size growth

N Adj-r2

0.297⁄⁄⁄ (0.091) 62,175 0.002

0.253⁄⁄ (0.096) 62,175 0.007

0.310⁄⁄⁄ (0.097) 62,175 0.016

0.141 (0.189) 62,175 0.159

0.290 (0.327) 62,175 0.257

0.229 (0.323) 62,175 0.264

Time f.e. Bank Firm f.e. Firm-time f.e. Firm-bank f.e.

No No No No No

Yes No No No No

Yes Yes No No No

Yes Yes Yes No No

No No No Yes No

No No No No Yes

Guarantees

This table presents the difference between the growth rate of loans issued with guarantees and the growth rate of similar loans issued without guarantees. Growth rate is estimated as the difference between the logarithm of the new loan amount logðnew loan amount þ 1Þ and the logarithm of the pre-existing credit amount logðpre-existing credit amount þ 1Þ. Columns differ in the types of fixed effects included in the estimation. Standard errors in parentheses are clustered at the bank level. ⁄⁄⁄, ⁄⁄, and ⁄ indicate significance at the 1%, 5%, and 10% level, respectively.

4.3. Guarantees and aggregated credit We estimate how insurance affects the aggregated credit that goes to SMEs by exploiting the nonlinear variation in the amount of guarantees generated in the bidding process. We also include a time fixed effect to control for the aggregated trend in the credit market and an institution fixed effect to control for the heterogeneity between banks. We estimate the extent to which the amount of insurance allocated to financial institution b affects the total amount of credit, the average size of new loans, and the number of new loans issued to SMEs by institution b. Specifically, we use the following specification:

DX bt ¼ K þ bDIbt þ lt þ cb þ bt ;

ð3Þ

where D is the first time-difference operator, X bt is the dependent variable (e.g., total credit, the number of new loans, and the average size of new loans) for institution b at time t; Ibt is the guarantees allocated to institution b at time t; lt captures time fixed effects, and cb captures bank fixed effects. In Table 10, we present the effect of the availability of insurance on the amount of credit issued to SMEs. In the first three columns,

107

K. Cowan et al. / Journal of Banking & Finance 59 (2015) 98–110 Table 10 Availability of credit guarantees and aggregated amount of credit for SMEs. Credit increase Total 0.792⁄⁄⁄ (0.101) 475 0.0382 No No

Insurance N Adj-r 2 Bank f.e. Time f.e.

New 0.787⁄⁄⁄ (0.104) 475 0.0070 Yes No

0.649⁄⁄⁄ (0.099) 475 0.1284 Yes Yes

0.253⁄⁄⁄ (0.048) 475 0.0509 No No

Renewed 0.253⁄⁄⁄ (0.049) 475 0.0169 Yes No

0.207⁄⁄⁄ (0.062) 475 0.1483 Yes Yes

0.539⁄⁄⁄ (0.062) 475 0.0237 No No

0.535⁄⁄⁄ (0.064) 475 0.0067 Yes No

0.442⁄⁄⁄ (0.075) 475 0.0692 Yes Yes

This table presents the effect of credit insurance on the aggregated amount of credit available for small- and medium-sized enterprises. The first three columns present the total increase in credit generated by one additional unit of insurance. The next three columns present the increase in credit to new clients generated by one additional unit of insurance, and the final three columns present the increase in credit to pre-existing clients generated by one unit of insurance. Standard errors in parentheses are clustered at the bank level. ⁄⁄⁄, ⁄⁄, and ⁄ indicate significance at the 1%, 5%, and 10% level, respectively.

Table 11 Availability of credit guarantees and number and size of loans to SMEs. Number of loans

Insurance N Adj-r2 Bank f.e. Time f.e.

Size of loans

Total loans

New loans

Renewed loans

All loans

New loans

Renewed loans

0.147⁄⁄⁄ (0.051) 475 0.0302 Yes Yes

0.022⁄⁄⁄ (0.005) 475 0.1298 Yes Yes

0.125⁄⁄⁄ (0.047) 475 0.0376 Yes Yes

0.219 (0.296) 452 0.0347 Yes Yes

0.194 (0.228) 450 0.0017 Yes Yes

0.173 (0.247) 440 0.0206 Yes Yes

This table presents the effect of credit insurance on the total number of loans issued to small- and medium-sized enterprises, and the effect of credit insurance on the average size of loans to SMEs. The first three columns present the increase in the number of loans generated by the addition of $1700 (Ch$ 1,000,000) of insurance, and the final three columns present the increase in loan size. Standard errors in parentheses are clustered at the bank level. ⁄⁄⁄, ⁄⁄, and ⁄ indicate significance at the 1%, 5%, and 10% level, respectively.

we present the effect on the total amount of credit. In the remaining columns, we separate the effect into two categories: credit for new clients and credit for pre-existing clients. We observe that for each additional unit of guarantees to institution b, its credit to SMEs increases by 0.65; credit to new clients increases by 0.21, while credit for pre-existing clients increases by 0.44. The increase in credit found in Table 10 can be explained by an increase in the size of new loans, an increase in the number of new loans, or both. In Table 11, we present the effect of guarantees on the number and average size of new loans and loan renewals. In the first, second, and third columns, we observe that the addition of guarantees for loans equivalent to US$ 1,700 (Chilean$ 1,000,000) is associated with an increase of 0.147 in the number of loans, 0.022 in the number of new loans, and 0.125 in the number of loan renewals; all changes are significant at the 1% level. In Section 4.2 we showed that for a given firm, credit size increases when its bank is allocated more guarantees. But we also

expect banks to increase the number of loans issued to SMEs (that borrow smaller amounts) when they are allocated more guarantees. In the fourth, fifth, and sixth columns of Table 11, we observe that the addition of guarantees does not change the average loan size, suggesting that the effects discussed earlier cancel each other. In specification (3), we control for time trends and for institutional fixed characteristics. However, there can still be time-varying institutional characteristics that are correlated with the amount of guarantees allocated to financial institutions. For example, banks might increase their bids for insurance when there is an increase in the demand for credit and that would create a spurious positive correlation between the amount of insurance and credit for SMEs. Because the auction data are at the institution-time level, we cannot include timeinstitution fixed effects to control this bias. However, we can address this concern by using the structure of the auction. As described in Section 2, the availability of insurance to institution b depends on the volume of guarantees offered by FOGAPE, the institution’s b bid in the auction, and the other participants’ bids in the auction. Institution b can affect the outcome of the auction by changing its own bid, but the other variables are beyond its control. We estimate the expected amount of insurance to institution b (from the perspective of institution b) as the fitted value of the following specification:

Ibt ¼ K þ xCRbt þ qGbt þ bt þ lt þ cb ;

ð4Þ

where Ibt is the amount of guarantees allocated to institution b at time t; CRbt is the coverage ratio requested by institution b at time t; Gbt is the volume of guarantees requested by institution b at time t; lt captures time fixed effects, and cb captures bank fixed effects. The residual of the OLS estimation of specification 4 can be interpreted as the unexpected amount of insurance allocated to institution b. As a robustness check to our findings in Table 10, we re-estimate Eq. (3) by substituting the amount of insurance

Table 12 Credit to SMEs as a function of unexpected changes in the availability of guarantees. Credit increase Total Residual N Adj-r2 Bank f.e. Time f.e.

0.671 (0.414) 475 0.0116 No No

New 0.665 (0.418) 475 0.0203 Yes No

0.613⁄ (0.339) 475 0.1127 Yes Yes

0.266⁄⁄ (0.118) 475 0.0257 No No

Renewed 0.265⁄⁄ (0.12) 475 0.0092 Yes No

0.234⁄⁄ (1.109) 475 0.1347 Yes Yes

0.405 (0.327) 475 0.0048 No No

0.400 (0.331) 475 0.0260 Yes No

0.378 (0.286) 475 0.0578 Yes Yes

This table presents the effect of unexpected changes in the availability of guarantees on the aggregated amount of credit available for small- and medium-sized enterprises. The first three columns present the total increase in credit generated by one unexpected additional unit of insurance. The next three columns present the increase in credit to new clients generated by one unexpected additional unit of insurance, and the final three columns present the increase in credit to pre-existing clients generated by one unexpected unit of insurance. Standard errors in parentheses are clustered at the bank level. ⁄⁄⁄, ⁄⁄, and ⁄ indicate significance at the 1%, 5%, and 10% level, respectively.

108

K. Cowan et al. / Journal of Banking & Finance 59 (2015) 98–110

Table A1 Variable definitions. Variable

Definition

Delinquency rate diblt

equals one if borrower i has amounts past due between 61 and 90 days on loan l at bank b at time t and zero otherwise is the number of months the borrower has been delinquent on a loan within the last n months equals the amount past due between 61 and 90 days divided by the loan size equals one if borrower i has amounts past due more than 90 days and zero otherwise is the number of months the borrower has been in default within the last n months equals the amount past due of more than 90 days divided by the credit size equals one if loan l is guaranteed and zero otherwise (and it is equivalent to g ibt eiblt ) equals one if borrower i has at least one guaranteed loan at bank b at time t and zero otherwise equals one if loan l of borrower i at bank b at time t is eligible for insurance and zero otherwise is the total credit to borrower i by bank b at time t and equals the summation of the commercial loans, the consumer loans, and the mortgages of borrower i at bank b is the total credit to SMEs by bank b at time t and equals the summation of all loans outstanding P to SMEs ( cibt ) is the total number of new loans issued to SMEs by institution b at time t is the average size of new loans to SMEs by institution b at time t equals one if borrower i has more than two bank lending relationship at time t and zero otherwise takes the value of one if the borrower i maintains both uninsured and insured loans in at least one financial institution at time t and zero otherwise equals one if borrower i asset size relative to other borrowers is larger than his credit size relative to other borrowers equals one if borrower i’s total credit is above the median firm total credit at time t is the amount of guarantees available to bank b at time t is the volume of guarantees requested by bank b at time t is the coverage ratio requested by bank b at time t is the constant variable in the regressions

Delinquent count Delinquent amount Default rate Default count Default amount Insurance insibl Guarantee g ibt Eligible eiblt Total credit cibt Total credit by bank C bt New loans by bank N bt  Average size by bank C bt Number of banks nbanksit or nit Simultaneous loans simit Assets ait Size sit Total guarantees Ibt Requested guarantees Gbt Coverage ratio CRbt K

Table A2 Within bank–borrower effect of credit guarantees on commercial loans and mortgages. Measures of delinquency after 12 months

g⁄e

(2) Count

(3) Amount

(4) Rate

(5) Count

(6) Amount

(7) Rate

(8) Count

(9) Amount

(10) Rate

(11) Count

(12) Amount

1.138⁄⁄⁄ (0.172)

0.022⁄⁄ (0.011)

0.153 (0.104)

0.590⁄⁄⁄ (0.152)

276,978 0.040

0.382 (0.264) 0.233 (0.217) 0.263 (0.232) 0.620⁄⁄⁄ (0.216) 276,978 0.043

0.208⁄⁄⁄ (0.028)

276,978 0.121

0.017 (0.030) 0.032 (0.024) 0.027 (0.026) 0.041⁄ (0.024) 276,978 0.123

1.547⁄⁄⁄ (0.221)

276,978 0.091

1.239⁄⁄⁄ (0.451) 0.054 (0.370) 1.083⁄⁄⁄ (0.395) 0.664⁄ (0.366) 276,978 0.094

276,978 0.091

276,978 0.119

276,978 0.055

0.143 (0.492) 0.293 (0.459) 0.249 (0.461) 1.989⁄⁄⁄ (0.423) 276,978 0.093

0.088 (0.068) 0.006 (0.060) 0.090 (0.063) 0.227⁄⁄⁄ (0.056) 276,978 0.123

0.495 (0.338) 0.021 (0.313) 0.422 (0.306) 1.176⁄⁄⁄ (0.304) 276,978 0.056

g⁄e⁄n g⁄e⁄a g⁄e⁄s N Adj-r2

Measures of delinquency after 24 months

(1) Rate

This table presents the results from a fixed effect estimation at the bank–borrower-loan level where the dependent variables are the delinquency rate that takes the value of one if the loan has past due payments between 61 and 90 days and zero otherwise, the number of months delinquent, and the delinquent amount (as a fraction of the loan size). Each column represents one regression. The independent variables are the interaction between g and e, where g takes the value of one if the borrower has at least one guaranteed loan with the bank and zero otherwise and e takes the value of one if the loan is eligible for guarantees and zero otherwise; n that takes the value of one if the firm has more than two bank lending relationships and zero otherwise; a that is the firm’s assets relative to its total credit amount and is described in Table A1 in the Appendix; s that is one if the firm’s credit size is above the average firm credit size and zero otherwise; and the interactions between these variables. Note that the interaction between g and e takes the value of one if the loan is guaranteed and zero otherwise and therefore captures the effect of guarantees. All the regressions include bank-firm-time fixed effects. Standard errors in parentheses are clustered at the borrower level. ⁄⁄⁄, ⁄⁄, and ⁄ indicate significance at the 1%, 5%, and 10% level, respectively. Some coefficients are omitted for the sake of brevity.

allocated to each institution with the residual of Eq. (4). This leads to the following specification:

DX bt ¼ K þ bDRbt þ lt þ cb ;

ð5Þ

which is similar to specification (3), except we use the residual of Eq. (4) Rbt instead of the allocated amount of insurance Ibt . The results of the OLS estimation of specification (5) are presented in Table 12. We see that one unit of unexpected insurance increases total credit by 0.61, credit to new clients increases by 0.23, and credit to pre-existing clients increases by 0.38. The results in Table 12 provide additional support to the idea that PCGs increase the aggregated amount of credit available to SMEs.

5. Conclusions and policy implications The results in this paper show that partial credit guarantees severely affect the delinquency rate of insured loans. In particular, insured loans are 1.67% more likely to be delinquent after twenty-four months than similar loans without insurance. We also find that this effect is more pronounced for larger borrowers but is not present for borrowers with high assets. The reduction in the repayment rate can undermine the borrower’s future credit capacity with other banks.23 Therefore, the importance of reducing PCG distortions to repayment is twofold: It 23 We are assuming a myopic borrower, which based on anecdotal information, is a sensible assumption for this type of borrower.

K. Cowan et al. / Journal of Banking & Finance 59 (2015) 98–110 Table A3 Within borrower effect of guarantees on loan delinquency rate, number of months delinquent, and delinquent amount. Measures of delinquency after 36 months

g

(1) Rate

(2) Count

(3) Amount

(4) Rate

(5) Count

(6) Amount

2.049⁄⁄⁄ (0.386)

0.353⁄⁄⁄ (0.066)

0.684⁄⁄⁄ (0.261)

2.306⁄⁄ (1.016) 0.577 (0.667) 2.287⁄⁄⁄ (0.682) 1.710⁄ (0.908)

0.217 (0.174) 0.134 (0.114) 0.238⁄⁄ (0.117) 0.418⁄⁄⁄ (0.156)

0.504 (0.687) 0.392 (0.452) 0.691 (0.462) 0.963 (0.614)

138,717 0.037

138,717 0.411

138,717 0.749

138,717 0.037

138,717 0.411

138,717 0.749

g⁄n g⁄a g⁄s N Adj-r2

This table presents the results from a fixed effect estimation at the bank–borrower level where the dependent variables are the delinquency rate that takes the value of one if the borrower has past due payments with the bank between 61 and 90 days and zero otherwise, the number of months delinquent, and the delinquent amount (as a fraction of total credit at the bank). Each column represents one regression. The independent variables are g that takes the value of one if the borrower has at least one guaranteed loan with the bank and zero otherwise, n that takes the value of one if the firm has more than two bank lending relationships and zero otherwise, a that is the firm’s assets relative to its total credit amount and is described in Table A1 in the Appendix, s that is one if the company’s credit size is above the average firm credit size and zero otherwise, and the interactions between these variables. All the regressions include firm-time and bank fixed effects. Standard errors in parentheses are clustered at the borrower level. ⁄⁄⁄, ⁄⁄, and ⁄ indicate significance at the 1%, 5%, and 10% level, respectively. Some coefficients are omitted for the sake of brevity.

Table A4 Within borrower effect of guarantees on loan default rate, number of months in default, and default amount. Measures of default after 36 months

g

(1) Rate

(2) Count

(3) Amount

(4) Rate

(5) Count

(6) Amount

0.269 (0.304)

0.068 (0.049)

0.771⁄⁄⁄ (0.215)

0.624 (0.801) 0.013 (0.526) 1.257⁄⁄ (0.537) 0.486 (0.716)

0.022 (0.130) 0.021 (0.085) 0.125 (0.087) 0.187 (0.116)

1.602⁄⁄⁄ (0.566) 0.617⁄ (0.373) 1.551⁄⁄⁄ (0.381) 0.228 (0.507)

138,717 0.100

138,717 0.395

138,717 0.820

138,717 0.100

138,717 0.395

138,717 0.820

g⁄n g⁄a g⁄s N Adj-r 2

This table presents the results from a fixed effect estimation at the bank–borrower level where the dependent variables are the default rate that takes the value of one if the borrower has past due payments with the bank of more than 90 days and zero otherwise, the number of months in default, and the default amount (as a fraction of total credit at the bank). Each column represents one regression. The independent variables include g that takes the value of one if the borrower has at least one guaranteed loan with the bank and zero otherwise, n that takes the value of one if the firm has more than two bank lending relationships and zero otherwise, a that is the firm’s assets relative to its total credit amount and is described in Table A1 in the Appendix, s that is one if the firm’s credit size is above the average firm credit size and zero otherwise, and the interactions between these variables. All the regressions include firm-time and bank fixed effects. Standard errors in parentheses are clustered at the borrower level. ⁄⁄⁄, ⁄⁄, and ⁄ indicate significance at the 1%, 5%, and 10% level, respectively. Some coefficients are omitted for the sake of brevity.

makes partial credit guarantees more sustainable, and more importantly, it positively affects the future credit capacity of the borrower that is one of the main objectives of partial credit guarantees. We also find that firms selected into the guarantee programs are 1.2% more likely to default than firms borrowing without guarantees. The analysis in the paper suggests the higher default is most likely caused by adverse selection. Given data limitations, we cannot determine whether the adverse selection is at the

109

borrower level, at the bank level or both. Considering that firms selected into the guarantee programs produce 3.7% of Chile’s GDP and generate 7.2% of the jobs in Chile, a 1.2% higher default rate might translate into a reduction in the GDP of $44 million and a destruction of approximately 7,000 jobs.24 We also confirm previous findings that guarantees are effective in increasing credit to SMEs. One unit of insurance is associated with an increase of 0.65 in the amount of credit issued to SMEs. Nonetheless, part of the insurance is allocated to loans that would have been issued anyway. A challenge for policy makers is to increase the fraction of the guarantees that are allocated to loans that would have not been issued without insurance. Appendix A See Tables A1–A4. References Ambrose, B.W., Buttimer, R.J., Capone, C.A., 1997. Pricing mortgage default and foreclosure delay. Journal of Money, Credit & Banking (Ohio State University Press) 29 (3). Ambrose, B.W., Capone, C.A., 1996. Cost-benefit analysis of single-family foreclosure alternatives. The Journal of Real Estate Finance and Economics 13 (2), 105–120. Ayuso, J., Pérez, D., Saurina, J., 2004. Are capital buffers pro-cyclical?: Evidence from spanish panel data. Journal of Financial Intermediation 13 (2), 249–264. Bannock, G., 2005. The economics and management of small business: an international perspective. Routledge. Beck, T., Demirgüç-Kunt, A., Maksimovic, V., 2005. Financial and legal constraints to growth: Does firm size matter? The Journal of Finance 60 (1), 137–177. Beck, T., Demirgüç-Kunt, A., Martinez Peria, M.S., 2007. Reaching out: Access to and use of banking services across countries. Journal of Financial Economics 85 (1), 234–266. Beck, T., Klapper, L.F., Mendoza, J.C., 2010. The typology of partial credit guarantee funds around the world. Journal of Financial Stability 6 (1), 10–25. Behr, P., Drexler, A.H., Gropp, R., Guettler, A., 2014. Financial incentives and loan officer behavior: Multitasking and allocation of effort under an incomplete contract. SAFE Working Paper. Berger, A.N., Espinosa-Vega, M.A., Frame, W.S., Miller, N.H., 2011a. Why do borrowers pledge collateral? new empirical evidence on the role of asymmetric information. Journal of Financial Intermediation 20 (1), 55–70. Berger, A.N., Scott Frame, W., Ioannidou, V., 2011b. Tests of ex ante versus ex post theories of collateral using private and public information. Journal of Financial Economics 100 (1), 85–97. Berger, A.N., Udell, G.F., 1990. Collateral, loan quality and bank risk. Journal of Monetary Economics 25 (1), 21–42. Besanko, D., Thakor, A.V., 1987. Collateral and rationing: sorting equilibria in monopolistic and competitive credit markets. International Economic Review, 671–689. Bester, H., 1985. Screening vs. rationing in credit markets with imperfect information. The American Economic Review 75 (4), 850–855. Bester, H., 1987. The role of collateral in credit markets with imperfect information. European Economic Review 31 (4), 887–899. Bolton, P., Scharfstein, D., 1996. Optimal Debt Structure and the Number of Creditors. Journal of Political Economy 104 (1), 1–25. Boocock, G., Shariff, M.N.M., 2005. Measuring the effectiveness of credit guarantee schemes: Evidence from Malaysia. International Small Business Journal 23 (4). Chan, Y.-S., Thakor, A.V., 1987. Collateral and competitive equilibria with moral hazard and private information. The Journal of Finance 42 (2), 345–363. Chaney, P.K., Thakor, A.V., 1985. Incentive effects of benevolent intervention: The case of government loan guarantees. Journal of Public Economics 26 (2), 169– 189. Cowling, M., 2010. The role of loan guarantee schemes in alleviating credit rationing in the UK. Journal of Financial Stability 6 (1), 36–44. Craig, B.R., Jackson, W.E., Thomson, J.B., 2007a. Small firm credit market discrimination, small business administration guaranteed lending, and local market economic performance. The Annals of the American Academy of Political and Social Science 613 (1), 73–94. Craig, B.R., Jackson, W.E., Thomson, J.B., 2007b. Small firm finance, credit rationing, and the impact of sba-guaranteed lending on local economic growth. Journal of Small Business Management 45 (1), 116–132. Craig, B.R., Jackson, W.E., Thomson, J.B., 2008. Credit market failure intervention: Do government sponsored small business credit programs enrich poorer areas? Small Business Economics 30, 345–360. Djankov, S., McLiesh, C., Shleifer, A., 2007. Private credit in 129 cobuntries. Journal of Financial Economics 84 (2), 299–329. 24 The GDP in Chile during the time of the study was US$ 100 billion, and the total number of jobs was approximately 8,000,000.

110

K. Cowan et al. / Journal of Banking & Finance 59 (2015) 98–110

Doblas-Madrid, A., Minetti, R., 2013. Sharing information in the credit market: Contract-level evidence from us firms. Journal of Financial Economics 109 (1), 198–223. Evans, D.S., Jovanovic, B., 1989. An estimated model of entrepreneurial choice under liquidity constraints. The Journal of Political Economy 97 (4), 808–827. Gale, W.G., 1990. Federal lending and the market for credit. Journal of Public Economics 42 (2), 177–193. Green, A., 2003. Credit guarantee schemes for small enterprises: an effective instrument to promote private sector-led growth? UNIDO, Programme Development and Technical Cooperation Division. Gropp, R., Gruendl, C., Guettler, A., 2013. The impact of public guarantees on bank risk-taking: Evidence from a natural experiment⁄. Review of Finance rft014. Gudger, M., 1998. Credit Guarantees: an assessment of the state of knowledge and new avenues of research. FAO Agricultural Services Bulletin. Honohan, P., 2010. Partial credit guarantees: Principles and practice. Journal of Financial Stability 6 (1), 1–9. Innes, R., 1991. Investment and government intervention in credit markets when there is asymmetric information. Journal of Public Economics 46 (3), 347–381. Jaramillo-Vallejo, J., Park, Y.C., Stiglitz, J.E., 1993. The role of the state in financial markets. World Bank Research Observer. Jiménez, G., Salas, V., Saurina, J., 2006. Determinants of collateral. Journal of Financial Economics 81 (2), 255–281. Khwaja, A.I., Mian, A., 2005. Do lenders favor politically connected firms? rent provision in an emerging financial market. The Quarterly Journal of Economics 120 (4), 1371–1411.

Kuniyoshi, S., Tsuruta, D., 2014. Adverse selection and moral hazard in the japanese public credit guarantee schemes for smes. CEPR’s Policy Portal. Manove, M., Padilla, A.J., Pagano, M., 2001. Collateral versus project screening: A model of lazy banks. RAND Journal of Economics, 726–744. Menkhoff, L., Neuberger, D., Suwanaporn, C., 2006. Collateral-based lending in emerging markets: Evidence from Thailand. Journal of Banking & Finance 30 (1), 1–21. Mhlanga, N., Miller, C., Zander, R., 2013. Credit Guarantee Systems for Agriculture and Rural Enterprise Development. Food and Agriculture Organization of the United Nations (FAO). Ongena, S., Smith, D.C., 2000. What determines the number of bank relationships? cross-country evidence. Journal of Financial Intermediation 9 (1), 26–56. Paravisini, D., 2008. Local bank financial constraints and firm access to external finance. The Journal of Finance 63 (5), 2161–2193. Rajan, R., Winton, A., 1995. Covenants and collateral as incentives to monitor. The Journal of Finance 50 (4), 1113–1146. Uesugi, I., Sakai, K., Yamashiro, G.M., 2010. The effectiveness of public credit guarantees in the japanese loan market. Journal of the Japanese and International Economies 24 (4), 457–480. Zia, B.H., 2008. Export incentives, financial constraints, and the (mis) allocation of credit: Micro-level evidence from subsidized export loans. Journal of Financial Economics 87 (2), 498–527.