Collateral requirements of SMEs: The evidence from less-developed countries Elmas Yaldız Hanedar a , Eleonora Broccardo a,b,⇑ , Flavio Bazzana a a Department of Economics and Management, University of Trento, Trento, Italy b Cefin (Centro Studi Banca e Finanza), Modena, Italy article info Article history: Received 29 October 2012 Accepted 27 September 2013 Available online 10 October 2013 JEL classification: G21 G32 O16 Keywords: Collateral SMEs Loan conditions Two-part models abstract This paper aims to investigate the determinants of collateral requirements for loans that are extended to small and medium-sized enterprises in less-developed countries. Our primary data source consists of the results from firms in Eastern Europe and Central Asia from the Business Environment and Enterprise Per- formance Survey, which is compiled by the World Bank and the European Bank for Reconstruction and Development. The results show that borrower-specific variables are more important than country-spe- cific variables in determining collateral requirements on loan contracts. The strongest evidence in our paper emphasises the importance of borrower risk and loan cost in collateral determinants. Ó 2013 Elsevier B.V. All rights reserved. 1. Introduction Borrowing difficulties can be one of the main obstacles to start- ing and running a business for small and medium-sized enterprises (SMEs), particularly in less-developed economies. Both the infor- mation asymmetry between lender (the bank) and borrower (the firm) (Baas and Schrooten, 2006) and the overall banking market structure (Petersen and Rajan, 2002; Berger and Udell, 2006) can influence lending terms and conditions. Pledging collateral is often an efficient solution to easing access to credit. Collateral require- ments are stringent in less-developed countries because the finan- cial environment in these countries typically involves opaque information and weak enforcement (Hainz, 2003; Menkhoff et al., 2006). In less-developed countries, borrowers have relatively low probabilities of holding collateralisable assets; thus, firms in these countries are more likely to experience difficulties in obtaining ac- cess to external financing (Menkhoff et al., 2006, 2012). Beck et al. (2006) use the World Business Environment Survey (WBES) to examine 12 financing obstacles. They report that collateral requirements are the third most important of these obstacles. The EBRD-World Bank Business Environment and Enterprise Per- formance Survey (BEEPS) results for firms in Eastern Europe and Central Asia indicate that high collateral requirements are the fourth most important reason firms do not apply for external loans; this factor ranked immediately below the issues of complex application processes and high interest rates in importance. 1 Therefore, collateralisation appears to be a crucial aspect of a firm’s access to external financing; this access can determine the eventual disappearance or survival of a firm. Due to poor data availability, however, there is a substantial lack of empirical evidence on collat- eral and its determinants in less-developed countries. Information about collateral for a cross-section of less-developed countries is even more difficult to obtain. Theoretical models addressing collater- alisation typically refer to financially developed economies, and their empirical verifications primarily use data from developed countries and largely focus on a single country. We aim to close this gap. An extensive body of research on developed countries 2 regards collateral as an efficient solution to the problems of information asymmetry with respect to the quality of borrowers (Berger et al., 2011a,b; Bester, 1987; Chan and Kanatas, 1985; Chan and Thakor, 0378-4266/$ - see front matter Ó 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jbankfin.2013.09.019 ⇑ Corresponding author at: Department of Economics and Management, University of Trento, Trento, Italy. Tel.: +39 0461 282104. E-mail addresses: [email protected](E. Yaldız Hanedar), eleonora.broccardo@ unitn.it (E. Broccardo), fl[email protected](F. Bazzana). 1 See Table A.1 of the Appendix A. 2 The wider literature on developed countries addresses many aspects of collat- eralisation. In this paper, we refer to only those studies on the role of collateral that are closely related to our analysis. Journal of Banking & Finance 38 (2014) 106–121 Contents lists available at ScienceDirect Journal of Banking & Finance journal homepage: www.elsevier.com/locate/jbf
Transcript
Journal of Banking & Finance 38 (2014) 106–121
Contents lists available at ScienceDirect
Journal of Banking & Finance
journal homepage: www.elsevier .com/locate / jbf
Collateral requirements of SMEs: The evidence from less-developedcountries
0378-4266/$ - see front matter � 2013 Elsevier B.V. All rights reserved.http://dx.doi.org/10.1016/j.jbankfin.2013.09.019
⇑ Corresponding author at: Department of Economics and Management,University of Trento, Trento, Italy. Tel.: +39 0461 282104.
This paper aims to investigate the determinants of collateral requirements for loans that are extended tosmall and medium-sized enterprises in less-developed countries. Our primary data source consists of theresults from firms in Eastern Europe and Central Asia from the Business Environment and Enterprise Per-formance Survey, which is compiled by the World Bank and the European Bank for Reconstruction andDevelopment. The results show that borrower-specific variables are more important than country-spe-cific variables in determining collateral requirements on loan contracts. The strongest evidence in ourpaper emphasises the importance of borrower risk and loan cost in collateral determinants.
� 2013 Elsevier B.V. All rights reserved.
1. Introduction
Borrowing difficulties can be one of the main obstacles to start-ing and running a business for small and medium-sized enterprises(SMEs), particularly in less-developed economies. Both the infor-mation asymmetry between lender (the bank) and borrower (thefirm) (Baas and Schrooten, 2006) and the overall banking marketstructure (Petersen and Rajan, 2002; Berger and Udell, 2006) caninfluence lending terms and conditions. Pledging collateral is oftenan efficient solution to easing access to credit. Collateral require-ments are stringent in less-developed countries because the finan-cial environment in these countries typically involves opaqueinformation and weak enforcement (Hainz, 2003; Menkhoff et al.,2006). In less-developed countries, borrowers have relatively lowprobabilities of holding collateralisable assets; thus, firms in thesecountries are more likely to experience difficulties in obtaining ac-cess to external financing (Menkhoff et al., 2006, 2012). Beck et al.(2006) use the World Business Environment Survey (WBES) toexamine 12 financing obstacles. They report that collateralrequirements are the third most important of these obstacles.
The EBRD-World Bank Business Environment and Enterprise Per-formance Survey (BEEPS) results for firms in Eastern Europe andCentral Asia indicate that high collateral requirements are thefourth most important reason firms do not apply for externalloans; this factor ranked immediately below the issues of complexapplication processes and high interest rates in importance.1
Therefore, collateralisation appears to be a crucial aspect of a firm’saccess to external financing; this access can determine the eventualdisappearance or survival of a firm. Due to poor data availability,however, there is a substantial lack of empirical evidence on collat-eral and its determinants in less-developed countries. Informationabout collateral for a cross-section of less-developed countries iseven more difficult to obtain. Theoretical models addressing collater-alisation typically refer to financially developed economies, and theirempirical verifications primarily use data from developed countriesand largely focus on a single country. We aim to close this gap.
An extensive body of research on developed countries2 regardscollateral as an efficient solution to the problems of informationasymmetry with respect to the quality of borrowers (Berger et al.,2011a,b; Bester, 1987; Chan and Kanatas, 1985; Chan and Thakor,
E. Yaldız Hanedar et al. / Journal of Banking & Finance 38 (2014) 106–121 107
1987; Coco, 1999). Another stream of literature investigates howbanking competition affects collateral requirements (Besanko andThakor, 1987; Hainz et al., 2013; Jiménez et al., 2011; Voordeckersand Steijvers, 2006). Because of the more serious information asym-metry and lower degree of bank intermediation in less-developedeconomies, the literature suggests that a further investigation of col-lateral requirements in these countries should be addressed. Re-search on less-developed countries is both scant and poor inquality, which makes it difficult to identify clear streams of litera-ture. As far as we know, this paper is the first to investigate thedeterminants of collateral with respect both to borrower risk andthe banking market structure by considering a sample consisting so-lely of less-developed economies.
To address this, we construct a cross-country sample consistingsolely of less-developed economies, including transition econo-mies from Eastern Europe and Central Asia. We draw on surveydata from the World Bank Business Environment and EnterprisePerformance Survey (BEEPS hereafter) and collect deep and widedata at both the firm and loan levels by focusing on a subsampleof SMEs. We investigate not only the presence of collateral but alsothe amount of collateral in loan contracts. The analysis specificallyinvestigates the importance of various firm- and country-specificfactors by testing (i) whether higher borrower quality reducesthe collateral-to-loan ratio; (ii) whether information sharingamong lenders affects collateralisation; and (iii) to what extentlending market and macroeconomic conditions affect the presenceof collateral in loan contracts. Our main contribution indicates thatcountry-specific variables are less important than borrower-spe-cific variables in determining collateral requirements. Accordingly,we find that loan variables and all the firm’s risk variables explainthe collateral determinants. We are unaware of a previous cross-country study on SMEs’ collateral determinants based solely on asample of less-developed countries. Thus, our paper yields new re-sults and important insights for businesses and policy makers thatoperate in these countries. The remainder of the paper is organisedas follows. Section 2 reviews the relevant literature on collateralwith regard to less developed countries. Section 3 introduces themodel employed in this study. Section 4 presents data and descrip-tive statistics. The results are discussed in Section 5, and Section 6concludes the paper.
2. Review of the literature on collateral in less developedeconomies
Caballero and Krishnamurthy (2001) is the first theoretical pa-per on collateral in emerging countries; however, the focus of theirmodel is specifically on financial crises in economies constrainedby limited international collateral. The model developed in Cabal-lero and Krishnamurthy (2001) indicates that firms in emergingcountries with limited domestic collateral and binding interna-tional collateral constraints systematically undervalue interna-tional collateral by taking actions that lead them to reduceinternational collateral during crises, which exacerbates the effectsof adverse shocks. Most of the studies addressing the role of collat-eral in emerging countries are empirical in nature. The scant avail-ability of information on both collateral and firms explains whymost of the empirical research consists of single-country analysesusing data that are mostly collected through surveys. One of thefirst empirical studies of developing economies is an investigationby Feder et al. (1988) that emphasises both the use of land collat-eral and credit availability in three developing rural countries(Thailand, India and Korea). The authors show that political, legaland social issues might influence the enforcement of land pledgedas collateral and affect the lending transaction. Where it is legal –in Thailand and in India – the extensive use of land collateral
reduces creditworthiness assessment costs. For Thailand, the avail-able data indicate that a credible threat of losing land makes collat-eral effective in accessing financing: pledging land collateralincreases the amount of credit offered by more than 40% comparedto loans without security. In the literature investigating how finan-cial reform affects both investment and credit allocation, Gelos andWerner (2002) explore the role of real estate used as a proxy forcollateralisable assets in enhancing investments following thefinancial liberalisation of the Mexican financial sector in late1988. Their main results show that the effect of collateral on cap-ital expenditures became more important after 1989. Moreover,after the financial liberalisation, banks appear to have continuedto rely heavily on collateral in their lending because of persistentinformational and enforcement problems. The primary effect offinancial liberalisation is not to lower the cost of credit but to in-crease its availability. In the debate over the effectiveness of rela-tionship lending in emerging economies, La Porta et al. (2003)examine the extent and the impact of related lending extendedby 17 Mexican banks in 1995. The so-called related borrowers con-sist of the bank’s shareholders, associates, family and firms con-trolled by the bank. The authors find that strong relationship-oriented loan transactions show a lower level of collateral and,simultaneously, higher default rates and lower recovery rates thanwith unrelated borrowers. Related lending (also called relationshiplending) in Mexico appears to be consistent with the pessimisticassessment – the so-called looting view – of such related lending.The authors suggest that scaling back on related lending might bethe best way to reduce the fragility of the financial system inemerging countries such as Mexico. Conversely, Menkhoff et al.(2006) find a positive relationship between collateralisation andrelationship banking. Their paper analyses credit transactions en-gaged in by nine Thai commercial banks during the 1992–1996period. In their analysis, relationship-oriented house banks de-mand higher collateral than non-house banks. This finding sug-gests that house banks appear more likely to be affected by ahold-up problem than by a reverse looting problem, as found byLa Porta et al. (2003) in Mexico. In a subsequent paper, Menkhoffet al. (2012) focus on the possible substitutes for collateral and findthat only 11% of households in northeastern Thailand are credit-constrained and that most borrow without pledging collateral. Tosolve this puzzle, they investigate how substitutes allow smallerfirms to access collateral-free financing. Their survey was con-ducted in 2007 and covered 2186 rural households. The resultsconfirm that the use of third-party guarantees decreases the prob-ability of collateral requirement by a factor of 0.54–0.57, whereasrelationship lending decreases the probability by only 0.07–0.10.Because even low-income households possess assets – typicallyland – that might serve as collateral suggests that ineffective col-lateral enforcement affects the lending conditions and explainsthe high number of loans with collateral substitutes. Thus far,the evidence indicates that empirical research on emerging coun-tries is narrow in its coverage, and most studies analyse a singlemarket. Three recent papers perform cross-country analyses onthe role of collateral in emerging markets. Although important re-sults are shown about collateral requirements in emerging mar-kets, we argue that these studies present some limitations withrespect to their observed sample. Liberti and Mian (2010) investi-gate how financial development affects collateral requirementsusing an original sample of loans extended to SMEs in 15 countries.The loans were extended by a single multinational bank, and thecost of financial underdevelopment is denoted in terms of collat-eral spreads measured by the difference in collateralisation ratesbetween high- and low-risk loans within the same economy. Theirresults show that an improvement in financial institutional devel-opment by one standard deviation reduces a country’s collateralspread by nearly one-half. The authors analyse a wide (and rare)
3 There is a clear dearth of substantial empirical support for the adverse-selectionypothesis. Although several studies support the role of collateral as a tool foritigating adverse-selection problems (Jiménez et al., 2006; Berger et al., 2011b),
ther investigations (Cressy and Toivanen, 2001) find evidence that risk and collateralre not significantly correlated. Instead, a positive relationship between collateral andan spread is consistently demonstrated, which confirms the observed-risk hypoth-
sis (Berger and Udell, 1990, 1995; Jiménez and Saurina, 2004; Gonas et al., 2004;hen, 2006; Menkhoff et al., 2006; Chakraborty and Hu, 2006; Brick and Palia, 2007).ithin this debate, some authors indicate that both hypotheses might be empiricallyconciled by examining the degree of information asymmetries in a country (Berger
t al., 2011a; Godlewski and Weill, 2011).
108 E. Yaldız Hanedar et al. / Journal of Banking & Finance 38 (2014) 106–121
set of data concerning collateral and banks’ ex-ante and ex-postrisk assessments related to borrowers. Because the data come froma single multinational bank, its specific lending policies might af-fect the results. However, as noted by the authors, this also allowsfor a more reliable comparison of the borrowers. Godlewski andWeill (2011) aim to test whether the degree of information asym-metry affects the relationship between collateral and the loanspread. Using Dealscan, they perform a cross-country analysis overa sample of 4940 large bank loans from 31 countries, almost half ofwhich are developed countries. They find a positive relationshipbetween the presence of collateral and the loan spread, thus vali-dating the observed-risk hypotheses. They also observe that thisrelationship may be (barely) negative in developing countries withstrong information asymmetries. As they suggest, further investi-gation is required on a sample composed only of emerging marketsin which information asymmetries may be severe. A limitation oftheir analysis is that large loans are less likely to be exposed toinformation asymmetry; however, this results from the absenceof a cross-country database for small loans in their cross-countryanalysis. To our knowledge, Hainz et al. (2013) is the first paperthat investigates the relationship between collateral and bankcompetition in a sample of several countries; however, fewer thanhalf are emerging countries. In a theoretical model, the authorsshow that firms located near a bank are best financed by screeningcontracts, whereas distant borrowers are required to pledge collat-eral. As bank competition increases, the proportion of firms fi-nanced by screening contracts increases, and the use of collateralconsequently decreases. By using Dealscan, these predictions areempirically tested in a cross-country analysis on a sample of4931 bank loans from 70 developed and emerging economies.Their results corroborate the theoretical prediction: collateral ismore likely to be required when bank competition is lower.
3. The model
We investigate how collateral requirements are related to firmcharacteristics and/or features of the credit market. We measurethe collateral requirement not only by the presence of collateralbut also by the collateral-to-loan ratio. With respect to firm char-acteristics, we analyse whether the risk profile of the borrowerpositively affects the collateral requirement (Hypothesis 1). Withrespect to market features, we investigate how information sharing(Hypothesis 2) and the concentration of the bank market (Hypoth-esis 3) affect collateral requirements. In the following, we describethe hypotheses that will be tested.
3.1. The hypotheses
H1. As the borrower risk increases, the presence of collateral inSME loan contracts becomes more likely, and collateral-to-loanratios will be higher for high-risk borrowers.
An extensive body of theoretical literature addresses collateralas a tool for resolving informational asymmetry problems regard-ing borrower quality in the context of either ex-ante adverse selec-tion or ex-post moral hazard (Stiglitz and Weiss, 1981; Bester,1987; Besanko and Thakor, 1987; Chan and Thakor, 1987). The ad-verse-selection hypothesis predicts that unobservable lower-riskborrowers will pledge more and better collateral than higher-riskborrowers because they have a lower likelihood of losing the col-lateral and because pledging the collateral is less costly. Nonethe-less, conventional wisdom suggests that when risk is observable,higher collateral requirements are most often associated withhigher-risk borrowers. Accordingly, the observed-risk hypothesispredicts that riskier borrowers will more likely be required to pro-vide collateral for loans to defray lender costs in the event of a
default. In a situation that involves hidden actions, collateral mayalign the interests of lender and borrower and act as a deterrentthat discourages the borrower from adopting ex-post unobservableopportunistic risk-shifting behaviour that can hinder the success ofthe project (Boot et al., 1991; Boot and Thakor, 1994). The crediblethreat of losing the pledged collateral disciplines the borrower’sactions by producing a higher level of effort to satisfy loan require-ments and therefore reduces the risk of the borrower’s default.3 Weexpect to find evidence of a positive relationship between the risksof SMEs and the collateral requirements to which these enterprisesare subjected, particularly given that SMEs typically display evenhigher perceived levels of risk in less-developed than in developedeconomies.
H2. The collateral requirements in SME loan contracts are lessrestrictive in countries that feature more intensive information-sharing mechanisms.
Information sharing among lenders allows banks to obtaininformation about the repayment histories and current debt expo-sure of loan applicants. Thus, information sharing is an importanttool for reducing informational asymmetries and eventuallydecreasing adverse-selection problems. Pagano and Jappelli(1993) demonstrate that information sharing increases the volumeof lending by easing loan conditions, particularly for situationsinvolving severe adverse-selection problems in the financial mar-kets. From an empirical perspective, Brown et al. (2009) reveal thatinformation sharing is associated with credit that is both moreavailable and less expensive for borrowers. How information-shar-ing mechanisms affect loan conditions – such as collateral require-ments – remains an open empirical question. In countries withweaker information-sharing mechanisms, lenders may experiencedifficulties measuring credit risk, particularly if they are unfamiliarwith the loan applicant prior to the loan application. Consequently,greater opaqueness regarding borrowers’ characteristics producesan increased probability of collateral requirements and a greateramount of collateral for any given loan. Therefore, if information-sharing mechanisms positively affect both financing access andloan conditions, we expect to find negative relationships betweeninformation sharing and collateral requirements (both the pres-ence of collateral in loans and the collateral-to-loan ratio).
H3a. Both the likelihood of the presence of collateral and thedegree of collateral in SME loan contracts are positively associatedwith banking concentration.
A stream of literature investigates the role of market competi-tion in collateralisation. One theoretical view asserts that bankcompetition may induce banks to focus more deeply on relation-ship lending, which may alleviate price competition pressures be-cause a client-driven lending system can help a bank become moreunique relative to its competitors (Boot and Thakor, 2000). Berlinand Butler (2002) demonstrate that as the competitive pressurein loan markets increases, lenders must relax contract terms, i.e.,lower their expected collateral ratios; thus, loan contracts becomeless stringent as competition increases. Voordeckers and Steijvers
hmoaloeCWree
E. Yaldız Hanedar et al. / Journal of Banking & Finance 38 (2014) 106–121 109
(2006) empirically demonstrate that if a company submits a creditrequest to more banks, the likelihood that the company will pledgeany type of collateral as part of its eventual loan diminishes. Hainzet al. (2013) indicate that collateral is more likely in loan contractsin less-competitive loan markets. Assuming that there is a negativeassociation between competition and concentration and if the rela-tionship argument prevails, we expect to find a positive relation-ship between the concentration of the credit markets and boththe presence of collateral in loans and the magnitude of the collat-eral-to-loan ratio.
H3b. Both the likelihood of the presence of collateral and thedegree of collateral in SME loan contracts are negatively associatedwith banking concentration.
A second theoretical view argues that as bank competition in-creases, the bank’s incentive to invest in information collectiondiminishes because the probability borrowers will switch to otherbanks also increases; under increasingly competitive conditions,therefore, a bank’s power to extract rent will be reduced, which in-creases the likelihood of the use of collateral (Besanko and Thakor,1987; Petersen and Rajan, 1995). In concentrated banking environ-ments, lenders possess an informational advantage over borrow-ers; this advantage produces collateral requirements that are lessstringent. Jiménez et al. (2006) find support for a negative relation-ship between the use of collateral and loan market concentrationand suggest that collateral and a bank’s market power appear tobe substitutes for one another. Assuming a negative association be-tween competition and concentration, if the rent extraction argu-ment prevails, we expect to find a negative relationship betweenthe concentration of the credit markets and both the presence ofcollateral in loans and the magnitude of the collateral-to-loan ratio.
6 Goldberger (1964) may be regarded as the first paper that addresses two-parmodels. However, Cragg (1971) is the first paper to use the term ‘two-part model’These models have been extensively used in consumption studies and healtheconomics research, particularly for situations involving cigarette and alcohoconsumption (Cragg, 1971; Jones, 1989; Yen and Jensen, 1996; Labeaga, 1999
3.2. Model specification and methodology
Most of the previous studies on collateralisation examine thepresence of collateral in loan contracts by using probit and logitmodels.4 However, these discrete choice models are unable to de-scribe the magnitude of collateral; for example, loan contracts with1% or over 100% collateral-to-loan value ratios are considered thesame and are coded identically. Only a few studies have examinedcollateral-to-loan ratios, and these investigations conventionallyuse Tobit models.5 The Tobit model in our study is:
y�icst ¼ x0ictsbþ uicst
uicst � Nð0;r2Þ
where y�icts is the latent variable representing the collateral-to-loanvalue ratio for firm i in country c from sector s and in year t, xicst
is the vector of independent variables and b is the correspondingvector of parameters to be estimated. Finally, uicst is the homosked-astic and normally distributed error term. As yicts is the actual collat-eral-to-loan-value ratio because this ratio cannot be negative. Thus,the relationship between y�icts and yicts is:
maxðy�icst;0Þ ¼ yicst
The loglikelihood function for the Tobit model is:
Log L ¼X
yicst¼0
ln 1� gx0icstbr
� �� �þX
yicst>0
ln1r
syicst � x0icstb
r
� �� �
4 Berger and Udell (1995), Degryse and van Cayseele (2000), Jiménez et al. (2009and Menkhoff et al. (2012).
5 Menkhoff et al. (2006) and Peltoniemi (2007).
Newman et al., 2003; Aristei et al., 2008; Madden, 2008). Two-part models are rarelyused in empirical finance studies. Dionne et al. (1996) use this model for crediscoring, and Moffatt (2005) employs this model for loan defaults. To the best of ourknowledge, two-part models have not yet been implemented in the empiricaliterature with respect to collateralisation.
)
where gð�Þ and sð�Þ are the standard normal cumulative distributionfunction and probability distribution function respectively.
In this paper, we argue that the Tobit model is restrictive be-cause of its assumptions. First, the maximum likelihood estimationfor the Tobit model assumes that errors are homoskedastic andpossess a normal distribution; if these assumptions are violated,the maximum likelihood estimator becomes inconsistent.Although there are several types of modified Tobit models (e.g.,the heteroskedasticity-robust Tobit estimator), Ramalho and Vidi-gal da Silva (2009) argue that none of these modifications producea single modified Tobit model that addresses all the issues with theTobit approach. Second, and more important to our study, is theassumption that the same data-generating process determinesboth the binary and continuous dependent variables, which in thisinstance are the presence of collateral and the collateral-to-loan ra-tio, respectively. We argue that the first decision is to impose ornot impose collateral on a loan contract, and the collateral-to-loanratio is decided next. Thus, the use of two-part model seems moreappropriate theoretically than the use of a single Tobit model.Wooldridge (2002, p.546) suggests a simple and informal way ofunderstanding the appropriateness of the Tobit model by estimat-ing a probit model over the dependent variable of the Tobit model.If the coefficient estimates from the probit model are close to thoseof the Tobit model divided by the estimated standard error of theTobit regression, then the Tobit model is not inappropriate. In addi-tion to this informal check, Greene (2000, p. 770) suggests a likeli-hood ratio test when reporting estimation results to compare theperformances of the two-part model and the Tobit model.
As originally formulated by Cragg (1971), double-hurdle or two-part models may be considered a generalised Tobit model.6 As thename ‘‘double-hurdle’’ suggests, the model from Cragg (1971) isbased on the assumption that households make two separate deci-sions about buying a durable good. In this model, each householdfirst decides whether to buy a durable good and subsequently deter-mines how much to spend on the purchase of the good in question.Thus, these decisions are determined by different data-generatingprocesses. As explained in Cragg (1971), to observe a positive levelof expenditure on a durable good, two separate hurdles must bepassed: the first hurdle is the participation decision (i.e., decidingwhether to buy the item), and the second is the consumption deci-sion (i.e., deciding how much to spend on the item). By using atwo-part model, we assume that implementing collateral or not ona loan contract and deciding on the value of the collateral are twoseparate contracting processes. Accordingly, these separate proce-dures (determining whether there is collateral in loan contractsand the degree of collateral measured by the collateral-to-loan-valueratio) should be modelled separately in two equations.
d�icts ¼ z0ictsaþ eicts
y��icts ¼ x0ictsbþ uicts
where d�icts is the latent variable that takes the value of 1 if the loancontract is secured by collateral and 0 otherwise, zicts is the vector ofexplanatory variables, and a is a vector of parameters to be esti-mated; y��icts represents the degree of collateral, xicts is the vector ofexplanatory variables, and b is a vector of parameters to be
t.
l;
t
l
Table 1Variable definitions and sources.
Variable Definition Source
CollateralColl1 Dummy = 1 if the firm has pledged collateral to obtain an external loan and zero otherwise BEEPSColl3 The ratio of collateral value to loan size (%) if Coll1 = 1 BEEPSColl2 The ratio of collateral value to loan size (%), including zeros BEEPS
Firm characteristicsLiq_risk The sales sold on credit over total sales (%) BEEPSOverdue Dummy = 1 if the firm has utility payments that are overdue by more than 90 days and zero otherwise BEEPSCrime Dummy = 1 if the SME has experienced any losses as a result of theft, robbery, vandalism or arson (and zero otherwise) BEEPSAge The number of years that the firm has been operating BEEPSSize The size of the firm, as measured by the number of full–time employees. BEEPSSoleown Dummy = 1 if the firm is owned by a sole owner and zero otherwise. BEEPSQuality Dummy = 1 if the firm has an internationally recognised quality certification, such as ISO 9000 or ISO 9002, and zero otherwise BEEPSCity Dummy = 1 if the firm is located in the capital or in a city with a population over one million and zero otherwise BEEPS
Information sharingInfo_sh Information sharing index that measures the presence and structure of public and private credit registries on a scale between 1 and 5 Brown et al.
(2009)
Lender market characteristicsCr The asset share of the three largest commercial banks within the commercial banking sector of the country as a measure of
concentration in the banking sector (%)Bankscope
Foreign The asset share of foreign banks in total banking system assets (%) EBRD
Macroeconomic variableLngdppc The logarithm of the GDP per capita in US dollars EBRD
This table presents variable definitions and the sources of study data. BEEPS stands for Business Environment and Enterprise Performance Survey, WB stands for the WorldBank, and EBRD stands for the European Bank for Reconstruction and Development.
110 E. Yaldız Hanedar et al. / Journal of Banking & Finance 38 (2014) 106–121
estimated. The two error terms ðeicts and uictsÞ are assumed to benormally and independently distributed. The first part is:
dicts ¼ 1 if d�icts > 0
dicts ¼ 0 if d�icts 6 0
the second part is:
y�icts ¼ maxðy��i ;0Þ
and the observed variable yicts is extracted from the equation:
yicts ¼ dictsy�icts
The log-likelihood function for the two-part model is:
Log L ¼X
yicst¼0
ln 1� gðz0icstaÞgx0icstbr
� �� �
þX
yicst>0
ln g z0icsta� � 1
rs
yicst � x0icstbr
� �� �
3.3. Variables
Following the two-part model, we first use a probit model to ex-plain the presence of collateral in loan contracts, which is ex-pressed by a dummy variable (COLL1). The information for thisdependent variable is extracted from the following question:‘‘Thinking of the most recent line of credit or loan you obtained froma financial institution, did the financing require collateral?’’ The COLL1variable takes the value of 1 if the firm reported a positive numberand 0 otherwise. Second, we use a truncated regression model toexplain the positive values of collateral-to-loan ratios by consider-ing only the firms that obtain a loan with collateral (COLL3).7 TheCOLL3 variables are extracted from the following question: ‘‘Referring
7 To simplify the notation, we refer to loan as ‘‘line of credit or loan’’ hereafter.
only to this most recent line of credit or loan, what was the approximatevalue of the collateral required as a percentage of the loan value?’’ TheCOLL3 variable measures the ratio of collateral value to loan size ifCOLL1 is equal to 1, i.e., only if the firm has pledged collateral.
To test the appropriateness of the two-part model, we use a To-bit model to explain the positive value of the collateral-to-loan ra-tio by including firms that obtained a loan without pledgingcollateral (COLL2). The COLL2 variable – which is extracted fromthe same question that extracts the COLL3 variable – measuresthe ratio of collateral value to loan size on the entire sample, thusincluding both the observations of firms that obtained a loan withcollateral (COLL1 equal to 1) and those of firms that obtained a loanwithout collateral (COLL1 equal to zero).
We model these firm-level dependent variables as functions ofborrower-specific and country-specific variables. To test ourhypotheses, we grouped the determinants of the presence of collat-eral in loan contracts and the collateral-to-loan ratios for thesecontracts into five categories. The first category refers to theloan-specific variables. The second refers to the firm-specific vari-ables to be tested in our first hypothesis. The third category relatesto information-sharing mechanisms and enables our secondhypothesis to be tested. The fourth category refers to the bankingmarket characteristics and is used to test the third hypothesis.8 Fi-nally, the fifth category includes LNGDPPC as a country-level macro-economic control variable. The definitions and sources of eachvariable are shown in Table 1.
The first group of variables includes two loan-specific factors:LOAN_DUR and LOAN_COST. LOAN_DUR measures the duration inmonths of the most recently obtained loan. A standard means toreduce information asymmetry is to shorten the duration of theloan and disincentivise the firm from engaging in risk-shiftingbehaviour. Thus, we expect short-term loans and collateral to besubstitutes for one another, and the coefficient estimation ofLOAN_DUR is, thus, expected to be positive. LOAN_COST is the
8 Because the majority of loans are borrowed from banks, we consider bankingctor characteristics to be a proxy for lending market characteristics.
se
E. Yaldız Hanedar et al. / Journal of Banking & Finance 38 (2014) 106–121 111
approximate value of the interest rate of the most recently ob-tained loan. According to the observed-risk hypothesis, we expectthat riskier firms are required to pledge more collateral and payhigher interest rates on loans. Therefore, we expect a positive signon LOAN_COST. With respect to firm-specific factors, we includethree proxies for borrower risk: LIQ_RISK, OVERDUE, and CRIME.LIQ_RISK measures the percentage of a firm’s sales that are soldon credit. This variable proxies the risk of the firm in terms of illi-quidity by assuming that the higher (lower) the incidence of salespaid on credit, the lower (higher) the availability of cash. We ex-pect that firms with a lower (higher) percentage of sales paid bycash tend to have smaller (larger) cash reserves and that this in-creases (mitigates) their risk assessment. OVERDUE is a dummyvariable that is set to 1 if a firm has utility payments that are over-due by more than 90 days and is 0 otherwise. CRIME is set to 1 ifthe SME experienced any losses due to crime, theft, or disorderin the previous year and is 0 otherwise. We expect lenders to beless willing to lend to a firm if the firm is less liquid, has unpaidutility bills and/or if the firm is located in environments with sig-nificant criminal activity. Accordingly, lenders are more likely toimplement more stringent loan conditions for these potentially ris-ky SMEs, including higher collateral requirements. Thus, we expecta positive association between our dependent variables and theLIQ_RISK, OVERDUE and CRIME variables. The remaining firm con-trol variables we use are as follows. AGE is the number of yearsthe firm has been operating. Older firms are more likely to havelonger relationship with lenders, as shown by Berger and Udell(1995). Thus, these more established firms may obtain loans withbetter conditions, i.e., lower interest rates and less collateral. Inour model, therefore, we expect to observe a negative coefficientfor AGE. SIZE represents the size of an examined firm and is mea-sured by the number of full-time employees of that firm.9 Becausewe expect that smaller firms are younger and have growing businessprospects – and are thus riskier – we expect a negative sign for SIZE.QUALITY is a dummy variable that is set to 1 if the firm has an inter-nationally recognised quality certification, such as ISO 9000 or ISO9002, and is 0 otherwise. Because higher values of this variable arereflective of higher borrower quality, we expect to find negativecoefficients for QUALITY. SOLEOWN is a dummy variable that has avalue of 1 if the SME is owned by a sole person and is 0 otherwise.Collateralisation becomes a less suitable tool if contracts cannot befully enforced. In those firms owned by more than a sole person,we expect the enforcement process to be weaker and collateralrequirements to be more stringent. We therefore expect a negativesign of SOLEOWN. Finally, because of transaction costs (for borrow-ers) and enforcement costs (for the bank), the locations of SMEsare important for determining the availability and cost of loans, inaddition to various terms of the loan contract, including collateral.CITY is a dummy variable that is set equal to 1 if the firm is locatedin a national capital or in a city with a population over 1,000,000inhabitants and is 0 otherwise. We expect loan contract conditionsto be less stringent in larger cities because financial centres are pri-marily located in such cities. Thus, we expect to find a negative asso-ciation between CITY and the dependent variables.10
To test for the effects of information-sharing mechanisms on
9 The number of employees might be depending on the industry a firm is operatingin and sales revenue could be another measure of size. However, we noted that salesare highly correlated with other variables of the model (such as LIQ_RISK andOVERDUE) and this affects the significance of our results. So, we believe that thenumber of full-time employees is a more appropriate measure of size in our analysisgiven the low correlation with the other variables of the model. However, weperformed a robustness check with revenues sales. As expected LIQ_RISK andOVERDUE lose significance, but the sign and significance of all other variables remainunchanged.
10 See Jiménez et al. (2009) for a discussion of the effect of location oncollateralisation.
11 Given that Brown et al. (2009) use average values, we contacted one of theauthors (Marco Pagano) to obtain the yearly information-sharing index.
12 The information sharing index by Brown et al. (2009) is the maximum of twoscores, one score for PCRs and one score for PCBs. These scores add 1 point forfulfilling each of the following five criteria: (i) both firms and individuals are covered(ii) positive and negative data are collected and distributed; (iii) the registrydistributed data that are at least two years old; (iv) the threshold for included loans isbelow per capita GDP; and (v) the registry has existed for more than three years.
13 In the robustness checks, we also consider a variable to include the presence ostate-owned banks. We use the shares of the total banking system assets that areowned by state-owned banks (STATE, expressed in terms of percentages), which dataare available only for 2005.
,
collateral requirements, we use the information-sharing index (IN-FO_SH) proposed and estimated by Brown et al. (2009).11 For eachcountry over the 1996–2005 period, the INFO_SH index measures thepresence and structure of public credit registries (PCRs) and privatecredit bureaus (PCBs) on a scale of 1–5.12 The higher the informa-tion-sharing index, the more developed the public credit registriesor private credit bureaus. As highlighted by the authors, this indexis similar to the ‘‘Credit Information Index’’ that is reported in the‘‘Doing Business’’ data of the World Bank. However, the ‘‘Credit Infor-mation Index’’ has been released by the World Bank only from 2004onwards. Because we expect that the level of information sharing ina country in year t � 1 affects financing terms in year t, we use thevalue of the information-sharing index for 2001 and 2004 for theanalysis of the 2002 BEEPS and the 2005 BEEPS, respectively. Brownet al. (2009) show that information sharing is associated with im-proved availability and lower cost of credit. Indeed, we aim to testwhether information-sharing mechanisms also mitigate collateralrequirements (expected negative sign of INFO_SH) or if the informa-tion-sharing mechanisms have a major impact on the decision to ex-tend the loan but not on the pledging of collateral (expected positivesign of INFO_SH).
To test our third hypothesis, we use country-level variables thatprovide information about the structure of the banking system. Weuse CR, which is the share of all commercial bank assets that areowned by the three largest commercial banks, to measure concen-tration in the lending market. To control for differences in owner-ship structure in the lending markets of the examined countries,we use the FOREIGN variable.13 Because foreign banks frequentlyface difficulties in accessing and evaluating subjective informationabout borrowers, they primarily use objective information andstandardised decision techniques in their lending decisions, whereasdomestic banks are more apt to use soft information and long-termrelationships (Berger and Udell, 1995; Berger et al., 2001; Petersenand Rajan, 2002). We use the shares of the total banking system as-sets that are owned by foreign banks (expressed in terms of percent-ages) as measures of the foreign–domestic ownership structure inlending markets; we expect a positive coefficient for FOREIGN to de-scribe the relationship of this variable to our dependent variables.
Finally, to control for macroeconomic conditions we useLNGDPPC, which represents the natural logarithm of the per capitagross domestic product. As LNGDPPC increases, we expect the pres-ence of collateral to decrease as the result of the possible occur-rence of credit expansion and the implementation of less-stringent loan conditions, which might lead to lower collateral-to-loan ratios and decreased collateralisation.
4. Data and descriptive statistics
The primary data set that is used in this study is provided by theBEEPS, which is a joint project of the European Bank for Recon-struction and Development (EBRD) and the World Bank (WB).The BEEPS is administered throughout 27 transition economiesfrom Eastern Europe and Central Asia (including Turkey) to assessthe business environments for private enterprises in the examined
;
f
Table 2The data sample for the different survey years.
112 E. Yaldız Hanedar et al. / Journal of Banking & Finance 38 (2014) 106–121
countries. The BEEPS have been conducted on an irregular basissince 2002 and have significantly changed over the years. Afterthe BEEPS 2005, several pieces of financial information are notavailable: among others, loan characteristics and firms’ externaldebt. Because collateralisation could also be determined by loan-specific factors our analysis is based on the pooled cross-sectiondata from the 2002 and 2005 surveys, only. BEEPS 2002 and2005 provide data on 6667 and 9656 firms, respectively.14 We ar-gue that these data sets possess a number of advantages comparedwith those used in previous studies. First, the data include firms inboth rural areas and large cities. Second, the dataset includes dataon SMEs and allows for fine-tuned size specification by classifyingfirms into medium-sized, small and micro categories.15 Third, theBEEPS collect a wide variety of information about collateral by inves-tigating both the presence of collateral and the collateral-to-loan va-lue. For our final sample of SMEs to be in accordance with bothBEEPS definitions and OECD conventions, we define SMEs to be firmsthat have a maximum of 250 full-time employees. Thus, we arrive ata total sample of 14,394 SMEs (see Table 2).16
Among these 14,349 SMEs, 5705 had obtained an externalloan, and 4728 (82.8%) had agreed to loan contracts that in-cluded collateral. In accordance with the BEEPS classifications,we define medium-sized firms as firms that have fewer than250 and more than 49 full-time employees, and small firms asthose with fewer than 50 and more than 9 full-time employees.In addition, we define micro firms as those with fewer than 10full-time employees. Among these 5705 SMEs that obtained aloan, 1582 are medium-sized firms, and 1350 (85.3%) had agreedto loan contracts that included collateral. Of the 2298 smallfirms that obtained a loan, 1972 (85.8%) had agreed to loan con-tracts that included collateral. Of the 1825 micro firms that ob-tained a loan, 2370 (77%) had agreed to loan contracts thatincluded collateral (see Table 3).
The average SME’s collateral-to-loan ratio (for the loan con-tracts that included collateral – COLL3) was 152.8%, with an aver-age standard deviation of nearly 79.7%. Of the loan contracts thatincluded collateral (COLL1 = 1), 68% required a quantity of collat-eral that was greater than the value of the loan (that is,COLL3 > 100).
Our data indicate that buildings, machinery and equipment arethe most-preferred types of collateral, whereas land is a secondarychoice for collateral in loan contracts.17 Table 3 provides detailedsummary statistics concerning the variables used in the empiricalanalysis.
Table 4 presents the summary statistics at the country level.Countries are sorted in descending order based on their averageCOLL1 values. The mean value for the presence of collateral is thelowest in Slovenia, where approximately half of commercial loansare secured by collateral. Albania is the country with the highestcollateralisation; 95% of the examined loans were secured by col-lateral. Georgia ranked first with respect to the degree of collateralrequired for loans, with an average collateral-to-loan ratio (as mea-sured by COLL3) of 223%. Among the examined countries, Turkeyhas the lowest collateral-to-loan ratio mean value, with COLL3equal to 82%.
Table 5 presents summary statistics for collateralisation in dif-ferent country groups. We grouped countries according to four re-gions: the European Union (EU), non-European Union (NON-EU),Commonwealth of Independent States (CIS), and Central and
14 See Appendix A for more information on the BEEPS sampling.15 By contrast, previous cross-country studies using the DealScan dataset focus on
large corporate loans (Godlewski and Weill, 2011; Hainz et al., 2013).16 The BEEPS definition of enterprise size is as follows: small = 2–49 employees,
medium = 50–249 employees, large = 250–9999 employees.17 See Table A.2 in the Appendix A.
Eastern Europe (CEE). This table reveals no major differencesamong the examined country groups with respect to the mean val-ues of collateralisation on loans that are extended to SMEs. We ob-serve that, compared with NON-EU countries, the EU countries inour sample have slightly lower collateralisation with respect toboth the presence of collateral and the collateral-to-loan ratio. Inall the assessed sub-groups of countries, we observe that microfirms demonstrate the lowest mean value for the presence of col-lateral (COLL1), and this mean value is similar for small and med-ium-sized firms.
5. Estimation results
5.1. Baseline results
Most of the previous studies on SMEs evaluate all SMEs as a sin-gle group of firms and do not distinguish micro, small and med-ium-sized firms. However, the determinants of collateralrequirements for these groups of firms may differ, so we performnot only regressions for the full sample of SMEs but also separateregressions for small, medium-sized, and micro firms.
Table 6 reports the estimation results. For all the examinedgroups of firms, we first provide the two-part model results. Theremaining column reports the estimations of the Tobit model. Inthe first column of the two-part model, we provide the probit mod-el estimation results for the probability of the presence of collateralin loan contracts (COLL1).18 In the second column, we report thetruncated regression results for COLL3. The average variance inflationfactor for the dependent variables is calculated as 1.24, which indi-cates the absence of multicollinearity. As we perform a standardlikelihood ratio test to assess the applicability of the two-part modelagainst the Tobit approach, we note that the Tobit model is too
18 We conducted logit models, and the results of these models were similar to thefindings previously presented.
Table 3Summary statistics.
Variable SMEs Medium firms Small firms Micro firms
# Mean Std. dev. Min Max # Mean Std. dev. Min Max # Mean Std. dev. Min Max # Mean Std. dev. Min Max
This table reports summary statistics for the pooled 2002–2005 sample.
E. Yaldız Hanedar et al. / Journal of Banking & Finance 38 (2014) 106–121 113
restrictive because of its assumptions.19 Following this statisticalevidence, we restrict our discussion of the results to the two-partmodel only (COLL1 and COLL3). We also underline that the Tobitanalysis includes those firm observations that receive a loan withoutcollateral (COLL1 = 0), which most likely consist mainly of lines ofcredit that normally do not require collateral. For this reason, wedecided that the truncated regression is the appropriate model toestimate the collateral-to-loan value by considering only those loanswith collateral.
The major finding consists of the positive and significant coeffi-cients for loan-specific and firm-specific characteristics about thepresence of collateral (COLL1). With regard to size specifications,these results are consistent for medium-sized firms. Conversely,the collateral-to-loan value (COLL3) is unaffected by loan charac-teristics and borrower riskiness, as shown by the insignificant coef-ficients for all variables in the truncated regression for all the firms’sizes. The only variable that yields the expected negative and sig-nificant coefficients for all specifications is CITY: all things beingequal, those firms located in larger cities in which financial centresare located are less likely to be required to pledge collateral with ahigh collateral-to-loan value.
With respect to the first hypothesis, the results show the expectedand significant coefficients for all borrower’s risk variables con-cerning COLL1. LIQ_RISK and OVERDUE have a positive and signifi-cant impact on the presence of collateral for SMEs and medium-sized firms. For CRIME, the observed positive sign is not valid formicro enterprises. However, the riskiness of the firm seems to af-fect whether collateral must be pledged but not in the degree ofthe collateral-to-loan value, as shown by the truncated regressionresults (COLL3) for all firm sizes.
19 The likelihood ratio statistic is calculated as LR ¼ �2½LTobit� ðLProbitþLTRUNÞ� � v2
k , where LTobit is the likelihood of the Tobit model; LProbit is thelikelihood of the probit model; LTRUN is the likelihood of the truncated regressionmodel; and k is the number of independent variables in the equations. Theformulation of the null hypothesis indicates that the Tobit model is an appropriatemodelling strategy to explain zero collateralisation; this null hypothesis is rejected inour regressions.
Turning to the firm-level control variables, AGE affects COLL3positively for the total sample of SMEs. This result is contradictoryto our predictions that loan contracts for older firms are less likelyto include greater collateral-to-loan value. However, AGE is insig-nificant in all size specifications. Our estimation results yield posi-tive coefficient estimates for the effect of SIZE on COLL1; this resultis more significant for micro firms. However, this relationship be-comes negative if COLL3 is regarded as the dependent variable. Thisfinding can be explained by the fact that smaller SMEs often lackcollateralisable assets and therefore apply for loans that do not re-quire collateral, such as loans from informal creditors or frommicrofinance institutions. When smaller SMEs obtain a collatera-lised loan, the collateral-to-loan ratio becomes higher as firm sizedecreases. By contrast, micro-sized firms have insignificant coeffi-cient estimates for SIZE. Examining the ownership structure ofSMEs, we note that the presence of collateral in loan contracts(COLL1) is less likely for SMEs that are established as sole propri-etorships than for SMEs that are corporations. SOLEOWN is signifi-cant for the entire SME sample and for micro firms. One possibleexplanation for this negative and significant relationship is thatthe enforcement process for firms owned by one person is ex-pected to be less costly and more effective. This leads to less strin-gent collateral requirements, as confirmed by the results. Theeffect of QUALITY is almost insignificant for all size specifications,both for COLL1 and COLL3. Finally, compared with SMEs locatedin smaller cities, SMEs located in capitals and/or large cities are lesslikely to obtain loans that require collateral (COLL1), and they ben-efit from lower collateral-to-loan ratios (COLL3). This effect isstronger for medium-sized firms. Regarding small and micro firms,the presence of collateral in loans (COLL1) is not dependent on thelocation of the firm; however, being located in large cities affectsthe collateral-to-loan value. In accordance with our expectations,this result demonstrates that the collateral requirements for SMEsare less stringent in capitals and large cities with a population ofover 1,000,000 inhabitants. This result may be explained by thefact that financial centres are typically located in these cities; thus,it is easier to switch to other lenders and search for loans that do
Table 4Summary statistics by country.
Country Statist. Coll1 Coll3 Coll2 Loan_dur Loan_cost Liq_risk Overdue Crime Age Size Soleown Quality City Info_sh Cr Foreign Lngdppc
This table reports summary statistics for the variables by country. The listed countries are ranked in descending order according to the mean value of COLL1.
E.YaldızH
anedaret
al./Journalof
Banking&
Finance38
(2014)106–
121115
Table 5Collateralisation in different country groups.
Region Var. SMEs Medium firms Small firms Micro firms
# Mean Std. dev. Min Max # Mean Std. dev. Min Max # Mean Std. dev. Min Max # Mean Std. dev. Min Max
This table presents summary statistics for collateralisation among different country groups. EU stands for European Union, CIS stands for Commonwealth of IndependentStates, and CEE stands for Central and Eastern Europe.
116 E. Yaldız Hanedar et al. / Journal of Banking & Finance 38 (2014) 106–121
not require collateral when the loan terms offered by one lenderbecome stricter.20
Overall, our results provide partial support for our first hypoth-esis (H1) regarding the presence of collateral. In fact, the positiveand significant coefficients for all borrower risk variables are fur-ther strengthened by the positive and significant coefficient ofLOAN_COST. Thus far, the research has provided some support forthe observed-risk hypothesis mainly by investigating the relation-ship between pledging collateral and the loan spread (Berger andUdell, 1990; Brick and Palia, 2007; Godlewski and Weill, 2011).In particular, on a sample of both developed and less-developedcountries, Godlewski and Weill (2011) observe that the relation-ship between the presence of collateral and the loan spread is po-sitive, but it tends to become weaker in developing countries inwhich information asymmetry is greater. The authors interpret thisresult as support for the adverse-selection hypothesis and as a sig-nificant motivation for further investigation into less-developedcountries. Our results show, by contrast, that the relationship be-tween collateralisation and the loan’s cost also remains positiveand significant for less-developed countries. Moreover, the positiveand significant value of the borrower’s riskiness variables corrobo-rates the observed-risk hypothesis. In less-developed countries, asthe risk of borrowers increases, the presence of collateral in SMEloan contracts becomes more likely.
With respect to the second hypothesis, INFO_SH is positive andsignificant for the presence of collateral (COLL1) in the SME sam-ple and in small and micro firms, whereas it yields an insignifi-cant coefficient for the degree of collateral (COLL3). Moreintensive information-sharing mechanisms do not seem to miti-gate the presence of or the degree of collateral. Instead, in coun-tries with more developed information-sharing mechanisms(higher levels of INFO_SH, which reflects a more complete andlonger track record of information on PCRs and PCBs), the bor-rower is more likely to be asked to pledge collateral on the loan.This result is more prevalent for small and micro firms. To inter-pret this apparently unexpected result, it must be underlined thatwe observe firms that previously obtained a loan and, therefore,have positively passed a credit evaluation by a bank. Therefore,
20 We created additional firm-specific control variables to verify our results. Inparticular, we used dummy variables to test the effects of being an innovative orexporting firm; however, we did not observe significant effects from these factors.
we decide to investigate whether information-sharing mecha-nisms positively affect the decision to extend the loan. In theBEEPS, firms were asked how problematic access to financing(as determined by collateral requirements and credit availability)is for their business. Similar to Brown et al. (2009), we code an-swers to these questions as a dummy (1 = major obstacle,0 = moderate, minor or no obstacles), and our dependent variableis coded as Probability to have a loan (PR_LOAN). We run a Probitregression over the independent variables, except for LOAN_DURand LOAN_COST. We obtain a negative and significant coefficientfor INFO_SH. Similar to Brown et al. (2009), this result thus showsthat credit access is less of a constraint for firms that are locatedin countries with better information-sharing mechanisms. How-ever, once the firms obtain the loan, our analysis of collateraldeterminants shows that the level of information sharing doesnot mitigate collateral requirements. We therefore reject H2.While information-sharing mechanisms are associated with im-proved availability of credit, the collateral requirements in SMEloan contracts are not less restrictive in countries that featuremore intensive information-sharing mechanisms.
With respect to the third hypothesis, we find no evidence indi-cating that banking concentration has a negative or positive im-pact on the presence (COLL1) and the degree of collateral(COLL3) for SMEs and all size specifications. Our results are con-sistent with those of Berger et al. (2011b), who demonstrate thatlending market concentration – which they use as a control var-iable – does not have a significant effect on the presence of col-lateral in loan contracts. With respect to the asset share offoreign banks in total banking system assets, which could be asignal of competition in the credit market, we find a positiveand significant coefficient of FOREIGN for COLL1 for the SMEs. Inthe size specification, this result is significant only for microfirms. The collateral-to-loan value (COLL3), however, is unaffectedby the presence of foreign banks. This result may be explained bythe fact that foreign banks engage in the intensive use of objec-tive information and standardised decision techniques in theirlending decisions because it is difficult for them to accuratelyevaluate subjective information about borrowers.
With respect to country-level control variables, our estimationresults demonstrate a negative and statistically significant associa-tion between LNGDPPC and collateralisation. This result revealsthat improvements in macroeconomic conditions help ease loancontract terms by relaxing collateral requirements.
Table 6Estimation results.
Variables SMEs Medium firms Small firms Micro firms
Two-part model Tobit Two-part model Tobit Two-part model Tobit Two-part model Tobit
This table presents the probit, and truncated regressions for coll1, and coll3, respectively, and Tobit regression results for coll2. All regressions include sector and year fixed effects. Due to a lack of observations, Tajikistan, Turkeyand Uzbekistan are excluded from these regressions. Robust standard errors are reported in parentheses. The Chi-square critical value v2(24) is 36.42 at 5%.* Statistical significance at the 5% level.** Statistical significance at the 1% levely.*** Statistical significance at the 0.1%.
E.YaldızH
anedaret
al./Journalof
Banking&
Finance38
(2014)106–
121117
Table A.1Why do firms not apply for new loans?
Main reason for not applying for a new loan # ofSMEs
Percentage
No need for a loan – the firm has sufficient capital 6300 69.65Interest rates are not favourable 1086 12.01Application procedures for loans or lines of credit are
complex496 5.48
Collateral requirements are too high 435 4.81Did not think that the loan would be approved 177 1.96The size or maturity times of available loans are
insufficient99 1.09
It is necessary to make informal payments to obtainbank loans
62 0.69
Other 256 2.83Don’t know 134 1.48
Total 9045 100
The data above represent the author’s calculations using 2005 BEEPS data.
Table A.2Forms of collateral in loans that are granted to SMEs.
Type of collateral 2002 2005
Land owned by the borrowing firm 721 343Buildings owned by the borrowing firm 1789 960Machinery 543Machinery and equipment, including movables 1010Accounts receivable 164Inventories 373Personal houses of an owner of the SME 279Personal assets (house, etc.) of an owner of the SME 585Other forms of collateral 244 378
This table presents the number of collateral types that were required. Thesenumbers are in accordance with Niinimäki (2009), indicating that real estate is themost common and dominant form of collateral across the examined countries. Thetable values reflect the author’s calculations using BEEPS data.
1 This type of consideration is in accordance with the approach of La Porta et al.997), a study that addresses the legal origins of countries as a source of difference in
nancial sectors and firm structures among countries. The countries in our sampleature three different legal origins: French, German, and socialist. Albania, Lithuania,omania, and Turkey adopted French law; Bosnia and Herzegovina, Bulgaria, Croatia,e Czech Republic, Hungary, Latvia, Macedonia FYR, Poland, Serbia and Montenegro,e Slovak Republik, and Slovenia adopted German law; and the remaining countries
dopted socialist laws.2 Due to lack of observations, we could not perform a robustness check for the CISuntries.
118 E. Yaldız Hanedar et al. / Journal of Banking & Finance 38 (2014) 106–121
5.2. Model robustness checks
In a first round of robustness checks, we changed the definitionsof the borrower risk proxies. We first used a categorical variable inplace of CRIME to account for the effect of risk that arises from thelocation of the SME. This variable was set the following values: ifcrime, theft and disorder are no obstacle to the current operationsof the firm, it was set at 0; it was set at 1 if these factors are a minorobstacle; it was set at 2 if they are a moderate obstacle; it was setat 3 if they are a major obstacle; or it was set at 4 if they are a se-vere obstacle. Although this new proxy of borrower risk is not sig-nificant, the signs and values of all the other variables do notchange. Second, we replaced the utility arrears with tax arrears.Our results remained unchanged.
In a second robustness check round, we added different controlvariables.
First, we considered the effect of the legal environment byexamining an index from the Doing Business project of the WorldBank that measures the strength of legal rights in a country. Thisindex ranges from 0 to 10, with higher scores indicating that collat-eral and bankruptcy laws protect the rights of borrowers and lend-ers and facilitate lending because better laws expand access tocredit. For the countries in our sample, the mean value for this var-iable is 5.85 (the median is 6), and the standard deviation is 2.17.This variable assumes its highest value for Latvia (a value of 10)and its lowest value for Uzbekistan (a value of 2). Our previousestimation results remain unchanged. The coefficient estimatesfor the legal rights index do not produce significant results forthe presence of collateral, as measured by COLL1; however, this in-dex has a statistically significant negative effect on the degree of
collateral in loan contracts, as measured by COLL3. As argued byBrown et al. (2009), better legal protection makes loan contractseasier to enforce and facilitates the issuance of a larger numberof loan contracts, and this legal protection may lead to lower col-lateral-to-loan ratios.
Second, to control for the effect of legal origin, we used a set ofdummy variables for the origin of the countries’ legal system(French, German, or Socialist).21 Our previous estimation results re-mained unchanged. These dummy variables yield an insignificantcoefficient for only for COLL1.
Third, because our sample consists of different country regions,we performed separate regressions for EU, non-EU, and CEE coun-tries.22 In the EU sample, almost all coefficients remain unchanged,except for CITY and INFO_SH, which became insignificant. We arguethat EU countries denote a higher and more homogeneous financialdevelopment compared to NON-EU countries; therefore, the locationof the firm does not affect collateral requirements. Regarding NON-EU countries, we observe a near loss of significance for the borrow-ers’ risk variables and the FOREIGN variable. For the CEE countries,the results remain unchanged.
Finally, we considered a variable that includes the presence ofstate-owned banks. We use the shares of the total banking systemassets that are owned by state-owned banks (STATE, expressed interms of percentages), which are available only for 2005. The re-sults remain unchanged. We find that STATE has a significant effectof reducing only the collateral-to-loan ratios. As Berger and Udell(2006) indicate, state-owned lenders frequently use governmentsupport in the form of subsidies to supply additional credit to SMEsand satisfy political purposes by easing the collateral requirementsfor loans.
5.3. Methodological robustness check
We also check for robustness regarding potential sample selec-tion bias. Menkhoff et al. (2012) and Chakraborty and Hu (2006)use the Heckman selection model to model the presence of collat-eral in loan contracts for which the selection equation is a loan ap-proval equation. Our paper differs from Menkhoff et al. (2012) andChakraborty and Hu (2006) in two methodological aspects. First,we are interested not only in the presence of collateralisation inloan contracts but also the degree of collateralisation. Second,our population of interest is SMEs with loans rather than thosewithout loans because we are interested in actual collateralisationand not in potential or latent collateralisation. Under these circum-stances, double-hurdle models are more appropriate for achievingthe purposes of our study. To address the potential bias due tosample selection, we use the Heckman selection model (Heckman,1979) following Menkhoff et al. (2012) and Chakraborty and Hu(2006). The dependent variable of the first stage is a dummy vari-able with a value of 1 if the firm has obtained a loan recently and 0if the firm does not have a loan and reported financial constraintsas a major obstacle to the operation of their businesses. We runthis first-stage regression over the independent variables, exceptfor LOAN_DUR and LOAN_COST. The dependent variable of the sec-ond-stage regression is COLL1. The lambda is �0.061 and is statis-tically insignificant, while the correlation coefficient between the
2
(1fifeRththa
2
co
Table A.3Pairwise correlation coefficients.
Coll1 Coll3 Coll2 Loan_dur Loan_cost Liquidity Overdue Crime Age Size Soleown Quality City Info_sh Cr Foreign Lgdppc
E. Yaldız Hanedar et al. / Journal of Banking & Finance 38 (2014) 106–121 119
regressions in which COLL1 is the dependent variable is only�0.182. We run another Heckman model in which the COLL2 isthe dependent variable of the second-stage regression and obtaina lambda that is �61.703 and statistically significant at 10%. More-over, the coefficient between the first and the second stages is�0.623. Relying on these results, we can conclude that sampleselection is not a serious problem (see Table A.3).
In our main model, if a firm is present in both 2002 and 2005,we consider two separate observations. Although there were only463 twice-interviewed firms, we perform the analysis as a robust-ness check by maintaining one firm observation and excluding theobservation from 2002. The results remain unchanged.
6. Conclusion
The objective of our paper is to investigate the determinants ofcollateral requirements on loans extended to SMEs in less-devel-oped countries through the examination of borrower- and coun-try-specific variables using data from the EBRD-World BankBusiness Environment and Enterprise Performance Survey(BEEPS). We focus both on the presence of collateral in loan con-tracts, and on the amount of collateral required for these con-tracts. Our analysis assesses borrower characteristics, i.e., loan-and firm-specific variables, and country-specific factors, i.e., infor-mation-sharing mechanisms and banking market variables, bothof which affect collateral requirements. Our results provide robustevidence that the presence of collateral in loan contracts is deter-mined mostly by the borrower’s characteristics. In particular, thefirm-risk and loan variables seem to play the most important role.A firm granted a loan with higher duration and cost, with lowerliquidity, with overdue utilities payments and in a crime-riddenarea has a greater probability of being required to pledge collat-eral. These risk variables affect the presence of collateral onlyfor medium-sized firms. The probability of pledging collateralcan be reduced if the firm is located in a larger city instead of arural area or if the firm is owned by a single person. The resultson the degree of collateral have less significant variables than onthe incidence. It seems that after a bank decides to grant the loanto a firm with collateral, the magnitude of the collateral follows acompetition process among banks. In fact, only the size and loca-tion of the firm negatively affect the degree of collateral. Largerfirms located in a larger city seem to have the contractual powerto mitigate the degree of collateral required. Only this location ef-fect is consistent for all firm sizes.
Regarding country-specific factors, we show that althoughinformation-sharing mechanisms are associated with improvedavailability of credit, the collateral requirements in SME loan con-tracts are not less restrictive in countries that feature more
intensive information-sharing mechanisms. This effect is evidentonly for small and micro firms. We find no evidence indicating thatbanking concentration has a negative or positive impact on thepresence and degree of collateral for SMEs through all size specifi-cations, which is consistent with Berger et al. (2011b). However,the presence of foreign banks in the market makes firms morelikely to pledge collateral.
Our study yields some implications for policy makers regardingthe role of mutual guarantee societies (MGSs). Because MGSs offerguarantees that are effective in mitigating the risks of banks whenthey are compliant with the Basel II guidelines, it appears thatenhancing the role of MGSs may be important to mitigating boththe presence, and the degree of collateral for SMEs. Regardingthe presence, because MGSs perform a risk analysis to grant guar-antees to the SME, this evaluation can produce a signalling effectfor the bank that can therefore reduce the asymmetric information.Consequently, the probability to pledge of collateral will be consis-tent with the risk of the firm. Regarding the degree of collateral, aBasel II compliant guarantee can improve SME financing condi-tions, in term of lower collateral requirements.
Acknowledgements
We would like to thank the anonymous referee. We also thankOliver Hart, Marco Pagano, Marco Bigelli, Sonia Silva, and the par-ticipants in the 2012 ADEIMF Annual Conference, the 2012 4thInternational IFABS Conference, and the 2012 7th Portuguese Fi-nance Network Conference for their comments on earlier versionsof this paper. We gratefully acknowledge financial support fromADEIMF. Any errors in the manuscript are ours.
Appendix A
The primary data set used in our paper is provided by the EBRD-World Bank Business Environment and Enterprise PerformanceSurvey (BEEPS), which is a joint project of the European Bank forReconstruction and Development (EBRD) and the World Bank.The survey was first taken in 1999–2000, when approximately4000 firms in 26 countries of Eastern Europe and Central Asia(including Turkey) were interviewed. In 2002, the survey wasadministered to approximately 6500 firms in 27 countries (includ-ing Turkey, but excluding Turkmenistan). In 2005, the survey in-cluded approximately 9500 enterprises in 28 countries. To set abenchmark for the transition countries, a survey was also con-ducted in 2004–2005 in Germany, Greece, Portugal, South Korea,Vietnam, Ireland and Spain. In 2008–2009, the survey coveredapproximately 11,800 enterprises in 29 countries (including, for
120 E. Yaldız Hanedar et al. / Journal of Banking & Finance 38 (2014) 106–121
the first time, Mongolia). In our analysis, we use the data from thesurveys conducted in 2002 and 2005 (see Table A.4).
The sample was designed to be representative of the number offirms within the industry and service sectors for each countryaccording to sector, size and location. Targeted distributional crite-ria were also defined to maintain a sufficient weight in the tail ofthe distribution. For example, at least 10% of the sample was smallfirms, and 10% was large firms. The interviews were conducted byphone or face-to-face, and the average success rate was approxi-mately 30% during the different years. Regarding the contents ofthe survey, the questions are grouped in the following principalareas: general information, infrastructure and services, sales andsupplies, degree of competition, innovation, capacity, inspections,certificates, land and permits, crime, finance, business-governmentrelations, labour, business environment and performance. For moredetailed information regarding the stratification of regions and foradditional features related to the sampling process, see http://www.ebrd.com/pages/research/economics/data/beeps.shtml.
References
Aristei, D., Perali, F., Pieroni, L., 2008. Cohort, age and time effects in alcoholconsumption by Italian households: a double-hurdle approach. EmpiricalEconomics 35, 29–61.
Baas, T., Schrooten, M., 2006. Relationship banking and SMEs: a theoretical analysis.Small Business Economics 27, 127–137.
Beck, T., Demirguc-Kunt, A., Laeven, L., Maksimovic, V., 2006. The determinants offinancing obstacles. Journal of International Money and Finance 25, 932–952.
Berger, A.N., Udell, G.F., 1990. Collateral, loan quality, and bank risk. Journal ofMonetary Economics 25, 21–42.
Berger, A.N., Udell, G.F., 1995. Relationship lending and lines of credit in small firmfinance. Journal of Business 68, 351–381.
Berger, A.N., Udell, G.F., 2006. A more complete conceptual framework for SMEfinance. Journal of Banking and Finance 30, 2945–2966.
Berger, A.N., Klapper, L.F., Udell, G.F., 2001. The ability of banks to lend toinformationally opaque small businesses. Journal of Banking and Finance 25,2127–2167.
Berger, A.N., Frame, W.S., Ioannidou, V., 2011a. Tests of ex ante versus ex posttheories of collateral using private and public information. Journal of FinancialEconomics 100, 85–97.
Berger, A.N., Espinosa-Vega, M.A., Frame, W.S., Miller, N.H., 2011b. Why doborrowers pledge collateral? New empirical evidence on the role ofasymmetric information. Journal of Financial Intermediation 20, 55–70.
Berlin, M., Butler, A.W., 2002. Collateral and Competition. Working Paper, FederalReserve Bank of Philadelphia.
Besanko, D., Thakor, A.V., 1987. Collateral and rationing: sorting equilibria inmonopolistic and competitive credit markets. International Economic Review28, 671–689.
Bester, H., 1987. The role of collateral in credit markets with imperfect information.European Economic Review 31, 887–899.
Boot, A.W.A., Thakor, A.V., 1994. Moral hazard and secured lending in an infinitelyrepeated credit market game. International Economic Review 35, 899–920.
Boot, A.W.A., Thakor, A.V., 2000. Can relationship banking survive competition? TheJournal of Finance 55, 679–713.
Boot, A.W.A., Thakor, A.V., Udell, G.F., 1991. Secured lending and default risk:equilibrium analysis, policy implications and empirical results. The EconomicJournal 101, 458–472.
Brown, M., Jappelli, T., Pagano, M., 2009. Information sharing and credit: firm-levelevidence from transition countries. Journal of Financial Intermediation 18, 151–172.
Brick, I.E., Palia, D., 2007. Evidence of jointness in the terms of relationship lending.Journal of Financial Intermediation 16, 452–476.
Caballero, R.J., Krishnamurthy, A., 2001. International and domestic collateralconstraints in a model of emerging market crises. Journal of MonetaryEconomics 48, 513–548.
Chakraborty, A., Hu, C.X., 2006. Lending relationships in line-of-credit and nonline-of-credit loans: evidence from collateral use in small business. Journal ofFinancial Intermediation 15, 86–107.
Chan, Y.S., Kanatas, G., 1985. Asymmetric valuations and the role of collateral inloan agreements. Journal of Money, Credit and Banking 17, 84–95.
Chan, Y.S., Thakor, A.V., 1987. Collateral and competitive equilibria with moralhazard and private information. The Journal of Finance 42, 345–363.
Chen, Y., 2006. Collateral, loan guarantees, and the lenders’ incentives to resolvefinancial distress. The Quarterly Review of Economics and Finance 46, 1–15.
Coco, G., 1999. Collateral, heterogeneity in risk attitude and the credit marketequilibrium. European Economic Review 43, 559–574.
Cragg, J., 1971. Some statistical models for limited dependent variables withapplication to the demand for durable goods. Econometrica 39, 829–844.
Cressy, R., Toivanen, O., 2001. Is there adverse selection in the credit market?Venture Capital 3, 215–238.
Degryse, H., Van Cayseele, P., 2000. Relationship lending within a bank-basedsystem: evidence from European small business data. Journal of FinancialIntermediation 9, 90–109.
Dionne, G., Artis, M., Guillen, M., 1996. Count data models for a credit scoringsystem. Journal of Empirical Finance 3, 303–325.
Feder, G., Tongroj, O., Tejaswi, R., 1988. Collateral, guaranties and rural credit indeveloping countries: evidence from Asia. Agricultural Economics 2, 231–245.
Godlewski, C.J., Weill, L., 2011. Does collateral help mitigate adverse selection? Across-country analysis. Journal of Financial Services Research 40, 49–78.
Goldberger, A.S., 1964. Econometric Theory. John Wiley, New York.Gonas, J., Highfield, M.J., Mullineaux, D.J., 2004. When are commercial loans
River, New Jersey.Hainz, C., 2003. Bank competition and credit markets in transition economies.
Journal of Comparative Economics 31, 223–245.Hainz, C., Weill, L., Godlewski, C.J., 2013. Bank competition and collateral: theory
and evidence. Journal of Financial Services Research 44, 131–148.Heckman, J., 1979. Sample selection bias as a specification error. Econometrica 47,
53–161.Jiménez, G., Salas, V., Saurina, J., 2006. Determinants of collateral. Journal of
Financial Economics 81, 255–281.Jiménez, G., Salas, V., Saurina, J., 2009. Organizational distance and use of collateral
for business loans. Journal of Banking and Finance 33, 234–243.Jiménez, G., Salas, V., Saurina, J., 2011. The effects of formal and informal
contracting in credit availability. Journal of Money, Credit and Banking 43,109–132.
Jiménez, G., Saurina, J., 2004. Collateral, type of lender and relationship banking asdeterminants of credit risk. Journal of Banking and Finance 28, 2191–2212.
Jones, A.M., 1989. A double-hurdle model of cigarette consumption. Journal ofApplied Econometrics 4, 23–39.
Labeaga, J.H.M., 1999. A double-hurdle rational addiction model withheterogeneity: estimating the demand for tobacco. Journal of Econometrics93, 49–72.
La Porta, R., López-de-Silanes, F., Shleifer, A., Vishny, R., 1997. Legal determinants ofexternal finance. The Journal of Finance 52, 1131–1150.
La Porta, R., López-de-Silanes, F., Zamarripa, G., 2003. Related lending. The QuarterlyJournal of Economics 118, 231–268.
E. Yaldız Hanedar et al. / Journal of Banking & Finance 38 (2014) 106–121 121
Madden, D., 2008. Sample selection versus two-part models revisited: the case offemale smoking and drinking. Journal of Health Economics 27, 300–307.
Menkhoff, L., Neuberger, D., Rungruxsirivorn, O., 2012. Collateral and its substitutesin emerging markets’ lending. Journal of Banking and Finance 36, 817–834.
Menkhoff, L., Neuberger, D., Suwanaporn, C., 2006. Collateral-based lending inemerging markets: evidence from Thailand. Journal of Banking and Finance 30,1–21.
Moffatt, P.G., 2005. Hurdle models of loan default. The Journal of the OperationalResearch Society 56, 1063–1071.
Newman, C., Henchion, M., Matthews, A., 2003. A double-hurdle model of Irishhousehold expenditure on prepared meals. Applied Economics 35, 1053–1061.
Niinimäki, J.-P., 2009. Does collateral fuel moral hazard in banking? Journal ofBanking and Finance 33, 514–521.
Pagano, M., Jappelli, T., 1993. Information sharing in credit markets. The Journal ofFinance 43, 1693–1718.
Petersen, M.A., Rajan, R.G., 1995. The effect of credit market competition on lendingrelationships. The Quarterly Journal of Economics 110, 407–443.
Petersen, M.A., Rajan, R.G., 2002. Does distance still matter? The informationrevolution in small business lending. The Journal of Finance 57, 2533–2570.
Peltoniemi, J., 2007. Collateral Requirements and relationship banking: empiricalevidence from unique Finnish credit-files. The Finnish Journal of BusinessEconomics 11, 393–414.
Ramalho, J.J.S., Vidigal da Silva, J., 2009. A two-part fractional regression model forthe financial leverage decisions of micro, small, medium and large. QuantitativeFinance 9, 621–636.
Stiglitz, J.E., Weiss, A., 1981. Credit rationing in markets with imperfect information.The American Economic Review 71, 393–410.
Voordeckers, W., Steijvers, T., 2006. Business collateral and personal commitmentsin SME lending. Journal of Banking and Finance 30, 3067–3086.
Wooldridge, J.M., 2002. Intoductory Econometrics: A Modern Approach. South-Western College Publishing, Boston.
Yen, S.T., Jensen, H.H., 1996. Determinants of household expenditures on alcohol.Journal of Consumer Affairs 30, 48–67.
