Access to credit, natural disasters, and relationship lending

J. Finan. Intermediation 21 (2012) 549–568

Contents lists available at SciVerse ScienceDirect

J. Finan. Intermediation j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / j fi

Access to credit, natural disasters, and relationship lending Gunhild Berg a,⇑,1, Jan Schrader b,2 a b

The World Bank, Africa Finance and Private Sector Development, 1818 H Street, NW, Washington, DC 20433, USA KfW Development Bank, Palmengartenstr. 5-9, 60325 Frankfurt, Germany

a r t i c l e

i n f o

Article history: Received 17 December 2009 Available online 1 June 2012 Keywords: Credit availability SME finance Natural disasters Relationship lending Developing economies

a b s t r a c t This paper analyzes the effect of unpredictable aggregate shocks on loan demand and access to credit by combining client-level information from an Ecuadorian microfinance institution with geophysical data on natural disasters, more specifically volcanic eruptions. The results of this ‘natural experiment’ show that while credit demand increases due to volcanic activity, access to credit is restricted. Yet, we also find that bank-borrower relationships can lower these lending restrictions and that clients who are known to the institution are about equally likely to receive loans after volcanic eruptions occurred. Ó 2012 Elsevier Inc. All rights reserved.

1. Introduction What determines access to credit in developing economies? And how does credit availability change in times of unpredictable aggregate shocks? These are the questions we address in this paper by making use of a unique dataset combining client-level information from a self-sustainable Ecuadorian microfinance institution (MFI) with geophysical data on natural disasters, more specifically volcanic eruptions. How easily and at what costs firms in emerging and developing economies receive access to external finance depends on a wide range of factors both internal and external to the firm. External factors are mainly connected to the institutional and regulatory environment. Important aspects include, among others, the enforceability of contracts, the protection of property rights, and the availability of credit registries (Beck et al., 2004, 2008; Brown et al., 2009). With respect to factors internal to a ⇑ Corresponding author. Fax: +1 202 522 1198. E-mail address: [email protected] (G. Berg). The findings, interpretations, and conclusions are entirely those of the authors and do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. 2 The views expressed in this paper are entirely those of the authors and do no necessarily represent those of KfW. 1

1042-9573/$ - see front matter Ó 2012 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.jfi.2012.05.003

550

G. Berg, J. Schrader / J. Finan. Intermediation 21 (2012) 549–568

firm, not only is the availability of own financial resources and collateral important, but also ’soft’ information on the potential borrower. Whether banks base their lending decision mainly on ’hard’ quantitative data such as financial accounts, denoted usually as transaction lending, or more on qualitative information, which is mainly referred to as relationship lending, depends on the types of clients and the environment in which the banks operate (Boot, 2000; Boot and Thakor, 2000). For instance, if clients cannot offer high collateral or if public information about the creditworthiness of potential borrowers is limited due to asymmetric information, relationship lending may be the only suitable option (Sharpe, 1990; Berger and Udell, 1995). Indeed, in such environments, Petersen and Rajan (1995) have shown that relationship lending may increase access to credit especially for small firms. Until today, no analysis has been undertaken regarding the question whether and to what extent access to credit changes after unpredictable aggregate shocks such as natural disasters. While Del Ninno et al. (2003) analyze credit demand after a major flood in Bangladesh in 1998 and find that the demand for loans increases, they cannot answer the question whether lending was restricted for some borrowers as they do not have bank-internal information. Apart from this study, there exists a wide literature on the potential micro-effects of credit-constraints arguing that access to credit can dilute the negative effects of aggregate shocks on consumption and similar outcomes (Jacoby and Skoufias, 1997; Beegle et al., 2006; Gitter and Barham, 2007; Khandker, 2007; Shoji, 2010; Becchetti and Castriota, 2011). However, for measuring access to credit, mostly proxy variables are used as the data that would be needed for analyzing the loan approval decision of a bank is usually not available. Khwaja and Mian (2008) analyze the effect of unforeseen liquidity shocks on access to credit and find that banks experiencing larger drops in liquidity are more restrictive in terms of lending afterwards and that smaller firms are less able to compensate this loss through additional borrowing. While this study shows that banks pass their liquidity shocks on to firms, they are not analyzing the question in which way a destruction of the capital stock of a firm affects the lending decision of a bank. With this paper, we attempt to close this gap by analyzing a unique dataset including information on all loan applications and subsequent approvals for ProCredit Ecuador between January 2002 and August 2007. As we combine this data with information on volcanic eruptions in Ecuador during the same time, we are able to exploit a ’natural experiment’ allowing us to clearly identify the determinants of access to credit in response to unpredictable aggregate shocks. In that regard our paper is also related to an emerging literature studying loan applications, most recently as affected by the financial crisis (Puri et al., 2010; Berg and Kirschenmann, 2012; Jimenez et al., 2012, among others). In order to come up with testable propositions for our empirical analysis, we present a simple theoretical model based on relationship lending that highlights the effect of volcanic eruptions on credit demand and access to credit. The concept of relationship lending best reflects the focus of the MFI we study and the setting of a developing country characterized by high degrees of asymmetric information. The predictions of our model are the following: First, after unforeseen aggregate shocks, the number of loans demanded will increase. Second, due to the higher risk involved, the fraction of credit applicants actually receiving a loan after aggregate shocks will decrease. And third, as the bank will attempt to differentiate between lower- and higher-risk clients, heterogeneous effects will be observable in a way that repeat clients will face less lending restrictions compared to new credit applicants who are unknown to the institution. Within our empirical analysis we first analyze the demand for loans and in a second step access to credit. The results show that credit demand increases significantly after volcanic eruptions which implies that there exists a need for additional financing after shocks occurred. When analyzing access to credit on the level of the individual credit applicant, the results suggest that high volcanic activity in the last months before the credit application leads to significant decreases in the probability to be approved for a loan. Yet, we also find that being a repeat client lowers these lending restrictions. More specifically, the results show that returning clients do not only have a higher probability to receive a loan in general, but that they are about equally likely to be approved for a loan after volcanic shocks occurred. Given that repeat clients are less risky due to the fact that the MFI has gathered relevant information on them during previous interactions, they face lower lending restrictions. New customers, however, are unknown to the institution and thus, in times of crises, it becomes even more difficult for them to receive financing, a result which can be directly associated with asymmetric information. The results are of particular interest as it is usually assumed that the existence of credit markets

G. Berg, J. Schrader / J. Finan. Intermediation 21 (2012) 549–568

551

can dilute the effects of shocks, for instance through the provision of emergency loans (Eswaran and Kotwal, 1989; Morduch, 1995). However, this presumption leaves the perspective of the financial institution aside which has to ensure its profitability. The results are robust to using different indicators for volcanic activity, i.e. volcanic eruptions or the seismic activity of the volcano, and alternative indicators for relationship. More precisely, while we focus our main results on the differences between new and repeat clients, we also use the number of previous loans as an indicator for the depth of the relationship. Furthermore, in order to analyze whether our results also hold for a sector highly affected by volcanic activity, we take a closer look at the agricultural sector and show that our main findings remain unchanged. To confirm that the effects depend on the proximity to the volcano, we further analyze the impact of volcanic activity in a region which is unlikely to be affected by volcanic eruptions and show that, indeed, no significant effects on access to credit can be found, further validating our main results. Finally, as far as the generalizability of our results is concerned, the findings should hold for comparable banks operating in settings with high degrees of asymmetric information and facing unpredictable aggregate shocks. The remainder of the paper is organized as follows. In Section 2, we present a theoretical model on the effects of aggregate shocks on the demand for loans and access to credit. Section 3 discusses the institution, the context, and the data we use and provides some descriptive statistics. The econometric models employed for testing the theoretical propositions are presented in Section 4. Section 5 is concerned with the empirical results. Finally, Section 6 closes the argument. 2. Theoretical framework Since we are analyzing a self-sustainable, semi profit-oriented MFI operating in a developing country characterized by high degrees of asymmetric information and potential clients who are often lacking collateral, we decided to base our theoretical framework on relationship lending. Considering the fact that the MFI we study is explicitly aiming at establishing long-lasting relationships with its clients in order to overcome the problems of asymmetric information, this concept seems to be appropriate in our setting. Relationship lending can be defined as a long-term implicit contract between a bank and its client (Boot, 2000). Due to repeated interaction with the borrower over time, the bank accumulates private information especially about his quality and creditworthiness, thus establishing close ties between the bank and its borrower (Elsas, 2005).3 Let us assume that we are in a risk neutral world. The population of entrepreneurs is normalized to 1. They start out with different amounts of capital An .4 The distribution of assets across firms is described by the cumulative distribution function GðAÞ indicating the fraction of firms with assets less R1 than A. Correspondingly, the total amount of firm capital is K ¼ 0 GðAÞdA. The demanded loan amount L0n is determined by the difference between the investment level I and the capital stock An . If An < I, a firm needs at least L0n ¼ I  An in external funds for being able to invest. The total fraction of entrepreneurs N in need of external financing is:

N ¼ GðIÞ

ð1Þ

Accordingly, the total loan amount demanded by the entrepreneurs is given by

L¼

Z

I

ðGðIÞI  GðAÞÞdA

ð2Þ

0

We model the mechanism of relationship lending as follows. There exist two types of entrepreneurs: good and bad entrepreneurs.5 A good entrepreneur has to borrow L0n for being able to invest in a safe project with a return of S1n in t ¼ 1. When the project ends, he will be able to invest I1n in another safe 3 Relationship lending can be distinguished from asset based lending where the lending decision of a bank depends on the value of the clients’ collateral, and transaction lending where the lending decision is based on credit scoring (Von Pischke, 2002). 4 We model the heterogeneity of clients’ asset levels along Holmstroem and Tirole (1997). 5 This part relies on the seminal paper by Petersen and Rajan (1995). The bad entrepreneurs are the incompetent, lazy, and dishonest in the group of potential entrepreneurs. The incompetent invest in bad projects and consequently waste the investment, the lazy do not put effort in their work and the dishonest steal the money or extract too many private benefits from the firm.

552

G. Berg, J. Schrader / J. Finan. Intermediation 21 (2012) 549–568

project with a return of S2n in t ¼ 2.6 In contrast, the projects of bad entrepreneurs who also have to borrow L0n will always fail and yield a return of zero at t ¼ 1. We assume that the revenue from the safe project is insufficient for financing the project in t ¼ 1: S1n < I1n . Also, safe projects have a positive net present value for each entrepreneur:

S1n þ S2n  I  I1n > 0

ð3Þ

Relationship banks are the only source of external finance in this market. Only agents know whether they are good or bad entrepreneurs. At t ¼ 0, the bank only knows that with a probability of h, an agent is a good entrepreneur. Thus, h is a measure of the ex ante credit quality of the agents. Through repeated interaction, the banks are able to accumulate soft information about the borrowers’ quality. We assume that at t ¼ 1, the bank becomes fully informed about the type of agent with whom it is dealing. The bank can charge an interest rate from the entrepreneurs in t ¼ 1 such that its expected return on loans is equal to R. We assume that R > 1. The banks’ cost of funds is zero since we are in a risk-neutral world. Both, the bank and the entrepreneur will only agree on a lending contract if the expected revenues are positive. From these participation constraints, the lowest credit quality h can be derived that the bank is able to finance without making losses given the initial loan amount I  An , the revenues of the projects and the investment in t ¼ 1:

hn ¼

I  An S1n þ S2n  An  I1n

ð4Þ

2.1. Aggregate shocks Aggregate shocks such as natural disasters can have various effects on the people and firms operating in their surroundings (Dercon, 2002). More precisely, a disaster can affect assets (direct damages), the flow for the production of goods and services (indirect losses), and the performance of the main macroeconomic aggregates of the affected country (macroeconomic effects). Direct damages such as the destruction of physical infrastructure usually occur at the moment of the disaster while the latter two types of losses can extend over a period of up to 5 yr (Economic Commission for Latin America and the Caribbean (ECLAC), 2003). The actual impact of a disaster depends, of course, on the nature of the shock and the economic sector affected. In the case of volcanic eruptions, direct effects on the manufacturing sector are unlikely since volcanoes are usually located in rural areas. The impact on the agricultural sector depends on the magnitude of the eruption. While ash falls and toxic gases cause only temporary damage in the case of small eruptions, full production recovery might be impossible after strong volcanic activity. Furthermore, if physical infrastructure such as roads is damaged, indirect effects on other sectors such as the commercial, transportation or tourism sector are possible as well. We assume that the shock occurs before new borrowers ask for a loan from the bank in t ¼ 0. An aggregate shock may be reflected in a reduction of the entrepreneurs’ capital An in the amount of DAn (direct damage), a reduction of the entrepreneurs’ revenue in the next period S1n in the amount of DS1n (indirect losses) or a reduction of the entrepreneurs’ revenue in later periods S2n in the amount of DS2n (macroeconomic effects). We assume that the effects of the shock DAn , DS1n and DS2n are not that high that entrepreneurs would no longer want to undertake the project and that the liquidity of the bank is not affected by withdrawals of savings or extensive default by affected clients.7 From our model, the following propositions can be derived. 6 In Petersen and Rajan (1995), good entrepreneurs also have the possibility to switch to a risky project in the first period. Since our main interest is not to analyze the effect of competition on relationship banking, we present a simplified version of their model abstracting from the risk shifting problems. However, even extending the model would not change our main results. 7 In the disaster management literature, (Pantoja, 2002) proposes the latter as an alternative channel. More specifically, bank capital could be insufficient to satisfy aggregate loan demand completely as, first of all, total loan demand after the shock Ls will rise. Second, it is possible that clients will withdraw their savings or save less. If these demands occur simultaneously, they will result in liquidity shortfalls for unprepared banks, especially if the client pool is not well diversified (Miamidian et al., 2005). However, this channel is unlikely to play a major role as long as the shock remains on a regional level and only affects some branches of the institution, which is usually the case for natural disasters.

G. Berg, J. Schrader / J. Finan. Intermediation 21 (2012) 549–568

553

Proposition 1. After aggregate shocks, the number of loans demanded increases.

Proof. A reduction of the entrepreneurs’ capital in the amount of DAn will shift the distribution function GðAÞ to the left. Consequently, for more entrepreneurs it will hold that An < I which, from Eq. (1), will result in a higher N. Therefore, the number of entrepreneurs in need of external finance will increase after shocks occurred. h Since some entrepreneurs who were able to finance their project completely with their own funds before the shock occurred ðAn > IÞ will now have to borrow as well ðAn  DAn < IÞ, the total loan amount applied for after the shock Ls will also increase. However, it is not clear whether the average loan amount applied for will rise as well since this will depend on the magnitude of the shock and the types of entrepreneurs affected.8 Proposition 2. After aggregate shocks, the fraction of credit applicants receiving a loan decreases.

Proof. An aggregate shock can be reflected in a reduction of An , S1n or S2n .9 From Eq. (4) it follows that all derivatives are <0 since it has to hold that I > An and that S1n þ S2n  I  I1n > 0. Otherwise, borrowing would either not be necessary or the entrepreneurs would not invest due to a negative net present value of the project (see Eq. (3)). Therefore, it follows that if the revenues of the entrepreneurs in t ¼ 1, respectively in t ¼ 2, go down, hn will rise. Then the bank has to charge a lower return rate R to ensure that the firm remains profitable. Yet, a lower return rate limits the ability of the bank to recover the losses from the first period. If an entrepreneurs’ asset level An is partially destroyed, hn will increase as well. The reason is that a destruction of the entrepreneurs’ assets increases the loan amount and therefore, the bank will have higher expected losses in the first period. Thus, the bank has to increase lending standards in order to obtain a client pool with a higher average credit quality hn (An , S1n , S2n ). Consequently, as the credit quality of the loan applicants h remains unaffected by the shock, the need for higher lending standards will result in a refusal of loan applications and especially low quality firms and those strongly affected by the shock will lose access to finance. h Proposition 3. After aggregate shocks, repeat borrowers face no lending constraints contrary to new credit applicants. Proof. Since it is assumed in the model that the bank becomes fully aware of the type of client at t ¼ 1, all borrowers who have repaid their first loan and ask for a second loan are good entrepreneurs by assumption. Thus, it is certain that repeat borrowers will repay their follow-up loans implying that it will always be profitable for the bank to finance their projects as long as they have a positive net present value.10 Since the costs of screening – the defaulted loans from period t ¼ 0 – can be considered sunk costs in t ¼ 1, the bank will always finance repeat borrowers in this model framework. This implies that contrary to new credit applicants, repeat clients will face no lending restrictions after aggregate shocks occurred. h Summing up, our theoretical analysis suggests that the demand for loans will increase after aggregate shocks and that relationship banks will restrict lending to new clients while repeat customers will still have access to finance. 8 If many entrepreneurs who had no need of external finance before the shock ðAn > IÞ are only slightly affected, but it holds that An  DAn < I, the average loan amount could even go down. 9 It is not very likely that the investment scale I is also altered by the shock since we assume that the shock will only temporarily affect the economy. 10 We assume that repeat and new clients are equally affected by the shock. Even though repeat clients might be wealthier and more successful than new clients and could therefore be losing more assets in the case of an eruption, they will also be able to insure themselves better against risks.

554

G. Berg, J. Schrader / J. Finan. Intermediation 21 (2012) 549–568

3. Description of the data For our analyses we use data from Banco ProCredit Ecuador from January 2002 to August 2007. The data was generated using the Management Information System (MIS) of ProCredit Ecuador and provides detailed information on all loan applications and subsequent approvals for all branches of the bank including comprehensive information on loan amounts and maturities (both demanded and approved), interest rates, and background information on the clients. ProCredit Ecuador was founded in October 2001, under the name Sociedad Financiera Ecuatorial, and received a full banking license in 2005. The bank is part of ProCredit Group which consists of 22 banks operating in transition economies and developing countries in Eastern Europe, Latin America and Africa and is led by ProCredit Holding AG, a holding company based in Germany. The shareholders of ProCredit Holding are development oriented.11 The group focuses on providing financing for small and medium sized enterprises (SMEs) and follows a development banking approach based on financial institution building. The group directs its lending toward lower income clients while covering costs and producing moderate profits at the same time. As ProCredit Ecuador is part of ProCredit Group with the majority shareholder ProCredit Holding (93.3% of the shares),12 it benefits from the synergies of this globally operating group, not only in terms of human resources and risk management but also with respect to refinancing. While ProCredit Ecuador has received various refinancing loans from public and private international financial institutions, being part of ProCredit Group has also ensured that the banks’ refinancing position was always strong. The mission of ProCredit Ecuador is to focus on lending to micro, small and medium sized enterprises and to offer a wide range of banking products as a development-oriented full-service bank. The primary aim when the bank was founded was to address the credit demand of micro and small businesses in a socially responsible way while also achieving a sustainable return on investment. However, as many self-sustainable microfinance institutions, the bank is not explicitly targeting the poorest of the poor, but supports those entrepreneurs with good business plans. The bank places great emphasis on the evaluation of the client’s debt capacity and willingness to repay so as to ensure that clients do not become overindebted. This approach is reflected in the high portfolio quality of the bank (arrears >30 days below 1.5%). The bank also explicitly aims at establishing lasting relationships with its clients and can therefore be considered a relationship bank as defined above.13 At the end of 2007, ProCredit Ecuador was operating 25 branches throughout the country and had granted approximately 58,000 loans with a total amount of $166 million.14 In order to analyze the effects of aggregate shocks on credit demand and access to credit we combine the data from ProCredit Ecuador with monthly data on the eruptions and seismic activity of Ecuador’s most active volcano Tungurahua provided by the Instituto Geofı´sico Ecuador. Ecuador is a country that has historically been strongly affected by natural disasters such as earthquakes and volcanic eruptions. Sitting atop five tectonic plates, the whole region of Latin America and the Caribbean is prone to intense seismic activity. Regarding active volcanoes, Ecuador has the second largest number in the region after Chile (Charvériat, 2000). Within Ecuador, volcanoes can be found in the Andean and in the jungle region while the coastal region differs considerably in terms of landscape, economic structure, and culture compared to the other two regions (Thoumi, 1990). As the distance to the volcanoes is also quite large in the coastal region, this area is usually unaffected by volcanic activity. The last severe eruption of Tungurahua took place in summer 2006. Even though the eruptive process started already in 1999 and various smaller outbreaks were recorded, the 2006 eruption was the most severe since the last significant period of activity from 1916 to 1925.15 During 11 ProCredit Holding is a public–private partership. The private shareholders are IPC GmbH, ipc-invest, the Dutch DOEN Foundation, the US pension fund TIAA-CREF, the US Omidyar-Tufts Microfinance Fund and the Swiss investment fund responsAbility. The public shareholders include KfW, IFC, FMO and BIO. 12 The remaining 6.7% of the shares are held by the Dutch DOEN Foundation. 13 In the Ecuadorian banking system, many banks use relationship lending as a lending technique in order to serve micro and small enterprises. Yet, in difference to ProCredit, they also use transaction lending at the same time. See Schrader (2009) for a classification of banks with respect to lending techniques in Ecuador. 14 See http://www.bancoprocredit.com.ec and http://www.procredit-holding.com for more information. 15 See the website of the Instituto Geofı´sico Ecuador (http://www.igepn.edu.ec) for more information.

555

G. Berg, J. Schrader / J. Finan. Intermediation 21 (2012) 549–568 Table 1 Summary statistics of credit applicants.

Demographic characteristics Male (%) Average age (yr) Married (%) Purpose of loan Agriculture (%) Business/trade (%) Livestock/fish breeding (%) Production/construction (%) Transportation (%) Tourism (%) No. applications No. approvals

Ambato and Riobamba

Coastal region

68.20 39.10 76.30

58.46 40.04 45.34

31.50 28.85 4.05 15.03 9.75 10.80 48,736 35,543

1.78 62.20 1.20 9.71 10.29 14.45 50,767 33,108

the eruption, pyroclastic flows went downhill threatening various smaller communities located at the base of the volcano. The 10 km high eruptive column was blown west and covered vast areas of the two provinces closest to the volcano, Chimborazo with the capital Riobamba and Tungurahua with the capital Ambato. Approximately 19,000 people had to be evacuated and the Ministry of Agriculture and Livestock reported that about 23,000 hectares of crops had been destroyed due to massive ash fall and that livestock experienced serious health problems from grazing in ash-covered pastures. Furthermore, as the Tungurahua region is a major tourist destination, the fact that all surrounding areas had to be evacuated and important infrastructure such as roads was partly destroyed led to a drop in tourism which strongly affected the economic situation in that region. Table 1 summarizes key demographic characteristics of loan applicants as well as typical purposes of loans for the whole period from January 2002 to August 2007. Since the two provinces closest to the volcano, Chimborazo and Tungurahua, were most strongly affected by the eruptions, we focus our analysis on the ProCredit branches operating in their capitals, i.e. Riobamba and Ambato, both located about 32 km away from the volcano. To confirm that the effects depend on the proximity to the volcano, however, we also compare the results to the effects of volcanic eruptions in the coastal region as it is unlikely that this area is affected by volcanic activity. In total, we observe 48,736 credit applications for Ambato and Riobamba and 50,767 for the branches in the coastal region between January 2002 and August 2007.16 Of these 48,736 (50,767) credit applications, in total 35,543 (33,108) were approved, representing 73% (65%) of all applications in that period. Table 1 shows that the majority of credit applicants in Ambato and Riobamba is male and married while that fraction is considerably lower in the coastal region. When it comes to the purpose of loans, the summary statistics clearly show that Ambato and Riobamba are focused on agriculture whereas the majority of loan applications is for business and trade in the coastal region, reflecting the different economic activities in the Andean and coastal regions, respectively.17 The financial information about the loans for Ambato and Riobamba is summarized in Table 2. It is clearly observable that the loan amount as well as the maturity applied for have, on average, increased over time. However, what is more interesting is that the amount applied for has, on average, always been considerably higher than the amount approved.18 The same result holds for the maturity. The maturity applied for is always higher than the maturity approved. These findings are intuitive since higher loan amounts and longer maturities imply higher risks for the bank as well. However, they also indicate that loan demand is not completely met by the bank, one explanation being that entrepreneurs do not have sufficient collateral or guarantees for receiving the amount or maturity they applied for. The

16 The data includes information on the loan terms requested and the loan terms granted, but no information on the balance sheets of the businesses. 17 The summary statistics for credit approvals are comparable to Table 1 and are thus not reported in detail. 18 For better comparability, the amount and maturity applied for are summarized only for future approvals.

556

G. Berg, J. Schrader / J. Finan. Intermediation 21 (2012) 549–568

Table 2 Financial information (averages) – Ambato and Riobamba.

Credit amount ($) Amount applied for Amount approved Maturity (months) Maturity applied for Maturity approved Repeat clients (%) Applied Approved No. applications No. approvals

2002

2003

2004

2005

2006

2007

2513 2185

2864 2508

2801 2617

3405 3094

3586 3404

3991 3705

11.1 10.4

13.2 12.3

14.3 13.6

16.3 15.6

17.8 17.1

19.6 19.1

21.54 22.10 3069 2462

34.37 38.66 5863 3968

47.63 51.19 7671 6589

58.94 64.88 8403 7016

47.98 59.42 11,946 8754

41.68 58.78 11,784 6754

Reported credit amounts were deflated to 2002 prices (Source: CIA World Fact Book). Figures for 2007 refer to the time from January to August.

Fig. 1. Seismic activity and eruptions of Tungurahua.

table also shows that the percentage of repeat clients within the overall pool of borrowers who apply for a loan is at about 50%. That the percentage of repeat clients actually receiving a loan (out of the pool of borrowers who are approved for credit) is considerably higher can be explained by the lower risk those entrepreneurs pose to the bank. Fig. 1 displays the seismic activity and eruptions of Tungurahua over time. Fig. 1a shows that the seismic activity has varied considerably over the years with a peak in mid 2006 when the severe eruption of the volcano occurred. This is even more apparent in Fig. 1b which depicts volcanic eruptions over time. The figure shows very clearly how severe the outbreak was compared to the years before and after this shock.

4. Econometric models For estimating the effect of unpredictable aggregate shocks, in our case volcanic eruptions, on credit demand and access to credit we use different econometric approaches. As volcanic eruptions are clearly an exogenous source of variation, we are able to exploit a ’natural experiment’ allowing us to identify causal relations between credit demand and volcanic eruptions on the one side and access to credit and volcanic activity on the other. First, in order to analyze whether the need for financing indeed increases after shocks occurred, we estimate a time series model in which we relate monthly credit demand, i.e. the number of loan

G. Berg, J. Schrader / J. Finan. Intermediation 21 (2012) 549–568

557

applications per month, to volcanic eruptions. According to the Dickey–Fuller and Phillips–Perron tests for stationarity, the time series are non-stationary and thus have to be stabilized first by using first differences. Having stabilized the time series, we identify the most adequate ARMA model specification by analyzing the autocorrelation (ACF) and partial autocorrelation functions (PACFs) and testing different ARMA models using the AIC and BIC criteria. The results suggest that an AR(1, 12) model outperforms all other specifications. Therefore, we estimate the following OLS model

C t ¼ h þ l1 C t1 þ l2 C t12 þ /1 St þ    þ /6 St5 þ v t ;

t ¼ 1; . . . ; T

ð5Þ

where C t is monthly credit demand. C t1 and C t12 are the relevant autoregressive terms and St to St5 correspond to the different lags of volcanic activity which are equal to monthly volcanic eruptions. Since the variables are used in first differences, St to St5 include six lags of the shock variables which are used in order to account for different effects over time.19 v t is the error term. The time series estimation will allow us to answer whether Proposition 1 from our theoretical model is fulfilled in this dataset. More precisely, if the lags for volcanic eruptions St to St5 are jointly significant and positive, we have a strong indication for increasing credit demand in response to volcanic activity which would imply that Proposition 1 is fulfilled. Second, in order to analyze whether access to credit changes due to volcanic activity, we use a random effects linear probability model (LPM) with branch fixed effects since the credit approval decision is a binary-choice variable and the data can best be represented as an unbalanced panel with multiple loan information for approximately half of the clients. However, it should be noted that the results are independent from the econometric approach chosen and a pooled cross-section linear probability model as well as a probit model yield the same results. The according latent variable model can be written as

Y ijt ¼ a þ St b þ X ijt d þ RLijt c þ SRLijt l þ cj þ ijt

ð6Þ

with the observed variable

n o Y ijt ¼ 1 Y ijt > 0 ;

i ¼ 1; . . . ; N;

j ¼ 1; . . . ; J;

t ¼ 1; . . . ; T:

ð7Þ

The dependent variable Y ijt equals one if the credit applicant j has received loan i at time t and zero otherwise. The vector St contains the aggregate shocks. In this specification, different indicators for volcanic activity are compared. Therefore, St is either monthly volcanic eruptions or the seismic activity of the volcano at the time of the credit application.20 We use up to six lags of the shock variables in this specification as well. However, since the results do not change when using the sum of the individual shock variables, we will usually refer to these estimates since they are easier to interpret. X ijt is a vector of demographic and loan characteristics including the age, marital status and gender of the loan applicant as well as the credit amount applied for. The vector RLijt is the indicator for relationship lending and shows accordingly whether entrepreneurs have received a loan from the bank before. SRLijt is an interaction term between volcanic activity and the relationship indicator displaying heterogenous effects between new and repeat clients. cj is the unobserved individual effect. Furthermore, purpose of loan as well as branch and year fixed effects are included in the regression. Due to the year fixed effects, St corresponds to within-year differential effects of volcanic activity. Finally, ijt is the error term. The main reason why we choose a random effects over an individual fixed effects model is that there is not enough intra-individual variation in the indicator for relationship lending so that an identification of the effect on access to credit would not be possible. Yet, as branch fixed effects are included throughout the regressions, at least time-constant unobserved effects on the branch level 19 We use six lags of volcanic activity since we assume that after 6 months, the effect of the shock is likely to die out. Entrepreneurs would have gone to the bank earlier if they were indeed affected by the shock and were in need of financing. However, a shorter time frame might be appropriate as well, yet not less than 3 months since firms need some adjustment time before being able to apply for a loan. Even though the results are not presented in detail here, it should be noted that they do not change considerably when using less than six time lags. 20 We use the levels of the shock variables in the analyses since it can be assumed that this is the effect that will be felt by the people living in the affected area.

558

G. Berg, J. Schrader / J. Finan. Intermediation 21 (2012) 549–568

can be controlled for. Eq. (6) and more precisely the coefficients b and l will allow us to ascertain whether Propositions 2 and 3 from our theoretical model are fulfilled in our dataset. More precisely, if b is negative and significant, access to credit is indeed negatively affected by volcanic eruptions indicating that Proposition 2 is fulfilled. Furthermore, if l is positive and significant and if the size of the coefficient is about equal or even greater than b, the results would imply that access to credit is only restricted for new clients while repeat borrowers can still access financing which would be in line with Proposition 3. If the coefficient should be significantly positive but smaller than b, the results would still imply preferential access for repeat clients, yet in that case access to finance would be restricted for all client groups. In the literature, different indicators are used for measuring relationship lending (Elsas, 2005). Within our theoretical model, we assume that the bank accumulates most of the valuable private information in the beginning of the relationship which is line with the findings of Cole (1998), among others. Despite of the distinction between new and repeat clients, the duration of the bank-client relationship and number of times a borrower has taken out a loan are frequently used as proxies for the depth of the relationship (Petersen and Rajan, 1994; Berger and Udell, 1995; Degryse and Ongena, 2001, among others). Therefore, in a third step and in order to further investigate the validity of our results, the number of loans obtained from the bank previous to the current credit application is used as an alternative indicator for relationship. Compared to the binary variable equal to one if the client has received a loan from the bank previously, the number of previous loans contains more information and may thus better reflect the extent and depth of the bank-client relationship. In principle, the estimation is identical to Eq. (6) except that RLijt and also the interaction effect SRLijt are now containing the number of loans obtained previous to the current loan application. An alternative indicator for the number of loans taken out before the current loan application would be the duration of the bank-client relationship, the indicator which is most often used in the literature (Elsas, 2005). Yet, we adhere to the first indicator as the information gains occur predominantly during the loan cycle so that the number of loans obtained should be a better indicator for relationship. However, it should be noted that the results remain qualitatively unchanged when using the duration of the relationship instead. Fourth, in order to analyze whether our results also hold for a sector highly affected by volcanic activity, we take a closer look at the agricultural sector. We therefore estimate

Y ijt ¼ a þ St b þ X ijt d þ RLijt c þ SRLijt l þ SAijt / þ SRLAijt w þ cj þ ijt

ð8Þ

which is similar to Eq. (6), but expanded by two additional interaction effects, first between volcanic activity and a binary variable indicating whether the loan was taken out for agricultural activity (SAijt ) and second, an interaction effect between volcanic activity, the indicator for the agricultural sector and the repeat client indicator ðSRLAijt Þ. If our assumption that the agricultural sector is one of the most strongly affected economic sectors is correct, / should be negative and significant. And similar to the interpretation above, if w is positive and significant and if the size of the coefficient is about equal or even greater than /, the results will again imply that access to credit remains unchanged or even improves for repeat clients while it is restricted for new credit applicants. Finally, we also estimate Eq. (6) for the coastal region, a region which is unlikely to experience any negative effects due to volcanic eruptions as it is located too far from the volcano. Hence, we expect b to be insignificant which is also why we exclude the interaction term between volcanic activity and the relationship indicator in this regression. Considering the structure of our dataset with the strong eruption of the volcano in 2006 and the available information for the affected as well as for the unaffected regions, it would also make intuitive sense to use a difference-in-difference approach. Yet, for the estimation on access to credit on the client-level this is not possible as discussed above. In principle, it would be feasible though to aggregate the information on credit approval rates on the branch level and analyze their development over time, yet this would leave us with only very few observations. Furthermore, if we only used the strong eruption in 2006, we would not be able to exploit the wealth of information available in our shock variables which is why, in the end, we decided to take the approach described above.

G. Berg, J. Schrader / J. Finan. Intermediation 21 (2012) 549–568

559

Fig. 2. Loan applications over time.

5. Empirical results We present the estimation results in a consecutive way by first analyzing loan demand in response to volcanic eruptions and then access to credit. Within the latter, we further differentiate between the two different indicators for relationship, new vs. repeat clients and the number of loans obtained previous to the current loan application. 5.1. Loan demand Before estimating the effects of volcanic eruptions on credit demand in a time series regression, we first analyze the development of loan applications over time for the branches of Ambato and Riobamba in a graphical way. The period of the strong eruption in 2006 is clearly indicated in the figure. From Fig. 2a it can be seen that loan applications in Ambato have been fluctuating over the years, but the positive trend is clearly visible. As the Riobamba branch has only been opened in the beginning of 2006, the time series is comparably short, yet the positive trend is visible for this branch as well (Fig. 2b). Whether or not volcanic eruptions had an effect on credit demand can, however, only be analyzed in a regression framework. As the number of observations is too low for estimating a meaningful time series regression for the Riobamba branch, we have to focus on Ambato in this case. The results of the regression of monthly credit demand on the autoregressive terms and the number of volcanic eruptions in the last 6 months are summarized in Table 3. The results show that the first, third and forth lag have the strongest effects on credit approval. Interestingly, the negative sign on the first lag suggests that volcanic activity, at first, leads to a decrease in credit demand while the positive effects become only significant after 3 months. This suggests that entrepreneurs need some adjustment time after volcanic eruptions occurred before applying for a loan. When calculating the combined effect of volcanic activity on credit demand, however, it is straightforward to see that the effect is positive.21 Furthermore, the F statistic indicates that the lagged terms for volcanic activity are jointly significant at the 1% level.22 Thus, the results show that Proposition 1 of our theoretical framework is indeed fulfilled in our dataset indicating that the number of loans demanded increases significantly in response to volcanic activity.23 21

This result holds independent of whether we take only the significant effects or the sum of all coefficients. These results hold not only for the overall client pool but also for the sub-samples of new and repeat clients. 23 From these results it follows that the total loan amount demanded will increase as well. Entrepreneurs who were in need of additional financing before a shock occurred will require higher loan amounts. Furthermore, some better off firms who would normally not have needed to apply for a loan will now be in need of additional capital as well. 22

560

G. Berg, J. Schrader / J. Finan. Intermediation 21 (2012) 549–568 Table 3 Time series regression for monthly credit demand. Variable

Coefficient

Autoregressive terms L1. L12. Volcanic activity Eruptions L1. L2. L3. L4. L5. Observations F-test for volcanic activity

0.48⁄⁄⁄ (0.15) 0.50⁄⁄⁄ (0.16) 0.01 (0.01) -0.02⁄⁄ (0.01) 0.00 (0.01) 0.05⁄⁄⁄ (0.01) 0.02⁄⁄⁄ (0.01) 0:01 (0.01) 55 20.94



At the 10% level. OLS regression for Ambato with robust standard errors. Standard errors in parentheses. Variables are in first differences. ⁄⁄ At the 5% level. ⁄⁄⁄ Denotes significant at the 1% level.

5.2. Access to credit – new vs. repeat clients Having found that credit demand increases after volcanic eruptions, we now turn to access to credit and first look at loan approval rates for the branches of Ambato and Riobamba over time (Fig. 3). Again, it can be seen that approval rates vary considerably over time, yet it also seems as if they were constantly declining from 2006 onwards dropping to about 60% in the case of Ambato and even 40% in Riobamba. During the time of the strong eruption in 2006, approval rates dropped sharply and even continued to decline in the aftermath of the outbreak.

Fig. 3. Loan approval rates over time.

G. Berg, J. Schrader / J. Finan. Intermediation 21 (2012) 549–568

561

Fig. 4. Loan approval rates over time – repeat vs. new clients.

Interestingly, Fig. 4 shows that the drop in approval rates was considerably stronger for new clients compared to repeat borrowers, both in Ambato and Riobamba. Whether those lower approval rates can be related to volcanic activity is analyzed in the random effects LPM regression of the credit approval decision on the covariates in Table 4. The two columns depict the estimation results for Ambato and Riobamba using the two different indicators for volcanic activity, i.e. the number of volcanic eruptions and the seismic activity of the volcano in the last 6 months.24 The coefficients reported in the table reflect percentage point changes in the probability to be approved for a loan since the credit approval indicator has been multiplied by 100 to allow for an easier interpretation of the results. The results of the estimations suggest that a higher number of eruptions as well as higher seismic activity in the last 6 months leads to a lower probability to be approved for a loan. The effects are equally strong in both regressions.25 For instance, when considering a one standard deviation increase in eruptions in regression (1), the effect implies a decrease in the probability to receive a loan of 4.9 percentage points.26 The results are the same when using the different time lags of volcanic activity individually. The results of regression (1) with the different lags of eruptions are displayed in Table A1 in the Appendix. It can be seen that the lags are highly significant as well and they are also jointly significant at the 1% level. Yet, since the aggregate estimates are easier to interpret and the results remain qualitatively the same when using the time lags individually, we refer to the results using the combined effects in the following. Table 4 also shows that the age of the applicant seems to have a negative but also nonlinear effect on the credit approval decision even though the effect is not significant in all specifications. Marriage, on the other side, increases the probability to receive a loan by about 5 percentage points. Interestingly, but not surprisingly, men have a lower probability to receive a loan than women. This could be explained by the fact that women are assumed to be more reliable when it comes to the repayment of loans (Armendáriz de Aghion and Morduch, 2005). Even though this effect may also be due to other reasons, it should be noted that ProCredit is not focusing explicitly on female customers. However, it

24 All indicators for volcanic activity have been divided by 100 in order to allow for a better interpretation of the rather small coefficients. 25 Since volcanic activity is an exogenous source of variation, it is unlikely to be influenced by any of the covariates. However, it could be argued that the coefficients may be biased since volcanic eruptions will have an effect on the loan amount demanded and typical purposes of loans as well. Yet, the effects of the shock remain qualitatively unchanged when excluding those variables from the regressions. The reason for focusing on the extended version of the regressions is that they show that the results even hold when controlling for variables that could be influenced by volcanic activity as well. 26 The standard deviation of ’’eruptions’’ is 37.66 with a mean of 23.57. The mean and standard deviation of ’’seismic activity’’ are 68.08 and 50.84, respectively.

562

G. Berg, J. Schrader / J. Finan. Intermediation 21 (2012) 549–568

Table 4 Random effects LPM for credit approval – new vs. repeat clients. Variable Volcanic activity Eruptions

(1) 0.13⁄⁄⁄ (0.01)

0.11⁄⁄⁄ (0.01)

Seismic activity Demographic and loan characteristics Age (Age)2 Married Male Amount applied for Relationship indicators Repeat client Repeat client⁄eruptions

0.31⁄⁄ (0.13) 3.50e03⁄⁄ ð1:55e  03Þ 4.91⁄⁄⁄ (0.55) 4.14⁄⁄⁄ (0.50) 1.50e04⁄⁄⁄ ð2:54e  05Þ

0.32⁄⁄⁄ (0.13) 4.44e03⁄⁄ ð1:65e  03Þ 4.86⁄⁄⁄ (0.55) 4.22⁄⁄⁄ (0.50) 1.50e04⁄⁄⁄ ð2:56e  05Þ

13.19⁄⁄⁄ (0.42) 0.10⁄⁄⁄ (0.01)

8.98⁄⁄⁄ (0.60)

Repeat client⁄seismic activity Year fixed effects Branch fixed effects Purpose of loan fixed effects Observations Client groups

(2)

0.10⁄⁄⁄ (0.01) Yes Yes Yes 47,477 22,454

Yes Yes Yes 47,477 22,454

Random effects GLS regression with standard errors clustered at the client level. Standard errors in parentheses. Denotes significant at the 1% level. ⁄⁄ At the 5% level.  At the 10% level.

⁄⁄⁄

could also be the case that female customers prepare their business plans more diligently once they decide to apply for a loan. As this is one of the most important criteria for the loan approval decision, this may thus be an alternative explanation for this finding. When looking at the effect of the credit amount applied for, it can be seen that the higher the amount, the lower the probability to receive a loan which can be explained by the higher risk of increasing loan amounts. The indicator whether the applicant is a repeat client has the strongest effect on the credit approval decision. The probability to receive a loan increases by about 13 percentage points compared to new credit applicants in regression (1). Interestingly, the interaction effects between volcanic activity and the repeat client indicator are indeed positive and highly significant in both regressions. Since the coefficients on the indicator for volcanic activity and the interaction effect have approximately the same size, this implies that repeat clients are about equally likely to receive a loan after volcanic shocks occurred. The results show that both, Propositions 2 and 3, are fulfilled in our dataset. We observe that access to credit is indeed negatively influenced by volcanic activity. Being a repeat client lowers these lending restrictions, however. For them, the probability to be approved for a loan remains more or less unchanged after volcanic shocks occurred.

5.3. Access to credit – depth of the relationship In order to analyze whether the results change when using an alternative indicator for relationship, namely the number of credits obtained previous to the current loan application, Table 5 summarizes

563

G. Berg, J. Schrader / J. Finan. Intermediation 21 (2012) 549–568 Table 5 Random effects LPM for credit approval – depth of the relationship. Variable Volcanic activity Eruptions

(1) 0.10⁄⁄⁄ (0.01)

0.07⁄⁄⁄ (0.01)

Seismic activity Demographic and loan characteristics Age (Age)2 Married Male Amount applied for Relationship indicators Number of previous loans Previous loans⁄eruption

0.41⁄⁄⁄ (0.14) 4.44e03⁄⁄⁄ ð1:65e  03Þ 4.77⁄⁄⁄ (0.57) 4.45⁄⁄⁄ (0.53) 1.66e04⁄⁄⁄ ð2:68e  05Þ

0.41⁄⁄⁄ (0.14) 4.46e03⁄⁄⁄ ð1:64e  03Þ 4.79⁄⁄⁄ (0.57) 4.49⁄⁄⁄ (0.53) 1.66e04⁄⁄⁄ ð2:69e  05Þ

5.25⁄⁄⁄ (0.22) 3.96e03⁄ ð2:14e  03Þ

5.00⁄⁄⁄ (0.23)

Previous loans⁄seismic activity Year fixed effects Branch fixed effects Purpose of loan fixed effects Observations Client groups

(2)

4.47e03⁄⁄ ð1:88e  03Þ Yes Yes Yes 47,477 22,454

Yes Yes Yes 47,477 22,454

Random effects GLS regression with standard errors clustered at the client level. Standard errors in parentheses. Denotes significant at the 1% level. ⁄⁄ At the 5% level. ⁄ At the 10% level.

⁄⁄⁄

the corresponding regressions using a random effects LPM as well. Similar to Table 4, the two columns depict the estimation results for Ambato and Riobamba using the two different indicators for volcanic activity. As can be seen in the table, the effects of volcanic activity on the probability to receive a loan remain strong and highly significant. Even though the size of the effects is slightly lower compared to the effects reported in Table 4, the difference is not high. The same holds for the demographic characteristics and the loan amount applied for. With respect to the number of previous loans, it can be seen that the effect is highly significant. The sign of the coefficient suggests that applicants are the more likely to receive a loan, the more credits they obtained from the bank before. With respect to the interaction effect between the number of previous loans and the indicators for volcanic activity, the results show that the more loans the client has obtained from the bank before, the less pronounced is the negative effect of volcanic activity on the probability to receive a loan. This result clearly shows that the stronger and deeper the relationship, the more the client benefits. These findings do not only corroborate the results obtained in Table 4 and Propositions 2 and 3, but show that the effect of volcanic activity on access to credit depends on the extent and depth of the bank-client relationship. 5.4. Access to credit – the agricultural sector The agricultural sector is likely to be the economic sector most strongly affected by volcanic eruptions. As described in Section 3 above, during the severe eruption of Tungurahua in 2006, 23,000 hectares

564

G. Berg, J. Schrader / J. Finan. Intermediation 21 (2012) 549–568

Table 6 Random effects LPM for credit approval – the agricultural sector. Variable Volcanic activity Eruptions

(1) 0.11⁄⁄⁄ (0.01)

0.11⁄⁄⁄ (0.01)

Seismic activity Demographic and loan characteristics Age (Age)2 Married Male Amount applied for Relationship indicators Repeat client Repeat client⁄eruption

0.31⁄⁄ (0.13) 0.00⁄⁄ (0.00) 4.94⁄⁄⁄ (0.55) 4.16⁄⁄⁄ (0.50) 1.48e04⁄⁄⁄ ð2:54e  05Þ

0.32⁄⁄⁄ (0.13) 0.00⁄⁄ (0.00) 4.88⁄⁄⁄ (0.55) 4.23⁄⁄⁄ (0.50) 1.48e04⁄⁄⁄ ð2:55e  05Þ

13.14⁄⁄⁄ (0.42) 0.08⁄⁄⁄ (0.01)

8.90⁄⁄⁄ (0.60)

Repeat client⁄seismic activity Agriculture and relationship Agriculture Agriculture⁄eruption Agriculture⁄eruption⁄repeat client

0.09⁄⁄⁄ (0.01) 4.81⁄⁄⁄ (0.82) 0.07⁄⁄⁄ (0.02) 0.07⁄⁄⁄ (0.02)

4.25⁄⁄⁄ (0.92)

0:02 (0.01) 0.04⁄⁄⁄ (0.01)

Agriculture⁄seismic activity Agriculture⁄seismic activity*repeat client Year fixed effects Branch fixed effects Purpose of loan fixed effects Observations Client groups

(2)

Yes Yes Yes 47,477 22,454

Yes Yes Yes 47,477 22,454

Random effects GLS regression with standard errors clustered at the client level. Standard errors in parentheses. Denotes significant at the 1% level. ⁄⁄ At the 5% level.  At the 10% level.

⁄⁄⁄

of crops were destroyed and livestock died because of health problems resulting from grazing in ashcovered pastures. It is therefore interesting to analyze whether our results remain valid when estimating the effect of volcanic activity on access to credit with a particular focus on the agricultural sector. Table 6 depicts the results from the estimations summarized in Table 4, but expanded by two interaction effects, first between volcanic activity and a binary variable indicating whether the loan was taken out for agricultural activity and second, an interaction effect between volcanic activity, the indicator for the agricultural sector and the repeat client indicator. The estimation results show that all previous results remain valid and that the finding that repeat clients are about equally likely to receive loans after volcanic shocks occurred is also true for loans taken out for agricultural activity.27 In general, loans for the

27 It should be noted that the results remain qualitatively unchanged when using the number of previous loans as the indicator for relationship.

565

G. Berg, J. Schrader / J. Finan. Intermediation 21 (2012) 549–568 Table 7 Random effects LPM for credit approval – coastal region. Variable Volcanic activity Eruptions

(1) 0.01 (0.01)

Seismic activity Demographic and loan characteristics Age (Age)2 Married Male Amount applied for Relationship indicator Repeat client Year fixed effects Branch fixed effects Purpose of loan fixed effects Observations Client groups

(2)

0.01 (0.01) 0.54⁄⁄⁄ (0.11) 0.01⁄⁄⁄ ð1:27e  03Þ 6.99⁄⁄⁄ (0.39) 1.91⁄⁄⁄ (0.40) 2.30e04⁄⁄⁄ ð5:13e  05Þ

0.55⁄⁄⁄ (0.11) 0.01⁄⁄⁄ ð1:27e  03Þ 6.99⁄⁄⁄ (0.39) 1.91⁄⁄⁄ (0.40) 2.31e04⁄⁄⁄ ð5:13e  05Þ

23.82⁄⁄⁄ (0.39)

23.85⁄⁄⁄ (0.39)

Yes Yes Yes 50,767 24,776

Yes Yes Yes 50,767 24,776

Random effects GLS regression with standard errors clustered at the client level. Standard errors in parentheses. Denotes significant at the 1% level.  At the 5% level,  at the 10% level. ⁄⁄⁄

agricultural sector tend to be more likely to be approved compared to loans for sectors such as production, construction, transport or commerce. The negative effect of volcanic activity, however, is more pronounced for the agricultural sector – in line with our presumption. Yet, when comparing the two newly introduced interaction effects it becomes clear that this negative effect again only holds for new credit applicants while repeat clients engaged in agricultural activity are about equally likely or even somewhat more likely to receive a loan. This implies that our results even hold for the sector most affected by volcanic activity. 5.5. Access to credit – the coastal region In order to plausibilize our results, we also estimate the effects of volcanic activity on access to credit in the coastal region as we assume that volcanic eruptions will have no impact in that area due to the fact that the distance from the volcano is too high.28 The results of the random effects LPM with the two indicators for volcanic activity are summarized in Table 7. As presumed, no significant effects of volcanic activity on the probability to receive a loan can be observed for the branches in the coastal region. Interestingly, the effects of marriage, the loan amount applied for and also of the relationship indicator are stronger in the coastal region while the gender effect is lower, yet still highly significant. These findings reflect the considerable economic, social, and cultural differences between the coastal and the Andean region. We explicitly excluded the interaction effect between volcanic activity and the relationship indicator. It should be noted, however, that the interaction effect is also insignificant if included in this specification while the remaining coefficients remain qualitatively unchanged. 28 The closest branch in the coastal region is located in Guayaquil with a distance of about 190 km to the volcano and the one furthest away is the Manta branch with 260 km distance.

566

G. Berg, J. Schrader / J. Finan. Intermediation 21 (2012) 549–568

Apart from the estimations for the coastal region, we also analyzed the effects of volcanic activity on access to credit for other branches in the Andean region as those may be affected at least by very strong volcanic eruptions as well, especially if they are located not too far from the volcano. In line with our previous findings, the results of these estimations show an effect of volcanic activity, yet compared to the effect in Ambato and Riobamba, it is considerably smaller.29 Finally, we also analyzed whether loan conditions were affected by the shock, but did not find any evidence that the shock might have had an impact on the loan amount requested or the difference between the loan amount requested and granted. Effects were also not observable for interest rates or the maturity of the loans. These findings might be explained by the strong standardization of loan contracts in the case of micro and SME finance. Thus, overall, the results from our plausibility analyses lend credibility to our presumption that volcanic activity will mainly be felt in the region closest to the volcano and that we are not picking up some other effects which, for some reason, might be closely related to the shock variables.

6. Conclusion The determinants of access to credit for firms in developing economies depend on a wide range of factors. Important aspects include, among others, the institutional and regulatory environment, the clients’ financial means as well as qualitative information on the potential borrower, particularly important in the case of relationship lending. While the drivers of credit availability have been analyzed from various perspectives, until today, there has been no study addressing the question whether and to what extent access to credit changes after unpredictable aggregate shocks such as natural disasters. This, however, is of particular interest as it is often assumed that the existence of credit markets can dilute the effects of shocks, for instance through the provision of emergency loans. With this paper we made a first step toward closing this gap by analyzing a dataset including all loan applications and subsequent approvals for a self-sustainable Ecuadorian MFI between January 2002 and August 2007. As we combine this data with information on volcanic eruptions in Ecuador during the same time, we are able to exploit a ’natural experiment’ allowing us to clearly identify the determinants of credit availability in response to unpredictable aggregate shocks. The empirical results show that credit demand increases significantly after volcanic eruptions suggesting that there exists a need for additional financing after shocks occurred. When analyzing access to credit on the level of the individual client, the results indicate that high volcanic activity in the last months before the credit application leads to significant decreases in the probability to be approved for a loan. Yet, we also find that being a repeat borrower lowers these lending constraints. More specifically, the results show that returning clients do not only have a higher probability to receive a loan in general, but that they are about equally likely to be approved for a loan after volcanic shocks occurred. As, in our setting, it is less risky for a MFI to lend to repeat clients who are already known to the institution, repeat borrowers face less lending restrictions. New clients, however, do not have this advantage and therefore, in times of crises, it becomes even more difficult for them to receive financing, a result which can be directly associated with asymmetric information. Our results therefore show that bank-borrower relationships are important determinants for increasing access to finance in times of unpredictable aggregate shocks. The findings are further validated given that our results also hold for the agricultural sector, the sector most affected by volcanic eruptions, and by the fact that no significant effects of volcanic activity on access to credit can be found in a region which is unlikely to be affected by volcanic eruptions. Finally, with respect to the generalizability of our results, the findings should hold for comparable banks operating in settings with high degrees of asymmetric information and facing unpredictable aggregate shocks. 29

The results of these estimations can be obtained from the authors upon request.

567

G. Berg, J. Schrader / J. Finan. Intermediation 21 (2012) 549–568 Table A1 Random effects LPM for credit approval – incl. lags. Variable Volcanic activity Eruptions L1. L2. L3. L4. L5. L6. Demographic and loan characteristics Age (Age)2 Married Male Amount applied for Relationship indicators Repeat client Repeat client⁄eruptions L1. L2. L3. L4. L5. L6. Year fixed effects Branch fixed effects Purpose of loan fixed effects Observations Client groups

Coefficient

(Std. error)

0.25⁄⁄⁄ 0.17⁄⁄⁄ 0.14⁄⁄⁄ 0.14⁄⁄⁄ 0.19⁄⁄⁄ 0.30⁄⁄⁄ 0.09⁄⁄⁄

(0.03) (0.04) (0.04) (0.04) (0.04) (0.04) (0.03)

0.32⁄⁄⁄ 0.00⁄⁄ 4.84⁄⁄⁄ 4.05⁄⁄⁄ 1.48e04⁄⁄⁄

(0.13) (0.00) (0.54) (0.50) ð2:53e  05Þ

13.15⁄⁄⁄ 0.18⁄⁄⁄ 0.15⁄⁄⁄ 0.10⁄⁄ 0.07 0.10⁄⁄ 0.13⁄⁄⁄ 0.05 Yes Yes Yes 47,477 22,454

(0.43) (0.04) (0.05) (0.05) (0.04) (0.05) (0.05) (0.04)

Random effects GLS regression with standard errors clustered at the client level. Denotes significant at the 1% level. ⁄⁄ Denotes significant at the 5% level.  Denotes significant at the 10% level.

⁄⁄⁄

Acknowledgments We are deeply indebted to Gabriel Schor from ProCredit Holding and the staff of Banco Pro Credit Ecuador, Cesar Vinueza and Oscar Villaseca among others, for their generous support. We are also grateful to the Instituto Geofı´sico Ecuador and Veronica Lopez for excellent research assistance in Ecuador. Special thanks are also due to Anja Baum, Bob Cull, Stefan Klonner, Rainer Klump, Robert Lensink, Pierre-Guillaume Méon, Friedhelm Pfeiffer, Andreas Roider, Eva Terberger, the participants of conferences and seminars in Brussels, Heidelberg, Frankfurt, and Rome, and two anonymous referees for valuable comments and support. Appendix A See Table A1. References Armendáriz de Aghion, B., Morduch, J., 2005. The Economics of Microfinance. The MIT Press, Cambridge, MA. Becchetti, L., Castriota, S., 2011. Does microfinance work as a recovery tool after disasters? Evidence from the 2004 Tsunami. World Dev. 39 (6), 898–912. Beck, T., Demirguc-Kunt, A., Maksimovic, V., 2004. Bank competition and access to finance: international evidence. J. Money, Credit, Bank. 36 (3), 627–648.

568

G. Berg, J. Schrader / J. Finan. Intermediation 21 (2012) 549–568

Beck, T., Demirguc-Kunt, A., Maksimovic, V., 2008. Financing patterns around the world: are small firms different? J. Finance Econ. 89 (3), 467–487. Beegle, K., Dehejia, R., Gatti, R., 2006. Child labor and agricultural shocks. J. Dev. Econ. 81, 80–96. Berg, G., Kirschenmann, K., 2012. Funding vs. real economy shock: the impact of the 2007–2009 crisis on small firms’ credit availability. World Bank Pol. Res. Working Paper 6030. Berger, A., Udell, G., 1995. Relationship lending and lines of credit in small firm finance. J. Bus. 68 (3), 351–381. Boot, A., 2000. Relationship banking: what do we know? J. Finance Intermed. 9, 7–25. Boot, A., Thakor, A., 2000. Can relationship banking survive competition? J. Finance 55 (2), 679–713. Brown, M., Jappelli, T., Pagano, M., 2009. Information sharing and credit: firm-level evidence from transition countries. J. Finance. Intermed. 18 (2), 151–172. Charvériat, C., 2000. Natural disasters in Latin America and the Caribbean: an overview of risk. Inter-American Development Bank Working Paper 434. Cole, R., 1998. The importance of relationships to the availability of credit. J. Bank. Finance 22, 959–977. Degryse, H., Ongena, S., 2001. Bank relationships and firm profitability. Finance Manage. 90, 9–34. Del Ninno, C., Dorosh, P., Smith, L., 2003. Social capital and coping with economic shocks: public policy, markets and household coping strategies in Bangladesh: avoiding a food security crisis following the 1998 floods. World Dev. 31 (7), 1221–1238. Dercon, S., 2002. Income risk, coping strategies, and safety nets. The World Bank Res. Obser 17 (2), 141–166. Economic Commission for Latin America and the Caribbean (ECLAC), 2003. Handbook for Estimating the Socio-Economic and Environmental Effects of Disasters. United Nations, ECLAC and International Bank for Reconstruction and Development, The World Bank. Elsas, R., 2005. Empirical determinants of relationship lending. J. Finance Intermed. 14 (1), 32–57. Eswaran, M., Kotwal, A., 1989. Credit as insurance in agrarian economies. J. Dev. Econ. 31, 37–53. Gitter, S., Barham, B., 2007. Credit, natural disasters, coffee, and educational attainment in rural honduras. World Dev. 35 (3), 498–511. Holmstroem, B., Tirole, J., 1997. Financial intermediation, loanable funds and the real sector. Quart. J. Econ. 112 (3), 663–691. Jacoby, H., Skoufias, E., 1997. Risk, financial markets, and human capital in a developing country. Rev. Econ. Stud. 64 (3), 311– 335. Jimenez, G., Ongena, S., Peydro, J.-L., Saurina, J., 2012. Credit supply versus demand: bank and firm balance-sheet channels in good and crisis times. EBC Discussion Paper No. 2012-003. Khandker, S., 2007. Coping with flood: role of institutions in Bangladesh. Agric. Econ. 36 (2), 169–180. Khwaja, A.I., Mian, A., 2008. Tracing the impact of bank liquidity shocks: evidence from an emerging market. Am. Econ. Rev. 98 (4), 1413–1442. Miamidian, E., Arnold, M., Burritt, K., Jacquand, M., 2005. Surviving Disasters and Supporting Recovery: A Guidebook for Microfinance Institutions. World Bank Disaster Risk Manage. Res. Working Paper Series No. 10. Morduch, J., 1995. Income smoothing and consumption smoothing. J. Econ. Perspect. 9 (3), 103–114. Pantoja, E., 2002. Microfinance and Disaster Risk Management. Experiences and Lessons Learned. Final Report, Provention Consortium, World Bank, UNDP, UNCDF. Petersen, M., Rajan, R., 1994. The benefits of lending relationships: evidence from small business data. J. Finance 49 (1), 3–37. Petersen, M., Rajan, R., 1995. The effect of credit market competition on lending relationships. Quart. J. Econ. 110 (2), 407–443. Puri, M., Rocholl, J., Steffen, S., 2010. Global retail lending in the aftermath of the us financial crisis: distinguishing between supply and demand effects. J. Finance Econ. 100, 556–578. Schrader, J., 2009. The Competition between Relationship-based Microfinance and Transaction Lending. University of Heidelberg, Mimeo. Sharpe, S., 1990. Asymmetric information, bank lending, and implicit contracts: a stylized model of customer relationships. J. Finance 45 (4), 1069–1087. Shoji, M., 2010. Does contingent repayment in microfinance help the poor during natural disasters? J. Dev. Stud. 46 (2), 191– 210. Thoumi, F., 1990. The hidden logic of ’’Irrational’’ economic policies in ecuador. J. Interam. Stud. World Aff. 32 (2), 43–68. Von Pischke, J., 2002. Innovation in finance and movement to client-centered credit. J. Int. Dev. 14 (3), 369–380.