WPS4772 Policy ReseaRch WoRking PaPeR 4772 How Does Geographic Distance Affect Credit Market Access in Niger? Jose Pedrosa Quy-Toan Do The World Bank Development Research Group Poverty Team November 2008 Policy ReseaRch WoRking PaPeR 4772 Abstract Distances involved in accessing basic services can To cope with the effects of geographical distance, constitute a major barrier to development. This microfinance institutions adapt their policies through paper analyzes the relationship between the distance more restrictive loan conditions, higher interest rates, separating households from microfinance institutions' and more intensive screening. The authors to discuss the offices in Niger, and the low levels of development and tension between access and sustainability in the context performance of the microfinance sector in the country. of financial services for the poor. This paper--a product of the Poverty Team, Development Research Group--is part of a larger effort in the department to understand access to finance for the poor. Policy Research Working Papers are also posted on the Web at http://econ. worldbank.org. The author may be contacted at edecastro@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team How Does Geographic Distance A¤ect Credit Market Access in Niger?1 Jose Pedrosa2 Quy-Toan Do3 1We would like to thank Hinh Truong Dinh, Amadou Ibrahim, and Peter Lan- jouw for helpful comments and discussions. The ...ndings, interpretations, and con- clusions expressed in this paper are entirely those of the authors. They do not necessarily reect the views of the World Bank, its Executive Directors, or the countries they represent. 2Corresponding author: Yale University. Economic Growth Center, 27 Hillhouse Avenue, New Haven, CT 06511, USA. jose.pedrosa@yale.edu 3Development Economics Research Group. The World Bank. MSN MC3-306. 1818 H street, NW. Washington DC 20433, USA. qdo@worldbank.org 1 Introduction This paper explicitly investigates the impact that distance can have on the development of micro...nance institutions (MFIs) in Niger. Distance is de- ...ned as the geographical space by road between households and the MFI's o¢ ce. Low population density might involve long distances for households to cover in order to access facilities and services such as micro...nance insti- tutions, health centers, or schools. The intensity of this e¤ect will be deter- mined by factors such as the development of the country's transportation infrastructure and services, eventually a¤ecting multiple aspects of house- holds'livelihoods. Our study will focus on Niger, a landlocked country in the Sahara-Sahel region of West Africa, with a vast surface of 1,267,000 square kilometers. The country is a member of the West African Economic and Monetary Union (WAEMU), and the Economic Community of West African States (ECOWAS). In 2006, Niger's population was estimated at 14.4 million. From 2000 to 2005, the population growth rate was 3.4%. In 2005, rural popula- tion accounted for 83% of the total (United Nations, 2006). The country's GDP grew at a 6% rate during the period 2005-2007. However, it remains one of the poorest nations in the world, with an average per capita income estimated at US$280 in 2006 with 61% of the population living on less than one dollar a day. The economy is led by the agricultural sector, which ac- counted for 47% of GDP in 2006. The main economic activities are livestock, mining (particularly uranium), and informal trade activities (World Bank, 2008). Social indicators are low, with a life expectancy at birth of 44 years, an infant mortality rate of 81 per 1,000 live births, a literacy rate of 28.7 percent in 2005, and a gross primary school enrollment rate of 54 % in 2006 (United Nations, 2006). In 2007, Niger ranked 174th out of 177 countries ac- cording to the United Nations Development Program's Human Development Index. In 2004, Niger obtained Highly Indebted Poor Country status (World Bank, 2008). One of the main features of Niger is its population density. In Table 1 we see that on average, the population density of 11 people per square kilometer makes Niger one of the lowest-density ECOWAS countries. Most of the population (90%) concentrates in the southern regions and the northern part of the country consists of the Ténéré Desert. The region with the lowest density is Agadez with 0.5 people per square kilometer; its terri- tory essentially belongs to the Sahara desert. The region with the highest population density is Maradi, with 40 people per square kilometer, which remains a low ...gure compared to other ECOWAS countries. As we can see in Table 1, some important characteristics of the micro- 1 ...nance sector in Niger are its low level of development compared to other ECOWAS countries. With the lowest number of MFI o¢ ces, the MFIs' networks are less developed than in neighboring countries. With the sec- ond highest rate of credit at risk per bene...ciary, MFIs portfolios are also of comparatively poor quality. When we look at the number of employees per 1,000 bene...ciaries we nevertheless see that the sector employs a relatively important labor force. With only one institution serving 26% of the clients, mobilizing 14% of the deposits and granting 10.6% of the loans, we could consider that the sector is fairly concentrated, even though Niger seems to have the least concentrated rural-...nance market of the ECOWAS countries (SFD BCEAO, 2004). All WAEMU countries have the same legal framework regarding ...nancial institutions; Niger adopted it in 1996. At the national level, the Ministry of Finance is in charge of its application and monitor- ing. In 2004 there were in Niger 61 MFIs registered (SFD BCEAO, 2004). The situation of the micro...nance sector remains very fragile, with very low ...nancial capacity MFIs and a contraction of donors'funds (SFD BCEAO, 2004). Niger's government has contributed to the dialogue of all the actors in the sector, particularly through the adoption of a National Strategy of Micro...nance in 2001. The policy debate has focused on restructuring the micro...nance sector in order to improve sustainability (République du Niger, 2001). Some organizations contend that distance may be an important factor restraining access to ...nance in some countries (Seep Network, 2006), and even in the United States (Petersen and Rajan, 2002). Some authors have also pointed at low population density as a factor restraining micro...nance development. Paxton (1995) studies the similarities and di¤erences in adapt- ing the Grameen group-lending model in Burkina Faso, a methodology that would permit higher outreach in sparsely populated areas. After di¤erent tests, it was determined that it was most convenient to work with women in groups of ...ve. Concentrating on female clients would target more disadvan- taged groups, which are also less likely to migrate to neighbor countries and, according to some MFI employees, entail lower transaction costs because they accept rules and regulations more readily. Paxton (1995) describes three group-dynamics mechanisms that inuence repayment: coordination, group solidarity, and peer pressure (ex-ante and ex-post). Coordination is- sues yield multiple equilibria whereby a borrower's incentives are to default (resp. repay) as long as others default (resp. repay). Group solidarity relates to the willingness of group members to pay for one of its members in case of a shock (insurance). Finally, peer pressure would also favor repayment, either ex-ante through selection of better co-borrowers, or ex-post to avoid 2 interruption of credit access. Paxton (1995) concludes that "the provision of micro...nance services has proven to be quite costly in the Sahel. The reasons are more related to the environment (low population density, poor infrastructure, poverty, illiteracy) than to the methodology of group lending itself." However, there is a general lack of studies focusing on population density and the costs associated with the distance that it may entail. Yet, the policy debate is rather focused on restructuring the micro...nance sector in order to improve sustainability (République du Niger, 2001). When considering an economic transaction between two agents -in our case between a household, and individual or group of individuals and a mi- cro...nance institution- the e¤ect of distance consists of the physical cost that one of the agents or both need to pay in order to be able to realize the trade. We analyze three models of credit markets: a complete information frame- work, one with adverse selection and one with moral hazard. We assume that distance a¤ects credit market equilibria in several ways. First, there is the direct transaction cost: the actual transportation cost to deliver ...nancial services to the borrower. We show that under competitive ...nancial markets, the costs are borne by the borrower in the form of more intense screening of projects and borrowers and higher interest rates. A second implication is an increase in monitoring costs: whether the lender needs to collect pre- loan-approval information on the lender (adverse selection), or monitor the borrower after the loan is made (moral hazard). Monitoring comes with costs that at the margin can inuence the decision whether to monitor or not. This in turn translates into even stricter lending restrictions and higher interest rates in equilibrium. Finally, we postulate that at larger distances, the demand for credit changes as underlying charateristics (e.g. education) are correlated with distance. Our selection models then predict patterns of borrower and loan pro...les that are the conjunction of MFI screening prac- tices and the underlying spatial distribution of characteristics. In particular, we ...nd evidence that distance is also associated with higher interest rates, lower loan amounts, but also lower frequencies of monitoring, lower default rates and a higher prevalence of female group lending despite the fact that female literacy is much lower than male literacy (the ratio of young liter- ate females to males is 83% for individuals aged 15-24, World Bank, 2008). We argue that these results are consistent with the view that (i) MFIs pass transaction costs onto their clients in the form of higher interest rates and more active screening, and (ii) credit markets are characterized by moral hazard, with an agency cost that is decreasing with distance. We postulate that an increasingly monopolistic power of MFIs acts as a disciplining device 3 as reputation for creditworthiness is more important when there are fewer alternative sources of credit. We would like to acknowledge several shortcomings of our paper. First, data quality has been a limitation in our study. The survey took place at the peak of the 2005 famine, and availability and attention of households have certainly a¤ected the quality of the data collected. The high costs related to transportation also limited the number of households that could be visited. Second, in both theoretical and empirical discussions, we will take the institution of group lending as given, and will not privilege one speci...c model of group lending. Thus, while the existence of group lending is consistent with our theory, we will be agnostic about the relationship between groups and geographic distance. Our paper belongs to the literature of geography-based economic devel- opment (see e.g. Fujita et al. 1999, Redding and Venables, 2002), and the interplay between geographical isolation and development. However, micro- econometric evidence is scarce. A large part of the literature dealing with isolation and its relationship to economic development has discussed the impact of infrastructure on access to public services and markets (see e.g. Jacoby, 2000). By its descriptive nature, our paper relates to Fafchamps and Wahba (2004) who look at the spatial distribution of child labor in Nepal. Fafchamps and Moser (2003) also ...nd that isolation is a source of weaker law and enforcement. Our paper is thus an attempt to look at the e¤ect of geographic isolation on access to ...nancial services. Finally, by testing credit market models, our paper is close to Edelberg (2004) who looked at evidence of adverse selection and moral hazard on the consumer loan market in the US or to other e¤orts to detect adverse selection in credit markets (see e.g. Calem and Mester, 1995, and Ausubel, 1999). The paper is organized as follows: Section 2 introduces di¤erent credit market models to describe the potential mechanisms that would induce dis- tance to a¤ect micro...nance development; Section 3 will present the data and empirical results, and Section 4 concludes. 2 Testing credit market models In this section, we present three di¤erent models of the credit market: a com- plete information model, and credit market models characterized by adverse selection and moral hazard, respectively. We emphasize the behavior of the equilibrium of the economy as geographical distance between borrowers and lenders increases. 4 2.1 A model of credit markets with complete information In this setting, there is complete information between lenders and borrowers. We use the representative borrower to model group lending. Entrepreneurs have no collateral and raise I units of capital from the micro...nance insti- tution - the lender. Borrowers are characterized by a vector of observable attributes that include education, distance to the micro...nance institution and other relevant characteristics. The project cycle is as follows: at time T = 0, a loan of size I is granted and invested. By construction, we as- sume that returns to capital drop to zero above a given threshold I: At time T = 1, the project is succesful with probability p ( ) and yields R ( ) I ( ), and fails with probability 1 p ( ) with zero returns. The probability of default on a loan is thus ( ) = 1 p ( ) : Under complete information, the lender observes perfectly so that loans can be made contingent on . Under the condition that lenders break-even in equilibrium, if r ( ) is the interest rate charged by lenders, credit supply is given by p ( ) r ( ) R ( ) = 1 Borrowers choose the amount of loan to apply for in order to maximize their surplus so that I ( ) = arg max p ( ) R ( ) I I; I which yields I if p ( )R ( ) 1 > 0 (1) I ( ) = ; 0 otherwise and when the individual can borrow 1 (2) r ( ) = : p ( ) R ( ) The step function behavior of the loan function I ( ) is essentially driven by the constant-returns-to-scale technology. By allowing decreasing returns, re- sults are qualitatively unaltered: loan amounts increase with , while interest rates decrease. We now look at the determinants of the key parameter . We suppose that is a function of two salient parameters: the distance d that separates the loan applicant to the micro...nance o¢ ce, and a vector of characteristics 5 e that includes the gender of the borrower, its education level, its sector of activity, etc. Thus, we model (d;e) with the structural assumptions that @ ( ) R ( ) < 0 and @ ( ) R ( ) > 0. The implicit function theorem @d @e implies that the marginal borrower de...ned by p ( )R ( ) = 1 is such that de > 0 : due to the selection process stemming from (1), characteristics such dd as education, ability to reimburse a loan are increasing as distance increases, even though the relationship for the average borrower is going in the opposite direction. To see this, let's suppose that e measures say education and let's write the average education of all borrowers as 1 E [eje e (d) ; d] = Ze e (d) f (ejd) de; where f (ejd) is the distribution of education levels at given distance d. For any d0 > d, we have (3) 1 E [eje e (d) ; d] E [eje e (d0) ; d0] = Z The ...rst term in (3) is the intensive margin|e¤ect, which compares the |eZe (d)e f (ejd) f ejd0 de intensive margin e (d)ef e{zd0 de }: (d0) j extensive margin {z } distributions of education levels as distance d increases. The second term is the extensive margin e¤ect, whereby the marginal borrower e (d) is such that e (d) > e (d0), provided that 0d< 0 and 0e> 0: Proposition 1: Suppose the distribution of characteristics e to be such that: (i) [f (ejd) f (ejd0)] < 0 for every e and distances d < d0 : Then intensive and extensive margin e¤ects reinforce each other and for any d0 > d; E [eje e (d) ; d] E eje e d0 ; d0 < 0 (ii) [f (ejd) f (ejd0)] > 0 for every e and distances d < d0 : Then intensive and extensive margin e¤ects o¤set each other and the net e¤ect is ambiguous. We summarize the other results below: 6 Proposition 2: Under the assumption that borrowers have access to decreasing-returns-to-scale technology, the following holds in equilibrium: (i) Loan amounts are non-increasing as distances increase: @I ( )=@d 0. (ii) The probability of default on a loan increases with distance: @ ( )=@d > 0. (iii) Interest rates charged to borrowers increase with distance: @r ( )=@d > 0. (iv) There is no scope for additional ex-ante or ex-post monitoring of the borrower by the lender as contracts are complete. 2.2 Adverse selection in credit markets We keep the same framework, i.e. borrowers are characterized by (e;d) that is observable by the lender. However, there are two types of borrowers: p or q such that p > q and the probability of success of their investments is now re- spectively p ( ) and q ( ). Types of borrowers are not directly observed by the lender. However, the distribution of types is common knowledge, and we denote ( ) the probability a potential borrower is of type p. The literature on joint liability (see e.g. Ghatak, 1999) has argued that joint liability could be an institutional response to adverse selection in credit markets. Thus, if adverse selection is more severe as distance increases, group lending is more likely to be observed further away from micro...nance o¢ ces. However, there is no clear prediction on group size or group composition. Furthermore, we assume that borrowers do not have su¢ cient funds to pledge as collateral, so that no separation can take place between p and q borrowers. Alterna- tively, lenders can decide to invest in a monitoring technology in order to observe types with probability 1, but such technology has cost ( ) I ( ). Monitoring will therefore take place if and only if p ( ) r ( ) R ( ) I ( ) ( ) I ( ) ( ) [ ( ) p + (1 ( )) q] r ( ) R ( ) I ( ) ; while lenders'participation is given by p ( ) r ( ) R ( ) I ( ) ( ) I ( ) I ( ) with monitoring and ( ) [ ( ) p + (1 ( )) q] r ( ) R ( ) 1 without. Then, supply of credit is equal to I ( ) = I if sup f ( ) [ ( ) p + (1 ( )) q] R ( ) ; [p ( ) R ( ) ( )]g 1 : 0 otherwise 7 Borrowers have expected probability of default ( ) = 1 s ( ) with s = p;q, and are charged interest rates 1 + ( ) r ( ) = p ( ) R ( ) with monitoring and 1 r ( ) = [ ( ) p + (1 ( )) q] ( ) R ( ) otherwise. The implications are thus qualitatively similar to the complete information framework as @ (:) > 0: Adverse selection in e¤ect imposes an @d extra monitoring cost to the lender that is then passed onto the borrower via higher interest rates and more stringent lending conditions. Loan applicants (who include recipients and those who were denied credit) will be visited by a lender prior to the loan decision with a likelihood that depends on the relative cost ( ) with respect to the "need"for screening ( ). Without further structural assumptions on these functions, the patterns of pre-loan visits are uncertain. However, conditional on receiving a loan, all individuals are equally likely to be monitored by the lender. We summarize the results below: Proposition 3: If credit markets are characterized by adverse selection: (i) Properties (i) and (ii) of Proposition 1, and properties (i)-(iii) of Propo- sition 2 still hold. (ii) Group lending is more likely to be observed at larger distances from micro...nance o¢ ces. (iii) Lenders undertake monitoring visits prior to the loan decision. The probability of visits is identical conditional on being a borrower; in particu- lar, it does not depend on d. 2.3 Moral hazard in credit markets Let's modify the complete-information model and add a T = 1 e¤ort stage, in which the borrower has the option to exert e¤ort. If the borrower ex- erts e¤ort, then the probability of success is p ( ) but the borrower does not enjoy any private bene...t. Otherwise, the borrower gets private bene...t B ( ) I ( ) but the probability of success drops down to q ( ), with q < p: 8 E¤ort is not contractible, but the lender can spend an amount ( )I ( ) in monitoring costs to bring the borrower's private bene...t down to 0: Borrow- ers'incentive-compatibility constraints are given by ( ) (4) r ( ) 1 (p q) ( ) R ( ) where ( ) 2 f0;B ( )g: The lender will exercise monitoring of the borrower if and only if p (5) ( ) B ( ) p q Assuming that monitoring costs increase with distance, monitoring is less likely to occur at further distances if [B ( ) b ( )] is non-increasing as dis- tance increases. Then, the lender will make a loan decision as follows: n pB( ); p ( ) R ( ) I ( ) = ( I if inf p ( )R ( ) (p q) 0 otherwise ( )o > 1 Furthermore, the literature on group lending also argues that group lending could be a response to moral hazard when group monitoring is more e¢ cient than individual monitoring. However, how group size will change as distance increases is uncertain as the tension between free-riding and insurance is unlikely to be systematically correlated with distance from MFIs'o¢ ces. Proposition 4: If credit markets are characterized by moral hazard: (i) Properties (i) and (ii) of Proposition 1, and properties (i), (ii) and (iv) of Proposition 2 still hold (ii) Group lending is more likely to be observed at larger distances from micro...nance o¢ ces (iii) If the informational rent (measured by private bene...t B ( )) is non- decreasing with distance, then interest rates r ( ) should decrease: @r ( )=@d < 0 (iv) If the informational rent (measured by private bene...t B ( )) is non- increasing with distance, then the probability of monitoring decreases with distance. 3 Empirical results In this section, we will test the predictions of the three stylized models analyzed previously. We summarize the main implications of the models: 9 1. The intensity of screening increases with distance: at the margin, bor- rower characteristics should improve as distance increases, while the predictions for the average borrower are ambiguous. 2. A pre-loan visit by MFI o¢ cials is evidence of adverse selection: across borrowers the probability of a pre-loan visit is independent of distance. 3. Post-loan monitoring visits by MFI o¢ cials are evidence of moral haz- ard: ceteris paribus, monitoring is less likely to occur as distance in- creases. 4. In the moral hazard context: If the informational rent is nondecreasing with distance: rates of de- fault increase with distance, while predictions on monitoring and in- terest rates are ambiguous. If the informational rent is nonincreasing with distance: monitoring de- creases with distance and interest rates increase with distance, while predictions on default are ambiguous. 3.1 Data Data were collected in July 2005. The speci...city of the data is that they both include information on the MFI's side, and the socio-economic infor- mation on clients: this approach allows to record information on clients' assets, education, and household composition, which are not collected by the MFI. The questionnaire was also especially designed to look at the issue of distance. The population of reference consists of MFI's clients in 5 of the 7 re- gions of Niger. Among the 59 MFIs with an authorization of the Ministry of Finance to o¤er micro...nance services (BCEAO, 2002), 10 were chosen based on availability and regional location. Moreover, as the focus of the present study was the investigation of the impact of distance on MFI perfor- mance, sampling of borrowers has been strati...ed by distance. Each sampled MFI branch was asked to sample clients by stratum. Of all the applications received by these 10 MFIs in the last 5 years, 191 loan applications were selected as survey sample. Those applications were requested according to di¤erent distance ranges to individuals and groups. Subsequently, a group and household questionnaire would be administered according to the infor- mation provided by the MFI to locate its clients. The ...nal sample consists of 161 clients who have had a ...nancial relationship with 10 MFIs. Statisti- cal weights were applied at the 3 levels of strati...cation: di¤erent regions of 10 the country, di¤erent sizes of MFI and di¤erent distribution of each MFI's clients. The ...rst weight accounts for the number of MFIs operating in a re- gion. The second addresses di¤erences in size between the MFIs, and ...nally the third weight corrects the selection of households in di¤erent distance ranges.1 Table 2 summarizes the main socio-economic variables for these individ- uals for di¤erent types of clients: as we can see, borrowers are around 50 years old and those in rural areas have lower household income per capita. On average, families are more numerous in urban contexts. Our analysis will restrict to group lending, as individual lending is found to be mostly restricted to urban areas, in the vincinity of micro...nance o¢ ces. The methodology we will follow consists of comparing groups characteristics, contractual forms, monitoring activities and outcomes as distance between clients and the MFI increases. 3.2 Regression results The canonical regression analysis consists of the following Yi = + disti + Xi + "i where Yi is the outcome of interest, disti is the distance that separates borrowers from the MFI's o¢ ce, and Xi is a vector of group characteristics. Observations are weighted to account for our sampling strategy, and our regressions account for heteroskedasticity. 3.2.1 Group characteristics First and foremost, groups are more likely to be found at larger distances (Table 1). While no individual lending is observed when the distance ex- ceeds 25 kilometers, group lending takes place as far as 230 kilometres from the MFI's o¢ ces. A complete information framework cannot explain the emergence of joint liability. We interpret the dominance of group lending as evidence of adverse selection or moral hazard that are more expensive to address as distance increases. However, there is no clearcut prediction on the relationship between group size and distance, and no empirical pattern emerges either (results not shown). 1The computation of the weight to be applied for group i is thus given by expression: TotalWeighti = RegionMFIweighti MFIsizeweighti Distancei 11 Table 3 looks at average group characteristics. In terms of our model, we are interested in looking at the behavior of E [eje e (d) ; d], as d in- creases, where e is a vector of group characteristics that are believed to be relevant for creditworthiness. Proposition 1 predicts that if E [ejd] is non- decreasing with e, then if selection actually takes place, we should observe @ e (d) ; d] > 0: @d E [eje Looking at gender, women in Niger are associated with higher creditwor- thiness, anecdotically because the stigma of default is more severe on them. In Paxton's terms we could say that women are more vulnerable to peer pressure. Furthermore, we do not expect to see the demand for credit to exhibit a gender gradient with distance, thus Proposition 1 predicts a higher likelihood that borrowers are female groups as distance increases. This pre- diction is supported by the positive correlation between group gender and distance displayed in Table 3, column (1). The coe¢ cient can be interpreted as follows: a group of borrowers at a 100 kilometer distance from the MFI center is 40 percent more likely to be a female group than the average group of borrowers at a 10 kilometer distance. On the other hand, if e measures education, we expect @ @dE [ejd] < 0 : average education levels in the population decreases as we move further away from urban centers, where MFI o¢ ces usually are. Thus, intensive margin and extensive margin e¤ects give ambiguous results in terms of education levels of the average borrower. This fact is consistent with the absence of signi...cant relationship between education and distance (Table 3, columns (2) to (7)). Finally, if we look at the sector of activity of the average borrower, cash generating activities are safer investments from a lender's standpoint. Thus, similarly to groups'gender, Proposition 1,3 and 4 all predict a bias of lending activities towards short term projects as distance increases. Columns (8) to (10) in Table 3 show a positive correlation between distance and the fact that the loan was primarily used to ...nance short-term trading activities. 3.2.2 Loan characteristics We now turn to the determinants of the ...nancial contract. in Table 4, columns (1) to (4) display the determinants of loan amounts, whether it is the reported amount that was applied for (columns (1) and (2)), or the amount granted (columns (3) and (4)). First, we remark that higher ed- ucation of the head of the group is associated with larger loans. Second, as distance increases, the amount of loan granted decreases. This result is consistent with the three proposed models of credit markets provided that 12 projects have decreasing net present value as distance increases. Columns (5) to (7) suggest that while pre-loan visits are common, their frequency is independent of the distance that separates the borrower from the lender. These results are consistent with the adverse selection case and Proposition 3, whereby all loan recipients should have equal probability of a visit, irrespectively of their distance. Finally, if we look at interest rates charged, we ...nd a strongly positive and statistically signi...cant relationship with distance. This result is con- sistent with Proposition 2 and 3, whereby interest rates internalize the net present value of the project. In the moral hazard case, Proposition 4 sug- gests that the only possibility for interest rates to increase with distance while net present values are decreasing, is that the information asymmetry B ( ) must decrease with distance. This is a plausible explanation if one assumes that B ( ) captures the outside option of the borrower in case of failure. When borrowers are further away, it is likely that reputation is more important as there is no alternative credit institution to turn to in case of default. Thus, if B ( ) decreases with distance d su¢ ciently rapidly, then (4) can yield a negative gradient betwen interest rates and geographic distance. 3.2.3 Loan performance Table 5 con...rms the presumption that B ( ) does not increase as d becomes large: columns (1) to (3) show that borrower monitoring is less likely as necessary traveled distances increase. This suggests that the bene...ts from monitoring (B (d) b (d)) do not increase as fast as monitoring costs (d). Yet, columns (4) to (11) in Table 5 suggest that in the absence of monitoring, repayment is still improved as distances increase. Thus, in an economy in which monitoring is taking place, for lower monitoring frequencies to be associated with higher repayment rates, it must be the case that information asymmetry is less severe as monitoring costs increase. We postulate that a rationale underlying decreasing moral hazard has to do with the increasing importance of reputation among borrowers: a default might be sanctioned by loan denial in the future as there are few ...nan- cial institutions far away from urban centers. Thus, since clients have less possibilities of getting access to ...nance, travel costs usually put MFIs in a situation of de facto monopoly power, that in turn acts as a monitoring device. 13 4 Conclusion We started the paper with the presumption that distance would impose a cost on micro...nance development. However, how the cost would be trans- mitted to borrowers was an open question. The analysis conducted here suggests that the cost is born in part by borrowers who face higher interest rates, and more constraints and delays to obtain a loan, but is certainly also faced by marginal borrowers who are then excluded from the semi-formal credit market. The marginal borrower is moving higher up in the income distribution as distance increases. Moreover, we argue that adverse selection and moral hazard also plague the credit market, increasing transaction costs. Finally, we speculate that distance is also associated with the rarity of mi- cro...nance institutions, putting providers of ...nance in a monopoly situation that acts as a monitoring tool for borrowers: reputation of creditworthiness is more important to preserve as distance increases. We do not make any normative statement on whether the micro...nance sector is developed enough or not. The results shown in this paper bring back the tension that exists between outreach and ...nancial sustainability. If micro...nance institutions need to be sustainable, they need to manage their portfolio carefully. Distance will de facto impose a risk on their portfolio, so that they will need to screen the demand for ...nance accordingly. There is therefore an intrinsic contradiction between outreach and sustainability that is exacerbated by low population density, making distance an important parameter in this tradeo¤. Our ...ndings also suggest that limiting outreach has important conse- quences, as distant clients are more likely to be traders, while producers (especially breeders) are more likely to be left out of the credit market. Be- yond the e¢ ciency concern, we also raise the equity concern whereby the poorest of the poor might be more likely to be excluded as they live further away from economic centers, and are engaged in activities that would not be considered creditworthy for MFIs. Screening policies implemented by MFIs to maintain the quality of their portfolios will potentially hurt the poor (see also Paxton and Fruman, 1997). The tradeo¤ between sustainability and outreach therefore deserves more attention in academic research and policy discussions. 14 References [1] Ausubel, L. (1999), "Adverse Selection in the Credit Card Markets", Working Paper, Department of Economics, University of Maryland. [2] Calem, P. and L. Mester (1995), "Consumer Behavior and the Stickiness of Credit-Card Interest Rates,"American Economic Review, 72(4): 836- 40. [3] Edelberg, W. (2004), "Testing for Adverse Selection and Moral Haz- ard in Consumer Loan Markets", Finance and Economics Discussion Series 2004-9, Board of Governors of the Federal Reserve System. [4] Fafchamps, M. and C. Moser (2003), "Crime, Isolation, and Law En- forcement", Journal of African Economies. [5] Fafchamps, M., and J. Wahba, (2004) "Child Labor, Urban Proximity, and Household Composition", Economics series Working paper, 213. University of Oxford. [6] Fujita, M., P. Krugman, and A. Venables, (1999). The Spatial Economy: Cities, Regions, and International Trade. Cambridge and London: MIT Press. [7] Ghatak, M. (1999), "Group Lending, Local Information and Peer Se- lection", Journal of Development Economics, 60(1). [8] Jacoby, H. (2000), "Access to Markets and the Bene...ts of Rural Roads", Economic Journal, 110: 713-37. [9] Paxton, J. (1995), "Case Study: Le Projet de Promotion du Petit Crédit Rural, Burkina-Faso", Sustainable Banking with the Poor, The World Bank. [10] Paxton, J. and C. Fruman (1997), "Sustainable Banking with the Poor ­Outreach and Sustainability: a Comparative Analysis of Saving-First vs. Credit-First Financial Institutions", Report # 24832, The World Bank. [11] Petersen, M. and R. Rajan (2002), "Does Distance Still Matter? The Information Revolution in Small Business Lending", The Journal of Finance, 57(6): 2533-70 15 [12] Redding, S. and A. Venables (2002), "The Economics of Isolation and Distance", Nordic Journal of Political Economy, 28(2): 93-108. [13] République du Niger (2001), "Stratégie Nationale de la Micro...nance", Version provisoire. [14] Systèmes Financiers Décentralisés de la Banque Centrale des Etats de l'Afrique de l'Ouest (BCEAO), Monographies 2003-2004; http://www.bceao.int/ [15] Systèmes Financiers Décentralisés de la Banque Centrale des Etats de l'Afrique de l'Ouest (BCEAO), Dispositif Institutionnel (2006); http://www.bceao.int/ [16] The Seep Network Poverty Outreach Working Group (2006) "Micro- ...nance and Non-Financial Services for Very Poor People: Digging Deeper to Find Keys to Success", http://www.seepnetwork.org/ [17] United Nations (2006), World Statistics Pocketbook, http://data.un.org [18] World Bank (2008), Country Brief, http://www.worldbank.org/ 16 Table 1: Population density* and microfinance development** indicators Indicator Benin Burkina Faso Cote d'Ivoire Mali Niger Senegal Togo Density population (People / km²) 77 50 57 11 11 61 111 Institutions 79 41 21 93 61 290 59 MFI offices 449 425 220 858 170 721 203 Clients 690,428 742,618 575,050 656,092 94,096 777,379 299,674 Deposits (Million FCFA) 39,240 34,988 46,651 26,511 3, 856 55,326 20,262 Outstanding loans (Million FCFA) 59,678 29,466 19,337 34,142 4,380 67,906 16,997 Risky credits (Million FCFA) 1,436 1,705 1,024 2,037 415 2,442 1,748 Net result (Million FCFA) 3, 536 532 -1,123 895 35 4,389 -703 Outstanding loans / Population (FCFA) 8,309 2,301 1,082 2,799 360 5,964 2,839 Deposits / Population (FCFA) 5,464 2,732 2,610 2,173 317 4,859 3,384 Clients/ Population (Ratio) 0.096 0.058 0.032 0.054 0.008 0.068 0.05 Outstanding loans / Clients (FCFA) 86,436 39,679 33,627 52,038 46,548 87,353 56,718 Deposits / Clients (FCFA) 56,834 47,114 81,125 40,407 40,979 71,170 67,613 Risky credit/ Clients (FCFA) 2,080 2,296 1,781 3,105 4,410 3,141 5,833 MFI offices/ 100 Clients (Ratio) 0.65 0.57 0.38 1.31 1.81 0.93 0.68 Employees / 1,000 Clients (Ratio) 2.26 2.76 1.46 3.92 3.59 2.47 3.13 * Source: UN, 2006 ** Source: SFD BCEAO 2003 & 2004 Table 2. Summary indicators Group clients Individual clients Difference Robust Variable Avrg. Std Dv. Min Max Obs Avrg. Std Dv. Min Max Obs Standard error Distance (km) 107.900 65.942 0 230 78 1.540 3.198 0 25 83 106.3*** 14.330 Age (number) 51.445 11.781 24 80 68 45.638 9.913 27 88 78 5.8* 2.948 Sex female (%) 0.987 0.113 0 1 75 0.187 0.392 0 1 83 0.000 0.000 Children (number) 5.739 2.041 0 19 66 7.801 4.498 0 21 82 -2.1** 0.836 Household income (thousand FCFA) 19.243 49.244 0 316,000 59 111.066 99.715 15,000 455,000 66 -91.82*** 23.720 Household income per person (thousand FCFA) 2.017 4.236 0 30,000 55 11.374 10.443 1,785.7 54,800.0 64 -9.4*** 2.310 Occupation commerce (%) 0.645 0.482 0 1 78 0.082 0.276 0 1 83 0.563*** 0.125 Occupation breeding (%) 0.059 0.237 0 1 78 0.000 0.000 0 0 83 0.059 0.043 Occupation agriculture (%) 0.099 0.301 0 1 78 0.017 0.130 0 1 83 0.082 0.051 Occupation handicraft (%) 0.001 0.026 0 1 78 0.051 0.222 0 1 83 -0.051 0.050 Object of loan commerce (%) 0.653 0.479 0 1 78 0.412 0.495 0 1 83 0.241 0.156 Object of loan breeding (%) 0.127 0.335 0 1 78 0.027 0.163 0 1 83 0.100 0.090 Object of loan agriculture (%) 0.376 0.488 0 1 78 0.151 0.360 0 1 83 0.23* 0.136 Object of loan handicraft (%) 0.002 0.044 0 1 78 0.000 0.000 0 0 83 0.002 0.001 Object of loan education (%) 0.000 0.000 0 0 78 0.111 0.317 0 1 83 -0.111 0.068 Object of loan social events (%) 0.000 0.000 0 0 78 0.130 0.338 0 1 83 -0.130* 0.068 Object of loan housing 0.000 0.000 0 0 78 0.094 0.294 0 1 83 -0.094* 0.052 Amount applied for (thousand FCFA) 755.346 650.470 100,000 11,085,000 77 225.227 348.436 10,000 2,400,000 83 530.12*** 85.740 Amount applied for, per group member (thousand FCFA) 57.623 30.579 6,383 363,636 73 225.227 348.436 10,000 2,400,000 83 -167.60*** 43.320 Amount applied for, per beneficiary (thousand FCFA) 58.756 33.416 6,383 318,182 63 225.227 348.436 10,000 2,400,000 83 -166.47*** 43.370 Amount granted (thousand FCFA) 705.258 668.290 100,000 9,000,000 75 219.204 294.057 9,000 1,700,000 81 486.05*** 79.840 Amount granted, per group member (thousand FCFA) 50.693 27.852 6,250 545,455 71 219.204 294.057 9,000 1,700,000 81 -168.51*** 36.640 Amount granted, per beneficiary (thousand FCFA) 51.403 30.821 6,250 318,182 61 219.204 294.057 9,000 1,700,000 81 -167.80*** 36.690 Interest rate (%) 2.797 0.435 1 3.125 30 2.495 0.052 2 2.5 64 0.303 0.192 Visited by officer before granting loan (%) 0.994 0.078 0 1 69 0.510 0.503 0 1 81 0.49*** 0.091 Processing time (days) 30.361 54.984 1 356 76 12.176 5.358 3 30 80 18.2** 8.264 Frequency of reimbursement (months) 5.898 0.751 1 12 77 1.408 1.538 1 12 67 4.5*** 0.209 Defaulted on loan (%) 0.878 0.331 0 1 47 0.664 0.477 0 1 52 0.214* 0.120 Delay in reimbursement (days) 1,034.820 801.233 2 1,770 29 58.956 47.738 1 390 31 975*** 229.636 Group members (number) 15.069 10.544 7 130 74 N/A N/A N/A N/A N/A Number of beneficiaries in group (number) 14.527 9.654 5 50 64 N/A N/A N/A N/A N/A * significant at 10%; ** significant at 5%; *** significant at 1% Table 3: Linear regression of the determinants of group characteristics Dependent variables Female group Attended school (1:yes,0:no) Years of education Purpose of loan trade (1:yes,0:no) Independent variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Logarithm of distance 0.038*** -0.045 -0.041 0.046 -0.047 0.040 0.707 0.358*** 0.375*** 0.083 [0.014] [0.047] [0.049] [0.098] [0.803] [0.833] [1.004] [0.127] [0.133] [0.166] Female group (1:yes,0:no) -0.108 0.149 -2.683 -0.609 -0.528 -1.442 [0.197] [0.295] [2.775] [3.927] [0.375] [0.911] Interaction distance and group gender dummy variable -0.090 -0.686 0.296 [0.110] [1.324] [0.285] Education of head of group (years) 0.020 0.019 0.021 [0.014] [0.014] [0.019] Log of group size -0.151 -0.168 -0.195 [0.187] [0.190] [0.239] Log of amount granted -0.033 [0.213] Number of observations 74 68 68 68 59 59 59 59 59 56 R-squared 0.06 0.01 0.02 0.02 0.00 0.01 0.01 0.28 0.29 0.24 Notes: Robust standard errors in brackets. ***, **, * indicate that the coefficients are statistically significant at the 1 percent, 5 percent and 10 percent confidence level respectively. Table 4: Linear regression of the determinants of the demand and supply of loans (groups only) Dependent variables Log amount applied for Log amount granted Credit officer made a Log of interest rate (per group member) (per group member) pre-loan visit (1:yes,0:no) Independent variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Logarithm of distance -0.363*** -0.346*** -0.267* -0.247 0.022 0.017 0.015 0.150*** 0.116*** 0.082** [0.120] [0.126] [0.142] [0.150] [0.014] [0.015] [0.019] [0.042] [0.023] [0.031] Female group (1:yes,0:no) -0.537 -0.466 0.176 0.200 0.305*** 0.251** [0.326] [0.462] [0.176] [0.177] [0.102] [0.108] Education of head of group (years) 0.031** 0.030* 0.058*** 0.057*** 0.000 0.000 -0.001 -0.003 0.002 0.008 [0.015] [0.015] [0.015] [0.015] [0.001] [0.001] [0.001] [0.006] [0.008] [0.008] Log of group size -0.818*** -0.836*** -0.829*** -0.849*** -0.007 -0.001 -0.008 -0.141 -0.105** 0.014 [0.165] [0.165] [0.172] [0.177] [0.010] [0.006] [0.009] [0.113] [0.046] [0.065] Log of amount granted 0.020 -0.062 [0.021] [0.038] Number of observations 58 58 56 56 53 53 50 22 22 21 R-squared 0.52 0.52 0.51 0.51 0.05 0.10 0.14 0.58 0.79 0.84 Notes: Robust standard errors in brackets. ***, **, * indicate that the coefficients are statistically significant at the 1 percent, 5 percen Table 5: Linear regressions of the determinants of loan performance (groups only) Dependent variables Loan monitoring by MFI? (1:yes,0:no) Any delayed repayment? (1:yes;0:no) Repayment delay (log of number of days) Independent variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Logarithm of distance -0.123*** -0.118*** 0.017 -0.005 -0.047 -0.044 -1.171* -1.172* -1.196* -1.196 [0.042] [0.043] [0.066] [0.065] [0.057] [0.059] [0.650] [0.670] [0.675] [0.697] Female group (1:yes,0:no) -0.605*** -0.649*** -0.341 -0.414 -0.864*** -0.746*** 4.309* 4.879** -0.454 0.198 [0.105] [0.120] [0.211] [0.246] [0.151] [0.240] [2.160] [2.099] [2.063] [2.117] Education of head of group (years) 0.015* 0.016* -0.026 -0.026 -0.022 -0.023 -0.145*** -0.145*** -0.144*** -0.145*** [0.008] [0.010] [0.029] [0.030] [0.030] [0.031] [0.038] [0.039] [0.039] [0.040] Log of group size 0.139** 0.132** -0.232 -0.217 -0.171 -0.178 -0.401 -0.405 -0.373 -0.377 [0.059] [0.060] [0.162] [0.169] [0.166] [0.179] [1.301] [1.343] [1.348] [1.394] Loan officer made a pre-loan visit (1:yes,0:no) 0.109 0.458 -0.049 -2.141*** -1.546 [0.243] [0.353] [0.180] [0.684] [1.422] Regular follow up activities by MFI(1:yes,0:no) -0.700** -0.579 -4.998* -4.747 [0.266] [0.358] [2.566] [2.816] Number of observations 56 50 34 31 33 30 22 22 22 22 R-squared 0.49 0.49 0.24 0.25 0.32 0.28 0.37 0.38 0.39 0.39 Notes: Robust standard errors in brackets. ***, **, * indicate that the coefficients are statistically significant at the 1 percent, 5 percent and 10 percent confidence level respectively.