Microfinance Default Rates in Ghana: Evidence from Individual-Liability Credit Contracts

Microfinance, however measured, has increased rapidly in Ghana since the start of the present decade, growing by 20-30 percent annually.  Microfinance Institutions (MFIs) currently provide financial services to an estimated 15 percent of the country’s total population compared with 10 percent for the commercial banking sector.

Rural and community banks account for the lion’s share of MFI activity in Ghana, representing more than half the total number of microfinance borrowers and a similar proportion of the sector’s total loan portfolio (Aryeetey:2008).  NGOs, by contrast, are comparatively unimportant: the average loan size is roughly one third that provided by rural and community banks and an even smaller fraction (25 percent) of the amount borrowed from savings and loan companies.  On the other hand, loan repayment rates, at a reported 99 percent, are considerably higher among financial NGOs than among other microfinance providers or government-sponsored lending programs; their loan loss exposure is also relatively modest.  

Finally, and perhaps not too surprisingly given their heavy dependence on donors or official sources of finance, financial NGOs have the worst record of achieving either operational or financial self-sufficiency, surpassed only by government-sponsored programs. The clear implication here is that without substantial subsidy, interest rates on loans provided by both NGOs and state-supported institutions would be significantly higher, with an attendant negative impact on repayment obligations. 

Ghana’s commercial microfinance operations have, by contrast, broadly achieved a degree of operational efficiency that compares favorably with medium sized African financial institutions or worldwide MFIs; even so, in terms of financial sustainability, many still have a long way to go.[1]  The purpose of this paper is to investigate repayment rates among MFIs that follow the individual-liability lending model; since this model closely approximates to that pursued by commercial banks, a close correspondence might be expected between the two approaches.[2]

Our analysis derives from data on loan repayments and borrower characteristics provided by ProCredit (Ghana), a local microfinance institution, which operates on the basis of individual liability lending. Individuals and micro-entrepreneurs that apply for ProCredit loans proceed through three stages prior to obtaining approval.

1.Preliminary Screening 

In this stage, loan applicants make contact with the institution and are carefully screened and asked to answer specific questions regarding the status of their business and household accounts, in order to establish whether they qualify under ProCredit‘s eligibility guidelines.

2.Loan Proposal and Credit Committee 

Loan applicants are assigned to specific loan officers. Applicants undergo a further review to verify the information taken at the initial stage, and a visit to the applicant’s businesses and household is arranged.  The information thus developed is organized into a formal loan proposal and presented to the lending institution’s credit committee for approval.  The loan amount and tenure are confirmed conditional on the adequacy of the cash flows generated by the borrower’s business, sufficient personal collateral and guarantors agreeing to co-sign the loan agreement.

3.Monitoring and Repayment

After disbursement, the account officer frequently visits the borrower’s business to ensure that the proceeds are being used for the specific purpose(s) for which the loan was granted, and to remind borrowers of their next repayment date.  Borrowers who miss payments are pressured at this stage; if the arrears continue, legal action is initiated against the borrower and guarantor(s) to recover any amounts owed, but usually after the designated collateral has been seized and liquidated.

1.  Loan Sample Data  

 The data used here are drawn from ProCredit’s lending files.  Six of the bank’s twelve branches, including one located outside the national capital, were selected for the study; the remaining branches were all established fairly recently and thus have relatively small loan portfolios.

The total sample consists of 960 loans made to local businessmen from the database of the institution’s six branches, five of which are located in Accra and one in Kumasi. The sample consists of loans granted and repaid (or not) between January 2002 and December 2007, and comprise 160 loans from each branch.  The sample thus consists of 720 repaid and 240 defaulted loans, with individual loans in each category chosen randomly.  To obtain a fair representation of the characteristics of defaulted borrowers, we deliberately over-sampled this category, a decision motivated by the low actual default rate.  The sample dataset was audited for errors and omissions to ensure consistency and uniformity. 

Twenty-four borrower characteristics were extracted from the data and grouped into four main categories cross-classified by borrower status (Table 1). 

(1) Individual borrowers’ household characteristics (gender, age, marital status, household income not generated from either the business or earnings of dependants).  Dependants consist of the number of people in the household who rely on the business income. Households with fewer dependants have a smaller claim on their business income, which should serve to reduce the default rate.  The borrower’s marital status is also expected to lower the likelihood of default; working spouses generate an independent income, thus increasing the financial resources available to service the loan, in contrast to borrowers who are single, divorced or widowed, where there are no supplementary earnings.  We also expect the probability of the loan being repaid to increase if the borrower is a woman, in keeping with empirical evidence to that effect.

(2) Savings behavior (default and non-default borrowers’ saving behavior during the term of their respective loans).  Borrowers who save during the term of the loan build up a cash reserve that can be used to service the debt during periods when the business is facing liquidity difficulties. The presence of savings is expected to increase the probability of the loan being repaid. 

(3) Business characteristics (business type, age of the business, location of the business [Branch]). The number of years the borrower has been in the same business should increase the probability of the loan being repaid; there is again ample evidence showing that established businesses are less prone to experiencing financial distress than are newly created ventures. 

(4) Loan characteristics (loan amount in Ghana cedis, loan purpose, loan monitoring, collateral type and value in Ghana cedis, term of loan, loan status, number of guarantors [co-signers]).  Each loan is unique in terms of loan amount, tenure, collateral and the number of co-signers who act as guarantors for the credit.  A greater number of guarantors and a high collateral-to-loan ratio should be consistent with lower default risk; so, too, should the intensity with which loans are monitored (but see below).   Loan status indicates whether the borrower had obtained prior loan(s).  Loan tenures are of variable length, though longer maturities appear consistent with a lower risk of default; for a given interest rate, longer maturities imply lower periodic installments.  Finally, loans used for working capital or stock accumulation appear less risky than then those used for acquisition of fixed assets.  

A larger percentage of defaulted borrowers in our sample are single or divorced and younger on average, and a relatively larger fraction is made up of women (48 percent in the defaulted group compared with 43 percent among borrowers who repaid their loans).  Defaulted borrowers also have a relatively larger number of dependents than their non-defaulted counterparts.  When the various household characteristics are subject to formal statistical analysis, the only variable shown to differ significantly is the borrower’s age; defaulted borrowers on average were eight years younger than non-defaulted borrowers.

The majority of borrowers in the default category have less than five years experience running their businesses.  Borrowing for the purpose of adding to stock accounted for 50.1 percent of all loans; working capital loans or loans to purchase fixed assets constitute 17.9 percent and 32 percent, respectively, of the total sample.  Statistical analysis confirms that number of years in business is an important determinant of default, in contrast to the purpose of the loan, though a higher incidence of working capital loans among repaid loans is marginally significant indicating a favorable impact on the probability of repayment.  

Loans offered fall into two broad categories: those above GHC1,000 are described as loans to small and medium enterprises (SME) and micro loans, while loans below GHC1,000 are known as ‘express’ loans. The majority of loans in the sample were express loans (55.2 percent), with a greater proportion of defaulted loans (54.2 percent) falling into the micro and SME loan categories; this compares with repaid express loans of 57.6 percent.  Loan status indicates whether the borrower is a new client obtaining his/her first loan or is a repeat borrower; 63 percent of borrowers fall into the former category.  Interestingly, the majority of defaulted borrowers were repeat not new clients, a statistically significant finding.

Collateral coverage is measured as the ratio of the collateral value to the loan amount. For the majority of clients in the sample (60.5 percent) this ratio exceeded 150 percent; coverage differences are highly significant.  Each loan was guaranteed by at least one guarantor, who also acted as a co-signer of the loan contract; 56.5 percent of borrowers in the total sample had their loans guaranteed by at least one guarantor.  Among borrowers that repaid their loans, nearly one half had more than one guarantor. 

Loan monitoring is part of the loan cycle: loan officers visit the residence and business of each borrower before and after loans are made to ensure that the proceeds are used only for the stated purpose and that the business/project is being run efficiently.  Regular visits also serve to strengthen the relationship with the borrower, encouraging repayment while simultaneously gathering information concerning the state of the business and household finances, all of which should be consistent with a lower default rate.  By contrast, more frequent visits could be taken as evidence that borrowers are experiencing repayment difficulties, higher frequency indicating greater severity.  The data appear more consistent with the second interpretation: defaulted loans were monitored more frequently than repaid loans, while statistical analysis confirms that the differences were significant.

Loan maturities range from 4-12 months, though fixed asset loans are sometimes extended for up to 18 months. Sector indicates whether the borrowers’ main business is in services, trade (buying and selling), or production (manufacturing).  The majority of borrowers operated in the trade sector (58.8 percent), with an average loan maturity of up to 12 months. Statistical analysis indicates that business sector does not matter, though the higher incidence of default among firms operating in the trade sector is marginally significant.  Loan maturities, too, do not appear to be important, with the slightly higher percentage of shorter maturities among defaulted loans being statistically insignificant.  Finally, while the data indicate that the percentage of borrowers who saved over the life of the loan was higher among repaid than defaulted loans, the differences are insignificant. 

Another way of assessing the extent to which borrower, business and loan characteristics affect repayment is to present Odds Ratios (OR) as shown in Figure 1.[3]  On this basis, borrowers having more than three dependents and operating in the service sector obtained larger loans with longer maturities, used the proceeds to finance fixed investments, lacked non-business income, and were at greater risk of default.  These findings confirm the bivariate results. 

2. Multivariate Results

Pair-wise comparisons are illustrative, but fail to take proper account of the interactions that exist among the explanatory variables.  Given that the main rationale for this study is to identify and analyze the factors that influence loan repayment rates in microfinance institutions, the way forward is to employ multivariate statistical procedures better able to achieve that objective.  The technique chosen, logistic regression, is perhaps the best of several statistical procedures that can be used when analyzing conditional data. 

In the present study, default probability, the dependent variable, is ascribed a value 1 if a given loan defaulted and 0 otherwise, with default related to the various independent variables enumerated above.  A direct logistic regression was fitted for each of the independent variables, except for the various branches (Tema, Madina, Kaneshie, Tudu, Kokomlemle and Suame) that were used in this study. The estimation results (not shown) indicate that seven of the independent variables are statistically significant at the 5 percent level or higher; other household, business and loan characteristics did not have any significant effect on the probability of loan default.

1. Other non-business income (OR = 0.5793). A unit increase in household non-business income leads to a reduction in the relative ratio of the default probability to repayment by a factor of 0.5793; that is, as the presence of other income separate from business income increases, the rate of credit default declines by 42 percent.  Given that the majority of borrowers were married, this suggests that in most instances their partners either operated income-generating businesses or were working in paid employment.[4]  This finding is consistent with a study undertaken among borrowers in Caja Los Andes, Bolivia, which indicates that borrowers with higher non-business income are less likely to default on their loan obligations (Vogelgesang: 2003). 

2. Loan status (OR = 0.1802). The OR implies an 82 percent decline in the default rate among new borrowers compared to repeat borrowers.  This may reflect an incentive effect, with access to future loans dependent upon successful repayment of the current loan.  Knowing this, new borrowers prove themselves to be a good credit risk, a finding consistent with Armendariz and Morduch (2000), Bolton and Sharfstein (1990), and Churchill (1999) among microfinance institutions that employ individual-liability schemes. The result is also consistent with Vogelgesang (2003), who shows that loan repayment rates among repeat borrowers deteriorate compared to new borrowers.

3. Working capital (OR = 0.5126). Loans used for the purpose of augmenting working capital reduce default probabilities by 49 percent; this compares with an increased default rate of 62 percent for loans used to finance fixed investment [see below (8)].

4. Guarantors (OR = 0.3732). A unit increase in the number of guarantors produces a decline in the default rate by 63 percent.  This may be due to social pressures that guarantors bring to bear on recalcitrant borrowers, and may also be seen as social collateral with its impact on loan repayment.  This result is consistent with Gine and Karlan (2006), who show in a related study that the use of collateral coupled with social pressure among borrowers reduces default while increasing repayment; it is also consistent with the findings of a study undertaken in Bolivia (Schreiner: 1999).

5. Number of years in business (OR = 0.7176). As the number of years a borrower has been in business increases, the probability of default declines by 28 percent.  This confirms that as borrowers gain commercial experience, the resulting improved productivity leads to a significant reduction in the likelihood of default compared to their less experienced counterparts.  Alternatively, the effect may indicate that established businesses, with their assured revenues and diversified cash flows, represent better credit risks than younger firms.  There is considerable evidence that firms with long operating histories are less prone to financial distress than are more recently established businesses.

6. Collateral to loan ratio (OR = 0.8437). A unit increase in the collateral demanded by lenders as security for the loan lowers the likelihood of default by 16 percent, a finding consistent with (Villas-Boas and Schmidt-Mohr: 1999), who argue that as competition increases, so too does the demand for additional collateral by MFIs.  On the other hand, the variable is significant at only around the 10 percent level.   

7. Number of dependants (OR = 1.2234).  For each additional dependant in the household the probability of loan default increases by about 22 percent.  As potential claims against business income increase, this is likely to encourage the diversion of resources to direct household purposes (paying school fees, funeral pledges, or other social commitments). 

8. Fixed assets (OR = 1.6180).  Loans made for the purpose of acquiring fixed assets increase the likelihood of default by 62 percent, a result that appears to connect with the relatively long gestation before fixed investments (machinery, plant and building) generate a satisfactory cash flow. Compared with loans used for inventory investment, default is reduced by 20 percent though this effect was not statistically significant.

9.  Monitoring (OR = 1.4786).  Monitoring increases the likelihood of default by 48 percent.  This may be due to excessive pressure from the institutions’ agents encouraging borrowers to invest in high-risk projects in order to generate higher cash flows to repay the loan.  It may also reflect ‘collusion’ between loan officers and borrowers; evidence of such behavior is known, or perhaps it may be due to outright fraud (Todd: 1996).

To test the robustness of the estimated coefficients and by extension the odds ratios, alternative logistic regressions were run that excluded all of the statistically insignificant variables. Also, variables that were closely correlated with each other were alternated to determine which had the greater ability to classify and predict default. Finally, branch dummies were introduced to control for regional or neighborhood effects. 

The alternative regressions were consistent with each other and with those estimated using all of the independent variables.  Nor can any systematic differences be detected in the pattern of branch lending, suggesting that screening and credit procedures were applied consistently and uniformly.  By any of the widely used goodness of fit criteria, the results are virtually identical; nor do the alternative specifications materially alter the percentage of observations correctly classified.  

All in all, we conclude that the probability of default increases with the number of dependents, whether the proceeds are used to acquire fixed assets, and the frequency of monitoring, and decreases with the availability of non-business income, years in business, the number of guarantors, whether the proceeds were used for working capital purposes, and whether the client is a first time borrower.  The ratio of collateral-to-loan value is also associated with an increase in the repayment rate, though none of the estimated coefficients are significant at the 0.05 level of higher.

References

Aryeetey (2008): “From Informal Finance to Formal Finance in Sub-Saharan Africa: Lessons from Linkage Efforts,” Paper presented at the High Level Seminar on African Finance for the 21^st Century , IMF and Joint Africa Institute, Tunis, Tunisia (March).

Armandariz and Morduch (2000):”Microfinance Beyond Group Lending” Economics of Transition, 8 (401-420).

Bank of Ghana (2007): “A Note on Microfinance in Ghana, Research Department Working Paper WP/BOG-07/01 (August).

Bolton and Sharfstein (1990): “A Theory of Predation Based on Agency Problems in Financial Contracting,” American Economic Review. 80 (93-106)

Churchill (1999): Client-focused Lending: The Art of Individual Lending, (Calmeadow).

Gine and Karlan (2007): Group versus Individual Liability: A Field Experiment in the Philippines, Yale University Economic Growth Centre Working Paper 940 (May).

Jha, Negi and Warriar (2004): “Ghana: Microfinance Investment Environment Profile,” unpublished ms., Princeton University.

Schreiner (1999): “Scoring Arrears at a Microlender in Bolivia,” Journal of Microfinance 6, (70-86).

Todd, H. (1996): Cloning Grameen Bank: Replicating a Poverty Reduction Model in Indonesia, Nepal and Vietnam (Intermediate Technology Publications).

Villas-Boas and Schmidt-Mohr (1999): “Oligopoly with Asymmetric Information: Differentiation in Credit Markets,” The Rand Journal of Economics.30 (375–396).

Vogelgesang (2003): “Microfinance in Times of Crisis: The Effect of Competition, Rising Indebtedness, and Economic Crisis on Repayment Behavior” World Development.31 (2085-2114).