77931

The Impact of Microfinance Loans on Small Informal Enterprises in Madagascar.A Panel Data Analysis

Flore Gubert, a* François Roubaudb

April 2011

Abstract

This paper analyzes the impact of a microfinance institution (MFI) serving small informal enterprises in Antananarivo (Madagascar). The methodology consists of comparing over time the situation of a representative sample of clients’ enterprises with a control group, constructed in an almost experimental way through a standard propensity-score matching technique. Overall, the results indicate a positive impact of the project. Taken as a snapshot, the evaluations successively conducted in 2001 and 2004 indicate that the clients’ enterprises recorded better average performance than enterprises without funding. With a dynamic perspective however, the results are more nuanced. If the positive effect of the project is clear during growth phases, its effect during economic recessions appears less certain.

Keywords: Microfinance, propensity score matching, difference-in-differences estimator, Informal Sector, Microentreprise, Madagascar.

JEL Code: 016, D24

a IRD, UMR 225 DIAL, University Paris Dauphine and Paris School of Economics

b IRD, UMR 225 DIAL, University Paris Dauphine

* Corresponding author:Flore Gubert, DIAL, 4 rue d'Enghien 75010, France, Phone: +33 1 53 24 14 66, Fax: +33 1 53 24 14 51, E-mail: gubert@dial.prd.fr

Acknowledgements

This research is part of a project entitled “Unlocking potential: Tackling economic, institutional and social constraints of informal entrepreneurship in Sub-Saharan Africa” (http://www.iss.nl/informality) funded by the Austrian, German, Norwegian, Korean and Swiss Governments through the World Bank’s Multi Donor Trust Fund Project: “Labor Markets, Job Creation, and Economic Growth, Scaling up Research, Capacity Building, and Action on the Ground”. The financial support is gratefully acknowledged. The project is led by the International Institute of Social Studies of Erasmus University Rotterdam, The Hague, The Netherlands. The other members of the research consortium are: AFRISTAT, Bamako, Mali, DIAL-IRD, Paris, France, the German Institute of Global and Area Studies, Hamburg, Germany and the Kiel Institute for the World Economy, Kiel, Germany.

Disclaimer

This is work in progress. Its dissemination should encourage the exchange of ideas about issues related to microfinance and entrepreneurship. The findings, interpretations and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the World Bank, the donors supporting the Trust Fund or those of the institutions that are part of the research consortium.

1. Introduction

After the pioneering experiences of Grameen Bank in Bangladesh and BancoSol in Bolivia, the microfinance sector blossomed in many countries serving the needs of around 155 million customers throughout the world. This rapid progression has been strongly encouraged and sponsored by multilateral and bilateral aid donors whose support found expression at various occasions. During the Microcredit Summit which was held in 1997, the decision was taken by 137 countries to provide 100 million of the world’s poorest families with credit and other financial services for self-employment activities by 2005. One year later, in 1998, the United Nations General Assembly designated the year 2005 as the International Year of Microcredit. During the 10th Summit of heads of state and government of countries using French as a common language which was held in Ouagadougou in 2004, participants agreed to support microfinance institutions (MFI) and to improve their integration in the developing financial sector. More recently, the 2006 Nobel peace prize for Mohammed Yunus and the Grameen Bank he created stood as a proof that microfinance has become the hottest idea for solving poverty. This enthusiasm for microfinance contrasts with the lack of knowledge about its achievements in terms of poverty alleviation. The questions of whether MFIs actually reach and empower the poor and/or whether microfinance is better than some other types of development project for the poor is not settled once and for all. Quoting Zeller and Meyer (2003) in their book devoted to microfinance, “MFI field operations have far surpassed the research capacity to analyze them, so excitement about the use of microfinance for poverty alleviation is not backed up with sound facts derived from rigorous research. Given the current state of knowledge, it is difficult to allocate confidently public resources to micro-finance development”. Part of this knowledge gap is due to the fact that the evaluation of the impact of microcredit is a particularly difficult problem. Selection issues that are common to nearly all statistical evaluations are indeed particularly poignant for microcredit, and because of the fungibility of money, loans provided to microentrepreneurs to expand their business may have no impact on the firms' outcomes while having strong impact on a wide range of household outcomes. In addition, when the focus of the evaluation is on business outcomes, additional difficulties come from the fact that microenterprises are highly vulnerable and that follow-up surveys aimed at collecting longitudinal data on clients and non clients generally suffer from strong level of (non random) attrition. In this complicated context, randomized control trials have been embraced as the gold standard to get clean estimates of the difference made by microfinance, and recent randomized evaluations properly addressing selection issues have been conducted in Morocco, urban India, South Africa and the Philippines. In urban India, for example, Banerjee, Duflo, Glennerster and Kinnan (2009) took profit of the expansion of a large microlender, namely Spandana, to measure what happens when microcredit becomes available in a new market. Overall, they report a mix of economic results but no strong average impacts. In particular, measured impacts on health, education, and women’s empowerment are negligible. Another experiment based on a quite different design in the Philippines find that expanding access to credit is not associated with an increase in business investment, but with an increase in profit, particularly for men and for men with higher incomes (Karlan and Zinman, 2010).

Yet, while it is true that RCTs can be powerful tools to establish causal relationships between interventions and impacts, they also have drawbacks and limits. Beyond ethical reasons, which make randomization not always desirable, properly designing an experiment is not always feasible. This is for example the case when the demand for evaluation emanates from a microfinance institution that has fully achieved both its pilot and expansion phases. Randomizing treatment is indeed inoperative in this context since the entire community (or at least the targeted one) has already been exposed to the treatment.

At the time when ADéFI, a microfinance institution serving small informal enterprises in the main cities of Madagascar, was created, in 1995, the managing team committed itself to assessing the impact of its intervention. However, when this commitment translated into concrete actions, some years later, the geographic expansion of ADéFI's activity was over, and the MFI had 31 branches in the six regions of Madagascar, with no plans of further expansion. Designing a randomized experiment was thus in this particular setting inappropriate and bound to fail. A protocol was thus designed in a pragmatic way to obtain an impact study as careful and credible as possible. It consists of comparing the situation of a representative sample of clients’ enterprises with a control group, constructed in an almost experimental way through a standard propensity-score matching (PSM) technique. It also includes two follow-up surveys conducted among the two samples of (matched) treated and non-treated microentrepreneurs. The panel structure of the data makes it possible to combine PSM with the difference-in-difference (DD) method and to eliminate time-invariant additive selection bias. It also contributes to bring light on the mortality rate of microenterprises in the Malagasy context.

This paper provides further details on the data collection phase and presents the main results of the evaluation. Overall, the results indicate a positive impact of the project. Taken as a snapshot, the evaluations successively conducted in 2001 and 2004 indicate that the clients’ enterprises recorded better average performance than enterprises without funding. With a dynamic perspective however, the results are more mixed. If the positive effect of the project is clear during growth phases, its effect during economic recessions appears less certain.

The paper is organized as follows. Section 2 describes the Malagasy context in the field of microfinance and describes ADéFI's lending activity and clientele. Section 3 describes the data collection methodology, the empirical strategy and the data used in the estimations. Section 4 presents the results. Section 5 concludes and suggests extensions to the present study.

2. Microfinance programmes in Madagascar

General patterns

With a PPP per capita income of 980 dollars in 2009, Madagascar (20 millions of inhabitants) belongs to the list of the least developed countries (LDCs). After a long recession from 1960 to 1995 during which per capita gross domestic product (GDP) and private consumption respectively fell by 36.8% and 46.8%, Madagascar started experiencing growth in 1996. Growth accelerated during the following years and picked up sharply during the period 1999-2003 excluding 2002 when there was a political crisis. Growth recovered after 2003, but was again put to a halt after the political turmoil of 2009.

Credit market imperfections have been one of the structural constraints impeding transition since 1996. Composed of seven foreign-owned commercial banks, the formal banking sector remains poorly developed in rural areas and inaccessible to small-scale producers. As a result, credit markets in villages are dominated by informal moneylenders (neighbouring farmers, merchants, traders, landlords, etc.) who grant farmers with financial problems loans in cash or in-kind (paddy) at annual interest rates ranging from 120% to 400%. The situation is also worrisome in urban areas where commercial banks are often reluctant to give substantial loan amounts to small-scale entrepreneurs. According to the latest estimates, only 35% of low income households (roughly 80% of the population) have access to depository services and 2% to credit (IMF, 2006). This situation has resulted in the creation, in 1990, of the first MFIs in Madagascar which have been strongly supported by both the government and the international donor community since then. Today, three types of microfinance institutions can be found in Madagascar: (i) membership-based credit unions and savings and credit cooperative associations whose services are limited exclusively or primarily to members (URCECAM, TIAVO, OTIV, AECA, ADéFI); (ii) client-based credit institutions (SIPEM, Vola Mahasoa, APEM/PAIQ, APEM Farahitso); and (iii) NGO or associations whose activities include lending operations. Despite differences in technology and in market niche among existing MFIs, most of them have several traits in common: they make small and short-term productive credit loans; they charge monthly interest rates ranging from 2% to 4%; they offer poorly diversified savings products; and they have succeeded in keeping arrears and loan losses low although the share of their portfolio “at risk” has been increasing since the political crisis of 2002. Since 1999, the microfinance sector has rapidly grown in Madagascar with a portfolio of 143.7 billion Ariary in 2009 (US$ 71.8 million) against 22.7 billion in 2002. However, with less than 65,600 active borrowers in 2009, the microfinance coverage in Madagascar remains thin with only 14% of households covered by microcredit programmes.1

From a demand-side perspective, recent surveys conducted in urban areas between 1995 and 2004 using representative samples of small-scale informal enterprises (SIEs) provide first-hand information on the credit needs of the informal private sector.2 To begin with, the 2004 survey provides figures on the coverage rate of microfinance institutions in the capital city of Antananarivo. Overall, while 46.5% of microentrepreneurs have ever heard of the existence of some microcredit programmes, only 3.1% of them have had direct contact with an MFI. Moreover, among those microentrepreneurs that asked for a loan, less than 40% actually obtained it. Turning to credit needs, the 2004 survey reveals that 86% of the sample microentrepreneurs declare facing some problems and that lack of access to credit and an excessive cost of credit respectively rank sixth and seventh in the list of difficulties they encounter (Table 1). As a direct consequence, a better access to credit is claimed by more than a third of SIEs whatever their sector of activity, and by 46% of SIEs operating in the trading sector. Last, when asked about how they would use their loans, 42% of the sample microentrepreneurs say they would create another SIE, among which more than 50% would do it in another sector of activity (extensive growth). The others would either improve their equipment (18.5%), their premises (15.9%), their stock of raw materials (14.0%) or spend the money elsewhere (6.7%). By contrast, no microentrepreneur would hire new employees. This suggests that any policy aimed at promoting SIEs through easing access to credit would have negligible effect on the level of employment.

1 Sources: http://www.madamicrofinance.mg/resultats.htm, 10 november 2010; http://www.mixmarket.org /mfi/country/Madagascar?order=products_and_clients_total_borrowers&sort=desc, 10 november 2010)2 More details on these surveys are provided in Section 3.

Table 1. Main dificulties faced by SIEs, by sector (ranked in decreasing order)Total Industry Commerce Services

1. Lack of demand 75.6% 67.7% 85.2% 73.9%2. Excessive competition 55.6% 52.3% 64.1% 50.5%3. Lack of tools and machinery 30.3% 52.3% 8.2% 30.4%4. Lack of working space 28.7% 27.0% 31.7% 27.2%5. Difficulties in accessing to raw materials 27.6% 33.0% 32.1% 17.5%6. Lack of access to credit 24.5% 24.2% 32.1% 17.2%7. Excessive cost of credit 14.2% 13.7% 18.1% 10.7%8. Organisational or management constraints 10.5% 13.9% 6.4% 11.3%9. Excessive regulations and taxes 9.9% 4.2% 13.7% 11.9%10. Technical difficulties 9.9% 18.7% 2.7% 8.4%11. Difficulties in hiring qualified people 3.7% 6.9% 1.5% 2.6%12. Other 2.4% 1.2% 1.2% 4.9%None 13.7% 10.7% 8.1% 21.3%Source: 1-2-3 2004 Survey, phase 2, DIAL,INSTAT/Direction des Statistiques des Ménages. Note: Total in columns can be higher than 100% due to multiple answers.

ADéFI: a membership-based microfinance institution

In what follows, we focus on ADéFI, a membership-based microfinance institution which has been serving microenterprises in urban areas since it was created in 1995.3 With six regional centres and 31 credit offices, ADéFi is specialized in financing urban microbusinesses providing one-year individual loans averaging 500 euros. Since 2002, it has also started providing longer-term loans (from 24 to 36 months) to small and medium-sized enterprises (SMEs) averaging 8,000 euros. Depending on loan duration, interest rates vary between 16 and 18% per year with the first repayment installment due 1 to 6 months after the borrowing date. At the time of the 2004 evaluation, ADéFI had 6,217 clients in Antananarivo among which 50% were active clients.4 Full access to the customer database allowed us to get a clear view of the main characteristics of the clients. In terms of activity, nearly 40% were engaged in the production of services, among which about a half were in the transport sector. The remaining 60% were equally distributed between the industrial and trading sectors. A closer look at the clients operating in the industrial sector revealed that two thirds were in the clothing industry. With regards to firm size, 80% of the clients were microentrepreneurs with less than three employees. In most cases (57%), SIEs' activity was taking place inside the home of the business owner, with strong variations between sectors (in the clothing industry, for e.g., this share was as high as 72%). Last, with regards to microentrepreneurs' education and qualifications, most microentrepreneurs in the 2004 database (66%) went at least to secondary school, while only 15% or so went only to primary school.

3 In May 2010, AdéFI was converted from a cooperative status into a joint stock company and renamed Agence de Crédit pour l’Entreprise Privée à Madagascar (ACEP Madagascar). Its main activity remains lending to small and medium entreprises in urban Madagascar. 4 Clients are considered as non active by ADéFI if they have not requested a new loan for at least 18 months.

The comparison of ADEFI's clientele with the aforementioned representative sample of SIEs interviewed in 2004 brings additional insights on the characteristics of ADéFI's clients as opposed to non clients. With regards to the sector of activity, ADéFI's clients in 2004 were clearly over-represented in the transport sector and clothing industry. They were also more engaged in the trading of primary products. By contrast, they were under-represented in the construction sector, and also much less engaged in the trading of transformed products and in the production of services to households and firms. ADéFI's clients were also bigger, with 3 employees on average against 1.4 for SIEs in general, a higher turnover, more physical capital, etc. The business owners themselves had a specific profile, with a higher share of females and of highly-educated individuals among clients than among non clients. All this suggests that while ADéFI does serve loans to small firms, its clients are over-represented among the biggest of these small firms.

3. Data and empirical methodology

Data for the evaluation

This research uses a unique dataset made up of the results of various surveys (Figure 1). The first survey provides detailed information on a representative sample of 198 microfinance clients of ADéFI.5 The second survey, known as the 1-2-3 Survey on Employment (Phase 1), Informal Sector (Phase 2), and Household Expenditure (Phase 3), provides data on a representative sample of small-scale informal enterprises located in the capital city of Antananarivo. The latter is used to build the control group. Both surveys were conducted simultaneously in 2001 using highly comparable questionnaires. Additional follow-up surveys were conducted in 2003 and 2004 to compare changes in measured outcomes between the sampled treatment group and the control group of non-participants to the microfinance programme. In addition, in 2004, the 1-2-3 Survey was conducted again and the questionnaire administered to a new representative sample of SIEs. This provides us with the opportunity to reassess the impact of AdéFI using new treatment and control groups. A second sample of 300 microfinance clients of ADéFI was thus randomly chosen in 2004 (out of 6,217 members). Thanks to this unique dataset, we are allowed to conduct two types of impact evaluation. We first assess the impact of ADéFI on various outcomes measures by comparing a sampled treatment group and a control group randomly chosen in 2001 and 2004. To enrich these « static » assessments, we also take advantage of the baseline surveys of 2001 and the additional follow-up surveys of 2003 and 2004 to compare our 2001 treatment and comparison groups in terms of outcome changes over time.

5 In 2001, AdéFI had a total membership of 4,300 small-scale enterprises located in Antananarivo.

Baseline Surveys (2001)

AdeFI Membership List(4,267 clients)

1-2-3 Survey (Phase2)(924 SIEs)

Follow-up Surveys (2003)

Static impac

t

Baseline Surveys (2004)+ Follow-up Surveys (2004)

AdeFI Membership List(6,217 clients)

1-2-3 Survey (Phase2)(1,009 SIEs)

Treatment group A (198 clients)

Treatment group A (198 clients)

Treatment group A (198 clients)

Dynamic impact

Dynamic impact

Static impact

Treatment group B(306 clients)

Control group B(167 SIEs)

Control group A(87 SIEs)

Control group A(87 SIEs)

Control group A(87 SIEs)

Figure 1. Protocol of baseline and follow-up surveys

While this kind of longitudinal datasets is now commonly used in non-experimental evaluations, the empirical literature assessing the ex post impact of some projects or programmes implemented in Sub-Saharan countries using such rich data is rather thin. Four additional features of our dataset make it particularly valuable for the purpose of our research.First, the control group is selected from a representative sample of small-scale informal enterprises located in Antananarivo which constitute the eligible beneficiaries of ADéFI. 1-2-3 surveys are mixed household-enterprise surveys whose basic principle is to construct a sampling frame of SIEs through a household survey operation. This household survey component makes it possible to cover all informal sector units irrespective of size, kind of activity and type of workplace. Moreover, the larger the sample size (around 1,000 SIEs in our case), the higher the quality of the matching between participants and non-participants to the microfinance programme. Second, information gathered by the questionnaires is of high quality and allow in particular to get reliable measures of various firms' outcomes. Since the majority of microenterprises in developing countries do not keep financial records, one has to rely on recall data on business expenses and revenues that generally lack precision given the fungibility of money and goods between the business and the household, the seasonality of most microenterprises’ activity, etc. (for a detailed discussion, see de Mel, McKenzie and Woodruff, 2009). In addition to directly asking firm owners for their profits in the last month, phase 2 of the 1-2-3 survey collects very detailed information on production level, sales and purchases of inputs in the last 12 months, as well as on expenses in each of the following categories: rent for buildings; wages and salaries for employees; water, gas, electricity and fuel; telephone charges; travelling expenses and insurance fees; maintenance and general repairs; rent for machinery and equipment; taxes; interest paid; etc. The survey also records detailed information on the

seasonal patterns of the activity over a one-year period and on the timing of transactions to account for potential lags between the time inputs are purchased and the time the products are sold. Thanks to all these data, we are able to get accurate measures of gross output, intermediate consumption, value added, capital formation, etc. and inconsistencies were checked with the respondents.Third, our survey protocol has appealing properties: the treatment and comparison groups come from the same economic environment (same local labour market); they were administered the same questionnaire by the same team of highly trained interviewers, etc. As shown by Heckman, LaLonde and Smith (1999), these conditions are required to properly assess the causal impact of a programme. Last, a number of quality controls have been made during the data collection process. In particular, in order to reduce attrition (which was high during the 2003 follow-up survey), interviewers in 2004 have been asked to track microentrepreneurs whose production units either moved away, recorded activity changes or closed down over the period under concern. Thanks to this tracking effort, we are able to analyse the dynamics of the treated and non-treated enterprises and to examine how they compare in terms of their survival rate. To our knowledge, this question has never been properly answered in existing microfinance impact assessments.

Identification strategy

Assessing the impact of an intervention is measuring the difference in the values of key variables between the outcomes on “agents” (individuals, enterprises, households, etc.) which have experienced the intervention against the values of those variables that would have occurred had there been no intervention. The fundamental problem, of course, is that no agent can simultaneously undergo and not undergo an intervention. Therefore it is necessary to construct a counterfactual measure of what would have happened if the programme had not been available. Formally, let i index the population under consideration. Yi1 is the value of the variable of interest when unit i is subject to the treatment (Ti = 1), and Yi0 the value of the same variable when the unit is exposed to the control (Ti = 0). The treatment effect for a single unit, i, is defined as i = Yi1 -Yi0. The average treatment effect on the treated is given by: E(Yi1 -Yi0 | Ti = 1). The problem of unobservability is summarized by the fact that we can estimate E(Yi1 | Ti = 1), but not E(Yi0 | Ti = 1). In what follows, this contrefactual measure is constructed using the propensity-score matching (PSM) method introduced by Rosenbaum and Rubin (1983). Formally, let P(Xi)=P(T=1|Xi) denote the propensity score, i.e. the probability of participating to the microfinance programme for unit i conditional on a vector of pre-exposure control variables. As shown by Rosenbaum and Rubin (1983), if the Ti ’s are independent over all i, and outcomes are independent of participation given Xi, then outcomes are also independent of participation given P(Xi). If these assumptions are valid it is thus possible to use non-participants to measure what participants would have earned had they not participated, provided we condition on the variables X. In what follows, the propensity score is calculated for each observation in the participant and the control-group samples using

standard logit models. The matching is then carried out using P̂i where P̂i is the estimated probability of participating to the microfinance programme for individual i given X. We then use two different matching estimators. The first one is the “nearest neighbor” estimator where we find the closest non-participant match for each participant and where the impact estimator is the difference in mean outcome between participants and their matched non-participants. The second estimator takes the average outcome measure of the closest five matched non-participants and compares this to the participant’s outcome measure.

To correct for possible bias due to the difference in time-invariant unobserved characteristics between the participant and control groups, we make use of the panel structure of our data and combine PSM with the difference-in-difference (DD) method.

4. Results

Propensity score models

Thanks to our statistical protocol, we have two sets of representative samples of ADéFI clients at our disposal, one for 2001 and one for 2004. For each sample of clients, we are able to associate one specific control group drawn from the phase 2 surveys conducted respectively in 2001 and 2004. As quoted above, the matching procedure has been applied, using standard propensity score models. Tables A.1 and A.2 in Appendix present the results of the two regressions used to estimate the propensity scores of clients and their matched controls in 2001 and 2004. In both estimations, the vector of observable characteristics include roughly the same set of variables: a set of individual characteristics of the business owner (sex, level of education, type of apprenticeship attended and age of the business owner) and a set of characteristics of the microenterprise (industry, type of premice, date of creation, size at the time of creation - labour and capital -, etc). However, in the estimation relating to year 2004, the values of the potentially time-varying regressors are either those of 1997, which corresponds to the year ADéFI started its activity in Antananarivo or those of later years for the microenterprises that were created after 1997. Such a specification guarantees that the differences in characteristics between clients and non clients at the time when ADéFI opened its first branches in Antananarivo are properly controlled for. For 2001, given that the information on the situation in 1997 was not available, the value of the time-varying regressors is measured at the date of creation of the microenterprise. Figure 2 shows the distribution of propensity scores by treatment status based on the 2004 estimates. The overlap in the propensity score distributions indicates that there are comparable clients and non clients in the data set and that the data can thus support comparisons between the two groups. Samples sizes in the regions of propensity score overlap for both years 2001 and 2004 are shown in Table 2.

Figure 2. Distribution of estimated Propensity Scores, by treatment status, 2004

0,00

- 0

,05

0,05

- 0

,1

0,1

- 0,

15

0,2

- 0,

25

0,25

- 0

,3

0,3

- 0,

35

0,35

- 0

,4

0,4

- 0,

45

0,45

- 0

,5

0,5

- 0,

55

0,55

- 0

,6

0,6

- 0,

65

0,65

- 0

,7

0,7

- 0,

75

0,75

- 0

,8

0,8

- 0,

85

0,85

- 0

,9

0,9

- 0,

95

0,95

- 1

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

Non clientsClients

Estimated Propensity Score

Table 2. Propensity score strata sample sizes, 2001 and 2004

Estimated Propensity Score 2001 2004Non clients Clients Non clients Clients

0,00 - 0,25 810 38 682 560,25 - 0,5 71 40 162 910,5 - 0,75 27 52 65 1040,75 - 1 5 54 12 55Total 913 172 921 306

After this first stage, each sample client of ADéFI has been matched with the nearest non client microenterprise in terms of propensity score with a caliper of +/-0.05 on the probability scale. Table A3 in Appendix lists all the covariates used in the regression and shows the level of imbalance of each covariate in both the unmatched and the matched samples. As suggested by the figures, the propensity matching procedure is quite effective in reducing covariate imbalance between the two groups: while a significant proportion of covariates are significantly different across exposure groups prior to matching, most of these differences disappear after the matching. After the matching procedure, we are thus able to compute simple tests comparing various variables of interest between the sample clients and their matched counterparts to assess the impact of ADéFI.

Impact of "being a client of ADéFI" in 2001 and 2004

A rich set of variables relative to the performance of the SIEs is considered to assess the impact of AdéFI: turnover, production, value added, gross surplus, number of workers, stock of physical capital and labour and capital productivity. The results are presented in Table 3. For each variable, the net gain is equal to the difference between the average performance of the clients and the control group, over the performance of the latter. We test the significance of differences in averages. The first panel deals with the impact in 2001, while the second deals with the impact in 2004. Whatever the indicator of performance considered, the impact of "being a client of ADéFI" is positive and strongly significant. For instance, in 2001, the turnover generated by ADéFI’s clients is 210% higher than the one generated by their matched counterparts. All the other indicators show the same pattern, in both 2001 and 2004. However, no significant differences are observed for labour inputs, capital stock and capital productivity.

Table 3. Net gains of the project (%), 2001 and 20042001 2004

OutputTurnover + 209,8 *** +91,5 ***Production + 179,3 *** + 71,5 ***Value added + 154,7 *** + 81,2 ***Gross operating profit marge + 166,8 *** + 84,4 ***Production factorsNumber of workers + 17,5 + 23,6Current value of capital stock + 48,4 + 192,1 ***ProductivityVA/L1 + 64,5 *** + 78,4 ***VA/L2 + 227,3 *** + 141,1 *VA/K + 1,0 - 11,9Sources: 1-2-3 Surveys, phase 2, 2001 & 2004, DIAL / INSTAT; Surveys on the MFI’s clients, 2001 & 2004; authors’ calculations.Notes: In 2001, the number of matched clients is 168 and the number of microenterprises in the control group is 87 (some of them have multiple peers). In 2004, the respective numbers are 306 and 167. VA: Value added; L1: Number of workers; L2: Number of hours worked per month. K: value of capital (at remplacement cost). *: significant at 1%; **: significant at 5%; ***: significant at 1%.

Overall, the estimated impact is lower in 2004 than in 2001. One technical reason may be at play here: the better specification of the covariates in the 2004 regression (Table A.2) may have resulted in a better matching procedure between clients and non clients for this year. The control group is thus probably more reliable in 2004 than in 2001. Being closer from the clients at the baseline (i.e. before ADéFI’s intervention), the control group is also closer at the endline (i.e. after the intervention) in terms of economic performances. Thus, the net gain estimates of the project are mechanically reduced.

Discussion about the matching procedure and robustness checks

In the previous section, the matching procedure is based on propensity scores estimated by a regression in which being a client of ADéFi is explained by a set of independent variables not including any ex ante output indicators (such as turnover, production, value added or gross operating profit marge). For 2001, this choice was dictated by the non availability of such indicators in our database. However, in 2004, the redesign of the questionnaire, with added

questions on the history of the microenterprise, made it possible to introduce the level of turnover in 1997 (i.e. at the time when ADéFI opened its first branches in Antananarivo) among the regressors of our participation model. The inclusion of this variable in the set of regressors allows us to better control for differences in the characteristics and level of activity of the SIEs prior to the implementation of ADéFI. This contributes to improve the matching procedure: in addition to ensure that the treated and non treated group presents the same profile in average (same age of microentrepreneurs, same level of schooling, same distribution by industry and type of premice, same volume of labour and capital stock in 1997, etc.), it also ensures that the pairs of matched microenterprises had the same level of activity prior to the start of ADéFI’s operations. In line with the preceeding results, the improvement in the matching process reduces the net gain of the project. To illustrate the point, Table 4 reports the results of Table 3 for 2004, based on a matching procedure which does not take into account the level of activity ex ante (matching 1). The second colomn reports the same indicators but relying (among others) on a matching procedure which controls for the level of activity ex ante (matching 2). The net gains of the projects still remain globally positive, but they are not always significant anymore. This suggests that some gaps in performance in 2004 (column 1) are just the perpetuation of gaps already existing in 1997. Nevertheless, in terms of turnover, stock of capital and labour productivity, the impact of the project remains positive and significant.

Table 4. Net gains of the project in 2004 (%)Matching 1 Matching 2

OutputTurnover +91,5 *** + 68,9 ***Production + 71,5 *** + 39,5Value added + 81,2 *** + 45,4Gross operating profit marge + 84,4 *** + 47,3Factors ofproductionNumber of workers + 23,6 + 9,1Value of capital + 192,1 *** + 128,7 ***ProductivityVA/L1 + 78,4 *** + 61,3 ***VA/L2 + 227,3 *** + 116,1VA/K + 1,0 - 35,4 Sources: 1-2-3 Survey, phase 2, 2004, DIAL / INSTAT; Surveys on the MFI’s clients, 2004; authors’ calculations.Notes: see Table 3. In the first column, the number of matched clients is 306 and the number of microenterprises in the control group is 167; in the second column, the respective numbers are 290 and 156, because of missing values in the 1997 turnover.

The previous discussion shows the importance (and the difficulty) of building up a control group. In this regard, to be rigorous, an evaluation protocole should consider collecting a large number of information about the situation ex ante of project’s members and non members to ensure that the observed differences between the two groups at the time of the evaluation can be truly imputed to the project and not to preexisting differences.

Impact of "being a client of ADéFI" before, during and after the crisis: a panel data analysis 2001-2004

In order to assess the impact of the project over the years, two follow-up surveys were conducted among the 168 sample clients and their 87 matched counterparts. The first one was launched in March 2003, nine months after the end of the political crisis, and the second one in September 2004. At that time, Madagascar was firmly engaged in a process of economic recovery. In what follows, we will successively mobilize these two rounds of surveys to assess the demographic dynamics of the sample SIEs, as well as the dynamic impact of ADéFI on the surviving ones, both in the short (2001-2003) and in the medium runs (2001-2004).

The demographic dynamics of the enterprises: 2001-2004

All in all, the first follow-up survey which took place in March 2003 allowed to track and reinterview 130 clients and 67 peers, which corresponds to attrition rates of 22% and 23% respectively. Unfortunately, we are not able to identify the precise reasons of such attrition (enterprise closure for economic reasons, changes in location of the owners, refusal to answer the survey, etc.). The only information we have concerns the clients of ADéFI. Out of the 38 clients who were not found, 40% have encountered repayment problems with ADéFI, while the managing team has had no news about the other 60%. Assuming that the refusal rate is not substantially different among the clients and their matched counterparts, which seems quite a reasonable assumption, it seems that being a client of ADéFI has had no impact on the survival rate during the crisis. Furthermore, attrition does not seem to be selective on our set of covariates: the characteristics of the SIEs we managed to tracked and reinterviewed do not significantly differ indeed from those of the full sample surveyed in 2001, with the only exception of the share of SIEs involved in trading activities, which is significantly lower in the panel.6

The second follow-up survey took place in Septembre 2004, exactly three years after the baseline survey of 2001, and a little more than two years after the end of the political crisis. In order to overcome the abovementioned shortcomings, this survey has been specifically designed to track and interview the SIEs (clients and peers) present in the 2003 follow-up survey, but also all those which had been surveyed in 2001 but not in 2003, whatever the reason. This tracking process, extremely demanding and time-consuming, is indeed the only way to rigorously assess the dynamics of SIEs in its double dimension: the demography of enterprises (here, their mortality rate) and the economic dynamics of the surviving ones.

The situation of the 255 panel enterprises as captured in the 2004 follow-up survey provides the precise picture of their trajectory. Firstly, the attrition’s process already registered in 2003 proceeded further, and probably at an accelerated pace. 23% of the enterprises definitively stopped their activity. 22% are in a fuzzier situation (the business owner is durably absent or dead, moved away, etc.), which corresponds at best to a sharp precariousness context, if not to an effective closure. Finally, a little more than one enterprise out of two (55%) can be clearly identified as still in activity (of which 90% have been surveyed). This low survival rate underlines the sharp instability of such enterprises that the crisis is likely to have weakened further.

6 The tests of attrition may be obtained upon request from the authors.

Secondly, the contrasted survival profiles between the clients and their matched counterparts put into light a rather counterintuitive result: the survival rate is lower among the clients than among the control group. The former are twice numerous as the latter to have stopped their activity (28% vs. 14%). Even if we are not able to precisely assess the respective role of ADéFI's financial services and the crisis in this respect, one of the reasons may be found in the fact that the crisis has had a stronger negative impact on the largest microenterprises, the latter being overrepresented among the clients (see below).

Two combined phenomena may explain the lower resistance of the largest microenterprises during the crisis: as global demand shrank, the households shifted towards smaller microenterprises which provide basic goods at lower prices, at the disadvantage of bigger enterprises which products have higher income elasticity. Additionally, the largest microenterprises usually operate with higher fixed costs (while the smallest ones often function with non wage workers and low investment), which make them less flexible to fluctuations in demand.

Table 6. Situation of the enterprises in the panel in September 2004Clients Non clients Total

In activity and surveyed 48.2% 54.0% 50.2%In activity and refused to answer 3.0% 9.2% 5.1%Business stopped 28.0% 13.8% 23.1%Business head absent 3.0% 2.3% 2.8%Moved 12.5% 10.3% 11.8%Others (business head died, etc.) 5.3% 10.4% 7.0%

Total 100% (168)

100% (87)

100%(255)

Sources: 1-2-3 Survey, phase 2, 2001; Client Survey, 2001 and Follow-up Surveys 2003 & 2004; MADIO, DIAL / INSTAT; authors’ calculation.

Table 7 presents the structure of the follow-up samples between 2001 and 2004. Among the 255 enterprises (clients and their matched counterparts) surveyed in 2001, only 111 have been interviewed in the three rounds. In fact, the number of « surviving » enterprises in 2004 is slightly higher (128) than in 2003, the difference coming from the fact that some SIEs not surveyed in 2003 have been tracked and found in 2004 (whether because they were temporary closed, or because the survey strategy to track the enterprises in 2003 was less efficient). As in the previous table, the survival rate is 50%. Apart from the mortality of enterprises, the matching procedure is also a source of reduction in the sample size. From the clients’ side, when her matched counterpart disappears, identifying a new peer is infeasible and so the observation is lost for all subsequent analyses. From the peers’ side, it is also possible that, again because of the matching procedure which allows replacements, the same non-treated enterprise is matched with several clients. That is for instance the case when two clients have the same propensity score. Finally, 128 enterprises have been surveyed both in 2001 and 2004, but only 107 among them are used in our panel analysis. Obviously, such variations in the samples according to the periods under concern is a source of great complexity, as we have to adjust our estimations to varying samples (this point is not specific to our case, but is a generic issue for panel data analysis). However, these variations in sample size can also be considered as robustness checks of our results. These will be more rigorously assessed, if they are consistent using different samples.

Table 7.Demographic dynamics of panel enterprises, 2001-2004Clients Peers Total

Surveyed Matched Surveyed Matched Surveyed MatchedImpact in 2001 168 168 87 87 255 255Follow-up 2001-2003 133 108 67 54 200 162Follow-up 2001-2004 81 71 47 36 128 107Follow-up 2001-2003-2004 70 56 41 39 111 95Sources: 1-2-3 Survey, phase 2, 2001; Client Survey, 2001; Follow-up Survey 2003 & 2004 ; MADIO, DIAL / INSTAT; authors’ calculation.

The economic dynamics in the short run (2001-2003)

Even though our follow-up protocole does not provide precise information on the demographic dynamics of the enterprises (mortality and creation rates), it allows to assess the impact of ADéFI on those microenterprises which were in activity before the crisis. For the 108 clients and their 54 corresponding peers composing the panel, we computed the net gains of the project, using the same methodology as before. The main interest of this exercice resides in assessing the impact before and after the 2002 crisis.

Table 8. Net gains of the project, 2001-2003 (constant euros of 2001)

Years Variations 2001/2003(double difference)

2001 2003 Average MedianOutput (annual)Turnover + 22,263 (***) + 15,940 (***) - 6,322 (n.s.) + 1,383Production + 14,346 (***) + 7,122 (***) - 7,224 (*) + 1,131Value added + 7,830 (***) + 4,756 (***) - 3,074 (n.s.) + 218Mixed income + 7,406 (***) + 4,288 (***) - 3,118 (n.s.) + 203Factors of productionNumber of workers + 1.69 (***) + 1.83 (***) + 0.15 (n.s.) - 0.5Capital value + 5,561 (n.s.) + 3,919 (***) - 1,642 (n.s.) - 71ProductivityVA/L1 + 116.5 (**) + 103.1 (**) - 13.4 (n.s.) + 1.9VA/L2 + 2.3 (n.s.) + 4.8 (n.s.) + 2.5 (n.s.) - 0.1VA/K - 0.09 (n.s.) - 0.06 (*) + 0.03 (n.s.) + 0.8Sources: 1-2-3 Survey, phase 2, 2001; Client survey, 2001; Follow-up Survey, 2003; MADIO, DIAL / INSTAT; authors’ calculation.Notes: see Table 4. The number of matched clients is 108 and the number of microenterprises in the control group is 54. In 2001, the exchange rate was 6,100 FMG for 1 €. *: significant at 1%; **: significant at 5%; ***: significant at 1%.

In 2003, the net gains of the project were still significant and positive, but were much smaller on average (Table 8). In a recessive economic context – between December 2001 and December 2002, household real income decreased by more than 20% (Ramilison, 2003) – and of enterprises’ decapitalization, the surviving clients suffered more than their counterparts. For example, the average value added generated by clients was 7,830 euros higher than that of non clients in 2001, but only 4,760 euros higher (in constant tems) in 2003. This diagnosis should nevertheless be nuanced. Indeed, when the focus is on medians – which are less sensitive to measurement errors – rather than on means, the clients are actually found to have performed better than the control group. Whatever the outcome indicator, the net gains of the project are positive and increasing between 2001 and 2003. This contrasted assessment when

median values are used instead of mean values could be explained by the fact that the crisis had a stronger negative impact on the largest SIEs, which are overrepresented among the clients. Although the net gains of the project have decreased between the two periods, the analysis in double difference suggests that most of the differences are not satistically significant. In other terms, ADéFI's clients had neither worse nor better performances than non clients during the crisis. This suggests that, while the impact of the project was still positive in 2003, this was entirely due to the benefits realized before the crisis, as if the economic recession that followed had stopped the positive impulse ADéFI gave to its clients during the previous growth period.

To investigate whether the economic performances of the clients vary along the distribution, we splitted the sample into quartiles of turnover in 2001. The analysis in double difference conducted on each sub-sample provides additional interesting insights. As sugested by Table 9 indeed, the absence of a significant gap between clients and non clients on the full sample masks a strong heterogeneity among clients according to their size. The relative performance of the clients is found to decrease with the level of their activity before the crisis. While the advantage of the clients from the two lower quartiles continued to grow (relatively to the control group), it remained stable for microenterprises of the third quartile, and substantially decreased for those of the fourth quartile.

Table 9. Nets gains of the project between 2001 and 2003 by quartile of turnover in 2001Turnover 2001 Double difference (DDIF)

1st quartile + 5,167 (***)

2nd quartile + 6,146 (*)

3rd quartile + 12,974 (n.s.)

4th quartile - 49,578 (**)Source: 1-2-3 Survey, phase 2, 2001; Client Survey 2001 and Follow-up Survey 2003; MADIO, DIAL / INSTAT; authors’ calculations.Notes: In 2001, the exchange rate was 6,100 FMG for 1 €. *: significant at 1%; **: significant at 5%; ***: significant at 1%.

To sum up, if the positive impact of the project seems to be clearly established in times of growth, its impact in times of recession is much less clear. And the fact that the performances of the clients were neither better (nor worse) than the control group between 2001 and 2003 does not clarify the issue. This last result would have been conclusive if SIEs in the control group and in the treated groups had been exactly similar when the crisis started. However, this condition was not met as the clients were significantly bigger thanks to the loans they got from ADéFI. Without any appropriate contrefactual, we are not able to assess whether the low performances of the clients during the crisis are due to their larger size (if it is true that the crisis had more impact on the biggest SIEs), or if they are due to the fact that being a client of AdéFI meant more constraints to cope with.

The economic dynamics in the medium run (2001-2004)

Repeated waves of data collection on a panel of firms raise a trade-off: on the one hand, they allow to get better insights on firm dynamics; on the other hand, they generally result in a

substantial reduction in sample sizes due to the attrition process, the latter being all the more important as the time lag between the baseline and the follow-up surveys is long. In what follows, we proceed in two steps. We first compute differences in differences between clients and non clients using the two ‘extreme’ survey rounds of 2001 and 2004. We then check the robustness of our results using the sample SIE's that were interviewed in each of the three survey rounds.

Impact 2001-2004

Out of the 168 clients and their 87 peers that were used for the 2001 evaluation, respectively 81 and 47 were still in activity in 2004. As mentioned above, such a high attrition highlights the extremely high vulnerability of microenterprises: over a three-year period, almost 60% of the SIEs have disappeared among the sample clients and their matched counterparts. A close examination of the 71 clients we managed to find, re-interview, and whose matched counterparts were also found and re-interviewed prompts the following observations:

− In 2004, the net gains of the project still remain positive and significant for almost all the variables considered;

− In constant euros, the difference of performance between clients and non clients in absolute terms is not different in 2004 than in 2001. The picture is however different in relative terms: for instance, while the average value added generated ADéFI's clients was 160% higher than that of the control group in 2001, the corresponding figure for 2004 is 280%;

− The results in 2001 on the sub-sample of SIEs that are still present in the 2004 database are highly similar to those obtained on the full sample, which we interpret as a proof of robustness.

Table 10. Net gains of the project 2001-2004 (in 2001 constant euros)

Years Variations 2001/2004(double difference)

2001 2004 (Average)Outputs (annual)Turnover + 20 570 (***) + 21 028 (***) + 459 (n.s.)Production + 13 741 (***) + 9 013 (***) - 4 728 (*)Value added + 6 199 (*) + 8 404 (***) + 2 205 (n.s.)Mixed income + 6 126 (*) + 8 633 (***) + 2 507 (n.s.)Factors of productionNumber of workers + 1,01 (n.s.) + 0,15 (n.s.) - 0,86 (n.s.)Value of capital + 2 222 (**) + 1 173 (n.s.) - 1 048 (n.s.)ProductivityVA/L1 + 133 (***) + 301 (***) + 168 (n.s.)VA/L2 + 0,36 (n.s.) + 2,07 (**) + 1,71 (n.s.)VA/K + 0,45 (n.s.) - 1,23 (n.s.) -1,68 (n.s.)Source: 1-2-3 Survey, phase 2, 2001; Client Survey 2001 and Follow-up Survey 2004; MADIO, DIAL / INSTAT; authors’ calculations.Notes: see Table 4. The number of matched clients is 71 and the number of microenterprises in the control group is 36.

Impact 2001-2003-2004

Let us now turn to the full assesment of the temporal dynamics, which takes into account all the survey rounds of surveys at our disposal (2001, 2003 and 2004). Our sample is now reduced to 56 clients matched with 39 peers surveyed in each of the three periods. Considered as consecutive cross-section surveys, the results confirm what we previously found on the full sample. For each of the three rounds, the project brings substantial net gains, in all the dimensions considered: output, inputs and productivity.

The impact is much less favorable, though, when we consider the evolution over time (double differences). Whatever the subperiod (2001-2003, 2003-2004 or 2001-2004), none of the variables of interest is significant anymore. This suggests that the trend followed by ADéFI's clients is similar as the one followed by control group? In other terms, ADéFI has not succeeded in accelerating the growth process of its members. As a matter of fact, the net gains observed in 2003 and 2004 were already achieved in 2001, and have just stabilized since then. This brings support to our previous assumption according to which the financial services offered by ADéFI have been more beneficial during the growth period than during the recession.

This conclusion is reinforced by the dynamics recorded during the different sub-periods. Even if not statistically significant, the relative performances of the clients have been inferior to those of the control group until 2003, while they have been rather better since, in a global context of growth recovery. One can reasonably assume that, if the business cycle is to keep with this upward trend, the impact of ADéFI will improve to finally become significantly positive.

Tableau 11. Dynamics of the net gains of the project 2001, 2003, 2004 (in 2001 constant euros)Years Variations

2001 2003 2004 DDIF 2001/2004

Outputs (annual)Turnover + 25,948 (***) + 23,375 (***) + 23,050 (***) - 2,898 (n.s.)Production + 18,190 (***) + 7,252 () + 10,616 (***) - 7,574 (n.s.)Value added + 8,993 (*) + 6,135 (**) + 10,108 (***) + 1,115 (n.s.)Mixed income + 8,684 (*) + 5,912 (**) + 9,615 (***) + 931 (n.s.)Factors of productionNumber of workers + 1.38 (***) + 1.55 (**) + 0.88 (n.s.) - 0.50 (n.s.)Value of capital + 3,530 (***) + 3,508 (***) + 2,065 (n.s.) - 1,465 (n.s.)ProductivityVA/L1 + 156 (***) + 161 (n.s.) + 331 (***) + 174 (n.s.)VA/L2 - 0.02 (n.s.) + 0.89 (**) + 2.35 (n.s.) + 2.37 (n.s.)VA/K + 0.24 (n.s.) - 1.37 (n.s.) - 2.72 (n.s.) - 2.97 (**)Source: 1-2-3 Survey, phase 2, 2001; Client Survey 2001 and Follow-up Surveys 2003 & 2004; MADIO, DIAL / INSTAT; authors’ calculation.Notes: see Table 4. The number of matched clients is 56 and the number of microenterprises in the control group is 39. None of the DDIF 2001/2003 and 2003/2004 (not reported here) are significant at 10%.

Conclusion

This paper leads to two types of concluding remarks.First, concerning ADéFI, the results of our analysis provide evidence that SIEs are positively impacted by their access to micro-loans. Using cross-sectional data collected in 2001 and 2004 among ADéFI's clients and non clients, our evaluation shows that SIEs which have benefited from loans record better performances on average than their matched counterparts, even if part of the observed gap between the two groups might have existed prior to the creation of ADéFI. Our dynamic analysis using panel data on clients and non clients provides nevertheless more ambiguous conclusions. If the positive impact of the project seems to be well-established in times of growth, it is much less obvious in a recessive economic context, partly because ADéFI mainly provides loans to the biggest microenterprises which are more fragile in times of economic recession because of higher fixed costs. This suggests that ADéFI has failed in protecting its clients against the detrimental impact of the 2002 political crisis.

Second, on methodological grounds, this paper has shown how challenging was a rigorous impact assessment of a microfinance programme. The very strong vulnerability of SIEs makes it particularly difficult to conduct follow-up surveys since the mortality rate of SIEs results in high attrition rates between successive survey rounds. Besides, the frequent changes in the domiciliation of SIEs makes the job of the surveyors even more complex since they require a tracking procedure that is both time-consuming and likely to improve recontact rate only marginally. The paper has also highlighted how dependent were the results on the quality of the matching procedure. What we learnt from this exercise is that a non-experimental evaluation protocol should always put great care on collecting as many informations as possible on the ex ante situation of participants and non participants to a project, in order to guarantee that the observed ex post differences are really due to the project and not to differences that were already prevailing ex ante.

References

Banerjee, A., Duflo, E., Glennerster, R., and C. Kinnan (2009), "The miracle of microfinance? Evidence from a randomized evaluation",

Heckman, J., Lalonde, R., and J. Smith (1999), "The economics and econometrics of active labour market programs", in Ashenfelter, A. and D. Card (eds), Handbook of Labor Economics, Volume 3A, Amsterdam: Elsevier Science.

IMF (2006), Republic of Madagascar: Financial System Stability Assessment, IMF Country Report No. 06/305, August, 37p.

Karlan, D., and J. Zinman (2010), "Expanding Credit Access: Using Randomized Supply Decisions To Estimate the Impacts", Review of Financial Studies, 23 (1), 433-464.

de Mel, S., McKenzie, D., and C. Woodruff (2009), "Measuring microenterprise profits: must we ask how the sausage is made?", Journal of Development Economics, 88(1), 19-31.

Ramilison, E. (2003), "Impact de la crise politico-économique de 2002 sur l’emploi et les conditions d’activité des ménages de l’agglomération d’Antananarivo", Projet MADIO, juin, Antananarivo.

Rosenbaum, P. and D. Rubin (1983), "The central role of the propensity score in observational studies for causal effects", Biometrika, 70(1), 41-55.

Zeller, M., and R. L. Meyer (eds) (2003), The Triangle of Microfinance: Financial Sustainability, Outreach, and Impact. The Johns Hopkins University Press, Baltimore, MD, USA, 424p.

APPENDIX

Table A.1. Probit model of being a client à ADéFI –Antananarivo, 2001Coefficient z P>|z|

Characteristics of business ownerEducation [ref. is tertiary education]

<CEPE (5 years of schooling) (d) -0.431 -1.90 0.06 *BEPC/CFEPCES (9 years of schooling) (d) -0.136 -0.67 0.50Pre-BAC/CFECP (11 years of schooling) (d) 0.038 0.17 0.87BAC (12 years of schooling) (d) 0.320 1.39 0.16

Male (d) -0.658 -4.63 0.00 ***Age [ref. is less than 31]

31-40 years old (d) -0.280 -1.68 0.09 *41-50 years old (d) 0.081 0.47 0.6451-61 years old (d) -0.527 -1.88 0.06 *

Apprenticeship [ref. is no apprenticeship]Vocational school (d) -0.495 -2.23 0.03 **On-the-job training in big firm (d) -0.412 -0.97 0.33On-the-job training in small firm (d) -0.168 -0.94 0.35Other (d) -0.125 -0.30 0.76

Characteristics of microenterpriseType of industry [ref. is BTP]

Agro-industry (d) 0.155 0.31 0.76Clothing and apparel (d) -0.033 -0.08 0.94Other industry (d) -0.017 -0.04 0.97Trade of primary products (d) -0.103 -0.23 0.82Trade of transformed products (d) -0.295 -0.63 0.53Services (d) -0.397 -0.96 0.34Hotels and restaurant (d) -1.483 -2.73 0.01 **Transport (d) 0.506 0.87 0.38

Type of premices [ref. is Itinerant]In a fixed street pitch (d) 0.661 1.30 0.19In vehicle (d) 0.760 1.48 0.14From the customer’s home (d) 0.184 0.35 0.73From home without having a dedicated set-up for the job (d) 1.383 3.59 0.00 ***From home with a dedicated set-up for the job (d) 0.817 2.08 0.04 **In a public market, not fixed (d) 0.910 1.81 0.07 *In a public market, fixed (d) 0.808 2.02 0.04 **In a shop, restaurant, hotel, etc. (d) 1.150 3.13 0.00 ***Other (d) 0.325 0.63 0.53

Date of creation [ref. is after 1999]Created between 1972 and 1981 (d) 0.061 0.22 0.82Created between 1982 and 1991 (d) 0.424 1.80 0.07 *Created between 1992 and 1996 (d) 0.586 2.62 0.01 **Created between 1997 and 1999 (d) 0.490 2.28 0.02 **

Number of workers at time of creation 0.082 1.57 0.12Value of physical capital at time of creation (in 1,000 Fmg) 0.008 2.82 0.01 **Date of creation [ref. is after 1999]

Created between 1972 and 1981 (d) 0.838 4.92 0.00 ***Created between 1982 and 1991 (d) 0.106 0.48 0.63

No financial records (d) -1.205 -8.94 0.00 ***Intercept -1.205 -8.94 0.00 ***Number of obs. 1 086LR chi2(38) 408.62PseudoR² 0.43

Table A.2. Probit model of being a client à ADéFI –Antananarivo, 2004Coefficient z P>|z|

Characteristics of business ownerMale (d) -0.271 2.55 0.01 **Age [ref. is less than 31]

31-40 years old (d) 0.031 0.28 0.7841-50 years old (d) 0.043 0.30 0.7651-61 years old (d) -0.203 -0.69 0.49

More than 61 years old (d) -0.542 -1.19 0.23Characteristics of microenterpriseType of premices [ref. is Itinerant]

In a fixed street pitch (d) 0.108 0.34 0.74In vehicle (d) 1.101 3.29 0.00 ***From the customer’s home (d) -0.245 -0.66 0.51From home without having a dedicated set-up for the job (d) 1.209 5.28 0.00 ***From home with a dedicated set-up for the job (d) 1.014 4.39 0.00 ***In a public market, not fixed (d) 0.744 2.41 0.02 **In a public market, fixed (d) 0.533 2.08 0.04 **In a shop, restaurant, hotel, etc. (d) 0.660 3.05 0.00 ***

Type of industry [ref. is Agro-industry]Clothing and apparel (d) 0.338 1.38 0.17Other industry (d) 0.022 0.08 0.94Trade of primary products (d) 0.861 3.35 0.00 ***Trade of transformed products (d) 0.326 1.16 0.25Services (d) 0.486 2.04 0.04 **Hotels and restaurant (d) 0.217 0.74 0.46Transport (d) 0.894 2.78 0.01 **

Conditions d'activitéWater (d) 0.110 0.72 0.47Electricity (d) 0.285 2.31 0.02 **Phone (d) -0.404 -1.44 0.15

Who created? [ref. : Someone else, non relative]Business owner, alone 0.016 0.03 0.98Business own, with other people 0.154 0.29 0.77A relative of the business owner -0.113 -0.21 0.83

Date of creation [ref. : Before 1972]Between 1972 and 1981 -0.095 -0.25 0.81Between 1982 and 1991 0.396 1.13 0.26Between 1992 and 1996 0.413 1.17 0.24Between 1997 and 1999 0.641 1.81 0.07 *Between 2000 and 2002 0.381 1.08 0.28After 2002 0.042 0.11 0.91

Number of workers at time of creation 0.164 4.00 0.00 ***Value of physical capital at time of creation (in 1,000 Fmg) 0.002 1.07 0.28Legal status / Registration

Commercial register (d) 0.031 0.16 0.87Trading license (d) 0.352 2.35 0.02 **Professional ID card (d) -0.043 -0.26 0.79CNaPS (d) -0.052 -0.10 0.92

Financial records (d) 0.947 -9.15 0.00 ***Intercept -2.399 -3.40 0.00 ***Number of obs. 1 244LR chi2(39) 446.2PseudoR² 0.32

Table A3. Covariate imbalances across groups prior to and after matching on the propensity score, 2001

CovariatesFull samples Matched pairs

Non clients Clients Non clients ClientsCharacteristics of business ownerMale (d) 0.59 0.37 0.39 0.37Education [ref. is no schooling]

<CEPE (5 years of schooling) (d) 0.39 0.11 0.08 0.11BEPC/CFEPCES (9 years of schooling) (d) 0.30 0.30 0.42 0.30Pre-BAC/CFECP (11 years of schooling) (d) 0.12 0.19 0.18 0.19BAC (12 years of schooling) (d) 0.09 0.22 0.19 0.22

Age [ref. is less than 31]31-40 years old (d) 0.34 0.36 0.45 0.3841-50 years old (d) 0.25 0.31 0.35 0.2951-61 years old (d) 0.10 0.05 0.02 0.05

Apprenticeship [ref. is no apprenticeship]Vocational school (d) 0.10 0.07 0.06 0.08On-the-job training in big firm (d) 0.03 0.01 0.02 0.02On-the-job training in small firm (d) 0.15 0.16 0.16 0.15Other (d) 0.03 0.02 0.03 0.02

Characteristics of microenterpriseType of industry [ref. is BTP]

Agro-industry (d) 0.03 0.05 0.07 0.05Clothing and apparel (d) 0.12 0.24 0.28 0.25Other industry (d) 0.10 0.07 0.05 0.06Trade of primary products (d) 0.14 0.17 0.19 0.17Trade of transformed products (d) 0.12 0.09 0.08 0.10Services (d) 0.24 0.10 0.13 0.12Hotels and restaurant (d) 0.07 0.04 0.00 0.02Transport (d) 0.08 0.21 0.19 0.20

Type of premices [ref. is Itinerant]In a fixed street pitch (d) 0.05 0.01 0.00 0.01In vehicle (d) 0.07 0.16 0.16 0.16From the customer’s home (d) 0.14 0.01 0.01 0.01From home without having a dedicated set-up for the job (d) 0.12 0.26 0.28 0.26

From home with a dedicated set-up for the job (d) 0.13 0.14 0.18 0.15In a public market, not fixed (d) 0.03 0.02 0.02 0.02In a public market, fixed (d) 0.06 0.09 0.03 0.09In a shop, restaurant, hotel, etc. (d) 0.18 0.27 0.24 0.27Other (d) 0.07 0.02 0.04 0.02

Date of creation [ref. is after 1999]Created between 1972 and 1981 (d) 0.14 0.06 0.13 0.07Created between 1982 and 1991 (d) 0.22 0.22 0.15 0.21Created between 1992 and 1996 (d) 0.20 0.29 0.36 0.29Created between 1997 and 1999 (d) 0.26 0.35 0.32 0.35

Who created? [ref. is business head alone]Created by business head with others (d) 0.09 0.34 0.25 0.31Created by a family member (d) 0.10 0.07 0.15 0.08

Number of workers at time of creation 1.51 2.36 1.95 2.24No financial records (d) 0.79 0.25 0.22 0.24Value of physical capital at time of creation (in 1,000 Fmg) 4.73 30.15 23.99 36.72