Thesis
Author: Rasmus Stenbäcken Accessor: Erica Lindahl Semester and Year: fall; 2005
Do Self-Sustainable MFI:s help alleviate relative poverty?
Acknowledgements
This thesis was made possible through the cooperation of Banco Solidario. I wish to thank my contact at Banco Solidario, Catherine Arrelano, with whom I had interesting discussions and from whom I got useful information. I also wish to thank Fransisco Díaz and Fabrizio Narvaez for among other things helping me plan the interviews. A special thanks goes out to the accessors of Banco Solidarios Recreo office: Marlon Rodriguez, Patricio Buenano, David Tenorio, Darwin Sanchez, Marcelo Morales, Vinicio Tipan, Cesar Jacome & Marco Juna, whom not only sacrificed their time to help me with this study, but also made me feel welcome and as one of the team.
I wish to thank Erica Lindahl for help throughout the whole process of writing this thesis. I also wish to thank Reijer Hendrikse for help with ideas and comments, and Peter Fredriksson for useful insights, on the early drafts of this paper. And to Anders Klevmarken for helping me with some of the models.
Finally I want to express my gratitude to SIDA for financing this study.
Abstract
The subject of this paper is microfinance and the question: Do self-sustainable MFI:s alleviate poverty?.
A MFI is a micro financial institution, a regular bank or a NGO that has transformed into a licensed financial institutions, focused on microenterprises. To answer the question data has been gathered in Ecuador, South America. South America have a large amount of self sustainable MFI:s. Ecuador was selected as the country to be studied as it has an intermediate level of market penetration in the micro financial sector. To determine relative poverty before and after the access to microcredit, interviews were used. The data retrieved in the interviews was used to determine the impact of micro credit on different aspects of relative poverty using the Difference in Difference method.
Significant differences are found between old and new clients as well as for the change over time. But no significant results are found for the difference in change over time for clients compared to the non-clients. The author argues that the insignificant result can either be a result of a too small sample size, disturbances in the sample selection or that this specific kind of institution have little or no affect on the current clients economical development.
Keywords:
Micro finance, MFI, Difference-in-differences, Principal component analysis, Ecuador.
Abbreviations
DiD Difference-in-Differences ME Microenterprise
MFI Micro financial Institution NGO Nongovernmental Organisation PCA Principal Component Analysis
Acknowledgements 1
Abstract 2
Keywords: 2 Abbreviations 2 1. Introduction 4
1.1 Aim 5
1.2 Microfinance 5 1.3 Country selection 6 1.4 Banco Solidario 7
2. Theory 9
2.1 Existing research on Micro Finance in Ecuador 9
2.2 Theory 10
3. Model 10
3.1 Principal Component Analysis 11
3.2 DiD 12
4. Method 13
4.1 Sample Selection 14
5. Results 15
5.1 Descriptive statistics 15 5.1.1 Demographics 16 5.1.2 Education 17 5.1.3 Income 18 5.1.4 Debts 19 5.1.5 Assets 20 5.1.6 Housing 20 5.1.7 Consumption 21 5.2 Calculations 21
5.2.1 t-Test 21 5.2.2 Estimating a poverty index through PCA 23
5.2.3 Change in the poverty index over time, DiD 26
6. Conclusions 27 6.1 Future studies 28 6.2 Authors thoughts 28
7. Sources 30
Appendix A 32 Appendix B 34
1. Introduction
About 2.7 billion people live on less than $2 a day, and 1.1 of them on less than $1 a day.1 That is almost half of the worlds population living on less than $2 a day while the richest countries2 have an average GDP per capita of $23,100, that is approximately $63 a day.3
These numbers do not take purchasing parity into account but the difference is still striking.
These numbers represents huge differences. In its millennium declaration the UN states the goal that by 2015 the proportion of the worlds people living on less then one dollar should be halved. The intent to half the population without access to safe drinking water and food is also stated.4
In 1998 the General Assembly of the UN proclaimed the year of 2005 to be the international year of microcredit. The UN claims that microcredit programs have contributed to alleviate poverty or even lifting people out of poverty around the world. "Recognizing that microcredit programmes have successfully contributed to lifting people out of poverty in many countries around the world“5 Ledgerwood (1999) claims that an estimate of 500 million economically active poor operate microenterprises in the world, of whom most have no access to formal credits.6
CGAP is an organisation consisting of public and private development agencies. They see microfinance as a strategy that combines a possible huge outreach with far-reaching impact.
And they claim that this together with the possibility to be financial sustainable make micro finance unique among development interventions.7
There are researchers, however, who do not share the certainty of the UN & CGAP. The microcredit programs may not help the poorest of the poor. If you lend money to a poor sick person you cannot expect them to invest it, but to treat the illness. Murdoch (1999) finds that the claims of poverty alleviation are mostly unsubstantiated.8 Even so he argues that it’s possible that microcredit may help other groups not extremely poor, but below the poverty
1 The World Bank (2004)
2 Includes the 30 countries in the OECD.
3 OECD (2004)
4 United Nations (2000)
5 United Nations (1998)
6 Ledgerwood (1999) p. 1
7 CGAP (2005)
8 Morduch Jonathan (1999) p. 1609
line.9 If microcredit really helps, government grants could help the poorest to reach a level where they are reachable for help by loans.
1.1 Aim
The aim of this study is to evaluate the impact of access to credit from a self-sufficient institution on the living standard of poor micro entrepreneurs. The institution investigated is Banco Solidario from Ecuador and the method used is difference-in-differences (DiD) on a poverty index computed through principal component analysis (PCA).
1.2 Microfinance
The Consultative Group to Assist the Poorest (CGAP), defines microfinance as “the supply of loans, savings and other basic financial services to the poor”10. In absence of microcredit the poor normally acquire financial services through the informal sector. These informal connections often charge a very high interest rate. Savings and insurances are provided by different methods, normally erratic and insecure. 11
It is hard to find an exact definition of microenterprises and thus also micro finance. USAID defines microenterprises as:
A “microenterprise” is a personal or family business in commerce, production, or services that employs fewer than 10 persons, is owned and operated by an individual, family, or group of individuals of relatively low income, whose owner exercises independent judgment on products, markets, and prices, and that is an important (if not the most important) source of income for the household.12
Please note that microfinance is no solution for the poorest of the poor. A borrower without food or a person who is sick might have no choice, but to use the loans for food or medicine.
The loans will not be used for investment in the micro enterprise. These clients should be helped by government or donor subsidies and grants.13
The organisations providing the credit ranges from NGO:s (non-government organisations), credit unions and cooperatives to state owned and commercial banks as well as insurance
9 Morduch Jonathan (1999) p. 1610
10 CGAP (2004)
11 CGAP (2004)
12 USAID (2005) p 4
13 Robinson (2001) p. 20
companies and other organisations. From the start NGO:s and other non-bank financial institutions led the development of microcredit.14 One of the problems with NGO:s are that they are normally not self-sustainable, but rely on direct subsidies as well as subsidies on the interest rate. This implies that they can not grow more than their donors wish to subsidize. A self-sustainable institution can on the other hand grow at the rate it wishes to.15
The reported success of early microfinance have attracted regular banks and in recent years some NGO:s have transformed themselves into licensed financial institutions, focused on microenterprises. This is especially true in parts of Latin America. There banks provide 29 percent of the funds to microenterprises and the transformed NGO:s together with other companies of that kind provide another 45 percent.16 These commercialised financial institutions in Latin America are characterized by stronger financial performance, not only compared to other microfinance institutions, but sometimes even stronger than regular banks.17
In some of the smaller countries of Latin America the access to self-sustainable microcredit institutions have managed to cover almost all of the estimated clients. But in the big countries such as Argentina, Brazil, Colombia, Uruguay and Venezuela, where 7 millions of the potential microcredit customers in Latin America live, there are almost no self-sustainable micro financial institutions (MFI).18 According to Robinson (2001) the reason for this lack of supply is the lack of information and efficient financial technology19.
That self-sufficiency is important for further growth of the microfinance sector is probable.
But if that growth are to be supported an answer to whether that kind of credit really alleviates poverty need to be found. This paper intends to answer that question: Do self-sustainable MFI:s alleviate poverty?
1.3 Country selection
The country chosen for this research should to some extent have established self-sustainable microcredit institutions that have penetrated the market. It is also preferable that the market in
14 CGAP (2004)
15 Morduch Jonathan (1999) p. 1592
16 CGAP (2001) p. 1
17 CGAP (2001) p. 2
18 CGAP (2001) p. 3
19 Robinson (2001) p. 34
the country are not too thoroughly penetrated. This to make it easier to find the “treat” and
“control” groups. This would make the countries Paraguay, Guatemala, Peru, Chile, Colombia, Dom. Rep., Ecuador and Costa Rica viable.20 Ecuador was choosen as the country of study because it had the characteristics demanded and because Banco Solidario, a suitable self-sustainable MFI, agreed on supporting this paper.
Ecuador is located on the pacific coast of South America. It neighbours Colombia to the north and Peru to the south and east. Slightly over 13 million people inhabit the relatively small country. It’s economy has traditionally been built up around the banana-industry but has in recent years come to rely on petroleum, which now account for 40% of the countries exports earnings. The country is quite dependent on its petroleum income, even though it has become more diversified with increasing export of shrimps and flowers. In 1999 Ecuador suffered an economic crisis driven mainly by natural disasters combined with falling world prices on petroleum. The result was a decline of the economy of more than 6%. This decline significantly increased poverty and decreased employment, making the need for self- employment in micro enterprises increase. This crisis led to a 70% depreciation of the Ecuadorian Sucre and the government decided to dollarize instead of risking hyperinflation.
The last years the Ecuadorian economy have benefited from stable inflation and stronger petroleum prices and the GDP growth for 2004 is an estimated 5.8%. 21
1.4 Banco Solidario
Banco Solidario was founded in 1995, at that moment 70% of the economically active population was without access to the traditional financial market.22 In difference from many other MFI:s (whom starts as NGO:s) started off as a private bank, but with a social mission.
Therefore, Banco Solidario has two official goals, social profitability as well as financial profitability. The bank started off financed with private capital. Today 51% of the shares are owned by Ecuadorians and the remaining 49% of the shares are owned by international organizations. These organizations are typically first world aid organizations or likewise23.24
20 CGAP (2001) Figure 2, p. 4
21 CIA (2005)
22 Sawyer M. (2004) p. 4
23 For example: PROFUND, SEED Capital, CAF, Swiss contact, ACCION gateway, ACCION Investments, Martin Conell, CARE, Oikocredit, SIDI, Stichting Hivos-Triodos Fond & Strichting Triodos-Doen.
24 Sawyer (2004) p. 3-5
The bank lends money to microentrepreneurs in different ways, the most common is the individual loan, which is the type of loan the clients in this study have received. The new clients of the bank normally don’t have credit history, especially the poorer ones. As this way of assessing clients will and ability to repay is not accessible the bank uses other ways. One way is through a stepwise loan size. The first loans lent to a micro finance client are typically small, between 400 and 600 dollars on a 8 to 12 months period for individual clients. As this loan is repaid larger amounts are paid out for every successful loan repaid. To be able to be eligible for an individual loan a client must have a minimum of 6 months of experience with his or her company. They also need to be between 21 and 70 years old.
The bank has had a steady and rapid growth since its founding. The last four years the size of its portfolio has grown from 40 million dollars in December 2000 to 174 million dollars in November 2004. The microfinancial clients account for 98% of the clients of the bank and 80% of the total lending. Of these 98% of the clients 52% are loans being lent to the client for investments in his or her micro businesses25. In November 2004 the average loan size of these loans to micro businessmen were 1350 dollars.26
Banco Solidario is not a NGO and have financial gains as part of their goal. The interest rates for micro finance are therefore set based on costs or market rates. Since micro finance are loans quite small per client the interest necessary to finance these loans are higher than for big loans, where economies of scale lowers the costs. The interest rate of Banco Solidarios microfinance loans were in November 2004 14.34%, but the effective cost of capital for the client also includes costs for commission and service costs amounting to 32.50%, the total cost being 46.84% in a country with inflation close to 2%. One reason for the high commissions and service fees compared relatively lower interest is that the Ecuadorian government has a maximum interest established by law and Banco Solidario and other banks chooses to put a higher effective interest through these mechanisms.27
This interest rate is in parity with what the traditional banks charge in the new micro finance sector, but above what the NGO:s do. The difference then between the kind of bank that
25 The other 47% are being lent either with gold as security 45% or for housing 2%.
26 Banco Solidario (2004)
27 Sawyer (2004) p. 11
Banco Solidario represents and the more traditional financial institutions is the focus on micro credit and also, according to the bank, its social mission.
2. Theory
2.1 Existing research on Micro Finance in Ecuador
In 1999 Brand studied the characteristics of new products developed by two MFI:s, one being Banco Solidario: pawn loans, a standardized high-margin product. Home equity loans, loans targeted to preferential or “premier” clients. Supplier–vendor loans, loans targeted at specific market niches. Her key findings are that the development is client-centred, that a diversified portfolio of products is offered to clients to increase market penetration and that a risk management techniques are used to preserve asset quality. The risk management is different for each product, but normally both back and front end. For example, on solidarity loans the bank manages its risk in two ways. By increasing loan size the bank identifies clients with
“character” collateral, and by notifying the clients supplier when the clients loan goes past due date, they put pressure on their clients to pay in time as well as help the suppliers28. Finally that a culture of innovation has been established to develop new products.29
In 2002 Banegas et al. conducted a study on the effects of financial services on microentrepreneurs. They study outreach and impact by two MFI:s, one being Banco Solidario. They also study the satisfaction of the clients. The clients studied are long time customers to the MFI and are compared to new clients as well as to non-clients. They find that the MFI:s has had a positive effect on it’s clients and that they in general are satisfied with the MFI:s. 30
In 2004 Development Alternatives, Inc conducted a major investigation on the situation of micro finance in Ecuador. Focus groups and existing data was added to the information collected from a survey of no less than 17,738 microentrepreneurs. The objectives of the study was to provide a benchmark of the microenterprise (ME) population to the MFI:s and other providers of services to the microentrepreneurs and to provide descriptions of the situation for the microentrepreneurs.31 They found that a large amount of the urban workforce
28 Brand (1999) p. 25
29 Brand (1999) pp. 1-5
30 Hivos (2000) pp. III-V
31 USAID (2005) pp. 1-2
(24%) are working in micro business. They also found that most of the micro entrepreneurs are not accessing financial services from MFI:s or NGO:s.32
2.2 Theory
Robinson (2001) identifies two reasons why microfinance is important. First microfinance provides the financial services, not accessible through the traditional financial system, needed by many people to increase and diversify their economical activities. Secondly it builds the self-confidence of the poor.33
One interesting observation from Development Alternatives, Incs study mentioned in 2.1 is that there is little indication that micro enterprises are expanding, new workers are hired when the business is started and very few workers have been hired after that.34 As has been mentioned in the introduction Murdoch (1999) argues that microfinance often supplement other incomes for the borrowers and rarely generate jobs for others.35
My study aim to add to the earlier studies by comparing clients and new clients evolution over time. It also draws its base of clients from the original client list at the start of the period instead of only including the present ones, that is, avoiding a problem of only counting the
"successful" samples.
3. Model
The aim of this paper is to identify the effect of microcredit of sustainable institutions on relative poverty of micro entrepreneurs. In order to achieve this different variables are used to sample poverty. The data used was gathered using interviews. To test for the effects of availability of microcredit on relative poverty a Difference-in-Differences (DiD) model is used. The DiD-analysis is performed on quantitative data. As “poverty” is not a directly quantifiable variable, an index for relative poverty is created. This index is created using Principal Component Analysis (PCA).
32 USAID (2005) p. 127
33 Robinson (2001) p. 37
34 USAID (2005) p. 128
35 Murdoch (1999) p. 1610
3.1 Principal Component Analysis
Principal Component Analysis is a method that tries to identify underlying variables, factors, that explain common correlation among observed variables36. It is used to try to most effectively find the underlying variables. The result of the PCA is an index that combines the information from the different variables. If indicators are related in more than one way, more then one component will be created. Each such component measures a similarity of the households, although only one measure relative poverty as is the variable of interest in this study.37
In this thesis the “scores” in the index represents the households relative poverty compared to the other households in the sample. In this thesis the PCA is used on indicators that show a significant correlation with the benchmark variable, “money spent on shoes & clothes last year, per capita”. This is done to get a stronger poverty component, that is, one that strongly associates with the variables included that measure poverty.38
Two poverty index scores is estimated for each client, one value for the present time period and one value for the older time period.39
To evaluate the quality of the components found in the PCA model, three different tests of the variables are investigated. The size of the Eigen value indicates the amount of variance in the PCA that is explained by the identified component. A higher value indicates a larger degree of explanation. The Eigen value needs to be at least 1 if the component is to be seen as an indicator of an underlying common variable.40
The second test checks the relative size of the communalities. They indicate how well the different indicators in the PCA combine to produce the underlying components. They are comparable to the R2-statsticva in a regression analysis, and as such they should not be close to zero if the component is to be used.41
36 SPSS
37 Henry Carla, Sharma Manohar, Lapenu Cecile & Zeller Manfred (2003) p. 130
38 Henry Carla, Sharma Manohar, Lapenu Cecile & Zeller Manfred (2003) p. 130
39 Henry Carla, Sharma Manohar, Lapenu Cecile & Zeller Manfred (2003) pp. 125-130
40 Henry Carla, Sharma Manohar, Lapenu Cecile & Zeller Manfred (2003) p. 137
41 Henry Carla, Sharma Manohar, Lapenu Cecile & Zeller Manfred (2003) p. 138
A Kaiser-Meyer-Olkin measure of sampling adequacy should be run to see if the model is adequate. The test compare the observed correlation coefficient with partial correlation coefficients. A higher value indicate a better model. A score above 0.60 is seen as adequate, above 0.70 are good, 0,80 commendable and above 0,90 exceptional.42
3.2 DiD
The DiD-method measures the difference of a variable over time for two different groups. It’s a powerful tool where group and time specific effects are allowed for. The individuals in one of the groups, the “treated” group are affected by the factor whose effects I want to investigate and the individuals in the “control” group are not. The difference over time in the “control”
group are subtracted from the difference in the “treated” group. This way the effects of time as well as the effect of structural changes not related to the effect investigated are explained and removed from the effect I investigate.43
DiD does not remove structural differences between groups exogenously imposed during the time period in question. In some cases explanatory variables are interacted to allow for partial effects to change over time.44
To use DiD data from two periods in time is needed. The first from just before the change, whose effects is studied. The second some time after the change. Here the time before the change is denoted with t=0 and for the period of time after the change t=1.
Two groups are also needed. One who are affected by the change, the “treated” group and one who are not, the “control” group. Here the “treated” group consist of the households who started borrowing money from Banco Solidario in 2002. They will be denoted with i=t. The households in the control group, here households who started to borrow money in 2005 will be denoted with i=c.
I produce the DiD-equation by using both the time and group variables individually and multiplied.
42 Henry Carla, Sharma Manohar, Lapenu Cecile & Zeller Manfred (2003) p. 139
43 Wooldridge (2002) pp. 129-130
44 Wooldridge (2002) p. 129
yit = α + βxit + γtimeit + δ(xit * timeit) + εit (Equation 1)
In the estimated equation δ is the variable of interest, the DiD-estimator. If the control group is assumed to be comparable to the treated group δ is the difference between the two groups depending only on the variable x, in this case the access to microcredit. However there might be other reasons for differences in relative poverty than micro credits, these differences will be caught by the time and group parameters.
4. Method
The data was gathered using personal interviews made February to April 2005. The households that was interviewed belonged to either of two groups. The “treated” group was made up of households that started receiving microcredits 3 years earlier45. The “control”
group was households that had recently been accepted for microcredits and just started receiving it46.
The strengths of this method are that it partly deals with the problems of self-selection and selection bias. Self-selection is not such as great problem here compared to an approach where clients would be compared to non-clients. The selection bias is avoided as all households would have made and been selected by the same selection, if it could be argued that selection methods are relatively stable over time. Finally, one additional positive effect of this method was that getting in contact with the chosen households was relatively easy, since they were are at least have earlier been customers of the bank.
One problem with this method might be that there could exist bias, due to the fact that the reasons for the “control” group to apply for credit might be that they have been doing bad for some time before and one would then receive an overestimation off the positive effect of the microcredit as the group have been doing worse than the average household.
To avoid making the mistake of only choosing the successful clients the base of “treated”
clients were made up using the new clients in January 2002 instead of current old clients. This problem somewhat remains as the old clients who had a relatively poor development since 2002 to a larger extent refused to be interviewed or had emigrated.
45 New client in January 2002.
46 New client in January 2005.
Some of the clients who where sorted into the control group turned out to be the clients of other financial institutes. This somewhat distort the control group and is taken into consideration in the conclusion.
To be able to use the DiD-method data from two periods is needed. This was achieved by asking the interviewed individuals about their situation three years ago. Although collecting data by recall, that is asking people about past times, has its flaws it is the only way to get the information. To use older data from another investigation for the first period would be to rely on that not only the same variables have been used to estimate poverty, but also that the emphasis put on each variable would be the same in the older data as in the new one. There are of course problems with using data by recal,l because people aren’t able to exactly recall their situation and what assets they had 3 years prior to the interview.47 To avoid this problem the intention has been to focus the questioning on factors that are easy to remember.
As a basis for the questionnaire the CGAP “Microfinance Poverty Assessment Tool”48 is used. To be able to use the recall method some questions have been deleted from the questionnaire. The per capita expenditure in period (t=1) on clothing and footwear will be used as the benchmark indicator. 49 Then a poverty index for period (t=1) will be retrieved and used to estimate poverty for the individuals in both periods.
To check for difference in relative poverty due to access to the microcredit the DiD-model will be used.
4.1 Sample Selection
To select what households to include in the sample the population of 75 treated clients and 100 control clients were each given an id-number. Using a random program 60 numbers from the treated group was drawn and 65 from the control group. Of these 125 households a total of 84 interviews were completed. That gives a missing value of 41. I have tried, but been unable to locate or interview 11 of the households in the control group and 30 of the households of the treated group. There is a difference in the percentage of the numbers of interviews
47 This is particularly a problem with the poorest clients who normally have a very low level of education and have a hard time putting dates and years to events.
48 Henry Carla, Sharma Manohar, Lapenu Cecile & Zeller Manfred (2003)
49 Henry Carla, Sharma Manohar, Lapenu Cecile & Zeller Manfred (2003) pp. 125-127
successfully made between the two groups. One reason is that the control group is made up from new clients, normally eager to be on a good stand with the bank. The treated group on the other hand include “morosos”, that is clients who has failed to repay their debts. These
“morosos” actively efforts to locate them for interviews, adding a problem of self-selection bias to the study. This might affect the result of this paper, because some of the “failures”
have removed themselves from the study. Some of the clients in the treated groups has actually left the country to work abroad, normally in Spain. The last decades a big percentage of the Ecuadorian population has emigrated. This of course make it quite hard to interview them.
There is always a trade-off between sample size and depth of questionnaire. To be able to maximize the number of clients interviewed, without letting the quality fall, a urban area was chosen for the interviews, namely the Recreo area in the poorer southern part of Quito. I was helped by the banks client assessors to contact and visit the chosen clients. This proved to be a time consuming business as not all the clients had telephones and could not be contacted before for setting up a meeting. Also sometimes the time looking for different houses and apartments could be quite long.
The questionnaire used include 46 questions 44 of these asked on the current situation and the situation 3 years ago. 2 of the questions were only asked on the current situation50.
5. Results
First descriptive statistics of the different areas of questioning are presented. Then the regressions necessary for constructing the poverty index. Last the results of the DiD-method are presented.
5.1 Descriptive statistics
The interviews are made up by questions51 questioning different aspects of the clients situation, among others their ability to generate income, consume and educate their children.52
50 Monthly spending on food and last years spending on shoes and clothes.
51 The complete questionnaire is presented in Appendix A & B
52 Not all of the categories are presented in this chapter. Answers to some of the questions has been removed as they did not add anything of interest to the project as the answers were either almost all zero or all yes. These indicators are incapacitated family members, pregnant female members of the family who are under 16, children labor, access to electricity, WC, piped water, number of meals a day consumed and ownership of radio.
A brief overview of the answers of the control and treated groups about the situation is presented below. The questions are presented divided into seven subcategories:
Demographics; Education; Income; Debts; Assets; Housing & Consumption. Each category will be represented in a table.
The first value of a indicator in this table is the average of the answers received, the second value is the standard deviation. All indicators have the sample size of 65 for the control group and 60 for the treated group. The missing value is 11 for the control group and 30 for the treated group in both time periods for all variables except money spent on food, money spent on clothes and shoes, kids in private/fiscal school and kids in public school.
Money spent on food and money spent on clothes & shoes were only asked on time period 1, the recent period. The indicators for kids participation in private or public school are depending on the fact that the family have kids in school and while the sample size remain the same the missing value for the treated group in the earlier period is 42 and in the more recent period 43. The missing value on these two parameters for the control group is 28 for the earlier period and 27 for the more recent one.
5.1.1 Demographics
Table 1.1 below features the average of the answers received from the clients from the control group and the treated group. T=0 is the values three years ago (Early 2002) and T=1 are the current values ( Early 2005).
The treated clients have a higher average value of females and males aggregated between 18- 55 years than the control group, but the control group have had a positive change on this indicator over time.
Table 1; Descriptive statistics for demographics
Demographics
Control Group Treated Group
T=0 T=1 Change T=0 T=1 Change Difference
Quantity of males 18-55 0,89 1,05 18% 1,24 1,21 -3% -21%
Std dev 0,68 0,70 3% 0,57 0,48 -15% -18%
Quantity of females 18-55 1,00 1,04 4% 1,10 1,14 3% -1%
Std dev 0,54 0,54 0% 0,66 0,63 -5% -5%
Quantity of Children 0,49 0,40 -19% 0,31 0,41 33% 52%
Std dev 0,71 0,68 -5% 0,53 0,67 26% 31%
Quantity of males 7-17 0,59 0,58 -2% 0,34 0,41 20% 22%
Std dev 0,62 0,68 9% 0,66 0,67 2% -7%
Quantity of females 7-17 0,35 0,42 21% 0,52 0,55 7% -14%
Std dev 0,55 0,62 14% 0,81 0,81 0% -14%
Quantity of males 55+ 0,13 0,13 0% 0,03 0,03 0% 0%
Std dev 0,33 0,33 0% 0,18 0,18 0% 0%
Quantity of females 55+ 0,13 0,13 0% 0,07 0,07 0% 0%
Std dev 0,33 0,33 0% 0,25 0,25 0% 0%
Female clients (% of whole) 0,45 0,45 0% 0,59 0,59 0% 0%
Std dev 0,50 0,50 0% 0,49 0,49 0% 0%
5.1.2 Education
Experience from the interviews made and from experiencing and observing the country indicate that a large majority of the kids in Ecuador attend school. Kids from upper and middle class attend private schools and there are quality differences, but most kids go to school. 99.5% of the relevant age group attend primary education and 78% reach 5:th grade.53 Public school is normally open from morning to mid day which, make it possible for the poorer kids to work in the afternoon. The result of the interviews goes well with the statistics.
One difference over time between the two groups are the number of kids attending private versus public school. For the treated group a shift from public to private school can be observed. No such shift is visible in the control group.
53 World bank (2005)
Table 2; Descriptive statistics for education
Education
Control Group Treated Group
T=0 T=1 Change T=0 T=1 Change Difference
Kids in school 1,09 1,09 0% 1,00 1,07 7% 7%
Std dev 1,01 0,98 -4% 1,02 1,05 3% 7%
Kids in public school* 0,68 0,69 1% 0,29 0,19 -34% -35%
Std dev 0,46 0,46 -1% 0,46 0,38 -17% -16%
Kids in private/fiscal school* 0,32 0,31 -3% 0,71 0,81 14% 17%
Std dev 0,46 0,46 -1% 0,46 0,38 -17% -16%
Monet spent on education 5,93 7,57 28% 7,79 8,28 6% -22%
Std dev 20,86 23,37 12% 23,81 17,20 -28% -40%
Schooling of client ** 1,64 1,62 -1% 1,66 1,66 0% 1%
Std dev 0,88 0,90 3% 0,60 0,60 0% -3%
University students 0,09 0,13 40% 0,21 0,17 -17% -57%
Std dev 0,29 0,33 16% 0,41 0,38 -7% -23%
* = Percent of kids in school that goes to that type. Families without children in school will here have the result 0.
** = Dummy variable, read as: 0 = No education, 1 = Primary school, 2 = Secondary school and 3 = Higher studies
5.1.3 Income
When it comes to income changes the control group claims a positive shift on average by 29%
during the last three years. The treated client on the other hand reports almost the double in income increase (50%). This is an interesting difference and even though some of the explanation with outliers that can no explain the whole difference. Both groups have on average increased the number of family members working in the micro enterprise. Partly this is due to a natural development of a family business growing larger at the same time as the family members reach working age. But this is also a result of a harsh economical situation and the difficulty of finding work outside the micro enterprise. For the treated group the number of family members with a wage has also increased. As the number of adults has been quite stable for that group the result is a bit surprising, but part of the answer lies in that some of the clients has been forced to have double work, both in the ME and with wage.
Table 3; Descriptive statistics for income
Income
Control Group Treated Group
T=0 T=1 Change T=0 T=1 Change Difference
Average Monthly income 526,82 679,09 29% 691,38 1 035,69 50% 21%
Std dev 373,69 509,89 36% 516,11 656,91 27% -9%
Family members in micro enterprise 1,16 1,35 16% 1,21 1,41 17% 2%
Std dev 0,76 0,69 -8% 0,76 0,67 -12% -3%
Family members with wage 0,64 0,60 -6% 0,59 0,69 18% 23%
Std dev 0,72 0,65 -10% 0,62 0,65 5% 15%
Amount of hired workers 0,36 0,56 55% 0,10 0,07 -33% -88%
Std dev 0,90 1,40 55% 0,40 0,36 -9% -64%
Money from family abroad 14,55 26,82 84% 8,62 8,62 0% -84%
Std dev 62,28 90,56 45% 37,30 37,30 0% -45%
Retirement grants 21,38 29,29 37% 0 0 0% -37%
Std dev 54,71 76,38 40% - - 0% -40%
5.1.4 Debts
Debts to Banco Solidario has substantially increased during the period for the control group.
This is not surprising as the control group is made up by new clients who have just received their first loan. The fact that the total debt has somewhat lowered in the treated group is due to the fact that while most of the clients remaining as clients has increased their loan size part of the treated group no longer remain as clients. The debts to informal lenders, friends and family are low. Debts to other financial institutions went up 54% in the treated group.
It should be stated that some of the households might take loans from Banco Solidario to repay loans from other institutions, and would then be quiet about that when asked about it.
Table 4; Descriptive statistics for debts
Debts
Control Group Treated Group
T=0 T=1 Change T=0 T=1 Change Difference
Debts to Banco Solidario - 1 170,91 N/A 1 720,69 1 694,83 -2% N/A Std dev 53,44 1 339,83 2407% 3 642,32 2 159,76 -41% #####
Debts to other fin. inst. 288,00 424,07 47% 224,14 344,83 54% 7%
Std dev 806,06 1 431,43 78% 805,01 821,52 2% -76%
Debts to informal lenders 54,55 3,64 -93% 34,48 34,48 0% 93%
Std dev 400,83 26,72 -93% 126,70 182,47 44% 137%
Debts to friends & family 276,36 300,00 9% 13,79 98,28 613% 604%
Std dev 2 003,81 2 010,43 0% 50,68 334,89 561% 560%
5.1.5 Assets
The value of machinery and vehicles has increased for both groups. The increase rate for telephones, mobile phones and computers is larger in the treated group.
Table 5; Descriptive statistics for assets
Assets
Control Group Treated Group
T=0 T=1 Change T=0 T=1 Change Difference
Value of Machinery 1 697,18 3 429,00 102% 3 088,97 5 623,45 82% -20%
Std dev 7 793,14 14 786,80 90% 11 057,47 16 462,69 49% -41%
Value of Vehicles 823,64 1 263,64 53% 1 660,00 3 417,24 106% 52%
Std dev 2 339,93 3 212,10 37% 5 282,55 5 948,81 13% -25%
Savings 96,36 74,55 -23% 51,72 93,79 81% 104%
Std dev 307,31 260,92 -15% 201,07 276,64 38% 53%
TV (dummy) 0,95 0,96 2% 1,00 0,97 -3% -5%
Std dev 0,23 0,19 -18% - 0,18 N/A N/A Telephone (dummy) 0,65 0,67 3% 0,86 1,30 50% 48%
Std dev 0,48 0,47 -1% 0,34 1,88 446% 447%
Mobile phones 0,31 1,09 253% 0,29 1,10 286% 33%
Std dev 0,63 1,12 78% 0,64 1,42 123% 46%
Computers 0,18 0,35 90% 0,17 0,38 120% 30%
Std dev 0,43 0,51 19% 0,46 0,55 20% 1%
Internet connection (dummy) 0 0,13 N/A 0 0,07 N/A N/A Std dev - 0,33 N/A - 0,25 N/A N/A
5.1.6 Housing
The interviews included several questions that afterward seem quite pointless in distinguishing changes between groups and over time. Electricity, access to piped water and water closets were such questions. They do serve the purpose of demonstrating the fact that the clients of the bank, treated or control normally have access to these services and are thus not part of the poorest part of population. As a comparison the fraction of urban population using adequate sanitation facilities in 2002 was 80 percent.54
Both groups have had a slight shift towards bigger homes. Both groups have also invested in home improvement. The positive change is bigger for the treated group then for the control group.
54 Unicef 2005
Table 6; Descriptive statistics for housing
Housing
Control Group Treated Group
T=0 T=1 Change T=0 T=1 Change Difference
Rooms 3,45 3,78 9% 4,14 4,21 2% -8%
Std dev 1,41 1,55 9% 1,41 1,40 -1% -10%
Value of recent house imp. 676,02 1 126,55 67% 235,71 1 592,41 576% 509%
Std dev 1 937,09 3 634,50 88% 836,62 5 802,78 594% 506%
Ownership of house (dummy) 0,55 0,55 0% 0,72 0,79 10% 10%
Std dev 0,50 0,50 0% 0,45 0,41 -9% -9%
5.1.7 Consumption
No big change can be noted nor any big differences over time. This is probably due to the fact that it is hard to remember your situation 3 years ago so precisely. That is the reason why the questions asked on money spent on food last week as well as clothes and shoes last year is only asked for the current time period. The results for the other two indicators strengthens the theory that the time span is too long for adequate answers on consumption matters.
There is quite a difference between the control and the treated group in money spent on food and on clothes & shoes. That can be seen as a indicator on the difference in relative poverty between the groups. But the reason for this and the change over time can not be decided by this data as there is no values for period one.
Table 7; Descriptive statistics for consumption
Consumption
Control Group Treated Group
T=0 T=1 Change T=0 T=1 Change Difference
Days with meat/fish/chicken a week 5,36 5,44 1% 5,72 5,72 0% -1%
Std dev 1,91 1,87 -2% 1,91 1,80 -6% -4%
Clothes & shoes last year N/A 496,04 N/A N/A 721,38 N/A N/A Std dev N/A 359,29 N/A N/A 683,98 N/A N/A Expenditure on food per week N/A 51,55 N/A N/A 72,90 N/A N/A Std dev N/A 26,46 N/A N/A 52,75 N/A N/A
5.2 Calculations
5.2.1 t-Test
In order to measure if the differences between the treated and control groups for some of the indicators are significant a t-test is used. The null hypothesis tested for each indicator is that
there is no difference between the mean for the control group and the mean for the treated group. The alterative hypothesis is that so is not the case. The tests are made on the indicators for the recent time period. Differences between the groups are interesting as one of the differences of the groups is that the treated group started being clients to the bank 3 years prior to the study.
The results of the hypothesis test55 are presented in table 8 below. There are 8 variables where the difference is significant on the 1%-nivel, Percentage of kids in school in public school, Percentage of kids in school in private or fiscal school, Average monthly income, Retirement or other grant, Value of vehicles, Telephone, Ownership of house & Money spent on food a week. On the 5%-nivel an extra three variables have significantly different means: Amount of hired workers in ME, Meals a day & Money spent on clothes & shoes last year.
Table 8; Two-sample t test with equal variances.
Two-sample t test with equal variances df
Mean Difference
Std. Error Difference t Quantity of grown ups: 82 - 0,08 0,20 - 0,41 Quantity of kids: 82 0,07 0,27 0,28 Clients sex: 82 - 0,16 0,11 - 1,37 Quantity of kids attending school: 82 0,08 0,23 0,34 Percentage of kids in school in public school: 53 0,47 0,13 3,60 Percentage of kids in school in private or fiscal school: 53 - 0,52 0,13 - 4,15 Schooling of client: 82 0,05 0,19 0,26 Familymembers in university: 82 - 0,04 0,08 - 0,46 Average monthly income: 82 - 324,02 131,16 - 2,47 Familymembers working in ME: 82 - 0,10 0,16 - 0,63 Family members working with wage: 82 - 0,06 0,15 - 0,37 Amount of hired workers in ME: 82 0,51 0,27 1,91 Income from family members abroad: 82 18,98 17,62 1,08 Retirement or other grant: 82 29,83 14,22 2,10 Debts to other financial institutions: 82 90,74 289,13 0,31 Debts to informal persons: 82 - 29,63 25,22 - 1,17 Debts to friends and family: 82 210,56 377,59 0,56 Value of companies machinery: 82 - 1 943,50 3 547,68 - 0,55 Value of vehicles: 82 - 2 016,30 1 006,92 - 2,00 Savings: 82 - 14,74 61,42 - 0,24 Radio: 82 - 0,00 0,04 - 0,09 TV: 82 - 0,00 0,04 - 0,09 Telephone: 82 - 0,27 0,09 - 2,85 Cell phone: 82 0,04 0,28 0,16 Computer: 82 - 0,01 0,12 - 0,12
55 Test statistic: [(xc– xe)2/xte] where: Xc = Observed frequency for control group; Xe = Expected frequency for control group.
Internet: 82 0,06 0,07 0,89 Meals a day: 82 - 0,09 0,05 - 1,73 Meals of fish, chicken or meat a week 82 - 0,10 0,43 - 0,23 Rooms: 82 - 0,32 0,35 - 0,92 Recent expenditure on house improvements: 82 - 433,41 1 037,88 - 0,42 Ownership of house: 82 - 0,26 0,11 - 2,45 Money spent on clothes & shoes last year: 82 - 199,89 114,85 - 1,74 Money spent on food a week: 82 - 18,74 8,75 - 2,14
The t-test indicate that there are differences between the households in the later time period studied. To investigate the variable I am interested in, change over time in relative poverty, a DiD-analysis is conducted. To be able to use the DiD-model on relative poverty an index for relative poverty is created using principal correlation analysis (PCA) in chapter 5.2.2 below.
5.2.2 Estimating a poverty index through PCA
The poverty index is estimated using principal correlation analysis (PCA). To decide what indicators to include in the PCA the indicators are checked for correlation with the benchmark variable, money spent on clothes & shoes last year per capita. The linear correlations coefficients between the benchmark indicator are presented in table 9 below. There are 4 variables with a statistically significant correlation on the 1%-level to the benchmark indicator. The percentage of kids in public school (-0,363); percentage of kids in school in private or fiscal school (0,355); average income (0,621); and numbers of cell phones in the household (0,301). On the 5%-level 3 more indicators can be added: computer in household (0,262); number of rooms (0,229) and value of vehicles (0,257).
The positive (negative) correlation indicates that for example a family with a higher percentage of kids in private or fiscal school (public school) spend more (less) money on clothes and shoes per capita, and will thus get a higher (lower) score in the poverty index.
Table 9; Linear Correlation Coefficient on “Money spent on clothes & shoes last year, per capita”
Dependent Variable: Money spent on clothes & shoes
Method: Pearson
Correlation
Sample size: 84
Variable Correlation Coef. Probability N Missing Value
Amount of hired workers in ME 0,173 0,115 84 0
Average monthly income 0,621 0,000 84 0
Computer in household (dummy) 0,262 0,016 84 0
Days a week with fish, chicken or meat 0,168 0,127 84 0
Debts to friends or family 0,150 0,175 84 0
Debts to informal persons -0,095 0,388 84 0
Debts to other financial institutions -0,037 0,740 84 0
Family members in University 0,068 0,539 84 0
Family members working in ME -0,197 0,072 84 0
Family members working outside ME -0,041 0,713 84 0
Internet in household (dummy) 0,073 0,510 84 0
Kids in school 0,034 0,761 84 0
Mobile phones 0,301 0,005 84 0
Money from family abroad -0,058 0,598 84 0
Money spent on education of kids 0,164 0,137 84 0
Money spent on recent home improvement -0,034 0,757 84 0
Number of grown ups (>=18) -0,082 0,459 84 0
Number of kids (<18) 0,000 0,997 84 0
Number of rooms disposed by family 0,229 0,036 84 0
Ownership of house (dummy) 0,047 0,669 84 0
Percentage of kids in private school 0,355 0,008 55 29
Percentage of kids in public school -0,363 0,006 55 29
Radio in household (dummy) 0,148 0,179 84 0
Retirement grants -0,111 0,313 84 0
Savings 0,167 0,128 84 0
Schooling of client 0,178 0,104 84 0
Telephone (Dummy) 0,214 0,051 84 0
TV in household (dummy) 0,071 0,521 84 0
Value of machinery in ME 0,082 0,461 84 0
Value of vehicles 0,257 0,018 84 0
The seven variables significantly correlated with the benchmark indicator on the 5%-level are used in a PCA56. The results from the tests performed are presented in table 10, 11 and 12 below, the component matrix is presented in table 13.
The first component of the PCA explains 35,73 % of total variance and has a Eigen value of 2,5. This model is constructed to find a poverty index and since the Eigen value is over 1 and
56 Households without children has been given the value “0” in both the indicator for children in private or fiscal school and children in public school.
the explained variance is relatively high the first component is considered the poverty index.
None of the communality indicators are close to zero and the Kaiser-Meyer-Olkin measure of sampling adequacy is 0,714, with scores above 0,7 seen as good. The model is seen as adequate.
Table 10; PCA, Kaiser-Meyer-Olkin Measure of Sampling Adequacy.
Kaiser-Meyer-Olkin Measure of Sampling Adequacy 0,7144753
Table 11; PCA, Communalities
Communalities Initial Extraction
Public 1 0,779969144
Pivate 1 0,609054037
Mobile 1 0,572030534
Income 1 0,581205292
Comp 1 0,396590002
Rooms 1 0,483886705
Val_vehic 1 0,267339776
Table 12; PCA, Total variance explained
Total Variance Explained
Component Eigen Values Percentage of variance explained Cumulative % 1 2,50 35,74 35,74 2 1,19 16,98 52,72 3 0,88 12,61 65,33 4 0,76 10,82 76,15 5 0,71 10,13 86,29 6 0,51 7,29 93,58 7 0,45 6,42 100,00
Component 1 in table 13 below is used to estimate the relative poverty index. Descriptive statistics for the poverty index is presented in table 14 below. Both the treated group and the control group get higher scores on the poverty index in 2005 then in 2002. The change over time is greater for the treated group than for the control group. This difference over time is analyzed more sensitively with a DiD-analysis in chapter 5.2.3.
Table 13; PCA, Component Matrix
Component
Variable 1 2
Public - 0,415 0,780 Pivate 0,603 - 0,495 Mobile 0,637 0,408 Income 0,756 0,095 Comp 0,592 0,215 Rooms 0,613 0,328 Val_vehic 0,512 - 0,075
Table 14; Descriptive statistics for Poverty Index, using PCA.
Poverty index, using PCA Descriptive Statistics
Samplesize: 168 Missing Value: 0
Group Time N Mean Maximum Minimum Std. Deviation
Treated 2002 30,00 0,00 1,67 -1,04 0,75
Treated 2005 30,00 0,64 4,05 -1,09 1,10
Control 2002 54,00 -0,39 1,86 -1,56 0,82
Control 2005 54,00 0,04 3,08 -1,56 1,06
Both Both periods 168 0,00 4,05 -1,56 1,00
5.2.3 Change in the poverty index over time, DiD
Each household now have one poverty index score for the older time period and one for the more recent one. To analyse what part of the change between these two time periods that adhere to the availability of loans from the MFI studied a DiD-analysis is conducted. The result is presented in table 15 below.
The variables for treated client and time period are both significant on the 5%-level. The indicator for treated client is positive and thus at either of the time periods a new client 2005 is relatively poorer than a new client in 2002. The time variable is also positive and indicates a positive relation between the passing of time and reduction of poverty. The DiD-coefficient is not statistically significant. The possibility that the effect of availability of credit on change over time in relative poverty is equal to zero cannot be dismissed due to this test.
The explanatory ability is quite low for the model (12%).
