Effectiveness of PRIME Interventions in Greater Rangpur at the Household level and Institutional level: A Longitudinal Approach 3rd Round Evaluation Report Prepared for: Palli Karma-Sahayak Foundation (PKSF) Prepared by Institute of Microfinance (InM) July 05, 2011 PRIME 3rd Round Report. Please do not cite or circulate. Project Team Atonu Rabbani Department of Economics University of Dhaka Mohammad Monirul Hasan Md. Mehadi Hasan Institute of Microfinance (InM) Institute of Microfinance (InM) Tunazzina Choudhury Mithun Aparna Howlader Project Officer - PRIME Project Officer - PRIME Page ii PRIME 3rd Round Report. Please do not cite or circulate. Contents List of Tables ..................................................................................................................................... v List of Figures ....................................................................................................................................vi Executive Summary ......................................................................................................................... viii Acknowledgement............................................................................................................................. xi 1. Introduction .............................................................................................................................. 1 1.1 Background ........................................................................................................................ 1 1.2 Evolution of the Program ................................................................................................... 4 1.3 Outreach of PRIME ............................................................................................................. 5 1.4 Overview of this Report...................................................................................................... 8 2. Methods .................................................................................................................................. 10 2.1 Sampling .......................................................................................................................... 10 2.2 Data ................................................................................................................................. 11 2.3 Identification.................................................................................................................... 13 2.4 Notes on Analytical Framework........................................................................................ 17 2.5 Discussion and Concluding Remarks on Identification....................................................... 18 3. Impact of PRIME on Selected Parameters ................................................................................ 19 3.1 Introduction ..................................................................................................................... 19 3.2 Consumption Ordering ..................................................................................................... 19 3.3 Other Outcome Parameters and Age of PRIME Participation ............................................ 24 3.4 Crisis Coping..................................................................................................................... 30 3.5 Health, Perception and Decision Related Aspects ............................................................. 33 3.5.1 Awareness about Health Care................................................................................... 33 3.5.2 Perception regarding Child Education ....................................................................... 35 3.5.3 Participation of Women in Decision Making ............................................................. 35 3.6 Discussion and Concluding Remarks on Impact of PRIME ................................................. 35 4. Overall Assessment of the Program ......................................................................................... 37 5. Concluding Remarks ................................................................................................................ 41 A. Operational Sustainability of the PRIME Branches.................................................................... 43 1. Preamble ............................................................................................................................. 43 2. Evaluation of the PRIME branches........................................................................................ 43 3. Performance Parameters ..................................................................................................... 44 Productivity and Efficiency ....................................................................................................... 45 Financial Spread and Operational Self-Sufficiency .................................................................... 45 4. Sample................................................................................................................................. 45 Page iii PRIME 3rd Round Report. Please do not cite or circulate. 5. Description of the Data ........................................................................................................ 47 6. Assessing Financial Performance against Age of the Program ............................................... 49 7. Dynamics of Branch Performance over the Age of the Program ........................................... 50 8. Financial Spread................................................................................................................... 52 9. Operational Self-Sufficiency (OSS) ........................................................................................ 53 10. Discussion ........................................................................................................................ 54 11. Concluding Remarks ......................................................................................................... 55 Page iv PRIME 3rd Round Report. Please do not cite or circulate. List of Tables Table 1: Income Generating Activities (Thousand IGAs) by Major Sub-Categories. ............................. 7 Table 2: Sampling ............................................................................................................................ 10 Table 3: Summary Statistics for Selected Variables from Third Phase of Household Survey. ............. 12 Table 4: Summary Statistics of Selected Household and Village Characteristics by Participation Status. ........................................................................................................................................................ 14 Table 5: Percentage distribution of PRIME membership duration by district .................................... 16 Table 6: Summary statistics of selected household characteristics by duration of PRIME membership (in years) ......................................................................................................................................... 16 Table 7: Comparisons between “Benchmark” Households and PRIME Participants with less than One Year in the Program. ........................................................................................................................ 25 Table 8: Land and Non-Land Assets against Age of PRIME Participation. .......................................... 26 Table 9: Asset value and Savings against Age of PRIME Participation. .............................................. 27 Table 10: Income against Age of PRIME Participation. ..................................................................... 28 Table 11: Food and Non-Food Expenditure against Age of PRIME Participation................................ 29 Table 12: Engagement in Agricultural and Non-Agricultural Self-Employment against Age of PRIME Participation. ................................................................................................................................... 30 Table 13: POs and Coverage ............................................................................................................ 46 Table 14: Summary Statistics for PRIME Branches............................................................................ 48 Table 15: Age of PRIME Implementation.......................................................................................... 49 Table 16: Age of PRIME Implementation over Financial Years 2008-2010. ....................................... 49 Table 17: Operational Self-Sufficiency (OSS) for different categories of PRIME POs. ......................... 55 Page v PRIME 3rd Round Report. Please do not cite or circulate. List of Figures Figure 1: Map of Bangladesh showing Rangpur division in red. .......................................................... 1 Figure 2: Loan Outstanding and Total Members, 2009-2010. ............................................................. 3 Figure 3: Average Loan Outstanding Per Borrower............................................................................. 4 Figure 4: Disbursement of Microcredit under PRIME Program. .......................................................... 4 Figure 5: Savings Mobilized under PRIME. ......................................................................................... 5 Figure 6: Income Generating Activities (IGAs) associated with Microcredit under PRIME Program. .... 5 Figure 7: Number of Remittance Services Rendered under PRIME. .................................................... 8 Figure 8: Number of Health Services (in Thousands) Rendered under PRIME. .................................... 8 Figure 9: Distribution of the Households over Consumption Ordering during “Normal” Time........... 20 Figure 10: Distribution of the Households over Consumption Ordering during Monga Time. ........... 20 Figure 11: Cumulative Distribution of Consumption Ordering during “Normal” Time ....................... 21 Figure 12: Cumulative Distribution of Consumption Ordering during Monga Time........................... 21 Figure 13: Fraction (%) of Household Reporting at least Three Meals/Day during “Normal” Time. ... 22 Figure 14: Fraction (%) of Household Reporting at least Three Meals/Day during Monga Time. ....... 22 Figure 15: Marginal Impact of the Program over Age of Participation on Reporting Three Meals/Day during “Normal” Time. .................................................................................................................... 23 Figure 16: Marginal Impact of the Program over Age of Participation on Reporting Three Meals/Day during Monga Time. ........................................................................................................................ 23 Figure 17: Fraction of Households Adopting Self-Coping Mechanism. .............................................. 31 Figure 18: Marginal Impact of Program on Reporting Self-Coping Mechanism to Cope The Shock . .. 31 Figure 19: Value of Housing by Duration of PRIME Participation. ..................................................... 32 Figure 20: Participation in Non-PRIME MFI Programs by Duration of PRIME Participation. ............... 32 Figure 19: Fraction (%) of Households Reporting to have Treatment from “Quack”. ........................ 33 Figure 20: Fraction (%) of Households Reporting Interest to Educate Their School-Going Aged Children........................................................................................................................................... 34 Figure 21: Fraction (%) of Household Reporting Mother’s Participation in the Decision of Children’s Marriage ......................................................................................................................................... 34 Figure 22: Growth Curve for Number of MFI Branches Implementing PRIME .................................. 46 Figure 23: Borrower Per Field Officer by Age of Branches. ............................................................... 50 Figure 24: Portfolio Outstanding Per Field Officer (Thousand Taka) and Age of Branches. ................ 51 Figure 25: Deposit Outstanding Per Credit Officer (Thousand Taka) and Age of Branches. ............... 51 Figure 26: Salary and Benefit as a % of Average Portfolio Outstanding and Age of Branches. ........... 51 Figure 27: Operating Cost Ratio as a % of Average Portfolio Outstanding and Age of Branches. ....... 51 Page vi PRIME 3rd Round Report. Please do not cite or circulate. Figure 28: Financial Spread and Age of Branches. ............................................................................ 52 Figure 29: Distribution of OSS as 2010 for Branches Established through Implementing PRIME (N = 132)................................................................................................................................................. 53 Figure 30: Distribution of OSS as 2010 for Pre-existing MFI Branches Implementing PRIME (N = 99) 53 Figure 31: Percentage of Branches with OSS above 100%. .............................................................. 54 Figure 32: Operational Self-Sufficiency (OSS, %) and Age of Branch. ................................................ 54 Page vii PRIME 3rd Round Report. Please do not cite or circulate. Executive Summary Introduction Palli Karma Sahayak Foundation (PKSF) initiated the project entitled Programmed Initiatives for Monga Eradication (PRIME) in the greater Rangpur region in 2006 to support ultra-poor households through both year-round and seasonal microcredit/microfinance services and other monetary (e.g. cash for work in the earlier period) and non-monetary (e.g. training and health services) instruments. The program started in 2006 in Lalmonirhat and later expanded to all other districts of the Greater Rangpur region (namely Gaibandga, Kurigram, Nilphamari and Rangpur). This report attempted to identify the impacts of the program on different markers of household welfare such as consumption ordering, food and non-expenditure, income, asset accumulation, self-employment etc. Findings One of the big challenges of the evaluation team was to identify a proper comparison group (control group). We used a sample of households that reported not participating with any NGO/MFI programs in any one of the three rounds of surveys spanning over 2008 through 2010 calendar years. This report used these non-participants as a “benchmark” group and compared many of the outcomes in comparison to this group. One should note that this group was not a proper control group because this non-participating group differed from the new participants significantly in many household characteristics. E.g. non-participating households were more likely to be in chars and female headed. Head of the benchmark households were more likely to older, unmarried and have less education. However, it was interesting to find that these households did not differ statistically for many of the outcome parameters. We looked at many of the outcome variables over years of households’ participation in PRIME. We found that food security among the sampled households improved on average over the three years of survey as typified by consumption ordering (i.e. having at least three meals per day). During the monga season in 2008, 58% of the households in our sample reported having at least three meals while this reached 75% during the similar season in 2010. This overall secular trend among all households (benchmark and PRIME participants as well households receiving participating outside PRIME program) actually made it chellenging to associate the improvement in food security with PRIME participation. If we looked over the years of PRIME participation, we found that the new participants reported a very similar level of food security (44% of the households in their first year of PRIME benefit having three meals a day) in comparison to the households that did not receive any support from any Page viii PRIME 3rd Round Report. Please do not cite or circulate. MFI/NGOs (44% having at least three meals). However, the consumption ordering increased over the age of PRIME participation and after four years or more in participation 60% of the households reported having at least three meals during the last monga. Other outcome variables also exhibited a similar pattern. It was interesting to find that most of the outcome variables were very similar (in level) between the “benchmark” households and the new participants (the households with less than one year of PRIME participation). Only food and nonfood expenditure, total savings and months engaged in non-agricultural self-employment exhibited differences that were statistically significant. It is possible to infer that PRIME may have immediate impacts on expenditure and savings before having significant impacts on assets, income and more importantly on overall food security (as we found above). We found that the households reported having 21% more land with three years of participation and 33% more with participating four years or more in compared with the “benchmark” households. The total number of livestock is also 15% higher for the households with three years and 23% higher with four or more years of participation. We also found similar pattern when we looked at different types of livestock and poultry. This suggested the households who participated into the program managed to accumulate assets over time and the value of their assets also increased over time. We found a similar “growth curve” over the years of PRIME participations. Total income for the households who are new to PRIME had 18% higher income compared with the “benchmark” households. As for households with three years of participation the income was 24% higher while it was 35% higher for households with four years or more of PRIME participations. Total income livestock was significantly high only for the households with three years (49% higher) and four years or more (66% higher) of PRIME participations. The food and non-food expenditure were also higher for the PRIME participants and this also exhibited an upward trend over the years of PRIME participations. Households that just entered the program showed an immediate jump in both types of expenditures. For the households with different length of participations until the third years showed a somewhat flat growth curve (around 17-18% in comparison with the “benchmark” group) which climbed to 29% for the households with four years of participation. We also found a similar pattern for the first three years of participation (the positive differences with the “benchmark” households were around 33-36%) while it went up to 58% for the households with four years or more of PRIME participation. As expected, self employment in both agriculture and non-agricultural sectors also increased over the age of participation. There was some evidence for improvement in better use of health care use Page ix PRIME 3rd Round Report. Please do not cite or circulate. as households showed lesser tendency to avail health services from the informal providers. The decision to send children or participating in family and/or community level did not vary over the age of the program participation as such we did not find any significant association with the time households spent with the program. We also found that total value of housing was higher for participants compared with the benchmark households and PRIME participants continued to participate in other non-PRIME MFI programs. We found no significant association between PRIME participation and migration of any household member. Concluding Remarks All these findings suggested that it was important to look at households over a period of time after they participated into ultra-poor programs such as PRIME because it might take a while for the ultrapoor households to rip benefits from such programs. This report concluded that inclusion of ultrapoor households to financial system (as offered by microfinance industry) along with training and perhaps monitoring could put such households in a path of development and improvement of their economic situations. We also explicitly addressed the cost of delivery and operational sustainability of the MFI branches implementing PRIME. Here, the picture showed a less optimistic situation. The cost of delivery of financial services to the hard-to-reach households was higher and the branches established through PRIME program had smaller loan size per borrower, smaller portfolio managed by the field officer and higher operating cost in comparison to the branches that were in microfinance operations before the PRIME was initiated. As such the branches did not show the capacity to be operationally sustainable even after three years or more into operations. As expected so-called “double bottomline” was especially stringent for these branches and reaching out to ultra-poor households in the remote areas in a sustainable manner would probably require innovations in financial service delivery mechanism. Page x PRIME 3rd Round Report. Please do not cite or circulate. Acknowledgement First, I would like to thank the members of my team without whose hard work this report would never be finished in time. I would specially like to thank Mehadi Hasan for looking over the household survey and successfully carrying it out in a timely manner. I would like to acknowledge the effort of Monirul Hasan for looking after the branch survey and also the GIS data collection. The research assistance from Tunazzina Chowdhury Mithun and Aparna Howalader has also been an invaluable asset throughout the life of this project. The research team also benefited immensely consulting with a number of individuals at the PKSF. Mr. Md. Golam Touhid (General Manager, Operations), Mr. A.Q.M. Golam Mawla (Deupty General Manager, Operations), Dr. Sharif Ahmed Chowdhury (Project Coordinator, PRIME) and Mr. A. K. M. Zahirul Haque (Deputy Manager, Operations) at PKSF has always been available to provide the research team with information and data on PRIME and with useful insights into the programs that has helped the research team to understand more about the program. The research team would also like to thank Devnet Ltd. (especially Mr. Abdullah Sumon) for efficiently designing the questionnaire in ICR format, capturing the data and ensuring completion of the project on time. The research team would also like to thank Dr. M. A. Baqui Khalily (Professor, Department of Finance and Banking, University of Dhaka), Dr. Meherun Ahmed (Assistant Professor, Asian University for Women), Dr. Tanweer Hasan (Professor of Finance, Roosevelt University) and especially Dr. Rashid Faruqee (Visiting Scholar, Institute of Microfinance) for helpful comments on many aspects of the project. The current version of the report also reflects some of the comments and suggestions the research team received during a seminar arranged by the InM. Lastly, the research team would specially like to thank the Institute of Microfinance (InM) for their continual support for carrying out this project. Special credit goes to Dr. Mosleh Uddin Sadeque (Interim Executive Director), Dr. M. A. Latif (Director, Research and Knowledge Management), Mr. Touhid Uz Zaman (Senior Deputy Director, Research & Knowledge Management), Mr. Md. Abdul Hye Mridha (Senior Deputy Director, Training & Administration) and Mr. Azahar Ali (Senior Assistant Director, Accounts) who have extended assistance and helping hand whenever the research team needed. Their support and InM’s hospitality are greatly acknowledged. Atonu Rabbani Team Leader PRIME Impact Evaluation (3rd Round) Page xi PRIME 3rd Round Report. Please do not cite or circulate. 1. Introduction 1.1 Background Palli Karma Sahayak Foundation (PKSF) initiated the project entitled Programmed Initiatives for Monga Eradication (PRIME) in the greater Rangpur region in 2006. Initially it was introduced in five upazillas (sub-districts) of Lalmonirhat and later PKSF expanded the program expanded to cover all five districts of the current Rangpur division (Gaibandha, Kurigram, Nilphamari and Rangpur being the other four districts). Monga is a situation of income deprivation resulting Figure 1: Map of Bangladesh showing Rangpur division in red. from lack of working opportunity during the pre-harvest situation during the months of September through November. Such lack of work opportunity (many a times aggravated by floods) in turn results in lack of food insecurity. As such, the main objective of the program was to support ultra-poor households through both yearround and seasonal microcredit/microfinance and other monetary (e.g. cash for work in the earlier period) and non-monetary (e.g. training and health services) instruments. As such, in 2007 and 2008 all 35 upazillas of the greater Rangpur region (in the aforementioned five districts) were covered under the program to reach an initial estimate of nine hundred thousand households living under both acute and Source: Internet. chronic state of deprivations – food, income or otherwise. One should note the acute underemployment and lack of work opportunities during the pre-harvest period pose the poor households in this region in an especially vulnerable situation which lead to severe seasonal food insecurity. This is traditionally known as monga. Having some of the major rivers (the Brahmaputra) and their feeders (e.g. Tista) also put the ultra poor households in an ecologically vulnerable state. This accentuates the severity of poverty of the households living in Page 1 PRIME 3rd Round Report. Please do not cite or circulate. chars and low-lying river banks because of floods. Dependence on agriculture, lack of industrialization and regular havocs caused by floods have caused immense hardship for these households which have drawn attention of policy-makers, development practitioners and researchers for a long time. The programs under PRIME epitomize the concerns of these various groups and resulted in a very comprehensive coverage of the ultra-poor households to mitigate the impact of monga and alleviate the economic hardship in the short to midterm while eradicating poverty in this region in the long run. Institute of Microfinance (InM) has been conducting the evaluation of these programs since 2008 following a sample of households comprising all five districts. The households were chosen in a stratified random fashion in a hierarchal fashion (see Section 2). Currently, InM in collaboration with PKSF is implementing third round of evaluation phases. This gives a unique opportunity to follow a large number of households over a period of three years and draw conclusions on impact of the program in a rigorous and sophisticated way. While household survey comprises of the largest component of the evaluation exercise, the PRIME evaluation team has also collected various supplementary information (e.g. village information, branch information and as a pilot study, GIS information for a subset of households in the same during the latest round of evaluation). This report was primary a result of the third phase of the evaluation to document the evolution of the program over time and identify the impacts of the programs where possible. The project team followed up the previous studies through a longitudinal dataset of about 6,500 households. The sample was drawn in such a fashion that researchers could address a number of issues such as selection, participation as well as identifying the impact of the programs on a number of parameters. Many of the outcomes continued from the previous studies to understand the dynamics of the impacts and their sustenance at the household level. Additionally, the research team also looked at some of the additional issues (as required by the Terms of Reference, see Box 1) dealing with (a) the operational sustainability of the program from the suppliers’ point-of-view and (b) some gender specific decision making outcomes at the household level. Page 2 PRIME 3rd Round Report. Please do not cite or circulate. Hence, the bulk of this report would address the Box 1: Objectives of the Study according to the Terms of References. The main objectives of the third round impact study will be to: a) assess the impact of the project interventions on the life and livelihood of the monga affected households, in both the social and economic terms; b) compare the current year (3rd round) impact of the project intervention with first round and second round impact. c) assess the impact on institutional capacity strengthening initiatives for Partner Organizations. d) assess the operational sustainability of prime branches and suggest possible way to attain operational sustainability. outcomes included in the PRIME evaluation reports of the previous rounds. Additionally, we would focus on some of the outcomes related with health seeking behaviors (as we will show PRIME included so called “credit plus” services in recent time) and also some decision making outcomes for women among PRIME beneficiary households. A major innovation of the present phase of the report is to address the issue of sustainability of the program in operational sense. PRIME is implemented through 16 partner organizations (POs) of PKSF (for a list of all POs responsible for implementing PRIME see Table 17 in the appendix). Currently, these POs have 231 branches in the greater Rangpur region implementing the program.1 An important dichotomy between these branches was some branches were established to implement PRIME and these branches were not likely to be in existence if PRIME did not induce establishment of these branches. Other branches implementing PRIME were already in operations and chosen primarily because they were operating in remote areas where many of the potential PRIME beneficiaries lived. The research team at InM collected financial information for all these branches and addressed the issue of productivity, efficiency and Figure 2: Loan Outstanding and Total Members, 2009-2010. 900 800 700 600 500 400 300 200 100 0 1 2 3 4 1 2009 2 3 4 2010 Flexible Loan Outstanding (Million Taka) Total Members (Thousand) operational sustainability of these branches in a very Source: InM Research Team’s Calculations using PKSF Administrative Data from PRIME Cell. comprehensive way. 1 As PRIME has evolved, so have operations of the programs. After the introduction of the program in all districts of Rangpur there were initially 235 branches. As per requirements, some the branches have merged and sometimes infused with other branches of the same PO. By design, only one PO was allowed to implement PRIME in one union. The same PO and other POs (whether or implementing PRIME) often had other ultra-poor programs in the same area (i.e. union). Page 3 PRIME 3rd Round Report. 1.2 Please do not cite or circulate. Evolution of the Program To facilitate the support for the ultra-poor in the Rangpur region PKSF initiated PRIME2 which was later supported by Department for International Development (DFID) under Promoting Financial Services for Poverty Reduction (PROSPER) from July, 2007. The program design has changed somewhat over the years while retaining many of the core aspects of the initial intervention at place. Microfinance and support through both microcredit (typically associated with some income generating activity or IGA) and mobilizing savings from the poor households has been the crux of the program all along. There were also multiple types of microcredit products covering both flexible and emergency in nature, albeit emergency loans constituting a small fraction of total loan disbursed. There were also some ancillary non-credit financial services such as remittance services offered to the PRIME beneficiaries. Figure 3: Average Loan Outstanding Per Borrower. 2,700 Figure 4: Disbursement of Microcredit under PRIME Program. 600 90 80 500 2,500 70 400 2,300 60 50 300 40 200 2,100 30 20 100 10 1,900 0 0 1 1,700 1 2 3 2009 4 1 2 3 4 2010 2 3 2009 4 1 2 3 4 2010 Total Disbursement (Million Taka) [Left Axis] Loan Outstanding per borrower (Taka) Total Number of Loans Disbursed (Thousand) [Right Axis] Source: InM Research Team’s Calculations using PKSF Administrative Data from PRIME Cell. Source: InM Research Team’s Calculations using PKSF Administrative Data from PRIME Cell. Initially there were more direct transfers through temporary wage employment or more traditional Cash for Work (CFW) type interventions. This was implemented in a substantial scale and was later discontinued because it overlapped with similar programs in the region (generally offered by the government).3 Member POs implementing PRIME kept records on the IGAs for which microcredit 2 Concurrently, PKSF also started another program called Learning Innovative Fund to Test New Ideas (LIFT) in the same area. 3 The discontinuation of CFW in the initial period was mostly caused by prevalence of similar programs initiated and implemented in the same regions by the Government of Bangladesh. Page 4 PRIME 3rd Round Report. Please do not cite or circulate. was provided.4 This was also important because PRIME program included training for the beneficiaries. Usually MFI branches arranged these training sessions through hiring government agricultural extension officers. In recent time MFI practitioners are focusing more and more on credit plus products to incorporate health and education related services to the beneficiaries. PRIME presupposes these roles of MFI/NGO by incorporating health services to the poor beneficiaries. PRIME is currently providing primary health care services to the beneficiary households. Such services include both consultation and medicines. Figure 5: Savings Mobilized under PRIME. Figure 6: Income Generating Activities (IGAs) associated with Microcredit under PRIME Program. 20 300 19 250 18 200 Thousand Million Taka 17 16 15 14 13 150 100 50 12 0 11 3 4 1 2 3 4 1 2 3 4 10 1 2 3 4 1 2 3 4 2008 Agricultural 2009 2010 Source: InM Research Team’s Calculations using PKSF Administrative Data from PRIME Cell. 1.3 2009 2010 Livestock Related Non-farm Source: InM Research Team’s Calculations using PKSF Administrative Data from PRIME Cell. Outreach of PRIME In this section, we will focus more on the aggregate outreach of the program and quantify some of the dynamic aspects of the outreach of the program under PRIME. As mentioned before PRIME is currently implemented in all 5 districts of Rangpur. As of December, 2010, 329.3 thousand members 4 This is clearly evident from the MIS data both PKSF and POs shared with the research team. The MIS data included very detailed IGA criteria. While fungibility of money makes it difficult to relate a use with the availability of credit within the household, the previous analysis using household data generally found that the bulk of the credit is diverted to many income generating activities. Page 5 PRIME 3rd Round Report. Please do not cite or circulate. were brought under PRIME and cumulatively a total of 829.2 million taka in (flexible) loan was outstanding by the end of 2010 calendar year (see Error! Reference source not found.). The maturity of the program is also visible in the size of loan outstanding per borrower as reported in the administrative data (see Error! Reference source not found.). In the beginning of 2009 (in the first quarter) the average outstanding loan size was Tk. 1,813 per borrower. The loan size reached Tk. 2,518 per borrower by the end of 2010 (about 40% growth over the same time). The increase in the average loan size was partly possible because the disbursement outpaced the total number of disbursement which one could see in Error! Reference source not found.. The total number of loan disbursed did not show any visible trend over these two years (2009 and 2010) while the total disbursed amount in aggregate exhibited an upward trend over the same period. One should note that there was a clearly seasonal pattern visible in the loan disbursement pattern (though there was an upward trend in the amount disbursed). This suggested a complex interplay between employment, agricultural productivity and credit demand stemming from typical seasonality. As of December, 2010 a total of Tk. 492.4 million was disbursed to about 67 thousand beneficiaries (see Error! Reference source not found.). The PRIME program had also been able to mobilize savings from the ultra-poor households (see Error! Reference source not found.). In the beginning of 2009 calendar year, a total Tk. 11.2 million was mobilized under the program. By the end of 2010 calendar year, this number reached Tk. 18.6 million (a 65% increase over the two years, reaching a total number of about 329 thousand beneficiaries in the program area). While it was not shown here, the average savings stayed surprisingly flat over this period. From the official data it was also possible to understand and follow the trends in income generating activities associated with microcredit disbursement toward the program beneficiaries. The official figures actually revealed that the IGAs pursued by the program beneficiaries could be classified into three broad categories. The agriculture related IGAs had been one of the major IGAs and as of December, 2010, a total of 134 thousand agriculture related IGAs were undertaken cumulatively by the program beneficiaries (see Figure 6). However, it always remained smaller compared with the other broad IGA categories. Livestock related IGAs (second broad category) dominated in the recent time and by the end of 2010 calendar year, 268 thousand livestock related IGAs were associated with microcredit borrowing under the program (see Figure 6 depicting a 22% growth over the entire two and half year period). Non-farm related IGAs (mostly for small businesses) were the third and most frequent categories Page 6 PRIME 3rd Round Report. Please do not cite or circulate. under which the credits were taken by the borrowers. However, this category experienced the lowest level of average monthly growth over the entire period. By the end of 2010 calendar year, a total number of about 223 thousand IGAs were associated with non-farm income activities. Table 1 showed the some of the major sub-categories of IGAs within all three broad categories. Within agriculture related IGAs, crop production dominated with about 97 thousand IGAs by the end of 2010. This was followed by homestead gardening cumulatively which stood at about 19 thousand IGAs at the end of 2010. Among livestock category, goat rearing garnered the most dominant IGA category with 77 thousand times with dairy cow and heifer rearing securing second and third most prominent IGAs under the livestock category with frequencies of about 68 thousand and 60 thousand frequencies. As we mentioned earlier small business dominated under non-farm related IGAs with rickshaw and van purchase (and maintenance) as the next most important category with a cumulative number 54 thousand times such IGA taken by the end of 2010. Table 1: Income Generating Activities (Thousand IGAs) by Major SubCategories. 2008 2009 2010 Agriculture Related IGAs of which: Crop Production (as per crop calendar) Homestead Gardening 38.5 81.2 134.4 23.9 10.0 57.6 14.0 96.7 18.9 Livestock Related IGAs of which: Goat rearing Dairy cow rearing Heifer rearing 54.8 158.2 268.3 26.7 10.5 6.8 53.6 39.7 29.2 77.0 68.4 59.6 Non-Farm Related IGAs of which: Small business Rickshaw/Van purchase/maintenance 82.1 151.8 222.9 51.2 21.1 95.1 38.4 137.2 54.2 Source: InM Research Team’s Calculations using PKSF Administrative Data from PRIME Cell. Note: Numbers are in thousands There were number of other services that PRIME offered to its beneficiaries. It provided remittance services (i.e. helping the beneficiary households receiving money transfer mostly from domestic sources where the income earning member of the household is living). Over the life of the program by the end of 2010 calendar year, the program assisted 8,334 times with such financial transactions. Page 7 PRIME 3rd Round Report. Please do not cite or circulate. Figure 7: Number of Remittance Services Figure 8: Number of Health Services (in Rendered under PRIME. Thousands) Rendered under PRIME. 9,000 1,200 8,000 1,000 7,000 6,000 800 5,000 600 4,000 3,000 400 2,000 200 1,000 0 0 3 4 1 2008 2 3 4 2009 1 2 3 4 3 2010 Source: InM Research Team’s Calculations using PKSF Administrative Data from PRIME Cell. 4 2008 1 2 3 2009 4 1 2 3 4 2010 Source: InM Research Team’s Calculations using PKSF Administrative Data from PRIME Cell. Health services had also become an important and integral part of the PRIME program in the recent year and we could also track the outreach of health services from the official data. Such data revealed that by the end of 2010 calendar year, 1.1 million units of health services (defined by primary care visits and consultations and drug dispensing in the recent time) were disbursed to the beneficiary households in the program areas. This particular component also experienced an unprecedented growth rate of more than 50% every month over the two and half years ending in 2010 calendar year. 1.4 Overview of this Report We addressed the impacts of the program on the household and focused on selected outcome variables most pertinent to the program and as guided by the Terms of References. As we mentioned earlier, we also focused on the supply-side factors such as productivity, efficiency and sustainability of the branches implementing PRIME. As such we would address the issue of sustainability with PRIME branch implementing PRIME as units of analysis.5 In Section 2, we described the data we would be using to identify the impacts of the program on selected household outcome variables. As one would see, we exploited the panel structure of the data set as much as possible to find different synthetic cohorts which differ by the intensity of 5 It is commonly mentioned in the literature that the MFIs face the double bottom-lines as they try to reach to poor households and provide them with financial services under the constraint that the program itself has to be self-sustaining in the operational sense. This is really important because this constraint will especially be binding for programs that try to reach to the ultra-poor households living in remote areas suggesting higher cost and lower productivity. Page 8 PRIME 3rd Round Report. Please do not cite or circulate. association of PRIME. The intensity was defined by the age of participation into the program regardless whether the household was a continuing member of the program or not. As our findings revealed, selection was an issue as the non-participants (precluding the households who were participating in the other NGO/MFI programs) were different from the PRIME participants (see Section 2.3). However, later we showed that the outcome variables were more similar on average (and these means were not different in the statistical sense) between the “benchmark” group (defined in Section 2) and households that were member into the program for less than one year (see Section 3.3). This suggested that the determinants of the selection worked more like natural barriers to participate into the program albeit being similar in many economic outcomes. In Section 3, we provided evidence on the effectiveness of the program on a number of parameters. We first looked at consumption ordering. It was evident from the data (by comparing all three rounds) that the monga situation in the greater Rangpur regions improved over last three years (i.e. between 2008 and 2010). However, we found a systematic association between age of PRIME participation and the likelihood of reporting a better food situation. This trend was later found to be present in many other household characteristics. In Section 4, we provided an overall assessment of the program bringing in both the impact we identified in the previous section and the sustainability issue which we addressed separately. The program was intended for hard-to-reach ultra poor households in the char areas. As such the branches exhibited lower productivity and were not able to become operationally self-sufficient after three years in operations. As such positive impacts of the program were especially important to justify the program. But it was important to realize that positive impacts of such program might take a long time to surface (two years or more). We mentioned few limitations of the present study in the conclusion. In the appendix, we analyzed branch level data we collected to address the issue of sustainability and other productivity parameters. We found that the branches implementing these programs are yet to reach a sustainable level of operations. The branches implementing PRIME program were less productive as their loan outstanding per borrower or borrowers per credit officer were lower than what it was required for a branch to recoup its operating cost. While the social returns from reaching out to hard-to-reach ultra-poor households in the char and other remote places probably justified continuation of the program, to sustain the program on its own, it would require innovative practices to pull down costs and be more efficient. Page 9 PRIME 3rd Round Report. Please do not cite or circulate. 2. Methods 2.1 Sampling PRIME extended in roughly three stages. In the first stage (during 2006), PRIME was implemented in 5 upazillas of Lalmonirhat district. In 2007-08, the program was extended to selected 18 upazillas of other 4 districts in the greater Rangpur region (namely Gaibandha, Kurigram, Nilphamari and Rangpur). First impact evaluation of PRIME program used census data6 of 482,984 poor households in these 23 upazillas as its population of program upazillas. The census covered a total of 23 upazillas, 209 unions, 2,531 villages and 482,948 households. Later, in the first phase of impact evaluation study, the research team at InM targeted a total of 5,240 households for a household survey from all five districts which consist of 16 upazillas, 61 unions and 271 villages (using multistage cluster sampling technique). A total of 5,308 households were surveyed eventually and the first evaluation report was based on this sample (including information from the baseline census where available). As the program was later extended in the other 12 upazillas (the third phase of the extension hence the program came to cover all 35 upazillas of the 5 districts of greater Rangpur region), the second phase of the impact evaluation targeted the households covered in the first survey along with additional about 2,500 households in the new area covered under PRIME. The second phase of the evaluation used information for all these households (a total of 7,212 out of 7,432 households included in the household survey) and assessed the impact of the program using a much longer period of data. One of the major findings was that the identification of the impact actually needed a longer period (usually two years or more) for the effects of the program to surface. Table 2: Sampling Phase 1 Phase 2 Phase 3 5 20 71 271 27 298 4,606 702 5,308 5 27 108 362 39 401 6,552 880 7,432 5 26 102 340 6,988 District Upazilla Union Program Village Control Village Total village Household in program village Household in Non-program village Total number of household Source: PRIME Household Surveys, Various Rounds. 6 The census data was primarily collected in order to identify and select households for program delivery. The information gathered for this purpose was later linked with the later rounds to assess the program in the first and second rounds. Page 10 PRIME 3rd Round Report. Please do not cite or circulate. By the year 2009, PKSF has extended the PRIME program to all 35 upazillas in 5 districts. So in third round of program evaluation study, there was no non-program village. In 2009 total number of villages was 401. Number of households in some villages was very small. Third phase of the impact evaluation have excluded those villages from the sample where number of households was less than 9. Using this exclusion criteria total number of village selected for survey was 340. Total number of households in these villages was 6,988. Details of sample size in all phases are summarized in Table 2. Field survey for the third phase started on December 1, 2010 and ended on January 15, 2011. The research team trained 50 field enumerators and 5 supervisors one week of extensive training on household questionnaire. From the list of 6,988 households, enumerators could reach for 6,830 households during the survey time. The field enumerators could not reach rest of the households because of permanent or temporarily migration or incorrect addresses. Rate of attrition was much smaller compared to any other years of impact evaluation. There were two possible explanations for high rate of reaching the households in the list by the enumerators: (i) last survey was conducted in February-March 2010. So the event of migration (or perhaps deaths) of household member was supposed to occur less frequently in these eight-nine months; and (ii) this year each enumerator was paid based on number of questionnaire completed properly. So the incentive to find the household was high for the enumerators. 2.2 Data This section describes some characteristics of all households in the sample. We have already mentioned in Section 2.1 that the sample was drawn from a group of households that were eligible to participate in PRIME program. The program was initiated targeting the ultra poor households of that region. So the summary statistics of the household characteristics in this section supposed to reflect the characteristics of ultra poor households. Table 3 showed that our sample households spent about 80 percent of their expenditure on food.7 This indicated that our sample households were mostly from the poorer group of the society. Most of the household head’s primary occupation was daily wage earning. Income from this occupation was more prone to shocks because of its relation with seasonality.8 Daily wage worker in agriculture also depend on the seasonality of different crop. In our sample household, those who are engaged in daily wage worker in agriculture, average working day in Bangla months of Vadro, Ashwin and Katrik 7 It is common to observe that the proportion of expenditure on food to total household expenditure falls as income of the household increases. This outcome is usually robust, so researchers often use the share of expenditure on food as the measure of poverty. For example, household that spend more than 60 percent of their expenditure on food might be considered as poor (see Haughton & Khandker, 2009). 8 For example, opportunity to work in brick-field as daily wage worker stopped in rainy season. Page 11 PRIME 3rd Round Report. Please do not cite or circulate. Table 3: Summary Statistics for Selected Variables from Third Phase of Household Survey. Mean Standard Deviation 12.4% 43.8 1.5 33% 13 2.6 Maximum education in a Household (years) 4.2 3.1 Family size 4.01 Sex Ratio (Number of female/Number of male) 1.2 1.5 0.9 Electricity access 8.8% 28.4% Access to tube-well or tape water 98.8% 10.7% Access to sanitary latrine 52.8% 49.9% 53.9% 49.9% 8.6% 28.1% 25.4% 43.6% 22.9% 42.0% 15.4% 36.2% Total land (decimal) 11.6 47.7 Total livestock (number) 4.5 5.8 Total number of cow (number) 0.6 1.1 Total number of goat (number) 0.3 0.8 Total number of poultry (number) 3.6 5.3 Total asset value including land (Taka) 1,09,330 2,47,608 Total asset value excluding land (Taka) 31,852 41,684 Savings (Taka) 2,459 10,191 Total Income (Taka) 49,104 1,14,294 Total Income from livestock (Taka) 1,513 5,686 Expenditure on food (Taka) 35,542 15,133 Nonfood expenditure (Taka) 8,752 12,991 Household head characteristics Household head is female Household head’s age (years) Household head’s education level (years) Family characteristics Infrastructure facilities Occupation and source of income Primary occupation of household head wage worker Primary occupation of household head self employment in agriculture Primary occupation of household head self employment in non agriculture Any of the household member migrate Household head migrate Physical and financial asset Household income (yearly) Household expenditure (yearly) Source: Household survey. The total number of observations is 6,830. Page 12 PRIME 3rd Round Report. Please do not cite or circulate. is about 10 days. Average working days in these three months were much lower than the average working days of others month in the year. So, the households that depended on the income from daily wage earning in agriculture faced seasonal income negative “shock” that affected consumption adversely in those months. Percentage of household heads involved in self-employed agriculture activities was very low because of little ownership of cultivable land. Average land holding of the household was very marginal, about 24 percent of the sample households are landless. Because of the unemployment in the lean season, some household members migrated to other districts of the country. About 23 percent households in our sample experienced the migration of at least one member in the last year, and in most of the cases household head was the one who migrated. Percentage of household having access to hygienic drinking water was high. Use of hygienic toilet was around the projected national average level but access to electricity was found much below the national average level. Another interesting feature was that sex ratio was higher than one, which meant that number of female was higher than male for most of the households. In our sample area wage rate of female was much lower than wage rate of male, so higher value of sex ratio actually indicated the lower earnings of the households. 2.3 Identification It was a big challenge to identify the impact of a particular program using non-random intervention (even randomization is not sufficient when the impacts take long time to surface). It was important to address the eligibility criteria of the program first. There were some selection criteria to identify who were eligible to participate in program. To participate in the PRIME program the household had to fulfill at least one of the three conditions: (i) Monthly income of household less than 1,500 Tk; (ii) Primary occupation of household head was daily wage earning; (iii) Land ownership was less than 50 decimal. However, meeting eligibility criteria themselves did not guarantee participation because from the households’ point of view it is voluntary. Field officer and households (or beneficiaries) jointly determined an outcome entailing whether participation took place or not. One restriction regarding the program participation was that households having intervention from programs offered by any NGO-MFIs (including PRIME POs) were not eligible to participate in the program. Still we found that there was a group of households who never participated in any sort of program initiated by any MFI/NGOs during the period covered under all three rounds of household surveys (i.e. 20082010 calendar years). We used these non-participants as a “benchmark” group.9 One should understand that even this group could not be considered a proper control group because their 9 We have systematically omitted the households that received non-PRIME MFI services during our period of study as suggested during any of the three rounds of studies. As such we have limited our analysis to 2,948 “benchmark” households and 1,353 PRIME participants only. Page 13 PRIME 3rd Round Report. Please do not cite or circulate. baseline characteristics were observationally different from the new participations (i.e. households who received the “treatment” within last one year, see Table 4). Table 4: Summary Statistics of Selected Household and Village Characteristics by Participation Status. Village characteristics Household in char Household head characteristics Female headed household Age of household head (years) Education of household head (years) Household head currently married Family characteristics Maximum education (years) Family size (number) Occupation and income source Household head migrate Household head's primary occupation wage worker HH head's primary occupation self-employed agriculture HH head's primary occupation self-employed non-agriculture Number of income source Infrastructural facilities Household have access to electricity Household drink from tube-well/tap Household use sanitary latrine Outcome variables Total income Total land Total livestock Source: Household survey. PRIME participant (Age of participation 0 to 1 year) N=535 Nonparticipant N=2,948 p-value 10.47% 22.25% < 0.01 6.54% 40.63 1.76 93.08% 18.66% 45.49 1.32 79.88% < 0.01 < 0.01 < 0.01 < 0.01 4.52 4.14 3.67 3.68 < 0.01 < 0.01 15.70% 56.45% 8.79% 26.54% 1.83 16.08% 57.60% 8.24% 17.61% 1.66 = 0.41 = 0.31 = 0.34 < 0.01 < 0.01 10.84% 98.32% 55.89% 5.87% 99.05% 47.66% < 0.01 = 0.06 < 0.01 46,933 10.37 4.33 45,712 11.65 4.25 = 0.43 = 0.32 = 0.38 Table 4 showed means and percentage comparisons between non-participants and new participants in PRIME. Most of the household characteristics revealed that these two groups differ significantly. For most of the cases these were the determinant variables that were associated with the decision whether a household would participate in any MFI/NGO initiated program or not. For example percentage of non-participant household living in char was higher compared to new participants. In char areas MFI/NGO activity was usually very limited because of inaccessibility of char area in terms of communication, so living in char increased the probability of not participating in the program. Different characteristics of household heads also determined whether a household would Page 14 PRIME 3rd Round Report. Please do not cite or circulate. participate or not. Table 4 shows that female headed households were less likely to participate. In our sample females were reported as household head not because they were taking decision over their male counterparts, rather in most of the cases they were either widow or separated. This was supported by current marital status of non-participant which showed that the percentage of widow or separated group was higher in non-participant group. Having multiple sources of income was important for the households to participate in the microfinance/microcredit programs. If a household participated in program and took a loan then they had to repay the loan in weekly installment. To avoid loan default the household should have smooth income source which was possible if the household diversified and had multiple income sources. Diversified income sources reduced the seasonal income shock which was very much observed in the sample area i.e. in Greater Rangpur Region. So number of different income source in household could determine the participation decision of the family. Table 4 also showed that lower age of household head increase the probability of joining the program. Lower aged people were more capable of generating higher earnings, which in turn increased the probability of participating in program. But the above mentioned criteria were not used to select the eligible group for participating in the program. Already we had mentioned that the key indicators of eligible group selection were income, land ownership and occupation of household head. The objective of the program was that over the year the participants would do better in terms of these indicators or outcome variables. Our assumption was that at the initial stage of participation they would be similar with those households who had never participated in any program in terms of those variables. Table 4 showed that these two groups did not significantly differ in terms of total income, land ownership and livestock. Also the percentage of household where primary occupation of household head was wage worker did not significantly differ between these two groups. These results from Table 4 indicated that these two groups were in same platform. So now we could use this non-participant group as our control group to see whether longer involvement in program had any impact on some selected outcome variable. As we mentioned in Section 2.1, PRIME was first implemented in 5 upazillas of Lalmonirhat district. Eventually all 35 upazillas in 5 districts were brought under coverage. Since selection was an issue (as evident from Table 4), we compared between households who had already been selected into the program. However, to identify the impact of the program we compared the households who had been with the program longer with the households who just started into the program. Thus, the age of participation into the program acted as the main identifier for the impact of the program. One should understand that the main purpose of this evaluation exercise was to estimate the treatment of the treated. The treatment in this case was such that it introduced the households to the Page 15 PRIME 3rd Round Report. Please do not cite or circulate. microfinance interventions. As such PRIME created a snowball effect for the participating household (on average) and established in higher growth curve (in income, asset and status of food security). So looking at PRIME participants with different length of participation into the program would give us the intensity of the impact and this was repeatedly identified throughout this report for different outcome variables. Different households participated in the program at different time. We divided the PRIME participant groups in four categories – households that were member for less than one year, households that were member of two years, households that were member of three years and those who were members for four or more than four years. Table 5 shows the distribution of PRIME membership duration by districts. Table 5: Percentage distribution of PRIME membership duration by district PRIME membership duration District 0-1 2 3 (N=535) (N=333) (N=268) 24.86% 25.23% 20.52% Gaibandha 4+ (N=217) 14.75% Kurigram 19.63% 23.12% 22.39% 18.89% Lalmonirhat 6.73% 17.12% 30.97% 38.71% Nilphmari 23.18% 14.71% 15.67% 11.52% Rangpur 25.61% 19.82% 10.45% 16.13% Source: Household survey. Table 5 showed that most of the members with duration of membership three or more were from Lalmonirhat. This was because PRIME was first implemented in Lalmonirhat district. But percentage of new participant (0-1) gave the indication that the rate of expansion of program in Lalmonirhat in recent year fell. Some basic summary statistics for these groups were presented in Table 6. Table 6: Summary statistics of selected household characteristics by duration of PRIME membership (in years) Household has electricity Household has access to tube-well/tap water Household has sanitary toilet Household head is female Household head’s age: years Household head’s education: years Household size Primary occupation of household head is wage worker Source: Household survey. 0-1 (N=535) 10.8% PRIME membership duration 2 3 (N=333) (N=268) 8.1% 7.8% 4+ (N=217) 13.8% 98.3% 99.1% 98.9% 99.1% 55.9% 6.5% 40.6 1.8 4.1 56.5% 9.0% 41.7 1.5 4.4 53.7% 9.3% 43.1 1.4 4.3 68.2% 6.9% 44.3 1.3 4.4 56.5% 53.8% 55.6% 48.9% Page 16 PRIME 3rd Round Report. Please do not cite or circulate. Table 6 showed that these four groups were very much similar in different household specific characteristics. Only in terms of household using sanitary latrine, the households in the earlier cohort of participants came up with good position (which itself might be result of the program). Also, they were in comparatively better position in terms of access to electricity. These variables were kind of outcome variable and this difference in percentage could be treated as the impact of longer involvement with the program. Table 6 also showed that percentage of household whose primary occupation was daily wage worker is less among the eldest group. It had been evident that daily wage worker were the most vulnerable group in this region because of irregularity in income in lean season. So the households that depended on daily wage earning were subject to consumption shock in lean season. PRIME encouraged their borrowers to use their loan in different income generating activities (IGA) like buying a rickshaw, or raising livestock, or meet the agriculture related expenses. PRIME also provide training on different IGA for their members. The idea was that promoting IGA might help them to move away from daily wage worker as their primary occupation to some selfsustaining activities. Table 6 showed that longer involvement in program could help them to change their primary occupation status. 2.4 Notes on Analytical Framework As evident from the previous section the analytical underpinning of this evaluation exercise relied on identifying the “treatment-on-the-treated”.10 There are two different ways we intended to do it in this report. If we denote a particular outcome variable as , where denotes a household and denotes receiving in the treatment (e.g. participating in PRIME); = 0 means non-participation. We intend to identify ( | = 1 means participation while = 1) − (( | = 1) which is defined as treatment on the treated. Unfortunately we do observe outcome for the same household in the absence of the treatment ( ). The source of bias stems from selection as participation depends on unobservables and they can potentially confound the outcome variables. As such we used the variation in the length of participation because the selection bias would be smaller for this group. We showed that the households not participating in any MFIs are different based on observed characteristics (hence likely based on unobservables as well). However, the differences in observed outcomes (which we expected would be affected by PRIME participation) are many times not significant. Hence, we assumed that while there were natural barriers for households to participate in PRIME they were very similar in terms household welfare (indicated by income and assets) that PRIME intervention intended to alter. Hence, comparing households with different 10 Please see Angrist, J. D., & Pischke, J.-S. (2008). Mostly Harmless Econometrics: An Empiricist's Companion. Princeton, NJ: Princeton University Pres. Page 17 PRIME 3rd Round Report. Please do not cite or circulate. duration in PRIME (excluding the households participating in other MFIs exclusively) and also with the “benchmark” households (i.e. the ones not participating in any NGO/MFI programs over last three years) would identify the treatment-on-the-treated we intended to identify. 2.5 Discussion and Concluding Remarks on Identification The main objective of PRIME program was to provide the ultra-poor households in the monga prone areas with access to finance and other support programs (such as IGA related training and in the recent time, health services). As such, this program brought households into the program who would not have access to microfinance otherwise. This resulted in “snowball” effects for the households that might lead way to a better economic situation. So the year(s) since participating into the program is an important marker to identify the efficacy and effectiveness of the program. If we assumed that at the time of first selection into the program, similar households participated into the program, then the new program participants could be used as a legitimate control group (however synthetic it might be) and we could compare the outcome of these group(s) with the more seasoned PRIME participants to identify the impact on the selected outcome variables. It was important to realize this reasoning would allow us to use households who did not continue to participate into the program. This was because, through providing the initial access to microcredit/microfinance, the PRIME program supposedly revealed that the participating households were capable of continuing in a rural credit program (offered by possibly another microcredit organization). Thus, we identified the total effects of the program through this comparison because later effects would not be possible but for PRIME. This was an important conceptual point that would help us interpreting the results we would turn to in the next section on impact of PRIME on the ultra-poor households. Page 18 PRIME 3rd Round Report. Please do not cite or circulate. 3. Impact of PRIME on Selected Parameters 3.1 Introduction This section would present the impacts on different parameters (as suggested by the Terms of References) that one could empirically associate with PRIME interventions. Using non-random assignment for an intervention it was challenging to identify impacts of a program and associate it with causal implications. On the other hand, as we saw in the previous reports it might take more than two years for impacts of programs such as PRIME to surface. It was operationally impossible to restrict a population (or a sample of population) to be refrained from getting microfinance interventions for three years in a country where penetration of such programs (PRIME or otherwise) had been very high. As we argued in the previous section, the age of the participants could (under certain assumptions) identify the intensity of the program’s impact across different households. A direct corollary of this conjecture was the positive “impacts” of the programs (e.g. higher consumption ordering or households’ asset accumulations) would be associated with earlier a household was associated with PRIME (defined as “age” in this report, see Section 2). In the next couple of sections, we would systematically show association of changes in the selected parameters with age of PRIME for surveyed households starting with consumption ordering. 3.2 Consumption Ordering One of the major problems of identifying impact of a program was to separate out the effect of a program from the secular changes in the outcome variables even if one had control and treatment groups. We saw this in the data where we documented the changes in consumption orderings for all households over all three rounds of household surveys (spanning over about three years of information) for most of the household. Figure 9 showed the changes in consumption ordering as reported during the time of the survey which one could regard as a more “normal” time (as opposed to a period when food deprivation was more pronounced i.e. during the monga time). We found that the incidence of “occasional starvation” was not common situation for the households during any of the rounds and it had become even for uncommon in the last round (i.e. in 2010). The most remarkable change was households reporting having three full meals increased dramatically (by 20 percentage points) between the last two rounds of surveys (i.e. between March, 2010 and December, 2010-January, 2011). It seemed that households previously reporting having three half meals were more likely to Page 19 PRIME 3rd Round Report. Please do not cite or circulate. report consuming three full meals during the “normal” periods (see Figure 9) during the 3rd round of the survey. We perhaps saw a similar change in consumption ordering during monga for the same set of households. There was a 12 percentage points increase in households reporting having three full meals every day and a similar increase for the households reporting a three half meals. The most dramatic change was evident in the “Occasional Starvation” category which experienced a decline of 19 percentage points between the second and the third rounds of households survey. Figure 9: Distribution of the Households over Figure 10: Distribution of the Households over Consumption Ordering during “Normal” Time. Consumption Ordering during Monga Time. 50% 60% 40% % of Households % of Households 50% 40% 30% 20% 30% 20% 10% 10% 0% 0% Occasional Two Three half Three full Starvation meals/day meals/day meals/day Round 1 (N = 5,308) Round 2 (N = 7,090) Round 3 (N = 6,758) Source: InM Research Team’s calculations from Household Surveys (Various Rounds). Occasional Two Three half Three full Starvation meals/day meals/day meals/day Round 1 (N = 5,308) Round 2 (N = 7,090) Round 3 (N = 6,758) Source: InM Research Team’s calculations from Household Surveys (Various Rounds). Another way to look at the same issue would be to look at cumulative distribution over the same consumption ordering so looking at the consumption up to a certain level of food intake (measured by number of meals each day). This also highlighted (see Figure 11) the fact that much of the improvement in number of meals were within the top two levels of consumption ordering while there was a significant decrease in people having less than three meals (full or half) as well. As Figure 12 suggested there was a general improvement in food intake (as measured by consumption ordering) during monga time as well over the three rounds of surveys (i.e. over 2008-10 period). Page 20 PRIME 3rd Round Report. Please do not cite or circulate. Figure 11: Cumulative Distribution of Figure 12: Cumulative Distribution Consumption Ordering during “Normal” Time Consumption Ordering during Monga Time 100% 100% 80% 80% 60% 60% 40% 40% 20% 20% 0% 0% Three full Three half Two Occasional meals/day meals/day meals/day Starvation Three full Three half Two Occasional meals/day meals/day meals/day Starvation At Least At Least Round 1 (N = 5,308) Round 2 (N = 7,090) Round 3 (N = 6,758) Source: InM Research Team’s calculations from Household Surveys (Various Rounds). Round 1 (N = 5,308) of Round 2 (N = 7,090) Round 3 (N = 6,758) Source: InM Research Team’s calculations from Household Surveys (Various Rounds). The analysis of the data using the consumption ordering suggested that it would be important to separate out the impact of PRIME on consumption ordering from the overall trend in the improvement in the everyday’s meal patterns. As we argued in the previous section, the age of participation in the PRIME program would provide us some indications whether the program worked on the targeted households or not. In the next set of figures we showed this. We found that the fraction of households reporting at least three meals per day was higher for households that participated earlier in the PRIME program compared to the households who were in to the programs for less than a year. The households in the latter category reported 74% of the time having at least three meals per day while the households who have been in the program for 4 or more years (mostly the households in the Lalmonirhat district) reported having at least three meals 82% of the time suggesting an increase of about eight percentage points. The “impact” increased with the age of PRIME participation. Page 21 PRIME 3rd Round Report. Please do not cite or circulate. Figure 13: Fraction (%) of Household Reporting Figure 14: Fraction (%) of Household Reporting at least Three Meals/Day during “Normal” at least Three Meals/Day during Monga Time. Time. Normal Time Monga Time 65% 84% 82% 60% 82% 60% 80% 80% 55% 52% 78% 50% 76% 76% 50% 46% 74% 45% 74% 72% 40% 0-1 2 3 0-1 4+ 2 3 4+ Duration of PRIME Participation (in Years) Duration of PRIME Participation (in Years) rd Source: InM Research Team’s estimates using 3 Round of household surveys. The benchmark indicates households who did not participate in any MFI/NGOs over 2008-2010 (N = 2,907). The vertical axis shows the fraction of household (%) reporting having at least three meals/day currently (i.e. at the time of the third round interview). rd Source: InM Research Team’s estimates using 3 Round of household surveys. The benchmark indicates households who did not participate in any MFI/NGOs over 2008-2010 (N = 2,907). The vertical axis shows the fraction of household (%) reporting having at least three meals/day during the last monga. We saw the similar pattern for consumption ordering during the time of the monga. While only 46% of the households reported having at least three meals every day among the PRIME participants with less than a year in the program, 60% reported having at least three meals that had been in the program for 4 to 5 years. This suggested that in comparison with the new participants (households without the influence of the program) the beneficiaries who experienced it earlier enjoyed a fourteen percentage points higher incidences of a better meal comprising of three meals per day. One of the important findings of this report was that we identified a group of households who had been consistently out of the programs offered by any NGO/MFIs. This gave us a cleaner “benchmark” to compare our results with.11 To qualify this claim we should expect to see that the benchmark households should behave similar to the households with less than one year of program participations. Interestingly, both for the “normal” time (basically the time of the survey that the research carried out between December and January) and monga time (the calendar months of September through middle of November), we found this to be the case. For the benchmark cases, 11 One should again be careful to interpret these households belonging to a control group because we found these households to be different from the households participating into the program. See Section 2 for a comparison between many baseline parameters between these two groups. Page 22 PRIME 3rd Round Report. Please do not cite or circulate. 73% of the households reported having at least three meals per day during the “normal” time (as oppose to 74% for the household with less than one year in the program) while 42% reported having at least three meals for the “benchmark” households during the last monga (in comparison to 44% of the households with less than one year in the program) . We further analyzed the data using the additional (marginal) probability of reporting three meals per day in linear probability models using the households without any participation with MFI/NGO over the three years of surveys as the baseline group. For “normal” time (see Figure 15), we found that the benchmark case and the households with less than one year of program participation were indistinguishable in the statistical sense when it came to reporting at least three meals per day. Interestingly, the marginal impact of age or participation (in PRIME) did not become statistically significant even after two years. However, the households with three or more years of participation gained significantly (seven and ten percentage points respectively, see Figure 15). Figure 15: Marginal Impact of the Program over Figure 16: Marginal Impact of the Program over Age of Participation on Reporting Three Age of Participation on Reporting Three Meals/Day during “Normal” Time. Meals/Day during Monga Time. 24% 16% 19% 12% 10% 8% 7% 4% 9% 9% 6% 3% 4% 2% 2% 0% -1% -4% -6% 0-1 16% 14% 2 3 4+ Age of PRIME Participation in Years 0-1 2 3 4+ Age of PRIME Participation in Years Source: InM Research Team’s estimates using household surveys. The benchmark indicates households who did not participate in any MFI/NGOs over 2008-2010 (N = 2,907). The vertical axis shows the coefficients for each category in the horizontal axis. The vertical line for each point shows the 95% confidence intervals. We saw a similar pattern for the case of consumption ordering during monga as well. However, the “marginal” impacts (as identified by the coefficients for each category) were little stronger and statistically significant after two years of participations into the program (as seen in Figure 16). Again, the “benchmark” households and households with less than one year of participation were statistically same when reporting at least three meals per day. However, the positive impacts of the program showed up after two years and the households with four to five years of participations in Page 23 PRIME 3rd Round Report. Please do not cite or circulate. the program would give an additional 16 percentage point higher probability of reporting at least three meals per day. Box 2: A Composite Dynamics of Consumption Ordering during Monga Season, 20082010. 60% 50% 0-1 2 40% 3 4+ 30% Never 20% 10% 2008 2009 2010 Rather than looking at one cross-section, we could potentially look at the outcome variables over last three years. Here we looked at consumption ordering during monga season over last three years (i.e. 2008, 2009 and 2010). We found that the households with four or more years of participation had a more impressive growth trajectory. With only exception of households who participated in PRIME for two years† as reported by the end of 2010, all the households improved their food security during monga over the calendar years of 20082010. It was also interesting to find that the households who did not report participating in any NGO/MFI programs and households who participated less than a year in the program reported very similarly over the study period. One should expect for these households to diverge in the coming years. †At this moment the research team did not have a satisfactory explanation for this anomaly. However, the households with duration of PRIME membership of two years coincide with the expansion of the program to new areas (13 new upazillas in the Greater Rangpur region) and these households were likely to be different from the benchmark and other households receiving benefit under PRIME. The severities of floods were higher in 2009 which might explain the deterioration of food security of these households between 2008 and 2009. However, this justification would require further analysis of the data. 3.3 Other Outcome Parameters and Age of PRIME Participation We used the same methodological scheme to identify the impact of the program for other outcome parameters. While we have found that the “never” participants (the households that did not participate with any NGO/MFIs over the three rounds of households surveys during the period of 2008-2010 calendar years) were observationally different. However, as we claimed (see Section 2) Page 24 PRIME 3rd Round Report. Please do not cite or circulate. these households signified a market with a further barrier to access for the MFIs to overcome and presented with a reasonable comparison group. Table 7: Comparisons between “Benchmark” Households and PRIME Participants with less than One Year in the Program. Land and Non-Land Assets Total Land Holdings (in Decimal) Total Number of Livestock Total Number of Cow(s) Total Number of Goat(s) Total Number of Poultry Asset Values Total Asset Values (Thousand Taka) Total Non-Land Asset (Thousand Taka) Total Savings (Taka) Total Housing Value (Taka) Income Total Income (Taka) Total Income from Livestock (Taka) Expenditure Food Expenditure (Taka) Non-Food Expenditure (Taka) Benchmark Households (Non-Par cipants) (N = 2,907) Households with less than one year in PRIME (N = 533) p-value 11.7 4.2 0.54 0.30 3.4 10.4 4.3 0.46 0.29 3.6 = 0.651 = 0.757 = 0.009 = 0.950 = 0.467 103.59 97.57 = 0.661 27,572.69 30,534.31 = 0.110 1,794.74 11,035.81 2,919.62 13,843.62 = 0.060 < 0.001 45,683.82 46,976.39 = 0.857 1,286.47 978.15 = 0.248 32,250.54 7,018.09 36,517.92 8,781.34 < 0.001 = 0.003 Self-Employment Months Engaged in Non1.93 3.30 < 0.001 Agricultural Self-Employment Days per Month Engaged in 9.48 10.81 = 0.181 Agricultural Self-Employment Source: InM Research Team’s estimates using household surveys. The benchmark indicates households who did not participate in any MFI/NGOs over 2008-2010. For the case of consumption ordering, we found that some of the baseline characteristics were different between the new PRIME participants and the “benchmark” households the program participants. But most of the outcome variables for the “benchmark” households and the households just entered the program with less than one year of participation were surprisingly similar (see Table 7). Most of the outcome parameters (such as asset and income) did not exhibit any between group differences which were statistically significant up to any reasonable levels of Page 25 PRIME 3rd Round Report. Please do not cite or circulate. confidence. Expenditure (both on food and non-food items) were different between the two groups and for both cases the participants, even with less than one of year of participation, exhibited higher spending. Interestingly, PRIME participants also spent higher number of months in non-agricultural self-employment implying incentives the participants received to engage in non-agricultural enterprises. Table 8: Land and Non-Land Assets against Age of PRIME Participation. Benchmark Level* [1] [2] [3] [4] [5] Log(Total Land) Total Number of Livestock Total Number of Cow(s) Total Number of Goat(s) Total Number of Poultry 11.7 4.2 0.54 0.30 3.41 -0.166* (-0.330 - -0.003) -0.067 (-0.264 - 0.130) 0.188+ (-0.014 - 0.390) 0.394** (0.163 - 0.624) -0.008 (-0.202 - 0.186) -0.405** (-0.683 - -0.126) 0.230+ (-0.015 - 0.476) 0.572** (0.307 - 0.837) 0.047 (-0.058 - 0.151) -0.026 (-0.157 - 0.104) 0.133+ (-0.009 - 0.275) 0.170+ (-0.001 - 0.341) Age of PRIME Participation (in years) 9.5 0.018 0-1 (-1.5 - 20.6) (-0.084 - 0.121) 9.4 -0.054 2 (-3.3 - 22.0) (-0.182 - 0.074) 20.5* 0.146* 3 (3.8 - 37.3) (0.007 - 0.285) 32.6** 0.233** 4+ (13.0 - 52.2) (0.066 - 0.400) 4,200 4,200 4,200 4,200 4,200 N The benchmark level corresponds to the set of households (N = 2,907) which did not receive any MFI support during any of the three rounds of household surveys. Robust 95% Confidence Intervals in parentheses. ** p<0.01, * p<0.05, + p < 0.1. The baseline group in this case is the benchmark households who did not participate in any NGO/MFI during 2008-2010 calendar years. For total land (in logs, Column [1]), we used simple OLS with robust standard errors. For the other variables which are count in nature, we used GLM models with negative binomial family and log link. * We found for the first two years of participation, the households did not gain significantly in comparison to the benchmark households, while after three years of participations, the households accumulated 21% higher total land and the households who participated for four years (and possibly moved on to other programs even before that) accumulated 33% more land assets (see Column [1] in Table 8)12. We saw a similar trend in other types of assets as well. The total number of livestock increased by 15% in comparison to the benchmark households after three years since participating into the program while it increased by 23% for the households after four years and more of participating into the program. For total number of cows, we again found for the first two years, the households did 12 One should notice that the reference group had only 11 decimals of land (see Table 7), so 21% and 33% translate into only 2-3 decimals of land for these household. Page 26 PRIME 3rd Round Report. Please do not cite or circulate. not exhibit any significant impact13. However, after three years since participating in to the program households accumulated 19% higher number of cows (significant at 10% level of confidence) compared with the “benchmark” case and 39% more after four or five years of participation in to the program. We saw a similar pattern for the other categories (goats and poultry, albeit a little weaker for the total number of poultry in the statistical sense) as well suggesting that the PRIME program established the participating households in a dynamic path of asset accumulation which usually surfaced after three years since participating in the program. This endorsed findings reported in the second evaluation report. Table 9: Asset value and Savings against Age of PRIME Participation. Log(Total Asset) Log(Total Asset Value excluding Land) Log(Savings) 103.59 27.57 1.79 24.6** (17.3 - 31.9) 14.4** (4.4 - 24.4) 23.4** (13.0 - 33.9) 37.8** (23.9 - 51.7) 157.8** (138.0 - 177.5) 162.4** (139.1 - 185.6) 142.5** (113.0 - 172.0) 188.0** (157.0 - 219.1) * Benchmark Level (Thousand Taka) Age of PRIME Participation (in years) 23.0** 0-1 (12.9 - 33.0) 17.7** 2 (5.1 - 30.4) 30.2** 3 (15.6 - 44.8) 42.1** 4+ (24.2 - 60.0) 4,200 4,200 4,200 N The benchmark level corresponds to the set of households (N = 2,907) which did not receive any MFI support during any of the three rounds of household surveys. Robust 95% Confidence Intervals in parentheses. ** p<0.01, * p<0.05, + p < 0.1. * We also looked at the accumulation of assets in value terms.14 Not surprisingly, we found the same dynamic pattern in asset accumulation in value terms as well. However, there were some nonlinearity in this case. The households who participated into the programs earlier higher asset values for all three categories of assets we looked into. Total assets grew by 30% after three years and 42% after four years or more since participating into the program. While exhibiting a U-shape pattern in non-land asset accumulation the early participants did distinctly better compared with the “benchmark” households. Having savings an integral part of the microfinance services offered to the participating households under PRIME, the savings were higher compared with the “benchmark” households and continued to grow as the years progressed since participating into the program. 13 As a matter of fact, for this particular case, the households who are new participants in to the program actually were worse off compared with the “benchmark” households. 14 These results are based on cross-sectional data for the same year. So it did not require correcting for price levels for the same current year. This qualification is applicable to all value or nominal figures in this report. Page 27 PRIME 3rd Round Report. Please do not cite or circulate. Table 10: Income against Age of PRIME Participation. Log (Total Income) Log (Total Income from Livestock) 45.68 1.29 18.1** (11.6 - 24.6) 17.9** (10.1 - 25.7) 24.0** (16.6 - 31.3) 34.7** (24.9 - 44.5) -5.8 (-36.5 - 25.0) -2.8 (-41.7 - 36.1) 49.1* (3.6 - 94.7) 66.2* (10.3 - 122.1) * Benchmark Level (Thousand Taka) Age of PRIME Participation (in years) 0-1 2 3 4+ 4,200 4,196 N The benchmark level corresponds to the set of households (N = 2,907) which did not receive any MFI support during any of the three rounds of household surveys. Robust 95% Confidence Intervals in parentheses. ** p<0.01, * p<0.05, + p < 0.1. * We saw a similar trend in total income as well (see Table 10). The participants did distinctly better compared with the “benchmark” households. Moreover, over the age of program participation the total income exhibited a pattern of continual increment. The first two years of participation we did not see much of a trend while after three years the total income increased by 24% and by four years the total income of the participants increased by 35%. For the first two years, the income from livestock did not change in any statistically significant way. However, three years after participation into the program the income from livestock increased by 49% and it continued to grow to 66% compared with the “benchmark” households. We also looked at the food and non-food expenditure separately (see Table 11). We found the same dynamic pattern in expenditure also. In the case of newer participants (such as less than two years of participation), the households showed a significantly higher levels for both kinds of expenditures however, there was no significant changes in the growth rates in food and non-food expenditures during the first three years of participations. However, there was a significant jump in the levels of expenditure after three years. Food expenditure grew 19% after three years and 29% after four years or more since first participating into the program. Similarly, non-food expenditure grew by 36% after three years and 58% after four years or more since participating into the program in comparison with the “benchmark” households. Page 28 PRIME 3rd Round Report. Please do not cite or circulate. Table 11: Food and Non-Food Expenditure against Age of PRIME Participation. Log(FoodExpenditure) Log(Non-Food Expenditure) 32.25 7.02 17.1** (13.6 - 20.6) 18.4** (13.5 - 23.2) 18.6** (13.4 - 23.9) 29.1** (22.9 - 35.4) 32.6** (25.3 - 39.8) 34.2** (24.9 - 43.4) 35.6** (26.1 - 45.2) 58.3** (44.9 - 71.6) * Benchmark Level (Thousand Taka) Age of PRIME Participation (in years) 0-1 2 3 4+ 4,198 4,087 N * The benchmark level corresponds to the set of households (N = 2,907) which did not receive any MFI support during any of the three rounds of household surveys. Robust 95% Confidence Intervals in parentheses. ** p<0.01, * p<0.05, + p < 0.1. We also found a similar dynamic pattern of impact in time spent in both agricultural and nonagricultural self-employment (see Table 12). Here we again had the “marginal impacts” of participation comparing to the “benchmark” level (i.e. no participation). Months engaged in nonagricultural self-employment increased by 1.4 months/year (about 54% more) engaged in nonagricultural self-employment compared with the “benchmark” level. But, this increases by 1.6 months/year after three years and 1.9 months/year after four years or more since participating into the program. Days per month engaged in agricultural self-employment increases by 1.3 days after three years and 4.9 days after four years or more since participating into the program which means an increase of nearly two months of agricultural employment in a whole year for the older participants. Page 29 PRIME 3rd Round Report. Please do not cite or circulate. Table 12: Engagement in Agricultural and Non-Agricultural Self-Employment against Age of PRIME Participation. [1] [2] Non-Agricultural Self-Employment Month(s)/Year 1.9 (1.8 - 2.0) Age of PRIME Participation (in years) 1.4*** 0-1 (1.0 - 1.7) 1.5*** 2 (1.0 - 1.9) 1.6*** 3 (1.1 - 2.1) 1.9*** 4+ (1.3 - 2.5) 4,200 N Percent --- [3] [4] Agricultural Self-Employment Day(s)/Month 9.5 (9.1 - 9.8) Percent --- 0.535*** (0.428 - 0.641) 0.565*** (0.434 - 0.697) 0.596*** (0.451 - 0.740) 0.684*** (0.511 - 0.856) 1.3** (0.3 - 2.4) -0.7 (-1.8 - 0.3) 1.3* (-0.1 - 2.8) 4.9*** (2.7 - 7.2) 0.132*** (0.035 - 0.228) -0.081 (-0.202 - 0.040) 0.133** (0.001 - 0.266) 0.418*** (0.260 - 0.577) 4,200 4,200 4,200 * The benchmark level corresponds to the set of households (N = 2,907) which did not receive any MFI support during any of the three rounds of household surveys. Robust 95% Confidence Intervals in parentheses. ** p<0.01, * p<0.05, + p < 0.1. 3.4 Crisis Coping In the current household survey, 2,693 households reported going through at least one unexpected shock in 2010. Total number of reported shocks or crisis was 2,998. About 50% of all shocks was in the form of accident or illness of household member. Other major shocks were natural disasters like flood, storm; livestock diseases etc. In 40% of the cases, households did not report any actions to cope up with these shocks. In rest of the cases, we found that self-insurance was the major coping mechanism adopted by the household. Other major coping mechanisms were borrowing and aid/help from different sources. Self coping mechanism included coping up the shock using either own savings or sell/mortgage of land or sell of livestock or other assets, etc. In the absence of more formal insurance schemes, a better asset position would allow a household to be in a better position to deal with negative shocks. We already saw (in Table 8 and Table 9) that PRIME participants were in a better position in terms of accumulating physical and financial assets. So the households which participated in PRIME for a longer time were supposed to have a better self-coping mechanism. Page 30 PRIME 3rd Round Report. Please do not cite or circulate. Figure 17: Fraction of Households Adopting Self- Figure 18: Marginal Impact of Program on Coping Mechanism. Reporting Self-Coping Mechanism to Cope The Shock . 20% PRIME Participants 46% 44.4% 15% 44% 11.7% 42% 10% 40% 37.8% 38% 36% 37.1% 5.0% 5% 35.5% 34% 4.3% 2.7% 0% 32% 0-1 30% 2 3 4+ -5% 0-1 2 3 4+ Age of PRIME Participation in year Source: InM Research Team’s estimates using household surveys. The benchmark indicates households who did not participate in any MFI/NGOs over 2008-2010 (N = 1,114). The vertical axis shows the fraction of household (%) reporting adopting selfcoping mechanism to cope the shock. Source: InM Research Team’s estimates using household surveys. The benchmark indicates households who did not participate in any MFI/NGOs over 2008-2010 (N = 1,114). The vertical axis shows the coefficients for each category in the horizontal axis. The vertical line for each point shows the 95% confidence intervals. Figure 17 showed that about 35% PRIME participants with less than a year in the program households reported to cope up with the shock using self-coping mechanism while this percentage is about 44% for those who had been in the program for 4 to 5 years. Figure 18 showed that up to participation age of three years, the impact of program on adopting self-coping mechanism was not that significant. But this marginal impact was statistically significant when the program participation age was four or more.15 15 Temporary within country migration is probably another mechanism by which households tend to cope up with negative income shocks. However, it was not ex ante obvious whether higher financial access and associated larger income and asset position would encourage or deter such migration behavior. So the research team decided to leave this issue outside the purview of this report. Page 31 PRIME 3rd Round Report. Please do not cite or circulate. Figure 19: Value of Housing by Duration of Figure 20: Participation in Non-PRIME MFI PRIME Participation. Programs by Duration of PRIME Participation. 13.3 13.5 79% 74% 11.0 71% 68% Benchmark 0-2 3+ Value of Housing (Thousand Taka) by Duration of PRIME Participation, Measured in Year(s) Source: InM Research Team’s estimates using household surveys. 0-1 2 3 4+ Duration of PRIME Membership, Measured in Year(s) Source: InM Research Team’s estimates using household surveys. We further looked at value of housing by duration of PRIME participation. Compared with the benchmark group (the households without any NGO/MFI participation), the total value of the housing of the PRIME participants was about Tk. 2,000 higher on average and the gap was statistically significant. However, there was no dynamic trend over the duration of membership in PRIME. We also looked at participation in other NGO/MFI programs. We found that over the duration of PRIME membership participation in other programs increased and seventy-five to seventy-nine percent of the participants with three or more years in PRIME retained participation in other NGO/MFI programs. Thus, PRIME worked as induction into the MFI industry which perhaps otherwise would not be possible.16 16 We also looked at migration and its association with PRIME membership duration. There was no meaningful association with duration of PRIME membership and also investigated level of migration (defined as if any of the household members lived outside home over the last year) for the benchmark households. We did not find any difference in level of migration between these groups (i.e. the benchmark households and PRIME participants with different lengths of durations) Page 32 PRIME 3rd Round Report. Please do not cite or circulate. Figure 21: Fraction (%) of Households Reporting to have Treatment from “Quack”. 65 62.02 60 55 49.61 50 49.05 45.7 45 40 0-1 2 3 4+ Source: InM Research Team’s estimates using Household surveys. The benchmark indicates Households who did not participate in any MFI/NGOs over 2008-2010 (N= 2,863). The vertical axis shows the fraction of household (%) reporting their interest to send their school-going aged children to school. 3.5 Health, Perception and Decision Related Aspects We also addressed some health case use and decision making capacity by women related parameters in the report (as required by The Terms of References for the third round of evaluation exercise). As such, we intended to find the impact of PRIME intervention program on access to health care and women’s participation at both family and community level decision making processed as well. 3.5.1 Awareness about Health Care To do that we restricted the target sample by the households who had at least one member suffering from any disease over the three months before the time the household was surveyed and had taken some treatment for the ailment. We classified different health service providers into two broad categories namely formal and informal sources of treatment. The formal sources included MBBS doctor, hospital or clinic, upazilla health complex, health worker from NGOs or any other formal sources. The Informal sources, on the contrary, included all types of informal providers of health care including quacks, kabiraj, compounder of a pharmacy or any other informal provider. The households living in the rural areas usually have a much limited access to formal health care so they are driven to seek health care from informal sources by consulting quacks or the village doctor in the first place whenever they encounter a disease, which sometimes lead them towards more sufferings as these village doctors are not qualified enough and often fake people in the name of Page 33 PRIME 3rd Round Report. Please do not cite or circulate. treatment. We tried to see whether the PRIME intervention had any impact on the awareness about health care among people. Like the previous sections, we also tried to see whether the age of participation in the PRIME program provided any indication whether the program improved the awareness about health care among the targeted households or not. We found that the fraction of households reporting treatment from a village doctor is lower for the households participating in the PRIME program earlier compared to the households who were in to the program for less than a year (Error! Reference source not found.). The latter members reported 62% of the time getting treatment from a quack, whereas the households who have been in the program for four or more years reported seeing a quack 46% of the time. This suggested a 16 percentage point decrease, implying an “impact” of the PRIME associated with the age of participation. Comparing with the benchmark, we found that the newest members of PRIME are somewhat close to those who were not in to the program yet (including any other NGO/MFIs). The “benchmark” level implied that about 60% of the non-participant households reported taking treatment from a village doctor, which is slightly lower than the percentage of household those have been under the program for 0 to 1 year, which is around 62%. But with the age of participation in PRIME, percentage of households reporting health care from a quack declined and for the households under the program for 4 or more years reported taking treatment from quack is the lowest. Figure 22: Fraction (%) of Households Reporting Figure 23: Fraction (%) of Household Reporting Interest to Educate Their School-Going Aged Mother’s Participation in the Decision of Children’s Children. Marriage 90.0 100.0 99.35% 99.5 89.5 99.25% 89.5% 99.0 89.0 98.5 88.5 98.0 97.89% 89.15% 88.0 97.73% 87.5 97.5 87.55% 87.0 97.0 0-1 2 3 4+ 87.08% 86.5 0-1 2 3 4+ Page 34 PRIME 3rd Round Report. Source: InM Research Team’s estimates using Household surveys. The benchmark indicates Households who did not participate in any MFI/NGOs over 2008-2010 (N=1,261). The vertical axis shows the fraction of household (%) reporting their interest to send their school-going aged children to school. 3.5.2 Please do not cite or circulate. Source: InM Research Team’s estimates using Household surveys. The benchmark indicates Households who did not participate in any MFI/NGOs over 2008-2010 (N= 2,863). The vertical axis shows the fraction of household (%) reporting their interest to send their school-going aged children to school. Perception regarding Child Education In order to capture the perception about children’s education we targeted the group of households having at least one child aged between 0 to 7 years. Although the percentage of households showing willingness to send their children to school was high for the “benchmark” group, yet the fraction was higher for the program participants. Even the newest members of the program reported 98% of the times that they were willing to send their children to school, whereas for the benchmark households this fraction was around 96%. We also found that the probability of such report improved with the age of participation. The households under PRIME for 4 or more than 4 years reported 99% of time willingness to educate their children, which was, in fact, a very inspiring figure. 3.5.3 Participation of Women in Decision Making In the patriarchal society of Bangladesh husband is usually the major decision-maker, especially on issues relating to family matters. In our analysis, we tried to see whether the PRIME program has any effect in improving women’s access in some major decisions like their children’s marriage or other major family issues. The targeted group was those who have children eligible to get married (aged between 13 to 25 years). We found that the benchmark households had always been below the PRIME participants in case of women’s participation in the decision making process of their children’s marriage. The following figure showed that for the oldest members of PRIME 89.15% of the time they reported that the decision of their children’s marriage was taken either by the mother alone or both the parents together. Although the participants aged between 2 and 3 years show a lower percentage value than the newest and oldest members, yet the fraction was higher than the “benchmark” level. 3.6 Discussion and Concluding Remarks on Impact of PRIME While it was challenging to identify the impact of the program from a non-random intervention, we carefully found pattern in the data that suggested systematically that inclusion of ultra-poor households to financial system (as offered by microfinance industry) along with training and perhaps monitoring could put such households in a path of development and improvement of their economic situations. One of the challenges was to find a control group comparable with the households that received the treatment (i.e. participated in the PRIME). In absence of defining such a group a priori, we synthetically defined a group that would be less prone to selection. We (as we did in the previous Page 35 PRIME 3rd Round Report. Please do not cite or circulate. 2nd phase report) compared the new entrants with the households who participated in to the program three or more years. People or households that participated first in to the program were probably similar (in terms of many baseline household characteristics) and selection was probably much less of an issue between these groups. We also identified and included a group of households who never participated in any NGO/MFI program. We found many outcome parameters were very similar between these “benchmark” and new entrants in to the program (i.e. households who were PRIME participants for less than a year) exhibited very similar levels for many of the outcome variables. Our findings endorsed that it was important to look at households over a period of time after they participated into ultra-poor programs such as PRIME because it might take a while for the ultra-poor households to rip benefits from such programs. The longer term dynamics of the program’s impact on the households was evident for most of the parameters. As for consumption ordering, we found similar results. It was important to notice that there was a general trend of improvement in consumption situation over last three years. However, when we compared households with different age of participation we found that households who participated in the program ended up with a much better consumption ordering compared to households that either were new entrants in the program or were in the “benchmark” group. Other outcome variables also exhibited a similar pattern. The households who participated into the program managed to accumulate assets over time and the value of their assets also increased over time. We also found total income and income from livestock also increased over the age of the PRIME participation. As expected, self employment in both agriculture and non-agricultural sectors also increased over the age of participation. There were some evidences for improvement in better use of health care use as households showed lesser tendency to avail health services from the informal providers. PRIME did have health services in its program design to provide some health consultation and medicines. However, implications for such services from the data should be followed up in later studies. The decision making by women showed little variances in the data. The decision to send children or participating in family and/or community level did not vary over the age of the program participation as such we did not find any significant association with the time households spent with the program. Page 36 PRIME 3rd Round Report. Please do not cite or circulate. 4. Overall Assessment of the Program It was challenging to cater to the demand for financial services to the low-income households and this issue typified the microfinance sector from the very beginning. While reaching out to a particular segment of the rural poor households had largely been successful, extending the breadth of the market to the segments that had traditionally been ignored posed some special problems. PRIME was one such extension and reaching out to the ultra-poor households in the monga prone areas in the greater Rangpur areas probably offered many of the same problems typical of microfinance serving to clients that are both hard-to-reach because of living in the remote areas (such as char). It was our assessment that one of the major achievements of the program was to bring in many households in these areas into the microfinance outreach, especially those who proved to become viable microcredit/microfinance clients. The participants, while they moved out of PRIME, continued to procure bigger loans and divert resources to income generating activities in the long-run. This issue was important in identifying the impact of program and interpreting the results we presented in this report. It is our assessment that by exposing and engaging the beneficiaries to microfinance practices the program led many of the beneficiaries into a higher growth curve in terms of better food security, higher income, more self-employment and larger asset accumulation. While one could argue that these sometimes just meant delivery of the intervention, but a proper discount of the total benefit would probably show that the total value accrued would far exceed the initial investment on these households. The true economic value of the investment (since the investment came from a subsidized program) would probably show a better assessment which was left outside the purview of this evaluation exercise. One of the main items of interest was food security which has always a burning issue for the poor households in the region. Our findings suggested that the households that were brought into the development path because of the program, on average, exhibited a better food security by reporting having meals at least three times a day during both “normal” (during the time of the survey) and monga periods. The dynamic pattern we found in the previous edition of the evaluation report continued this year as well as evident from the data. While overall monga situation improved over time and this secular trend for all types of households (categorized by MFI participation including non-participants) we could systematically associate improvement in consumption ordering with years since an individual household participated in PRIME. One very interesting facts of the dynamic impact of the program was to find that the households who never participated in any of the program (the “benchmark” households) and the households Page 37 PRIME 3rd Round Report. Please do not cite or circulate. who were recently selected into the programs (i.e. less than one year into the program) were surprisingly similar in many of the outcome variables we looked at. This suggested while there were probably many non-participating households in a similar economic situation (as indicated by the outcome variables) but because of various natural (e.g. char) and socio-economic constrain could not participate in the program. “In spite of having sufficient staff, the PRIME We ventured into the selection issue and did find branches are facing higher cost than revenue some systematic associations between program earned because of lower average loan size. participation Average loan size should be more than BDT and some selected household characteristics. For example, the non-participating households were more likely to be in the chars. As such, geographic barriers and living in the remote places did provide some hindrance against 8,000. In this regard the supply of money should be ensured. According to the demand various kind of loan program can be offered like RMC, UMC, ME in the PRIME branches.” - A PRIME coordinator. participating in the program (PRIME or otherwise). We also found the non-participating households were more likely to be female-headed, had older household heads with less education and were also less likely to be currently married. These nonparticipating households also had lesser number of income sources and spent less time in nonagricultural self employment. Similarly, the relative economic deprivation was also evident from never participating households having lesser access to electricity and sanitary latrine. All these in combinations suggested that the program could further reach out to households who perhaps had “In order to be sustainable branch drop out of some use of microfinance products however, at the the borrowers should be stopped. They should time of the survey did not have any “access” to any microfinance program. This led us to look into the institutional and operational side of the story because PRIME be offered various kind of non-credit program. To sustain the borrowers they should be offered after loan support like IGA implementation monitoring and other support.” A PRIME coordinator. program was implemented under a particular institutional set-up. While the program kept the cost of fund lower by the whole-sale credit provider (i.e. PKSF) providing the fund at a subsidized rate, the POs also faced the problem of dealing with a situation where the branches perhaps would exhibit less productivity because of smaller number of clients per field officer and also smaller loan size. So the double bottom-lines were more stringent for the branches under the program and we found this to be largely true in the data. For example, from a group meeting with the PRIME coordinators from all the POs implementing PRIME, the research team gained some important insights into the program from an operational Page 38 PRIME 3rd Round Report. Please do not cite or circulate. point of view. The participants of the meeting raised concerns about the financial viability of the branches implementing PRIME (which we found to be true from the data as well). The main incentive to run these branches were pressure from PKSF and according to them even the subsidy itself was not enough to cover the cost of operation for these branches. So, for the POs it was loans tied with conditionality and they were cross-subsidizing from more profitable operations in other areas. One of the major concerns was high turn-over rate and high drop-out rates. It was not that the dropped out beneficiaries stopped taking loans altogether (rather they moved on to other more lucrative offers indicating competitive pressure from other MFIs even in such targeted areas). One of the major reasons for such high turn-over was the cap on loan-size. Many beneficiaries showed capacity to take larger loans and divert it to useful income generating activity in a productive way. However, since these beneficiaries were new comers to the microfinance sector, as a policy MFIs were reluctant to provide larger loans to “As PRIME is designed for the very vulnerable households and the loan management capacity of these people are low the average loan size them hence they many-a-times found such loans becomes low. So it needs minimum three years not so useful and dropped out. It was important to to get operationally self-sufficient for a branch. note larger loan per borrower was also a way to Big organization can provide subsidy but the be more efficient and productive. This was one small MFIs can’t. So to open a branch the aspect of the program the designer should look into more. sufficient amount of money should be ensured at first.” A PRIME coordinator. A related issue was offering of non-credit program with the more traditional microfinance services (credit and/or savings). The coordinators expressed concern that it was common for them to compete with bigger NGO/MFIs because of the larger one offer so called “credit plus” services (e.g. BRAC). So the more traditional credit services were facing more and more competitive pressure. There were ancillary services offered under PRIME however more product diversifications and along with need-based program along with microcredit/ microfinance might help retain clients. It was not also obvious (to the researchers) what policies were in place for the graduating clients which the again the designers should address for the overall welfare for both the clients and MFIs. Since the program itself was yet to be operationally (hence financially) sustainable, we needed to focus and highlight the social returns from the program which we found to be largely positive. In almost all of the parameters we found the dynamic pattern of improvement with some nonlinearity. The PRIME participants continued to gain in terms of asset accumulation both in kind and in values. The households (under PRIME) found themselves in a higher income growth path and we Page 39 PRIME 3rd Round Report. Please do not cite or circulate. found evident that the income from self-employment (in both agricultural and non-agricultural sector) and direct income from livestock grew accordingly. We also looked at few health-related parameters. Health service had been made an integral part of the program and the team met with medical officer (a registered physician) employed under the program. The program gave out primary health services (consultation) and some medicines to the beneficiary households. One limitation of the study was that it did not explicitly look at this aspect of the program in details. However, the questionnaire did include a specific section on the overall health care use. We found that taking health care services from the informal sector came down with the age of the program. However, we feel that health care related issues should be addressed more closely in the coming rounds because the poor households are especially vulnerable to disease and many-a-times health related shocks are considered as one of the major reasons for households to fall back to poverty through loss of income and asset depletion. We also looked at few parameters which would indicate improvement in decision making by the women in the beneficiary households. However, we did not find any significant association with the age of the program. In most cases, there were not much variations in the data to identify the impact. We found some “soft” association of mother being involved in child’s marriage over the age of the program. But sociological gain from the program was hard to measure to begin with from a household survey and perhaps more ethnographic study was required to identify and document these aspects of the program which was outside the purview of this exercise. Page 40 PRIME 3rd Round Report. Please do not cite or circulate. 5. Concluding Remarks One should not shy away from noticing the limitations of this study. Firstly, the issue of selection was not embedded in the study design itself. As such the estimates were expected to be biased. While we were careful enough to construct synthetic cohorts by varying intensity of impact and also a “benchmark” group, this perhaps still would not replicate identifying the average treatment effect of the program in a clear way. However, the overall research design would allow a researcher to comment on the selection issue itself in a more general framework of participation which could potentially turn out to be strength of the research design. Secondly, like most other evaluation exercises, this research was based on a household survey asking interviewees to recall past experience and activities (e.g. consumption). This was bound to introduce some recall bias. While for few variables (e.g. loans) this error was expected to be minimal (because most households report these information from written passbook documents), for consumption and days worked throughout the year was bound to have some measurement errors in them. Thirdly, for some of the parameters like decision making and women empowerment household survey was probably not the right research design. A more “anthropological” approach or at least embedding those features in a questionnaire format remained challenge which should be addressed in the future rounds. Fourthly, while we explicitly included questions regarding health care use we needed to address the specific program related health care products and use in the later rounds of the study. Fifthly, the project appraisal could benefit greatly if there were some intermediate assessments in a higher frequency especially addressing the monga situation in real time. Sixthly, operational assessment would benefit from a more systematic reporting. For example, the research team found it difficult to measure parameters like loan loss provision (and the expense related to it) because of absence of comparable reporting across different MFIs. The improvement of MIS can reduce the cost (thus increasing the efficiency and productivity of the branches) of branch management. Even after these limitations, we made a consistent effort through a systematic data mining exercise to understand the impacts of the program in a dynamic setting. As we mentioned the concern in the industry about the “double bottom-lines” was especially salient for this program. The branches dealt with this issue and a frequent turnover probably sustained a low level of productivity given the constraint of mainly ultra-poor households as beneficiaries in hard-to-reach remote areas. Investing in productivity of the branches should be a top priority through a better data management process and retaining more productive employees. However, the cross-subsidization at the institutional level should be maintained through proper incentives. These policies are all the more important because of the long-term positive impact of PRIME program as clear evidence of a sustaining growth path in Page 41 PRIME 3rd Round Report. Please do not cite or circulate. income, asset accumulation and engagement in productive enterprises were exhibited throughout this evaluation exercise. Page 42 PRIME 3rd Round Report. Please do not cite or circulate. A. Operational Sustainability of the PRIME Branches 1. Preamble Making financial services available to the poorest people is recognized as important tool for poverty reduction. While “increasing outreach” has been the catch-cry as long as microfinance as its current delivery mode came into existence for the last thirty years, the present delivery models were not quite meeting the challenge, especially when it came to serving communities in remote locations characterized by low population density. One reasonable conclusion might be that reaching out to the poor people living in the ecologically vulnerable areas (such as char) would involve subsidized delivery mechanism in the short run and sustainability of the MFI branches serving such areas should receive some special attention. The quest for sustainability and eventual self-sufficiency is widely regarded as a best practice in the microfinance industry. Self-sufficiency is seen as an appropriate mechanism for achieving the longterm viability of the microfinance sector. First, available resources and subsidies are too small to provide microfinance to all who might benefit from it. Second, a focus on self-sufficiency can lead to decreased costs through increased efficiency. Third, leverage is more easily attained by organizations that generate the means to repay debt. Finally, reliance on subsidies might alter a firm’s incentive structure in ways that could increase the likelihood of a negative event. 2. Evaluation of the PRIME branches One novelty of the present evaluation report was to explicitly address the issue of sustainability of the program from the financial point-of-view. One of the major achievements of the microfinance programs (subsidized or otherwise) was to reaching the poor though financial instrument (e.g. microcredit) using a mode of delivery which also addressed the issue of self-sufficiency. This, of course, put an additional constraint on a MFI’s outreach motive and delivering financial product to the hard-to-reach clients subject to operational and financial viability received its due attention. The PRIME program actually addressed this issue. But one should keep in mind that in its current form it was still subsidized and reaching toward program sustainability would first require the PRIME branches to show operational self-sufficiency. As a prerequisite to reach the goal of operational selfsufficiency it would also be important to assess the efficiency, productivity and sustainability of the branches through which PRIME had been initiated and implemented. Page 43 PRIME 3rd Round Report. Please do not cite or circulate. While there were many different aspects for branch performances, we, for the purpose of this evaluation of the program would focus on a limited set of parameters. As per Terms of Reference, the research team was required to assess the sustainability of the program. This was the first time the research team explicitly focused on the sustainability of the program using branch level information extensively. The collected data (through a structured questionnaire for each branch, see sections on data and sampling below) allowed the research team to answer some very specific questions such as: According to the TOR with PKSF, dealing with the institutional level is a part of PRIME evaluation. At the institutional level, the objective is to assess: How productive are the PRIME branches? How efficiently performing? are Are the PRIME sufficient? branches self (i) Impact on institutional capacity strengthening initiatives for POs How the previous measures behaved over the age of a branch? What factors are associated with the observed variations in the above measures? (ii) Operational sustainability of prime branches and suggest possible way to attain operational sustainability. the PRIME branches operationally Answering the above questions would provide guidance how to formulate the future path of the program so that the program was both operationally and financially viable. Increasing client outreach provided economies of scale that in turn made the microfinance program more efficient and therefore more sustainable, at least in the immediate financial terms. However, by design PRIME worked in areas where household traditionally had little access to financial services. The cost of delivering such services to ultra-poor households was generally high (because of being at river islands or “char”). Thus solely focusing on self-sufficiency in financial terms might cause such programs to move away from the initial aims of supporting the deprived households coping with monga. Yet the financial status of the programs might shed some light on designing the program more efficiently for future growth (possibly in other areas with similar situation). 3. Performance Parameters In order to analyze the sustainability and efficiency of the program, we looked at the individual PRIME branches. This was reasonable because the branches were the lowest units through which interface with the program beneficiaries actually took place. New branches were established to implement the program to reach out to the targeted households while some of the existing branches, strategically placed to reach the ultra-poor households were brought under the program. We emphasized on few parameters which can loosely be categorized under the following categories. Page 44 PRIME 3rd Round Report. Please do not cite or circulate. Productivity and Efficiency Expenses Financial Spread Operational Self-Sufficiency Productivity and Efficiency To measure the efficiency of human resource utilisation or staff productivity ratios – clients per member of staff and outstanding portfolio per member of staff – were the two key indicators. This analysis used the client-to-loan officer ratio and portfolio-to-loan officer ratio with some qualifications. It was difficult to classify staff as loan officers across MFIs because in most of the cases field officers played multiple roles which might go beyond credit management for borrowers. Many MFIs gave field officers responsibility for all functions related to microfinance groups. In this situation the definition of a loan officer was not clear. Financial Spread and Operational Self-Sufficiency For measuring the portfolio quality and financial viability, we looked at financial spreads and the operational self-sufficiency of all PRIME branches. Financial spread showed the ratio of net income (i.e. total income less financial cost) and average portfolio outstanding. Operational Self-Sufficiency (OSS) measured the financial viability of an institution. OSS showed the required amount of revenue required to meet the cost of the branch. OSS is a fraction (expressed in percent) which depicted the ratio of income and expenses of any institution. As such, a self-sufficient branch (in the operational sense) would necessarily have OSS of 100% or more. A MFI branch continuing under 100% of OSS is not considered operationally viable in the long-run. However, most branches (and organizations) initially exhibited an OSS below the threshold level of 100% in which case the trend in OSS over the age of the branch implied whether the branch would be operationally self-sufficient over time. 4. Sample We selected all 231 PRIME branches of five districts of Greater Rangpur currently under operations.17 These branches belonged to 16 different MFIs all of whom were partner organizations of PKSF and these were selected because of their prior performance and experience in implementing ultra-poor households in the program areas. As such we expected to draw a comprehensive operational and to some extent financial picture of the entire program from the supply-side. One should keep in mind, many of the MFI branches later started regular microcredit programs and currently only 41 branches were involved with micro-credit/micro-finance programs targeting only for ultra-poor households. 17 The branches of the POs implementing PRIME have changed over time in the sense that few branches have merged. There were 235 branches over the PRIME existence however after the branch mergers there are 231 branches under operation as of now. Page 45 PRIME 3rd Round Report. Please do not cite or circulate. Table 13: POs and Coverage 1 Partner Organizations ASOD 2 ESDO 3 RDRS 4 5 6 TMSS SKS GUK SL 7 8 9 10 11 12 13 14 15 16 PMUK POPI UDDIPON SSS S-SUS Gonoshastho GBK SHARP JCF BEES Distric(s) covered Rangpur & Lalmonirhat Nilphamari, Rangpur, Gaibandha & Lalmonirhat Nilphamari,Kurigram & Lalmonirhat Rangpur,Kurigram, Gaibandha & Lalmonirhat Gaibandha, Rangpur Gaibandha Nilphamari,Rangpur, Gaibandha & Lalmonirhat Nilphamari, Lalmonirhat Kurigram, Rangpur Rangpur & Kurigram Rangpur Gaibandha Rangpur & Saidpur Nilphamari Rangpur Rangpur, Gaibandha Number of Upazilas Covered* 6 Number of Unions Covered 20 8 43 29 15 66 53 10 8 2 44 50 7 27 35 4 10 2 6 4 2 4 4 1 1 2 29 12 28 11 12 8 9 4 3 8 12 4 18 8 6 7 5 4 2 6 354 231 All POs Number of Branches 11 Source: Construction from PKSF administrative data by the InM Research Team. *Upazilas covered here are double counted as the different POs are operating in the same upazila. But uniquely the 16 POs covered the 35 upazilas. Later on some POs had merged some of their branches. So the number of branched may change overtime. Up to March 2010, it is reported that there are 231 branches are operating in the field. The research team collected official year-end financial data for all PRIME branches. The Figure 24: Growth Curve for Number of MFI Branches Implementing PRIME program coordinator and the MIS officer 231 205 responsible especially for PRIME program for each PO were entrusted with the data 137 collection using a structured questionnaire prepared by the research team. The questionnaire asked for last three financial years’ data including number of borrowers, 21 loan outstanding, disbursement and savings. The questionnaire also included forms regarding cost and income of the branch over 2006 2007 2008 2009 Source: PKSF Administrative Data from PRIME Cell. There are few instances of branch mergers with are not reported here. Page 46 PRIME 3rd Round Report. Please do not cite or circulate. the last three financial years. InM research team conducted a training session in Rangpur and also conducted a focus group discussion. The day long training basically provided the concern POs how to fill up the data and what they thought about the entire PRIME program. They also provided insights on the sustainability issues of the PRIME branches. 5. Description of the Data We have data for last three financial years of every PRIME branch where available.18 PRIME starting years varied among branches. Initially 21 branches in 2006 started PRIME program. Later a rapid expansion took place in 2007 and in 2008. There were 231 branches operating PRIME program in Greater Rangpur. There were 132 branches (57% of total) started with PRIME and among them some branches were operating PRIME with some other programs. At the time of writing this report, 41 branches were running only PRIME or ultra-poor microcredit/microfinance program. The other 99 branches existed before PRIME program started and were chosen to implement with PRIME because of their geographical and programmatic advantage to reach the appropriate clientele. We analyzed the financial information of these two types of branches separately (see Table 14). We presented the summery statistics for these two types of branches in Table 14 and one should note that there were some differences between them exhibiting differences in financial performances between the two groups. The number of active borrowers per field officer was higher for the branches established through implementing PRIME (1,282 active borrowers per branch) than the pre-existing branches (1,478 active borrowers per branch). The portfolio outstanding per field officer for pre-existing branches (Tk. 1,248 thousand/field officer) was higher than newly initiated PRIME branches (Tk. 1,059 thousand/field officer). The number of active depositor per field officer was higher for PRIME initiated branches than the pre-existing branches implementing PRIME. The deposit outstanding per field officer and savings collected per field officer was higher for existing branches rather than newly initiated PRIME branches. The newly established branches also showed a lower level of efficiency (in the operational sense) compared with the pre-existing branches implementing PRIME. The operating cost as a percentage of average portfolio outstanding was much lower for the pre-existing branches (16%) than that of newly initiated branches (31%). The salary and benefit as a percentage of average portfolio outstanding was also smaller for pre-existing branches (13%) than the new branches of PRIME (24%). Loan loss provision was almost same for two groups. Interestingly, financial spread (as measured by 18 Some of the branches were established later so for these branches financial information were not available in the earlier years. Page 47 PRIME 3rd Round Report. Please do not cite or circulate. financial income less operating cost expressed as percent of average loan outstanding) was higher for the new branches (17.5%) as opposed to the pre-existing branches (15.7%). Table 14: Summary Statistics for PRIME Branches. Productivity Ratios Number of Active Borrowers per Branch Number of Active Borrowers per Field Officer Portfolio Outstanding per Field Officer (Thousand Taka) Amount Disbursed per Field Officer (Thousand Taka) Number of Active Depositors per Field Officer (Number) Deposit Outstanding per Field Officer (Thousand Taka) Savings Collected per Field Officer (Thousand Taka) Efficiency Ratios Operating Cost Ratio (% of Average Portfolio Outstanding) Salary and Benefit (% of Average Portfolio Outstanding) Loan Loss Provision (% of Average Portfolio Outstanding) Financial Spread (% of Average Portfolio Outstanding) All (N = 231) Pre-existing MFI Branches Implementing PRIME (N = 99) Branches Established through Implementing PRIME (N = 132) 1,366 1,478 1,282 241 232 247 1,140 1,248 1,059 2,207 2,746 1,804 272 260 281 385 476 316 242 250 236 24.5 (N=228) 15.9 (N=97) 30.9 (N=131) 19.1 (N=228) 12.8 (N=97) 23.8 (N=131) 4.5 (N=228) 4.2 (N=97) 4.7 (N=131) 16.7 15.7 17.5 Operational Self-Sufficiency (%) 78.3 93.1 Percentage of Branches with OSS >= 100% 24.2 40.4 Source: InM PRIME Research Team’s calculations from primary branch level data. 67.1 12.1 Lastly, we looked at operational self-sufficiency (OSS) of all the branches and also by the two types of branches defined above. One striking feature was that the average OSS over all branches was just 78.3% as of 2010 financial year. While the pre-existing branches were at a higher OSS (about 93%), the average OSS for branches established for PRIME were at a meager 67%. Even after 5 years into the program only twelve percent of the branches (16 out of 132 branches included in the analysis) reached operational self-sufficiency of 100% as oppose to forty percent of pre-existing branches (40 out of 99 branches included in the analysis) reached operational self-sufficiency. Page 48 PRIME 3rd Round Report. Please do not cite or circulate. 6. Assessing Financial Performance against Age of the Program It was not expected that a bank branch (microcredit or otherwise) would reach certain level of efficiency and self-sufficiency right away. It would usually take time for a branch to be financially viable. This convexity might exert a fixed cost for a branch start-up and justified investing on branch set-up in areas where MFIs would not go otherwise. As such we should expect better efficiency over time and the branch to be more productive, efficient and self-sufficient over the age of the program which we would address now. In Table 15, we classified the branches Table 15: Age of PRIME Implementation. according to the previous types and showed the distribution of the branches by age of the program. One should note that while for the PRIME initiated branches the age of the program and age of the branch coincided, for the pre-existing branches, this was not true. In these cases, we continued to classify the branches by the Years 0 1 2 3+ All (N = 231) Pre-existing MFI Branches Implementing PRIME (N = 99) 26 69 115 21 10 23 60 6 Branches Established through Implementing PRIME (N = 132) 16 46 55 15 Source: InM PRIME Research Team’s calculations from primary branch level data 2010. “Age of branch” measures the number of years the branch is implementing PRIME. age of the PRIME program, not by the age of the branch. The average age of the program for a typical branch was 2.55 years. For the preexisting branch implementing PRIME was 2.62 years while it is 2.52 years for new branch established for implementing the program. In this section, we Table 16: Age of PRIME Implementation over Financial Years 2008-2010. also exploited the information available for all three financial years and the unit of observation in this case was branch-year. Over three years, financial we had All (N = 640) Pre-existing MFI Branches Implementing PRIME (N = 297) Branches Established through Implementing PRIME (N = 343) Years 0 68 43 25 1 210 93 117 2 205 89 116 3+ 157 72 85 Source: InM PRIME Research Team’s calculations from primary branch level data 2010. “Age of branch” measures the number of years the branch is implementing PRIME. N=640 means that a branch is being counted in different years in different age groups. unbalanced panel of 640 observations (not all branches existed all the time hence the unequal number of observations over the years). Table 16 showed the distribution of age over the branch-years for all three years. To Page 49 PRIME 3rd Round Report. Please do not cite or circulate. have enough observations in each group of age (measured in years), we stratified the sample into 0, 1, 2 and 3 plus years of age. 7. Dynamics of Branch Performance over the Age of the Program We found that (see Figure 25) the number of Figure 25: Borrower Per Field Officer by Age of active borrowers per field officer increased with Branches. ages for both types of branches (hence in 300 aggregate, see Figure 25). It was much higher 250 for existing branches up to age of 2 years but 200 after 2 years into the program, the PRIME initiated branches caught up with the preexisting branches in terms of borrowers per 150 100 50 field/loan officers. 0 0 We saw a similar trend in portfolio outstanding per field officer however the PRIME initiated branch never reached the same level as the pre- 1 2 3+ All Pre-existing MFI Branches Implementing PRIME Branches Established through Implementing PRIME existing branches and even for the branches that were in the program for three years or more had a difference of Tk. 256 thousand in Source: InM PRIME Research Team’s calculations from primary branch level data 2010. “Age of branch” measures the number of years the branch is implementing PRIME. portfolio outstanding per field officer. If we compared this result with the ones we saw for the number of borrowers per field officer, the field officers at the pre-existing branches are maintaining a larger loan outstanding per borrower over a similar number of borrowers making them more efficient. We also saw a similar result for savings deposit outstanding (see Figure 27). The pre-existing branches had a much larger savings outstanding per field officer as a result pre-existing branches show a better capacity for resource mobilization over all (we should reemphasize that these branches probably had a more diversified portfolio). It was worth noticing that the branches managed to maintain a larger savings deposit over the age of the program on average. Page 50 PRIME 3rd Round Report. Please do not cite or circulate. Figure 26: Portfolio Outstanding Per Field Figure 27: Deposit Outstanding Per Credit Officer (Thousand Taka) and Age of Branches. Officer (Thousand Taka) and Age of Branches. 1,400 600 1,200 500 1,000 400 800 300 600 200 400 200 100 0 0 0 1 2 3+ 0 1 2 3+ All All Pre-existing MFI Branches Implementing PRIME Pre-existing MFI Branches Implementing PRIME Branches Established through Implementing PRIME Branches Established through Implementing PRIME Source: InM PRIME Research Team’s calculations from primary branch level data 2010. “Age of branch” measures the number of years the branch is implementing PRIME. Source: InM PRIME Research Team’s calculations from primary branch level data 2010. “Age of branch” measures the number of years the branch is implementing PRIME. Figure 28: Salary and Benefit as a % of Average Figure 29: Operating Cost Ratio as a % of Portfolio Outstanding and Age of Branches. Average Portfolio Outstanding and Age of Branches. 30% 60% 50% 40% 20% 30% 20% 10% 10% 0% 0% 0 1 2 0 3+ 1 2 3+ All All Pre-existing MFI Branches Implementing PRIME Pre-existing MFI Branches Implementing PRIME Branches Established through Implementing PRIME Branches Established through Implementing PRIME Source: InM PRIME Research Team’s calculations from primary branch level data 2010. “Age of branch” measures the number of years the branch is implementing PRIME. Source: InM PRIME Research Team’s calculations from primary branch level data 2010. “Age of branch” measures the number of years the branch is implementing PRIME. Page 51 PRIME 3rd Round Report. Please do not cite or circulate. The personnel cost (as measured by salary and benefit) as a fraction of total portfolio outstanding exhibited a hump shape as the cost (as percent of portfolio outstanding) went up after one year then came back to the previous level after that (i.e. after two years). We again saw that the pre-existing branches performed better in this dimension as well. Since salary and benefits were major part of the operating cost we found the similar trend in overall operating cost as well. The hump shape dynamics was especially visible among the pre-existing branches implementing PRIME and after two years the operating cost as percent of loan outstanding came down to pre-PRIME level. Not surprisingly, the PRIME initiated branches started with a much higher level of operating cost and this came down after two years of operations. So there were evidences of efficiency gain over time. 8. Financial Spread The average financial spread of the branches implementing PRIME showed an improvement over time. While in the first year of operation the financial spread stood at 7.4% while it reached 17.7% by the fourth year of operation. Financial spreads between two groups were somewhat distinguishable. The financial spreads for the branches established through PRIME showed a Figure 30: Financial Spread and Age of Branches. 30% 25% 20% more consistent dynamic over time. These branches started with a much lower level of financial spread but exhibited a secular as they 15% 10% aged over the years. In the first year of 5% operation, the average financial spread of 0% these branches was at 1.6% while it increased 0 1 2 3+ All to 19.8% by four years. The average financial spread for the pre-existing branch Implementing PRIME showed a more erratic pattern. The average financial spread of the Pre-existing MFI Branches Implementing PRIME Branches Established through Implementing PRIME Source: InM PRIME Research Team’s calculations from primary branch level data 2010. “Age of branch” measures the number of years the branch is implementing PRIME. pre-existing in the first year was 10.8% and by the end of fourth year it stood at 15.1%. It was interesting to observe that the by the end of fourth year (the last possible year of observation) in terms of financial spread the branches established through PRIME actually performed better. Page 52 PRIME 3rd Round Report. Please do not cite or circulate. 9. Operational Self-Sufficiency (OSS) Operating self-sufficiency is a ration (expressed in percent), which indicates whether or not enough revenue has been earned to cover a branch’s total cost which constitute operational expenses, loan loss provisions and financial costs. It was an important measure of sustainability of the lending operations. The OSS is measured by dividing Operating Income (Loans + Investments) by costs (Operating Costs + Loan Loss Provisions + Financing Costs). 0 0 .005 Density .005 Density .01 .01 .015 Figure 31: Distribution of OSS as 2010 for Figure 32: Distribution of OSS as 2010 for PreBranches Established through Implementing existing MFI Branches Implementing PRIME (N = PRIME (N = 132) 99) 0 200 400 oss 600 800 Source: InM PRIME Research Team’s calculations from primary branch level data. 0 100 200 oss 300 400 Source: InM PRIME Research Team’s calculations from primary branch level data. We showed the distribution of OSS for two types of branches separately In Figure 31 and Figure 32. The distribution of OSS for the branches established through PRIME were more concentrated toward the left side of the support suggesting our previous findings of lower percent of branches exhibiting operational self-sufficiency compared to the pre-existing branches implementing PRIME. Figure 33 showed that the pre-existing MFI branches implementing PRIME had a higher proportion of branches reporting OSS more than 100%. In 2010, 40 percent of the pre-existing branches had OSS greater than 100% whereas only 12% of the newly initiated PRIME branches had OSS greater than 100%. Latly, we looked at the OSS over the age of the program (see Figure 34). The overall average OSS for all the branches did not show any discernible trend over time. Initially the MFI branches implementing PRIME had an OSS of 80%. It came down after that before exhibiting an improvement over time. It reached 81% by the end of four years in implementing PRIME. The branches established through PRIME started with a low level of OSS of 23% while it increased to 70% by the end of four years. One should note that by the end of fourth year in operation these branches did not get Page 53 PRIME 3rd Round Report. Please do not cite or circulate. anywhere near self-sufficiency. Also more interesting was to notice that the OSS for pre-existing branches actually came down after PRIME was initiated through these branches (an average level of 113% in the first year to 95% by the end of the fourth year). Figure 33: Percentage of Branches with OSS above 100%. 120 40.4 38.4 Figure 34: Operational Self-Sufficiency (OSS, %) and Age of Branch. 100 32.3 80 60 12.1 6.3 40 20 6.1 0 2008 2009 0 2010 Pre-existing MFI Branches Implementing PRIME (N = 99) 1 2 3+ All Pre-existing MFI Branches Implementing PRIME Branches Established through Implementing PRIME (N = 132) Source: InM PRIME Research Team’s calculations from primary branch level data. Branches Established through Implementing PRIME Source: InM PRIME Research Team’s calculations from primary branch level data 2010. “Age of branch” measures the number of years the branch is implementing PRIME. OSS is measured as ratio of Operating Income (Loans + Investments) and costs (Operating Costs + Loan Loss Provisions + Financing Costs). We also classified the branches by the size of the MFIs (as reported by PKSF). There were no clear patterns between average branch OSS under a MFI and its size. Branches of only four MFIs were above or at least close to operational self-sufficiency (see Table 17). There were some instances for which the OSS actually declined. 10. Discussion The findings in this section suggested that operationally it remained a challenge for MFIs to reach out to ultra-poor households. While the branches on average maintained an average number of borrowers between 1,200 to 1,300 active borrowers (which is usually considered to be sufficient for a branch to break even), the branches on average were not able to show the level activities to selfsustain themselves. Low size of loans and high operating costs did not allow the branches to reach a level which would require to cover the cost of operations through service charges and other financial and non-financial revenues. So we observed that the PRIME branches increased their client outreach by increasing membership with small loan size which was reflected in the portfolio outstanding per Page 54 PRIME 3rd Round Report. Please do not cite or circulate. field officer and the amount disbursed per field officer. The loan ceiling of PRIME was much lower than the non-PRIME program. Table 17: Operational Self-Sufficiency (OSS) for different categories of PRIME POs. Size of MFIs PRIME POs Operational Self-Sustainability (OSS, %) 2008 2009 2010 JCF 15.3 72.5 PMUK 28.8 66.6 61.0 RDRS 85.1 79.8 97.2 Large SSS 25.3 35.8 52.5 TMSS 63.4 73.2 77.4 UDDIPAN 23.8 26.6 58.6 ASOD 84.5 50.2 37.8 BEES 97.1 81.4 94.9 Medium ESDO 22.4 38.3 53.0 POPI 35.9 47.5 83.0 SKS 135.1 119.5 111.7 GBK 74.1 55.1 57.2 GUK 1.5 21.9 60.1 Small GSK 36.8 58.9 60.7 S-SUS 152.5 148.8 103.9 SHARP 4.0 43.1 Source: InM PRIME Research Team’s calculations from primary branch level data 2010. The operational self-sufficiency ratios depended on productivity of the branch and also the efficiency. We found that the branches established through PRIME typically exhibited lower loan size and higher cost relatively in comparison with the branches existed before the PRIME was introduced. However, the ultra-poor programs evidently put some additional constraints on the performance of the MFI branches implementing PRIME. It was important to notice the dynamics of the financial performance of the branches under study. While it is generally expected that MFI branches to become operationally self-sufficient after three years, the PRIME branches did not indicate to reach this threshold level after three years. So increasing the productivity and efficiency of the branches should attract renewed attention. 11. Concluding Remarks Increasing client outreach provided economies of scale that in turn made the microfinance programs more efficient and therefore more sustainable, at least in immediate financial terms. However, emphasizing more on the higher operational sustainability (OSS) might eventually cause “mission drift” for such programs targeted toward special market segments. Page 55 PRIME 3rd Round Report. Please do not cite or circulate. PRIME especially focused on the ultra-poor households who otherwise did not have access to the financial services. On the other hand, if “increasing outreach” was taken to mean “targeting hard-toreach clients” such as people living in remote areas, then “outreach” and “sustainability” might effectively turn out to be competing terms. Reaching clients in remote areas was relatively costly, which made the microfinance program less productive and therefore less sustainable. This was the real outreach challenge for microfinance programs because it required new, as yet unproven business models and processes, including technological innovation. These would remain challenges for the sector in general and special programs such as PRIME in near future. Page 56