Determinants of Credit Participation and Its Impact on Household Consumption: Evidence From Rural Vietnam *

Transcription

1 CENTRE FOR ECONOMIC REFORM AND TRANSFORMATION School of Management and Languages, Heriot-Watt University, Edinburgh, EH14 4AS Tel: Fax: World-Wide Web: Determinants of Credit Participation and Its Impact on Household Consumption: Evidence From Rural Vietnam * Cuong H. Nguyen April 2007 Discussion Paper 2007/03 Abstract This paper analyses the Vietnam s rural credit market to understand the determinants of credit choices and to measure impacts of borrowing activities on borrower s consumption in the period. There are three main results. First, there exists uniform access to formal credit among rural households in Vietnam. Households financial activity is found to be determined by household size and agricultural work rather than education or distance from the commune to the nearest bank branch. Education level seems to have an inverse U-shape effect on credit taking possibility; the least and the most educated households borrow least. Second, there is evidence of money lenders being crowded out by formal institutions via competition. Finally, we apply fix-effected regression and propensity score matching estimation on cross-sectional and panel data to assess impact of credit taking on household consumption. Our study demonstrates that formal credit positively affects borrowers consumption while informal finance has mixed results. Keywords: rural credit, credit participation, Vietnam JEL Classification: O12, O16, O17 * Paper presented at the 3rd Leicester PhD Conference on Economics, May It won Prize for the Best Paper at the conference.. I am grateful to Mark Schaffer and Julian Fennema for their discussion and comments throughout this research. Financial support from CERT and School of Management & Languages, Heriot-Watt Universtiy is greatly appreciated. Any remaining errors in the paper are those of the author. CERT, Heriot-Watt University. Contact address: Department of Accountancy and Finance, School of Management and Languages, Heriot-Watt University, Edinburgh EH14 4AS, UK. cn5@hw.ac.uk.

2 1. INTRODUCTION Microfinance has been one of the attractive topics in economic researches for the last two decades. Inspired by successes of the Grameen Bank in Bangladesh and its replications around the world, economic researchers are seeking theoretical and empirical evidences to understand what make microfinance works. At some level, theoretical studies successfully model the mechanism of grouplending (peer monitoring), dynamic incentives (threaten to ban further credit if default) to show how these lending methods help microfinance institutions overcome problems of adverse selection and moral hazard. However, assessing credit impacts is often problematic and a consensus is still far from reach. This is mainly because of the inherent difficulty in evaluating programmes with self-selection and endogeneity. E.g. a borrowing household is performing better than a non-borrowing household not because she takes a loan but because she lives in a better economic environment, an area with easier access to credit or transportation, or she has a better entrepreneurial and managerial skills. The challenge that empiric researchers need to tackle is to eliminate those potential endogenous biases arise from lending procedures (non-randomness of credit program placement), credit participation process (selfselection) and unobserved characteristics. Different econometric models have been applied and results are also diverse. The most cite paper is from Pitt and Khandker (1998), in which the authors apply fixed effect cross-sectional analysis on household data from Bangladesh and use land holding size clear-cut as a exogenous condition of credit eligibility to generate control group. They find that the programme credit has a positive effect and women get more benefit when participating credit programmes than men. Khandker (2003) also find positive impact of credit on borrowers using panel data. However, using the same data to Pitt and Khandker (1998), Morduch (1998) considers a difference-in-differences estimation and finds the impact of credit programmes is small and negligible. Many other empirical studies have been published but very few of them seriously address problems of endogenous biases. The objective of this paper is to analyse the role of credit in Vietnam rural areas; to understand the determinants of credit choices; to measure the impact of borrowing activities on borrowers consumption. The paper employs commune-level fixed-effect and propensity score matching model as the key empirical strategy to address estimation biases. The structure of the paper is as follows. Firstly, we provide an overview of rural credit market in developing countries and in Vietnam in section 2. Section 3 is some basic characteristics of data samples. Determinants of households credit activities are analysed in section 4. The last section will be an empirical study of impacts of credit participation on households consumption. Discussion and conclusion are presented at the end of the paper. 1

3 1. RURAL CREDIT MARKET IN DEVELOPING COUNTRIES 2.1 OVERVIEW Production is time lag from input to output and when the input is acquired, if people are unable to finance the project themselves, they have to borrow from outside either formal lending institutions or informal money lenders. Their choices of borrowing are really depends on how they can access credit providers and how can they obtain the loan. The rural credit markets in developing countries are often described as repressed, imperfect and fragmented (or absented) in the sense that different segments of borrowers are observed to have different level of access to certain types of loans and certain types of credit institutions (Hoff and Stiglitz, 1990). In some markets, would-be borrowers may find themselves excluded or dissuaded from obtaining access to certain credit instruments, or rationed to loan amounts smaller than they might have optimally chosen, by collateral requirements and other non-price terms. They may then adjust by turning to substitute, but possibly more expensive financing sources or may modify their first best allocation plans in other ways (Conning and Udry, 2005). This explains for the fact that informal money lenders still exist (even dominated the market in some places) although they are accused of taking their position to exploit the poor by charging high interest rate and short term loan, which is sometimes pushing borrowers to state of unable to repay. Aleem (1990) reports that money lenders in rural credit market in Pakistan charged an average interest of 80 percent per annum with standard deviation of 40 percent. However, assessing to formal credit is hard for many rural people. Reasons are not only limited availability of formal finance but also characteristics of the lending contracts and requirements of formal lenders. As results of asymmetric information and weak enforcement system, formal credit institutions in developing countries seem to seek for some kinds of collaterals to make sure their loan will be repay. These can be land using rights, fixed assets, guarantee from local officer (this is often an extra required collateral but not the only) or some other regulations making accessibility to formal credit sometime impossible (i.e. borrowers need official identification or documents which are not always available for poor farmers). In contrast to formal finance, informal finance provides a much easier conditions to get loan. Borrowers and lenders often live in the same areas or easy to get to know each others. Lenders often do not require any collaterals or documents, they treat each other by informal laws. Borrowers are incentive to repay basing on the relationship and trust or sometime borrowers are threatened to repay by violence. The difference level of accessible to loan explain why there co-exist formal and informal finance in rural developing countries. The high cost and limited enforcement are other features of rural credit market. While informal lenders are charging higher interest rate, they are cheaper to access and it may be quick in disbursement. 2

4 It is an opposite case for formal credit. Higher cost, lower interest rate and slower to get the loan disbursed (Hoff and Stiglitz 1990). Using data from Thailand, Gine (2005) estimates that cost of accessing a formal institution is approximately US$30 while the cost of accessing an informal source is relatively small and negligible. Before the rise of microfinance as an alternative strategy for poverty reduction in developing countries, governments were attempted to provide credit to poor farmers and other rural producers by direct-lending schemes via formal financial institutions. Many such attempts have failed. A large literature has made clear that direct-credit often created extremely poor incentives to make sensible loans. These polices also had a high cost of delivering credit to target households and failed to enforcing contracts or mitigating problems of asymmetric information which all leaded to very high default rates (Hoff and Stiglitz 1990, McKinnon 1973). The microfinance movement aims to solve those problems and find the way to reach the poor people in a more efficient way. There are growing number of success stories and microfinance institutions have reported reaching over 80 million poor people in developing countries. 2.2 VIETNAM The economic reform in Vietnam started in 1986 and two years later, the banking reform was also implemented. During the first phase of the reform, (when the Asian Financial crisis began), the GDP growth rate in Vietnam was over 7 percent annual, achieved the 2nd highest growth rate in the region (behind China). After 1997, learning the lessons from financial crisis in neighbour countries, Vietnam sped up its financial reform. New law/regulations on banking operation have been amended or built up aiming at a more efficient legal framework for banking system. A number of joint-stock banks, joint-venture banks and foreign bank are allowed to join the market. Interest rate and exchange rate regime is eased. Recapitalisation of State Bank, State Own Commercial Banks and joint-stock banks has been implemented. National stock exchange was established in the year Vietnam has kept the growth rate at a fairly high level of average 6-7% annual since The rural credit market in Vietnam is repressed and segmented, where formal and informal finance exist side by side. The formal sector is led by the Vietnam Bank for Agriculture and Rural Development (VBARD) and the Vietnam Bank for the Poor (now restructured as Vietnam Bank for Social Policies) while the informal financial network is including private money lenders, relatives, friends, and local rotating savings and credit associations. By 1997/1998, the formal credit had accounted for 49 percent share of the market and continue to spread to around 70 percent by

5 2. THE DATA AND BASIC ANALYSIS The data for this study is the Vietnam Living Standard Survey (VLSS) conducted in 1992/93 and 1997/98 by the General Statistical Office of the Government of Vietnam, and funded by United Nation Development (UNDP) and the World Bank. The surveys contain detailed information of 4,800 households from 150 communes in VLSS93 and 6,000 households from 194 communes in VLSS98. Samples were weighted basing on the statistics of Vietnam Population Census in 1989 with approximately 80% of Vietnamese households lived in rural areas. The communes were randomly selected from a total of proximately 10,000 communes in 646 districts, and 64 provinces and cities in Vietnam, and then an average of 30 households were randomly selected for interview in each commune. The household survey has 15 sections to provide information on household characteristics including household size and composition, education, health, employment, agricultural productions, housing, expenditures, and borrowing and lending activities. The commune survey provides information on commune general conditions, location, business and production activities, and credit institution availability. Because our interests are in rural areas only, therefore all urban households and communes are dropped out, leaving only minor-urban and rural samples for the analysis BASIC CHARACTERISTICS OF HOUSEHOLDS Table 1 displays the general characteristics of households in the data samples. Male heads of household are accounting for 76 percent is not a surprise. The average education level of household heads is lower secondary school and there are 496 household heads (10 percent of the samples) never attended school. This could be explained as a result of the war in Vietnam which only ended in Following economic development, the rate of 80 percent households working in agriculture in VLSS93 reduces to 70 percent in 1998 with the average cultivatable land holding size increases from 1508 to 3,600 square meter respectively. Supported by the 1993 Land Law which granted five rights to the household: the right to transfer, exchange, inherit, rent and mortgage; the land and house ownership would give great opportunities for households to have access to formal credit market that they might never been able to approach before. 4

6 Table 1: Summary Data for Household Characteristics Characteristics of Household VLSS92/93(n=3840) VLSS97/98 (n=4,886) General Age of head 44.8 (14.7) 47.5 (13.78) Gender of head (1:Male; 0:Female) 0.77 (0.41) 0.76 (0.42) Education of head - Highest diploma (1.38) 3.1 (1.8) - Years in school 6.6 (4.1) Marital status (1=Married, 0=Single) 0.82 (0.38) 0.81 (0.38) Size 4.9 (2.11) 4.8 (1.9) Agricultural work (1=Yes; 0=No) 0.82 (0.37) 0.7 (0.46) Assets House ownership (0.17) 1.05 (0.29) Land holding size (m 2 ) 1508 (1836.7) (5884.3) Note: (1) Education highest diploma ranging from 1-6 including pre-school, primary school, lower secondary school, upper secondary school, vocational training, university. (2) VLSS93: 1=owner, 0=rent; VLSS98: House ownership: 1=whole property, 2=part of property, 3=rent. The land holding size is calculation of land for agricultural production purpose only. Source: The Vietnam Living Standard Survey 1992/93 & 1997/ BASIC CHARACTERISTICS OF CREDIT In this section we will provide some general information on loan attributes such as interest rate, duration and sources, and characteristics of borrowing households. We divide the sources of loan into formal and informal. The formal sources then break down to loans from Government banks (Bank for Agriculture and Rural Development, Bank for the Poor and other Government banks) and loans from Government programmes (Poverty alleviation programmes, Job creation programmes, Cooperative, and others). This is because of the domination of the Bank for Agriculture and Rural Development (BARD) and Vietnam Bank for the Poor (VBP) in rural credit market. BARD reportedly lent to 30 percent of rural households in

7 Table 2: Loan characteristics VLSS 1992/93 VLSS 1997/98 Finance sources Borrowing Borrowing households 1 Loan size 2 Annual interest rate households 1 Loan size 2 Annual interest rate Formal - BARD & VBP - Government programmes 26.5 (13.7) (35.6) 55 (29.1) 14.7 (7.7) Informal 85.8 (44.4) (27) Both sources - Formal only - Informal only 12.3 (6.4) (9.7) Total (51.69) (52.91) Note: 1. Percentage within borrowing households. Numbers in brackets is percentage of the whole samples. 2. Loan amounts are in 000 VND. Interest rate are annualised. Table 3: Distribution of loans by collateral Source of loans VLSS 1992/93 VLSS 1997/98 by household by loan number by loan amount by household by loan number by loan amount Formal - BARD & VBP - Government programmes Informal Total (17.2) (10.8) (18.7) (43.3) (35.9) (37.7) Note: "by household" shows the percentage of collateralized households by credit sources. "by loan number" reports the percentage of collateralized loan number over total loan numbers from the same source. The same calculation is set for "by loan amount". Number in parentheses is percentage of total borrowing households. 6

8 By 1998, nearly 53 percent of households in rural areas had been borrowing, in which 82 percent of them had loans from only one source, either formal or informal, leaving 18 percent of households borrowed from both sources. This is interesting. When a household decides how to finance their investment, they have four choices: self-finance, borrowing from a formal or informal lender or borrowing from both sources. If that household is credit constraint then borrowing is the only way he cans go and he will choose the best available option. Questions: why these households had to borrow from both sources? Are they credit rationed from formal source and must seek for more capital from informal lender to fulfill their demand? Study to find out what factors and how these factors affect household s financial decision would be worth to carry out. The table 3 shows that around 83 percent of borrowers in VLSS93, including both formal and informal finance, do not need collateral for their loans. This number reduces to 57 percent after 5 years in VLSS98. In the rest of 43 percent who need collaterals, 98 percent of them borrows from formal source. This is not a surprise. Due to asymmetric information problems on potential borrowers, formal credit institutions often use collaterals to reduce moral hazard and then improving repayment rate. This requirement from formal finance is the main reasons why rural households in developing countries have very limited access to formal credit market and normally depend on relatives or local money lenders whenever they need extra capital. That is it, 60 percent of formal loans (by number) were required to place collateral. This is not as high as we fear but why the rest 40 percent of formal loans were not asked for collateral. What made they different from others? A study to understand how formal credit institutions made their decision on loan applications is also interesting. Table 4: Distribution of collateralised loans by collateral assets (n=1,119) Collateralised loans Land House Others By households By loan number By loan amount Source: VLSS 1997/98 Table 4 reports type of assets households used as collaterals. It turns out that most of collaterals were land which accounts for 60 percent of collateralised households, 61 percent of collateralised loan number and 67 percent of collateralised loan amount. 3. DETERMINANTS OF HOUSEHOLD BORROWING In this section, we are going to explore the determinants of household credit decision basing on the VLSS1997/98. We use two measures of credit participation: (1) a yes/no answer for those who participated in a credit programme; and (2) the size of the loan received. We apply Probit model to 7

9 estimate credit programme participation and Tobit model to estimate loan amount received. The dependent variables of credit programme participation will have two value (1): household borrows or not (1=Yes, 0=No) regardless of the financial source, and (2): household borrows from formal source only (1=Yes, 0=No). Tobit regression will also run for two values of loan amount: total loan amount and loan amount from formal source only. By separating the source of loan, we expect that the determinants of credit participation will be different as the eligible requirements for borrowing are different between sources. Following Pitt and Khandker (1998) and Morduch (1998), among others, we specify credit programme participation or the loan size as a function of household characteristics including gender of household head, age of household head, number of household members, education level of household head, agricultural work, value of house and land holding size; and of commune characteristics including distance from commune to the nearest Government banks and distance from commune to the nearest BARD. The Probit and Tobit regression results are reported in Table 5. As we discussed before, there are about 60 percent of formal borrowing households and formal loan numbers have to use collateral and the collateral policy does not apply for all households. This is captured in both Probit and Tobit regression where the coefficients of land holding size and value of household s house, which are the most used as collateral, are small and significant. The coefficients of value of household s house is improved in the Tobit estimation for loan amount. Distance from commune to nearest Government bank does not have impact to determine whether a household get loan from formal source. We found no correlation between the household borrowing activities and distance to BARD or other Government banks or programmes. This suggests an uniform access to formal credit at the commune level in rural Vietnam. A number of control variables are found to affect participation. Number of members in a household is found to have a large and significant affect on credit participation, especially from formal source. Farm work is also significant. Household is more likely to borrow if head of household is working in agriculture or self-business. Amount of loan from formal source is higher with farm work too. The demand of labor during farm work could be the reasons why larger number of household members affect probability of getting loans. Without hiring other people, small family does not have motivation and capacity to expand the family business conducing to a less participation. Among other results, there is no strong evidence of women participation in credit activities, as borrowing probability is significantly determined by man as head of household. Education level is not important and age of head of household has a negative affect. Having a closer look at the correlation and distribution of education and borrowing activities we found that households having its head s education level at primary and lower secondary school are main borrowers. Households heads who have university degree or never attended school do not borrow much. This U-shape relationship could be because higher 8

10 education may help head of households easier to find a paid job and then do not have to work in farm or self-business. Vice versa for those who never attended school, they just simply lack of skills and abilities. In sum, the determinants of taking credit of a household are quite complicated. While farm work and number of members of household show a large and significant affect, value of house and land holding size display a small but significant impact on household borrowing activities. It is interesting to find out that higher education does not determine credit participation and credit access is uniform at commune level in term of distance to formal credit institutions. 9

11 Table 5: Probit and Tobit results (n=4289) Independent variables Probit household borrowing Probit Borrowing from formal source Tobit Loan amount Tobit Loan amount from Gov banks Gender - household head (1:Male; 0:Female).070 (.050).047 (.055) ( ) ( ) Age household head -.014*** -.006*** *** *** (.001) (.001) (14.482) (17.219) Household size.096***.104*** *** (.010) (.011) (97.206) ( ) Education level schooling years of household head (.005).007 (0.005) (48.96) (57.474) Agricultural work (1=Yes; 0=No).153***.2323*** *** (.045) (.049) (404.96) ( ) House value * *** 8.534*** (.0003) (.0006) (2.742) (3.100) Land holding size **.00001***.159***.233*** (0.000) (.000) (.030) (.033) Distance to Government banks *** (.007) (.008) (70.084) (79.591) Distance to BARD (.007) (.008) (65.026) (72.586) Pseudo R Notes: Numbers in parentheses are standard errors. Significance level 10%, 5% and 1% are marked by *, ** and *** Source: VLSS 1997/98 10

12 4. IMPACTS EVALUATION 5.1. EMPIRICAL STRATEGY To measure the impact of credit participation on household consumption, ideally, we would like to compare consumption/expenditure between two groups of identical households who differ only in borrowing activities, one takes a loan (treatment group) and the other don t (control group). However, when we look at the borrowing amount of a household, we actually have the information from both sides of demand and supply. The credit demand theory tells us that households characteristics would determine the credit demand of households but problem is lending institutions also base on these to decide credit supply. Our estimation is likely to give bias results if credit participation are correlated with other unobserved variables, which also affect household consumption. For example, more productive households may have higher demand for capital and then more likely to participate in a credit programme but at the same time, productive households may consume more. That is demand for credit and consumption are jointly determined by household and commune attributes. In general, to obtain un-biased estimations studies of impact evaluation using non-experimental data have to carefully address these problems of endogeneity. The potential endogenous biases include nonrandom placement of credit programme, unmeasured village attributes, and unmeasured household characteristics (selfselection). Our impact evaluation strategy will first run a simple OLS regression of the household consumption on the household independent control variables such as age of head of household, gender of head, education level of head, employment, house value and land holding size to provide a starting point. The results could be biased as we have not included any control for endogenous variables. In our discussion above, we show that communes in rural Vietnam have uniform access to formal credit which means the problem of non-random placement of credit programme could be relaxed. Formal institutions are available to households in all commune samples. To address endogenous unmeasured village attributes which may affect both credit demand of households in the village and household consumption, we apply village-level fixed-effect estimation. The regression equation is written as: Y ij = β o + β 1 X ij + γc ij + μ j + ε ij where Y ij indicates the outcome of interest - consumption of household i of village j, C ij denotes household's receipt of credit, X ij are the observed household and village characteristics, μ j is unobserved village-level fixed-effect that determines household credit and ε ij are the error terms. The estimation results are shown in table 6 & 7. 11

13 Although the fixed-effect regression have been able to control for the unobserved village characteristics that could affect households financial decision, it still uses general non-participating households as control group without a appropriate selection. This bases on a strong assumption that participating and non-participating households would have made the same change in consumption if both of them do not take credit. By this assumption, the fixed-effect regression seems not to rule out selection biases. If participating households expected to have a larger (smaller) change in consumption comparing to non-participating households without credit programmes then the fixed-effect estimation would be biased upward (downward). In other words, simply using non-participating households as control group will not be a solution to address selection biases. To identify the credit effect more accurately, we need a control group of non-participating households that would be more similar to the group of participants had the credit program not been available. In order to identify such a control group, the best strategy is to find out exogenous eligible conditions used by lenders in selecting a borrower. These exogenous requirements will help to define who among non-participating households are comparable with participants. Pitt and Khandker (1998), Morduch (1998), Khandker (2003) use the Grameen Bank s eligible requirement of maximum landholding of 0.5 acre to be such a convincing instrument. However, we do not found that sort of exogenous eligibility in our data currently. Instead, we apply the recently developed identification technique: the propensity score matching. Following Rosenbaum and Rubin (1983), Heckman et al (1998), Dehejia and Wahba (2002), the propensity score matching methods pair each programme participant with a single non-participant, where pairs are chosen base on the degree of similarity in the estimated probabilities of participating in a programme. The average impact of the programme is then estimated by the mean difference in outcomes of the matched pairs. In our estimation, we first perform a Probit regression to obtain participation probability of each household in the sample. Households characteristics which is analysed in section 4 will be used. We acknowledge that the bigger set of household and commune control variables, the better estimation we obtain. However, at this preliminary state only main variables are applied. We then match the participation probability of each observation in participant group to the observation in the non-participant group to find the best match. Each matching pairs will be separated into treatment and control group. The impact estimation of credit taking using score matching methods will be reported in Table RESULTS AND DISCUSSION The results display in table 6 are coefficients from OLS and FE regression of household consumption on credit taking, and households and commune characteristics. The impact is reported positive and significant. To control the problem of contemporaneous correlation and causation we apply 12

14 the model on panel data. We run the regression of consumption in VLSS1997/98 on the financial activities in VLSS1992/93. The estimation shows that borrowing in 1992/93 will positively improve the consumption in 1997/98. When we control for financial source, the impact of informal credit is also positive. Although informal money lenders provide capital at interest rate 4 times higher than formal source and with shorter duration (Table 2), small-amount, short-term and high-interest-rate informal loans to some extent help poor borrowers to smoothing their consumption and stabilizing production over idiosyncratic shocks. The regression also reveals the higher and better impact of formal credit on consumption comparing to impact of borrowing from informal source. Providing loan with longer duration, lower interest rate and bigger amount, formal institutions are the main source of external capital for household s investment which is likely to improve household s economic condition. Other household characteristics like household size and education of household head also have a significant impact on consumptions. The agricultural work coefficient is negative because our consumption measurement does not include self-production consumption. The commune-level fixed-effect regression which is controlling for potential biases arise from commune characteristics also reports consistent results with OLS regression. The coefficients of impact are all found to be significant at level of 1 percent. Our results of credit impact estimation using propensity score matching method report in Table 8 & 9. The pool of samples for this estimation is dropped from 4528 to 4289 observations in table 8 due to condition of matching between participating household and non-participating household. The table 8, which is running a cross-section propensity matching on VLSS1997/98 data, and the table 9, which is running a matching estimation on VLSS1992/93 and comparing consumption of treated and control group in VLSS 1997/988, shows the average treatment on the treated. The results are different from the OLS and FE regressions. While formal credit still have positive and significant impacts on household s consumption, the borrowing activities in total and the informal credit shows a negative impacts. Annually, formal finance borrowers in 1997/98 would consume 280,000 VND more than non-borrower, and borrowers in 1992/93 would be 790,000 VND higher. 5. CONCLUSION AND FURTHER WORK In general, our findings are consistent with other results from researches in the field. There are two main results. First, there exists a uniform access to credit at commune level in rural Vietnam. Households' financial activity is found to be more determined by household size and agricultural work than education or distance from commune to the nearest bank. Second, we find a positive and significant impacts of formal credit on household consumption. The impact from informal credit is 13

15 mixed and need further study to have conclusion. The positive result of formal credit is confirming the important role of financial reform and restructuring rural credit market in Vietnam. In further research, we would like to explore the panel data analysis, taking more control of loan characteristics. The impact of credit should be measured in more outcomes like household productions, off-farm work or child s education. 14

16 Table 6: OLS and fixed-effect regression VLSS1997/98 Independent variables Credit Impact coeff. OLS (n= 4528) Commune FE (com. n=158) Total loan Formal Loan Informal loan Total loan Formal loan Informal loan.164***.258***.141***.133***.211***.045** (.014) (.02) (.013) (.014) (.02) (.02) Gender - household head Age household head -5.1 (242.03) *** (7.15) (241.9) *** (7.58) (227.51) 57.75*** (7.21) (253.5) 48.51*** (7.75) ( ) *** (7.76) (255.9) 47.9*** Household size *** *** *** *** *** *** (52.06) (52.48) (51.042) (50.81) (53.51) (57.23) Education level sch. years of HH head *** (26.52) *** (26.41) *** (27.129) *** (26.98) *** (26.12) *** (113.6) Agricultural work *** *** *** *** *** *** (1=Yes; 0=No) (218.05) (216.64) (241.3) (240.27) (221.5) (267.8) House value *** *** 38.49*** *** *** 39.38*** (1.588) (1.576) (1.56) (1.55) (1.6) (1.64) Land holding size.106***.103***.096***.1***.1***.12*** (.017) (.017) (.018) (.018) (.019) (.02) _cons R Numbers in parentheses are standard errors. Significance level 10%, 5% and 1% are marked by *, ** and ***. FF stands for Formal Finance, IF for Informal Finance, LA for Loan Amount, FA for Formal Loan Amount, IFA for Informal Loan Amount (7.8) 15

17 Table 7: OLS and FE regression results The impact of 1992/93 borrowing on 1997/98 household consumption Independent variables OLS Commune FE (n= 2916) (com. n=158) Total loan Formal Loan Informal loan Total loan Formal loan Informal loan 1992/93 borrowing.36***.76***.029***.29***.48***.025*** Impact coefficience (0.04) (.1) (.04) (.041) (.11) (.045) Gender - household head (294.2) (295.22) (295.8) (276.9) (278.21) (277.7) Age household head 40.1*** 42.1*** 41.2*** 42.4*** 43.6*** 43.5*** (8.7) (8.8) (8.8) (8.38) (8.4) (8.4) Household size *** *** *** *** *** *** (63.5) (63.41) (63.8) (62.5) (62.6) (62.7) Education level *** 804.9*** 853.8*** *** *** *** number of sch. years of (125.47) (125.9) (126.2) (124.5) (125.23) (124.8) HH head Agricultural work (1=Yes; 0=No) *** (263.97) *** (264.6) *** (265.1) *** (280.7) *** (282.05) *** (281.6) House value 97.5*** 99.5*** 98.18*** 84.3*** 86.1*** 84.7*** (29.78) (3.27) (3.29) (3.5) (3.5) (3.51) Land holding size.11***.11*** 0.12*** 0.11***.11***.12*** (.021) (.021) (.021) (.022) (.023) (.023) _cons R

18 Table 8: Credit participation impact on household consumption, Propensity score matching estimation (n=4289) Mean Consumption Credit participant Controls Difference Credit taking Formal source only Informal source only ** ( ) ** ( ) *** ( ) Numbers in parentheses are standard errors. Significance level 1%, 5% is marked by ***, and ** Table 9: Credit impact on household consumption Propensity score matching esimation of VLSS92/93 financial activities and VLSS97/98 consumption Mean Consumption Credit participant Controls Difference Credit taking Formal source only Informal source only * ( ) ( ) ** (413.84) Numbers in parentheses are standard errors. Significance level 1%, 5%, and 10% is marked by ***, ** and * 17

19 BIBLIOGRAPHY Aghion, B. A. D. and Morduch, J.: 2000, Microfinance beyond group lending, Economics of Transition 8, Aghion, B. A. D. and Morduch, J, 2005, The economics of microfinance, The MIT Press Conning J. and C. Udry, 2005, Rural financial markets in developing countries, Discussion paper No. 914, Yale University Dehejia R and S. Wahba, 2002, Propensity score matching methods for nonexperimental causal studies, Review of Economics and Statistics 84(1), pp Gine X., 2004, Accessing to capital in rural Thailand: an estimated model of formal vs. informal credit, The World Bank working paper Heckman J., Ichimura H., and P. Todd, 1998, Matching as an econometric evaluation estimator, Review of Economic Studies 65(2), pp Hoff, K. and Stiglitz, J. E., 1990, Imperfect information and rural credit markets: Puzzles and policy perspectives, The World Bank Economic Review 4, Khandker, R. and Faruqee, R., 2003, The impact of farm credit in Pakistan, Agricultural Economics 28, Khandker, S., 2001, Does microfinance really benefit the poor? Evidences from Bangladesh, Asia and Pacific forum on poverty, Asian Development Bank. McCarty, A., 2001, Microfinance in Vietnam: A survey of schemes and issues, DFID and State Bank of Vietnam Report, Mekong Economics McKinnon R., 1973, Money and Capital in economic development, Brookings Institution, Washington DC Morduch, J., 1998, Does microfinance really help the poor? new evidence from flagship programmes in Bangladesh, Working paper Morduch, J., 1999, Microfinance promise, Journal of Economic Literature Vol. XXXVII. Pitt, M. and Khandker, S., 1998, The impact of group-based credit programs on poor households in Bangladesh: Does the gender of participants matter?, Journal of Political Economy 106, Smith J. and P. Todd, 2004, Does matching overcome LaLonde s critique of nonexperimental estimators? Journal of Econometrics, 125(1-2), pp Rosenbaum P. and D. Rubin, 1983, The central role of the propensity score in observational studies for causal effects, Biometrika 70, pp