Effects of Microfinance on Poverty Reduction In Vietnam: A Pseudo-Panel Data Analysis

Transcription

1 Journal of Accounting, Finance and Economics Vol. 4. No. 2. December Pp Effects of Microfinance on Poverty Reduction In Vietnam: A Pseudo-Panel Data Analysis Hoai An Duong 1 and Hong Son Nghiem 2 The development of microfinance in Vietnam since 1990s has coincided with a remarkable progress in poverty reduction. Numerous descriptive studies have illustrated that microfinance is an effective tool to eradicate poverty in Vietnam but evidence from quantitative studies is mixed. This study contributes to the literature by providing new evidence on the impact of microfinance to poverty reduction in Vietnam using the repeated crosssectional data from the Vietnam Living Standards Survey (VLSS) during period Our results show that micro-loans contribute significantly to household consumption. JEL Classification: D13, D14 and O12 1. Introduction Microfinance has been recognised as a potentially effective tool to fight against poverty. The spread of microfinance appears to coincide with a sharp decrease in poverty rate across countries. However, empirical studies have not reached a consensus about the extent to which microfinance contributes to poverty reduction. For example, one of the most widely cited studies Pitt and Khandker (1998) suggested that microfinance has a significant and positive contribution to poverty reduction in Bangladesh whilst Roodman and Morduch (2009) found insignificant effects, using the same data set. In Vietnam, the story is similar. Pham and Izumida (2002), Quach, Mullineux and Murinde (2005), and Doan, Gibson and Holmes (2014) have found that microfinance created positive impacts on poverty reduction while Nghiem, Coelli and Rao (2012) argued that the effect of microfinance on poverty reduction was negligible. Thus, more concrete evidence on impacts of microfinance, especially using wider ranges of data and/or methodologies, is needed. This study examines the effects of microfinance on poverty reduction, using data sets from the Vietnam Living Standard Surveys (VLSS) from 1992 to To the best of our knowledge, no previous study has investigated the development and effects of microfinance for such a long time frame. We were also able to mitigate the self-selection issue in microfinance participation using a fixed-effects estimator with a pseudo panel data set, and a mixed-process seemingly unrelated (SUR) estimator. Thus, this study will expand the stock of knowledge on microfinance impacts, especially in the context of Vietnam. In particular, most previous study in Vietnam only focused on shorter time frames such as of Quach, Mullineux and Murinde (2005), or of Nguyen, Nordman and Roubaud (2013) but no study investigated trends of microfinance impacts over the period of Our results show that access to microfinance is associated with an increase of income per adult equivalent of 42 per cent and consumption per adult equivalent of almost 100 per cent. We also found that loan volumes matter: households with large loans are associated 1 PhD student in Griffith University Australia. hoaian.duong@griffithuni.edu.au 2 Research fellow, National Centre for Disability and Rehabilitation Medicine (CONROD), the University of Queensland Australia. h.nghiem@uq.edu.au

2 with a higher level of consumption and income. However, we did not find significant effects of access to microfinance and reduction of poverty incidence. The remainder of this paper is organised as follow. Section 2 reviews the literature on microfinance impact studies. Methodologies and data are presented in Section 3 whilst results and discussions are presented in Section 4 and Section 5 concludes. 2. Literature Review This section reviews studies of microfinance impacts in Vietnam. We classify their study according to the methods used to address the selection problems: panel data, quasiexperimental surveys and randomised experiments. 2.1 Panel Data Quach, Mullineux and Murinde (2005) used data from the Vietnam Living Standard Survey (VLSS) data in 1993 and 1998, in which more than one thousand households were sampled repeatedly in both periods. The author applied a probit regression to estimate determinants of credit and the Heckman two-step method to estimate the impact of credit on household welfare. The findings showed that access to formal credit has a positive impact on consumption per capita but the magnitude of this impact is modest. For example, one per cent increase in the volume of credit, ceteris paribus, led to 0.07 and 0.06 per cent increase in consumption per capita in 1993 and 1998, respectively. Lensink and Pham (2012) used panel data with a sample of about 3,200 households, obtained from VLSS 2002 and 2004 to evaluate the impact of microcredit provided by VBSP on self-employment profits in Vietnam. The findings indicated that microfinance had positive and significant impacts on self-employment profits of the borrowers. More importantly, microfinance had positive impacts on poverty reduction and these impacts were more significant for the poorest households. In addition, the authors did not find direct impacts of credit access on investments. Despite the achievement, some limitations still can be found in this study. First, the study only focused on VBSP one of the current microfinance service providers in the nation. Second, the research also focused only on rural areas while VLSS data provide information of households in urban areas as well. Finally, although VLSS data provide a wide range of information of household characteristics, the data were not designed for specific research purposes like microfinance assessment. In short, panel data are very powerful for producing reliable estimates. However previous studies only took a short panel that contain only two periods, and hence, interesting questions such as the dynamic interrelationship between access to microfinance and livelihood of the poor could not be answered. 2.2 Quasi-Experimental Surveys Swain, Nguyen and Vo (2008) employed the 2006 household survey data to estimate the impact of microfinance on poverty reduction and compare the different impacts between three groups in the Mekong Delta of Vietnam. The survey targeted three groups covering 134 households in the research area: successful members (those who successfully 59

3 escaped from poverty at the time of the research), poor members (members who remained poor), and non-members (who did not join in any microfinance program). In order to obtain in-depth data, the researchers used participatory rural appraisal (PRA) techniques: focus group discussions and in-depth interviews. The findings showed that average income per capita of the member households was up to three times higher than that of non-members. The authors also found that members had accumulated more assets and better quality assets than non-members. Member households also had opportunities to access training programs and improve their social positions. The relatively small sample size of this study may limit its representation to microfinance operations in the Mekong Delta region. Doan, Gibson and Holmes (2014) conducted a study to evaluate the impacts of household credit on education and health care spending by the poor in peri-urban areas of Ho Chi Minh City. The study employed two major methods namely propensity score matching and multiple treatment effects. The sample included 411 member and non-member households. The findings showed that formal credit had positive impacts on education and health care expenditures but impacts of informal credit were insignificant. For example, on average, member households had higher levels of education and health care consumption than non-member did by roughly 90 thousand VND (about $US 4.5) per month. We believe that effects of microfinance on education and health often occur at longer time frame than that covered in this study. The advantage of a quasi-experimental approach is that it allows researchers to apply many techniques to avoid or mitigate self-selection and program placement biases in microfinance impact assessment. The downside of this approach is that conducting a quasi-experiment is time consuming and costly. 2.3 Randomised Experiments Despite much effort spent, we did not find any research using randomised experiments to evaluate impacts of microfinance in Vietnam. Thus, we have to review studies using this approach in other countries. Karlan, Harigaya and Nadel (2006) conducted a randomized control study in Peru to estimate the impacts of a microfinance package including a business-training program and credit programs, by assigning the programs in three groups: mandatory treatment, optional treatment, and control. The sample size of these three groups was 183 banks. The results showed that clients in treatment groups had better business knowledge and business practices. Also, the business-training program had strong and positive impacts on business results among the treatment groups. For example, sales of participating businesses were 16 per cent to 28 per cent higher than those of control groups. The authors suggested some improvements such as venue of the training should be more convenient for most of the clients, and hired professionals to train the trainees should be taken into account to attract the attention of the trainees. One criticism of the program is that the program placement may not be strictly random: it targets experienced businesswomen, and thus results are more likely to be positive. Banerjee et al. (2013) conducted a randomized study to evaluate the impacts of introducing microcredit in Hyderabad, India with Spadana the largest microfinance 60

4 institution in the country at that time. The study selected 2,800 households from 120 areas of the organisation at the baseline survey. The design followed three steps: small pilot, full pilot, and full launch. They found that micro loans are associated with increased profits but they found no influence of the program on health, education and women s decision-making empowerment. Overall, randomised experiments provide clean estimates of microfinance impacts but the time and resource constraints of this study prohibit us from following this path. 3. Methodology and Data 3.1 Methodologies The main challenge of estimating impacts of microfinance is selection bias. Individuals who decide to participate in microfinance programs may have some unobserved characteristics (e.g., risk attitude, entrepreneurial ability) that may also affect impacts of microfinance on indicators, such as income or consumption. In this study, we exploit the availability of seven repeated cross-sectional surveys to address the selection bias issue using a pseudo panel data approach a second best option when genuine panel data are not available. Tracking individuals overtime is time and resource consuming, hence panel data on development projects are often rare (Deaton 1985), (Verbeek 2008). But with the availability of repeated cross-sectional surveys one can create pseudo panel data that allow analysis with panel data methods. Pseudo panel data have an advantage that researchers can combine data from different sources into a single data set, as long as the cohorts in each source are comparable (Deaton 1985). Attrition problems often found in genuine panel data can be minimised as long as other participants in the same cohort remained in the survey. The characteristics of cohort averages can soften impacts of response errors. In order to construct pseudo panel data, particular individual characteristics that are steady over time (e.g., ethnicity, year of birth, and residential locations) should be used to define cohorts. The average values of interested variables by these characteristics are used as observations of the pseudo panel data. In this study, we apply the fixed-effects estimator to control for unobserved effects at provincial level, which is equivalent to constructing a pseudo-panel using provincial average but it has the advantage of keeping the original number of observations. Also, by using provincial fixed-effects we can examine effects of individual time-invariant exogenous characteristics such as gender, ethnicity and educational background. We use the following equation to measure the impacts of participation in microfinance on welfare indicators of households. Where Y is a set of interested outcomes such as income and consumption; MF is the binary variable, representing the access to microfinance (proxied by a dummy variable for loan volume less than the average loan volume of MFI in 2010, which is VND 4.2 million 61

5 and borrowed from concession sources such as credit cooperatives, poverty alleviation programs and subsidized loans from government banks such as Vietnam Bank for the Poor); L represents the slope effect of access to credit (proxied by loan volumes); X is a vector of household and community characteristics (e.g., household size and access to electricity), represent is the set of variables that capture time-invariant unobserved characteristics of individuals; and is random error. We expect that parameters 1 and 2 are positive as participation in microfinance and larger value of loans would be associated with higher income or consumption. The parameter 1 represents the contribution of the access to microfinance to selected outcomes. Meanwhile, parameter 2 designates the impact of loan volume to selected outcomes. It is expected that the access to microfinance services will improve the members welfare. We also expected that parameter 3, which represents the effects of loans to microfinance participants to have opposite signs than with 1 and 2 due to the law of diminishing marginal returns. Parameter 4 represents the relationships between selected household characteristics and welfare indicators. Parameter captures unobservable household characteristics, which can affect both the decision to join microfinance programs and outcomes of interest (e.g., income and consumption). With the availability of panel (or pseudo-panel) data as in this study, unobserved individual characteristics can be eliminated using fixed-effect regressions. For sensitivity analysis, we apply a mixed-process, seemingly unrelated estimator (SUR) proposed by Roodmand (2014). In particular, this estimator can mitigate effects of selection bias by estimating both the microfinance impact in Equation 1 and the microfinance participation equation: Where Z is a set of exogenous variables, which can include part of set X from Equation 1, is the error term, and other notations are defined as previously. The selection bias (if any) will be represented by the significance of the correlation coefficient of the error terms in equations 1 and 2 as both include the individual unobservable characteristics. The significance of this correlation coefficient can be used as a test for selection bias in Equation 1. The main advantage of this estimator compared with the traditional SUR is that the impact equation and participant equation can have different formats (e.g., continuous, binary or categorical). 3.2 Data The data used in this study are drawn from the Vietnam Living Standard Survey (VLSS) carried out in 1992, 1998, 2002, 2004, 2006, 2008 and 2010 respectively. The VLSS is a general-purpose survey series, conducted by the General Statistics Office (GSO) of Vietnam with financial and technical assistance from the United Nations Development Programme (UNDP) and the World Bank. The surveys cover a variety of characteristics of households such as demographic information, income, consumption, expenditure, employment, health, migration, credit and insurance. (2) 62

6 The sample size of the VLSS 1992 covered 4,799 households in 300 villages, of which 3,839 households (80 per cent) were from rural areas. The VLSS 1998 sampled 5,900 households, including most of those surveyed in 1992 but covered wider areas with a further 388 villages. To make data more representative, the GSO increased the sample size of the VLSS to 29,532 households in 2002 but this attempt was so costly that the sample size was reduced to approximately 9,189 households in subsequent surveys (i.e., 2004, 2006, 2008 and 2010). However, since the VLSS are general-purpose surveys, questionnaires about financial information (e.g., loans and savings) were only conducted with smaller set of sample (e.g., 1000 households in 2002 and 2010). Thus, our final data set include only 15,000 households with valid credit information. Table 1: Descriptive statistics Variables Mean Std. Min Max Age of household head (years) Gender of household head (male=1) Location of household (urban=1) School years of household head Ethnicity (Vietnamese/Chinese=1) Household size (persons) Income/adult equivalent (VND1,000/year) Consumption/adult equivalent (VND 1000/year) Poverty status ($1.25 PPP/day) Poverty status (National poverty line) Loan volume (VND1,000) Microfinance access (1=yes) Savings (VND1,000) Assets (VND1,000) Durables goods (VND1,000) (Calculated from VLSS Constant prices at 2004 are used) Despite the sufficiently large sample, the VLSS data are far from ideal for this study. The surveys were designed for general purposes, which do not include detailed information on issues such as characteristics of microfinance institutions. Also, only some proportion of the VLSS in , periods were surveyed repeatedly, thus the data series can be considered as repeated cross-sectional (the 2010 survey did not repeat the sample from the previous survey). The descriptive statistics show that only 24 per cent of households surveyed live in urban areas, reflecting the population distribution in the census. In addition, 73 per cent of households surveyed were headed by a male with an average age of 47 years. On average, household heads spent 7 years at school. Also, 86 per cent of households were Vietnamese or Chinese, which are generally better-off than other ethnic groups (Baulch et al. 2007). Households in Vietnam in the study period earned 13.7 million VND per adultequivalent per year, on average (or $US2.4/day). The average consumption per capita was slightly lower, at 10.5 million VND per adult-equivalent per year (US$2.0/day). Note that the original data consisted of households with minimum income far beyond subsistent level (e.g., VND1000 or $US0.2/year). We believe that such figures were reporting errors, and hence, we excluded those households from the analysis. On average, 16 per cent of 63

7 households in the sample access to microfinance. The poverty incidence, using the national and international poverty lines, was 29 and 17 per cent, respectively. The international poverty line was calculated using a GDP deflator and exchange rate information from the Penn World Table version 7.1 by Heston and Summers (2002) and (2012). 4. Results and Discussions The fixed-effects results show that both the value of loans and the access to microfinance have positive impacts on income and consumption per adult equivalent. Since we take the natural logarithm on both side of the equation to mitigate effects of their skewed distribution, parameters of continuous variables are interpreted as elasticity. In particular, the results show that a one per cent increase in loan volume lead to 0.15 per cent increase in income and 0.23 per cent increase in consumption per adult-equivalent (see Table 2). In addition, compared to other households, microfinance members had their income per adult-equivalent higher by 41 per cent (i.e., e ) and their consumption higher by 99 per cent (i.e., e ). This evidence suggests that consumption smoothing may be a stronger impact of microfinance, compared to income generating. But we believe that the low consumption level (about $700 per adult equivalent per year) of households in the sample would make a substantial increase in percentage possible (i.e., credit input follows the law of diminishing marginal return in the household production function). The interactions of microfinance and loan volume for income and consumption confirm the diminishing marginal return effects as they are negative and significant. Positive effects of most other exogenous parameters on income and consumption are also as expected: Vietnamese and Chinese households have higher income and consumption than that of other ethnic groups. Similarly, households that reside in urban areas have higher income and consumption than rural dwellers. The increasing magnitude of year dummies also suggest that the living standard (proxied by real income and consumption) improved over time during the study period (The first survey year in 1992 was selected as the reference period). Some exceptions included 1998 where the rate of poverty reduction was not significantly different from the reference year; 2002 where real income per adult equivalent was lower than that of 1992 by 4.7 per cent (i.e., e ); and 2010 where the magnitude of poverty reduction was lower (in absolute value) than that of

8 Table 2: Impacts of microfinance (pseudo-panel fixed effects) Variables Income Consumption Poverty status Microfinance access (1=yes) ***0.347 *** Log of loan volume ***0.149 ***0.231 *** Loans volumes * microfinance *** *** Gender of household head (1=male) *** Age of household head (log) ***0.237 ***0.246 *** School years of household heads ***0.035 ***0.037 *** Ethnicity (Vietnamese/Chinese=1) ***0.193 ***0.197 *** Residential location (1=urban) ***0.272 ***0.170 *** Household size (log) ***0.100 ***0.060 ***0.216 Year dummy for 1998 ***0.081 *** Year dummy for 2002 ** ***0.578 *** Year dummy for 2004 ***0.378 ***1.096 *** Year dummy for 2006 ***0.519 ***1.258 *** Year dummy for 2008 ***0.672 ***1.335 *** Year dummy for 2010 ***1.019 ***1.489 *** Constant ***5.728 ***4.464 N Adj. R * p<0.1, ** p<0.05, *** p<0.01. Marginal effects are reported for the last column. Robust standard errors are not reported for brevity. The results, however, suggest that access to microfinance create no significant effects on the poverty status of households, using the international poverty line of $1.25 PPP/person/day. Similar results are found when the national poverty line is applied but we did not report them here for brevity. This result may sound counter intuitive and in contradiction with previous studies such as Quach, Mullineux and Murinde (2005) but they did not control for the trend via year dummies as we applied in this study. Thus, one explanation is that effects of poverty reduction may be outweighed by the growth of the economy, proxied by year dummies. However, loan volume is associated significantly with the reduction in the probability of being poor. Since parameters of probit regressions are difficult to interpret, we report marginal effects for this result. Since values of some continuous variables are log-transformations, the marginal effects for these parameters can be interpreted using e-based (2.7) instead of using unit based (1.0). For example, an additional unit of VND 2,700 increase in loan volume is associated with 0.11 percentage point reduction in the probability of being poor. But the additional effect of loans for microfinance to households (i.e., the interaction term) is insignificant. The negative sign of most other parameters is as expected. For example, advantaged households (urban residents, belonging to Vietnamese or Chinese ethnic groups, and who spent more time at school) are significantly less likely to be poor. For stativity analysis, we also performed a mixed process seemingly related (SUR) estimator and the results were almost the same as that produced by the fixed-effect estimator (see Table 3). In particular, we also found that participation in microfinance is associated with an increase in income and consumption but its impact on poverty status is not significant. The correlation coefficient ( ) of the error terms in equations 1 and 2 is significant for income and consumption but not significant on poverty status. Thus, using 65

9 standard regressions to estimate effects of microfinance on income and consumption may produce biased results. Table 3: Impacts of microfinance (mixed process SUR estimator) Variables Income Consumption Poverty status Microfinance access (1=yes) ***0.395 *** Loan volume (log) ***0.176 ***0.245 *** Loans volumes microfinance ** Gender of household head (1=male) Age of household head (log) ***0.235 ***0.244 *** School years of household heads ***0.036 ***0.037 *** Ethnicity (Vietnamese/Chinese=1) ***0.191 ***0.196 *** Residential location (1=urban) ***0.273 ***0.171 *** Household size (log) ***0.100 ***0.060 ***0.142 Year dummy for 1998 ***0.077 *** Year dummy for 2002 *** ***0.467 ** Year dummy for 2004 ***0.404 ***1.110 *** Year dummy for 2006 ***0.560 ***1.279 *** Year dummy for 2008 ***0.715 ***1.358 *** Year dummy for 2010 ***1.040 ***1.500 *** Constant ***5.416 ***4.301 N *** *** * p<0.1, ** p<0.05, *** p<0.01. Marginal effects are reported for the last column. Robust standard errors are not reported for brevity. 5. Concluding Remarks This paper has examined impacts of microfinance on poverty reduction in Vietnam, using the data from the Vietnam Living Standard Survey series from Our results suggest that microfinance produces positive and significant contributions to household income and consumption. However, microfinance participation creates no significant impact on the poverty status of households. Our results are robust in the choices of estimate in order to address the selection bias in microfinance due to unobserved individual effects. These results are in line with previous studies such as Pitt and Khandker (1998), Quach, Mullineux and Murinde (2005) and Khandker and Samad (2014) that microfinance creates positive effects for the poor. We extend the knowledge of microfinance impacts in Vietnam by exploring the longest possible time frame ( ) and applied relevant econometric methods to address the selection bias issue. Also, unlike previous studies on microfinance in Vietnam, we used the international poverty line, which make our results compare easily with international studies. Based on this result we suggest that microfinance provides a useful tool for poor households to smooth-out consumption during critical periods and improve income. However, to lift the poor out of poverty on a sustainable basis, Vietnamese authorities would seek to coordinate the provision of microfinance with other factors such as job creation, infrastructure development and agricultural extension. The main limitation of this study is that the data were drawn from general-purpose surveys, hence the finance 66

10 questionnaire only covers a small proportion of the sample. In addition, we did not have a genuine panel data in this study; thus, standard panel methods could not be applied. References Banerjee, A, Esther, D, Rachel, G & Cynthia, K, 2013, The Miracle of Microfinance? Evidence from a Randomized Evaluation, The MIT Department of Economics, Working paper, no , pp Baulch, B, Truong, TKC, Dominique, H & Jonathan, H, 2007, Ethnic Minority Development in Vietnam, The Journal of Development Studies, vol. 43, no. 7, pp Deaton, A, 1985, Panel Data from Time Series of Cross-Sections, Journal of Econometrics, vol. 30, pp Doan, T, John, G & Mark, H, 2014, Impact of Household Credit on Education and Healthcare Spending by the Poor in Peri-Urban Areas, Vietnam, Journal of Southeast Asian Economies (JSEAE), vol. 31, no. 1, pp Heston, A, Summers, R & Aten, B, 2012, Penn World Table Version 7.1 Center of Comparisons of Production, Income and Prices, the University of Pennsylvania. Heston, A, Summers, R & Aten, B, 2002, Penn World Table, Center for International Comparisons, the University of Pennsylvania. Karlan, D, Tomoko, H & Sara, N, 2006, Evaluating Microfinance Program Innovation with Randomized Controlled Trials: Examples from Business Training and Group Versus Individual Liability, Global Microcredit Summit 2006, pp Khandker, SR & Hussain, AS, 2014, Dynamic Effects of Microcredit in Bangladesh, The World Bank, Development Research Group, Working Paper, no. 6821, pp Lensink, R & Pham, TTT, 2012, The Impact of Microcredit on Self-Employment Profits in Vietnam, Economics of transition, vol. 20, pp Nghiem, S, Tim, C & Prasada, R, 2012, Assessing the Welfare Effects of Microfinance in Vietnam: Empirical Results from a Quasi-Experimental Survey, Journal of Development studies, vol. 48, no. 5, pp Nguyen, HC, Christophe, JN & François, R, 2013, Who Suffers the Penalty?: A Panel Data Analysis of Earnings Gaps in Vietnam, Journal of Development Studies, vol. 49, no. 12, ppl Pham, BD & Yoichi, I, 2002, Rural Development Finance in Vietnam: A Microeconometric Analysis of Household Surveys, World Development, vol. 30, no. 2, pp Pitt, MM & Shahidur, RK, 1998, The Impact of Group-Based Credit Programs on Poor Households in Bangladesh: Does the Gender of Participants Matter?, Journal of Political Economy, vol. 106, pp Quach, M, Mullineux, A & Murinde, V, 2005, Access to Credit and Household Poverty Reduction in Rural Vietnam: A Cross-Sectional Study, The Birmingham Business School, The University of Birmingham Edgbaston, pp Roodman, D & Jonathan, M, 2009, The Impact of Microcredit on the Poor in Bangladesh: Revisiting the Evidence, Center for Global Development, Working Paper, no. 174, pp Roodman, D, 2014, Cmp: Stata Module to Implement Conditional (Recursive) Mixed Process Estimator, Statistical software components. Swain, RB, Nguyen, VS & Vo, VT, 2008, Microfinance and Poverty Reduction in the Mekong Delta in Vietnam, African and Asian Studies, vol. 7, no. 2&3, pp Verbeek, M, 2008, Pseudo-Panels and Repeated Cross-Sections, The Econometrics of Panel Data: Springer, pp