81036 Bringing the Bank to the Doorstep: Does Financial Education Influence Savings Behavior among the Poor? Evidence from a Randomized Financial Literacy Program in India 1 Leopold Sarr 2 Santadarshan Sadhu 3 Nathan Fiala 4 November 2012 Draft results. Please do not cite or quote. Abstract One of the obstacles to the use of branchless banking has been the low level of familiarity and trust with the technology behind electronic cards and mobile phone banking among the poor. As a result, the banking correspondent (BC), or “doorstep banking” model was introduced in India to bring basic banking services to rural people. Clients of BC programs include mainly house- holds with very low incomes and poor access to the formal banking system. This paper explores the uptake of branchless banking in one of the largest BC programs in the world, FINO, which 1 Acknowledgements: We thank Tarun Agarwal and Prakash Lal of FINO for their support in implementing the fi- nancial literacy training and giving us access to their client transaction database. We gratefully acknowledge funding from the World Bank’s Russian Evaluation Trust Fund. Anup Roy, Sitaram Mukherjee, and Mudita Tiwari provid- ed excellent field coordination through the Centre for Micro Finance (CMF). Ajay Tannirkulam of CMF has provid- ed valuable inputs for designing the randomization strategy. We also would like to acknowledge Bilal Zia for his invaluable inputs, and Christian Salas Pauliac, consultant at the World Bank for his excellent research assistance. All findings and interpretations in this paper are those of the authors, and do not necessarily represent the views of the World Bank or the CMF. 2 Leopold Sarr (corresponding author): The WorldBank Group, South Asia Human Development Unit, lsarr2@worldbank.org. 3 Centre for Micro Finance, Institute for Financial Management and Research, santadarshan.sadhu@ifmr.ac.in 4 Research fellow at the German Institute for Economic Research (DIW), nfiala@diw.de 1 currently has over 48 million activated savings accounts across India. Despite being open, many savings accounts have remained dormant, thus raising the question of whether access alone can result in real financial inclusion. In this paper, we present the results of a randomized financial literacy training program offered to FINO clients on the transaction activities in their (no-frills) savings account. About 3000 clients, in two districts of the state of Uttar Pradesh, were randomly assigned to a control and a treatment groups out of whom 1500 treatment clients received a two- day financial literacy training. Using the historical transaction data from their savings account, we estimate the short run impacts of the financial literacy training on account usage. Our results show a persistent treatment effect on account usage in the short run. Further, when we control for heterogeneity, it appears that, the treatment effect is more pronounced for female clients, while treatment clients who contracted outstanding loans at baseline, made more deposits and transac- tions in the post intervention period. Overall, the results suggest that financial education can in- crease usage of no-frills savings accounts and consequently could go a long way in improving financial inclusion for the poor. Given FINO’s mandate to expand access and use of financial services to the poor, an effective financial literacy training, coupled with adequate incentives for the Bandhus, have the potential to influence savings behaviors, and hence, improve the financial welfare of low income families. 2 Table of Contents 1. Introduction ............................................................................................................................... 2 2. The FINO Program and Intervention ..................................................................................... 6 2.1 Doorstep banking.............................................................................................. 6 2.2 Financial education intervention..................................................................... 7 3. Experimental Design ................................................................................................................. 8 3.1 Methodology ..................................................................................................... 8 3.2 Sampling ........................................................................................................... 9 3.3 Data Collection ............................................................................................... 10 3.4 Summary Statistics and Balance Test ............................................................ 11 4. Empirical Analysis of Short Run Treatment Effects of the FE training .......................... 12 4.1 Graphical Analysis ......................................................................................... 14 4.1.1 Trends in FINO Account Usage: From Pre to Post Intervention Analysis .. 14 4.1.2 Trend in Non-Trivial Accounts .................................................................... 22 4.2 Econometric Analysis: Estimating the Treatment effect on FINO Account Usage ............................................................................................................................... 23 4.2.1 Econometric Specification ............................................................................ 24 4.2.2 Estimated Treatment Effects ......................................................................... 25 4.3 Heterogeneity Results...................................................................................... 26 4.3.1 Econometric Specification ............................................................................ 26 4.3.2 Estimated Heterogeneity effect ..................................................................... 27 5. Conclusions .............................................................................................................................. 28 References .................................................................................................................................... 30 Regression Results ...................................................................................................................... 31 1 1. Introduction Financial literacy programs deal with the knowledge of basic financial concepts and the skills to translate this knowledge into improved financial behaviors. An increasingly common approach to fighting poverty is to provide training in financial literacy under the assumption that the poor need to fully understand the basics of the financial world as they often don’t. The hope is that participants of financial literacy programs walk away with an increased awareness and compre- hension of financial concepts, especially those focused on saving. Arguably, with the right knowledge, the poor can avoid scams, stay out of debt, build assets, create and maintain financial independence. This paper evaluates the short term impacts of a financial literacy program on the use of FINO smart card. A baseline survey covering 1500 individuals in control and 1500 individuals in treatment villages and including detailed questions on individual and household de- mographics, income, savings behavior, risk and time preferences, was carried out. The baseline information along with banking transactions made by FINO clients will help underscore the po- tential for financial literacy to influence savings behavior. We describe the data collection activi- ties, the balance of baseline variables between households in treatment and control villages, and explore some additional hypotheses to better understand who is making use of the FINO smart card. Using data on FINO client transaction activities, we look at the impact of financial literacy training on account usage in the short run, with a focus on the heterogeneity of impacts for gen- der, age, education level and baseline financial literacy or exposure to formal savings. Despite the growing interest financial literacy has had in individuals’ livelihood, the cur- rent literature on financial literacy training is sparse. Much of the interest in financial literacy has been driven by correlations and studies in developed countries. For instance, Lusardia and Mitchell (2007) find that financial literacy is correlated with wealth levels at retirement. To the best of our knowledge, the only published experimental trial of financial literacy education is the study by Cole, Sampson and Zia (2011). Working with unbanked households, they explore the impact of financial education and monetary incentives for opening bank ac- counts and find that training works for adults with low education and low financial knowledge, but not for other groups. Comparing education to simple payments for opening accounts, they find that payments have large effects on the full sample, and are significantly cheaper than the education program. These effects last after two years, though there is no effect on whether indi- 2 viduals keep savings, except for those who got both high incentives and financial literacy train- ing. The researchers worked with a local bank at no cost, but the overall take up was very low, with only 10% at most opening an account. These results suggest there may not be much hope for the call for greater financial literacy. An unpublished paper by Cole, Shapiro, Carpena and Zia finds financial education does not prepare people to make good choices between complicated financial options, though it can make them more aware of financial products. They propose that numeracy is the limiting key factor. However, there is evidence that financial education can work in specific circumstances. Duflo and Saez (2003) randomly encouraged staff at a university to attend retirement account information sessions. They find that enrollment in the account increases, though by a small amount. The effect of training on business owners also looks more positive. Valdiva and Karlan (2010) find, in Peru, some evidence of effect of business training for entrepreneurs on business practices, but no effect on profit. Working with business owners in Bosnia and Herzegovina, Bruhn and Zia (2011) find that training leads to improvements in knowledge and attitudes, and success of surviving firms, but does not increase the likelihood of survival. The results are driven by surviving businesses investing more into their businesses and refinancing more favorably. Again, looking at the heterogeneity of impacts, those with better baseline financial literacy knowledge had more profit, but with no effect on survival or default rates. The authors conclude that the lack of business knowledge is unlikely to be the major constraint for new businesses. Rather than focusing on the standard workshop method of teaching financial literacy to businesses, Drexler et al. (2010) compare a “rule of thumb” program, which focused most heavi- ly on the need to keep separate records between home and business, and the more common class- room financial literacy training. They find that rule-of-thumb training has some effect on wheth- er business owners kept accounting records at all, how they calculated revenue and if they kept separate books for business and home. Businesses also had better sales during bad weeks, sug- gesting training can help with adverse shocks. Follow-up training had modest improvements for those in classroom training. 3 There is some macro evidence that utilizing banking has significant implications for de- velopment. Beck et al (2007) find a correlation between financial depth and poverty across coun- tries, while Levine (2005) finds a correlation between financial depth and economic growth. This macroeconomic evidence seems to be consistent with the fact that most poor people in developing countries simply don’t bank. About 90% of the 2.5 billion people around the world making less than $2 per day don’t have a bank account (FAI and McKinsey 2009). Many of them participate in other savings options, such as ROSCAs, but most don’t use formal savings options. This may be due in part to the lack of knowledge about the value of formal banking, and a lack of easy access to formal savings services. Most banks simply are not located where the poor live or near them and they do not offer services for low depositors. Reducing transaction costs for banks -local branches and ATMs are expensive, especially when working with very small amounts of money- and for customers –given that fees, travel and wait times can be costly- could be a solution to this lack of access. Doorstep banking is also often called “last mile” banking as the bank reaches out to those who can’t make it to the banks themselves. This is sometimes done in retail shops, other times by agents who live in or near the villages, such as FINO, or through mobile banking vehicles or even mobile phones, such as those being pioneered by MPESA and M-Kesho. The value of saving for the poor can be numerous. Savings can be used to generate lump sum cash to invest in businesses or mitigate risk, such as adverse shocks to employment, health or crops. Also, given that income for poor people comes often very irregularly, a place to store cash could be used to smooth consumption. All of these issues can be solved through loans and microfinance, though interest rates can often be beyond the poor’s ability to repay. As Murdoch et al. (2010) show, the poor often use a mix of options, which includes savings and loans at the same time. The lack of full usage of savings is often attributed to a mix of psychological com- mitment issues, hyperbolic discounting, and the value of risk sharing. Saving, whether formal or informal, is hard. Dupas and Robinson (2011) find that de- mands for transfers to others and unplanned luxury expenditures are the two biggest reasons for people not saving. Reducing the barriers to acquiring a formal savings option significantly re- duced both of these issues. Their results suggest that self-control issues can be overcome through savings devices. 4 There are some advantages to formal banking. Unlike village savings programs, banks of- fer privacy from villagers and family members, both of which can present significant demands on cash holdings, decreased risk of theft or default from other savings members and reliability, if the banking agent is regularly available. They can also, when financed by NGOs or through gov- ernment regulation, be lower cost or even free of any charges. Dupas and Robinson (2010) find that giving micro enterprises in Kenya access to a low cost savings account increases savings, productive investment and food expenditures for women, but not for men. The accounts also helped to mitigate health shocks. Even with a de facto nega- tive interest rate, usage was high, though heterogeneous, with only about half utilizing the ac- count in the first six months. Sometimes, unique savings programs offer the best chance for households. Duflo, Kre- mer and Robinson (2009) experiment with alternative money storage by encouraging farmers at harvest time to spend money on fertilizer for next season, which was then delivered for free. The program was found to increase fertilizer usage. Brune et al (2010) gave Malawi farmers either normal savings accounts or commitment savings accounts where the farmers specified when money could be withdrawn. The rates of deposits were high for commitment savings accounts at almost twice that of the normal account. Ashraf et al (2006, 2010) also introduced commitment savings accounts in the Philippines for those who already had savings accounts. They find in- creased savings rates. Natural experimental evidence utilizing banks expansions also suggests a value to offer- ing savings accounts to individuals, though it is hard to disentangle all of the effects. Aportela (1999) studies the expansion of a Mexican savings program in post offices in communities. Sav- ings rates in the areas increased, though it is possible they came at the expense of other savings options. Two other studies that look at bank expansion, Burgess and Pande (2005) in India and Bruhn and Love (2009) in Mexico, find increases in welfare, though they can’t distinguish be- tween the effect of increased banking, or increased or subsidized credit opportunities. The results of previous research on financial education and savings account access pro- vide the main impetus for the research described here. By randomly providing financial literacy training to those with formal savings options, we hope to increase the knowledge of what works and what does not in financial literacy training. 5 The remainder of the paper is organized as follows. The next section outlines the FINO program and why it presents a unique opportunity to study the intersection of financial literacy and financial access. In section 3, we describe the experimental design, sampling and data, in- cluding issues that arose during the survey implementation. Section 4 presents the graphical analysis using transaction data on account activity followed by an econometric estimation of the treatment effects and in section 5, we provide concluding remarks. 2. The FINO Program and Intervention While the benefits of banking access and financial literacy are well acknowledged, the evidence of their impact for most people, as previously discussed, is lacking. This could be due to a num- ber of reasons, including people’s low interest in utilizing banking. This section discusses the FINO banking and financial literacy training programs. 2.1 Doorstep banking In 2006, the government of India instituted a requirement of banks that 20% of all bank accounts in India must be held by the poor. FINO was thus developed in order to help banks bring ac- counts to the poorest people. FINO works with partner banks to establish financial distribution platforms in rural vil- lages, ensuring that people in previously hard to reach areas have access to bank accounts. This outreach is done through a hierarchical system, with the bandhu, or business correspondent, be- ing based in the villages and hence being most in contact with individual clients. Ideally, the bandhus would be interacting with clients multiple times in a week, though in practice, as will be discussed, this occurred significantly less. There is, however, very little regulation on such programs, especially given the fact that the law rests on the number of accounts held by the poor instead of the activity level in their ac- counts. While FINO has carried out an incredibly aggressive campaign to reach out to the poor- est, across India, through the FINO smart card, much more than access is needed for the poor to reap the benefits of branchless banking. Today, mobile money operators, who are growing in popularity and offering an increasing range of services, focus on active accounts, not just regis- tered accounts. FINO accounts are no frills, and offer interest but not any additional benefit other than deposit and withdrawal facilities. 6 FINO has trained more than 10,000 bandhus and has over 48 million customers and it is growing by up to 1 million clients per month. While this doorstep banking model looks very at- tractive, the reality on the ground is actually more complicated than the model would suggest. Of the baseline sample of 3000 clients, 88% were found to have not made any transaction during the pre-intervention period (March- May 10 2011), whereas around 10% held positive balances in FINO account as of May10, 2012, when the training began. FINO had only recently begun opening accounts in the areas of this study, but after 12 months, only 10% had maintained more than Rs. 50 of balances in their accounts as of April 2012. This is partly due to the fact that, many bandhus are not catering to the needs of clients in the villages, despite living with them. As the account activity crucially depends on the Bandhu’s visit in the village/neighborhood, a follow up interview was conducted on a subsample of clients to collect information on Bandhu presence. Out of a total of 1363 clients who were sampled for the follow up survey, only 28% report having seen their bandhus during a monitoring survey conducted between October and November 2011. 2.2 Financial education intervention In partnership with the World Bank evaluation team, FINO developed and implemented a pilot financial education training program. The program was designed to support the increased use of FINO’s savings accounts to encourage and facilitate saving. It focused on teaching the knowledge and skills required to adopt good money management practices for household budget- ing and spending, and for saving. The program took place in two districts in Uttar Pradesh over the course of several months. Originally, the financial education program was to be delivered by the bandhus, but to ensure that the quality of the training is maintained and uniform, a team of seasoned trainers was deployed for the various FE workshops. The training consisted of 2-day financial education workshop for the beneficiaries. Each day, people were given 2 to 3 hours of training with up to 30 people in the sessions. The beneficiaries of the training were the sampled clients who were assigned to the treatment group and administered the baseline survey. Initially, client attendance was very low as information regarding the program was not spread well and many clients are poor farmers who rely heavily on daily wage earnings for their subsistence. To ensure that the treatment clients attend the training, CMF assisted FINO in the information campaign by making 7 door to door visits before the training workshop was carried out. In addition, a small remunera- tion, equal to the daily wage, was given to clients as an incentive for attending the training. The initial attendance level was only 46% of the total sample, but rose significantly after the addi- tional campaigning was carried out. About 71% of all treatment clients attended both days of the program, 12% attended one of the two days of training and 17% did not attend any of the ses- sions. The training material consisted of a video program shown to people by projector, roll playing and discussions after relevant topical sections are presented. The sections included (1) the role of banking in people’s lives, (2) borrowing and spending, including a discussion of in- terest rates, and (3) cash management. 3. Experimental Design 3.1 Methodology In order to unpack the causal impact of the financial literacy training program, the experiment was conducted on a random sample of individuals in villages where FINO operates. Villages were randomly selected to either receive the training, or receive no training. Individuals from treatment villages that had FINO smart cards were then randomly selected to be offered the fi- nancial literacy training. To decrease contamination, randomization was done at bandhu level, i.e., at the village level. When treatment assignment is randomized and compliance with treatment assignment is perfect, all those assigned to the training complete it, and all those in the comparison group do not pursue training by other means – then the average treatment effect, or ATE, is simply the dif- ference in performance among the individuals in the treatment versus control groups. With base- line data on particular outcomes, one can also calculate an ATE on the differential improvement over time between treatment and control individuals. In the real world, it is likely that some individuals selected for the training would not at- tend it, and those not selected could find alternative means to receive the “treatment”. Under such circumstances, an instrumental variables approach is the ideal estimation method, where being treated is instrumented by being assigned to treatment. This is sometimes referred to as a local ATE, or LATE. However, what the LATE estimate does not tell us is the impact of the 8 program on individuals who would have found a way to enroll in training in any case, or those who would never enroll regardless of assignment (i.e., non-compliers). One might argue that the impact of treatment on the compliers is a key policy parameter of interest. It will not, however, be representative of the average impact on all participants. Another set of parameters of interest are the conditional ATEs—the average impacts of the program on individuals with different initial characteristics, such as sex, literacy, education, etc. To identify these heterogeneous impacts, treatment can be interacted with initial values of these characteristics and the conditional impact identified. In these cases, however, it will be im- portant to recognize that many initial characteristics are inter-correlated (i.e. high education with urban presence and high family incomes), and so attribution of the conditional effect to a particu- lar initial trait must be done with care, primarily by controlling for the maximum number of such interactions. 3.2 Sampling The program was rolled out with the clients of 200 bandhus who were working in 244 villages in the two experiment districts, Varanasi and Azamgarh. A description of the sample size require- ments is presented in Appendix A. These 200 bandhus were selected from the list of all FINO bandhus who work in these districts using a distance based randomized bandhu dropping method. Under this method, in or- der to prevent contamination in control and treatment groups due to contiguous bandhu service areas, the evaluation team decided to adopt a random dropping method in which, from a pair of bandhus who are very close to each other (less than 1.25 KMs), one bandhu was randomly dropped to minimize spillovers) and bandhus whose own service areas are far apart (more than 10 KMs) were also dropped (in order to make data collection and training easier). Using the GPS coordinates of bandhu service areas, distances between the service areas of each bandhu were calculated and a drooping rule was applied to drop bandhus based on the calculated distances. In the next step, these 200 bandhus were randomly assigned into treatment and control. In total, 108 bandhus were kept as treatment and the remaining, 92, as control using the following procedure: from the list of 200 bandhus, 25 clients were randomly selected from each bandhu using FINOs 9 account records. 1 Using the random treatment assignment, a pre-baseline randomization check was undertaken to ensure that the sample was well balanced with respect to available de- mographics and account activity information. From the FINO client database, information on age, gender, and account activity status (whether a client has made at least one transaction during the 6 month period before February 2011) was collected. These parameters (percentage of fe- male; share of clients in the age groups 18-24, 25-59, 60 and above; and share of clients who made at least one transaction in the 6 months period before February 2011) were individually regressed on the treatment dummy. The regression results showed that, in all cases, the treatment dummy was statistically insignificant indicating that there was no observable difference between the treatment and control bandhus with respect to these parameters before the baseline was im- plemented. Finally, a sample of 15 clients per bandhu was drawn for the survey interview. 3.3 Data Collection A questionnaire was designed by the evaluation team to understand clients’ current knowledge of financial tools and their current financial behaviors. The questionnaire also collected detailed information on various variables that are assumed to play an important role in household behav- ior and financial wellbeing. These include: • Household demographics such as the number of family members, age, educational at- tainment, primary, secondary and tertiary occupation, income earned in the preceding 14 days; • Household income from various sources; • Household ownership of financial and non-financial assets; • Household savings and borrowings; • Household expenditures; • Respondent’s perception towards budgeting; • Measure of respondent’s numeracy; 1 Buffers of 10 clients per bandhu were kept to ensure that, for each bandhu, the target of 15 clients could be surveyed. The first 15 clients (based on sorted client ids) per bandhu was treated as the priority and the buffer only used in the extreme case where, in spite of making every effort deployed, the survey team was unable to find the client from the original list. 10 • Respondent’s involvement and knowledge regarding household financial matters such as savings, investment and insurance etc. • Respondent’s time preference and preference for risk. The survey was conducted using a Samsung mobile device with Windows Mobile 6.5 op- erating system. The questionnaire was programmed into the mobile device using C++ program- ming language. After the completion of the survey, the entire database was exported from the mobile devise into CSV files and a baseline survey database was created in excel and STATA format. As the data was collected using a mobile device, the software enabled standard logical checks and, as a result, no further cleaning was necessary. However, to eliminate the possibility of data entry error and to ascertain quality and consistency of data, 10% of the responses were selected randomly and values were crosschecked through telephonic verification from the re- spondents. Additionally, for most of the important variables, a thorough outlier checking was con- ducted to eliminate the possibility of data entry error. Extreme values in the top and bottom of the distribution were crosschecked by telephonic verification and in case of any mismatch, the incorrect responses of the survey data were overridden by the values provided by the respondents over phone. 3.4 Summary Statistics and Balance Test In this section, we first present a brief summary of the major baseline variables and then discuss the results of the balance test. The majority of households own livestock, although most did not receive income from them, in the week preceding the interview. Half of the respondents have national bank accounts, which suggests that, while banking is not easily accessible in the areas where FINO operates, some in- dividuals/households are interested enough in obtaining formal savings that they will go through the effort. Children represent about half of the household size at just over 3 minors per house- hold. Literacy also appears to be low with only half of household heads reporting that they are literate. The household heads are on average about 45 years old. 11 Interestingly, only 86% of households report having a FINO account. This suggests that some of the population are either not aware they have accounts, or were not aware of what they were signing up for when they opened their account. As the next step, we have conducted the balance test to identify the differences that existed be- tween the treatment and control samples before the intervention occurred. The results of the bal- ance test are presented in Appendix B. Most of the baseline variables were balanced except for the following: per-capita expenditure, standardized index for numeracy, standardized index 2 for financial literacy, dummy for having outstanding loans from formal sources, dummy for clients who completed at least secondary level of education. Thus, in order to avoid the bias that might arise in estimating treatment effects, we will use these variables as controls in the empirical analysis. 4. Empirical Analysis of Short Run Treatment Effects of the FE training In this section, we investigate whether the FINO financial literacy intervention has had any in- fluence on the usage of FINO smart card by the clients, in the short run. For this, a unique dataset of transactions made by the clients was collected on a regular basis, from March 2011 up to April 2012. To the best of our knowledge, this is the first transaction database with such detailed in- formation made available for research to understand the impact of financial literacy on savings behaviors. In particular, the FINO transaction data provides the number of debit and credit trans- actions reported on the client’s account for a stipulated time period. It also gives the amount of balance held at any specific date. For the purpose of our analysis, activity of the client account has been captured in various peri- ods covering the pre-intervention period through the post intervention period. The pre- interven- tion data went from March to mid May 2011 whereas the post- intervention period started from 2 The combined score achieved in financial literacy related questions was standardized (by subtracting sample mean from the observed value and then dividing the difference by the standard deviation of the distribution) to remove the scale effect. 12 August 2011 to April 2012 data. In the meantime, the financial literacy intervention took place between May and August 2011. From the transaction data, we constructed measures of average daily account usage. First, the total number (amount) of transactions, for given period, was computed as the sum of the total number (amount) of debit and credit transactions during that period. Then, a measure of average number (amount) of daily transactions was calculated by dividing the total number (amount) of transactions by the total number of days in that period and then converted into monthly average by multiplying by 30. Similarly, the monthly average number (amount) deposit, and the monthly average number (amount) of withdrawals were calculated for various reference periods. While constructing the monthly average values of the dependent variables, we have trimmed the distri- bution of deposit, withdrawal and total transactions at the 99 percentile to remove the outliers. The table below presents the summary statistics of the variables of interest, including the set of dependent variables and the set of control variables used in the estimations: Table 1: Summary of Variables used in Analysis Number of Standard Observations Mean Deviation Dependent Variables Before Intervention Monthly average deposits: March- May 2011 * 2971 2.05 10.51 Monthly average withdrawal March- May 2011 * 2973 0.98 6.01 Monthly average total transactions (sum of deposits and withdrawals) March- May 2011 2964 2.74 13.29 During Intervention Monthly average deposits: May- August 2011* 2973 6.04 27.65 Monthly average withdrawals May- August 2011* 2973 3.83 22.39 Monthly average total transactions (sum of deposits and withdrawals): May- August 2011 * 2968 9.45 45.45 Post Intervention Monthly average deposits: August 2011 - April 2012 * 2973 7.33 36.21 Monthly average withdrawals: August 2011 - April 2012) * 2973 5.84 32.05 Monthly average total transactions (sum of deposits and withdrawals): August 2011 - April 2012 * 2973 14.68 75.12 Explanatory Variables Dummy for having loan outstanding with formal financial institutions (Bank, MFI, SHG, NBFC) 3004 0.11 0.31 Number of female members in the household 3004 3 1.93 13 Per capita total expenditure:14 days prior to survey * 2964 249 165 Standardized index of competency in numeracy 2986 0 1 Standardized index of competency in financial literacy 3004 0 1 Dummy: client has at least secondary education level 2992 0.24 0.43 Dummy: client is female 2992 0.40 0.49 Dummy for having a non-FINO savings/post office bank account at base- line 3004 0.56 0.50 Balance held in FINO account as of May 2011 * 2973 3.89 18.40 * Indicates that variables are capped at 99 percentile. As can be seen from Table 1, even though the last three variables were found to be balanced in the baseline, we will use them as controls in the empirical analysis since they may have im- portant bearing on account activity. Using the transaction data, we will explore how the usage of the no frills account offered by FI- NO has been influenced by the financial literacy intervention. For this purpose, we start with a graphical analysis of account usage in the pre, during and post intervention periods and then pre- sent an econometric estimation of the treatment effects. 4.1 Graphical Analysis 4.1.1 Trends in FINO Account Usage: From Pre to Post Intervention Analysis As mentioned previously, one of the key features of this intervention is the availability of client transaction data. This section presents a graphical analysis of account activity where we clubbed various time periods in such a way that trends in pre and post intervention periods could be readily compared. Figure 1 in the next page shows the monthly average transaction amount from March 2011 to April 2012. The period March-May11 represents the pre-intervention period whereas May- Aug11 corresponds to the intervention period and Aug11- Apr12 represents the window of entire post-intervention period. The treatment and control means show that, the monthly average 14 amount of transaction was quite similar 3 for both treatment and control clients before the inter- vention, suggesting that the overall sample is balanced across observable and unobservable char- acteristics of clients. On the other hand, the monthly average amount of transaction increased significantly during the intervention, in May-Aug11, for both groups, but the jump in the month- ly average amount of transactions of the treatment group is much greater than that in the control group. This observed increase seems to be transitory and could be due to seasonality effect, as can be seen from the post intervention period Aug11-April 12. However, the immediate effect of the financial literacy intervention appears quite prominent. Figure 1 Treatment Effect on Monthly Average Transaction Amount Sum of all transactions in INR 15 14 Monthly Average transaction amount 10 5 5 5 3 3 2 0 Pre(Mar-May11) During(May-Aug11) Post(Aug11-Apr12) Control Treatment 3 A ttest shows no difference in treatment and control mean for the period March- May. 15 The data from the post intervention period (Aug11- Apr12) quite clearly shows that the treatment sample does significantly greater amount of transactions than the control sample, as can be seen 4 from the last two bars, despite the decreasing effect of the training over time. In addition to the amount of transactions, we look at the number of transactions made by FINO clients. Figure 2 shows the monthly average number of transactions generated between March 2011 and April 2012. Figure 2 Treatment Effect on Monthly Average Transaction Count Sum of all transactions Number .4 Monthly Average transaction count .343 .3 .201 .2 .112 .107 .102 .103 .1 0 Pre(Mar-May11) During(May-Aug11) Post(Aug11-Apr12) Control Treatment The monthly average number of transaction data also demonstrates the treatment effects. At baseline, we observe similar values and when the training program got implemented, the data shows significant increase in the monthly average number of transactions, for the treatment 4 A ttest on the difference between treatment and control mean in the post intervention period (Aug11- Apr12) con- firms this. 16 sample, as compared to the control sample. This effect did linger, even during the post interven- 5 tion period, suggesting some persistence in the impact of the training. To be able to unpack the effect of the total transactions made in the client account, we decom- pose the transactions into deposits and withdrawals. We start with Figure 3 which shows the monthly average amount of deposits made between March 2011 and April 2012. Figure 3 Treatment Effect on Monthly Average Deposit Amount Credit transactions in INR 10 8.9 Monthly Average Deposit amount 6 8 4.2 4 3.2 2.4 2.2 1.8 2 0 Pre(Mar-May11) During(May-Aug11) Post(Aug11-Apr12) Control Treatment As can be observed from Figure 3, the treatment clients started depositing more money during and after the intervention. Starting from similar values before the intervention, the monthly aver- 5 A ttest on the difference in treatment and control mean in the post intervention period (Aug11- Apr12) confirms this. 17 age amount of deposits has increased by more than four times, for the treatment sample, during the intervention, while it only increased by around 32 % for the control sample during the same period. 6 Further, treatment clients deposited, on average, significantly greater amounts, as com- 7 pared to the control sample over the entire post intervention period. A similar trend is shown by the monthly average number of deposits – starting from similar pre- intervention values, the treatment mean records a big spike during the FE intervention and re- mains way above the control mean (Figure 4) during the post intervention, as the difference be- tween control and treatment appears to be highly statistically significant. 8 Figure 4 Treatment Effect on Monthly Average Deposit Count Credit transactions Number .25 .24 Monthly Average Deposit count .2 .16 .15 .1 .1 .085 .09 .072 .05 0 Pre(Mar-May11) During(May-Aug11) Post(Aug11-Apr12) Control Treatment 6 The increase in both groups could indicate some seasonality effect in account usage 7 A ttest on the difference in treatment and control mean in the post intervention period (Aug11- Apr12) confirms this 8 A ttest on the difference in treatment and control mean in the post intervention period (Aug11- Apr12) confirms this. 18 We also look at the withdrawal activity in the client account, to see whether there were any sig- nificant differences between treatment and control groups. The data on the monthly average withdrawal amount has been plotted in Figure 5. Both treatment and control means increase during the intervention. But, the increase in treatment mean is very steep, while the post intervention data (Aug11- Apr12) shows that treatment clients withdrew significantly larger amounts of cash, as compared to the control clients. 9 What explains such behavior? Does this cancel out the positive impact of the training program on the total de- posits made by FINO clients? We will attempt to address these questions in the next sections. Figure 5 Treatment Effect on Monthly Average Withdrawal Amount Debit transactions in INR 5 Monthly Average Withdrawal amount 3 4 5 2 2 2 1 1 1 1 0 Pre(Mar-May11) During(May-Aug11) Post(Aug11-Apr12) Control Treatment 9 A ttest on the difference in treatment and control mean in the post intervention period (Aug11- Apr12) confirms this. 19 Finally, Figure 6 presents the data on the monthly average number of withdrawals. It again shows a similar trend. Throughout the post intervention period, treatment clients withdrew al- most twice as much as they did before the intervention. Figure 6 Treatment Effect on Monthly Average Withdrawal Count Debit transactions Number .15 Monthly Average Withdrawal count .13 .1 .072 .047 .05 .043 .039 .036 0 Pre(Mar-May11) During(May-Aug11) Post(Aug11-Apr12) Control Treatment Comparison of deposit and withdrawal data show quite clearly that the treatment clients are mak- ing more of both credit and debit transactions after the intervention as compared to the control sample. However, the amount deposited outweighs the amount withdrawn, on average, suggest- ing some positive effect of the training program in the sense that it has induced change in the savings behaviors of FINO clients. The persistent treatment effect suggests some effectiveness in the financial education program implemented. This effect will be further investigated when the endline data become available. 20 Finally, we explore whether the treatment had any effect on the amount of balances held in the savings account served by FINO. We have collected information on balances held as of a given cut off dates in the period before, immediately after and seven months after the intervention (May 15th, August 27th, 2011 and April 10th, 2012 respectively). Figure 7 Treatment Effect on Amount of Balances Held Balances held in a given date in INR 25 25 20 Balance as of date 15 13 12 10 6.4 4.3 5 3.5 0 Before(May11) Immediately after(Aug11) After 7 months(Apr12) Control Treatment The above figure shows that the amount of balances held in the savings account increases by around four times for the treatment in August 2011, immediately after completion of the inter- vention, while it increased marginally for the control group. In the long run (after 7 months from the completion of intervention), though the balances held by both groups have increased over- 10 time, treatment group , on average, held twice of the balances held by the control group. 10 Tests of difference in mean (ttests) show that the balances held by the treatment group in August and April 2012 were statistically significantly greater than that of the control group, though there was no statistically significant difference between the treatment and control groups in May 2011. 21 Thus, looking at the balances held at a specific date, we find a persistent treatment effect as well, which indicates that the treatment group not only undertook more transactions, they also main- tain larger balances in the post intervention period. 4.1.2 Trend in Non-Trivial Accounts The no-frills savings accounts served by the business correspondents often remain dormant – maintaining just minimum or very negligible balances. To examine whether the financial literacy training has had any impact on the fraction of non-trivial accounts, we first classify the non trivi- al accounts as those that maintain balances of more than Rs. 50, at a given date (the cutoff date when the data for a given period was collected) and also conduct at least two transactions per month, on average, in the given period. Then, we plot the data on the percentage of non-trivial accounts, for both treatment and control samples, during the same time period, as shown in Fig- ure 8. Figure 8 Treatment Effect on Percentage of Non Trivial Accounts Balance>Rs. 50 & at least 2 transaction during a given period 4 Percentage Non Trivial Account 3.3 3.3 3 2.7 2.7 2.5 2.5 2 1.8 1.5 1.6 1.2 1.2 1.1 1 1 1.1 1 .53 0 May Aug Nov Dec Jan Feb Mar Apr Control Treatment 22 Figure 8 shows that, before the intervention, both treatment and control groups had around 1% of non-trivial accounts. But, the intervention has induced a large increase in the share of non-trivial accounts in the treatment sample. In the period immediately following the intervention, that share has increased more than two times (at 2.7%) for the treatment sample, then remained most- ly in that range until February 2012, except in December, which could be due to seasonality, and increased to be at more than 3% non-trivial balance, in the subsequent periods. To sum up, in the months following the intervention, the percentage of non-trivial accounts among treatment cli- 11 ents, has remained significantly greater than in the control sample as seen in the graph. This seems to indicate that the financial literacy training has not only had significant potential to in- crease the fraction of the no-frills accounts that maintain non-trivial balances, but it may have also allowed client transactions to remain particularly active, for the treatment sample. 4.2 Econometric Analysis: Estimating the Treatment effect on FINO Account Usage Before proceeding with the econometric analysis, it is important to reiterate that we have con- ducted a detailed analysis to explore whether, at baseline, the treatment and control samples were balanced with respect to important variables. As mentioned in Section 3, we have identified the variables which were not balanced at baseline to include them as controls in the regression anal- ysis. To explore whether the financial literacy treatment affects the usage of no-frills accounts (hereaf- ter NFAs perhaps we should introduce this much earlier in the text) served by FINO in the short run, we estimate the treatment effect on total transactions, total deposits and withdrawals using three specifications: in the first specification, we only estimate the basic treatment effect; in the 11 Also, ttests of difference in mean show that for each of the post intervention periods, the treatment mean was sta- tistically significantly greater than the control mean. 23 second specification, we incorporate interactions between the treatment variable and some im- portant variables to estimate their effect on the treatment, and, in the final specification, as a measure of robustness check of the heterogeneity effects observed, we include these interaction terms simultaneously along with a set of baseline controls to pick up the heterogeneity effects. 4.2.1 Econometric Specification To estimate the treatment effect on the transactions in NFA served by FINO, we estimate the fol- lowing regression: Zi =a0+a1Ti + a2Xi +ei (1) Where Zi represents the dependent variable of interest for client i, Ti represents the treatment dummy (=1 if the client was assigned into treatment group and zero otherwise), and Xi includes a set of independent variables that were found to be imbalanced at baseline. 12 The error term ei is an iid random error variable with a zero mean and the standard errors are clustered around the unit of randomization (at the agent level). We will estimate the above equation for three dependent variables – total deposits, total with- drawal amounts and total transaction amounts, which are the sum of deposits and withdrawals in a given period. The coefficient a1 estimates the average treatment effect. However, since some of the clients who were assigned to the treatment did not actually attend the financial literacy training, we will pre- sent instead the Intent-to-treat (ITT) effect. Further, to reduce biases that might arise from differences in key baseline variables between treatment and control samples, we use the latter as controls in the above regression. Additionally, we include, as controls, client’s gender, education and amount of balances held in the account 12 Bruhn and McKenzie (2009) 24 before the intervention. All estimations have standard errors clustered at the bandhu level (equivalent to the village level). 4.2.2 Estimated Treatment Effects Tables 1 to 6 present the regression results with the coefficient estimates of equation 1 in which the dependent variables are the monthly average amount and number of deposits, withdrawals, and of total transactions. For each of the regression equations, we present the estimates of the effect during the intervention period (May- August 2011), and the post treatment period (August 2011- April 2012) to explore the overall treatment effect. Column 1 of Table 1 presents the results of the treatment effect on the amount of deposit during the intervention period, May- August 2011. The coefficient estimate on the treatment variable shows that the treatment effect is positive and statistically significant: on average, the treatment clients deposited Rs. 5.4 more than the control clients during the intervention period. Column 2 shows that the treatment effect persists in the post intervention period: during the en- tire post intervention period of August 2011- April 2012, the monthly average deposit of the treatment clients was Rs. 2 more than the monthly deposits of the control clients. In light of these results, it seems that the treatment clients made significantly more deposits after they were exposed to the financial literacy training. Looking at the magnitude of the treatment effect during and post intervention, it appears that the strongest treatment effect was observed during the intervention period. Tables 2 and 3 present the estimates of treatment effect on monthly withdrawals and total trans- actions respectively. Similar to the treatment effects on monthly average deposits, we find a sta- tistically significant impact of the treatment on withdrawals and total transactions during and af- ter the intervention, with the strongest effect observed during the intervention period. When we look into the effect of treatment on the monthly number of transactions undertaken during and after the intervention, we find similar results. Tables 4- 6 show that the monthly aver- 25 age number of deposits, withdrawals and total transactions of the treatment clients are statistical- ly significantly greater than those of the control clients during and after the intervention. Also, the magnitudes of the treatment effect follow the same trend as before – largest effect being ob- served during the FE intervention with a fading effect over time. Overall, these results suggest that the financial literacy training has been quite effective at induc- ing participants to increase the use of their FINO smart card. 4.3 Heterogeneity Results 4.3.1 Econometric Specification After having singled out the estimated treatment effects, we now explore, in this section, the het- erogeneity effects by estimating the following equation: Zi =a0+a1Ti + a2Ti*Yi +ei (2) Where Yi is the variable of interest, which is interacted with the treatment to identify heterogene- ity in the treatment effects. This specification allows to identify the independent effect of key baseline variables. For this purpose, we interact one variable at a time with the treatment and measure the heterogeneity ef- fect. Given the difference in the effect that gender and education can have, we include interactions of client gender and education with the treatment. As earlier, we include interactions with variables that were found to be imbalanced at baseline. Additionally, to capture the effect of exposure to formal savings instruments, we interact the dummy of having non-FINO savings bank account with the treatment. Finally, we include an interaction of baseline measure of financial literacy with the treatment to identify whether initial difference in financial literacy results in heteroge- neous treatment effect. 26 4.3.2 Estimated Heterogeneity effect The coefficient estimates of equation 2 are presented in Tables 7-12. Columns 1-14 in Table 7 present the heterogeneity effects on monthly average deposits. Columns 1 and 8 present the het- erogeneity with client’s gender. The coefficient estimates on client gender and interaction term with treatment indicate that the deposits of female treatment clients were statistically significant- ly larger amount in their FINO savings account during the post intervention period as compared to the male clients (irrespective of their treatment status) and as compared to female control cli- ents. This result clearly indicates that the financial literacy treatment has been effective in in- creasing deposits for FINO female clients. The heterogeneity results with client’s educational attainment are presented in columns 2 and 9 in Table 7. The coefficient estimates indicate that being a client with at least secondary education level does not induce any heterogeneous effect on their monthly deposits. Similarly, the results shown in columns 5 &12 and 6 & 13 indicate that measures of per-capita expenditure, and competency in numeracy do not have any heterogeneous effects on treatment. Interaction of baseline financial literacy measure with treatment (presented in columns 7 & 14) is also found to be statistically insignificant, suggesting that the treatment effect is independent of the pre-existing financial literacy status. We next consider whether having previous exposures to other non-FINO savings bank account and previous exposures to formal loans have had any heterogeneous effect. Results shown in column 3 and 10 in Table 7 indicate that treatment clients who had loan outstanding at baseline made more deposits during the post intervention period of August 11 – April 12. As shown in column 10, treatment clients’ with outstanding loan deposited Rs. 2.76 more than the control cli- ents, although this effect was absent at the beginning of the intervention. Columns 4 and 11 of Table 7 indicate that, treatment clients with non-FINO savings account made an additional de- posit of Rs. 2.58 compared to the control clients in the August11- April12 period, possibly sug- 27 gesting that financial literacy training is more effective for clients with pre-existing exposure to formal savings account, compared to those who have not been exposed. We also estimate the treatment heterogeneity effect on withdrawals and total transactions. Tables 8 & 9 present the heterogeneity effects. Table 8 shows that gender has no heterogeneity effect on treatment, as can be seen from the coefficient estimate in the post intervention period (column 8). Clients with formal loan outstanding make more withdrawals and total transactions in the post intervention period. Additionally, results reported in column 11 of tables 8 and 9 indicate that, having a non FINO formal savings account does not make any difference on client’s withdrawal and total transactions in the post intervention period. Finally, as a robustness check, we estimate a specification in which all interactions found to be statistically significant are simultaneously regressed on the dependent variable along with other key baseline controls. Tables 13- 18 present the results of such specification. The coefficient es- timates of this nested model grossly support the findings discussed in the previous section. 5. Conclusions Basic financial literacy is viewed as a critical step in enabling poor households to improve their financial status. Though there are studies indicating financial education can lead to increased awareness about financial products and services, there are limited studies evaluating the impact of financial education on financial behavior. However, financial literacy interventions, in the ab- sence of easy and secure access to formal financial services, might not be sufficient in generating changes in the financial behavior of beneficiaries. On the other hand, although innovative deliv- ery channels of financial services could help solve the access issue, low levels of financial litera- cy might result in sub-optimal use of formal financial products that are made available through these delivery channels. 28 In this paper, we explore whether financial literacy interventions could affect the usage of no- frills savings bank accounts that are made available at the door-step of low income households by one of the largest Business Correspondents in India. Using a unique dataset on transactions in savings account, we estimate the short run impacts of financial literacy training on account us- age. Results of our experimental study indicate persistent treatment effects where usage of no- frills savings account by the treatment group significantly increases in the post intervention peri- od. When controlling for heterogeneity, we find that, the treatment effect is more pronounced for female clients, while treatment clients who contracted outstanding loans at baseline, made more deposits and transactions in the post intervention period. Overall, the results suggest that finan- cial literacy education can increase usage of no-frills savings accounts and consequently could go a long way in improving financial inclusion for the poor. Furthermore, given FINO’s mandate to expand access and use of financial services to the poor, our results suggest that, an effective financial literacy training, coupled with adequate incentives for the bandhus, have the potential to influence savings behaviors, and hence, improve the financial welfare of low income families. 29 References Aportela, F. (1999). Effects of Financial Access on Savings by Low-Income People. (Unpublished doctoral dissertation) MIT Department of Economics, Boston. Ashraf, N., Karlan, D., & Yin, W. (2006). Tying Odysseus to the Mast: Evidence from a Commitment Savings Product in the Philippines. The Quarterly Journal of Economics, MIT Press, 121, 635-672. Ashraf, N., Karlan, D., & Yin, W. (2009). Female Empowerment: Impact of a Commitment Savings Product in the Phil- ippines. World Development, 38, 333-344. Beck, T., Demirguc-Kunt, A., & Peria, M. (2005). Reaching out: Access to and use of banking services across countries. World Bank Policy Research Working Paper 3754, The World Bank Group. Bruhn, M. & Love, I. (2009). The Economic Impact of Banking the Unbanked: Evidence from Mexico. World Bank Policy Research Working Paper 4981, The World Bank Group. Bruhn, M., & Zia, B. (2011). Stimulating Managerial Capital in Emerging Markets: The Impact of Business and Finan- cial Literacy for Young Entrepreneurs. World Bank Policy Research Working Paper 5642, The World Bank Group. Brune, L., Gine, X., Goldberg, J., & Yang, D. (2010). Commitments to Save: A Field Experiment in Rural Malawi. World Bank Policy Research Working Paper 5748, The World Bank. Burgess, R., & Pande, R. (2005). Do Rural Banks Matter? Evidence from the Indian Social Banking Experiment. Amer- ican Economic Review, 95, 780-795. Carpena, F., Cole, S., Shapiro, J., & Zia, B. (2011). Unpacking the Causal Chain of Financial Literacy. World Bank Pol- icy Research Working Paper 5798, The World Bank Group. Cole, S., & Shastry, G.K. (2008). If You Are So Smart, Why Aren’t You Rich? The Effects of Education, Financial Lit- eracy and Cognitive Ability on Financial Market Participation. Working Paper 09-071, Harvard Business School. Cole, S., Sampson, T., & Zia, B. (2009). Prices or Knowledge? What Drives Demand for Financial Services in Emerg- ing Markets?. Working Paper 09-117, Harvard Business School. Collins, D., Morduch, J., Rutherford, S., & Ruthven, O. (2009). Portfolios of the Poor: How the World's Poor Live on $2 a Day. Princeton, NJ: Princeton University Press. Duflo, E., & Saez, E. (2003). The Role of Information and Social Interactions in Retirement Plan Decisions: Evidence from a Randomized Experiment. Quarterly Journal of Economics, 118, 815-842. Duflo, E., Kremer, M., & Robinson, J. (2009). Nudging Farmers to Use Fertilizer: Theory and Experimental Evidence from Kenya. American Economic Review, 101, 2350-2390. Dupas, P., & Robinson, J. (2009). Savings Constraints and Microenterprise Development: Evidence from a Field Exper- iment in Kenya. Working Paper 14693, NBER. Dupas, P., Green, S., & Robinson, J. (2012). Challenges in Banking The Rural Poor: Evidence From Kenya’s Western Province. NBER Africa Project Conference Volume, forthcoming. Karlan, D., & Valdivia, M. (2011). Teaching Entrepreneurship: Impact of Business Training on Microfinance Clients and Institutions. Review of Economics and Statistics, 93, 510-27. Levine, R. (2005). Finance and Growth: Theory and Evidence. In P. Aghion & S. Durlauf (Ed.), Handbook of Economic Growth (pp. 865-934), Elsevier. Lusardi, A., & Mitchell, O. (2007). Baby Boomer retirement security: The roles of planning, financial literacy, and housing wealth. Journal of Monetary Economics 54, 205–224. Lusardi, A., & Mitchell, O. (2007). Financial Literacy and Retirement Planning: New Evidence from the Rand Ameri- can Life Panel. Working Paper 2007-157, University of Michigan Retirement Research Center. Regression Results Table 1: Amount of Deposits (monthly average - Trimmed at the 99pct) This note describes tables 1 to 6: Each column presents monthly averages of each variable using the periods that are indicated, periods which are named in reference to the implementation of the financial literacy training (during and post), which took place on May-Aug 2011. ’Transactions’ include both deposits and withdrawals. All regressions control for: Client’s gender, Client has secondary and above education, Household had a loan outstanding with formal sources, Household had a non-FINO savings bank account, Per capita total expenditure, Numeracy index, Number of female members in household, Amount of balances held as of May 11 in FINO savings account, An indicator variable of the Varanasi district and Standardized index of financial literacy at the baseline. Robust s.e. in parenthesis, clustered at the agent level. Levels of significance: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 (1) (2) DUR- POST ING Aug May- - Apr Aug Treatment 5.382∗∗∗ 2.006∗∗∗ (1.434) (0.664) Control 3.190 2.221 Table 2: Amount of Withdrawals (monthly average - Trimmed at the 99pct) (1) (2) DUR- POST ING Aug May- - Apr Aug Treatment 3.248∗∗∗ 1.208∗∗∗ (1.037) (0.401) Control 2.369 1.170 Table 3: Amount of Transactions (monthly average - Trimmed at the 99pct) (1) (2) DUR- POST ING Aug May- - Apr Aug Treatment 8.939∗∗∗ 2.717∗∗∗ (2.308) (0.848) Control 5.159 2.855 Table 4: Number of Deposits (monthly average - Trimmed at the 99pct) (1) (2) DUR- POST ING Aug May- - Apr Aug Treatment 0.130∗∗∗ 0.068∗∗∗ (0.036) (0.024) Control 0.101 0.091 Table 5: Number of Withdrawals (monthly average - Trimmed at the 99pct) (1) (2) DUR- POST ING Aug May- - Apr Aug Treatment 0.089∗∗∗ 0.036∗∗∗ (0.017) (0.011) Control 0.040 0.036 Table 6: Number of Transactions (monthly average - Trimmed at the 99pct) (1) (2) DUR- POST ING Aug May- - Apr Aug Treatment 0.235∗∗∗ 0.098∗∗∗ (0.042) (0.029) Control 0.102 0.103 Table 7: Heterogeneity Effect: Amount of Deposits (monthly average - Trimmed at the 99pct) This note describes tables 7 to 12: Each column presents monthly averages of each variable using the periods that are indicated, periods which are named in reference to the implementation of the financial literacy training (during and post), which took place on May-Aug 2011. ’Transactions’ include both deposits and withdrawals. Robust s.e. in parenthesis, clustered at the agent level. Levels of significance: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) DURING DURING DURING DURING DURING DURING DURING POST POST POST POST POST POST POST May- May- May- May- May- May- May- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug Aug Aug Aug Aug Aug Aug Apr Apr Apr Apr Apr Apr Apr Treatment 4.659∗∗ 5.985∗∗∗ 5.424∗∗∗ 4.362∗∗ 5.573∗∗∗ 5.819∗∗∗ 5.766∗∗∗ 1.044 1.372∗∗ 1.688∗∗ 0.551 2.785∗∗ 2.061∗∗∗ 1.997∗∗∗ (1.804) (1.408) (1.557) (1.763) (1.661) (1.515) (1.484) (0.713) (0.668) (0.692) (0.893) (1.071) (0.670) (0.675) Treatment X Female 2.947 2.296∗∗ (1.873) (1.008) Treatment X Secondary or above -0.376 2.852 (2.629) (1.945) Treatment X Had Loan 2.972 2.765∗ (3.363) (1.553) Treatment X Had Non-FINO sav acc 2.573 2.585∗∗ (2.156) (1.125) Treatment X Per cap Exp 0.001 -0.003 (0.005) (0.004) Treatment X numeracy 0.904 0.206 (1.251) (0.497) Treatment X financ literacy -0.484 0.322 (1.154) (0.471) Control Mean 3.165 3.165 3.162 3.162 3.173 3.177 3.162 2.204 2.204 2.198 2.198 2.206 2.208 2.198 Table 8: Heterogeneity Effect: Amount of Withdrawals (monthly average - Trimmed at the 99pct) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) DURING DURING DURING DURING DURING DURING DURING POST POST POST POST POST POST POST May- May- May- May- May- May- May- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug Aug Aug Aug Aug Aug Aug Apr Apr Apr Apr Apr Apr Apr Treatment 1.173 2.521∗∗ 2.719∗∗ 1.046 3.090∗∗ 3.123∗∗∗ 3.037∗∗∗ 0.848∗ 1.054∗∗∗ 0.869∗∗ 0.834 1.212∗∗∗ 1.197∗∗∗ 1.187∗∗∗ (1.466) (0.996) (1.173) (1.241) (1.246) (1.135) (1.120) (0.459) (0.401) (0.397) (0.506) (0.429) (0.409) (0.412) Treatment X Female 4.615∗∗∗ 0.806 (1.732) (0.635) Treatment X Secondary or above 2.469 0.578 (2.553) (1.017) Treatment X Had Loan 2.031 2.687∗ (3.416) (1.428) Treatment X Had Non-FINO sav acc 3.502∗∗ 0.609 (1.654) (0.699) Treatment X Per cap Exp -0.000 -0.000 (0.005) (0.001) Treatment X numeracy 0.667 0.030 (0.844) (0.286) Treatment X financ literacy 0.061 -0.083 (0.853) (0.312) Control Mean 2.350 2.350 2.344 2.344 2.352 2.355 2.344 1.161 1.161 1.159 1.159 1.163 1.164 1.159 Table 9: Heterogeneity Effect: Amount of Transactions (monthly average - Trimmed at the 99pct) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) DURING DURING DURING DURING DURING DURING DURING POST POST POST POST POST POST POST May- May- May- May- May- May- May- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug Aug Aug Aug Aug Aug Aug Apr Apr Apr Apr Apr Apr Apr Treatment 5.274∗ 8.503∗∗∗ 8.195∗∗∗ 5.765∗∗ 8.088∗∗∗ 8.953∗∗∗ 8.789∗∗∗ 1.589∗ 2.192∗∗∗ 2.048∗∗ 1.457 3.051∗∗∗ 2.683∗∗∗ 2.655∗∗∗ (3.102) (2.172) (2.567) (2.624) (2.599) (2.491) (2.454) (0.933) (0.799) (0.837) (1.068) (0.959) (0.858) (0.863) Treatment X Female 8.848∗∗∗ 2.526∗ (3.350) (1.332) Treatment X Secondary or above 1.887 2.056 (4.824) (2.193) Treatment X Had Loan 4.156 5.089∗∗ (6.569) (2.546) Treatment X Had Non-FINO sav acc 5.358 2.104 (3.398) (1.414) Treatment X Per cap Exp 0.002 -0.002 (0.009) (0.002) Treatment X numeracy 0.783 -0.017 (1.793) (0.570) Treatment X financ literacy -0.665 -0.086 (1.873) (0.604) Control Mean 5.118 5.118 5.111 5.111 5.128 5.135 5.111 2.833 2.833 2.827 2.827 2.837 2.839 2.827 Table 10: Heterogeneity Effect: Number of Deposits (monthly average - Trimmed at the 99pct) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) DURING DURING DURING DURING DURING DURING DURING POST POST POST POST POST POST POST May- May- May- May- May- May- May- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug Aug Aug Aug Aug Aug Aug Apr Apr Apr Apr Apr Apr Apr Treatment 0.100∗∗ 0.151∗∗∗ 0.128∗∗∗ 0.107∗∗ 0.165∗∗∗ 0.139∗∗∗ 0.137∗∗∗ 0.046∗ 0.083∗∗∗ 0.066∗∗∗ 0.060∗∗ 0.071∗∗ 0.072∗∗∗ 0.067∗∗∗ (0.039) (0.040) (0.039) (0.046) (0.050) (0.038) (0.038) (0.028) (0.026) (0.025) (0.031) (0.028) (0.025) (0.025) Treatment X Female 0.094∗∗ 0.057∗ (0.045) (0.031) Treatment X Secondary or above -0.061 -0.058∗ (0.047) (0.032) Treatment X Had Loan 0.089∗ 0.025 (0.050) (0.038) Treatment X Had Non-FINO sav acc 0.053 0.017 (0.046) (0.038) Treatment X Per cap Exp -0.000 -0.000 (0.000) (0.000) Treatment X numeracy 0.037 0.001 (0.024) (0.015) Treatment X financ literacy 0.026 -0.019 (0.028) (0.016) Control Mean 0.100 0.100 0.100 0.100 0.101 0.101 0.100 0.090 0.090 0.090 0.090 0.090 0.090 0.090 Table 11: Heterogeneity Effect: Number of Withdrawals (monthly average - Trimmed at the 99pct) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) DURING DURING DURING DURING DURING DURING DURING POST POST POST POST POST POST POST May- May- May- May- May- May- May- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug Aug Aug Aug Aug Aug Aug Apr Apr Apr Apr Apr Apr Apr Treatment 0.078∗∗∗ 0.099∗∗∗ 0.083∗∗∗ 0.084∗∗∗ 0.117∗∗∗ 0.092∗∗∗ 0.090∗∗∗ 0.026∗∗ 0.036∗∗∗ 0.032∗∗∗ 0.029∗∗ 0.044∗∗∗ 0.036∗∗∗ 0.035∗∗∗ (0.018) (0.018) (0.017) (0.022) (0.022) (0.017) (0.017) (0.011) (0.013) (0.012) (0.013) (0.012) (0.012) (0.011) Treatment X Female 0.032 0.024∗ (0.020) (0.013) Treatment X Secondary or above -0.041∗ -0.006 (0.021) (0.013) Treatment X Had Loan 0.060∗∗ 0.029 (0.025) (0.020) Treatment X Had Non-FINO sav acc 0.011 0.013 (0.021) (0.014) Treatment X Per cap Exp -0.000∗ -0.000 (0.000) (0.000) Treatment X numeracy 0.018 0.004 (0.011) (0.006) Treatment X financ literacy 0.004 -0.002 (0.011) (0.007) Control Mean 0.039 0.039 0.039 0.039 0.040 0.040 0.039 0.036 0.036 0.036 0.036 0.036 0.036 0.036 Table 12: Heterogeneity Effect: Number of Transactions (monthly average - Trimmed at the 99pct) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) DURING DURING DURING DURING DURING DURING DURING POST POST POST POST POST POST POST May- May- May- May- May- May- May- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug Aug Aug Aug Aug Aug Aug Apr Apr Apr Apr Apr Apr Apr Treatment 0.219∗∗∗ 0.263∗∗∗ 0.230∗∗∗ 0.216∗∗∗ 0.288∗∗∗ 0.246∗∗∗ 0.243∗∗∗ 0.075∗∗ 0.103∗∗∗ 0.092∗∗∗ 0.073∗∗ 0.117∗∗∗ 0.100∗∗∗ 0.097∗∗∗ (0.042) (0.048) (0.045) (0.052) (0.059) (0.044) (0.044) (0.030) (0.033) (0.031) (0.035) (0.033) (0.031) (0.031) Treatment X Female 0.058 0.057 (0.051) (0.038) Treatment X Secondary or above -0.098∗ -0.025 (0.056) (0.035) Treatment X Had Loan 0.107∗ 0.058 (0.062) (0.048) Treatment X Had Non-FINO sav acc 0.045 0.046 (0.053) (0.044) Treatment X Per cap Exp -0.000 -0.000 (0.000) (0.000) Treatment X numeracy 0.055∗ 0.008 (0.029) (0.017) Treatment X financ literacy 0.034 -0.009 (0.033) (0.018) Control Mean 0.101 0.101 0.102 0.102 0.102 0.102 0.102 0.103 0.103 0.103 0.103 0.103 0.103 0.103 Table 13: Heterogeneity Effect: Amount of Deposits (monthly average - Trimmed at the 99pct) This note describes tables 13 to 18: Each column presents monthly averages of each variable using the periods that are indicated, periods which are named in reference to the implementation of the financial literacy training (during and post), which took place on May-Aug 2011. ’Transactions’ include both deposits and withdrawals. All regressions control for: Client’s gender, Client has secondary and above education, Household had a loan outstanding with formal sources, Household had a non-FINO savings bank account, Per capita total expenditure, Numeracy index, Number of female members in household, Amount of balances held as of May 11 in FINO savings account, An indicator variable of the Varanasi district and Standardized index of financial literacy at the baseline. Robust s.e. in parenthesis, clustered at the agent level. Levels of significance: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 (1) (2) DURING POST Aug- May-Aug Apr Treatment 2.553 -1.448 (2.060) (1.161) Treatment X Female 2.015 3.117∗∗∗ (1.928) (1.138) Treatment X Secondary or above -1.402 3.232 (2.502) (1.982) Treatment X Had Loan 2.697 2.695∗ (3.009) (1.534) Treatment X Had Non-FINO sav acc 3.750∗ 2.067∗ (2.052) (1.072) Control Mean 3.190 2.221 Table 14: Heterogeneity Effect: Amount of Withdrawals (monthly average - Trimmed at the 99pct) (1) (2) DURING POST Aug- May-Aug Apr Treatment -1.111 0.224 (1.633) (0.674) Treatment X Female 3.562∗∗ 0.783 (1.495) (0.717) Treatment X Secondary or above 2.549 0.743 (2.456) (1.104) Treatment X Had Loan 1.766 2.620∗∗ (2.698) (1.316) Treatment X Had Non-FINO sav acc 3.848∗∗ 0.398 (1.556) (0.662) Control Mean 2.369 1.170 Table 15: Heterogeneity Effect: Amount of Transactions (monthly average - Trimmed at the 99pct) (1) (2) DURING POST Aug- May-Aug Apr Treatment 2.105 -0.409 (3.257) (1.308) Treatment X Female 6.270∗∗ 2.862∗∗ (2.985) (1.401) Treatment X Secondary or above 2.080 2.764 (4.612) (2.310) Treatment X Had Loan 3.988 4.954∗∗ (5.356) (2.408) Treatment X Had Non-FINO sav acc 6.140∗ 1.445 (3.235) (1.344) Control Mean 5.159 2.855 Table 16: Heterogeneity Effect: Number of Deposits (monthly average - Trimmed at the 99pct) (1) (2) DURING POST Aug- May-Aug Apr Treatment 0.063 0.049 (0.059) (0.030) Treatment X Female 0.076 0.044 (0.052) (0.032) Treatment X Secondary or above -0.062 -0.053 (0.051) (0.034) Treatment X Had Loan 0.089∗ 0.023 (0.050) (0.038) Treatment X Had Non-FINO sav acc 0.076 0.020 (0.047) (0.037) Control Mean 0.101 0.091 Table 17: Heterogeneity Effect: Number of Withdrawals (monthly average - Trimmed at the 99pct) (1) (2) DURING POST Aug- May-Aug Apr Treatment 0.074∗∗∗ 0.018 (0.024) (0.014) Treatment X Female 0.021 0.025∗ (0.022) (0.013) Treatment X Secondary or above -0.044∗ 0.001 (0.023) (0.013) Treatment X Had Loan 0.056∗∗ 0.028 (0.025) (0.021) Treatment X Had Non-FINO sav acc 0.019 0.008 (0.021) (0.013) Control Mean 0.040 0.036 Table 18: Heterogeneity Effect: Number of Transactions (monthly average - Trimmed at the 99pct) (1) (2) DURING POST Aug- May-Aug Apr Treatment 0.204∗∗∗ 0.053 (0.058) (0.035) Treatment X Female 0.026 0.056 (0.055) (0.039) Treatment X Secondary or above -0.120∗∗ -0.017 (0.059) (0.037) Treatment X Had Loan 0.106∗ 0.055 (0.062) (0.048) Treatment X Had Non-FINO sav acc 0.069 0.037 (0.053) (0.042) Control Mean 0.102 0.103 Appendix A: Power Calculations In statistics, power is the ability to identify if a program has impact. A concern with any evalua- tion is if we falsely reject an impact because of low statistical significance. This can happen if the effect of the program is small, and the number of people interviewed is also small. With any evaluation, it is important in the design phase to attempt to avoid being “under powered”, i.e. having too few observations to detect an effect. Based on previous unpublished evaluations of financial literacy training, the impact on individ- ual’s knowledge of financial tools is expected to be very high, while the impact on behaviors and wellbeing is expected to be very low, though potentially still of an important size. Power calculations were done in the program Optimal Design and assume that the pro- gram will change behaviors by between 10% and 15% with a power of 0.8 and significance level of 0.05. Based on these calculations, a conservative number of individuals to follow in both treatment and control villages was determined to be 15 per village, thus requiring 15*200=3000 individuals to follow. 1 Appendix B: Balance Test Table B1: Results of Balance Test Control Treatment Balanced Variables Mean Mean P value at 10% Number of members in the household 6.74 6.96 0.17 Yes Number of female household members 3.18 3.35 0.06 No Number of male household members 3.56 3.61 0.63 Yes Number of minors in the household 2.71 2.86 0.08 No Number of adults >=18 in the household 4.03 4.1 0.49 Yes Dummy: Head of household has at least secondary education 0.23 0.2 0.19 Yes Dummy: Head of household is illiterate 0.43 0.43 0.90 Yes Age of Head of household 44.68 45.52 0.20 Yes Dummy: client has at least secondary education 0.27 0.21 0.01 No Dummy: client is female 0.39 0.42 0.11 Yes Dummy: client is the head of household 0.44 0.42 0.32 Yes Client age 37.61 38.18 0.34 Yes Whether belong to general caste Dummy 0.11 0.13 0.63 Yes Whether belong to schedule caste Dummy 0.3 0.35 0.16 Yes Whether belong to schedule tribe Dummy 0.04 0.04 0.53 Yes Whether belong to other backward community Dummy 0.54 0.49 0.17 Yes Whether religion is Hindu Dummy 0.95 0.94 0.77 Yes Whether religion is Muslim Dummy 0.05 0.06 0.80 Yes Whether has land Dummy 0.75 0.75 0.89 Yes Total landholding 24.45 25.54 0.42 Yes Asset Index 1st Principal component 0.02 -0.02 0.45 Yes Dummy for having a non-FINO savings/post office bank account at baseline 0.56 0.55 0.94 Yes Total amount of formal savings 2182.82 2117.05 0.90 Yes Total amount of savings 5522.19 4297.28 0.34 Yes Dummy for having a loan outstanding with formal sources Bank/MFI/SHG 0.09 0.12 0.02 No Total outstanding formal loan amount 1661.89 1859.35 0.63 Yes Per capita Household income: 14 days prior to survey, capped at 99 percen- tile 154.35 147.3 0.66 Yes Per-capita Household expenditures: 14 days prior to survey, capped at 99 percentile 271.43 249.26 0.05 No Whether plan to save for upcoming expenses Dummy 0.84 0.85 0.69 Yes Normalized index of competency in numeracy 0.08 -0.08 0.03 No Normalized index of financial literacy 0.08 -0.08 0.03 No Whether client is risk averse 0.47 0.44 0.11 Yes Whether client is patient 0.26 0.24 0.46 Yes Amount of Balance held in FINO account as of May 2011 Rs. 8.23 14.78 0.15 Yes 2