BruneGineGoldbergYang MalawiSavingsJDE

Transcript

1 ∗ Commitments to Save: A Fiel d Experiment in Rural Malawi Lasse Brune Department of Economics, University of Michigan Xavier Giné Development Economics Research Group, World Bank is and Development (BREAD) and Bureau for Economic Analys Jessica Goldberg Department of Economics, University of Maryland Dean Yang of Economics, University of Michigan, Ford School of Public Policy and Department Bureau for Economic Analysis and Development (BREAD), and National Bureau of Economic Research (NBER) February 2013 Abstract randomly offered Malawian smallholder farmers We report the results of a field experiment that y treatments, offering e ither: 1) “ordinary” formal savings accounts. We tested two primar accounts, or 2) both ordinary and “commitm ent” accounts. Commitment accounts allowed customers to restrict access to their own funds until a future date of their choosing. A control group was not offered any account but was tracked alongside the treatment groups. The commitment treatment led to increases in deposits at the partner bank, and over the next agricultural year cause d increases in agricultural input use, crop sales, and household expenditures. The effects of the commitment trea tment are not due to literally “tying the hands” of farmers, since farmers in that treatment mo stly saved in ordinary accounts (rather than commitment accounts). We discuss other possi ble channels through which the commitment treatment’s effects may have opera ted, such as reduced sharing with one’s social network as well as other psychological channels. ences, self-control, sh aring norms, mental Keywords: savings, commitment, hyperbolic prefer accounting. JEL codes: D03, D91, O16, Q14. ∗ Brune: [email protected] . Giné: [email protected] . Goldberg: [email protected] Yang: . We thank Niall Keleher, Lutamyo Mwamlima a nd the IPA staff in Malawi; Steve Mgwadira, [email protected] Mathews Kapelemera, and Webster Mbekeani of OIBM; and the OIBM management and staff of Kasungu, Mponela and Lilongwe branches. Matt Basilico and Britni Must provi ance. We are grateful to ded excellent research assist Beatriz Armendariz, Orazio Attanasio, Oriana Bandiera, Abhijit Banerjee, Luc Behagel, Marcel Fafchamps, Maitreesh Ghatak, Marc Gurgand, Sylvie Lambert, Kim Lehrer, Rocco Macchiavello, Lou Maccini, Sharon Maccini, Marco Manacorda, Cost as Meghir, Rohini Pande, Albert Park, Imran Rasul, Chris Woodruff, Bilal Zia, Andrew Zeitlin, and seminar participants at the FAI Micr ofinance Innovation Conference, Ohio State, London School of Economics, Warwick, Institute for Fiscal Stud ies, Paris School of Economics, and Oxford for helpful comments. We appreciate the support of David Rohrbach (World Bank) and Jake Kendall (Bill & Melinda Gates Foundation). We are grateful for research funding from the World Bank Research Committee and the Bill & Melinda Gates Foundation. The views expressed in this paper are those of the authors and should not be attributed to the World Bank, its executive directors, or the countries they represent. 0

2 1. Introduction Recent experimental studies have found high marginal returns to capital in developing nd Woodruff, 2008; Fafchamps et rprises (de Mel, McKenzie a countries in non-agricultural ente emer and Robinson, 2008). These high returns stand al., 2011) as well as in agriculture (Duflo, Kr in contrast to low utilization of modern inputs such as fertilizer in many low-income countries, particularly in sub-Sahara n Africa (World Bank, 2008). governments and donors many developing country To raise input utilization in agriculture, have implemented large-scale input subsidies. However, the scale of such programs takes a 1 Another ng doubt on their long-term sustainability. heavy toll on government budgets, casti microcredit programs. In 2009, the Microcredit popular response has been the introduction of Summit estimated that there were more than 3,500 microfinance institutions around the world with 150 million clients (Daley-Harris 2009). Wh ile these outreach numbers are impressive, microcredit today is largely devoted to non-ag ricultural activities (M orduch 1999; Armendariz 2 agricultural lending. challenges inherent in de Aghion and Morduch 2005) due to the substantial Given the limited supply of credit for agricultur e, many donors and academics (for example, Deaton, 1990; Robinson, 2001 and more recently th e Bill and Melinda Gates Foundation) have 3 easing access to formal savings. emphasized the potential for incr Low-income individuals, however, have diffi culty saving in formal banking institutions. nsive and riskier methods to sa Instead, they rely on more expe ve informally (Rutherford, 2000 and Collins, Morduch, Rutherford and Ruthven, 2009). These alternatives include cash held at home, purchases of durable assets such as livest ock with risky returns, participation in ROSCAs or the use of deposit collectors (such as (rotating savings and credit associations), collectors susu in West Africa). 1 For example, the cost of Malawi’s large-scale fertili zer subsidy program amounted to 11 percent of the total government budget in the 2010-11 fiscal year. 2 personal identification leads to asymmetric information Giné, Goldberg, and Yang (2012) find that imperfect problems (both adverse selection and moral h azard) in the rural Ma lawian credit market. 3 Aportela (1999) uses data from an expansion of branches set up in post offices in the end of 1993. He finds that the expansion resulted in an averag e increase in savings rate of 3 to 5 percentage points, with higher effects (up to 7 percentage points) for low-income individuals compared to other low-income households located in towns without the expansion. Burgess and Pande (2005) find that a policy-driven expansion of rural banking reduced poverty in India, and provide sugges tive evidence that deposit mobilization and credit access were intermediating channels. Despite positive social effects, the program was discontinued in 2001 due to losses from defaults. Bruhn and Love (2009) examine the opening of bank branches in consumer durable stores in Mexico in 2002 and find an increase in the number of informal business owners by 7.6 percent, in total employment by 1.4 percent, and in average income by about 7 percent. The effects are concentrated among low income households and in municipalities with lower pre-existing bank penetration. 1

3 A number of explanations have been advan ced for low levels of formal savings in savings may be high for a variety of reasons, developing countries. Transaction costs for formal including substantial distances to branches, costly and unreliable transport, and mistrust towards event households from ncial illiteracy may pr formal financial institutions. In addition, fina the benefits of formal savings and lack of opening accounts due to a lack of knowledge about familiarity with account-opening procedur es (Cole, Sampson and Zia, 2011). Psychological factors, such as impatience (a strong preference for the present over the future) and issues of se lf-control (competing prefer ences that dictate differe nt actions at different is evidence from both de times) may also lead to lower savings. There veloped and developing countries that self-aware indivi r options in anticipation of future self- duals seek to limit thei 2006) investigate demand for and impacts of a control problems. Ashraf, Karlan, and Yin ( and find that demand for such commitment commitment savings device in the Philippines ng present-biased time preferences. Duflo, devices is concentrated among women exhibiti a small, time-limited discount on fertilizer Kremer and Robinson (2011) find that offering immediately after harvest has an effect on fertilizer use that is comparable to that of much larger discounts offered later, around planting time. Giné et al. (2012) find that Malawian tobacco farmers with present-biased preferences are more likely to revise a plan about how to use future income, even when that plan is made under commitment. Another potential explanation for low savings leve ls in rural communities is the pressure to ee, e.g., Anderson and Baland 2 share income with spouses (s 002; Ashraf 2009; Schaner, 2012), relatives and friends (see, e.g., Platteau, 2000; Maranz, 2001; Ligon, Thomas, and Worall, 2002; Hoff and Sen, 2006; Baland, Guirkinger and Mali, 2011; Jakiela and Ozier, 2011). Sharing erting effort or accumulating assets, and may obligations may discourage individuals from ex encourage them to spend resources hastily before income is dissipated through demands from others. People who anticipate pressure to share cash with others in th eir social network may spend that money quickly in order to pre- empt requests for transfers (Goldberg 2011). An important point is that comm itment devices, by tying the hands of individuals, may also make it easier to resist demands for sharing with their social network. In other words, commitment devices may assist with “other”-co ntrol as well as self-c ontrol problems. The existing literature has only partially investig ated whether the demand for and impact of commitment devices is (at least in part) due to other-control problem s (Hertzberg 2010). This discussion brings to the fore three interrelated que stions that are the fo cus of this paper. First, does merely offering easy access to form al savings accounts improve savings and other 2

4 household outcomes? Second, are such impacts magnified when the savings accounts offered have commitment features, such as an option to own ability to make voluntarily restrict one’s if offering accounts with commitment features withdrawals for a defined period of time? Third, ough which the effect operates? leads to larger impacts, what is the underlying mechanism thr field experiment among smallholder cash crop To answer these questions, we implemented a farmers in Malawi. We are able to shed light most clearly on the first tw o questions. With respect to the third question (on mechanisms of commitm ent impacts), we provide clear evidence against the self-control channel, discuss some evidence on the other-control channel, and speculate on other psychological channels that might be at work. p with a local microfinance institution, we In our experiment, conducted in partnershi randomized offers of account-opening and deposit assistance for formal savings accounts. One randomly-selected group of farmers was simply o individual “ordinary” ffered assistance opening features. This treatment sheds savings accounts with standard light on the impact of simply facilitating access to savings accounts. To test the importan ce of offering accounts with group of farmers was offered, in addition to the commitment features, another randomly-selected ount that allowed account holders to request “ordinary” account, a “commitment” savings acc that funds be frozen until a specified date (e.g., until the next planting season, so that funds could be preserved for farm input purchases). Other fa rmers were randomly assigned to a control group that was surveyed but not offered assistance with opening either type of savings account. This design allows us to test the re lative impact of offering accounts with commitment features versus savings accounts. offering only ordinary pressure to share with one’s social network We designed a sub-experiment to test whether reduces savings. Among farmers who were offered the savings treatments, we cross-randomized dividual savings account balances. If the public an intervention that provided a public signal of in 4 revelation of balances induces greater pressure balances may be lower. to share, then saving Our findings are distinguished from those in the existing literature in two ways. First, we are among the first to show impacts of commitment savi ngs offers (as opposed to offers of ordinary 5 accounts) on important economic outcomes beyond savings. Previous research has often 4 Flory (2011) conducts a field experiment in rural Malawi where households in treatment villages were encouraged to open savings accounts. He finds that transfers to poor households increase in treatment villages, perhaps because everyone in the village knew who had savings accounts and thus access to funds. 5 As a follow-up to Ashraf, Karlan, and Yin (2006), Ashraf, Karlan, and Yin (2010) show impacts of commitment account offers on fema le empowerment in the same Ph ilippine experimental sample. 3

5 focused on the mechanical effects of savings prod ucts on levels of savings, but we have longer directly. The commitment treatment had run outcomes that measure economic impacts more and withdrawals at our partner large positive effects on a range of outcomes of interest: deposits under cultivation (an increase the next planting season, land institution immediately prior to use in that planting (27.4% group mean), agricultural input amounting to 9.8% of the control in the subsequent harvest (21.8% increase), increase over the control group mean), crop output increase). While the and household expenditures in the months immediately after harvest (17.4% ordinary treatment’s effect on deposits and with drawals was similar to that of the commitment treatment effect, the ordinary treatment effects inputs and subsequent outcomes on agricultural are uniformly smaller than those of the comm itment treatment, and are never statistically the impact of the commitment significantly different from zero. A joint hypothesis test finds that nditure outcomes is sta tistically significantly account offer on the set of agricultural and expe the ordinary account offer. larger than the effect of demonstrate that if an offer of commitment The second key contribution of this paper is to savings accounts has substantial impacts (e.g., on later outcomes such as investment and household productive output), it doe s not have to operate via so lving individuals’ self-control problems. The basic facts in our experiment are striking: the vast majority (89.0%) of deposits among individuals offered commitment accounts were in ordinary as opposed to commitment accounts. The average amount deposited in co mmitment accounts was about an order of magnitude smaller than the commitment treatme nt’s later impact on input use. Clearly, the commitment treatment did not have its impact so lely by literally “tying the hands” of farmers by preventing them from withdrawi planting time. Impacts on total ng money in the months prior to deposits (in ordinary and commitment accounts co mbined), by contrast, do exceed the impact on later reported increases in inputs, so the measur ed increase in inputs could have been funded by total deposited funds (just not by the funds deposited into commitment accounts alone). Through what other channels might the effects of the commitment treatment have operated? We explore two alternative mechanisms. First, the commitment treatment could have helped farmers solve “other-control” problems, by allowing them to better resist social network 6 sive evidence in support of this As it turns out, we do not find conclu demands for their savings. 6 Even though only a small minority of deposits went into commitment accounts, farm ers might have been able to claim to others in their social network that their funds were tied up, since the distribution of funds across ordinary and commitment accounts was not public knowledge. The cross-randomized raffle treatments awarded raffle tickets 4

6 hypothesis. The commitment treatment did not redu ce reported transfers to other households; in created public revelation of savi ngs balances did not lead to addition, the sub-experiment that 7 That said, it is still possible that the commitment treatment allowed lower savings as expected. study participants to keep funds from others wi thin the household, or to refrain from consuming resources early in anticipation of future requests from others (as in Goldberg 2011). We therefore an important focus in future research. believe the other-control channel should remain Second, the commitment treatment may have led to changes in behavior via other psychological channels. In the commitment treatm ent we asked farmers to specify in advance how much money from their crop sa deposited into their ordinary les they wanted to be directly and commitment accounts. This mere elicitation of farmers’ intentions may have influenced their bb and Sheeran 2006, Zwane et al, 2011). later behavior (Feldman and Lynch 1988, We deposited into commitment accounts may have Relatedly, the act of stating amounts to be 1990), although the accounts were not actively created a investment mental account (Thaler, labeled. Unfortunately, we can offer no direct evidence to support or contradict that such psychological channels may have been at work. Fu ture research should prioritize investigation of these and potentially othe r psychological channels. This paper contributes to the burgeoning literature on the effects of formal savings accounts, and in particular of making offers of commitment savings. Dupas and Robinson (2012a) offer ordinary savings accounts to finding positive impacts on Kenyan urban entrepreneurs, by contrast, we test the investment and income for women. In this paper, differential impacts of offering commitment savings versus ordinary sa vings accounts. Prina (2011) finds that random ds in Nepal leads to increases in financial assignment of basic savings account access to househol stments. Atkinson et al. (2010) offer microcredit borrowers in assets and in human capital inve Guatemala savings accounts with different f eatures, including reminders about a monthly commitment to save and a default of 10% of loan repayment as a suggested monthly savings target. They find that both features increase sa vings balances substantially. Dupas and Robinson (2012b) test the impact of comm vings in western Kenyan ROSCAs; itment features for health sa on the basis of total funds across all accounts, so this treatment also did not reveal how much was saved in commitment accounts. 7 The public revelation treatment may have had little e ffect because withdrawals from the accounts occurred earlier than we had expected. Public revelation of balances occurred after most funds had already been withdrawn, which likely led to substantially attenuated effects. We ther efore cannot rule out that public revelation of savings balances may have had significant effects if it had occurred earlier in time. 5

7 their qualitative findings from a post-intervention survey are s uggestive of a mental accounting channel. follows. Section 2 explains the study design and The remainder of this paper is organized as le. Section 3 describes the estimation strategy. briefly describes the characteristics of the samp Section 4 presents the main empirica l results and Sec tion 5 concludes. 2. Experimental design and survey data The experiment was a collaborative effort of Opportunity Internat ional Bank of Malawi (OIBM), Alliance One, Limbe Leaf, the Univ ersity of Michigan and the World Bank. l is a private microfinance institu tion operating in 24 countries that Opportunity Internationa e and Limbe Leaf are two large private agri- offers savings and credit products. Alliance On and high-quality inputs to smallholder farmers business companies that offer extension services 8 Farmers in the study were organized by the tobacco via an out-grower tobacco scheme. companies into clubs of 10-15 members and a ll had group liability tobacco production loans the central Malawi region we study, tobacco from OIBM prior to enrollment in the study. In 9 to those of non-tobacco-producing households. farmers have similar poverty and income levels While all farmers in the study were loan customer s of OIBM at the start of the project, the loans provided a fixed input package that for the majority of farmers fell short of optimal levels 10 of fertilizer use on their tobacco plots. This is important because it suggests that there is room for a savings intervention to in tion, while a minority of farmers crease input utilization. In addi was using optimal levels of fertilizer at baseline, even such farmers could use savings generated by the intervention to obtain addi r tobacco cultivation, or shift tional inputs and expand land unde 8 Tobacco is central to the Malawian economy, as it is the country’s main cas h crop. About 70% of the country’s foreign exchange earnings come from tobacco sales, and a large share of the labor force works in tobacco and related industries. 9 Based on authors’ calculations from the 2004 Malawi Integrated Household Survey (IHS), individuals in tobacco farming rural househol ds in central Malawi live on PPP$1.48/day on average, while the average for central Malawian rural households overall is PPP$1.51/day. 10 area. As a result, 60.4% of farmers we re applying The input package was designed for a smaller cultivated less than the recommended amount of nitrog en on their tobacco plots at baselin e. The figures for the two other key nutrients for tobacco are even more striking: 83.2% and 84.7% of farmers used less than the recommended amount of phosphorus and potassium, respectively. For each of the three nutrient s, among farmers using less than recommended levels, the mean ratio of actual use to optimal use was about 0.7. Optimal use levels were determined by Alliance One and Limbe Leaf in collaboration with Malawi’s Agricultural Research and Extension Trust (ARET), and are similar to nutrie nt level recommendations in the Un ited States (Pearce et al. 2011). 6

8 land from other crops towards toba cco. Finally, the savings interven tion could also affect use of 11 in staple crop in Ma fertilizer and other inputs on maize (the ma lawi) and other crops. e household and farmer club characteristics. Table 1 presents summary statistics of baselin All variables expressed in money terms are in Malawi Kwacha (MK145/USD during the study period). Baseline survey respondents own an average of 4.7 acres of land and are mostly male (only six percent were female). Respondents are on average 45 years old. They have an average 12 Sixty three percent ve low levels of financial literacy. of 5.5 years of formal education, and ha 13 The average with a formal bank (mostly with OIBM). of farmers at baseline had an account baseline in bank accounts was MK 2,083 (USD 14), reported savings balance at the time of the d in the form of cash at home. with an additional MK 1,244 (USD 9) save Figure 1 presents the timing of the experiment with reference to the Malawian agricultural were administered in April and May 2009, season. The baseline survey and interventions immediately before the 2009 harvest. Financial Education Session After the baseline was administered, all clubs (ordinary and commitment treatments as well as control) attended a financial education session that reviewed basic elements of budgeting and explained the benefits of formal savings accounts, in particular how they could be used to set aside funds for future expenses. The full script of the financial education session can be found in Appendix A. The financial education session was deliberatel y provided to both treatment and control groups so that treatment effects could be attributed solely to the provision of the financial education (for example, strategies for improved products, abstracting from the effect of financial r. For this reason, we can estimate neither budgeting) implicitly provided during the product offe 11 At baseline, 89.5% and 99.9% of farmers were applying less than the recommended amount of nitrogen and phosphorus, respectively, on their maize plots. Among farmers applying less than the recommended amount of nitrogen (phosphorus) on maize, the ratio of actual use to optimal use was 0.48 (0.14). Potassium is not recommended for maize. Nutrient recommendations for are from Benson (1999). maize in central Malawi 12 In particular, 42% of respondents were able to compute 10% of 10,000, 63% were able to divide MK 20,000 by five and only 27% could apply a yearly interest rate of 10% to an initial balance to compute the total savings balance after a year. 13 This number includes a nu mber of “payroll” accounts opened in a previous season by OIBM and one of the tobacco buyer companies as a payment system for crop proceeds, and which do not actually allow for savings accumulation. Our baseline survey unfortunately did not properly distinguish be tween these two types of accounts. 7

9 the impact of the ordinary and commitment treatm ents without such financial education, nor the 14 impact of the financial education alone. Ordinary and Commitment Treatments ree savings treatment conditions. To minimize Farmers were randomly assigned to one of th carried out at the level of farmer clubs (of cross-treatment contamination, randomization was which there were 299, described further below). The first experimental group was the control group and only received the financial education session described above. vantage of the existing system of depositing Implementation of the savings treatment took ad crop sale proceeds into OIBM bank accounts. In the control group, the process followed the status quo, as follows. At harvest, farmers sold their tobacco to the company at the price 15 The proceeds from the sale were then prevailing on the nearest tobacco auction floor. ducted the loan repaym electronically transferred to OIBM, which de ent (plus fees and surcharges) of all borrowers in the club, and then credited th e remaining balance to a club account at OIBM. Club members authorized to acce ss the club account (usually the chairman or the treasurer) came to OIBM branches and withdrew the funds in cash. Farmers in the savings treatment groups were given the same financial education session provided to the control group, and in addition we re also given account opening assistance and offered the opportunity to have their harvest proceed s (net of loan repayment) directly deposited into individual accounts in their names (see Figure 2 for a schematic illustration of the money flows). After their crop was sold, farmers traveled to the closest OIBM branch to confirm that the club level. Authorized members of the positive proceeds net of repayment were available at clubs (often together with other club members) then filled out a sheet specifying the division of the total amount between farmers. Depending on whether a club member had opted for the were then either transferred to the individual’s account(s) individual accounts or not, funds rvention, prior to the harvest, or paid out in cash. opened at the time of the inte There were two savings treatme nt conditions. In the first, farmers were offered only an ordinary savings account (the “ord inary” treatment). In the sec ond, farmers were offered both an 14 Ghana where eligible individuals who already have a Karlan et al. (2012) conduct a field experiment in savings account are allowed to open and label a second account. They find that sa vings in this group is 31 % larger. A subset of the individuals that opened the second account was asked to state a savings goal for this account, but they find that setting a savings goal had no impact on savings balance, suggesting that it was the opening of the second account and labeling it what was driving the results. 15 The tobacco growing regions are di vided among the two tobacco buyer comp anies. In their coverage area each buyer company organizes farmers into clubs an d provides them with basic extension services. 8

10 ordinary and a commitment sa vings account (the “commitment” treatment). Farmers who chose required to have an ordinary account where to open a commitment savings account were also uncommitted funds would be deposited. Farmers in the control group and the “ordinary” treatment group who could have learned about and requested commitment accounts were not denied those accounts, but they were not give n information about or assistance in opening 16 them. with an annual interest rate An ordinary savings account is a regular OIBM savings account same interest rate but allows farmers to of 2.5%. The commitment savings account has the 17 the bank would allow access to the funds. specify an amount and a “release date” when farmers stated how much they wanted deposited in the During the account opening process, ordinary and commitment savings accounts after the sale of their tobacco crops. For example, if a farmer stated that that he wanted MK 40,000 in an ordinary account and MK 25,000 in a deposited into the ordinary account until MK commitment savings account, funds would first be 40,000 had been deposited, then into the commitment savings account for up to MK 25,000, with any remainder being deposited back into the ordinary account. Raffle Treatments To study the impact of public informati on on savings and investment behavior, we implemented a cross-cutting randomization of a savi ngs-linked raffle. Participants in each of our two savings treatments were randomly assigned to one of three raffle conditions. These raffles provided a mechanism for revealing public. We distributed tickets individual savings balances in for a raffle to win a bicycle, where the numbe r of tickets each participant received was determined by his or her savings balance as of pre-announced dates. Every MK 1,000 saved with OIBM (in total across ordinary and commitment savings accounts) entitled a participant to one buted twice. The first distributi on took place in early September, raffle ticket. Tickets were distri cond distribution took place in November, and and was based on savings as of August 19. The se was based on savings as of October 22. By varyin g the way in which tickets were distributed, we sought to exogenously vary the information that club members had about each other’s savings. 16 up, nobody requested an ordinary or a commitment account during the Among farmers in the control gro savings training at baseline. According to OIBM admini strative records, eight farmers in the control group had commitment accounts by the end of Octo ber 2009 (opened withou t our assistance or enc ouragement), but none of these had any transactions in the accounts. 17 By design, funds in the commitment account could not be accessed before the release date. In a small number of cases OIBM staff allowed premature withdrawals of funds when clients presented evidence of emergency needs, e.g. health or funeral expenditures. 9

11 Because the raffle itself could provide an incentiv e to save or could serve as a reminder to an, 2010 and Kast, Meier and Pomeranz, 2012), save (Karlan, McConnell, Mullainathan, Zinm vings accounts was randomly one third of all clubs assigned to either ordi nary or commitment sa kets (and was not told about the raffle). Another determined to be ineligible to receive raffle tic one third of clubs with savings accounts was random ly selected to have ra ffle tickets distributed privately. Study participants were called to a meeting for raffle ticket distribution but were handed their tickets out of view of other study participants. The final third of clubs with savings bution of raffle tickets. In these clubs, each accounts was randomly selected for public distri eceived was announced verbally to everyone that participant’s name and the number of tickets r attended the raffle meeting. Because of the simple formula for determining the number of tickets, farmers in clubs sily estimate how much other members of the where tickets were distributed publicly could ea , though, did not reveal information about club had saved. Private distribution of tickets raffle scheme was explained to pa rticipants at the time of the individuals’ account balances. The baseline survey using a simulation. Members were first given hypothetical balances, and then given raffle tickets in a manner that corresponded to the distribution mechanism for the treatment condition to which the club was assigned. In cl ubs assigned to privat e distribution, members were called up one by one and given tickets in private (out of sight of other club members). In bution, members were called up and clubs assigned to public distri their number of tickets was announced to the group. The design of the project, therefore, incl udes seven treatment conditions: a pure control condition without savings account offers or raffles; ordinary savings accounts with no raffles, ribution of raffle tickets; and with private distribution of raffle tickets, and with public dist commitment savings accounts with no raffles, with private distribution of raffle tickets, and with public distribution of raffle tickets (see Table 2). As mentioned, the randomization was carried out at the club level. The list of tobacco clubs in central Malawi (all of which had existi OIBM) was provided by ng production loans with OIBM in cooperation with the two tobacco buyer companies. Prior to randomization, treatment 18 or dark-fire) and week of tobacco type (burley, flue-cured clubs were stratified by location, 18 “Locations” are the tobacco buying companies’ geographically-defined administrative units within which extension services and contract buying activities are coordinated. 10

12 scheduled interview. The stratification of treatment assignment resulted in 19 distinct location/tobacco-type/week stratification cells. The sample consists of 299 clubs with 3,150 farmers surveyed at baseline, and 298 clubs 19 Attrition from the baseline to the endline survey was with 2,835 farmers surveyed at endline. ent status (as shown in Appendix Table 1). 10.0% and does not vary substantially by treatm While attrition is uncorrelated with treatment as signment for five out of the six treatment groups, tage point lower rate farmers in the ordinary (private raffle) treatment group have a three percen of attrition from baseline to endline survey, comp ared to the control group, and this difference is statistically significant at the 10% level (p-value 0.085 in the specification with full baseline controls). Since the difference is very small, we do not view this as an important concern. Balance of baseline characterist ics across treatment conditions tments achieved balance in pre-treatment To examine whether randomization across trea in means of 17 baseline variables for the six characteristics, Table 3 presents the differences atistical inference about the differences in treatment groups vis-a-vis the control group. For st means we estimate a regression of each baseline two savings treatment variable against the indicators (Commitment and Ordina ry), the four respective interac tions with the raffle treatment indicators (Public or Private) and the stratification cell dummies. With a few exceptions, baseline variables for the ordinary and commit ment (without raffle) treatment groups are well balanced with the cont rol group. The exceptions ar e that individuals in likely to be married (column 2), the ordinary group are more likely to be female (column 1), less later” (column 14); a nd individuals in the and less likely to be “patient now, impatient . Overall, however, for both the ordinary and commitment group are more likely to be female s of all 17 baseline variables commitment (no raffle) groups we cannot reject the null that mean are jointly equal to those in the control group (see p-values of F-tests at the bottom of Table 3). The situation is similar for the coefficients on the interactions between the savings and raffle treatments – most outcomes are balanced vis-à-vis the corresponding “no raffle” savings treatment, with a scattering of statistically signi ficant differences that ar e not too different from 19 60 clubs in two locations had to be excluded from the sample because of serious implementation irregularities. Clubs in Kasungu Central were discovered to contain substantial numbers of “ghost” (nonexistent) club members and served as vehicles for larger landowners to fraudulently obtain very large loans from our partner institution; survey data collected for these individuals is thus likely to be fictitious. Clubs in Mndolera were excluded because of clerical and comm unications errors that led to ambiguity in treatment assignment. In the two locations subject to these issues, we excluded all clubs (amounting to three stratification cells) from the sample. Because entire stratification cells were excluded, inference among the remaining stratification cells yields internally valid results. 11

13 what would likely have arisen by chance. Again, for none of the raffle sub-treatments can we onal levels that the full set of baseli ne variables is jointly equal to the reject the null at conventi mean for the corresponding “no raffle” treatment. baseline imbalance may be driving our results, At any rate, and to alleviate any concern that we follow Bruhn and McKenzie (2009) and include the full set of baseline characteristics in 20 addition to the stratification cell fixed effects. Table 3 as controls in our main regressions, in Estimation strategy 3. awals prior to A number of dependent variables are of interest, such as de posits and withdr and sales in the next the next planting season, inputs used in the ne xt planting, crop output planting, and household expenditure s after the next harvest. To estimate the impact of the treatments we estimate the following regression equation, similar to the specification used in Table 3 when checking for balance across baseline characteristics: (1) Y = + α Ordinary Commitment α + δ j 1 j ij 2 α + Com_Raf + α Com_PubRaf j j 4 3 α + Ord_Raf + Ord_PubRaf α 6 5 j j + β ’ X ε + ij ij i in club j . Commitment is an indicator is the dependent variable of interest for farmer Y ij i and Ordinary is an indicator variable for variable for assignment to the commitment treatment i is an indicator for assignment to any raffle assignment to the ordinary treatment. Com_Raf j treatment (either private or public) for the commitment savings treatment group, and Com_PubRaf is an indicator for assignment to the public raffle treatment specifically. Ord_Raf j j and Ord_PubRaf is a are defined analogously, but for th e ordinary savings treatment group. X ij j on cell dummies and control variab les measured in the baseline vector that includes stratificati is a mean-zero error. survey, prior to treatment (the 17 baseline variables in Table 3) and ε ij is the club, standard errors are clustered at this level (Moulton Because the unit of randomization 1986). Coefficients α and α measure the difference in means of the dependent variable between 1 2 the commitment treatment and the ordinary trea ithout additional raffle tment, respectively (w 20 Results turn out to be very similar when only stra tification cell fixed effects are included. See Appendix Tables 2, 3 and 4. 12

14 treatments) vis-à-vis the control group. The difference ( α - α ) represents the difference in means 1 2 nary treatment (each without layered-on raffle between the commitment treatment and the ordi is the difference in means between the no-raffle commitment treatments). The coefficient α 3 treatment group and the commitment treatment comb ined with the private raffle treatment. The represents the difference in means betw een the commitment treatment with a coefficient α 4 private raffle. Put differently, coefficient public raffle and the commitment treatment group with α represents the additional impact s of making the raffle public, ove r and above the private raffle 4 measures the treatment for the commitment savings group. Similarly, the coefficient α 5 difference in means between the oup and the ordinary treatment no-raffle ordinary treatment gr is the difference in means between the ordinary group with private raffle. The coefficient α 6 nary treatment group with private raffle. treatment with a public raffle and the ordi Therefore, α + α is the total impact of the commitmen t treatment with private raffle, and α 1 3 1 + α is the total + α α is the total impact of the commi tment treatment with public raffle. α + 2 4 3 5 α is the total impact of the + α impact of the ordinary treatment with private raffle, and + α 6 2 5 ordinary treatment with public raffle. because not every club member offered account We focus on intent-to-treat (ITT) estimates opening assistance decided to op en the account. We do not report average treatment on the treated (TOT) estimates because it is plausible that members without accounts are influenced by the training script itself or by members who do open accounts in the same club, either of which would violate the Stable Unit Treatment Va lue Assumption (SUTVA) (Angrist, Imbens and Rubin, 1996). Empirical results: impact of treatments 4. To understand the impacts of access to formal savings, we first study the extent to which funds flowed into and out of the savings accounts in the pre-planting and planting periods. Then we examine impacts on agricultural inputs, farm output, household expenditures and other household outcomes. A. Impacts on savings transactions (deposit s and withdrawals) and savings balances Table 4 presents regression results from estim ating Equation 1. The dependent variables are various deposit and withdrawal outcomes from ad ministrative records of our partner institution, OIBM. The first column presents results in which the dependent variable is an indicator variable 13

15 for whether any transfers were made from th e club account to the farmer’s individual account id. This is essentially an indicat or for “take-up” of the savings after the group loan had been repa treatments. Columns 2 to 7 present results for three types of savings behaviors in March to October 2009, the “pre-planting” period: total depo dinary, commitment and sits (separately for or other accounts, as well as the sum across all acc ounts), total withdrawals, and net deposits into 21 This pre-planting period is when funds are accumulated from the previous OIBM accounts. season’s harvest and when inputs are pur chased for the 2009-2010 growing season. Results from column 1 show that while none of the farmers in the control group transferred money via direct deposit into an OIBM account (since they were not offered direct deposit nor account opening assistance), 16% of farmers in the ordinary account, no raffle treatment did larger at 21% for farmers in the commitment transfer money. This percentage is somewhat tically significant differential effects of either savings treatment without raffle. There are no statis the public or private raffle on this take -up indicator. Among the individual baseline ucation, household size an characteristics in Table 1, we find that age, ed d having a prior account with OIBM correlate positively and significantly with receiving a deposit into the ordinary account. Interestingly, net transfer s made during the year prior to the intervention are also correlated with this take-up indi cator. These correlations sugges t that richer farmers are more likely to receive a direct deposit into their ordi nary accounts. deposits, both ordinary and commitment Turning to dependent variables related to higher total withdrawal s during the pre-planting treatments led to higher total deposits as well as period compared to the control group. Coeffici ents on both types of savings treatments are statistically significantly different 2) as well as for withdrawals from zero for deposits (column (column 3). The coefficient on the commitment (no raffle) treatment is nearly identical to the coefficient on the ordinary (no raffle) treatment. We note that the private raffle leads to lower de posits as well as lower withdrawals for both types of treatments (columns 2 and 3). This resu lt is surprising as we had expected that the private raffle, by providing an incentive to save tion of the individual’s without direct revela savings balance, would have a positive effect, or at worst a zero effect. However, because only a minority of participants had pos itive deposits at the time the ra ffles were conducted, individuals who had no savings did not attend the raffle mee ting. As a result, we speculate that simply 21 Net deposits are deposits minus with drawals. For accounts op ened in March or later (i.e., for all accounts opened by our project), “net deposits” is equal to the acc ount balance at the end of the time period (Oct 2009). 14

16 showing up at the meeting would have been a signal that the individual had positive savings of at 22 sary to receive a raffle ticket). least MK1,000 (the minimum amount neces The coefficient on vs. the private raffle for ect of the public raffle “Ordinary x Raffle x Public” is the differential eff e coefficients are about the same magnitude as the ordinary treatment. In columns 2 and 3 thes the corresponding coefficients on “Ordinar y x Raffle” but of the opposite sign, and are statistically significantly different from zero at the 10% level. Th ese results indicate that the ry treatment effect does not hold for the public negative effect of the private raffle on the ordina raffle. It is possible that the public treatment may have led to differe ntially higher savings by fostering competition or social comparisons wi th others in the group, offsetting the negative 23 The coefficients on the corresponding raffle indicators for the effect of the private raffle. are uniformly smaller in magnitude and are not commitment treatment are of similar signs, but different from zero. statistically significantly s on deposits and withdr awals are unexpected, Overall, the private and public raffle result and so our interpretation of the patterns is somewh at speculative. In subsequent tables nearly all coefficients on the raffle interaction terms are not statistically significan tly different from zero, 24 possibly due to the lack of power given th at few farmers participated in the raffle. We therefore limit subsequent discussion of the raffle results in this paper. parately examine impacts on three different To further explore the impact on deposits, we se : deposits into ordinary accounts (column 4), components of deposits in the pre-planting period commitment accounts (column 5), and other account s not set up by the project (column 6). It is clear that most of pre-planting deposits go into ordinary accounts, even among farmers in the zes of the coefficients on the commitment (no commitment (no raffle) treatment. The relative si that 89% of pre-planting deposits (MK 19,464.30 raffle) treatment in columns 4 and 5 indicate 22 There are other possible explanations for the negative effect of the private raffle on savings. Study participants may have overestimated the expected value of th eir raffle tickets, and saved less as a result. In addition, individuals may have planned to deposit funds at some later date (because the raffle tickets would be awarded based on balances as of specific dates in August and October), but may have ended up depleting their cash stocks in the interim or otherwise failing to make those later deposits. 23 At the same time, the private raffle treatment may have “primed” individuals to worry about demands from others who might learn they were saving, while the public treatment may have primed individuals to raise their social status by saving more than others. Key references in the psychology literature on priming include Bornstein (1989), Bettman and Sujan (1987), and Zajonc (1968), who document situations where decisions can be influenced by highly local or transitory influences, su ch as the introduction of certain concepts. 24 One could also argue that the null ef fects of the raffle on subsequent outc omes are due to the irrelevance of the raffle because other club members are not part of one’s risk sharing network or the irrelevance of the public vs private treatment because club members ar e familiar with each other balances as they jointly filled out the sheet that specified how total club sales had to be divided among members. 15

17 out of total deposits of MK 21,861.22) by farmer s in the commitment (no raffle) treatment ounts, rather than the commitment accounts. actually went into the ordinary savings acc This finding that most of the savings in the commitment (no raffle) treatment were actually paper, and helps rule out an deposited in ordinary accounts is one of the key results of the ough which the commitment treatm ent may have had its effects. important potential channel thr Because the amount deposited in the commitment account was several times smaller than the ented in the next section), it cannot be the case increase of input usage on average (to be docum that the commitment account helped farmers d eal with their self-control problem by literally “tying their hands”. In column 7 we turn to net deposits during the pre-planting period. The commitment savings (no raffle) treatment led to a small and statistica lly significant increase in net deposits, while the ffle) treatment was not statistic effect of the ordinary (no ra ally different from zero. The difference in coefficients between ordinary a nd commitment treatments is not statistically from zero, however. significantly different ril 2010, is conveniently divided into the The “growing” period, from November 2009 to Ap “planting” period from November to December 2009, when land is prepared, seeds are sown and fertilizer is applied, and January through April 2010, which is the lean or “hungry” season when households may have depleted stoc ks of maize from the previous season’s harvest and have not yet harvested crops or received payments for the 2010 harvest. In column 8 of Table 4, we onths of the growing season, November and examine net deposits during the first two m on net, led to higher withdrawals as did December 2009. The commitment (no raffle) treatment, the ordinary (no raffle) treatment. The coeffici ents are small in magnitude, however, indicating r both treatments during this time period. impacts on net withdrawals of around MK1,000 fo deposits in the January to April 2010 lean or In column 9, the dependent variable is net “hungry” season; coefficients on the commitment and ordinary (no raffle) treatments are small and not statistically significantly different from zero. Neither treatment appears to have led to more access to saved resources during the 2010 lean season. Time patterns of deposits and withdrawals nd withdrawals from OIBM accounts in the 2009 Table 4 documents that both deposits into a pre-planting period were substantial for both th e commitment and ordinary treatments. An open question is whether most funds remained deposit ed in the accounts until the planting period. As it turns out, most funds were wit hdrawn not long after being depos ited. Figure 3 presents average deposits into and withdrawals from ordinary and other (non-commitment) accounts, by month, 16

18 25 from March 2009 to April 2010. cludes all individuals in a The sample in Figure 3.a in the raffle treatments), while the samples for commitment treatment (whether no-raffle or one of with or without raffle). For Figure 3.b include all individuals in an ordinary treatment (again the control group. udes all individuals in comparison, the sample used in Figure 3.c incl rred in June, July, and August 2009, coinciding The figures indicate that peak deposits occu with the peak tobacco sales months. Average depo sits in every month for individuals in both the similar in magnitude to average withdrawals, commitment and ordinary treatments are quite indicating that the majority of deposited funds were withdrawn soon thereafter. As a result, savings balances during the pre-planting period were much lower than accumulated deposited 26 amounts, explaining why most farmers did not participate in the raffle. accounts were withdraw One likely reason why funds in the ordinary n at once soon after they had been deposited has to do with transactions costs. Farmers lived on average 20 27 kilometers away from the bank branch and w ould typically travel by foot, bus, or bicycle. In addition to the commuting time, farmers report a me dian waiting time at the branch to withdraw money of one hour. In contrast to the time pattern of the or dinary accounts, funds into commitment accounts do stay deposited for longer periods of time as expected. Figure 4 disp lays average deposits into and withdrawals from commitment accounts, by mont h, for all individuals in a commitment the peak months are June, July, and August, treatment (with or without raffle). For deposits, ordinary accounts. But withdrawals from the coinciding with the peak deposit months for the commitment accounts are delayed substantiall y, occurring in October, November, and December, coinciding with the key months when agricultural inputs must be purchased and applied on fields. nt accounts is consiste nt with the release This time pattern of withdrawals from commitme dates chosen by users of commitment accounts. Fi gure 5 presents the histogram of commitment account release dates (when commitment account funds would be “unlocked” and funds made available to farmers) that farmers chose dur ing account opening. About 60% of farmers chose release dates in the months of October to D ecember while others chose to have access to the funds in January and February, dur ing the lean or “hungry” season. 25 The data presented are the sum of the dependent variables in columns 4 and 6 of Table 4. 26 The pattern is similar for individuals in the control group, but levels are much lower owing to the fact that direct deposit from the tobacco au ction floor into farmer accounts was not enabled for that group. 27 The median round-trip bus fare is MK 400 for a two hour ride one way. 17

19 B. Inputs, crop sales, and expenditures We now turn to impacts of the treatments on land cultivated, inputs, crop output, and household expenditures in Table 5. Across the six dependent variables the coefficient on the cant in five out of six commitment (no raffle) treatment is large, positive, and statistically signifi columns (the exception is column 5 for farm profits, where the coefficient is marginally fficients on the ordinary savings significant). In comparison, the coe (no raffle) indicator are all smaller in magnitude and none ar e statistically significantly diffe rent from zero at conventional levels. Furthermore, the private raffle treatment never leads to a statistically significantly corresponding no-raffle treatment. Similarly, in none of the different effect vis-à-vis the regressions does the public raffle have effects that are statistically significantly different from the 28 private raffle. The first two columns of the table reveal that the commitment (no raffle) treatment had a atistically significant eff large positive and st ation and the total value ect on both land under cultiv of inputs used (which include seed, fertilizer, pesticides, hired labor, transport and firewood for 29 curing) in the late-2009 planting. Farmers in the commitment group cultivated on average 0.42 more acres of land than the control group (whi ch had 4.28 acres of land under cultivation). The tly different (p-value 0.05 commitment coefficient is statistically significan 7) from the ordinary coefficient of 0.05 (which in turn is not statistic ally significantly different from zero). Compared to MK 60,372 in inputs used by control group farmers on average, commitment treatment , while the coefficient on the ordinary (no farmers used MK16,534 (or 27.4%) more. By contrast bout half the magnitude of the commitment (no raffle) treatment is also positive, it is only a raffle) treatment coefficient and it is not sta tistically significantly different from zero. The difference in the coefficients on the two treatme nts in column 2, however, is not statistically different from zero at conventional levels. It is noteworthy that the impact on input use is substantially larger than total savings balances (or net deposits) at the r to the typical start of the end of October 2009, immediately prio planting season (column 7 of Table 4). Examina tion of the timing of w ithdrawals in Figure 3.b 28 The public raffle treatment in the ordinary treatment group is significant different than the control group for most variables but as we can see from Table 3, this group is also the one that suffers from most imbalance. 29 We note that we report the cash value of inputs at baseline instead of the more comprehensive measure of total value of inputs, which is only available at follow-up. We find similar results when we use the cash value of inputs computed at follow-up (results not shown). 18

20 helps shed light on when funds were likely to have been accessed for input purchases. Most funds used to purchase inputs must have come from withdrawals before the end of October: column 3 of Table 4 indicates that the commitment (no raffle) treatment led to total withdrawals in the period leading up to and including October amounting to more than MK 20,000, which e) treatment on inputs used. This time pattern exceeds the impact of the commitment (no raffl e planting season, either accumulating funds held suggests that farmers withdrew funds prior to th ases, or purchasing inputs in outside of the bank (e.g., stored at home) for later input purch 30 advance of the planting season. The increase in input use due to the commitment (no raffle) treatment is 8.3 times the impact e-planting period (MK 16,534 from column 2 of on deposits in commitment accounts in the pr e 4), but is well with Table 5 divided by MK 1,994 from column 5 of Tabl in the total amount of ngs accounts (MK 21,861 from column 2 of Table deposits into ordinary and commitment savi 31 The bulk of funds used to purchase inputs, ther efore, must have come from ordinary rather 4). than commitment savings, and thus were availa the pre-planting period, ble to farmers during instead of physically being locked away at th e bank. This result is inconsistent with the hypothesis that the commitment accounts helped to solve farmers’ self-control problems by keeping them from accessing the funds prior to the planting season. The fact that such a large proportion of observed deposits at OIBM was allocated towards input purchases in the commitment treatment gr oup suggests that the funds were deposited in e outset that they would be used for input purchases. It should accounts with the expectation at th r only half of all household income, and so be noted that revenues from tobacco account fo farmers have other sources of funds (incompletely observed by us) that are used to pay for other household expenditures. Columns 3 and 4 indicate that the larger input use caused by the commitment treatment resulted in higher proceeds from the sale of crops as well as tota l crop output in the 2010 32 harvest. Both coefficients on the commitment (no raffle) treatment in these regressions are large in magnitude and statistically significantly di 5% level. The increase fferent from zero at the 30 Duflo, Kremer, and Robinson (2011) show that interventions encouraging advance fertilizer purchases raise fertilizer use in western Kenya. 31 As we shall see in the next section, the increase in input use does not appear to be driven by higher borrowing. 32 Since the baseline was conducted right before the harvest, we only collected the proceeds from crop sales for the 2008 season. The value of crop output (sold and unsold) is only available for the follow-up survey after the harvest of 2010. 19

21 in crop sales (MK 22,962.78) comes exclusively from tobacco sales rather than maize sales since ease in total value of crop output (MK 33,968) the latter do increase significantly. The incr efficient on the ordinary (no amounts to 21.8% of mean crop va lue in the control group. The co ive but its magnitude is much smaller than the raffle) treatment in this column is also posit coefficient on the commitment treatment, and it is cantly different from not statistically signifi zero. The difference between the ordinary (no raffle) and commitment (no raffle) coefficients in this column is sta tistically different from zero at the 10% level (p-value 0.081). on farm profits, defined as the difference Column 5 shows the impact of the treatments between the total value of crop output (dependent variable of column 4) and the total value of 33 The coefficient on the commitment treatment is inputs used (dependent variable of column 2). lly significant. The coefficient for the ordinary large in economic terms and marginally statistica cally significant, and the differe nce vis-a-vis the commitment account is small and not statisti account is marginally si gnificant (p-value 0.142). Column 6 examines the impact of the treatments on total household expenditures in the endline (post-harvest) survey (fielded in July to September 2010). The commitment (no raffle) treatment coefficient is positive and statistically significantly different from zero at the 5% level, while the coefficient on the ordi nary (no raffle) treatment is substantially smaller and not statistically significantly differe nt from zero. The commitment (no raffle) treatment effect represents a 17.4% increase total expenditures over the last 30 days compared to the control 34 group. In order to examine further whether the co mmitment accounts treatment had a differential impact vis-a-vis the ordinary accounts across the set of outcomes in Table 5, we follow Kling, Liebman and Katz (2007) and presen t p-values of three F-tests at the bottom of Table 5 that are based on seemingly unrelated regression (SUR) estimation. We si multaneously estimate equation 35 The test that the coefficient on the 1 with the dependent variab les of columns 1, 2, 4 and 6. 33 The coefficients of column 5 are identical to the difference between not exactly (though nearly) numerically the coefficients from column 4 and 2 since survey va riables are winsorized (see Appendix B for details). 34 We also check whether the results are driven by th ose that take-up the accounts and receive a deposit into their accounts among all those offered the accounts. In partic ular, we use specification (1) and include interactions of treatment dummies with an indicator of take-up (the dependent variable in column 1, Table 4). As expected, we find that the interaction of commitment (no raffle) dummy with take-up is positive and significant in three out of the five variables of Table 5 (results not shown). This provides suggestive evidence that the results are driven by the compliers. 35 We restrict attention to just the regressions for the four outcomes in columns 1, 2, 4, and 6 of Table 5 because farm profit in column 5 is simply the difference between the dependent variables in columns 2 and 4 and 20

22 commitment (no raffle) treatment is jointly equal to zero across the four regressions is rejected at conventional levels of st ce (p-value 0.042). In contrast , we cannot reject that atistical significan tment is jointly equal to zero across the four the coefficient for the ordinary (no raffle) trea the coefficients on the ordinary regressions (p-value 0.694). We al so fail to reject however that the commitment (no raffle) treatment across the (no raffle) treatment equals the coefficients on regressions of columns 1, 2, 4, and 6 (p-value 0.254). Other outcomes and mechanisms C. Table 6 presents regression results on the impacts of the treatm ents on household size, taken out to finance the agricultural investment transfers to and from the social network, loans and demand for fixed deposit accounts, measured at the endline survey. Column 1 shows that the interv ention had no effect on household size. This implies that the impacts presented in Table 5 are driven by changes in agricultural decisions and outcomes rather than changes in household composition. interest because a poten tial channel of the Transfers sent and received are of particular observed impact of the commitment treatment ma y be an increased ability to resist demands hough net balances in the commitment accounts were small, the from the social network. Alt provided an excuse to turn dow n requests for assistance from existence of the account may have the social network by claiming that savings were inaccessible. Even though most of farmers’ funds were in ordinary accounts, this could have been a credible claim because the division of an individual’s funds between ordi nary and commitment accounts was not directly observable to 36 others. In columns 2, 3 and 4 of Table 6 we examine th e sums of transfers made, transfers received 37 and net transfers over th e last twelve months. reduced net transfers for We find no evidence of ere is a small positive, effect on net transfers the commitment (no raffle) treatment. If anything, th 38 made (column 4). This result fails to support the hypot hesis that the channel through which commitment accounts led to increased input use wa s via an increased abi lity to resist sharing cannot therefore be separately determin les are included in the value of crop ed. Similarly, the proceeds from crop sa sold in column 4. 36 total balances at OIBM, not how Even the public raffle treatments only provided a signal of an individual’s those savings were split between ordinary and commitment accounts. 37 The coefficients of column 4 are not exactly (though nearly) numerically identical to the difference between the coefficients from column 3 and 2 since survey va riables are winsorized (see Appendix B for details). 38 In a very different context, this result is similar to th at of Chandrasekhar et al. (2012), who show in a lab-in- the-field experiment among Indian villagers that sa vings access does not crowd out transfers to others. 21

23 with one’s social network. We note that the transfers studied inter -household transfers, in columns 2, 3 and 4 refer to intra -household transfers. It is and do not capture any changes in possible that the commitment ers by study participants to spouses at least in part by reducing transf treatment effect operated y, the null effect on total transfers in the same household. Alternativel and other individuals with (column 2) could be the result of lower transf ers made during the pre-planting season while the commitment account was active (and thus the excu se that funds were lo cked up valid) but higher tion had been larger. Similarly, farmers in the after the harvest given that agricultural produc t savings quickly to avoid the ordinary treatment may have spen pressure to share them with Unfortunately, we lack the data needed to test others, leading to no differences in transfers made. these hypotheses. Column 5 examines the largest source of borrowing for agricultural investment in inputs, 39 After all, the increase in total value of namely loans provided by a lender to the tobacco club. inputs for the commitment treatment group could be driven by a higher loan size and not by the increased ability to keep the funds until planting. Column 5 shows that this is not the case. The commitment (no raffle) and ordi nary (no raffle) treatment groups report loan amounts from the 40 tobacco club that are similar to those in the control group. Finally, we present data on s ubsequent opening of fixed deposit accounts (column 5) at the deposit accounts in Malawi typi time of the endline survey. Fixed cally have a duration of three or six months. The client makes an initial one-time deposit of pre-specified amounts, typically in multiples of MK10,000. During the three- or six- month duration the client cannot make a also cannot increase th withdrawal from the fixed deposit account and e savings balance. Interestingly, we find that ow nership of fixed deposit account s is six percentage points higher and significant at the 1% level in the commitment (no raffl e) group, and three percentage points higher in the ordinary (no raffle) group (s ignificant at the 5% level) compared to the control group (this difference in treatment effect s across the ordinary and commitment treatments is not significantly different from zero at conventional levels). The positive impact on subsequent ownership of fixed deposit accoun ts suggests that the commitment treatment caused farmers to raise thei r perceived value of commitment features in 39 Loans from informal lenders and friends and family account for a small fraction of total borrowings. At any rate, total credit instead of tobacco credit yield very similar results. 40 Similarly, we find no differen ce across treatment and control groups in the probability of accessing a loan (results not shown). 22

24 formal savings accounts. This rein forces the interpretation of our results as causal effects of the rrelations, insofar as hi gher demand for fixed commitment treatment rather than spurious co deposit accounts reflects farmers’ own recognitio n that such accounts have some benefits. However, this evidence does not help differentia te between self- and other-control problems as sources of demand for commitment, and also doe s not rule out the possibility that other psychological channels may be at work. Conclusion 5. We find that offering commitment savings acco unts to smallholder cash crop farmers in deposits and withdrawals prior to the next Malawi has substantial impacts on formal bank s applied in the next planting season, crop sales at the next planting season, agricultural input itures after the next harvest. Wh ile offers of “ordinary” bank harvest, and total household expend udes, effects on agricultu ral input use and other accounts also lead to deposits of similar magnit subsequent outcomes are smaller and statisticall y insignificant. Given the large impacts of the commitment tr eatment, it is important to ask why the treatment had such substantial effects, while the ordinary treatment did not. Several possible may have helped farmers solve their self- mechanisms exist. First, the commitment account maintain their balances until the next planting control problems, giving them the discipline to season when they could be used for agricultura l inputs. Alternatively, the commitment accounts ing with others in their social network. An may have helped farmers to refrain from shar additional possibility is that the commitment accounts may have increased later input use via some other psychological channel. We provide evidence against the hypothesis th at the commitment treatment helped via solving farmers’ self-control problems. The actual amounts saved in the commitment accounts were very low (about an order of magnitude lower than the obse rved increase in inputs), with ordinary accounts receiving the va erved increase in input use due st majority of deposits. The obs to the commitment treatment plausibly could have been funded out of deposits into ordinary accounts, but is much too large to have been funded purely out of deposits in commitment accounts. This rules out that the impacts of the commitment treatment were due to literally “tying the hands” of treated farmers by preventing them from spending their profits earlier in the year. We also find no evidence that the commitmen t treatment helped solve “other-control” problems (demands for sharing of resources w ith one’s social network). The commitment 23

25 treatment did not reduce net transfer s to other households (and in fact had a small positive impact on such transfers). Relatedly, a sub-experiment testing the impact of making one’s account balances public to others also did not find that public revelation of ba lances reduced savings this context is not nce of other-control in deposits. That said, the case against the importa ables we examined refer to inter -household transfers, and do not conclusive. The transfer vari intra -household transfers. It remain s possible that the commitment capture any changes in at least in part by reducing transf treatment effect operated ers by study participants to spouses in the same household. and other individuals with r-control results is th at though we do not find evidence that Another caveat regarding the othe commitment savings accounts reduced transfers to other members of the social network, the accounts may have helped farmers increase their input use by mitigating another possible consequence of social pressure to share. Individuals who know they will be subject to demands from others in their social network can preven t others from claiming their money by spending it preemptively. Rapidly consuming income makes it unavailable to others; it is consistent with signaling a high marginal utility of consump tion in a model where income is taxed and redistributed from those with low marginal utility of consumption to high marginal utility of consumption. Goldberg (2011) found support for such a model in an experiment that who received money in public settings spent demonstrated that Malawian cash crop farmers significantly more of that money immediately than farmers who received money in private not have the high-frequency consumption data necessary to test settings. In this project, we do counts were less likely to engage in hasty whether farmers with commitment savings ac consumption than farmers without such accounts. However, a reduction in sub-optimally timed consumption is a channel through which offers of commitment accounts could have led to increased use of inputs and improvements in output, profits, and household expenditures. In future related work we will examine whether commitment savings offers affect such timing and composition of expenditures in the “anticipatory consumption” by examining the post-harvest months. Future research is thus st ill needed to shed light on the importance other- control problems as a potential hindrance to savings. We can provide no empirical evidence as to whether mental accounting or some other psychological phenomenon may be behind the impact of the commitment treatment. Investigating this possible channel for the effects of the commitment treatment is also an important area for future research. 24

26 While the well-being of farmers offered co mmitment accounts is likely to have improved, in the community. An initial worry was that we do not shed light directly on impacts on others the commitment accounts led farmers to make fewer transfers to others in the community in the ngements (for example, to help ot hers cope with shocks). As it context of informal insurance arra turns out, we do not find any negative impacts of the commitment treatment on net transfers to other households. That said, redu ced “anticipatory consumption” in the months immediately after the intervention may have had negative im pacts via reduced demand for goods and services produced by others in the community. We view investigation of longer-term impacts on study participants and on others in the community as an important area for future research. 25

27 References Economics Of Roscas And Intrahousehold Anderson, S. & Baland, J.-M. (2002), 'The 117 (3), 963-995. Resource Allocation', The Quarterly Journal of Economics & Rubin, D. B. (1996), 'Identifi Angrist, J. D.; Imbens, G. W. cation of Causal Effects Using Statistical Association 91 (434), 444-455. Instrumental Variables', Journal of the American l Access on Savings by Low-Income People', Aportela, F. (1999), 'Effects of Financia mimeo, Banco de Mexico. Armendariz de Aghion, B. & Morduch, J. (2005), The economics of microfinance , MIT Press, Cambridge, Mass. household Decision Making: An Experimental Ashraf, N. (2009), 'Spousal Control and Intra- American Economic Review 99 (4), 1245-77. Study in the Philippines', Ashraf, N.; Karlan, D. & Yin, W. (2010), 'Fem ale Empowerment: Impact of a Commitment World Development 38 Savings Product in the Philippines', (3), 333-344. st: Evidence from a Ashraf, N.; Karlan, D. & Yin, W. (2006), 'Tying Odysseus to the Ma Commitment Savings Produc t in the Philippines', 121 (2), The Quarterly Journal of Economics 635-672. Atkinson, J.; de Janvry, A.; Mc Intosh, C. & Sadoulet, E. (2010), 'Creating Incentives To oral Experiment From Guatemala', mimeo, Save Among Microfinance Borrowers: A Behavi University of California at Berkeley. (2011), 'Pretending to Be Poor: Borrowing to Baland, J.-M.; Guirkinger, C. & Mali, C. Escape Forced Solidarity in Cameroon', 60 (1), 1- Economic Development and Cultural Change 16. Benson, T. (1999), 'Area-specific fertilizer id maize grown by recommendations for hybr l for field assistants', Maize Commodity Team, Chitedze Malawian smallholders: A manua Agricultural Research Station, Malawi. Bettman, J. R. & Sujan, M. (1987), 'Effects of Framing on Evaluation of Comparable and Noncomparable Altern atives by Expert and Novice Consumers', Journal of Consumer Research 14 (2), 141-154. Bornstein, R. F. (1989), 'Exposure and affect: Overview and meta-analysis of research, 1968-1987', 106 (2), 265-289. Psychological Bulletin Bruhn, M. & Love, I. (2009), 'The economic im pact of banking the unbanked: evidence from Mexico', mimeo, The World Bank, Working Paper No. 4981. Bruhn, M. & McKenzie, D. (2009), 'In Pursuit of Balance: Randomization in Practice in Development Field Experiments', American Economic Jour nal: Applied Economics 1 (4), 200- 26

28 232. Burgess, R. & Pande, R. (2005), 'Do Rural Banks Matter? Evidence from the Indian Social Banking Experiment', 95 (3), 780-795. The American Economic Review H. (2012), 'Informal Insurance, Social Chandrasekhar, A.; Kinnan, C. & Larreguy, a Lab Experiment in the Field', mimeo, Networks, and Savings Access: Evidence from Northwestern University. Cole, S.; Sampson, T & Zia, B. (2011), ‘Prices or Knowledge? What Drives Demand for Financial Services in Journal of Finance , 66(6), 1933-1967. Emerging Markets?’ Collins, D.; Morduch, J.; Rutherford, S. & Ruthven, O. (2009), Portfolios of the Poor: How the World's Poor Live on $2 a Day , Princeton University Press. e Microcredit Summit Campaign Report 2009', Daley-Harris, S. (2009), 'State of th Microcredit Summit Campaign, Washington, DC. Deaton, A. (1990), ‘Saving in Developing Countries: Theory and Review’, Proceedings of the World Bank Annual Conference on Development Economics 1989. Supplement to The World Bank Economic Review and the World Research Observer , pp. 61-96. ow High Are Rates of Return to Fertilizer? Duflo, E.; Kremer, M. & Robinson, J. (2008), 'H Evidence from Field Experiments in Kenya', The American Economic Review Papers and Proceedings, 98 (2), 482-488. Duflo, E.; Kremer, M. & Robi nson, J. (2011), 'Nudging Farmers to Use Fertilizer: Theory and Experimental Evidence from Kenya', American Economic Review 101 (6), 2350-90. Dupas, P. & Robinson, J. (2012), 'Savings C onstraints and Microenterprise Development: Evidence from a Field Experiment in Kenya', mimeo, UCLA. e Poor Save More? Evidence from Health Dupas, P. & Robinson, J. (2012), 'Why Don’t th mimeo, Stanford University. Savings Experiments in Kenya', Fafchamps, M.; McKenzie, D.; Quinn, S. R. & Woodruff, C. (2011), 'When is capital enough to get female microenterpr ises growing? Evidence from a randomized experiment in Ghana', NBER Working Paper No. 17207. Feldman, J. M. & Lynch, J. G. (1988), 'Sel f-generated validity and other effects of measurement on belief, attitude Journal of Applied Psychology 73 (3), , intention, and behavior', 421-435. Flory, J. A. (2011), 'Micro-Savings & Inform al Insurance in Villages: How Financial Deepening Affects Safety Nets of the Poor - A Natural Field Experiment', Becker Friedman Institute for Research in Economics, Univ ersity of Chicago, Working Paper No. 2011-008. Giné, X.; Goldberg, J. & Yang, D. (2012), 'C redit Market Consequences of Improved Personal Identification: Field Expe rimental Evidence from Malawi', American Economic Review 27

29 102(6): 2923-2954. The Roles of Self-Interest and Impatience in Goldberg, J. (2011), 'The Lesser of Two Evils: Consumption Decisions', mime o, University of Maryland. Individuals, Hyperbolic H Hertzberg, A. (2010), 'Exponential ouseholds', mimeo, Columbia University. Hoff, Karla and Arijit Sen (2006), “The Kin System as a Poverty Trap?” in Poverty Traps Bowles, Samuel, Steven Durlauf and Karla Hoff (eds), Princeton University Press, pp. 95-115. Jakiela, P. & Ozier, O. (2011), 'Does Africa Need a Rotten Kin Theorem? Experimental Evidence from Village Economies', mi meo, University of Maryland. Kahneman, D. & Tversky, A. (1979), 'Prospect Theory: An Analysis of Decision under Econometrica 47 Risk', (2), 263-91. & Zinman, J. (2010), 'Getting to the Top of Karlan, D.; McConnell, M.; Mullainathan, S. Mind: How Reminders Increase Saving', NBER, Working Paper No. 16205. Kast, F.; Meier, S. & Pomeranz, D. (2012) , 'Under-Savers Anonymous: Evidence on Self- Help Groups and Peer Pressure as a Savings Commitment Device', Institute for the Study of Labor (IZA), Working Paper No. 6311. Kling, J. R.; Liebman, J. B. & Katz, L. F. (2007), 'Experimental Analysis of Neighborhood Effects', Econometrica 75 (1), 83-119. Ligon, E.; Thomas, J. P. & Worrall, T. (2002 ), 'Informal Insurance Arrangements with Limited Commitment: Theory and Evidence from Village Economies', The Review of Economic Studies (1), 209-244. 69 Maranz, D. E. (2001), Observations from Africa African Friends and Money Matters: (Publications in Ethnography, Vol. 37) , SIL International. , 'Returns to Capital de Mel, S.; McKenzie, D. & Woodruff, C. (2008) in Microenterprises: Evidence from a Field Experiment', The Quarterly Journal of Economics 123 (4), 1329-1372. Morduch, J. (1999), 'The Microfinance Promise', Journal of Economic Literature 37 (4), 1569-1614. Moulton, B. R. (1986), 'Random group effects and the precisi on of regression estimates', Journal of Econometrics , 32(3), 385-397. Pearce, B., Denton, P., Schwab, G., See bold, K. & Pearce, B., eds., (2011), 2011-2012 Kentucky & Tennessee Tobacco Production Guide , University of Tennessee and University of Kentucky, chapter Field Selection, Tillage, and Fertilization, pp. 24-29. Platteau, J.-P. (2000), Institutions, Social Norms and Economic Development (Fundamentals of Development Economics) , Routledge. 28

30 Prina, S. (2011), “Do Basic Savings Accounts He lp the Poor to Save? Evidence from a Field Experiment in Nepal,” mimeo, Case Western Reserve University. tainable Finance for the Poor , Robinson, M. (2001), The Microfinance Revolution: Sus World Bank Publications. The poor and their money , Oxford University Press, New Delhi; Rutherford, S. (2000), Oxford. ce Heterogeneity, Commitment, and Strategic Schaner, S. (2012), 'Intrahousehold Preferen Savings', mimeo, Dartmouth College. ngibility, and Mental Accounts', Thaler, R. H. (1990), 'Saving, Fu Journal of Economic 4 Perspectives (1), 193-205. Webb, T. L. & Sheeran, P. (2006), 'Does changi ng behavioral intenti ons engender behavior change? A meta-analysis of the experimental evidence', Psychological Bulletin 132 (2), 249-268. World, B. (2008), 'World Development Report: Agriculture for Development', World Bank, mimeo, World Bank, Washington D.C. Zajonc, R. B. (1968), 'Attidunial Effects of Mere Exposure', Journal of Personality and 9 (2, Pt.2), 1-27. Social Psychology Zwane, A. P.; Zinman, J.; Van Dusen, E.; Pari ente, W.; Null, C.; Miguel, E.; Kremer, M.; Karlan, D. S.; Hornbeck, R.; Giné, X.; Duflo, E.; Devoto, F.; Crepon, B. & Banerjee, A. (2011), “Being surveyed can change later behavi or and related parameter estimates,” Proceedings of the National Academy of Sciences 108 (5), 1821-1826. 29

31 Figure 1: Project timing Figure 2 : Tobacco Sales and Bank Transactions 30

32 31

33 32

34 : Distribution of commit ment savings release dates grouped by month Figure 5 30.0% 24.6% 25.0% 21.2% 20.0% 14.5% 15.0% 11.7% 10.9% 10.0% 6.5% 3.9% 5.0% 2.7% 2.5% .4% .4% .4% .3% .1% .1% .0% .0% 0.0% 33

35 Appendix A: Account details and full text of training script Savings account details In this experiment we offered farmers training and account opening assistance for two types of accounts depending on the treatment status (control, ordinary savings or commitment savings). The “ordinary” account referred to in th e main text is OIBM’s Kasupe account. Kasupe accounts had an account opening of MK500, no monthly fee, three free withdrawals transactions via ATM per month, and a MK25 fee per ATM wit hdrawal thereafter (all withdrawals at the Kasupe accounts was MK1,000 and there was an teller were free). The minimum balance for account closing fee of MK1,000. Kasupe accounts paid an interest rate of 2.5% p.a. with interest accruing quarterly. Deposit transactions into Kasupe accounts were free. proceeds directly deposited into an existing Farmers were given the option to have their ith OIBM. Another type of savings account not account if they already had a savings account w OIBM’s product portfolio was standard savings actively marketed in this experiment but part of ructure: an opening fee of MK500; a monthly fee of MK75; no accounts with the following fee st withdrawal fees; minimum balan ce of MK1,000; a closing fee of MK1,000; an interest rate of 6.5% p.a. with quarterly accrual. This less co cluded in the category mmon account type is in “ordinary” accounts together with Kasupe accounts. The “commitment” account referred to in the ma in text was an account newly developed for the project called “SavePlan.” Save Plan accounts paid the same interest rate as Kasupe accounts, but had no minimum balance requirement. SavePlan accounts also had no account opening or closing fees. Deposit transactions into Save Plan accounts were free. The only withdrawals permitted for SavePlan accounts were transfers to ordinary (Kasupe or other) savings accounts, for which no fee was charged. ccount offers, and raffle training Scripts for savings training, a (Scripts were administered in club meeting immediately following administration of baseline survey. Malawian research project staff played the roles of Persons 1 and 2.) Section 1: Savings Accounts (All Clubs) Person 1: Saving money in an individual bank account is a very smart way to protect your money and improve your wellbeing. As you know , OIBM has Kasupe accounts that are easy and affordable to use. Person 2: But I already have a savings account with my cl ub. What is better ab out this Kasupe account? First ask the group to list things that are good about the Kasupe account. When the group has come up with several suggestions, move on to the next line: Person 1: The Kasupe account is yours alone. You don’t share it with the rest of your club members. You are the only one who can take money out of the account and the only one who knows how much money you have saved in the account. Person 2: What are the details of the account? How much does it cost, and what is the interest? Person 1: MK 500 for smartcard, MK 500 for initial depos it, no monthly charge, MK 25 transaction charge (ATM fee, withdrawal fee). Person 2: But I can just keep money at home. What are some of the benefits of saving my money in a Kasupe account instead of at home? 34

36 Let the group make suggestions. After several things have been suggested, agree with the group and then move on to the next line. Person 1: Money is safer in a bank account than at home. If you keep your money at home, it could be stolen or lost in a fire. If you keep it at the bank, it is protected . Also, if you keep money at home, you may feel obligated to give money to your family or friends if they ask for it. If your money is in the bank, you can say that you don’t have any money to give. Person 2: That is interesting, but I think my money is safe at home. Ask the group: “Do you think money is safe at home?” Let the group come up with answers, then move on. Person 1: There are other reasons to keep money in the bank, too. Keeping money in a bank account can help you save for the future. If you have money at home, it is easy to be tempted to spend it on food or drinks or household items. If you have money in the bank, you will think twice about taking it out to spend. Instead, you can hool fees or buying fertilizer or accumulating the deposit leave it in the bank to save for important purchases like sc in case you have an emergency in the future, like someone for a new loan. Also, you can be sure to put away money gets sick and needs to go to the hospital. r the future (All Clubs) Section 2: Saving fo Person 2: It would be good to save for the future, but I have many needs now. How can I afford to save? It is important to make a plan for how to spend your money. One way to do this is to divide the Person 1: money you will have after selling your t obacco and paying your loans into two amounts. One amount is to use now, and the other amount is to use in the future. Then, you can commit to keeping the future amount safe, and not touching it now. Person 2: How can I do that? Person 1: Think about how much money you will have after you sell your tobacco and repay your loan to OIBM. Then, think about expenses you have immediately. Have the group list things they need to spend money on immediately. Get a list of 5-6 things, then move on. Person 2: Yes, I will have to pay someone who has done weeding for me. Also, I need to buy some soap and other household goods. My children need new clothes, too. Person 1: Yes, these are the kinds of things you need to spend money on right away, when you get paid. But now think of things you will need to spend money on in the future. What do you want to be absolutely sure you can afford? Ask the group to list things they want to save for in the future. Make sure they are thinking of long-term things or expenses that will happen in a few months. Get the group to list 5-6 things, then move on. Person 2: I can think of many things. I will need to pay school fees. Also, I want to make sure I can buy fertilizer for my maize. And I want to have money for food next year during the hungry season. Person 1: These are important expenses. You should plan to protect some of your money so that it is available for those expenses. You can do that by committing to locking it away until a date in the future, when you will need it. What is a date that makes sense? Choose a time that is close to when you will need the money for the reasons you just described, so that you aren’t tempted to spend it on other things. Ask the group: “When do you think you want to access money you would save for the future?” Let the group discuss several dates. Make sure they consider purchas ing inputs, and also food during the hunger season. Person 2: Hmmm. November 1 is probably a good time. That will be in time for me to buy fertilizer and pay my loan deposit. 35

37 Person 1: ecide how to divide your money between things you Now that you have chosen a date, you have to d the future. This is an important choice. You have to will buy before that date, and the things you are saving for in make sure that you have enough money for your immediat e needs and things you will have to buy before the date you have chosen. You also have to estimate how much money you will need for the things you want to buy in the y you will have after you sell your tobacco and repay your future. Start with money you need soon. Of the mone loan, how much do you need to have available for spending before November 1, which is the date you have chosen? Have the group suggest amounts of money they will spend on immediate expenses. Person 2: Well, I need to pay someone for ganyu. And I need to buy clothes, and some household items right away. I will also need to spend some money after the harves t season on small things like soap. I will need to spend MK 25,000 between when I get money and November 1. Person 1: Ok. How much do you want to make sure to have for the future, after that date you have chosen? I will need MK 4,500 for fertilizer, and MK 3,000 for a deposit on a new loan. Also, I want to keep Person 2: MK 2,000 for food in the hungry season. That is MK 9,500 total. So in total, your plan is to spend at least MK 25,000 now, and MK 9,500 in the future. That is MK Person 1: profit after selling your tobacco and repaying your loan? 34,500. Do you think you will have at least that much Person 2: Yes, I think I will have about MK 40,000. Good. If you earn that much, then the extra m oney can be available immediately. Then you can Person 1: commit to saving MK 9,500 for the future, and keep your other money available to spend sooner. You don’t have to spend it all before your date of November 1, of course, but it will be available while you are committing to lock away MK 9,500 until then. You made three decisions: You decided how much money you needed immediately, you decided how much money to lock away for the future, and you decided when you needed to access that locked away money. Person 2: Yes. Those weren’t hard decisions. But let’s demonstrate how it would work if I had chosen different options. Section 3: Account Allocati on Demonstration (All Clubs) In this section, the two enumerators will work together to do a demonstration with bottle caps. You will need 12 bottle caps for this demonstration. Draw two big circles in the dirt, and make sure everyone can see them. These circles represent money available for use immediately (point at one circle) and money committed to be (point at the other circle). These bottle caps represent money. Think of each cap as MK 1,000. saved for the future So, the 12 caps I have here represent MK 12,000 that someone has after selling his crop and repaying his loan. Now, if I need MK 3,000 now and commit to saving MK 5,000 for the future, then the first MK 3,000 I earn goes in this circle, for use immediately (put 3 bottle caps in the immediate use circle). Then, the next MK 5,000 I earn gets locked away for the future ( put 5 bottle caps in the future circle). Any extra money is available for use in the future, even though I don’t have to spend it immediately it is not locked away ( put the remaining 8 bottle caps in the immediate use circle). ( Collect all of the bottle caps). Think of this like a debt. I owe the ordinary account 3 bottle caps, and I owe ry account first, before I pay the commitment account. the commitment account 5 bottle caps. I must pay the ordina Suppose I get 10 bottle caps after I sell my tobacco and repay my loans. (Hold up 10 caps). First, I put 3 for immediate use. ( .) Next, I lock 5 away for use in the Put 3 caps in the immediate use circle future. ( Put 5 caps in the future use circle .) Then, since I’ve met the targets for immediate use and future use, I put all the other caps in the immediate use circle. (Put the remaining 2 caps in the immediate use circle .) 36

38 What if I only get 3 caps? ( Have someone come up to demonstrate. Give the person 3 caps. See where he puts them. All 3 should go in the immediate circle, and none in the future circle. If he gets this wrong, ask if anyone has a different idea. Explain if necessary.) (Enumerator, if farmers don’t understand the demonstration you just performed, please skip back to the start of the demonstration and explain the bottle caps idea again.) Have a volunteer come up and give him 6 caps. Correct answer: 3 in immediate, 3 in What if I get 6 caps? ( future.) What if you get 12 caps? ( Have another volunteer come up, etc. Co rrect answer: first put 3 in immediate, .) then 5 in future, then 4 more in immediate. Total is 7 immediate, 5 in future Dividing the bottle caps between the two circles is just like the spending plan you made before. You decide how much money you need to have available for immediate us e. When you get money, it is first made available for Point at the immediate use circle). Then, you decide how much to save for immediate use, up to the goal you set. ( Point at the the future. After making sure you have money for immediate use, you protect money for the future. ( Then, if there is money left after you meet both your immediate and future goals, that extra future use circle). Point at the immediate use circle). This way, you can money remains available for use whenever you choose. ( make a plan for how to divide your money between money you need now, and money you can commit to saving for the future, even when you don’t know exactly how much you will earn. y) Accounts (All Clubs Except Group 0) Section 4: Offer of Kasupe (Ordinar We have talked a lot about how to make a budget that gives you enough money for immediate needs Person 1: and commits you to saving money for the future. Also, we’ve discussed why saving at the bank is useful. Person 2: Yes. I can make a plan about the amount of money I need for the short term, an amount I want to be sure to save for the future, and a date in the future when I will want that money. But how am I to use the bank? Person 1: Usually, when you are paid for your tobacco, money is put into your group account. Then, the club officers give you your share of the cash. You leave it in the group account if you want. Or, you can save it at the bank, but to do that, you have to take your cash to the bank and deposit it into your individual account. Yes. It is inconvenient to have to take the money back to the bank, and often, I am tempted to Person 2: spend the money as soon as I receive it. Person 1: This season, we are offering you a new option. You can sign up to have your money transferred That means that when your bales of tobacco clear the auction floor, OIBM directly into your own Kasupe account. would automatically put the money you have earned after repaying your loan into your own Kasupe account. Person 2: How would OIBM know which money was mine and which money belongs to others in my club? Person 1: You would have to agree that OIBM could get a copy of your seller sheet from Auction Holdings. OIBM would use the information on th e seller sheet to figure out how much money should go into your account. Person 2: So if I agree to this, what do I have to do? Person 1: The first thing to do is to open a Kasupe account, if you don’t already have one. We can help with filling out the forms. The next thing to do is to sign a form authorizing the direct deposit. You can do both of those things today. Person 2: That’s all I have to do? Yes. It is very easy. If you open an account or already have one, and fill out the form for direct Person 1: deposit, then your mone y will be put into your indivi dual account automatically when your tobacc o is sold and your loan has been recovered. 37

39 Ask the group if there are any questions about how to sign up for direct deposit. What if I decide I don’t want to try this system and I would rather have my money go into the club Person 2: account? You can still open a Kasupe account. Just don’t fill out the [BLUE] form. Then, you will continue Person 1: to get your money from the club officers, who will withdraw it from the club account for you. But if you do choose to have the money sent directly to your individual accoun t, then ALL of your money fo r tobacco this season will go to the individual account. You can’t change your mind part way through the season. Ok. I think I want the direct deposit. If I sign up for that, how do I get my cash? Person 2: Person 1: You can withdraw cash from the bank. You can either use your smartcard, or make the withdrawal by talking to a teller. You can do this at the branch or kiosk, or when the mobile bank comes to town. The closest awal is __ place to make a withdr ____________. So I can take money out whenever I want? Person 2: Person 1: Yes, you can, but you should remember the commitment you thought about to save money for a date in the future. ent) Accounts (Commitment Clubs Only) Section 5: Offer of SavePlan (Commitm Is there a way that OIBM can help me keep that commitment? Person 2: Yes. You can open a special “SavePlan” account in addition to your Kasupe account. Person 1: Person 2: How would that work? Person 1: Opening a SavePlan just tells the bank to follow the plan you made before. You will fill out a form s you made earlier: how much money you need to have available for immediate use, the with the three decision amount of money you want to lock away for the future, and the date you want that money released. Person 2: That is easy. It’s just writing down decisions I’ve already thought about. What happens after I fill out the form? Once you fill out the form, OIBM will use it to put the money you are saving for the future in a Person 1: You won’t be able to take money out of that account until the date you special, individual, commitment account. have chosen, and you can’t change your mind about the date or the amount of money. Person 2: Do I earn interest on m oney in this special account? Person 1: Yes. You earn the same interest on money in the commitment account as in the ordinary Kasupe account. The only difference is that th e money in the commitment account is locked away until the date you have chosen. Person 2: What if I earn more or less money than I thought I would have? Person 1: It works just like the bottle caps. After the loan is recovered, money first goes into your ordinary Kasupe account, up to the amount you said you needed to have available immediately. Then, money goes to the SavePlan to be locked away for the future. When you have reached your target for saving for the future, extra money earned after that amount goes back to the ordinary Kasupe account. So if I don’t earn as much as I thought, I will still have money available immediately? Person 2: Person 1: Yes. Money goes to the Kasupe account first, a nd you can withdraw from that whenever you want. It only goes to the special commitmen t account when you have reached your target for immediate spending. 38

40 Person 2: So this form just tells the bank to stick to the commitment I made to myself about how much to save for the future, and when I can use that money. Person 1: That’s right. You can choose any amount and date you want, and OIBM will hold it for you so that special account in addition to if you would like to use this you stick to the plan. We can help you fill out the form the regular Kasupe account. Section 6: Raffle (All Raffle Clubs) As an extra incentive to save money, there will be a raffle draw where some farmers in this project may have a chance to win a prize. You have to save to have a chance to win, and the more you save, the better your chance to win. There will be two prizes in each district. The first prize will be a new bicycle, and the second prize will be a 50 kg bag of D-compound. y you save in your bank account. The prizes will be The raffle tickets will be based on the amount of mone two times before then. The first time will be in August awarded in November. The raffle tickets will be given out at when we will come back and give you tickets based on the money you have saved between July 1 and August 1. OIBM will calculate the average balance in your savings account for those 30 days and the number of tickets you will get will be based on this amount. The second time we ha nd out tickets will be in Oc tober. OIBM will calculate your average balance from September 1 to October 1, and give you additional tickets based on that balance. Each d on their account balance. The prize is for individuals and not for the club. person will get individual tickets base You can increase your chance of winning by saving mo re money and saving it for a longer time. You will get one ticket for every MK 1000 in your average balance. If you put MK 10000 in your account by July 1 and keep it there until at least August 1, then you will get 10 tickets. If you don’t have any money in your account from July 1 to July 14, and then put MK 10000 into your account on July 15 and keep it there until at least August 1, you will only get five tickets. If anyone here has two accounts with OIBM, we will add up the balance in both accounts. Money saved with other banks will not count for the raffle, though. Section 7A: Public Raffle (Public Raffle Clubs Only) We will hand out the raffle tickets in August and October during group meetings like the one we are having so your friends will know how many tickets you are getting. today. We will give out the tickets in front of others, I will demonstrate how tickets will be handed out. I am going to hand you a piece of paper with a number on it. Pretend that is your average account balance from July 1 to August 1. No one but you and OIBM knows this number, so don’t tell anyone! (Distribute the papers with fake account balances to 5 volunteers) Now, I will give you the number of raffle tickets you get for that balance. Come up one at a time and show me your piece of paper, so I can give you your tickets. (Have the farmers come up one at a time. Look at the paper and hand out tickets. Make sure to say out loud for every farmer how many tickets he gets. Make sure th at the other farmers are pa ying attention to this.) ill work the same way. Yo u will each be called up one When we hand out tickets in August and October, it w at a time to receive tickets based on the amount you ha ve saved, and your club w ill see how many tickets you receive. Section 7B: Private Raffle (Private Raffle Clubs Only) We will hand out the raffle tickets in August and October during group meetings like the one we are having today. We will give out the tickets one at a time, so no one will know how many tickets you are getting. I will demonstrate how tickets will be handed out. I am going to hand you a piece of paper with a number on it. Pretend that is your average account balance from July 1 to August 1. No one but you and OIBM knows this number, so don’t tell anyone! 39

41 (Distribute the papers with fake account balances to 5 volunteers) Now, I will give you the number of raffle tickets you get for that balance. Come up one at a time and show me your piece of paper, so I can give you your tickets. (Have the farmers come up one at a time. Look at th e paper and hand out tickets. Make sure no one sees how many tickets you hand to each person.) When we hand out tickets in August and October, it w ill work the same way. Yo u will each be called up one at a time to receive tickets based on the amount you have saved, and no one will know how many tickets you have received. 40

42 Appendix B: Variable definitions Data used in this paper come from two surveys as well as from administrative records of our partner financial institution (OIBM). We conducted a baseline survey from March to April 2009 and an endline survey from July to September 2010. data are top coded at the 99th percentile for All variables that are created from survey variables with a positive range and bottom a nd top coded at the 1st and 99th percentile respectively for variables with a range that spans both negative and positive values. All figures in money terms are in Malawi Kwacha (MK). Baseline characteristics (from baseline survey): Number of members per club is the number of listed club members per information provided by the buyer companies (Alliance One and Limbe Leaf). Not all club members were interviewed. equals 1 for female respondent s and 0 for male respondents. Female equals 1 for married respondents and 0 Married for respondents who are single, widowed, or divorced. Age is respondent’s age in years. Years of education is the respondent’s years of completed schooling. Household size is the number of people counted as members of the respondent’s household at the time of the baseline survey. is an index based on the fi rst principal component of the number of items owned Asset index of 14 common non-financial, non-lives tock assets and indicators of presence of 4 major types of housing characteristics (iron sheet roof, glass wi ndows, concrete floor, el ectricity connection). Livestock index is an index based on the first princi pal component of the number of animals owned of 7 common t ypes of livestock. is the total of area of land under Land under cultivation cultivation, measured in acres, for the late-2008 planting season. Proceeds from crop sales is the sum of sales from the two main cash crops, maize and tobacco, in the 2008 harvest. Cash spent on inputs is the total amount of cash spen t – excluding the value of input packages that are part of a loan -- on seeds, fertilizer, pesticides, and hired labor for the 2008- 2009 planting season Has bank account is 1 if a household member has an account with a formal financial institution, and 0 if not. Savings in accounts and cash is the sum of current savings with formal institutions and in cash at home. Hyperbolic is 1 if the respondent exhibited stric tly more patience in one month, hypothetical monetary trade-offs set 12 months in the future than in the same trade-offs set in the present, and 0 otherwise. See section 5 above for more details. Patient now, impatient later is 1 if the respondent exhibited strictly less patience in one month, hypothetical monetary trade-o ffs set 12 months in the future than in the same trade-offs set in the presence and 0 otherwise. Net transfers made in past 12m is the total of transfers made to the social network minus the sum of transfers received from the social netw tegories (social events, ork, summed across six ca health shocks, education of children, agri cultural inputs, hired labor and ‘other’). Missing value for formal savings and cash is 1 if the variable “ Savings in accounts and cash” is missing and 0 if it has valid values. 41

43 Missing value for time preferences is 1 if the respondent has missing values for the time Hyperbolic ” and “ ”) is missing, and 0 if these preferences variables (“ Patient now, impatient later variables have valid values. Transactions with Partner Instituti on (from internal records of OIBM): es any deposit from his or her Any transfer via direct deposit is 1 if the respondent receiv tobacco club’s account to his or her individual savings account, and 0 if not. is the sum of (positive) transactions into the Deposits into ordinary accounts, pre-planting respondent’s OIBM ordinary savings accounts during the period of March to October 2009. Deposits into commitment accounts, pre-planting is the sum of (positive) transactions into nt savings accounts during the the respondent’s OIBM commitme period of March to October 2009. is the sum of (positive) transactions into the Deposits into other accounts, pre-planting the period of March respondent’s OIBM non-ordinary, non-commitment savings accounts during to October 2009. Total deposits into accounts, pre-planting is the sum of transactions into the respondent’s OIBM accounts (sum across all accounts) dur ing the period of March to October 2009. Total withdrawals from is the sum of transactions out of the accounts, pre-planting respondent’s OIBM accounts (sum across all account s) during the period of March to October 2009. Net deposits, pre-planting is the difference between all deposits and withdrawals in the respondent’s OIBM accounts during the period of March to October 2009. Net deposits, Nov-Dec is the difference between all de posits and withdrawals in the respondent’s OIBM accounts during the pe riod of November and December 2009. is the difference between all de Net deposits, Jan-Apr posits and withdrawals in the respondent’s OIBM accounts during the period of January through April 2010. Agricultural outcomes, household expenditure, a nd other variables, from endline survey (all planting and harvest variables refer to the 2009-2010 planting season): Land under cultivation is the total area of land under cultivation, measured in acres. Cash spent on inputs is the total amount of cash spen t – excluding the value of input packages that are part of a loan -- on seeds, fertilizer, pesticides, and hired labor for the 2009- 2010 planting season. Total value of inputs is the sum of cash spent on agricult ural inputs plus the value of inputs included in-kind in loan packages for the 2009-2010 planting season. Input categories include seeds, pesticides, fertilizer, hired labor, transport and firewood (for curing tobacco). Proceeds from crop sales is the sum of sales from the two main cash crops, maize and tobacco for the 2009-10 planting season. is the sum of revenue from crop sales and the value of Value of crop output (sold & not sold) the unsold crop for seven main crops (maize, bur ley tobacco, dark fire tobacco, flue-cured tobacco, ground nuts, beans, soya). Value of harvest not sold equals the kilograms of crops not sold multiplied by the price/kilogram, summed across the seven main crops. Price/kilogram for each crop is obtained by calcula ting crop-specific revenue/kilogr am for each observation in the sample and then taking the sample average. Farm profit (output - input) is the difference between “Value of crop output” and “Total value of inputs” defined above. 42

44 Total expenditure in last 30 days is the sum of three categories household expenditures e last 30 days prior to the endline survey. (food, non-food household items and transport) over th is the number of people counted as members of the respondent’s household Household size at the time of the endline survey. is the total of transfers made to the social network over the 12 months Total transfers made x categories (social events, health shocks, prior to the endline interview, summed across si education of children, agricultural inputs, hired labor and ‘other’). Total transfers received is the total of transfers received fr om the social network over the 12 months prior to the endline interview, summed across six categories (social events, health shocks, education of children, agricultur bor and ‘other’). al inputs, hired la Total net transfers made is the difference between “Tot al transfers made” and “Total transfers received” defined above. Tobacco club loan is the total amount owed as part of a tobacco club lo an for the 2009-2010 planting season. Not interviewed in endline is 1 if the respond ent was not interviewed and is 0 if the line survey of July to September 2010. respondent was interviewed during the end 43

45 Table : Summary Statistics 1 Standard 10th 90th Median Observations Mean Percentile Percentile Deviation conditions Treatment 0.135 group 0 1 3150 Control 0.341 0 0.493 0 1 3150 0.417 0 Commitment Account 0.497 0 0 1 3150 Ordinary Account 0.448 1 Raffle 0.450 0 0 0.282 3150 x Commitment Raffle Commitment Public 0.139 0.346 0 0 1 3150 x x x Raffle 0.302 0.459 0 0 1 3150 Ordinary 0 x Public 0.153 0.360 0 Raffle 1 3150 Ordinary x Characteristics Baseline of members per Number club 13.88 6.44 9.00 11.00 23.00 299 0.063 0.000 0.000 0.000 3150 Female 0.243 0.208 1.000 1.000 3150 Married 0.955 1.000 45.02 28.00 44.00 64.00 3150 [years] 13.61 Age Years 5.45 3.53 0.00 6.00 10.00 3150 of education size 5.79 1.99 3.00 6.00 9.00 3150 Household index 0.02 1.86 ‐ 1.59 ‐ 0.67 2.46 3150 Asset ‐ index ‐ 0.03 1.15 ‐ 1.00 0.36 1.37 3150 Livestock ‐ 2.14 cultivation [acres] 4.67 2.50 4.03 7.50 3150 Land under spent on inputs [MK] 25168.64 Cash 0.00 10000.00 64500.00 3150 41228.45 3150 sales [MK] crop 174977.36 7000.00 67000.00 300000.00 125656.65 Proceeds from bank account 0.634 0.482 0.000 1.000 1.000 3150 Has 1243.79 in at home [MK] cash 3895.00 0.00 0.00 3000.00 3150 Savings in bank accounts [MK] 2083.16 8264.71 Savings 0.00 3000.00 2949 0.00 0.102 0.000 0.000 1.000 3117 Hyperbolic 0.303 0.460 now, later 0.304 0.000 0.000 1.000 3117 Patient impatient transfers made in past 12m [MK] 1753.46 7645.07 ‐ 2990.00 500.00 8100.00 3150 Net for formal Missing savings and cash 0.064 0.244 0.000 0.000 0.000 3150 value 0.000 for time preferences 0.010 0.102 0.000 0.000 3150 Missing value with Partner Institution Transactions transfer via direct deposit 0. 1 54 0.36 1 0.000 0.000 1 Any 3 1 50 .000 Any transfer via direct deposit 0.154 0.361 0.000 0.000 1.000 3150 planting ordinary accounts, pre ‐ Deposits [MK] 18472.21 82395.98 0.00 0.00 38907.23 3150 into Deposits into commitment accounts, pre ‐ planting [MK] 614.68 5366.73 0.00 0.00 0.00 3150 Deposits other accounts, pre ‐ planting [MK] 296.37 3804.48 0.00 0.00 0.00 3150 into Total into accounts, pre ‐ planting [MK] 19383.25 84483.49 0.00 0.00 40694.26 3150 deposits [MK] withdrawals accounts, pre ‐ planting from 18600.00 82744.16 38600.00 0.00 0.00 3150 Total of all transactions, pre ‐ planting Net 762.21 13857.42 0.00 0.00 649.45 3150 [MK] all transactions, Nov ‐ Dec of [MK] ‐ 848.00 6870.17 0.00 0.00 1.93 3150 Net 3.83 0.00 0.00 4032.43 269.49 3150 ‐ Jan ‐ Apr [MK] transactions, all of Net Survey Outcomes Endline 2.00 under (acres) 4.52 2.66 cultivation 4.00 8.00 2835 Land spent Cash on inputs [MK] 21631.74 32853.49 500.00 11000.00 51500.00 2835 value of inputs [MK] 68045.83 84014.11 Total 43750.00 157272.22 2835 1500.00 Proceeds from crop sales [MK] 109603.80 162579.79 0.00 56000.00 270000.00 2835 not Value output (sold & of sold) [MK] 177746.78 201130.84 27480.27 115581.93 387203.19 2835 crop Farm profit (output ‐ intput) [MK] 110703.16 156746.96 0.00 70371.97 264952.88 2835 Total in last 30 days [MK] 11904.75 13219.35 2250.00 7500.00 26000.00 2835 expenditure Household 2.15 3.00 6.00 9.00 2835 size 5.80 5098.9 transfers [MK] 3152.4 0.0 1300.0 8000.0 2835 Total made 2835 6050.0 500.0 0.0 4377.3 [MK] 2204.0 transfers Total received net transfers made [MK] 938.7 5896.4 ‐ 3000.0 350.0 5750.0 2835 Total Tobacco amount [MK] 40787.2 77961.5 0.0 0.0 130000.0 2835 loan 0.067 Has deposit account fixed 0.250 0.000 0.000 0.000 2835 Not interviewed in follow ‐ up 0.100 0.300 0.000 0.000 0.500 3150 our Data based on two surveys conducted in February to April 2009 (baseline) and July to September 2010 (endline), and on administrative records of partner definitions. variable for B Appendix See numbers. negative during study period). Withdrawals presented as kwacha US$1 = (MK145 Malawi is "MK" institution.

46 Table : Assignment of clubs to treatment conditions 2 Savings intervention: Savings intervention: ordinary and ordinary savings No intervention accounts commitment accounts offered offered No raffle Group 0: 42 clubs Group 1: 43 clubs Group 4: 42 clubs Public of raffle tickets N/A Group 2: distribution 44 clubs Group 5: 43 clubs clubs distribution of raffle tickets N/A Group 3: 43 clubs Group 6: 42 Private

47 ‐ the for prefe 0.00 0.01 0.01 (17) 0.00 0.01 0.00 0.01 ‐ ‐ ‐ 3,150 0.302 tests (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) rences Missing value time Raffle" and for No 0.01 0.01 (16) 0.01 0.07 0.02 0.01 cash 0.04* ‐ ‐ 3,150 0.947 (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) ‐ formal Missing value savings past Commitment, [MK] in = Net (15) 3,150 0.409 81.78 48.57 313.63 185.57 151.71 336.31 ‐ (496.27) (535.56) (560.04) (528.38) (541.27) (547.16) 1,655.33 transfers 12m made Raffle No now, 04) 0.06 0.04 (14) 0.35 0.01 0.02 0.07 later 0.08* ‐ ‐ 3,117 0.592 (0. (0.05) (0.04) (0.04) (0.05) (0.05) ‐ impatient Patient "Ordinary, variables. 0.01 0.03 (13) 0.01 0.00 0.03 0.00 0.10 ‐ ‐ baseline 3,117 0.293 (0.03) (0.02) (0.03) (0.02) (0.03) (0.02) regressions: Hyperbolic 17 all in on column cash 1.61 (12) for [MK] ‐ 2,949 0.531 309.33 535.10 515.36 769.73 200.15 ‐ ‐ (737.33) (857.36) (727.54) (885.36) (730.59) and 3,235.37 accounts Savings (1,009.58) dummies tests ‐ F 05) 05) 05) 04) 04) 05) bank 0.02 0.02 0.04 (11) 0.05 0.01 0.66 ‐ ‐ ‐ 3,150 0.265 (0. (0. (0. (0. (0. (0. 0.09** effects. treatment ‐ account Has fixed cell respective spent inputs 798.46 (10) of [MK] 316.48 3,150 0.981 2,925.17 6,525.43* on ‐ ‐ Cash stratification regression crop [MK] (9) in 3,150 0.791 5,823.24 2,093.43 3,052.18 3,623.95 6,474.56* 7,888.38 2, Proceeds 14,051.63 include ‐ ‐ from 11,741.81 2,232.79 sales ‐ 117,494.92 21, significance under (8) 0.02 0.14 0.01 0.25 0.11 4.67 regressions ‐ ‐ ‐ 3,150 0.530 (0.18) (12,806.36) (3,594.99) (0.18) (13,121.34) (3,750.31) (0.16) (12,340.76) (3,594.52) (0.18) (13,191.49) (3,414.65) (0.20) (12,765.47) (2,625.86) (0.19) (13,686.32) (3,917.08) [acres] 0.60*** joint ‐ cultivation All Land of 145. test MK (7) 0.05 0.01 0.02 0.04 0.08 0.03 0.03 ‐ ‐ ‐ ‐ ‐ ‐ index 3,150 0.826 (0.09) (0.11) (0.08) (0.12) (0.10) (0.12) ca. Livestock is 1 significance: USD index joint (6) 0.08 0.11 0.26 0.13 0.24 0.04 0.27* ‐ ‐ ‐ 3,150 0.532 of (0.17) (0.16) (0.15) (0.15) (0.16) (0.18) ‐ level. Asset club tests ‐ F the at 16) 15) 15) 16) 16) 15) (5) 0.09 0.08 0.11 Size 0.18 5.81 ‐ ‐ ‐ 3,150 0.28* 0.824 (0. (0. (0. (0. (0. (0. 0.45*** ‐ Household treatments. clustered of are raffle (4) 0.08 0.39 0.19 0.14 0.38 0.03 5.31 ‐ ‐ 3,150 0.169 (0.28) (0.29) (0.25) (0.24) (0.28) (0.26) Years education errors additional Standard [years] (3) 0.40 0.57 1.17 0.93 0.88 0.11 ‐ ‐ ‐ ‐ ‐ 3,150 0.122 0.896 0.140 0.637 46.23 0.832 0.158 0.473 (1.22) (1.11) (1.02) (1.09) (1.25) (0.95) without levels. each (***) (2) 0.02 0.02 0.02 0.97 0.01 groups ‐ ‐ ‐ 3,150 0.360 1% (0.01) (0.02) (0.01) (0.01) (0.01) (0.02) 0.03** 0.03** ‐ Married Age variables: and 17: ‐ 1 (**), treatment baseline (1) 5% 0.02 0.00 0.02 0.03 0.00 ‐ ‐ 3,150 0.03* 0.519 (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) of Female 0.05*** (*), Characteristics columns in 10% commitment at and Baseline significance = Public in x Public regressions joint x Public Raffle significance ordinary x Public for for x Control Raffle Raffle in Balance No x x Account Raffle Raffle Raffle in Account x x of tests tests Raffle Raffle ‐ ‐ variable: No indicate F F observations x x Account Raffle Raffle Var means x x Account of of of Test of : Stars Dep 3 values values Ordinary Ordinary Commitment Commitment Ordinary Ordinary, Commitment, Commitment ‐ ‐ Ordinary Ordinary Mean P Number P Commitment Dependent Table Ordinary Commitment Commitment equality Notes:

48 [MK] 2008 2010 45) . the with and (9) deposits 0.440 3,150 39.32 80.06 56.17 Apr 352.23 255.61 361.32 147.938 ‐ ‐ ‐ (327 45) (327 (201.25) (357.64) (351.51) (224.86) (298.18) ‐ variables: deposits are during Jan without replaced hyperbolic sales each baseline in deposits values [MK] Net maize Net 2009 value groups 91) . and (8) following 9.835 Dec). Dec 0.817 3,150 254.58 110.77 130.66 ‐ ‐ (missing 434.73 ‐ 941.67* ‐ ‐ ‐ (544 91) (544 (617.48) (655.39) (397.70) (401.48) (481.11) ‐ the deposits 1,065.54*** missing ‐ Nov (Nov treatment tobacco and for hyperbolic from season effects [MK] Net for dummy ordinary 2009 36) . fixed and (7) Oct planting 114 0.678 3,150 52.71 proceeds , ‐ 24.692 627.91 399.44 346.81 1,235.14 (928.21) (972.55) (669.29) dummy cell ‐ 1,120.77* of (1 114 36) (1 (1,173.40) (1,027.00) amount; deposits Mar Net start zeros); savings commitment and cultivation; in with in stratification into 2009 10) under value . before) (6) accounts means Oct 39.22 0.846 3,150 28.82 [MK] 180.46 ‐ ‐ 329.26 403.38 485.93* ‐ include replaced 174.087 (271 10) (271 (249.67) (120.20) (366.93) (347.93) (278.44) land ‐ of and Deposits missing Mar other (Oct values index; for equality regressions period the missing All into dummy [MK] 2009 livestock 17) . tests (5) 145. 1.19 Oct 95.34 0.009 0.000 3,150 45.42 (with 698.37 100.46 ‐ ‐ planting ‐ ‐ (253 (253 17) (548.96) (219.45) (910.62) (788.59) (234.94) ‐ index; 1,993.55** MK months; commitment Deposits Mar cash pre accounts Raffle" 12 ca. or asset is No both 1 over in bank USD in into 2009 39) . members; occur Ordinary, accounts (4) network = level. Oct 274 0.826 3,150 [MK] , ‐ 4,453.27 savings 1,781.13 3107.045 ‐ (7 274 39) (7 (6,714.08) (7,777.86) (7,192.16) (6,282.23) (6,962.06) 14,133.97* 13,914.46* ‐ 21,367.01*** 19,464.30*** club of Deposits social Mar Raffle ordinary to purchases the household No at of amount made 2009 Fertilizer 60) . number clustered (3) accounts Oct 032 account; [MK] 0.979 3,150 Dec. , ‐ withdrawals ‐ 1,315.18 transfers 3256.440 5,384.82 ‐ "Commitment, ‐ are (7 032 60) (7 (6,509.39) (7,891.30) (7,546.98) (6,829.20) (6,750.73) 15,155.79* ‐ 14,965.51** 20,740.45*** 20,967.89*** ‐ ‐ Nov Mar from net bank Total in tests: education; errors ‐ F formal zeros); B. into occurs [MK] any 2009 53) with . Standard of completed (2) Oct 365 0.977 3,150 , ‐ of Appendix 5,332.11 deposits 1,714.62 3281.132 ‐ (7 (7 365 53) (6,897.70) (7,829.14) (7,847.77) (6,884.88) (7,073.18) 13,920.65* 14,618.70** levels. ‐ 21,861.22*** 21,595.81*** application see replaced Mar accounts years Total ownership (***) values for 1% years; via Fertilizer Apr in definitions, ‐ Withdrawals and Up ‐ Apr. ‐ 05) age Deposit . dummy (1) (missing 0.08 0.03 0.01 0.04 ‐ and 0.000 3,150 0.333 2010 (0 (0 05) (0.06) (0.06) (0.06) (0.05) (0.05) (**), 2009 0.16*** 0.21*** Transfer Nov (Take Indicator) variable is 5% later" Direct Mar Any season; married; (*), for Deposits season 2009 10% complete on at impatient the For = dummy for Planting now, Public later". x Raffle inputs significance Treatments Public No on Control x "patient Raffle of respondent; Raffle Raffle treatments. in for Account x x No impatient test: variable: spent ‐ indicate observations Raffle Raffle Var male F raffle x x Account of Impact of now, for : cash dummy Stars Dep 4 withdrawals. period: Commitment, Ordinary, value ‐ Commitment Dependent P Number Table Mean Ordinary Ordinary Commitment Commitment Ordinary Time Dummy "patient additional season; minus zeros); Notes:

49 in for club to of the of inputs [MK] at prior replaced on (6) amount; across 2,835 0.141 177.14 532.12 years 876.83 412.76 277.95 equality ‐ ‐ (968.76) (888.51) (979.02) (856.78) (873.19) (876.40) 10678.42 expenditure days 1,859.77** survey values the spent 30 Unrelated clustered savings years; Total in in cash tests are ‐ (missing Commitment age value on Seemingly errors Raffle" season; on (output No [MK] married; missing 2008 (5) hyperbolic 2,835 0.142 184.16 for based 2,063.48 1,888.26 3,046.85 Standard for 95209.65 19,205.08 profit 15,371.41 the coefficient for (12,941.66) (13,481.03) (11,737.00) (12,398.64) (12,985.04) (11,194.60) input) are Ordinary, the = Farm levels. dummy during dummy dummy equals Raffle (***) sales No 1% sold) zeros); months; output 35 maize 12 respondent; and not Ordinary with (4) crop [MK] 2,835 0.081 and 936.87 on Harvest ‐ 6,340.87 over and 7,844.33 9,197.65 9,486. of (**), ‐ male 155684.93 (15,460.54) (15,941.24) (15,159.76) (15,115.21) (15,750.11) (14,805.72) 33,967.76** 5% "Ordinary=Commitment" for 2010 replaced "Commitment, (sold Value tobacco (*), and network coefficient after tests: values Dummy 10% from ‐ F the social at crop if B. to and missing [MK] from "Ordinary=0" proceeds (3) variables: 2,835 0.254 0.694 0.042 0.297 made 414.80 of zero Expenditure 8,208.04 13,509.38 91746.78 11,704.31 10,305.56 (with Appendix ‐ (13,818.39) (13,774.27) (12,204.97) (11,613.70) (11,935.35) (12,609.42) sales 22,962.78** significance tests see from ‐ cash baseline Proceeds F transfers or cultivation; indicate 6: net Household bank different under inputs following and Stars in definitions, and 4 of 6: treatments. zeros); the land 2, (2) 2010. [MK] 2,835 0.276 and 8,507.97 1,381.03 7,337.26 6: 1, 8,521.34 7,223.72 savings and with 60371.80 ‐ ‐ ‐ value (6,224.95) (7,214.78) (6,571.42) (6,394.38) (5,865.43) (6,272.71) raffle variable 4 Season 16,533.52** of index; significantly 2, and For 4 Total 1, effects 2010 ‐ 2, replaced columns September jointly amount additional later". is livestock 1, to of fixed 2009 5. values July in cell columns index; cultivation without and of account; Ordinary (1) 0.13 0.01 3 columns 4.28 0.05 0.25 0.16 impatient ‐ ‐ 0.057 2,835 (0.19) (0.19) (0.21) (0.20) (0.19) (0.20) 0.42** [acres] during 2, on asset of (missing each regressions bank under 1, now, Outcomes stratification later" Land groups across formal regressions column conducted members; coefficient "patient any regressions include the of from across and if impatient survey Agricultural 0 treatment up across test on ‐ household 0 now, of variables = and regressions ownership hyperbolic follow ordinary All for in Public "patient the x and Raffle number 145. Commitment= Ordinary= Ordinary=Commitment Public Treatments for value No x Control estimation from of of of Raffle MK of dummy independent Raffle Raffle in Account x x No ca. test test: test test variable: ‐ ‐ ‐ ‐ (SUR) observations is come dummy with Raffle Raffle Var missing F F F F education; 1 x Account x commitment of Impact season; of of of of for : in Data Dep 5 USD zeros); 2009 Ordinary, Commitment, value value value value ‐ ‐ ‐ ‐ Commitment Dependent P Ordinary Table P P Ordinary Commitment Commitment Ordinary Number P Mean the level. means regressions with completed Regressions Notes: dummy

50 in sales the cash value deposit 02) or . (missing years; (6) 0.02 tests 0.04 0.02 0.01 maize ‐ 2,835 0.250 (0.02) (0.03) (0.02) (0.02) (0.02) (0 02) (0 0.03** 0.03** in ‐ 0.06*** All account fixed missing bank and later" age in 145. Raffle" Has for No MK tobacco ca. savings married; impatient dummy loan is 23) of . 1 for from (5) Ordinary, now, 728 2,835 0.903 , = 5,156.98 8,524.39 3,784.89 4,508.14 5,578.51 3,686.22 amount USD ‐ ‐ 40,147.26 (6,809.44) (6,552.34) (5,705.79) (6,613.50) (7,190.95) (6 728 23) (6 amount; amount Tobacco dummy proceeds Raffle level. "patient No savings for club account; in made the net 25) . respondent; cultivation; bank at (4) value dummy [MK] 2,835 0.846 445.54 417.89 120.96 641.82 148.57 714.81* ‐ 724.38* (436.42) (403.40) (405.43) (421.69) (442.23) (390 25) (390 ‐ Total male under formal transfers treatments. "Commitment, for zeros); missing clustered land any for raffle tests: with are of ‐ F Dummy [MK] demand B. index; 52) . errors dummy (3) transfers 6.12 44.68 2,835 0.299 205.87 156.28 ‐ replaced 490.38* ‐ ‐ (322.20) (251.87) (290.27) (257.67) (310.07) (293 (293 52) 2,492.13 ‐ additional 512.32** ownership variables: deposit received livestock Total Appendix for values months; Standard see without 12 fixed index; baseline levels. dummy each and over (missing 69) [MK] asset . (2) transfers (***) 2,835 0.553 45.63 97.01 118.62 490.80 278.24 556.49* definitions, ‐ (352.28) (380.59) (330.70) (387.17) (333.87) (322 (322 69) 2,871.70 ‐ season; groups following 1% made network Total transfers the hyperbolic and members; 2009 variable social for and size, the (**), to treatment size For for 5% 11) effects dummy . household made (1) 0.05 0.13 5.72 0.03 0.06 0.13 0.05 (*), ‐ 0.952 2,835 (0.11) (0 11) (0 (0.12) (0.11) (0.12) (0.11) later". of inputs ordinary household fixed 10% on zeros); Household and on at cell transfers number = with impatient spent net Public cash now, x Raffle zeros); significance commitment Public replaced treatments stratification education; No Control x Raffle in of Raffle Raffle season; in with "patient Account x x No values test: variable: ‐ indicate observations Raffle Raffle Var include F means and 2008 Account x x of Impact of of completed : Stars the Dep replaced 6 of missing Ordinary, Commitment, value ‐ years hyperbolic (with regressions values during equality Commitment Dependent Notes: Ordinary Commitment Number Ordinary Commitment Ordinary P Mean Table

51 Appendix 1 : Attrition from Baseline to Endline Survey Table (A) (B) Including Control s No Baseline controls Baseline during interviewed during endline endline Not interviewed Not Dependent variable: survey survey 0.01 0.00 Account Commitment (0.02) (0.03) Account ‐ 0.00 0.00 Ordinary (0.03) (0.03) Commitment Raffle ‐ 0.01 ‐ 0.02 x (0.02) (0.02) Raffle x Public x 0.00 0.01 Commitment ‐ (0.02) (0.02) x Raffle ‐ Ordinary ‐ 0.04 0.03 (0.02) (0.02) Public x Raffle x Ordinary 0.05** 0.05** (0.02) (0.02) Dep Var in Control 0.10 0.10 Mean Number observations 3,150 3,150 of value of F ‐ test: ‐ P No Raffle Commitment, = No Raffle Ordinary, 0.802 0.790 Stars indicate significance at 10% (*), 5% (**), and 1% (***) levels. Standard errors are clustered at the club level. Notes: stratification cell fixed effects. Column A includes the following include baseline variables: Dummy for male Regressions household members; asset index; age in years; years of completed education; number of respondent; dummy for married; livestock land under cultivation; proceeds from tobacco and maize sales during the 2008 season; cash spent on index; the 2009 season; dummy for ownership of any formal inputs bank account; amount of savings in bank or cash for (with now, "patient for dummy zeros); with (missing replaced values zeros); for dummy hyperbolic with replaced values missing impatient later" (missing values replaced with zeros); net transfers made to social network over 12 months; dummy for missing in savings amount; dummy for missing value in value hyperbolic and "patient now, impatient later". For variable in means of equality the tests: "Commitment, No Raffle = Ordinary, No Raffle" tests ‐ B. Appendix see definitions, F commitment and ordinary treatment groups each without additional raffle treatments.

52 the [MK] of tests 2010 26 . (9) start 3,150 0.430 50.70 48.24 83.42 Apr 276.00 147.94 426 426 26 ‐ ‐ ‐ ‐ 411.61* (230.70) (320.09) (295.59) (201.39) (355.86) (355.71) Raffle" deposits and Jan No before) and [MK] Net 2009 Commitment, (Oct 09 . = (8) 9.84 42.25 51.06 ‐ Dec 3,150 0.824 201.21 ‐ ‐ 377 09 377 ‐ 966.40* ‐ (397.27) (533.49) (491.51) (621.41) (411.08) (644.27) ‐ deposits period 1,090.51*** Raffle ‐ Nov No planting ‐ pre [MK] Net "Ordinary, 2009 07 both . (7) in tests: Oct 3,150 24.69 0.721 ‐ ‐ 672.70 177.56 240.01 308 308 07 1,285.14 F (944.96) (674.72) (966.01) 1,104.33 ‐ (1,072.73) (1,099.16) (1,033.92) deposits Mar occur Net effects. fixed purchases other cell [MK] 2009 45** into . (6) Oct 26.34 83.21 3,150 0.829 139.70 ‐ ‐ ‐ Fertilizer 411.24 174.09 329.61 ‐ (117.53) (290.73) (249.12) (263.08) (351.68) (384.74) 502 502 45** ‐ Mar accounts Dec. ‐ stratification Deposits Nov in include into [MK] 2009 occurs 93 . (5) 0.00 Oct 77.07 26 26 93 3,150 78.64 0.010 168.93 660.97 ‐ ‐ ‐ ‐ ‐ (214.49) (558.89) (248.51) (792.99) (229.74) (907.34) regressions 1,981.98** commitment Deposits Mar accounts All application 145. MK Fertilizer ca. into 82* 2009 is . Apr. ‐ accounts 1 (4) 265 Oct , [MK] 3,150 0.844 ‐ Nov 2,264.35 3,317.70 3107.05 ‐ ‐ USD 12,712.39 (8,230.31) (6,886.74) (7,956.91) (6,544.53) (7,744.01) (7,265.29) 13 265 82* 13 ‐ is 21,061.32*** 22,927.66*** Deposits Mar ordinary level. season club [MK] the at Planting 27* 2009 . (3) 103 Oct , 3,150 0.987 ‐ withdrawals 2,756.50 4,295.93 3256.44 ‐ ‐ accounts (8,375.91) (6,699.59) (7,732.94) (7,525.24) (7,097.81) (7,664.60) 14 103 27* 14 14,049.83* ‐ 22,350.21*** 22,507.50*** clustered Mar Total treatments. from are controls raffle errors into baseline [MK] 20* 2009 . (2) additional 795 Oct Standard , 3,150 0.978 ‐ 2,516.49 4,118.37 3281.13 deposits ‐ ‐ 12,764.69 (8,291.97) (7,079.88) (8,051.13) (7,845.12) (7,191.44) (7,950.51) 13 795 20* 13 ‐ 23,454.54*** 23,180.20*** Mar accounts additional levels. Total without Withdrawals ‐ (***) withdrawals. and groups 2010 1% without via (Take minus and Apr but 00 ‐ . (1) 0.09 0 0.03 000 0.04 0.00 ‐ Deposits 0.361 3,150 (0.06) (0.06) (0.05) (0.05) (0.05) (0.06) Indicator) (**), 0.22*** 0.17*** Transfer treatment Deposit 2009 on effects Up 5% deposits Any (*), are Mar Direct fixed 10% cell commitment at Treatments deposits = and of Net Public x Raffle Dec). significance ‐ ordinary Public Impact stratification : No Control x in Raffle 2 e Raffle Raffle in (Nov x x Account ffl No with test: a variable: ‐ indicate observations Var R ffl Raffle R F Table means x Account x of season of of Stars Dep period: nary Commitment, Ordinary, di value r ‐ Time Odi Dependent P Ordinary Commitment Commitment Number O Appendix Ordinary Mean Regressions Commitment equality planting Notes:

53 in on to effects. without [MK] from prior (6) 2,835 0.210 378.69 each fixed 398.97 831.83 524.17 ‐ 1,022.16 coefficient (883.06) (955.61) 10678.42 ‐ expenditure (1,062.25) (1,042.70) (1,006.45) (1,038.19) days 2,260.29** survey cell the 30 variables if groups Total ‐ test and stratification treatment (output independent [MK] (5) 2,835 0.219 857.43 with ‐ 9,902.10 7,219.16 2,230.97 include profit 95209.65 ‐ 15,159.52 24,495.51* (14,078.16) (11,710.71) (14,733.64) (13,212.61) (13,772.67) (15,148.96) estimation ordinary input) and Farm (SUR) regressions regressions Harvest sold) All output across not commitment Regressions 2010 145. (4) crop in 2,835 0.180 [MK] 1,002.69 3,992.97 and MK 21,033.26 of ‐ 11,251.43 18,494.18) 17,862.93) 15,223.08 17,982.76) 19,841.45) 155684.93 ‐ ( ( ( (17,745.21) (15,827.62) ( 41,143.38** after ca. is means (sold Unrelated Value 1 Commitment of on USD crop Expenditure equality Seemingly level. [MK] on the from coefficient (3) club 2,835 0.095 0.615 0.563 0.385 2,431.04 4,879.92 22,797.51 91746.78 11,076.62 15,407.48 the tests ‐ (16,227.86) (14,518.41) (13,661.24) (14,311.60) (12,993.12) (16,557.75) the sales 28,808.13** based at Household are Proceeds equals Raffle" and No 6: clustered inputs Ordinary controls are and Season of 4 on 6: Ordinary, (2) 2, = 2,835 0.394 [MK] 2010 errors 5,402.06 6,391.76 7,451.19 ‐ 4,771.61 60371.80 ‐ ‐ ‐ and 6: 10,922.89 1, (7,125.71) (7,647.77) (7,551.69) (6,659.53) (7,049.34) (8,679.48) value 18,560.76** baseline 4 Raffle 2, 2009 and coefficient "Ordinary=Commitment" Total 4 in No 1, Standard 2, the columns and if 1, of additional levels. and columns [acres] Outcomes of under (***) zero (1) 0.07 0.29 0.11 0.10 0.25 4.28 columns ‐ ‐ 0.145 2,835 "Commitment, (0.22) (0.23) (0.22) (0.24) (0.23) (0.23) without 0.46** "Ordinary=0" 1% of regressions from Land but and tests: ‐ cultivation F B. Agricultural (**), across regressions different effects on 5% regressions (*), across fixed "Commitment=0", 0 Appendix of 10% across cell see significantly at 0 Treatments tests = ‐ F of jointly Public is x Raffle definitions, Commitment= Ordinary=Commitment Ordinary= significance Public 7. stratification No Impact x Control of of of Raffle : Raffle Raffle 3 treatments. in and Account x x No with test test test: test variable (Ordinary) 5 variable: ‐ ‐ ‐ ‐ observations indicate Raffle Raffle Var F F F F 3, raffle Account x x Table of 1, of of of of Dep Stars complete Ordinary, Commitment, value value value value ‐ ‐ ‐ ‐ Dependent Regressions Commitment Commitment Ordinary P Ordinary Mean P Commitment Number P Ordinary P Apendix Commitment For additional columns Notes:

54 fixed ** deposit each cell (6) 0.02 0.02 0.01 ‐ 2,835 0.039 0.292 (0.02) (0.03) (0.02) (0.02) (0.02) (0.02) 004 0.04 0.06** 0.03** ‐ account fixed groups Has stratification treatment include amount ordinary (5) loan 2,835 0.958 4 824 74 4,824.74 3,834.78 4,558.31 5,158.28 4,209.79 40147.26 ‐ (7,207.73) (6,915.37) (7,003.51) (6,785.92) (7,535.90) (5,923.72) 11,806.25* and ‐ regressions All Tobacco 145. commitment MK in ca. [MK] transfers is (4) 2,835 0.830 47.37 1 370.38 370 38 156.38 means 889.61* ‐ 417.893 778.80* net (390.29) (459.38) (418.47) (467.84) (438.87) (419.60) ‐ 875.05** of made USD Total level. equality club the the [MK] tests at (3) 5.04 transfers 33.77 33 77 ‐ 2,835 0.340 176.91 459.24 193.97 ‐ ‐ ‐ ‐ 429.88* (301.94) (241.61) (320.99) (255.57) (320.88) (283.53) 2492.134 Raffle" received Total clustered No controls are deposit Ordinary, errors made fixed = baseline and (2) 84.06 2,835 Raffle 0.631 [MK] 460.58 460 58 370.87 ‐ 120.40 470.16 ‐ ‐ 650.96* (311.28) (416.42) (344.48) (426.51) (334.95) (335.84) Standard 2871.702 transfers No additional transfers levels. Total size, (***) without 1% size "Commitment, but and (1) 0.04 0.01 0.00 023 0.23 tests: ‐ ‐ 2,835 0.849 5.717 (0.18) (0.17) (0.19) (0.16) (0.19) (0.19) 0.45** ‐ 0.34** ‐ household F (**), B. effects on 5% Household (*), fixed Appendix 10% cell at treatments see = of Public treatments. x Raffle significance Public Impact definitions, stratification No : x Control raffle Raffle 4 Raffle Raffle in Account x x No with test: variable: ‐ indicate observations Raffle Raffle Raffle Var variable F Table x Account x x of of additional For Stars Dep Ordinary, Commitment, value ‐ Regressions Dependent Commitment P Ordinary Ordinary Appendix Ordinary Mean Number Commitment Commitment Ordinary without effects. Notes:

Related documents