Policy Research Working Paper 9248 Locking Crops to Unlock Investment Experimental Evidence on Warrantage in Burkina Faso Clara Delavallade Susan Godlonton Africa Region Gender Innovation Lab May 2020 Policy Research Working Paper 9248 Abstract Financial market imperfections remain pervasive in is high (94 percent), while credit take-up is moderate (38 developing countries, constraining potentially profitable percent). Households with access to warrantage primarily investment decisions, especially for rural smallholder farm- store sorghum and maize and sell their production over an ers. Warrantage is an innovative model of rural finance extended period of time, earning higher average prices and with the potential to overcome credit, storage, and com- resulting in higher sales revenue ($248, or 33 percent, on mitment constraints through a localized inventory credit average). Increased incomes are spent on long-term invest- scheme. Exploiting random variations in household access ments, including human capital expenditures (education), to warrantage and intensity of access across villages, this livestock purchases, and investment in agricultural inputs paper studies the direct impact of this scheme on bene- for the subsequent year. ficiaries as well as its spillover effects. Take-up of storage This paper is a product of the Gender Innovation Lab, Africa Region. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at cdelavallade@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Locking Crops to Unlock Investment: Experimental Evidence on Warrantage in Burkina Faso Clara Delavallade and Susan Godlonton1 Keywords: Rural finance, inventory credit, storage, Burkina Faso JEL Classification: O13, O16, Q12 1 Corresponding author: Susan Godlonton (Email: Susan.Godlonton@williams.edu), Assistant Professor at Williams College. Clara Delavallade (Email: cdelavallade@worldbank.org), Economist at the World Bank Africa Gender Innovation Lab. This project was funded by the Wellspring Foundation and received support from the CGIAR Research Program on Policies, Institutions and Markets, as well as Williams College. Alexia Pretari and Leila Bugha- Nee provided exceptional project management oversight and assistance with data-related activities. We acknowledge the extensive contributions of the field team led by IPA Francophone West Africa, Alassane Koulibaly and Kone Soumaila, and the data management and programming support provided by Michael Murphy. This study would not have been possible without the close collaboration of Felicite Kambou and the COPSA-C staff nor without the commitment of Gauthier Malnoury and the CISV. We also thank several individuals who made contributions during various stages of the project, including Sonia Cheung, Naila El Haouari, Samuel Lewis, and Michael Navarrete. We thank seminar participants at the University of Cape Town, and at the 4th IPA Financial Inclusion and Social Protection researcher gathering as well Jessica Goldberg and Lauren Falcao Bergquist for providing valuable feedback. 1. Introduction Smallholder farmers in developing countries are vulnerable to volatile crop prices. Credit market imperfections, inadequate storage, and behavioral challenges all inhibit smallholders’ ability to save their crops and sell them later in the season when prices are higher, resulting in seasonal liquidity constraints that drive suboptimal investment decisions. Academics and policy makers alike have long emphasized financial inclusion as key to smoothing investment and consumption and reducing poverty. Yet the recent literature on microfinance finds that realized impacts are often more limited than expected (Banerjee et al., 2015; Karlan and Morduch, 2009). An emergent literature instead considers modalities of rural finance with complementary features that may enhance the efficacy of credit by jointly addressing smallholders’ credit constraints and behavioral constraints.2 In this paper, we study one such innovation: warrantage, an inventory credit system that offers farmers the opportunity to store their crop production and access credit simultaneously. Crop storage, primarily of grain, acts as a commitment device, restricting farmers from accessing their stored grains for a fixed duration. Storage also serves to realize profits from seasonal price increases (MkNelly and Kevane, 2002) and provides collateral for farmers who wish to access credit. We examine the demand for and impact of warrantage when introduced in new villages in the South-West of Burkina Faso by the Cooperative de Prestation de Services Agricoles Coobsa (COPSA-C). For this, we exploit the random variation of access to warrantage among interested households, as well as intensity of household access across villages. This two-level randomization results in 67% of interested households being offered access to warrantage (referred to as high 2 Such constraints include social pressures within the community (Jakiela and Ozier, 2016; Goldberg, 2017), mistrust of financial institutions (Casaburi et al., 2013), limited bandwidth (Shafir and Mullainathan, 2013), and present-biased time preferences (Duflo et al., 2011; Frederick et al., 2002). 2 exposure villages), and 33% of interested households being offered access to warrantage (referred to as low exposure villages). Baseline and follow-up data were collected from 528 households, who participated in the lottery to gain access to warrantage (hereafter referred to as lottery participant households). We also collected data from 183 households who did not partake in the household randomization (hereafter referred to as non-participant households). Using the household-level randomization, we find that treated households choose primarily to store sorghum and maize grains through the warrantage storage mechanism. These crops are common in the area and exhibit significant historical price volatility. Take-up of the storage element of warrantage was close to universal and enabled these households to shift from home storage to storage in village warehouses. Credit utilization, on the other hand, was far lower, with only 38 percent of treated households taking up the credit offer. Through the increased storage of key grains, we find suggestive evidence of fewer sales immediately after harvest and more sales at the time of destockage, when prices are higher. Higher sales prices translate into higher crop revenues among treated households. With more cash on hand, treated households invest more in human capital, specifically education expenditure. We find no impact on food expenditure and no significant impact on a series of food security indicators, although the general pattern of results is positive. Alongside the educational investments, treated households increased their livestock holdings and increased input expenditures for the subsequent agricultural season. To examine the spillover effects of the introduction of warrantage, we compared non- treated households in low exposure villages to those in high exposure villages. Spillover effects attributable to greater village-level exposure to warrantage are limited. Also, in contrast to anecdotal claims, access to warrantage does not meaningfully affect village-level relations or sharing norms within the community. We conduct a simple cost-effectiveness calculation which 3 reveals that the warrantage program, implemented using existing storage infrastructure, is highly cost-effective, showing gains in production value approximately nine times higher than program costs. Although a large literature exists on a myriad of microfinance innovations, few studies examine products that combine storage and credit access.3 Two key exceptions include Bergquist et al. (2018), who examine a similar context in Kenya, and Casaburi et al. (2013), who examine a product targeted specifically to palm oils in Sierra Leone. In contrast to our setting, Bergquist et al. (2018) focus exclusively on maize and rely on home storage and monitoring in lieu of third- party storage. In this setting, through experimentally varied density in credit offers, they find significant impacts of credit on farmer profitability, operating through reduced volatility of market prices. Casaburi et al. (2013), on the other hand, find very different results. Their study is more comparable to ours in that storage is held by a third party as explicit backing for credit access; however, their approach differs in that storage is restricted to palm oils. They find limited uptake of the program (approximately 30 percent) and no positive impacts on farmers’ livelihoods. They speculate that this may be attributable to mistrust between farmers and the financial institution, as well as to the existence of adequate storage prior to the intervention. While we observe similar credit take-up rates, storage is used almost universally by treated households. Our study does not restrict the crop designated for storage, and our farmers exhibit far higher baseline trust in warrantage-related institutions than the farmers studied in Sierra Leone; both factors may in part explain why we observed impacts on profitability while they do not. Our results help reconcile the findings in these papers and further show that, at least in the Burkina Faso context, these profit 3 Coulter and Shepherd (1995) and Coulter (2012) provide an overview and in-depth case studies on the existing implementation of various modalities of inventory credit across Africa. 4 gains are invested productively: in education, livestock, and inputs for the subsequent agricultural season. In addition, we documented limited spillover impacts on both control households and non- participants, specifically addressing concerns raised by local policy makers that warrantage may undermine village relations and sharing norms. Our findings also expand our understanding of how savings constraints limit the ability of smallholder farmers to take advantage of arbitrage opportunities in commodity markets.4 To date, two key papers find evidence in favor of binding savings constraints in the context of agriculture. In rural Timor-Leste, Basu and Wong (2015) find that the offer of storage drums and sacks improves consumption and income indices. In Busia, Kenya, Aggarwal et al. (2018) find that offering ROSCAs a communal storage mechanism that included the provision of PICS bags and a subsidized stand to elevate bags off the ground increased storage, sales, and sales prices. They also tested an individual storage saving mechanism using a labeled savings account for inputs and found no impact on increased input usage. These results are consistent with findings in the health literature (e.g. Dupas and Robinson, 2013) that have highlighted the important role that social commitments play in saving collectively. Our experiment complements the findings in the Aggarwal et al. (2018) study in several ways. We find that a household savings mechanism coupled with a credit offer can also increase storage and sales and result in productive investments and input adoption in the subsequent season. Importantly, the storage restricts farmers’ access to stored crops, which may function in a similar way as the social commitment aspect of the group savings mechanism in Aggarwal et al. (2018). 4 Much of the previous literature focused on understanding the role of credit and liquidity constraints on intertemporal arbitrage opportunities (e.g. Stephens and Barrett, 2011; Dillon, 2017). However, recent work by Karlan et al. (2014) shows that offers of insurance rather than credit are more impactful in Northern Ghana, and Delavallade et al. (2015) show that offers of insurance rather than savings are more impactful in Senegal and Burkina Faso. Both studies imply that risk is more important than credit. 5 The paper proceeds as follows. In Section 2, we describe the experimental design and provide relevant background information. Section 3 describes the data and empirical approach undertaken, while Sections 4 and 5 present the take-up and impact results, respectively. We discuss scale-up and cost-effectiveness considerations in Sections 6 and 7, while Section 8 concludes. 2. Experimental Design 2.1 Warrantage communautaire (Smallholder farmer inventory credit) In Burkina Faso, warrantage communautaire, or smallholder farmer inventory credit, is a particular form of warehouse receipt finance jointly managed by farmers’ organizations and financial institutions. Farmers deposit their grains at harvest time in a (small) warehouse, typically located in their own village, as collateral for a loan of up to 80 percent of the value of the grains stored for a pre-determined storage period, at an annual interest rate of 9.75 percent. This system provides no tradable receipts. After the storage period, the producer may collect the grain, repay the loan, and sell the grain at will; alternatively, a buyer may collect the grain and directly repay the financial institution, a representative of which would be present at the warehouse (FAO, 2012; Ghione et al., 2013; Malnoury, 2011; Pala, 2012; Cortès and Carrasco, 2012). Inventory credit was first introduced to Burkina Faso in 2005; by 2013, approximately 6,000 smallholder farmers had accessed inventory credit, storing 4,000 tons of crops and taking out around 300 million FCFA in loans from local microfinance institutions. Warrantage has the potential to overcome many constraints faced by smallholder farmers. Collateral provided through the storage of grains allows financial institutions to provide credit to rural farmers who typically are unbanked and unable to access credit. Storage at home, particularly of grains, can increase the chance of post-harvest losses due to pests (although post-harvest losses 6 at the farm level appear to be lower than initially thought; see Kaminski and Christaensen, 2014; Ambler et al., 2018). By enabling farmers to store grain in an appropriately constructed and ventilated facility, warrantage may further reduce such losses. Warrantage also acts as a commitment savings device: farmers can only access grains on a pre-determined date rather than at will. Thus, households can address potential self-control problems by tying their hands through storage.5 Also, the storage feature of warrantage may increase a household’s ability to withstand social pressures to share their agricultural output during the harvest season. Individuals face considerable pressure to share in villages (for example, see Jakiela and Ozier, 2016). As it is practiced currently, warrantage could potentially increase this burden by shifting the timing of grain on hand from the harvest to the lean season; households may then face increased pressure to share grains after storage. This may offer greater protection to the community but may result in negative impacts on warrantage participants. Finally, and arguably most importantly, farmers can use warrantage to take advantage of arbitrage opportunities by shifting the timing of their sales to a period in which prices are high (See Figure 1) but still maintaining cash on hand through the credit provision portion. Figure 1. Price and sales seasonality in the west Burkina Faso (sample of 70 villages) Source: FARMAF project (CIRAD) 7 2.2 Experiment and Project Timeline COPSA-C, one of the main farmers’ organizations facilitating community-based warrantage in Burkina Faso, led the study intervention in coordination with Coris Bank (a financial institution with locations spread across all major towns of Burkina Faso) and with technical assistance provided by Comunità Impegno Servizio Volontariato (CISV). At the time of implementation, COPSA-C facilitated 29 village warehouses in southwest Burkina Faso. The cooperative conducted sensitization activities, negotiated loan terms with Coris Bank on behalf of the beneficiaries and coordinated storage and de-storage activities. Figure 2 provides a timeline of activities. Figure 2: Timeline of intervention and data collection activities Our sample includes 38 villages in close proximity to one of 13 village warehouses, all of which had experienced consistent under-utilization prior to the intervention. Targeted villages had no previous access to warrantage. We conducted a census in September 2014, during which time we administered a basic household roster to every household, gathering information on demographic characteristics, household size, and warrantage interest. The number of households per village varies substantially in this area, ranging from 14 to 208 households per village (median: 86; mode: 50). Using the census data, we randomly selected up to 35 households that had declared an interest in warrantage and up to 15 households that had no interest in warrantage. Due to the variability in the number of total households in a village, this sampling approach yields approximately 40 households per village. The proportion of interested households (as measured at baseline) varies in the villages, from 45 to 100 percent of all households within a village. 5 Commitment devices have been shown to be effective in triggering a wide range of behavior changes (for a full review, see Bryan et al., 2010). 8 Before the commencement of the warrantage program, we conducted a census and a baseline survey in September and November 2014, respectively. These surveys provided extensive information on socio-demographic characteristics, consumption, food security and diversity, agricultural investments and production, access to credit, and savings. Concurrently, COPSA-C implemented several sensitization activities with CISV support. Immediately after harvest in December 2014 and early January 2015, we implemented public lotteries that operationalized the randomization. Specifically, we used a two-stage randomization process, with the first stage conducted at the village level and the second at the household level. Villages were randomly allocated to either the high exposure or the low exposure treatment condition, in which the probability of access to warrantage was 67 percent or 33 percent, respectively, for interested households. Village-level randomization was stratified on village storehouses (i.e. multiple villages store in the same nearby village storehouse), and randomization was implemented in Stata. Thereafter, we implemented the household-level randomization. In each village, we conducted a public lottery in which each interested household drew a ball from a bag to determine their treatment status.6 In high exposure villages, the bag contained 67 percent of winning (green) balls, while low exposure villages had only 33 percent of winning balls. Crops were stored after the lottery until May 2015, when bags were redistributed to the farmers. At the time of both storage and redistribution, we collected administrative data documenting the types of crops stored, as well as the number of bags owned by each farmer. Finally, we conducted an endline survey in August 2015, several months into the subsequent agricultural growing season. 3. Data and Empirical Strategy 3.1 Data Sources and Description 9 We utilized two main sources of data: COPSA-C administrative data regarding storage and related loan information (take-up, loan amount, and repayment) and independent survey data collected at several stages of implementation (Census, Baseline, Stockage, Destockage, and Endline). Table 1 provides summary information of the sample. To describe the profile of the villages in which the intervention took place, we show average demographics from the census in column 1. Most households are male-headed. About one-quarter of households are polygamous; thus, households tend to be relatively large, with an average of eight household members. Knowledge of warrantage was high at 81 percent; this is probably a result of the concurrent sensitization activities run by COPSA-C. Columns 2 and 3 present means of selected census and baseline variables for lottery non-participants and lottery participants respectively. “Non-participants” refers to households in the village that were not interested in warrantage and thus chose not to participate in the lottery; participants, on the other hand, are those households that did participate in the lottery (i.e. both treated and control households). Comparing columns 2 and 3 thus sheds light on selection into participation in the warrantage lottery. Column 4 shows the p-value for the pairwise tests of equality between non-participants and participants. Lottery participants tend to come from larger households with slightly younger household heads and are more likely to be familiar with warrantage. Participant households also reported higher hypothetical interest in storing key grains (sorghum and maize) using warrantage. Despite these differences, using an omnibus F-test, we find that the set of variables available at the census presented in Table 1 do not jointly predict participation in the lottery (p-value: 0.152). 6 Consent to take part in the lottery was then obtained. The village head held an introductory speech emphasizing the importance of complying with the protocol (including forbidding storing other household members’ bags). Household representatives were asked to declare the number of bags they would be willing to store if allocated storage space; this number was capped at 10. Participants’ names were randomly picked from the name bag to vary the order in which household representatives participated in the lottery. The selected participant was then invited to blindly draw a ball from a basket containing green (win) and blue (lose) balls (the warrantage bag). 10 Previous work has documented that trust in managing institutions can be a constraint to adoption (for example, see Casaburi et al., 2013). To examine these findings, we used several proxies to measure trust and confidence in warrantage-related institutions (neighboring villagers, banks, NGOs, and MFIs). Storage occurs in a pre-identified neighboring village storehouse; therefore, we asked participants about the frequency of their interactions with neighboring villagers in the village in which their warehouse is located, as well as whether they would be confident storing in the neighboring village and whether they think a neighboring villager would return a lost wallet. Across measures, there is little difference across participant and non-participant households. Households also rated on a scale from one (representing complete confidence) to five (representing 11 no confidence) various warrantage-related entities, including fellow villagers, banks, relevant NGOs and MFIs. Using these four indicators, we constructed an index to measure confidence in warrantage-related institutions. Participant households have a lower trust index compared to non- participant households; however, the differences are not statistically significant. Overall, the limited differences between participant and non-participant households imply that there are non- observable reasons influencing a households’ decision to participate in the lottery. 3.2 Balance Table 1 further shows that the household randomization was reasonably successful in generating comparable groups. Columns 5 through 7 are restricted to the sample of lottery participants. Columns 5 and 6 present the means of selected baseline and census variables for the Control and Treatment households, respectively, while column 7 presents the p-values for tests of equality between the treatment and control averages. We observed few statistically significant differences. Lottery winners are slightly larger in household size, less likely to trust members of the neighboring village to return a lost wallet (p-value of 0.1), and interested in storing a larger number of bags under warrantage. We also found no evidence that the set of census variables (p- value=0.548) or the combination of the census and baseline variables (p-value=0.933) jointly predict household treatment status. Finally, restricting the sample to all non-treated households, thus including both control households and lottery non-participant households, we compare household characteristics of those resident in Low Access and High Access treatment villages. These results are presented in columns 8 through 10. We similarly found limited differences in village-level characteristics between the High Access and Low Access treatment groups. 12 3.3 Estimation Approach We use two core sets of analyses to estimate the impacts of warrantage. First, we set out to quantify the effects of receiving access to warrantage. Second, we estimated the spillover impacts of warrantage by examining impacts on non-treated households by comparing non-participant households in low and high exposure villages. To begin, we quantified the effects of warrantage among households that expressed interest in participating. To do so, we estimated the following model: ℎ1 = + 1ℎ + ℎ0 + ℎ0 + + ℎ (1) where h indexes households and v indexes villages. The sample was restricted to all lottery participants. ℎ is a binary indicator equal to one if the household won the lottery and thus received access to warrantage. Yh1 is the outcome measure as measured during the endline survey. Yh0 is the baseline value of the outcome measure, included whenever available. Continuous variables are presented in FCFA, winsorized at the 99th percentile.7 Xh0 is a vector of household controls, all measured at baseline, including indicator variables for whether the household head is married or is a polygamist, as well as the number of total spouses, household size, the value of warrantage crops, amount of land owned, and an indicator variable for whether the household head was familiar with warrantage.8 Due to village-level lotteries, we include village fixed effects, denoted by . We clustered standard errors by village to account for within-village correlation of unobserved variables. Here, 1 measures the impact of receiving access to warrantage among 7 Results are qualitatively similar when using alternative specifications, such as using log transformations of continuous variables. 8 Note that all results are qualitatively similar when we exclude covariates from the analysis. 13 households that expressed interest in participating in warrantage at the time of the village lottery. We present only the intention to treat estimates. Second, we examined the spillover impact of warrantage access in a village for non-treated households (that is, households that did not express interest in participating in warrantage at the time of the lottery and the control households). Here we estimated: ℎ1 = + 5ℎ + ℎ0 + ℎ0 + ℎ (3) where ℎ is a binary indicator for whether the village was assigned to the high exposure treatment condition. Variation in village size and in village interest in warrantage participation imply that the village-level randomization does not result in precisely one-third and two-thirds of the village receiving access. Figure 3 plots the cumulative distribution functions (CDF) of the proportion of households, based on the census, treated for villages assigned to low exposure (solid line) and high exposure (dotted line), respectively. The CDF for the high exposure treatment condition clearly shifted to the right; that is, in high exposure villages, a larger fraction of the village obtained access to warrantage. The figure also clearly demonstrates variation in the fraction of the village treated: 40 percent less of all households in a village are treated in 20 percent of the high exposure villages. Figure 3: Cumulative distribution functions (by village level treatment) of share of village treated 1 .8 .6 .4 14 Our main tables present our output in a similar manner across outcomes. In each of our main tables, where appropriate, Panels A and B present results from equations (1) and (2), respectively. 4. Take-Up Results In this experiment, treated households were offered both features of warrantage: the credit feature and the storage feature with fairly strong commitments attached to the savings component (namely locking crops for about four months). To better understand to what extent farmers’ investment decisions are constrained by credit market imperfections and commitment issues (to self and others), we begin by examining warrantage take-up among lottery winners and investigate both the storage and credit features of warrantage. 4.1 Storage Utilization Overall awareness of warrantage is relatively high at baseline, with 81 percent of all households claiming they have heard of warrantage. Yet, 49 percent of the sample incorrectly believed that warrantage offered interest-free credit. Knowledge of transportation fees and the storage mechanism were better understood by households at 81 percent and 90 percent, respectively, among those claiming to have heard about warrantage previously. To understand the relative importance of warrantage activities, we asked respondents to rate different features of warrantage, including the secure location for cereal storage, the possibility of higher prices for stored cereals, and the possibility for more food in the lean season. The least important feature identified by farmers was the ability to restrict grains to friends and families, with 20 percent of the sample stating this was not a feature of warrantage and another 11 percent stating it was not at all an important feature. Table 2 reports summary statistics of warrantage-related take-up activities. Columns 1 and 15 2 present the means of key storage-related variables among control and treated households, while column 3 presents the p-value corresponding to the test of equality of means. Encouragingly, all households participating in the lottery intended to stock approximately three bags if they were to win the lottery. At the time of storage, however, treated households stored on average 3 bags while control households stored 0.065. Ninety-four percent of treated households chose to stock at least one bag, whereas only 1.1 percent of control households stored at least one bag using warrantage. The two key crops stored were sorghum and maize. Almost one-third of households stored sorghum and 60 percent stored maize, storing on average 77 and 178 kilograms of sorghum and maize, respectively. Finally, it is important to understand whether there are meaningful differences in observed warrantage behavior in the lottery sample across high exposure and low exposure villages. Table 2, columns 4 and 5 present the means of key storage variables for participating households in high exposure and low exposure villages, respectively. Column 6 presents the p-value corresponding to the test of equality of the group means. In general, the pattern of results shows that the village treatment is meaningful, that is, take-up and usage of warrantage is greater in high exposure villages as compared to low exposure villages.9 16 4.2 Credit Take-up Interestingly, while most households stored crops, only 38 percent of households took up the option to use credit backed by their grain collateral (Table 2). Evidently, farmers do not view storage through warrantage simply as a means to gain access to credit but many do so as a preferred storage mechanism for some of their grain production. Among households utilizing the credit option, the average loan size was 23,725 FCFA, approximately US$40 at the time. The credit repayment rate was universal among lottery winners at the time of destockage. 17 A key difference across village-treatment conditions is that households in low exposure villages are also more likely to take up the credit option (conditional on winning the lottery) and take out larger loans on average. Due to the sizeable difference in the utilization of the credit feature, the total overall amount of credit taken up in high exposure and low exposure villages is approximately the same. To explore the sources of variation in credit take-up among warrantage participants, we examine the determinants of credit take up among the treated households (lottery winners). Table 3 presents these results.10 Credit take-up is significantly higher among households with more educated household heads: Household heads with a complete secondary education are between 39 and 58 percentage points more likely to opt for the credit feature. Households more familiar with warrantage and with the expressed intention to store more bags immediately prior to the lottery also correlate significantly and positively with credit take-up. Credit take-up is also suggestively higher among households with a credit history, but this result is sensitive to the inclusion of village fixed effects (column 2). Finally, and contrary to our expectations, those displaying more trust in neighboring village members and receiving transfers from individuals within the past year are less likely to take up the credit offer, perhaps suggesting that those relying on the storage feature only are more confident in their ability to rely on informal channels to meet their financial needs. 9 In results not presented, these household-level differences, within the lottery sample, translate into differences in village-level aggregates across high and low exposure villages. 18 19 In sum, we observed strong compliance with the lottery and significant take-up of the storage aspect of warrantage, but more moderate use of the credit feature, particularly among households in the high exposure treatment villages and among households unfamiliar with warrantage and with lower completed years of schooling. At the aggregate village level, villages in the high exposure treatment condition store more relative to villages in the low exposure condition, but the total credit issued remains approximately the same. 5. Impact Results 5.1 Impact on Storage and Credit We begin our analysis of the impacts of the intervention by examining whether treatment induced changes in overall storage behavior and credit among treated households. To this end, we estimate whether access to warrantage increased storage activities at the extensive margin (Table 4, Panel A). Warrantage crops include all crops of which at least one household stored any bags during warrantage. The full list includes sorghum, maize, millet, groundnuts, rice, and beans. Among control households, 95.7% stored at least one of these crops, thus leaving little room for warrantage to modify storage behavior on the extensive margin. Unsurprisingly, we found no evidence of changes to household storage behavior on this margin (column 1). Inspecting household storage behaviors further, households substitute away from home storage into storage at storehouses for warrantage crops. 10 Table 3 presents selected characteristics from the regression. Additional controls not presented are indicators for whether the household head is married, the age of the household head, and the share of sorghum and maize production consumed. 20 On the credit side, we found no evidence that warrantage increased access to credit among treated households, as measured by the total credit taken out in the past year or by current debt (Table 4, columns 6 through 8). These results suggest that the credit utilization was insufficient to impact (on average) overall credit utilization in the villages of interest. 5.2 Direct Benefits of Warrantage Next, we examined whether access to warrantage changed household behavior with respect to how output was managed. That is, did warrantage impact the share of own production consumed, sold and given away to others? These results are presented in Table 5. Treated households consume 4.4 percentage points (6 percent) less and sell 3.6 percentage points (17 percent) of their warrantage crop output combined (columns 1 and 2). To address the concern held by policy makers that treated households will share less with their neighbors, we investigate the treatment impact on the proportion of warrantage crop output shared; the coefficient is positive but insignificant and very small (column 3). Our results demonstrate that any increased demands on treated households in the lean season are negligible. 21 The finding that treated households sell more of their own production of key warrantage crops is interesting and suggests that warrantage provided households with a self- control mechanism reducing consumption to increase sales. However, due to the fact that treated households have restricted access to their stored grains, we further investigate whether warrantage impacted other farmers’ sales behavior; namely the timing of and price received for their crop sales. The vast majority of crops stored were sorghum and maize and thus we limit the sales and price analysis to these two crops. Figures 4a and 4b demonstrate the treatment effects from a series of regressions, using equation (1), estimating whether the household sold any of its crop production in a particular month for sorghum and maize, respectively. Treated households are more likely to sell later on in the season and somewhat less likely to sell immediately post- harvest in January and February (although the coefficients are noisy).11 Further, treated households earn a higher price on average. 22 Using a standardized price index to combine both sorghum and maize prices, we find that treated households receive prices 0.15 standard deviation higher than control households. Figure 5 plots the distribution of this standardized price index separately by household-level treatment assignment. The distribution of prices for treated households is clearly shifted to the right of the .2 corresponding distribution for control households. Using a Kolmogorov- Smirnov test, we found 0 that we could reject the null that the empirical distributions come from the same underlying distribution (p-value: 0.08). Figure 4: Impact of warrantage on sales seasonality of Sorghum and Maize .1 .1 .05 .05 0 0 -.05 -.05 -.1 Treated Treated Sep Oct Sep Oct Nov Dec Nov Dec Jan Feb Jan Feb Mar Apr Mar Apr May Jun May Jun Jul Aug Jul Aug a. Sorghum b. Maize Note: The outcome variable is defined as equal to one if the household conducted any sales of a) Sorghum b) Maize in that particular month. The graph plots the estimated treatment coefficients from regression (1). 11 We do see an uptick in sales beginning in the month of April, prior to the redistribution activities. Households may be selling off more of the grains they have in home storage at this time; alternatively, this result may indicate recall error. We do not measure sales every month and instead rely on a retrospective sales history. Retrospective agricultural data have been shown to be heavily influenced by current agricultural experiences (see for example, Godlonton et al., 2018); a similar issue with recall bias may influence these timing results. 23 Figure 5: Distribution of average standardized prices for key warrantage crops .8 .8 .6 .6 Density Density .4 .4 .2 .2 0 0 a. by household treatment status b. by village treatment status (lottery sample) (non-beneficiary households). This distributional shift in prices, together with the earlier finding that access to warrantage increases the amount of the key crops sold (Table 5), translate into treated households earning FCFA 147,188 ($248) more on average from their crop production (Table 6, Column 1). This is measured by the total value of agricultural output of all warrantage crops, defined as all crops in which at least one household stored the crop in a warrantage village warehouse.12 The realization of the lottery occurred after harvest; thus we cannot attribute these impacts to input adoption or other farm management decisions made during the agricultural season. The magnitude of the difference is large, albeit with large standard errors. Taken at face value, the coefficient implies a 33-percent increase in the total value of crop production. Warrantage crops account for about half of the total value of agricultural output of these farmers; thus, the increased revenue from crop sales resulted in meaningful increases to cash-on-hand for the treated households. With so much additional cash-on-hand, what did households do? To answer this question, we examined household expenditures, specifically whether the increase in crop production income translates into any improvement in overall welfare, as measured by a standard consumption module and a range of food security metrics (see Table 6). To measure consumption, we implemented the LSMS food consumption module, which includes a seven-day recall period for food expenditures. We combined this with a non-food expenditure module covering a 30-day recall period. Total expenditure per capita increased by 13 percent, primarily driven by increases to non-food expenditures. Households comprise 9.3 members on average thus the net impact on total 12 While the vast majority store either sorghum or maize or sorghum and maize, other crops stored include rice, groundnuts, millet, and assorted beans. This is consistent with warrantage in the region (Bambara et al., 2008). expenditures of 143,050 FCFA (15,262*9.373) is roughly equivalent to the treatment impact on the value of the warrantage crops. We further disaggregated non-food expenditures in Table 7 to examine whether and to what extent different non-food expenditure categories are affected. Treated households exhibit higher non-food expenditures, primarily due to larger investments in education and additional expenditures on personal effects (e.g. clothing). Unlike non-food expenditures, we observe no impact on total food expenditure. Nonetheless, we also examine three standard indices common in the literature to verify this result: the food consumption score (FCS) developed by the World Food Programme, which measures both dietary diversity and food frequency; a food security score (FSS) also constructed according to WFP guidelines that takes values from -1 to -4 depending on the severity of food shortages; and the inverse of the WFP coping strategies index (CSI), referred to as the resilience index which weights the frequency and severity of actions taken by households to cope with food shortages (following Goldberg, 2017). Across outcomes we find no evidence and importantly no negative impact of warrantage on any of the food security indices. Food expenditures are however, measured once off, and do not capture food expenditures during the time of storage. Thus, while we see no net impact on food expenditures at the time of the endline, it is possible that during the storage period households experienced short-term food insecurity, we are unable to directly test this. Alongside the additional non-food expenditures, we also consider whether other savings behavior change among treated households (Table 8). Treated households are no more likely to have savings; but we do observe a sizeable and robust increase in livestock holdings, a popular informal savings mechanism in this area, among treated households (see, for example, Kazianga and Udry, 2006). We use tropical livestock units (TLSU), an aggregate measure that combines all animals into one summary measure assigning different weights to each animal type, as well as the total value of all livestock held.13 Both indicators suggest large increases to livestock holdings among treated households, which likely will lead to sustained impacts over the medium term. Interestingly, the size of these impacts are similar in magnitude to those in Ambler et al. (2019) which examined the effects of lumpy cash transfer payments of $100 coupled with a farm management plan intervention in Senegal. Finally, we are also able to examine whether treated households change their agricultural investments in the subsequent agricultural season. Table 9 presents these results. In addition to the educational and livestock investments, treated households spend more on inputs in the following year. Across the board, we see positive coefficients, although, we only note a statistically significant impact for seed expenditures. Unlike seed purchases, farmers would not have completed their purchases of chemical fertilizers or tools or paid labor costs for the agricultural season at the time of the endline survey.14 In sum, the results suggest that treated households store some of their sorghum and maize and delay the sale of some grains in order to take advantage of higher prices later on in the season. This led treated households to earn more for their warrantage crops relative to control households. With the additional revenue, households invest in education and their farms and increase their buffer savings through livestock purchases. 5.3 Spillover Impacts of Warrantage To measure the spillover effects of warrantage in the villages studied, we considered two non- beneficiary groups: households that were interested in participating in warrantage but were designated to the control group (control households) and households that indicated no interest in warrantage at the time of the lottery (non-participant households). We examine these groups jointly and broadly consider two sets of spillover effects: those driven by market-level forces, and behavioral adaptations. 13 We followed the literature to calculate a standardized measure of livestock and compute the tropical livestock units. We assigned different weights to different animals as follows: 0.8 for horses, 0.7 for cattle, 0.3 for pigs and donkeys, 0.1 for sheep and goats, and 0.01 for chickens, rabbits, guinea fowl, ducks, geese and pigeons. Results disaggregated by animal type are presented in Appendix Table 1. 14 In Appendix Table 2, we consider other income generating activities. We show that access to warrantage had no impact on such activities in the preceding 12 months. Sales prices received by non-beneficiary households from our endline survey suggest prices received for the two main warrantage crops were higher in high exposure villages. Figure 5b plots the distribution from which we observe a clear rightward shift of the average standardized price index in high exposure villages. Increased access to warrantage also lowered the variance in prices received by non-beneficiary households from 1.089 in Low Access communities to 0.831 in high exposure communities. These moderate gains do not however translate into discernible positive spillovers with respect to the sales value of warrantage crops in high exposure villages. The coefficient is positive, but noisy (Table 6, Panel B). In consultations with implementers and local policy makers, we repeatedly heard concerns that warrantage may undermine village relations, particularly with respect to sharing in times of food insecurity. To address this concern, we examine whether non-beneficiary households in high exposure villages changed how much of their warrantage crops were sold, consumed and/or given away. Table 5 Panel B provides suggestive evidence supporting these anecdotal concerns, as non- beneficiary households in high treatment communities do share more with their neighbors. The estimated effect size is small in magnitude, 3.8 percentage points, but is a meaningful change as it translates into these households sharing 177% more than those in low access villages. Despite the large treatment impact, the overall levels of sharing are very low, suggesting that the overall impact on the affected households is limited. To complement these findings, we also examine the three food security indices (Table 6), which suggests that the negative impacts of sharing more with others by non-treated households is negligible. We pursued this concern further and examined whether various proxies for village relations are affected. We used several proxies similar to those measuring confidence in warrantage- associated institutions; that is, whether the household had confidence storing crop production within their home village, and whether a village member would return a lost wallet. In addition, we conducted trust experiments to elicit an incentivized trust indicator. Within a sub-sample of villages, we implemented a simple variant of a dictator game with 400 FCFA. Figure 6a shows very similar offers were made across treated, control, and non-participant households; Figure 6b shows near identical distribution of offers in low and high exposure villages. Table 10 columns 1 through 4 corroborates these findings, showing no evidence that warrantage undermines nor improves village relations. Figure 6: Dictator game offers (trust game sample) a. by household type b. by village level randomization 6. Scaling Considerations: Targeting, Trust, and Village Relations Our results demonstrate benefits to treated households and negligible impacts on control and non- participant households in communities with greater access to warrantage. The next step is to determine whether warrantage communautaire is scalable and, if so, to understand why we have not seen more experimentation with rural finance of this nature. In this section, we attempt to shed light on these important questions. To effectively scale, organizations and/or governments need to be able to effectively identify interested participants in relatively localized areas in order to minimize costs of implementation. Throughout the project, we elicited interest in participating in warrantage at the time of the census, baseline, lottery, and endline. One goal was to assess whether we can predict warrantage participation and if so, at what point we are able to do so effectively. Overall, the level of expressed interest is consistently high. Yet, correlation in interest across the different stages is relatively low (correlation coefficient = 0.19); this is not terribly surprising, as inferring actual behavior using hypothetical questions is difficult. Using lottery participation (the only elicitation indicator that is incentivized), we examined the determinants of interest in warrantage in order to identify how to target sensitization efforts (Appendix Table 3). Familiarity with warrantage is an important determinant, demonstrating that sensitization activities prior to scale-up are important. Larger households with younger household heads and households with an unmarried household head are also more likely to participate. More generally, our results suggest that it is hard to predict participation based on household characteristics, implying that it is likely to be challenging to identify interested households to target sensitization efforts and reduce implementation costs. Casaburi et al. (2013) identify two additional challenges for scale, particularly with respect to low take-up: alternative and preferred home storage options and trust in the institutions involved. In that study, households were limited to storing palm oils exclusively and had appropriate (or at least preferred) storage mechanisms for their palm oil at home. Our results suggest that providing households with a choice of which crops to store can overcome this constraint, which is encouraging. Building trust, on the other hand is more complex, particularly given the wide range of institutions required to operate such an intervention effectively. Bergquist et al. (2018) mitigate both constraints noted in the Casaburi et al. study by allowing households to store crops at home with regular monitoring by field agents. They find that home storage, monitoring, and bank (or MFI) openness to home storage can work in cases in which these factors are feasible and cost- effective. Unfortunately, in many settings, such as the context of our study, regular monitoring would be too costly and financial institutions require that collateral be held with a third party. Thus, we considered whether participation in warrantage, and different levels of exposure of village participation in warrantage, can build trust over time. To do so, we focused on trust proxies specific to the various institutions involved in delivering warrantage: trust in neighboring villagers (specifically the ones the village where the storehouse for warrantage is located), banks, MFIs, and NGOs. In our case, storage of the grains in nearby villages constituted an important anecdotal consideration. While we could not test the extent to which this factor matters, we did see that even at baseline, confidence in storage in the neighboring village was high though not universal (over 70 percent). Using a wide range of trust proxies, we examined whether trust is likely to grow within villages in which warrantage is implemented (Table 10). We observe limited difference between treated and control households at endline, suggesting that participation in warrantage does not influence confidence in the institutions (Panel A). There is also no evidence that confidence in warrantage-associated institutions was impacted among control and non-participant households in high exposure villages (Panel B, columns 3 through 5). 7. Discussion on Cost-Effectiveness The large effects of warrantage on income and investment in various types of capital naturally raise the question of cost. In this section, we document the program’s costs and benefits using actual program costs from the implementing partner. The costs of the intervention are based on roll-out of the intervention as implemented: that is relying on existing infrastructure for village warehouses. The project distributed 821 bags to 251 farming households, at a cost of $3 per bag. Prior to the distribution of bags, the project team conducted a one-day sensitization campaign in each of the 38 villages under study. The campaign costs for each village encompass the daily cost of an NGO representative in charge of the sensitization ($10), as well as transportation costs ($5). The sensitization campaign costs thus totaled $570, i.e. $2.30 per household or $0.70 per bag stored. Each bag needed to be transported to and from the warrantage warehouse. The average cost of transportation was $2 per bag per trip; totaling $4 per bag, for storage and destorage. In addition, stored bags needed to be checked once a month to mitigate storage-related hazards such as mold growth, insect infestation, and rodent attacks. Storage monitoring costs roughly $90 per warehouse over six months, or $1.50 per bag. The overall cost of granting access to warrantage thus amounts to $9.20 per bag or $30.10 per household. We first measured benefits through gains in production value. The estimated program effects indicate a 33-percent increase in the total production value of warrantage crops for storing households, corresponding to a gain of FCFA147,188 or $277. Roughly speaking, the gains in production value are nine times higher than the program costs. These production value gains trickle down to three main sectors of household budget investment: education, livestock, and agricultural inputs. When we measure benefits in terms of investment in human capital, households storing crops spend an additional $10 on education in the 30 days preceding the follow-up survey, which took place in July and August. Assuming that (i) warrantage started translating into additional education expenses when stored crops were sold at destorage in April-May, (ii) the gains averaged $10 per month between then and the time of survey, and (iii) gains in education expenses were exhausted by the time of the follow-up survey, we found a lower bound for gains in human capital investment of $30 per household. This conservative calculation indicates that gains in education investment solely approximate the costs of the program. Using effectiveness estimates from the livestock value, we found even higher gains. We roughly estimated $351 in livestock value gains per household against $30.10 in costs, for a ratio of approximately 12/1. This is higher than the gains in production value and indicates that additional investments resulting from gains in production value yield positive returns within three months after destorage. Finally, assessing cost-effectiveness based on input purchases only, we found the gains to be slightly lower than the program costs ($23.60). We should note that we did not account for non-monetary costs, such as psychological costs arising from uncertainty related to storing crops in a neighboring village and those attributable to anxiety related to price volatility while crops are stored and individuals are unable to sell their stored grains. In addition, implementers designed the program explicitly to emulate how to use warehouses at full capacity. The program consisted of extending access to warrantage in existing warehouses for neighboring farmers who did not have prior access. Most importantly, the program costs therefore do not encompass the costs of introducing warrantage into new villages, which would involve building warehouses. Accounting for the building of warehouses would dramatically change the cost-effectiveness of such an intervention. 8. Conclusion This paper shows that smallholder inventory credit can have positive impacts on farmers’ investment behavior. Take-up of storage is close to universal. Treated households elect to store primarily sorghum and maize and sell their grain production over an extended period of time, earning higher average prices and resulting in higher sales revenue. Increased incomes are spent on long-term investments, including human capital expenditures (education), livestock purchases, and agricultural inputs in the subsequent year. Take-up of credit is more limited, suggesting that the changes are driven primarily by farmers’ storage behavior and that relaxing behavioral constraints may be at least as critical as relaxing credit constraints for improving farmers’ productivity and welfare. However, our experimental design does not allow for disentangling the extent to which the effects are driven by storage alone versus storage combined with the credit feature. Further, we found limited spillover effects of increased warrantage access within our sample among control households and non-participant households. In particular, we addressed anecdotal concerns voiced by potential stakeholders that warrantage may affect village relations, specifically as they pertain to village sharing norms. We found limited evidence to support these concerns. Warrantage prevalence did little to alter measures of institutional trust among either participating or non-participating households. Our results should be interpreted in the local context. We observed relatively high levels of baseline trust in the institutions involved in carrying out warrantage. This is most likely due to the strong relationships the organization facilitating warrantage has built with village chiefs and local villagers over the several years it has worked in the study villages. Without trust, take-up is likely to be low, as evidenced in other contexts (e.g. Casaburi et al., 2013). Further, the magnitude of the impacts may be affected by dampened crop price volatility due either to higher take-up of warrantage that results in price smoothing or to policy interference in grain markets. In this experiment, we did not attempt to unbundle mechanisms or to provide rigorous evidence regarding credit versus storage pathways. However, our findings suggest that the storage aspect of warrantage is important in explaining the results, highlighting the welfare potential of policies alleviating behavioral constraints to rural farmers’ storage and saving. Although these constraints include, but are not limited to, social pressure and self-control issues, we found no evidence of adverse consequences that could undermine sharing norms or trust within the community. References Aggarwal, S., Francis, E., and Robinson, J. (2018). Grain Today, Gain Tomorrow: Evidence from a Storage Experiment with ROSCAs in Kenya. Journal of Development Economics, 134: 1-15. Ambler, K., de Brauw, A., and Godlonton, S. (2019). Cash transfers and management advice for agriculture: Evidence from Senegal. World Bank Economic Review, forthcoming. Ambler, K., de Brauw, A., and Godlonton, S. (2018). Measuring Postharvest Losses at the Farm Level in Malawi. Australian Journal of Agriculture and Resource Economics, 62(1): 139-160. Bambara D., Zoundi J.S., and Tiendrebeogo J.-P. (2008). Association céréale/légumineuse et intégration agriculture-élevage en zone soudano-sahélienne. Cahiers Agricultures, 17(3): 297-301. Banerjee, A. V., Duflo, E., Glennerster, R., and Kinnan C. (2015). The miracle of microfinance? Evidence from a randomized evaluation. American Economic Journal: Applied Economics, 7(1): 22-53. Basu, K. and Wong, M. (2015). Evaluating seasonal food storage and credit programs in East Indonesia, Journal of Development Economics, 115: 200-216. Bergquist, L. F., Burke, M., and Miguel, E. (2018). Selling Low and Buying High: an Arbitrage Puzzle in Kenyan villages. Forthcoming, Quarterly Journal of Economics. Bryan G., Karlan D., and Nelson S. (2010). Commitment Devices. Annual Review of Economics, 2(1): 671-698. Casaburi, L., Glennerster, Kumara, S., and R., Suri, T. (2013). Providing Collateral and Improving Product Market Access for Smallholder farmers: A Randomized evaluation of Inventory Credit in Sierra Leone. 3ie Grantee Final Report. Cortès, G. P., and Carrasco, I. G. (2012). First Line of Defence: Assessing the potential of local food reserves in the Sahel. Oxfam Research Reports. Coulter, J. (2012). Cross-Country Review of Warehouse Receipts Systems in East Africa and their role in development of commodity exchanges. Natural Resources Institute. Mimeo. Coulter, J., and Shepherd, A. W. (1995). Inventory credit: an approach to developing markets. FAO Agricultural Services Bulletin, Issue 120. Delavallade C., Dizon, F., Hill R.V., and Petraud J.-P. (2015). Managing Risk with Insurance and Savings: Experimental Evidence for Male and Female Farm Managers in West Africa, World Bank Policy Research Working Paper Series WPS7176. Dillon, B. (2017). Selling crops early to pay for school: a large scale natural experiment in Malawi. Working Paper Series No. 243. Duflo, E., Kremer, M., and Robinson, J. (2011). Nudging farmers to use fertilizers: Theory and experimental evidence from Kenya. The American Economic Review, 3350-3390. Dupas, P., Karlan, D., Robinson, J, and Ubfal, D. (2018). Banking the Unbanked? Evidence from three countries. American Economic Journal: Applied Microeconomics, 10(2): 257-297. Dupas, P. and Robinson, J. (2013). Why Don’t the Poor Save More? Evidence from Health Savings Experiments. American Economic Review, 103(4): 1138-1171. FAO. (2012). Secteur Porcin Burkina Faso. Revues nationales de l’élevage de la division de la production et de la santé animales de la FAO. No 1. Rome. Frederick, S., Lowenstein, G., and O'Donoghue, T. (2002). Time discounting and time preference: a critical overview. Journal of Economic Literature, 40(2), 351-401. Ghione, A., Kambou, F., Le Cotty, T., d'Hôtel, E. M., and Malnoury, G. (2013). Le warrantage paysan, un outil de protection des ressources. Rapport 52-62. Grain de sel. Godlonton, S., Hernandez, M., and Murphy, M. (2018). Anchoring bias in recall data: evidence from Central America. American Journal of Agricultural Economics, Vol. 100 (2):479-501. Goldberg, J. (2017, November). The Effect of Social Pressures in Malawi. Journal of Economic Behavior and Organization, 143: 173-185. Jakiela, P., and Ozier, O. (2016). Does Africa need a Rotten Kin Theorem? Experimental evidence from village economies. Review of Economic Studies, 83(1): 231-268. Kaminski, J., and Christaensen, L. (2014). Post-Harvest Loss in sub-Saharan Africa — what do farmers say? Global Food Security, 3: 149-158. Karlan, D., and Morduch, J. (2009). Access to Finance. In Handbook of Development Economics, Vol. 5 (Chapter 2). Karlan, D., Osei, R., Osei-Akoto, I., and Udry, C. (2014). Agricultural decisions after relaxing credit and risk constraints. The Quarterly Journal of Economics, 597-652. Kazianga, H. and C. Udry. 2006. Consumption Smoothing? Livestock, Insurance, and Drought in Rural Burkina Faso. Journal of Development Economics, 79(2): 413-46. Malnoury, G. (2011). Analyse du système de warrantage et de son impact sur les ménages ruraux au Burkina Faso: Provinces du Tuy et du Ioba. Thèse de Master, Gembloux Agro Bio Tech, Université de Liège. MkNelly, B., and Kevane, M. (2002). Improving Design and Performance of Group Lending: Suggestions from Burkina Faso. World Development, 30(11): 2017-2032. Pala, L. (2012). L'impact du warrantage : tentative d’évaluation sur le bien-être socioéconomique des ménages bénéficiaires des provinces du Tuy et Ioba au Burkina Faso. Thèse de Master, AGRINOVIA, Université de Ouagadougou (UFR/SH). Shafir, E., and Mullainathan, S. (2013). Scarcity: Why having too little means so much. Times Books. Henry Holt and Company, LLC. Stephens, E. and C. Barrett. (2011). Incomplete credit markets and commodity marketing behavior. Journal of Agricultural Economics, 62 (1): 1-24. Appendix Table 1. Livestock purchases impacts Number of: Cattle Sheep Goats Donkeys Pigs Household randomization (1) (2) (3) (4) (5) Access to warrantage 1.481** 1.395** 0.450 -0.061 0.315 (0.578) (0.658) (0.569) (0.096) (0.384) Observations 518 518 518 518 518 Control mean 3.677 4.759 6.770 0.463 2.922 Notes: The sample is restricted to lottery participants. Standard errors are clustered by village and presented in parentheses. *, **, *** indicate significance at the 1, 5, and 10 percent respectively. Appendix Table 2. Other income Income from Income from non- agricultural agricultural Total income activities activities Total Other of livestock by- Income from within the within the wincself_villag wincself_outsi Income products livestock village village e_ml de_ml Panel A: Household randomization (1) (2) (3) (4) (5) (6) (7) Access to warrantage 1,150.920 -780.016 21.886 -102.367 354.900 6,691.880 -1,337.871 (6,509.090) (544.454) (2,513.449) (210.157) (316.254) (4,469.189) (836.356) Observations 518 518 447 514 487 473 498 Control mean 21504 1900 7241 690.6 497.9 6229 2715 Notes: The sample is restricted to lottery participants. Standard errors are clustered by village and presented in parentheses. *, **, *** indicate significance at the 1, 5, and 10 percent respectively. Appendix Table 3. Determinants of lottery participation Dependent variable: Participated in lottery All villages (1) (2) (3) HHH is male -0.099 -0.111 -0.105 (0.100) (0.097) (0.098) HHH age -0.004*** -0.004*** -0.004*** (0.001) (0.001) (0.001) HHH is married -0.099* -0.106* -0.102* (0.054) (0.054) (0.053) HHH is polygamist -0.037 -0.014 -0.027 (0.047) (0.044) (0.043) # of spouses of HHH -0.021 -0.037 -0.032 (0.034) (0.033) (0.033) Household size 0.018*** 0.018*** 0.018*** (0.004) (0.004) (0.004) Ever heard of warrantage 0.124*** 0.112** 0.120*** (0.044) (0.043) (0.043) Number of crops grown 0.016 0.017 (0.010) (0.010) Value of total produtction (in USD) -0.000 -0.000 (0.000) (0.000) Land owned (in hectares) 0.004*** 0.004*** (0.001) (0.001) Total per capita expenditures (in USD) 0.000 0.000 (0.000) (0.000) Interacts with neighbouring village monthly 0.006 (0.041) Willingness to store in neighboring village 0.053* (0.027) Neigboring village member would return lost wallet -0.003 (0.036) Confidence in warrantage related institutions -0.009 (0.008) Constant 0.789*** 0.729*** 0.670*** (0.096) (0.096) (0.096) Observations 717 717 717 R-squared 0.058 0.070 0.076 F-test 0.639 0.695 0.156 Prob > F 0.429 0.410 0.695 Notes: Sample includes both lottery participant and non-participant households. All dependent variables measured from the census or baseline survey. Standard errors are clustered by village and presented in parentheses. *, **, *** indicate significance at the 1, 5, and 10 percent respectively.