Seasonal liquidity constraints and agricultural productivity: Evidence from Zambia

Transcription

1 Seasonal liquidity constraints and agricultural productivity: Evidence from Zambia Günther Fink Harvard School of Public Health B. Kelsey Jack Tufts University Felix Masiye University of Zambia preliminary and incomplete: do not circulate, do not cite Abstract Small-scale farming remains the primary source of income for a majority of the population in developing countries. In rainfed agricultural settings, income typically arrives only once a year. To meet consumption and investment needs over the subsequent months in the absence of formal financial markets, households adopt a range of coping strategies. Two common strategies are reducing food consumption and selling labor to other farms, both of which improve short term liquidity, but may harm subsequent harvest outcomes. To investigate whether farm productivity is increased by access to short-term credit, we conducted a randomized controlled trial across 175 villages rural Zambia, which provided selected households with access to approximately US$ 40 of loans during the lean season, with repayment due after harvest. We find that access to credit during the lean season increases harvest output and revenue by around 10 percent relative to the control group, an increase similar in magnitude to the amount owed on the loan. This impact is driven mostly by increases in food consumption and labor hiring, as well as decreases in the frequency of selling labor to other farms. We observe no statistical impact on other agricultural input expenditures, and no effect on other consumption smoothing strategies that are less seasonal in nature, such as asset sales. Our results suggest that both the seasonal consumption and labor allocation of small-scale farmers are affected by frictions in the capital market. The high take up and repayment rates for the loans as well as reduced prevalence of hunger and improvements in self-reported well-being suggest that the overall welfare gains associated with increased access to credit may be substantial. We thank audience members at IGC Growth Week for comments and suggestions. We are grateful to the Growth and Labor Markets in Low Income Countries (GLM-LIC), the International Growth Centre, the Agricultural Technology Adoption Initiative (JPAL/CEGA) and an anonymous donor for financial support, and to Innovations for Poverty Action for logistical support. Many thanks to Rachel Levenson for management of the field work and to Chantelle Boudreaux and Carlos Riumallo Herl for assistance with the data. 1

2 1 Introduction In Zambia, like in much of Sub-Saharan Africa, agriculture employs the vast majority of the rural population, with generally low levels of productivity and farming income. 1 Alackofirrigation infrastructure combined with a long dry season means that harvest income arrives only once per year, and must cover household needs for the subsequent months. across seasons is difficult if access to capital markets is limited. 2 Distributing resources In the absence of functional capital markets, households may turn to alternative strategies for smoothing consumption, including livestock and asset sales (Rosenzweig and Wolpin 1993, Janzen and Carter 2014), off-farm labor (Kochar 1995, 1997; Jayachandran 2006; Ito and Kurosaki 2009) migration (Halliday 2012; Bryan et al. 2013), or lowering food intake (Kazianga and Udry 2006; Kaminski et al. 2014). 3 All of these mechanisms tend to be costly. In the case of livestock or asset sales, high transaction costs along with seasonal price fluctuations lead to financial losses to the household; in the case of hunger or off-farm labor, farmers may suffer from physical and emotional hardship and lower subsequent harvest outcomes. The high cost of these smoothing mechanisms implies that both the marginal cost of consumption and investment varies substantially across the agricultural season. Anticipating this, utility-maximizing farms will alter both the total quantity of land used and the crop mix chosen and deviate from the optimal production plan in an unconstrained environment (Fafchamps 1993; Rosenzweig and Binswanger 1993). These extensive margin or ex ante inefficiencies can be further compounded by intensive margin inefficiencies if households experience unanticipated income or expenditure shocks during the farming season. To cover liquidity needs associated with these shocks, farming households may further deviate from their original (adjusted) production plan by reallocating inputs, such as household labor, to meet consumption needs (Kochar 1995, 1999; Rose 2001; Ito and Kurosaki 2009). 4 To investigate the degree to which agricultural productivity can be improved through short-term credit during the lean season, we conducted a cluster-randomized experiment with 3140 small-scale farmers from 175 villages in rural Zambia. Zambia s agricultural cycle is centered around the rainy season from November to April. Harvest takes place in May and June, and generates income that would ideally cover all consumption and investment needs in the subsequent year. As illustrated 1 A recent study in the region of proposed research shows average gross production value of USD 500 for a family of six (Fink and Masiye 2012). Once capital input, land and labor costs are considered, net profits for many of these families may be negative. 2 The general relationship between income seasonality and consumption smoothing is not well established in the literature. While some studies suggest that precautionary savings are sufficient to smooth consumption even if income is highly seasonal (Paxson 1993; Chaudhuri and Paxson 2002; Jacoby and Skoufias 1998), others have highlighted the pronounced consumption differences over the year (Dercon and Krishnan 2000; Khandker 2012). Kaminski et al. (2014) point to recent evidence from Africa that seasonal consumption patterns are closely linked to seasonality in staple crop prices, which suggests that households are not able to adequately smooth consumption. 3 For summaries of the consumption smoothing literature, see Morduch (1995) and Besley (1995). 4 Typical income shocks in the study area include the loss of stored food reserves due to pests or theft; expenditure shocks include funerals, school uniforms and medical costs. 2

3 in Figure 1, household (food) reserves gradually decline between July and December, and are most scarce from January to March before early crops become available for consumption. The January to March period is generally referred to as the lean season or the hungry season by farmers, and is the period we directly targeted with the intervention. For the experiment, villages were randomly divided into three groups. In treatment group 1 (59 villages), households could borrow 200 Kwacha (approximately USD 40) of cash in January. In the second group (58 villages), farmers could borrow three bags of maize. The three bags of maize were roughly equivalent in terms of financial value to the cash loan, and theoretically provide a sufficient amount of calories to feed a family of five for the at least two months. 5 In both loan groups, repayment was due after the harvest in late June to early July. All borrowing households were given the option to either repay in cash (260 Kwacha) or in kind (4 bags of maize). 6 The remaining third of the villages were assigned to a control group. 7 Both the demand for and the willingness to repay loans was high, with around 98 percent take up among eligible households and close to 95 percent repayment in both treatment arms. Though households in both treatment arms were told that they were free to repay in whichever modality (cash or maize) they preferred, households were more likely to repay cash loans with cash and maize loans with maize. To assess the impact of loans on agricultural productivity, we develop a series of predictions through a simple multi-period agricultural production model, and test them empirically using our experimental data. Consistent with the model, we find that agricultural output increases in villages where loans were available, with an estimated intention-to-treat effect of KR 271 or 8.7 percent, marginally (but not statistically significantly) higher than the loan repayment amount of KR 260. To investigate the causal mechanisms underlying these results, we examine impacts on food intake and nutrition, as well as asset and livestock retention during the hungry season. We also examine program impact on household labor allocation and short-term labor hiring, as well as household investment in productive inputs such as fertilizer and pesticides. Overall, we find no evidence of loan programs affecting inputs (seeds, pesticides, fertilizer), which may partially be a result of the delivery of loans relatively late in the cropping season in January, when the need for fertilizers and pesticides is limited. We do, however, find relatively large impacts on food consumption and labor: on average, farms in the loan treatment arms were on average 15 percentage points less likely to experience hunger during the peak hungry season (January to March) and consumed on average 0.2 more meals per day. In terms of labor allocation, farms eligible for a loan were 4 percentage points less likely to sell labor to other farms, and 6.5 percentage points more likely to hire additonal labor for their own farms. We also test whether the nature of the loan (cash or in kind) matters in this setting. Our results 5 One kilogram of maize provides approximately 3600 kcal. Three bags would thus provide 5 household members about 1800 calories per day over a 60 day period 6 Official rates set one 50 kg bag of maize at 65 Kwacha. However, local seasonal fluctuations in maize prices affect the relative value of the two loan offers. In Section 4, we calibrate the interest rate and value of each loan. 7 In a small sub-sample of the control villages, households were given a gift of 60 Kwacha, which serves as a control for any income-effects that the loan may generate. 3

4 suggest that the overall utilization of the additional resources provided differed across the two arms: while the maize loan induced larger increases in food consumption than the cash loan, cash loans seem to have induced more labor hiring, and overall higher labor inputs on farms. The effects of the maize loan program on agricultural output are smaller than in the cash treatment arm, and are generally not statistically significantly different from either the control group or the cash loan villages. Recent evidence on the impacts of capital access interventions on agricultural productivity is mixed. 8 In Ghana, Karlan et al. (2014) find no evidence that liquidity constraints impede agricultural investments. Beaman et al. (2014) find that relaxing credit constraints through grants increases agricultural investment and yields among rice farmers in Mali, but that the same is not true for loan programs. Both studies focus on farming inputs (farm expenditure on seeds, fertilizer or pesticides) as the primary mechanism through which credit impacts yields. The results presented in this study suggest that loans may have an impact on farm productivity via their effect on smoothing strategies including labor allocation and nutrition. In settings where access to capital markets are limited, farms engage in a range of costly smoothing strategies both to finance consumption in the lean part of the season and to deal with unanticipated liquidity needs (Kochar 1995, 1999; Rose 2001; Ito and Kurosaki 2009). The results presented in this paper suggest that reducing food intake and selling labor to other farms are the most common strategies chosen in the setting studied. A large literature on nutrition and productivity debates whether farmers constrained nutritional intake leads to suboptimal production (Pitt and Rosenzweig 1986; Strauss 1986; Behrman et al. 1997), an idea also supported by recent evidence from India (Schofield 2013). Similarly, a growing literature suggests that selling off-farm labor is not consistent with income maximization, and likely to lower overall farm productivity (Kerr 2005; Bryceson 2006; Orr et al. 2009; Michaelowa et al. 2010; Cole and Hoon 2013). Our results suggest that loans are effective in increasing staple crop consumption, an effect which appears to be particularly large for maize loans. Cash loan programs appear to be more effective in moving farms towards the income-maximizing level of labor input. This second adjustment seems to be more relevant in terms of the overall impact on output; however, the benefits of improved nutrition may clearly go beyond total agricultural productivity. The results in this study are also linked to a growing literature investigating the impact of seasonal transfer or loan programs. Burke (2014) offered farmers in Kenya a loan product that allowed them to exploit seasonal variation in maize prices and finds significant effects on total maize revenues and household expenditures. Bryan, Chowdhury and Mobarak (2013) provide credit and grants for short run seasonal labor migration in Bangladesh and argue that credit market failures and highly uncertain returns likely keep long-distance labor supply below welfare-maximizing levels. Most similar to our study, Basu and Wong (2014) study a seasonal food credit and improved storage 8 An earlier literature uses observables to define whether households are credit constrained, and compares productivity and consumption across constrained and unconstrained households (e.g. Feder et al. 1990). 4

5 program in Indonesia and find that food loans increase non-staple food consumption during the lean season and income from crop sales at harvest, but do not analyze impacts on yields. Our findings contribute to that literature by providing the first direct evidence that capital market interventions timed to coincide with the hungry season can increase agricultural productivity. The paper proceeds as follows. In the next section, we present a simple model of agricultural production in the presence of credit constraints. Section 4 describes the study context, experimental design and implementation. Section 5 sets up the identification strategy and presents the results, and Section 6 concludes. 2 Conceptual framework Consider a simple model of agricultural production, where rational farming households maximize their utility over consumption and leisure. Households start off with an endowment consisting of previous assets and their most recent harvest, and maximize their utility by optimally allocating resources to investment and consumption. The agricultural season is divided into three periods: period 1, the post-harvest season (July to September in the Zambian context), where farm activities are limited due to lack of rainfall; period 2, when fields need to be prepared and crops need to be planted (October - December); and period 3, where fields need to be weeded and maintained (January - April). Forward-looking farmers maximize the following utility function: u(c 1 )+ u(c 2 )+ 2 u(c 3 )+ 3 V (Y,B) (1) where u(.) is a generic concave utility function, < 1 is the subjective discount rate, and V (.) is the indirect utility derived from the final net harvest value Y, net of borrowed resources. We measure all inputs and outputs in monetary units and normalize all prices to one. Farms have an initial capital endowment A 0, which comprises previous assets and savings as well as net harvest outcomes from the most recent season. Farms can earn a return r s on savings, and can borrow locally at a rate r b > 1. 9 In period 1, farms can consume or save. In periods 2 and 3, farms can consume, save or invest into their field. The investment in both periods 2 and 3 (I 2,I 3 ) can be financed by loans, which need to be fully repaid at the end of the season. Given this, the consumption constraints in periods t =1, 2, 3 are given by c t + I t + S t apple S t 1 r s + B t (2) The total debt payments due at the end of the season are given by B = B 1 r 3 b + B 2 r 2 b + B 3 r b, and net harvest income is given by Y (I 2,I 3 ) B. 9 Note that r is therefore equal to the interest rate on borrowing or savings plus one. 5

6 Perfect Capital Market Equilibrium In the absence of credit market frictions, farmers can save and borrow at a constant return r = r s = r b. With unrestricted capital access, the optimal amount of investment in periods one and two is such that the marginal product of on-farm inputs, including labor, equals the market rate of return, i.e. Y 0 (I t )=r (1+t), (3) where Y 0 is the marginal product (harvest) generated by the period-specific investment, t =2, 3, and r is the market interest rate. We shall denote the output achieved under this optimal investment plan by Y = Y (I2,I 3 ). Consumption will be such that the marginal rate of substitution across the three period equals r, which implies u 0 (c 1 )=r u 0 (c 2 )=r 2 2 u 0 (c 3 )=r 3 3 V 0 (Y,B). (4) Capital Market Frictions and Interventions Similar to most other developing country settings, capital market access is limited in Zambia, with low (or no) interest earned on savings, and generally very high interest rates on borrowing, so that r s <r<r b. The model is set up to allow for a range of borrowing mechanisms: in principle, farms can borrow from informal and formal money lenders; they can sell livestock or household assets and repurchase it after the harvest at future dates; they can also take on work on other farms (ganyu), giving up some of their own future production. All borrowing mechanisms are likely to be very costly in practice either due to high interest rates, high transaction cost, high risk or a combination of all factors. This friction in the credit market leads to non-separability of investment and consumption decisions, with two important production adjustments relative to the optimum with perfect capital markets. First, r s <rimplies that farms shift resources towards consumption in the first period, lowering subsequent investment and final harvest output. Second, higher cost of borrowing imply - by optimality condition (3) - that farms will invest less in both periods 2 and 3, which results in lower harvest outcomes, and lower net incomes compared to the unrestricted capital market model. This stylized model generates several testable hypotheses for the roll-out of credit programs like the ones described in the study: H1: Access to credit markets increases average small scale farmer output and welfare. The smaller the starting endowment of the farmers, the larger the impact on output and welfare. Hypothesis 1 is relatively straightforward: access to credit markets implies that small scale farmers are able to save at a higher return or to borrow at lower cost, which in turn allows them 6

7 to invest more into their plots and achieve higher yields. If access to credit is in fact limited, we expect willingness to participate in loan programs to be high among small scale farmers. We expect smaller impacts for farmers with larger initial endowments who should for a constant plot size be better able to self-finance consumption and investment needs. H2: Loans announced and made available during period 3 will increase period 3 consumption, period 3 investment and final harvest output. H3: Loans announced during period 2 and made available period 3 will increase consumption and investment in both periods, and lead to larger output increases than loans announced in period 3. Hypotheses 2 and 3 highlight the importance of anticipating credit availability. Given that a substantial fraction of farming decisions are taken early on in the agricultural cycle, if loans are not announced until late in the agricultural cycle, they can only be used to adjust one margin of the production process, with accordingly smaller productivity impact than loans announced at the beginning of the agricultural cycle, which allow for changes in both early and late decisions. Earlier knowledge of loan availability allows farmers to increase consumption and investment in period 2 as well as period 3. Given the adjustments on both margins with early announcement, we expect consumption increases in period 3 to be smaller with early announcement than with late announcement. H4: The long run productivity impact of single-period loan programs increases with the the marginal return to investment and decreases with the farmer s discount rate. Hypothesis 4 is more complex, highlighting the importance of the loan allocation chosen by farmer. While loans should unambiguously increase output and overall utility (as described in hypothesis 1), the long-run benefits for farmers are less clear, since increases in output may be more than offset by increases in total cost. With high degrees of myopia (small ) it may be optimal for the farm to use the full loan amount for consumption, and invest very little in additional inputs; farmers would then be better off overall in terms of the discounted net present value of their utility, but worse off in terms of future availability of resources. Similarly, lower marginal return to investment (for example in case of small farms, or in case of loans arriving late in the season) will mean a relatively larger share of the additional resources allocated to consumption, with an accordingly smaller amount of net resources available for period 3. The model resembles in many ways a classic poverty trap setup: low initial endowments (high poverty rates) combined with capital market frictions lead to a suboptimal production plan including suboptimal nutrition and suboptimal labor inputs on fields. Improved access to capital could thus in theory not only improve production in the short run, but also raise longer run output by reducing farm s dependency on external capital in subsequent years. 7

8 3 Background and context The study was implemented over the course of a year in Chipata District in Eastern Zambia. Chipata District is located at the southeastern border of Zambia, with an estimated population of 456,000 in 2010 (CSO 2010). Approximately 100,000 people live in Chipata town, the district and provincial capital; the remaining population lives in rural areas, with small-scale farming as primary source of income. According to the 2010 Living Conditions and Measurement Survey, rural households in Chipata are on average poorer than in the rest of the country, with 47 percent of household classified as very poor in the district overall, and 63 percent of households classified as very poor in the rural parts of Chipata. Average monthly expenditure of rural households is about one third of the national average, and access to electricity and piped water close to zero in rural areas (see Appendix table A.1 for a summary of differences between Chipata and the rest of Zambia). 3.1 Local credit and labor markets The conceptual framework builds on several contextual features, namely local capital and labor markets. We provide additional qualitative background on these features of the study setting. As described in greater detail below, the study sample was limited to small farmers those with land holdings of 5 hectares (12 acres) or less. The attribute of small-scale is somewhat misleading since it suggests that these farmers are unusually small; in fact, small-scale farmers represent the overwhelming majority of households in rural villages in Zambia. In our study villages, over 90 percent of listed households fall into this category. Capital markets In terms of borrowing opportunities, the study setting is also fairly representative of rural areas in developing countries, where credit markets are absent or very costly to access. In the baseline survey, 2 percent of household respondents report accessing formal loans for something other than inputs. 10 Input loans are more common: around forty percent of baseline respondents accessed an in-kind input loan, typically provided by companies purchasing cash crops like cotton and tobacco from small scale farmers. For accessing cash, informal borrowing channels are slightly more common: around 7 percent report taking high interest loans, locally referred to as kaloba, with interest rates over 100 percent. Loans between friends and family are reported by around 8.5 percent of baseline respondents. Rotating savings and credit associations (ROSCAs) are very rare in the study setting, reported by around 1 percent of baseline respondents, as are village savings and loan associations (VSLAs), also reported by around 1 percent of baseline respondents. Rates are similarly low for savings. Only 5.6 percent report saving in a bank; slightly more (9.1 percent) report saving with friends, family or employers. By far the most common savings strategy, reported by 76.7 percent of households, is saving money at home, while only 8 percent of 10 Formal lenders include banks, credit unions, government sources, NGOs, and agricultural companies 8

9 baseline respondents report zero savings over the past year. The median self reported cash savings (a measure likely to be reported with substantial error) at baseline, at the start of the planting season, is 80 Kwacha or around 16 USD. Savings also occurs through grain storage, which typically occurs in a thatch (28 percent of respondents) or bamboo (62 percent of respondents) granary. Sixty percent of households report storage losses and the median grain in storage at baseline is only four bags, or enough to last a family of four until February or March at most. Thus, both cash savings and grain storage are insufficient to last most households until the next harvest. Ganyu labor Local wage earning opportunities for study households are defined largely by casual or piecewise labor locally referred to as ganyu. In focus groups, a majority of small-scale farmers in Chipata described ganyu labor both as the most common strategy to cope with temporary liquidity shortages, as well as an activity most farmers would rather avoid if possible. In the baseline survey, the most common response to why an individual in the household worked ganyu during the previous agricultural season was to obtain food. The second most common reason was to access cash for a personal purchase, and the third was to deal with an emergency. When asked what the household will do in the coming year if they run out of food, 56 percent report that they will do ganyu. The next most common answers include borrow from friends or family (28 percent), using savings (22 percent) and sell assets or livestock (17 percent). Households appear reasonably accurate in their forecast of whether they will have to engage in ganyu in a given year. Among control group households that predicted at baseline that they would have to do ganyu in the coming year, around 76 percent did; among those that predicted not doing ganyu, around 41 percent ended up working off-farm. Households that sell ganyu one year are not necessarily sellers in all years. Among control group households that did not engage in ganyu the year before the study, 40 percent sold ganyu the following year. The model that we present in Section 2 simplifies a complex rural labor market. In the study setting, road infrastructure is extremely bad, there is no motorized public transport and distances between villages are substantial. Most casual labor takes place in or near the worker s own village. In the labor survey, over 60 percent of reported ganyu incidents occurred in the respondents own village, and almost 90 percent were for another small farmer (i.e. fewer than 5 hectares of land). This highlights that fact that the boundaries between ganyu buyers and sellers are fluid, and the same farm may sell ganyu at one point in the year, and purchase labor at another when more resources are available. 3.2 Study sample The study sample was constructed to be representative of Chipata District. The district is divided into 8 administrative blocks, each of which contains a number of camps. We randomly sampled 5 villages from 50 of the 53 camps in the district, omitting the camps that contain Chipata town. 9

10 The village list was assembled from the Ministry of Agriculture s farm registry, which includes all registered farms in the district. To facilitate sampling, villages with less than 20 or more than 100 farms listed were excluded. IPA enumerators visited the sampled villages in order, recording the number of households, farm sizes and screening for eligibility. Villages were ineligible if: (1) IPA had worked there before, (2) the village bordered a village that was in the study pilot, (3) the village bordered a village already listed, (4) the village had fewer than 17 households, or (5) it was impossible to get a 4x4 vehicle within a 5km radius of the village during rainy season. These eligibility criteria eliminated more villages than expected, and an additional 150 villages were sampled randomly across all camps to supplement the list. Enumerator screening visits stopped once 201 villages met all eligibility criteria. During the baseline survey, 25 additional villages were eliminated for a failure to meet one or more of the eligibility criteria that had been overlooked during the screening process. In addition, one village refused to participate in the baseline survey. This left us with a sample of 175 villages for the study. Within each eligible village, households were sampled from the village rosters collected during the screening visits. Only small farmers less than 5 hectares according to the Zambian Ministry of Agriculture were eligible for the program. 11 Eligible households were randomly sorted and the first 22 selected for the baseline survey. A total of 3,701 households were sampled for the baseline and 3,141 were surveyed (84.9 percent). 12 We describe attrition, conditional on being in the baseline sample, in Section 4.4 below. 4 Experimental design Study implementation began in October 2013 and will last for two years. We describe the experimental design for both years but show results only for year Loan treatments The main objective of the project was to estimate the productivity impact of short run loans offered during the hungry season on household-level outcomes. In January 2014,two types of loans were offered to randomly selected subsets of households: a maize loan and a cash loan. During year 1 of the program, 58 villages (1033 farms) were assigned to a control group, which received no intervention, 58 villages (1092 farms) were assigned to a cash loan program, and 59 villages ( We restricted our sample to households with at least 2 acres of land to distinguish households with very small scale home gardens from households engaged in crop production, and also to increase the likelihood of sufficient harvest to repay the loan. 12 The most common reasons that listed households were not surveyed were that they were temporarily or permanently away from the village (N=219) or that they were ineligible when land size was verified with the household head (N=146). 10

11 farms) were assigned to a maize loan program. In the second year of the program, the treatment groups will be rotated to identify persistent effects of the loan treatments. The timing of the loan announcement was also varied, with half of the treated villages receiving notification before the start of the planting season, in September. The design details for the intervention are described in Appendix table A.1. The loan treatments are summarized in Table 1. In both treatment arms, the loan offer was announced in early January 2014, at the start of the hungry season. In the maize loan treatment arm households were offered three 50-kilogram mags of unpounded maize, enough to feed a family of five for at least two months. In the cash loan treatment arm households were offered 200 Kwacha (~ USD 40), an amount equivalent to three bags of unpounded maize at government prices. In both treatment arms, repayment was due in July, toward the end of the harvest period, and households could repay either 4 bags of maize of 260 Kwacha (or a mix at K65 = 1 bag). While both treatment arms were designed to reflect an interest rate of about 30 percent, actual interest rates are hard to compute due to substantial seasonal price fluctuations in major crop prices. The calculation is further complicated by the transaction costs associated with buying and selling maize, which is often unavailable in the village during the lean season. As shown in Table 1, the interest rate in the maize arm is between -11 and 33 percent (excluding transaction costs), depending on the calculation, and also depending on the repayment modality chosen by farmers.some further discussion of the comparability of the maize and cash loan sizes is warranted. While the value of the maize loan may appear higher than the value of the cash loan in January based on locally reported seasonal prices, few maize transactions take place in January, because most households are severely liquidity constrained during this period. To make the two loan programs as comparable as possible, we conducted a series of hypothetical choice experiments in non-target villages within the district in November In these choice experiments, respondents (N=72) were asked a series of dichotomous choice questions on whether they would prefer a loan for three bags of maize over a cash loan of x Kwacha, with x varied between 50 and 600 Kwacha percent of respondents preferred a maize loan over a cash loan of 175 Kwacha, while 36 percent preferred the maize loan over the next choice value a cash loan of 250 Kwacha. In a second set of questions, respondents were asked if they would take up a maize (cash) loan that paid 3 bags (200 Kwacha) in January with a repayment of 4 bags (265 Kwacha) due in x month, with x varied between February and December. For both hypothetical choice sets, acceptance rates jumped from 27.8 and 20.8 to 81.9 and 83.3 in June the maize and cash loan questions, respectively. The hypothetical choice experiments therefore provide support for indifference between the loan options around the values chosen for implementation. Further detail on the implementation of the choice experiments is provided in Appendix A.2. Treatments were assigned at the village level using min-max T randomization (Bruhn and 13 Hypothetical loan dates were consistent with program offered (pay out in January and repayment in June), but the hypothetical loans involved no interest. 11

12 McKenzie 2009), checking balance on both household and village characteristics. The approach relies on repeated village-level assignment to treatment and selects the draw that results in the smallest maximum t-statistic for any pairwise comparison across treatment arms. Balance was tested for 14 household level variables, village size and geographic block dummies, with results described in Section 4.4. The smallest p-value for the pairwise comparisons observed in the final draw was p= In addition to the main loan treatments, a small number (N=6) of villages were assigned to an income effect control, which provided a cash gift of 60 Kwacha, to capture potential income effects of the loan program. 14 villages assigned to the control group. 4.2 Implementation These were selected by random draw within geographic block from among The loan was administered under the project name Chipata Loan Project (CLP) to distinguish it from the surveys, which were being run by IPA. This distinction was intended to minimize strategic responses to the survey questions, but the relationship between CLP and IPA was not denied if a participant asked. The loan intervention was announced to households during a village meeting to which eligible households were invited. 15 At the meeting, project staff began by describing eligibility for the program to clarify why only some households were invited to the meeting. The terms of the loan were then described, followed by details on how the loan distribution would be organized. Loan enrollment and consent forms were provided to eligible households. If a household wished to join the program, they were required to present both forms with a signature of the household head when picking up the loan. Loans were distributed between 3 days and one week after the village meeting at a location convenient for transportation, selected in cooperation with the village headman. Project staff registered attendees, confirmed their identity using the national registration card, 16 and collected their signed enrollment and consent forms. Before finalizing the transaction, project staff confirmed that the participant understood the terms of the loan. The loans (cash or maize) were handed over and a receipt was provided to the household and kept for project records. Repayment was due six months later, in late June to early July. Villages were notified in advance about the location and date of repayment. Households were provided with a repayment receipt upon 14 The size of the cash gift in the income effect control was calibrated using choice experiments described in greater detail in Appendix A Ineligible households were not barred from listening in. Eligible households could send an adult representative if the household head was not available to attend. All village headmen were eligible for the loan, even if they were not sampled for the baseline survey (and are therefore not in our study sample). In addition, the baseline data for 3 households who were surveyed was lost. They are dropped from the sample. 16 In select cases, a household representative picked up the loan. In these cases, the representative needed to carry the loan-holder s NRC card with him or her. 12

13 full repayment, and a second visit was made to selected villages to follow up on loans that were not fully repaid during the first visit. Throughout, households were told that the program might or might not continue for a second year. Further summary statistics on repayment patterns are described in Section Data We rely on both survey and administrative data in our analysis. Administrative records include loan take up and repayment, two key outcomes in our data. Survey data come from the following sources: 1. Baseline survey (November-December 2013): Survey of up to 22 households per village, conducted with household heads. The baseline survey includes sections on household demographics (including individual roster, employment roster of working household members, general household information about assets owned and food insecurity faced, farming information for season, expected farming activity for season, risk and time preferences), 2. Labor survey (January 2014-ongoing): Rolling survey of ~70 households per week (7 of the baseline households in 2 villages per day). The list of baseline households for each village were randomized and the first ~7 households interviewed, in cases where a household can t be interviewed (temporarily busy, moved, etc.), the household is skipped and the next household on list visited. Survey asks one week and one day recall questions on household labor allocation, ganyu earnings, and consumption (including consumption of green maize). 3. Employer survey (January 2014-ongoing): Rolling survey of ~10 ganyu employers per week. Sampling is based on Labor survey records of where households in a village report doing ganyu. Additional sampling is done in a snowball method where employers interviewed then provide names of other employers of ganyu that they know. The employer survey tracks the labor survey by geographic block and rotates through villages rather than targeting an explicit sample. 4. Midline maize assessment (February-March 2014): On-field assessments of maize height (measurement) and visual records (photographs) for a sample of 380 households in 64 villages. Only households with their nearest field within a 30 minute walk were eligible. 5. Midline survey (February-March 2014): Hungry season survey of 1200 randomly selected households, stratified on treatment. One week and one month recall questions on labor supply, ganyu earnings, consumption, basic strength and anthropometric measurement. 6. Harvest survey, year 1 (July-September 2014): Survey of all baseline households. Includes sections on changes to household composition, shocks experienced by the household, agricultural 13

14 productivity. Includes anthropometric measures for adults and children. 7. Year 2 data: Data collection will be repeated in Year 2, including the labor and employment surveys, and an abbreviated harvest survey. The midline survey will be collected for a reduced sample size. 4.4 Identification We estimate intention-to-treat regressions, including all households regardless of whether they selected into the loan. Our primary specification for evaluating the overall effect of relaxing credit constraints is: y ivt = + loan vt + X iv + t + u ivt (5) where y ivt is an outcome of interest for household or individual i located in village v and month or season t. loan vt indicates that the village was assigned to either the cash or the maize loan treatment, X iv is a vector of controls measured at baseline and t are month-year or season-year dummies to capture seasonal effects. Treatment assignment varies over time according to the treatment rotation between years, as described in Section4.1. Errors are clustered at the level of the randomization unit, the village v. We can also break out the treatment effect by treatment arm, and estimate separate coefficients for the cash and maize loans. In much of the analysis, we analyze self-reported outcomes from the midline or harvest surveys by collapsing equation (5) into a cross section and controlling for lagged outcomes measured at baseline. We also estimate time-specific treatment effects by interacting treatment indicators with time dummies and including village fixed-effects, : y ivt = + loan vt t + X iv + v + u ivt, (6) which delivers a vector of coefficients for each treatment arm by month or season. Given the large number of causal mechanisms and pathways explored, we show both unadjusted p-values and p- values correcting for multiple testing. Specifically, we show significance level with the very restrictive family-wise error rate (FWER) correction as well as under the less restrictive false discovery rate (FDR) adjustments originally developed by Benjamini and Hochberg (1995) and Benjamini and Yekutieli (2001). See Fink et al. (2014) for further details on these methods. Balance and summary statistics The coefficients presented in the subsequent analysis identify the causal effect of the loan under the identifying assumption that treatment assignment is orthogonal to u iv. Table 2 presents the means and standard deviations of baseline survey characteristics among study households, by treatment 14

15 arm (columns 1-3). Column 4 shows the largest pairwise t-statistic and column 5 the largest pairwise normalized difference. While the t-statistic on household assets suggests a significant difference between the maize loan treatment arm and the control group, the normalized difference is nevertheless small. Overall, the randomization successfully balanced households across treatment arms. The variables shown in Table 2 are our household-level controls, used throughout the analysis. We also test for balance in other surveys that rely on a sub-sample of the study households. Appendix tables A.2 and A.3 show the covariate balance for the midline sample and the rotating labor sample, respectively. 17 While a couple of the pairwise t-statistics suggest significant differences, none of the pairwise normalized differences exceed the 0.25 rule of thumb threshold for balance. Attrition and selection The main identifying assumption of our empirical analysis could be violated if households select into eligibility status or drop out of the loan program differentially across treatments. Households could exit the study during year 1 both during the midline survey and the endline survey. Overall attrition rates are low as shown in Table 3. The table reports means and standard deviations at baseline in column 1 and the coefficients from a regression of a binary interview indicator for each survey round on our main set of controls, with standard errors clustered at the village level. While the individual coefficients suggest some selection into the midline and endline, particularly among households with children, the overall attrition rate was low across both interview rounds. In the midline, of the 1223 households sampled for interviews, 97.6 percent were surveyed. In the year 1 endline, 96.5 percent of the full sample was surveyed. 5 Results Following our pre-analysis plan, we organize our analysis and results into five groups. The main objective of the study is to assess whether relaxing frictions in the capital market can improve agricultural productivity. Given this objective, the primary outcome of interest is agricultural yields. Based on the existing literature as well as the evidence compiled as part of the pilot study concluded in 2013, we hypothesize that credit constraints affect productivity through several causal pathways. We will begin with this core set of outcomes, and report adjusted p-values for multiple hypothesis testing. The second main question of interest addressed by the study and the main rationale underlying its 3-arm design is whether in-kind maize loans are more effective than cash loans for improving productivity. Last, we will analyze spillovers of the intervention on market prices in order to be able to (at least partially) address the general welfare implications of the program, and calculate preliminary cost effectiveness comparisons. 17 Results are similar if we break the labor sample into the hungry season (January - March) and harvest season (April - June). 15

16 Before turning to these hypotheses, we provide results on administrative outcomes: loan take up and repayment. 5.1 Take up and repayment We begin with an analysis of take up by treatment group. Take up is 98.6 percent among eligible households in the cash loan treatment arm and is statistically indistinguishable from take up in the maize treatment arm, as shown in columns 1 and 2 of Table 4. The remaining columns of Table 4 show repayment rates. 93 percent of the 2008 households that took out a cash loan repaid it in full (columns 3 and 4). Full repayment does not differ between the cash and maize loan treatment arms. Some households partially repaid, bringing overall average repayment to 94.3 percent (columns 5 and 6). Finally, in the cash loan treatment arm, 35 percent of households with loans made at least some of their repayment in cash. In the maize treatment arm, the probability of repaying any cash was 22 percentage points lower (columns 7 and 8). That said, in both treatment groups, some households repaid in cash, some in kind and some with a combination of the two. Consistent with our conceptual framework, demand for the loans is high. 5.2 Yield impacts To the extent that credit and savings constraints affect consumption and input decisions, as described in the conceptual framework (Section 2), both types of loans (cash and maize) are likely to affect agricultural yields. We first establish the impact on yields before turning to causal pathways and secondary outcomes of interest. Agricultural yields are measured through self-reported quantity estimates collected as part of the harvest survey and endline for all crops. Under the strong assumption that crop mix is unaffected by the loan treatment, which is plausible if the loan announcement occurs after all cropping decisions are made, we can use the raw output measure, number of kilograms, as an outcome. However, given that the crop mix varies considerably across farmers, and typical yields and values per kilogram vary substantially across crops, we compute a combined production value across all crops. For this purpose, farmers are asked to report the total quantity of all crops on their field, including the quantities used for own (early) consumption, quantities sold or used for payment, quantities stored and quantities still on the field. To convert total harvests into monetary terms, all reported quantities are multiplied by the self reported price to capture heterogeneity in value based on location. We also use median prices in the treatment group to reduce measurement error. We do not attempt to calculate profits, nor do we calculate revenue net of the loans, since the loans offered through the program may have substituted for other higher (or lower) interest borrowing. We instead examine these other responses, including impacts on borrowing and agricultural inputs (fertilizer, labor, pesticides), below. Revenue net of the program loans can, of course, be approximated by 16

17 subtracting around 260 Kwacha from the harvest revenues. We estimate equation (5) for these three outcome measures, and report results in Table 5. Panel A shows treatment effects for any loan type, which we discuss first. We return to the effects by treatment arm (Panel B) in Section 4.1. We control for household-level baseline variables as shown in Table 2 and geographic block dummies in even numbered columns. By controlling for previous year outcomes, baseline controls substantially improve precision, therefore we focus on the specifications with controls in our discussion of the results (even numbered columns). Panel A indicates that the loans had a positive effect on the total kilograms of output across all crops (column 2) and harvest value at median prices (column 6). At own prices (column 4), the coefficient on the loan intervention is slightly smaller than at median prices and less precisely estimated. Effects at median prices are consistent with a 8.7 percent increase in harvest value, while the effect on total output in kilograms is around an 8 percent increase over the control group. Consistent with the conceptual framework, the gross impact of relaxing credit constraints on the value of agricultural output is positive. Table 5 also reports the effect, measured as part of the midline survey, of treatment on maize height. Maize height is self-reported and measured according to a standard meter stick in centimeters. While the measures are noisy, they show statistically insignificant increases in self-reported maize height overall in the treatment groups. 5.3 Causal mechanisms A number of potential causal pathways underlie the effect of the loans on agricultural productivity. In addition, many of these pathways represent outcomes in their own right. For example, even if yields do not increase as a result of the loan, better consumption smoothing may make households better off. We break our analysis down into five groups, as described in our pre-analysis plan: (i) food intake and nutrition, (ii) labor supply, (iii) other productive inputs, (iv) cognition and decision making and (v) the ex ante selection of the crop mix. We investigate (i) through (iii) in the current draft and leave (iv) and (v) until year 2 data are available Food intake and nutrition As stated in Hypothesis 2, the availability of loans during the hungry season will increase hungry season consumption. 18 In our data, we expect to see an increase in food consumption (a reduction in food shortages and missed meals), improved strength and endurance and potentially also an increase in body weight and musculature. We also expect to see less engagement in costly consumptionsmoothing strategies, such as livestock and asset sales and consumption of green maize before harvest. We focus on the consumption measures in the current analysis. 18 Hypothesis 2 predicts that if the availability of loans is announced in advance, then consumption will adjust even before loans become available. We will test this hypothesis with the variation in the timing of loan announcement in year 2 of the intervention. 17