Implementing Effective Warehouse Receipt Financing Systems: Lessons from a pilot WRS project in the Senegal River Valley Guigonan Serge Adjognona , Aram Gassamaa , Jonas Guthoffa , Victor Pouliquenb a The World Bank Group, Development Economics Impact Evaluation Unit (DECIE) b Paris School of Economics Abstract Warehouse Receipt Financing Systems (WRS) are financial arrangements which al- low farmers to store their agricultural production in a certified warehouse in ex- change for a warehousing receipt that can serve as a collateral for receiving credit in a formal financial institution (bank or micro-finance institution). WRS have received increasing attention in recent years as a way to release capital constraints for farmers during the post-harvest period and allow them to take advantage of potential price arbitrage opportunities, while reducing post-harvest losses, and thereby increasing farm income. Given the limited available evidence on the effectiveness of WRS, we embedded an experimental impact evaluation study in a pilot WRS project in the Senegal rice sector, to shed light on how smallholder farmers might benefit. In a sample of 1079 rice producers, of which 363 were offered access to a WRS, we ob- served a very low take-up (2%), which compromised the opportunity to uncover the impacts of WRS. We therefore focus on the reasons for non participation and find that large transaction costs, limited availability of marketable surplus, and limited market arbitrage opportunities in the rice sector, have reduced the potential bene- fits from participation and affected farmers’ decision to use the WRS. The findings suggest that the implementation of effective WRS warrants a careful consideration of costs factors and expected price arbitrage opportunities to ensure farmers would benefit. 1 1. Introduction Lack of access to credit and storage infrastructures considerably limits farmers’ ability to take advantage of interseasonal variations in commodity prices in rural areas of Sub Saharan Africa (SSA). Farmers generally sell most of their production right after harvests, when prices are the lowest, to satisfy immediate cash needs, only to buy again at higher prices later for their own consumption (Stephens and Barrett, 2011). For this reason, it has been hypothesized that access to post harvest credit would offer more flexibility to farmers to decide the timing of sales of their products and earn higher revenues from their farming activities (Burke et al., 2018). But the credit market in rural SSA suffers critical failures due primarily to information asymmetries and risks. The lack of conventional collateral makes it harder for farmers in rural SSA to have access to formal loans from financial institutions. Warehousing Receipts Systems (WRS), whereby farmers can use the products stored in a certified warehouse as collateral for the loans, have gained popularity in recent years, as institutional innovation to solve the above-described failure in rural credit markets (Coulter and Shepherd, 1995). Meanwhile, little is known about the actual impacts of WRS on access to finance and ultimately the welfare of smallholder farmers (Burke et al., 2018, Casaburi et al., 2014). Proponents of WR financing systems stress the potential for releasing the collateral value of agricultural harvests, and yielding the following impacts among others (Onumah (2003);Casaburi et al. (2014)): (i) mobilization of credit to agriculture by creating secure collaterals for the farmer, processor, and traders; (ii) smoothing of market prices by facilitating sales throughout the year rather than just after harvests; (iii) upgrading the standards and transparency of the storage industry since it requires better regulation and inspection; (iv) contributing to lower post-harvest losses due to better storage conditions. On 2 the other hand, the critics argue that it places farmers in a speculator role, which exposes them to increased price risks they may not be able to absorb. We contribute to this debate by embedding an impact evaluation study in a pilot WRS intervention in Senegal. The pilot was led by the Ministry of Com- merce (MoC)1 , with technical assistance from the International Financial Corpora- tion (IFC), focusing on the rice sector in the Senegal River Valley. The study sought to understand the impact of participation in WRS on rice farmers’ access to credit, rice sales and storage, and agricultural income, as well as the factors that influence participation in WRS by smallholders farmers. After offering access to a WRS to a randomly selected group of rice farmers, we observed a very low take-up for WRS, suggesting a low demand for the system. Based on a mix of quantitative, qualitative, and geolocation data, covering farmers sale and storage patterns in the post harvest period, we conclude that high transaction costs due primarily to poor transportation infrastructures, low marketable surpluses, and limited arbitrage opportunity in the rice sector, coalesced to hamper farmers participation in the system. The results align with the discussions in Miranda et al. (2017), and suggest important considera- tions for the effective implementation of WRS as a way to relieve capital constraints for smallholder farmers in developing countries. This paper summarizes the lessons learned from such pilot and aims at providing actionable evidence to inform the implementation of WRS in developing countries. Section 2 describes the pilot WRS in Senegal; section 3 illustrates the theoretical rationale for such an intervention; section 4 details the methods and data collection strategies used in this study. Section 5 summarizes the findings and section 6 con- cludes. 1 MoC=Ministère du Commerce, de la Consommation, du Secteur informel et des PME. 3 2. The WRS pilot in the Senegal River Valley The government of Senegal through the Ministry of Commerce, with technical assistance from the International Financial Corporation (IFC), have been working, since 2014, on a WRS project in the country. The overall objective is to support increased access to finance to value chain actors in the agricultural sector of Senegal, favoring the development of better storage facilities and building capacities of private and public stakeholders. The main activities of this project include (i) creating a legal and regulatory environment for WHR, (ii) provision of technical expertise and assistance to the WRS regulatory unit during start-up phase, (iii) sensitization and training of stakeholders on the WRS, (iv) assistance to the warehousing industry, (v) stakeholder engagement for a warehouse receipts trading platform. This pilot is very important for the government as lessons learned will be used to inform scale up to other value chains and other regions. In addition to supporting the process leading to the adoption of the law2 regu- lating WRS in Senegal, the project has also implemented a pilot WRS intervention, in 2018, during the post harvest period of the dry production season3 , to serve as demonstration for the various actors involved in the chain. As part of the pilot, selected farmers were offered the opportunity to bring their rice production to a des- ignated warehouse4 , and received a warehousing receipt that they could take to a 2 Loi numéro 2017-29 du 14 juillet 2017 portant sur le Système de Récépissé d’Entrepôt de Marchandises au Sénégal. 3 The dry production season, locally called “saison sèche chaude”, goes from January to July and its commercialization lasts generally between July and December. 4 The warehouse was rented specifically for the purpose of this pilot. 4 partner bank.5 At the bank, farmers could use their receipt against access to credit for up to 80% of the value of the product in storage, at a 5% annual interest rate, pro- rated to the duration of the credit. The warehouse was chosen based on its central location, and also because it meets the quality conditions defined by the WRS regu- latory framework adopted by the parliament in 2017. At the warehouse, a collateral manager was operating and was responsible for providing farmers with warehousing receipts after quality control checks, as well as for the appropriate monitoring of the paddy stored to preserve the collateral value (Figure 1). Given the limited capacity of the warehouse compared to the rice farmer popula- tion in the area, there was a selection process implemented as follows: (i) a sensitiza- tion campaign to educate farmers about the WRS and stimulate their interests; (ii) a listing of farmers via the rice farmers’ organizations in the area, during which farm- ers had indicated their interest in participating in the pilot; (iii) a computer-based lottery process to select the farmers who would be given access to the warehouse; (iv) a high frequency data collection starting right after harvest and spanning the whole post harvest period, to capture rice sales, prices, and storage from individual farmers. 3. Conceptual Framework Are rice farmers in the Senegal River Valley feeling forced to sell their production immediately after harvest due primarily to pressing needs for capital? If yes, we would expect that, when offered access to the pilot WRS described above, farmers would take advantage, and delay their sales, so as to benefit from the potential 5 CNCAS = Caisse Nationale du Crédit Agricole du Sénégal partenered with the MoC to support the pilot WRS. 5 Figure 1: Operations at the warehouse (a) Paddy rice on pallets (b) Collateral manager (c) List of eligible farmers (d) Precision balance (e) Mini husker (f) Humidity meter Note : Piling of paddy on pallets on norm, is done by a collateral manager, based and operating at the warehouse, who checks the identity of the rice producer owing the paddy, based on the list provided by the impact evaluation team. The collateral manager then proceeds with the test of paddy quality with the illustrated material and determines if the rice is eligible to be stored or not. The paddy, to reach the minimum quality standards determined for the pilot, shall have a humidity rate between 12 and 16%, an impurity rate of less than 2% and be of the variety Sahel 108 or Sahel 134. time arbitrage during the post-harvest period. As indicated in the Figure 2, which summarizes the theory of change underpinning a WRS project, this should then lead to increased sales price, farm income, and ultimately better standards of living for the farmers. However, if farmers respond differently after being offered access to the system, or if the hypothesized effects do not materialize, this could suggest that, there may be limited arbitrage opportunity to take advantage of in the first place, or that farmers face additional constraints preventing them from taking advantage of such system. A WRS is expected to thrive under some institutional and legal conditions, no- 6 Figure 2: Theory of Change tably the existence of a well-established and transparent regulatory framework, a regulatory agency, the monitoring of warehouses by an independent operator, which is connected to the existence of a performance guarantee, the existence of an in- surance regime and the familiarity of banks with the system. Moreover, some eco- nomic conditions related to the quality of the commodity (storability, grade), market transparency and price volatility, need to be met (Lacroix and Varangis (1996), Gio- vannucci et al. (2000), Höllinger et al. (2009)). It is important to note that our experimental context lacks many of the regulatory features listed above. It was meant to serve only as a demonstration and pilot from which the lessons learned would inform a full scale-up. Miranda et al. (2017) describes an economic model depicting the process through which farmers make the decision to participate or not in a WRS. The model con- 7 cludes that more often than not, WRS system under-deliver on their underlying goals because of high transaction costs and risks transfers, which might reduce the perceived benefits from participation. As a results, existing WRS in Africa have served mostly large traders and processors (see William et al. (2015) for Tanzania and Miranda et al. (2017) for Ghana cases). 4. Research Design 4.1. Research question, design and sampling This pilot is designed as a Randomized Controlled Trial (RCT), to answer the following primary question: What are the impacts of a WRS on access to finance, sales prices, and revenues of rice farmers ? The random selection insures that we can construct similar groups of treatment and control individuals so that the comparison between these groups after the program implementation provides unbiased estimates of the treatment effects (Athey and Imbens, 2017, Khandker et al., 2009). As shown in Figure 3, there was a two step randomization process involving a total of 1066 farmers from 120 producers’ organizations (POs)6 . In the first stage, listed POs were randomly assigned to treatment and control groups. The control group received no intervention which means that the individuals members of those groups were pure control farmers. Meanwhile, in the treatment group, farmers inside each PO were assigned (randomly) to full treatment and contamination groups. The full treatment farmers were officially offered the opportunity to bring their rice to the designated warehouse in exchange for a receipt, while farmers in the contamination 6 Furthermore, 57 private producers were included in the study, thus the overall sample sums up to 1123 producers. Nevertheless, those private producers present very different characteristics from the typical farmers of the sample and their limited number cannot help draw information on their population. Thus, we will keep them out of the following analysis. 8 groups belong to the same POs as the treatment farmers but were not allowed to bring their products to the warehouse. The contamination groups are primarily useful for capturing within POs spillover effects from the treatment. This two-stage randomization process yielded overall 363 farmers in treatment, 242 in contamination groups, and 461 in pure control groups. Figure 3: Study Design Treatment Farmers Treatment PO Contamination Eligible PO Farmers Control PO 4.2. Data The data used for this project were collected in multiple phases including: • A first short baseline survey where key farmers’ characteristics were collected. These allowed us to create more homogeneous strata of farmers within which the randomization was performed • A high frequency data collection for collecting data on storage and sales pat- terns. Between June 2018, right after the harvest of the dry season, to Decem- 9 ber 2018 right before the harvest of the next production season, we contacted farmers physically and via mobile phone, at a monthly average frequency. This allowed us to track sales and storage patterns through the entire post harvest period. • An endline survey, with a multi-module survey instrument capturing a rich set of information including risk and time preferences, kinship pressure and willingness to pay to hide income, etc. Table 1 presents the balancing tests between the different groups at baseline. Most characteristics are balanced across the 3 groups, suggesting that the random- ization process was successful. The table also indicates that the farmers included in the study were relatively small, with an average surface cultivated of 3.4 hectares. The majority of the farmers are male (around 90%), with an average of 49 years old. About 87% of the respondents are head of their household, 27% have at least completed primary school, and 65% earn additional incomes from non agricultural activities. At baseline, about a third of the farmers had declared having had an easy access to some storage infrastructures in the previous 12 months, 15% reported having an individual bank account at the CNCAS, 31% from another bank than CNCAS. Access to credit was available to about 33% of the farmers. 10 Table 1: Baseline characteristics (1) (2) (3) (4) T-test PO - Pure control PO - Contamination PO - Treatment Total Difference Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE N/[Clusters] Mean/SE N/[Clusters] Mean/SE (1)-(2) (1)-(3) (2)-(3) No specific position in the PO (Yes=1, No=0) 441 0.580 235 0.519 350 0.560 1026 0.559 0.061* 0.020 -0.041 [47] (0.027) [67] (0.035) [70] (0.031) [119] (0.019) Age 441 50.370 235 49.209 350 49.697 1026 49.874 1.161 0.672 -0.489 [47] (0.815) [67] (0.940) [70] (0.835) [119] (0.509) Female (Yes=1, No=0) 441 0.086 235 0.111 350 0.097 1026 0.096 -0.024 -0.011 0.013 [47] (0.025) [67] (0.032) [70] (0.030) [119] (0.020) Wolof (Yes=1, No=0) 441 0.615 235 0.570 350 0.577 1026 0.592 0.044 0.037 -0.007 [47] (0.067) [67] (0.059) [70] (0.057) [119] (0.043) Head of household (Yes=1, No=0) 441 0.873 235 0.838 350 0.869 1026 0.864 0.035 0.004 -0.030 [47] (0.029) [67] (0.031) [70] (0.027) [119] (0.019) Has been to primary school (Yes=1, No=0) 441 0.460 235 0.443 350 0.466 1026 0.458 0.018 -0.005 -0.023 [47] (0.046) [67] (0.043) [70] (0.042) [119] (0.029) Has at least completed primary school (Yes=1, No=0) 441 0.268 235 0.247 350 0.271 1026 0.264 0.021 -0.004 -0.025 [47] (0.035) [67] (0.033) [70] (0.029) [119] (0.021) Has been harvesting other thing than rice in the last 12 months (Yes=1, No=0) 441 0.488 235 0.494 350 0.500 1026 0.493 -0.006 -0.012 -0.006 [47] (0.040) [67] (0.046) [70] (0.038) [119] (0.028) Has another source of revenues beside agriculture (Yes=1, No=0) 441 0.653 235 0.681 350 0.643 1026 0.656 -0.028 0.010 0.038 [47] (0.026) [67] (0.030) [70] (0.031) [119] (0.018) Land harvested within PO during dry season 2018(in hectare) 441 2.094 235 1.455 350 1.704 1026 1.815 0.639 0.390 -0.248 [47] (0.393) [67] (0.126) [70] (0.140) [119] (0.184) Land harvested privately during dry season 2018(in hectare) 441 4.151 235 2.551 350 3.171 1026 3.450 1.600 0.980 -0.619 [47] (0.971) [67] (0.718) [70] (0.565) [119] (0.512) Has signed contracts to sell to millers (Yes=1, No=0) 441 0.456 235 0.366 350 0.406 1026 0.418 0.090 0.050 -0.040 [47] (0.048) [67] (0.042) [70] (0.038) [119] (0.029) Has had easily access to a storage infrastructure in the last 12 months (Yes=1, 441 0.286 235 0.391 350 0.366 1026 0.337 -0.106** -0.080* 0.026 [47] (0.029) [67] (0.037) [70] (0.033) [119] (0.022) Has rented some storage infrastructures in the last 12 months (Yes=1, No=0) 441 0.293 235 0.340 350 0.329 1026 0.316 -0.048 -0.036 0.012 [47] (0.039) [67] (0.039) [70] (0.038) [119] (0.025) Has had some important post-harvest losses in the last 12 months (Yes=1, No=0) 441 0.095 235 0.102 350 0.123 1026 0.106 -0.007 -0.028 -0.021 [47] (0.018) [67] (0.022) [70] (0.021) [119] (0.013) Has an individual bank acoount at the CNCAS (Yes=1, No=0) 441 0.190 235 0.115 350 0.143 1026 0.157 0.076* 0.048 -0.028 [47] (0.033) [67] (0.020) [70] (0.023) [119] (0.018) Has an individual bank account in a bank other than CNCAS (Yes=1, No=0) 441 0.315 235 0.311 350 0.323 1026 0.317 0.005 -0.008 -0.012 [47] (0.032) [67] (0.033) [70] (0.030) [119] (0.020) Has benefitted from an individual credit from a bank(Yes=1, No=0) 441 0.338 235 0.306 350 0.337 1026 0.330 0.031 0.001 -0.031 [47] (0.038) [67] (0.035) [70] (0.030) [119] (0.022) Is planning to produce rice during the rainy season 2019 (Yes=1, No=0) 441 0.846 235 0.813 350 0.809 1026 0.826 0.033 0.037 0.004 [47] (0.035) [67] (0.037) [70] (0.038) [119] (0.025) Average surface harvested (in hectares) 441 4.223 235 3.446 350 3.658 1026 3.852 0.777 0.565 -0.212 [47] (0.647) [67] (0.565) [70] (0.404) [119] (0.352) Total quantity in storage before harvest (in kg) 441 1001.696 235 93.885 350 263.297 1026 541.876 907.811** 738.399* -169.412 [47] (514.805) [67] (26.084) [70] (100.393) [119] (227.740) Notes : This Table displays randomization results for participants in the treatment group. The value displayed for t-tests are the differences in the means across the groups. Standard errors are clustered at GIE level. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 11 At the end of the production season, farmers in our sample, harvested on aver- age 25 tons of paddy rice individually (Table 2). During the post-harvest period, we observe that farmers split their rice production mostly between inputs credit re- imbursement7 , as well as labor and other services used during the production and mostly the harvest period. As shown in Figure 4, the largest share of paddy rice production (almost 40% - measured in the pure control group) goes to reimbursing credits to CNCAS and other sources. The marketable surplus, defined as the remain- ing stock of paddy after in-kind reimbursement for factors of production received on credit was estimated at 10 Tons, or 43% of the paddy harvested on average, and was managed during the post-harvest period between households’ own consumption, sales, and storage. With the high frequency surveys, we monitored sales and storage pattern through- out the post harvest period between June 2018 and December 2018. Tables 3 and 4 shows that for all treatment arms, the majority of rice sales were done in August, shortly after harvests were completed; and the key buyers were the informal collec- tors (called banas-banas locally). Banas-banas are mobile and tend to purchase the rice directly at at the farm gate, which allows farmers to avoid having to bare trans- portation costs to sell their production. They are also known to pay immediately, which allow farmers to have access to liquidity for their cash needs. Farmers’ storage patterns show that right after harvest, thus in July and August, the share of rice held in storage by producers was around 20-25% of the harvest while it went down to around 5% by the end of the season (Table 5). 7 All producers belonging to unions in our sample, have received a group credit by CNCAS, through their PO, for financing the production process and at harvest reimburse this credit in nature. Additionally, some farmers tend to borrow some additional amounts or materials from other sources to cover completely the production and harvest operations. 12 13 Table 2: Key farming characteristics (1) (2) (3) (4) PO - Pure control PO - Contamination PO - Treatment Total Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE N/[Clusters] Mean/SE N/[Clusters] Mean/SE Overall quantity harvested (in kg) 438 28616.100 234 21294.902 349 23043.140 1021 25033.217 [47] (4692.272) [67] (2910.071) [70] (2235.855) [119] (2381.818) Yield (in kg/ha) 438 8165.900 234 7596.780 349 7707.157 1021 7878.657 [47] (267.966) [67] (293.708) [70] (298.926) [119] (185.648) Marketable surplus (in kg) 441 12023.067 235 8097.271 350 8873.183 1026 10049.362 [47] (1385.322) [67] (860.998) [70] (728.814) [119] (735.346) Marketable surplus as share of overall harvest(in pct) 438 0.455 234 0.408 349 0.420 1021 0.433 [47] (0.016) [67] (0.019) [70] (0.018) [119] (0.011) Notes : This Table displays key indicators after randomization for participants in the treatment group. The value displayed for t-tests are the differences in the means across the groups. Standard errors are clustered at GIE level. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 7.14% 17.46% 30.89% 0.77% 17.35% 9.00% 1.58% 6.73%1.09% 8.00% CNCAS Credits from other banks Labor Machines Transport Gifts and zakat Sales Losses Stored Other uses Figure 4: Use of paddy rice Table 3: Month of sale (1) (2) (3) (4) T-test PO - Pure control PO - Contamination PO - Treatment Total Difference Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE N/[Clusters] Mean/SE N/[Clusters] Mean/SE (1)-(2) (1)-(3) (2)-(3) Share sold in June 441 0.008 235 0.013 350 0.007 1026 0.008 -0.005 0.001 0.006 [47] (0.003) [67] (0.005) [70] (0.003) [119] (0.002) Share sold in July 441 0.163 235 0.204 350 0.178 1026 0.177 -0.041 -0.015 0.026 [47] (0.017) [67] (0.026) [70] (0.022) [119] (0.014) Share sold in August 441 0.512 235 0.488 350 0.438 1026 0.481 0.025 0.074** 0.050 [47] (0.024) [67] (0.028) [70] (0.029) [119] (0.017) Share sold in September 441 0.106 235 0.091 350 0.115 1026 0.106 0.015 -0.009 -0.024 [47] (0.013) [67] (0.016) [70] (0.017) [119] (0.010) Share sold in October 441 0.053 235 0.051 350 0.058 1026 0.054 0.002 -0.005 -0.007 [47] (0.007) [67] (0.011) [70] (0.009) [119] (0.005) Share sold in November 441 0.050 235 0.030 350 0.040 1026 0.042 0.020** 0.010 -0.010 [47] (0.006) [67] (0.007) [70] (0.007) [119] (0.004) Share sold in December 441 0.004 235 0.001 350 0.005 1026 0.003 0.003 -0.001 -0.004 [47] (0.002) [67] (0.001) [70] (0.003) [119] (0.001) Notes : This Table displays summary statistics by treatement status for producers within unions. The value displayed for t-tests are the differences in the means across the groups. Standard errors are clustered at variable q6. Fixed effects using variable strata_gie_num are included in all estimation regressions. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. Table 4: Buyer type (1) (2) (3) (4) T-test PO - Pure control PO - Contamination PO - Treatment Total Difference Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE N/[Clusters] Mean/SE N/[Clusters] Mean/SE (1)-(2) (1)-(3) (2)-(3) Sold to banas-banas (Yes/No) 441 0.787 235 0.779 350 0.754 1026 0.774 0.008 0.033 0.024 [47] (0.021) [67] (0.035) [70] (0.029) [119] (0.018) Sold to millers (Yes/No) 441 0.306 235 0.238 350 0.214 1026 0.259 0.068 0.092** 0.024 [47] (0.036) [67] (0.033) [70] (0.028) [119] (0.022) Sold to other (Yes/No) 441 0.102 235 0.094 350 0.106 1026 0.101 0.008 -0.004 -0.012 [47] (0.015) [67] (0.021) [70] (0.018) [119] (0.010) Notes : This Table displays summary statistics by treatement status for producers within unions. The value displayed for t-tests are the differences in the means across the groups. Standard errors are clustered at variable q6. Fixed effects using variable strata_gie_num are included in all estimation regressions. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. Table 5: Share of harvest in storage (1) (2) (3) (4) T-test PO - Pure control PO - Contamination PO - Treatment Total Difference Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE N/[Clusters] Mean/SE N/[Clusters] Mean/SE (1)-(2) (1)-(3) (2)-(3) Share of harvest in storage in July 142 0.246 68 0.289 107 0.234 317 0.251 -0.043 0.012 0.055 [36] (0.029) [36] (0.043) [46] (0.034) [87] (0.021) Share of harvest in storage in August 145 0.216 68 0.213 102 0.208 315 0.212 0.003 0.008 0.005 [37] (0.020) [36] (0.029) [46] (0.025) [88] (0.015) Share of harvest in storage in September 391 0.162 213 0.120 325 0.157 929 0.151 0.043*** 0.005 -0.038** [45] (0.012) [62] (0.011) [70] (0.015) [116] (0.008) Share of harvest in storage in October 436 0.107 234 0.077 348 0.097 1018 0.097 0.030*** 0.010 -0.020 [47] (0.010) [67] (0.009) [70] (0.009) [119] (0.006) Share of harvest in storage in November 435 0.067 231 0.045 345 0.061 1011 0.060 0.022** 0.006 -0.016 [47] (0.009) [67] (0.009) [70] (0.008) [119] (0.005) Share of harvest in storage in December 336 0.047 167 0.035 248 0.033 751 0.040 0.011 0.013 0.002 [46] (0.011) [61] (0.008) [67] (0.007) [115] (0.006) Notes : This Table displays summary statistics by treatement status for producers within unions. The value displayed for t-tests are the differences in the means across the groups. Standard errors are clustered at variable q6. Fixed effects using variable strata_gie_num are included in all estimation regressions. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 5. Results The main result from this experiment is that there was a critically low take up by farmers. Out of the 363 farmers assigned to the full treatment group, only 8 used the warehouse to deposit their rice (around 2%). There has been multiple recent attempts by researchers to implement similar experimental designs and evaluate the impacts of WR or related systems, which have yielded mixed take up rates, and never as low as in our case. For example, Casaburi et al. (2014) found a take-up rate of around 25% for inventory credit8 in an experiment conducted in 2011 in Sierra Leone, with palm oil farmers. Burke et al. (2018), tracking farmers for two consecutive years between 2012 and 2015, found a take-up of more than 60% for maize farmers in Kenya. With such a low participation rate, we could not exploit the research design and answer the intended questions by comparing treatment and control individuals. How- ever, we leveraged the large amount of quantitative and qualitative data collected, to explore factors that might have affected farmers’ willingness to participate. First, we looked at key characteristics of participants and compared them with the rest of the sample, with no attempt to draw any inference (see table 6). We noticed that compliers were on average less educated, owned and harvested smaller land surfaces, had a largest share of surplus available from their harvest, had more limited access to storage infrastructures, and were located in a closer distance to the selected warehouse than the average of the sample. Their sales prices were slightly higher than the average in the sample but not sufficiently large to back profitable returns on participation. 8 Inventory credits are very close to WRS conceptually, except that they offer community level storage facilities and they do not allow to trade receipts (Edelman et al., 2015) 16 We then explored further a few alternative hypotheses for the low participation rate, and found that limited arbitrage opportunities in the rice sector, high transac- tion costs, and limited marketable surplus, may have curbed farmers’ participation in the WRS pilot. Table 6: Compliers versus rest of the sample (1) (2) Non compliers Compliers Variable N Mean/SD N Mean/SD Has at least completed primary school (Yes=1, No=0) 1071 0.265 8 0.125 (0.442) (0.354) Has another source of revenues beside agriculture (Yes=1, No=0) 1071 0.652 8 0.750 (0.477) (0.463) Overall land owned (in hectare) 1071 7.420 8 4.925 (16.680) (10.173) Average surface harvested (in hectares) 1071 4.401 8 2.199 (13.606) (3.476) Overall quantity harvested (in kg) 1066 26007.955 8 48071.500 (45907.175) (1.16e+05) Has had easily access to a storage infrastructure in the last 12 months (Yes=1, 1071 0.338 8 0.125 (0.473) (0.354) Average price of sale (FCFA/kg) 930 125.777 7 126.021 (37.419) (8.716) Marketable surplus (in kg) 1071 10311.089 8 13754.940 (13915.257) (17799.892) Marketable surplus as share of overall harvest(in pct) 1066 0.434 8 0.681 (0.266) (0.305) Share of marketable surplus sold(in pct) 1000 0.693 8 0.619 (0.325) (0.384) Distance from warehouse (in km) 1071 29.471 8 17.387 (19.175) (11.027) Notes : This table shows the average characteristics of people having used the WRS vs those not having used it. Limited arbitrage opportunities In a classical storage model, rational farmers are expected to store as long as the discounted expected future price is greater than the current price plus the unit storage costs (Saha and Stroud, 1994). This suggests that if farmers do not expect 17 prices to increase sufficiently during the post-harvest period to justify the costs of bringing their rice to the warehouse and go to the bank to exchange the WR against some credit, they will not have the incentive to participate. We explore this using the high frequency farmers sales price and, as shown in Figure 5, we found that average paddy sales prices varied between 124 and 126 FCFA/kg during the post harvest period (August-December). A simple cost-benefit analysis, considering the transac- tion costs of the 8 farmers who actually participated, and assuming no discounting, revealed that expected future prices should have been at least 133FCFA/kg, to jus- tify participation in the WRS (Figure 6). Gilbert et al. (2017) confirms this, using data from Ghana, Malawi, Uganda, and Tanzania, showing that the rice sector of- fers less post-harvest arbitrage opportunities than other commodities such as maize, fruits, vegetables, etc. They ascribed part of this to the high volumes of international rice imports into African markets. The low participation rates in the WRS seems consistent with a rational expectation by farmers about limited increase in future price trends. Nevertheless, there were still cases of paddy sales during all our survey months, suggesting that some farmers did hold positive storage, but at home. 18 Figure 5: Rice sales prices during post harvest season Figure 6: Cost-Benefit Analysis with different price levels High transaction costs High costs of using the WRS, including transportation costs, handling and storage fees, interest rate, etc., could all affect farmers’ participation in the WRS. GPS data 19 collected on farmers locations indicated that the location of the warehouse rented was relatively far from the key production areas (Figure 7). The median distance between farmers locations and the warehouse was 33.74 km. On the other side, the median distance for producers who participated in the pilot was 17.68 km, which is much lower than the overall sample one. Figure 8 shows that compliers, those who were offered to participate in the WRS and who actually brought their rice to the warehouse, were generally located close to the warehouse. This suggests that distance played a crucial role in the decision to use the WRS. In addition, the rural feeder roads in the pilot area were in bad conditions and, with heavy and continu- ous rain during the post-harvest period, they were absolutely non-accessible. Those conditions may have increased transactions costs so high as to completely deter rice farmers from participating in the WRS. Distribution: Distance from warehouse Median = 33.74 km .03 Share of farmers .01 0 .02 0 50 100 150 200 250 distance in km kernel = epanechnikov, bandwidth = 4.2645 Figure 7: Distance from warehouse 20 Figure 8: Map with compliers versus the whole sample. The central point shows the position of the warehouse, red points show position of compliers, while the small blue dots show the position of other farmers in the sample but who did not deposit their rice Moreover, in an endline survey implemented at the end of the season, we asked producers to self report why they did not participate in the WRS. The figure 9 summarizes the reasons reported by farmers. In addition to the factors reported above, many of them mentioned not having enough marketable surplus. After paying their debts in-kind for inputs and labor received on credit, many farmers were left with just enough paddy for auto-consumption. Therefore the limited quantities they had, could not justify economies of scale large enough to make the participation in the WRS worthwhile. Figure 9: Reasons reported by respondents for non participating in the pilot 6. Conclusion We embedded an experimental design in a WR Financing pilot, in the rice sector, in the Senegal River Valley, to understand the impacts of such system on farmers’ access to finance, storage patterns, sales prices, and income. Due to a low take up of the system (2%), we lacked the statistical power to answer the primary questions 22 of interest. However, we drew important insights for WRS implementations, based on the qualitative and quantitative data collected: • The post harvest arbitrage opportunities justifying WR Financing systems do not always materialize for all types of commodities in all contexts. In cases where post-harvest prices do not increase sufficiently to compensate for the costs of participation, farmers will likely not participate in a WRS. Therefore, a thorough investigation of price trends and forecasts is important for guiding the types of commodities to target with a WRS. • When designing WRS, it is important to take into account the transaction costs faced by the potential users. Large transactions costs arising from remoteness, poor transportation infrastructures, and other fees related to storage and credit access from the bank, increase the costs of participation relative to the benefits and discourage farmers from participating. Having warehouses located near the production areas, for easy access, has the potential to decrease significantly those transaction costs and increase participation by smallholder farmers. • When introducing a WRS in a new context, with farmers who might not all be financially savvy, it is important to put in place a sound communication and financial literacy strategy that not only explains sufficiently the benefits from adopting the system, but also provides enough information to make farmers comfortable and trust the system enough to want to experiment with it. Be- cause a WRS system can appear complex, and even risky to many farmers, a local and trusted focal point who can answer all questions timely should be available and easily accessible by farmers. 23 Andrews, R. and Munro, R. (2007). Building a warehouse receipts program that works for all stakeholders–notes from the field no. 1. Athey, S. and Imbens, G. W. (2017). The state of applied econometrics: Causality and policy evaluation. Journal of Economic Perspectives, 31 (2), 3–32. Barrett, C. B. (2007). Displaced distortions: Financial market failures and seem- ingly inefficient resource allocation in low-income rural communities. Development economics between markets and institutions: Incentives for growth, food security and sustainable use of the environment, pp. 73–86. Besley, T. (1994). How do market failures justify interventions in rural credit markets? The World Bank Research Observer, 9 (1), 27–47. Boltz, M., Marazyan, K. and Villar, P. (2019). Income hiding and informal redistribution: A lab-in-the-field experiment in senegal. Journal of Development Economics, 137, 78–92. Boltz-Laemmel, M. (2013). How can i say no? savings and forced solidarity in the extended family: Evidence from senegal. Tech. rep., Working Paper DIAL. Burke, M., Bergquist, L. F. and Miguel, E. (2018). Sell Low and Buy High: Arbitrage and Local Price Effects in Kenyan Markets*. The Quarterly Journal of Economics, 134 (2), 785–842. Casaburi, L., Glennerster, R., Suri, T. and Kamara, S. (2014). Provid- ing collateral and improving product market access for smallholder farmers. a randomised evaluation of inventory credit in sierra leone. 3ie Impact Evaluation Report, 14. 24 Coulter, J. and Onumah, G. (2002). The role of warehouse receipt systems in enhanced commodity marketing and rural livelihoods in africa. Food policy, 27 (4), 319–337. — and Shepherd, A. (1995). Inventory credit: an approach to developing agricul- tural markets. 120, Food & Agriculture Org. Dorward, A., Kydd, J. and Poulton, C. (1998). Smallholder cash crop produc- tion under market liberalisation: A new institutional economics perspective. CAB Internat. Duflo, E. and Banerjee, A. (2017). Handbook of Field Experiments, vol. 1. Else- vier. —, Glennerster, R. and Kremer, M. (2007). Using randomization in devel- opment economics research: A toolkit. Handbook of development economics, 4, 3895–3962. Dupas, P. and Robinson, J. (2013). Why don’t the poor save more? evidence from health savings experiments. American Economic Review, 103 (4), 1138–71. Edelman, B., Lee, H. L., Mabiso, A., Pauw, K. et al. (2015). Strengthening storage, credit, and food security linkages: The role and potential impact of ware- house receipt systems in Malawi. Tech. rep., International Food Policy Research Institute (IFPRI). Gilbert, C. L., Christiaensen, L. and Kaminski, J. (2017). Food price season- ality in africa: Measurement and extent. Food policy, 67, 119–132. 25 Giné, X. and Karlan, D. S. (2014). Group versus individual liability: Short and long term evidence from philippine microcredit lending groups. Journal of devel- opment Economics, 107, 65–83. Giovannucci, D., Varangis, P. and Larson, D. F. (2000). Warehouse receipts: facilitating credit and commodity markets. Washington DC, World Bank Group. Gondwe, A., Baulch, B. et al. (2017). The case for structured markets in malawi. Policy Note, (29). Höllinger, F., Rutten, L. and Kiriakov, K. (2009). The use of warehouse receipt finance in agriculture in transition countries. FAO Investment Centre. Kelly, V., Adesina, A. A. and Gordon, A. (2003). Expanding access to agri- cultural inputs in africa: a review of recent market development experience. Food Policy, 28 (4), 379–404. Khandker, S., B. Koolwal, G. and Samad, H. (2009). Handbook on impact evaluation: quantitative methods and practices. The World Bank. Lacroix, R. and Varangis, P. (1996). Using warehouse receipts in developing and transition economies. Finance and Development, 33, 36–39. McKenzie, D. and Puerto, S. (2017). Growing markets through business training for female entrepreneurs: A market-level randomized experiment in Kenya. The World Bank. Miranda, M., Mulangu, F., Kemeze, F. H., Kolavalli, S. L. et al. (2018). Does warehouse receipt financing benefit Ghanaian smallholders? Tech. rep., In- ternational Food Policy Research Institute (IFPRI). 26 Miranda, M. J., Mulangu, F. M. and Kemeze, F. H. (2017). Warehouse re- ceipt financing for smallholders in developing countries: Short on logic, long on imagination. Mtanda, A. M. (2015). Farmers’ attitude towards warehouse receipt system in Newala district. Ph.D. thesis, Sokoine University of Agriculture. Onumah, G. (2010). Implementing warehouse receipt systems in africa potential and challenges. In fourth African agricultural markets program policy symposium, organized by the Alliance for Commodity Trade in Eastern and Southern Africa (ACTESA) of the Common Market for Eastern and Southern Africa (COMESA). Onumah, G. E. (2003). Improving access to rural finance through regulated warehouse receipt systems in africa. In United States Agency for International Development–World Council of Credit Unions conference on Paving the Way For- ward for Rural Finance: An International Conference on Best Practices. Wash- ington, DC, June, pp. 2–4. Saha, A. and Stroud, J. (1994). A household model of on-farm storage under price risk. American Journal of Agricultural Economics, 76 (3), 522–534. Squires, M. (2017). Kinship taxation as an impediment to growth: experimental evidence from Kenyan microenterprises. Tech. rep., Working Paper, University of British Columbia. Stephens, E. C. and Barrett, C. B. (2011). Incomplete credit markets and commodity marketing behaviour. Journal of agricultural economics, 62 (1), 1–24. Varangis, P. and Saint-Geours, J. E. A. (2017). Using commodities as collat- 27 eral for finance (commodity-backed finance) (English). Tech. rep., 000263971:Juan Buchenau:jbuchenau@worldbank.org; Washington, D.C. : World Bank Group. William, J., Kaserwa, N. et al. (2015). Improving smallholder farmers access to finance through warehouse receipt system in tanzania. International Journal of Economics and Financial Research, 1 (3), 41–49. 28