Building Market Linkages for Smallholder Farmers in Uganda
How to cite
McIntosh, Craig; Bergquist, Lauren. 2015. "Building Market Linkages for Smallholder Farmers in Uganda." AEA RCT Registry. AEARCTR-0000749. https://doi.org/10.1257/rct.749-2.0
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
Shallow markets in African food grains impose substantial welfare costs; they are a direct contributor to food insecurity and may dampen the incentives to invest in productivity-enhancing inputs to agriculture. We will implement a suite of interventions in a large Randomized Controlled Trial involving 240 markets in Uganda, intended to improve market depth directly. We are training and certifying 210 Commission Agents who will operate in treatment markets using a number of technology-based tools to facilitate trade in maize, beans, bananas, and tomatoes. Agents and buyers can use mobile-phone based SMS to post bids and asks into a new digital trading platform that matches supply with demand and facilitates long-distance trade. We are gathering price data in all study markets and using a number of SMS-based approaches to push out marketing and price information to farmers, traders, and buyers in treatment areas. To investigate the role that contract risk plays in inhibiting trade we will implement a sub-experiment in which buyers are randomly offered a basic or comprehensive guarantee that will reimburse part of their transport costs if a deal falls through. To investigate the role of credit constraints in the supply chain we will randomize access to Cash on Bag credit for a subset of agents. The study aims to shed light on the relative contributions of the major causes of shallow markets that have emerged from the theoretical literature.
Intervention
The main purpose of the program is to overcome search barriers, facilitating new matches and ideally cutting out the middlemen who profligate along the supply chain in the presence of information frictions. To achieve this, we are fielding a multi-pronged intervention:
1. AgriNet CAs: We will train and certify 210 Commission Agents from AgriNet. These are individuals who are already existing agricultural traders in the treatment communities, but as a part of the project they will be given training in how to work inside the AgriNet model, including how to bulk and quality grade, how to interact with AgriNet’s head office to facilitate transactions, how to use Cash on Bag credit, and how to use the new digital trading platform.
2. Kudu trading platform: We will dramatically scale up Kudu, a digital platform written by computer scientists at Makerere University that allows buyers to post bids, sellers (CAs) to post asks, and then performs a back-end matching on crop, price, and distance to identify the three ‘best’ sales lots for each bidder. The system sends these lots by SMS to the buyer; if the buyer is interested in pursuing a purchase then she calls AgriNet who facilitate the transaction and collect a 10% commission on the sale, to be split between the CA, the Network Manager, and AgriNet headquarters.
3. Price Information: We will implement an ‘SMS Blast’ system that will use push SMS to saturate a randomized subset of the treatment areas with information on local prices, national prices, prices currently available through the Kudu platform, and how to contact the CA to arrange for Kudu transactions. The numbers used for this will be harvested from the baseline household and trader surveys, as well as from the business relationships of the CAs themselves. The SMS blast system will be randomized at the individual level among farmers in the household baseline to create variation in exposure to information. There are five information types in the SMS Blast system:
a. ‘Downstream’ price information: price information for your local market, the local hub, and the local superhub. This is expected to be particularly salient for farmers, because it gives them a sense of the margins their local traders are making, but it will be sent to CAs and non-CA traders as well.
b. ‘Random Blast’ price information: each week we will randomly sample five treatment TCs and circulate price information on these markets to the entire treatment set of CAs, traders, and buyers.
c. Kudu Promotion information: advertising messages for Kudu, and information on how to contact the local CA to use it, sent to farmers (in future seasons we will also use this system to encourage non-CA traders to independently register on Kudu).
d. Kudu price information: sent to those receiving the Kudu Promotional information
e. Extra AgriNet price information such as FIT-Uganda prices for markets all across the country, to be sent only to CAs.
Then, there are four types of users who will receive the SMS Blast:
a. CAs: all CAs will receive price information at high frequency through the Blast system.
b. Farmers: 3/4s of farmers in treatment TCs surveyed in the baseline who have mobile phones will receive the SMS Blast. The remaining 1/4th of these farmers will receive no messages through the SMS Blast system, to create household-level variation. Control TC farmers receive no SMS information.
c. Non-CA Traders.
d. Buyers.
4. Transport Guarantees: We will use project funds to introduce two types of randomized guarantees to the Kudu system. These guarantees are aimed at mitigating the impact of contractual risk for buyers, thereby inducing them to accept deals for which the risk would otherwise be too great. The randomization occurs at the level of the Kudu sales match; meaning that each time a buyer enters the system and is matched to sellers there will be an independent randomization as whether that bundle of contracts receives a guarantee. There are two levels of guarantee: the Basic Guarantee, which covers the buyer against any shortfalls in quantity that occur when they arrive in the village to buy, and the Comprehensive Guarantee which additionally covers against shortfalls in quality or attempts by the seller to renegotiate price. The cost of the guarantees during the pilot phase will be covered by a grant from IGC. 38% of sales matches will receive the Basic and 15% the Comprehensive Guarantee. The guarantees reimburse buyer losses according to a distance-related percentage:
a. 0-100 km: 25%
b. 100-200 km: 50%
c. 200-300 km: 75%
d. 300+ km: 100%
Losses are calculated relative to the contract terms as agreed before the buyer set out, and the specific losses covered are as per the type of guarantee. The fact that the reimbursements are a discontinuous function of distance allows us to use Regression Discontinuity to assess the impact of transport risk on buyer willingness to transact.
5. Trader Credit: Cash on Bag (COB) is a service for extending credit to CAs, who in turn may use this credit to pay cash-constrained farmers for 50% of the value of their crop upon bulking with the CAs and 50% upon sale to the buyer. It is designed to allow CAs to bulk a greater volume from a larger number of traders and thereby engage in more trade. We will randomize the allocation of Cash on Bag (COB) credit among the 60 most reliable CAs in the AgriNet system, whereby 30 will randomly receive credit and 30 will not. This randomization will be done at the subcounty level to prevent spillovers from treated to untreated CAs. Credit will be delivered at the beginning of the harvest season to be used for trading throughout that season. The expectation is that CAs can make 6 successive transactions within a season with the COB credit, and their use of it will be overseen by (and guaranteed by) the Network Managers.
The study is focusing on four crops: maize, dried nambale beans, matooke bananas, and tomatoes. The first two are the primary storable food crops in Uganda and are major export crops; matooke bananas are the core subsistence food across most of southern Uganda, and tomatoes are the most heavily traded highly perishable crop in the country.
Experimental design
Selection of Trading Center Sample
We begin sample selection from a group of 110 subcounties in 11 districts of Uganda that were identified by AgriNet as being opportune markets for implementation of the project. These are areas that are generally food surplus and are close enough to national road infrastructure to be feasible for long-distance trade, while not being already well integrated into the national commercial food supply system.
Within these 110 subcounties we did a census of markets, focusing particularly on the full-time ‘trading centers’ (TCs) at which local intermediaries have their warehouses, rather than the weekly ‘markets’ that are typically used by locals for buying perishable foods but are not centers of selling grain harvests. On average we located just over two such trading centers per subcounty.
We then divided these TCs into two types: ‘hub’ markets, which are connected to national trunk infrastructure and serve as the core trading location for their areas, and ‘spokes’ that are local markets trading primarily with their respective hub. We eliminated TCs that lack access to mobile networks (a core component of our interventions), and eliminated spokes for which there was no obvious hub (unless there were fewer than 2 other TCs per perish, in which case these unmapped TCs were kept in our sample), and sampled so that we have no more than two spokes per parish. The final sample consists of 240 TCs that are divided into 20 ‘wheels’, where a wheel consists of a hub and its component spokes.
We drew in 24 additional TCs to be able to collect price data on them; 20 of these are a set of hubs and spokes in districts that are near to the study districts but not directly connected to them by market linkages. These are intended to serve as a ‘pure control’ to check for contamination of study control markets by activities in the treatment. We also drew in the four major national markets so as to have comparable price data in the most important national trading centers for the SMS Blast system; these are Kampala (capital), Kabale (border with Rwanda), Busia (border with Kenya), and Arua (border with DRC & South Sudan). The total number of markets reporting the biweekly Market Survey is thus 264.
Experimental design.
The unit of randomization is the sub-county. In order to create a 2x2 design at the spoke level (is the spoke treated, is the hub treated) we blocked the design by whether the sub-county contains a hub (17%) or not (83%), and we stratified by a sub-county level price index (mean of the z-scores of the prices of each of our four crops at the trading centers in each sub-county).
This creates a design in which half of the hubs are treated and half are not, but there is random variation in the fraction of spokes for each hub that are treated. The final sample then consists of 55 treated sub-counties and 55 control sub-counties. With 210 CAs located in these 55 treated subcounties, on average there are 3.8 CAs per treated sub-county.
Trader and Farmer Survey Sampling.
The 240 study trading centers serve as the Primary Sampling Unit for a set of farming household and trader surveys. On average we conducted 5 trader surveys and 11 households per trading center. Traders were randomly sampled from the universe of individuals who operated as permanent entrepreneurs in each trading center. In addition, we conduct the trader survey with the 210 AgriNet CAs who are participating in the study.
The household survey was conducted as follows. First, for each study subcounty, we listed all the Local Councils (LC1s) in the sub-county. We then divided these into the LC1s that contain a study trading center (which are likely to be more urban) and those that do not (which are rural). We then selected the LC1 containing the trading center, and randomly sampled one of the LC1s that do not contain the trading center. For these two LC1s, we then listed all the households based on administrative records held by the LC1 chairperson, and randomly sampled households from these lists. We imposed two eligibility criteria: the household had to be engaged in agriculture, and had in the previous year to have sold some quantity of any of the four crops included in the study.
Design details
Cited by
No citation data available for this registration yet.