Quality Upgrading and Competition - Theory and Evidence from the Coffee Sector in Uganda

The aim of this project is to identify variation in quality premiums along the supply chain, from downstream exporters to intermediaries to upstream producers, as well as understand the underlying causes and their efficiency and distributional implications. We partner with a large coffee exporter in Uganda (hereafter referred to as “the company”). We start by mapping out the supply chains emanating from this company and observing how quality premiums vary along the chain (more below). We then conduct randomized control trials with both traders and farmers along those chains. We offer both farmers and traders randomized coffee production contracts – which vary not just in whether they are offered at all, but also the quantity and quality – to induce quality-specific demand shifters. For both, we collect respondent’s “minimum willingness to accept” (WTA) the contract via a Becker-DeGroot-Marschak (BDM) mechanism. We will use this experiment to estimate trader costs and markdowns to understand how much of the declining quality premium is due to intermediary market power vs. joint quality production between farmers and traders.

In the experimental component of this paper, we offer randomized coffee contracts. These experiments are run with both traders and farmers. Contracts vary randomly on quality (high vs. low) and quantity (three possible levels). Our contracts are based on the Becker-DeGroot-Marschak (BDM) mechanism. Specifically, we start by asking respondents for their willingness-to-accept (WTA) a contract to sell a randomized quantity and quality of coffee. Then, respondents receive a scratch card with prices in a randomized order and covered by stickers. Respondents choose a sticker to scratch off to reveal the underlying price; if the uncovered price is weakly higher than their willingness-to-accept (WTA), then the coffee producer or intermediary receives a contract for the randomized quantity and quality to be sold at the uncovered price. After the two weeks have elapsed, we return and purchase the coffee if it meets the quantity and quality specified in the contract. This approach has two benefits. First, conditional on respondents’ WTA, whether they win the contract is random (i.e. based on the scratched off price); this induces random variation in whether the contract is offered. Second, we can use the resulting induced exogenous shift in quantity, along with the WTA price, to identify agents’ underlying cost structure, which we allow to vary by quality type and by upstream vs. downstream agents. This will allow us to identify costs and markdowns at different points in the supply chain and by quality type, which will give insight into whether imperfect competition or productive substitutability is driving the declining quality gradient along the supply chain.

2022-02-01

2024-12-31

1. Quality-premium (how price varies by quality) and heterogeneity in this premium along the value chain
2. Quality provision (quantity of each quality type) in response to randomized contracts

These outcomes will be measured separately at each point in the value chain

1. Using our quality sampling survey and samples, we will measure the quality gradient (how prices vary with quality), and heterogeneity in this quality gradient along the supply chain (distance to world markets). Specifically, we will first measure supply chain length by the number of times coffee changes hand until it reaches the export-gate. We will then conduct high-frequency quality sampling visits to the field throughout the harvest season to collect physical samples of coffee that was transacted. Specifically, we ask traders and farmers to keep samples of their main coffee transactions (purchase and sale) throughout the season. For each sample, we ask traders and farmers to record the supplier/buyer of the transaction, price paid/received, and quantities purchased/sold. We will send the physical samples to a lab to measure quality along the following dimensions: moisture level, foreign matter and total defects. These measurers will help us compute an “outturn” measure. This will allow us to have an accurate and objective measure of prices paid/received and quantities purchased/sold by quality level. Next, we will measure quality both continuously (using the outturn index used by our exporting partner) and a binary measure of high vs. low quality using the cutoff used by our exporting partner (defects under 33%, foreign matter under 4%, and moisture under 14.5%).

2. Quality provision: during the season, we will offer both farmers and traders randomized coffee production contracts – which vary not just in whether they are offered at all, but also the quantity and quality. We will combine the offer of these contracts with the high-frequency sampling visits to measure whether traders and farmers’ total production of high and low quality coffee changes during the period when they win the randomized coffee production contracts.

1. Minimum willingness to accept (WTA) the BDM contract
2. Sourcing behavior by traders
3. Processing activities behaviors by farmers and traders

1. Minimum willingness to accept (WTA): we will collect respondents’ minimum willingness-to-accept (WTA) (i.e., the minimum price at which they will accept) the contract to sell a randomized quantity and quality of coffee. Then, respondents receive a scratch card with prices in a randomized order and covered by stickers. Respondents choose a sticker to scratch off to reveal the underlying price; if the uncovered price is weakly higher than their willingness-to-accept (WTA), then the coffee producer or intermediary receives a contract for the randomized quantity and quality to be sold at the uncovered price. This mechanism has been shown to be incentive compatible.
2. To measure sourcing behavior by traders, we will collect information about coffee purchases. We will collect physical samples of the inputs purchased used to measure whether the downstream agent provides high quality by sourcing higher quality from upstream or by upgrading the quality himself. Relatedly, we will also ask respondents about the upgrading activities they have done to the input purchased. We will also ask about identity of the supplier from whom the trader sourced to fulfill the contract, including whether it is an existing supplier or new supplier.
3. We also add questions in the survey to ask the processing activities and relevant assets owned by traders and farmers, to gain insights on the quality investment behaviors along the chain.

This project includes two parts: 1) detailed data collection on the supply chain network and high-frequency transaction level data and 2) randomized contract offers throughout the supply chain.

We begin by mapping the complete supply chain, from the farm gate to export gate, and collect detailed transaction-level data on prices, quantities, and laboratory quality measures along the chain. We document the layers of domestic intermediation and the price-quality gradient as we move up the supply chain.

We explore the role of two mechanisms that may diminish the quality premium upstream: one is markdowns resulting from differential market power by quality, and the second is productivity substitutability, the ability of downstream agents to make quality investments that substitute for those upstream. To differentiate between these two mechanisms and quantify their role in contributing to the diminishing quality gradient, we conduct experiments that offer randomized coffee production contracts throughout the supply chain (to both traders and farmers). These contracts induce quality-specific demand shocks that allow us to separately identify costs from markdowns, as well as understand how agents along the supply chain would respond to varying quality premia on investment in quality. We use the estimated model to examine the efficiency implications and distributional consequences of barriers to entry and productive substitutability, as well as how they interact to affect quality production and surplus distribution along the chain. Finally, we use the model to investigate counterfactual policies to induce quality upgrading.

Supply Chain Mapping and High-Frequency Transaction Level Data

We begin by mapping out the supply chain network to identify the suppliers of a given trader who sells directly to the company (referred to as the “downstream trader”). We will then follow up with the suppliers to identify the suppliers’ suppliers, until we reach the producers. Once we have the baseline network, we will select a single main supply chain for a given downstream trader, which potentially includes other upstream traders and a farmer. This will form our trader sample and farmer sample. For this sample of traders and farmers, we will collect detailed information on quality, prices and sales through surveys and high-frequency field visit. Through high-frequency visits, we will collect transaction-level data on quality, prices and quantities. Specifically, we ask traders and farmers to keep samples of their main coffee transactions throughout the season. For each coffee sample, we ask traders and farmers to record the supplier/buyer of the transaction, price paid/received, and quantities purchased/sold. We will send the physical samples to a lab to measure quality along the following dimensions: moisture level, foreign matter, total defects, and outturn. This will allow us to have an accurate and objective measure of quality. We will also ask what quality improvement activities they have conducted.

This will allow us to document two facts: (1) typical supply chain length and structure; and (2) the quality premium (gap in price for high vs. low quality coffee) along the supply chain. Specifically, we aim to document how long supply chains are and whether quality gradients diminish up the supply chain.

Experimental Contracts

In the experimental component of this paper, we offer randomized coffee contracts. These experiments are run with both traders and farmers. Contracts vary randomly on quality (high vs. low) and quantity (three possible levels). Our contracts are based on the Becker-DeGroot-Marschak (BDM) mechanism. Specifically, we start by asking respondents for their willingness-to-accept (WTA) a contract to sell a randomized quantity and quality of coffee. Then, respondents receive a scratch card with prices in a randomized order and covered by stickers. Respondents choose a sticker to scratch off to reveal the underlying price; if the uncovered price is weakly higher than their willingness-to-accept (WTA), then the coffee producer or intermediary receives a contract for the randomized quantity and quality to be sold at the uncovered price. After the two weeks have elapsed, we return and purchase the coffee if it meets the quantity and quality specified in the contract. This approach has two benefits. First, conditional on respondents’ WTA, whether they win the contract is random (i.e. based on the scratched off price); this induces random variation in whether the contract is offered. Second, we can use the resulting induced exogenous shift in quantity, along with the WTA price, to identify agents’ underlying cost structure, which we allow to vary by quality type and by upstream vs. downstream agents. This will allow us to identify costs and markdowns at different points in the supply chain and by quality type, which will give insight into whether imperfect competition or productive substitutability is driving the declining quality gradient along the supply chain.

We will begin by mapping out the supply chain network for traders who sell directly to the company (the downstream traders). To do so, we call all the downstream traders to administer a short survey to map out their main areas of operation, i.e., main parishes of purchasing and selling coffee. We will also collect their supplier information, and then call their suppliers to ask about the suppliers’ suppliers, until we reach a farmer. Additionally, we collect information on a randomly selected supplier for a quarter of all downstream suppliers, and follow the same procedure of call that randomly selected supplier and asking about the suppliers’ suppliers until we reach a farmer. We also collect data on the physical market centers where traders store their coffee among the primary parishes where the company is active. At each market center, we perform a census listing exercise where we collect information on all traders active in the market centers. From this listing, we randomly select a subset of parishes and add an additional downstream trader to the supply chain mapping exercise.

We then select supply chains to include in the high-frequency transaction level data collection through the season, using both main and randomly selected supply chains. Additionally, we incorporate additional downstream traders selected randomly from the market listing activities, stratified by parish.

All agents (traders and farmers) along selected main supply chains will be in the pool that is potentially eligible for the randomized contracts. The randomization of the experimental contracts will come from the BDM mechanism. Conditional on WTA, the award of the contract is randomized through the randomized price scratched off. For each contract, the quality (high vs. low) and quantity (three levels) is also randomized, with quality and quantity assignment stratified by chain length.

Randomization will occur at the individual respondent level.

No

N/A

We expect to have about 400 respondents (drawn from 200 supply chains) in the experiments.

All agents (traders and farmers) along these selected supply chains will be in the pool that is potentially eligible for the randomized contracts. The randomization of the experimental contracts will come through the BDM mechanism. Conditional on WTA, the award of the contract is randomized based on the randomized price scratched off. For each contract, the quality (high vs. low) and quantity (three levels) is also randomized, with quality and quantity assignment stratified by chain length.