Is the local wheat market a ‘market for lemons’? Certifying the supply of individual wheat farmers in Ethiopia

Abstract

1. Introduction

Agricultural development in low-income countries is an important policy priority. Establishing pro-poor value chains, linking smallholders to input and output markets, is a key element of most strategies seeking to modernise farming. A large literature focuses on value chain development and smallholder inclusion (e.g. Reardon et al., 2009; Swinnen et al., 2015; Devaux et al., 2016; Abate et al., 2018; Bellemare and Lim, 2018). A long-standing concern in this literature is that transaction costs exclude smallholders from markets, for example, because fixed costs erode the profitability of trading small quantities (e.g. World Bank, 2007). Another impediment to market participation is the low quality of smallholder supply—impeding competition with imports and access to high-value international value chains (Prieto et al., 2020).¹ Low quality is explained by the imperfect adoption of proper inputs and practices at the production stage or post-harvest stage.

Problems emanating from high transaction costs and low quality are intimately related. For example, in an effort to reduce transaction costs, agricultural development strategies often involve efforts to ‘bulk’ (and, hence, mix) smallholder output at the local level. Aggregation may occur through producer cooperatives, farmer groups or local traders. However, aggregation of output prior to individual quality assessment implies that incentives for supplying quality are diluted.² Smallholders supply high-quality output if the crop price they receive is based on the quality of their own output—not the average quality level after bulking and mixing the supply of many farmers. If individual quality is not rewarded, producers will under-invest in quality and supply the market with lemons (Akerlof, 1970; Viscusi, 1978).

This expectation is supported by empirical observations. For example, Saenger, Torero and Qaim (2014) demonstrate that quality measurement of individual supply incentivised dairy farmers in Vietnam to produce more milk and milk of higher quality. Bernard et al. (2017) show that individual grading and certification in the onion value chain in Senegal induced smallholders to raise output quality. Treurniet (2021) evaluates the impact of transitioning from group-based quality testing to individual testing in the context of the Indonesian dairy market and also finds positive effects on milk quality supplied by smallholders.

The insight that product grading at the level of individual farmers could be a pathway for agricultural value chain development is now gaining momentum (Saenger, Torero and Qaim, 2014, Abate and Bernard, 2017; Bernard et al., 2017; Ruben, Bekele and Lenjiso, 2017). Moreover, technological advances reduce the cost of quality assessment, so grading the supply of individual smallholders may be financially viable. For third-party grading and certification to develop as a market-based solution, two conditions should be satisfied simultaneously (Swinnen et al., 2015). First, quality assessment and certification should ‘pay’ for smallholders, or the price premium they receive for supplying high-quality output should at least cover the extra production costs due to the adoption of quality-enhancing technologies and practices as well as the costs of certification. Second, assessment and certification should ‘pay’ for entrepreneurs entering this new niche, or the fee they can charge for their service should cover fixed and variable costs. For most commodities, it is unclear whether these conditions are satisfied. The promise of independent quality assessment and certification as a market-mediated approach to create value in rural value chains has so far remained largely untested.

We piloted the introduction of independent quality assessment and certification shops (‘booths’) in local wheat markets in Ethiopia and evaluated how smallholders value this service. Specifically, we set up grading and certification booths in four important wheat-producing areas, using quality standards by the government standard agency.³ We use Becker–DeGroot–Marschak (BDM) auctions to gauge willingness-to-pay (WTP) for grading and certification and certify output of the farmers bidding higher than the strike price. Importantly, we measure WTP for grading twice: (i) when farmers were uninformed about the presence of the booth on the market (surprise presence), so that the quality of wheat supplied is exogenous and (ii) when farmers knew 1 week in advance that the grading and certification service would be available. We consider whether farmers can ‘choose’ output quality through post-harvest crop management (e.g. cleaning and sorting) and look at the effect of crop quality on WTP for grading and certification. We also measure (current) WTP for grading and certification 1 year in the future, using a hypothetical auction design. A 1-year delay enables smallholders to invest in post-harvest crop management as well as to adopt quality-enhancing inputs and practices.

In sum, we hypothesise that a key barrier to the adoption of quality-enhancing inputs and practices is that local markets do not reward quality. This may be addressed by introducing third-party quality certification at the farmer level, transmitting price incentives to individual producers. The objective of this paper is threefold. First, we test whether farmers are aware of the quality of the output they currently supply, and we measure quality using a standard quality grading system. Second, we examine whether farmers can improve the quality of their output if they are incentivised to supply quality through post-harvest crop management (in the short term) and through adoption of quality-enhancing technologies and practices (in the long term). These two objectives directly speak to the conceptualisation of local wheat markets as markets for lemons. Third, we measure smallholder WTP for quality assessment and certification and probe whether certification can emerge as a financially viable service on Ethiopian wheat markets.

We obtain the following results. First, farmers are somewhat but only imperfectly informed about the quality of the wheat they supply. While WTP for grading and certification is positively correlated with own wheat quality—farmers bringing better wheat to the market are willing to pay more for a quality certificate—the level of understanding of quality levels is imperfect. If we provide informal feedback about quality to a (random) sub-sample of farmers who did not ‘win’ in the BDM auction, we find that their future quality is affected. This suggests that the quality signal contains informational value.

Second, farmers can improve the quality of their output. Even in the short term, they are able to respond to (expected) monetary incentives. If farmers have prior knowledge about the presence of the booth, on average, they supply wheat of higher grade and bid higher for certification. This follows from our analysis based on incentive-compatible BDM auction data, focusing on post-harvest management. Moreover, an auxiliary analysis based on hypothetical WTP for grading after the next cropping season suggests that further quality improvements through on-farm management are within reach.

Third, an important caveat to these encouraging findings is that our ‘back-of-the-envelope’ analysis suggests that independent grading and certification may not be a financially viable activity for all local markets. Profit margins for booth owners are small and likely negative for owners servicing small rural markets. This outcome emerges because the majority of smallholders is unwilling to pay the profit-maximising fee for the booth, which may reflect farmers’ uncertainty about the premium they will receive for certified output. One solution would be to implement ‘mobile booths’ that follow traders to multiple markets to increase the size of the client base.

The main contribution to the literature is twofold. First, we document that grading and certification of the supply of individual smallholders attenuates ‘market for lemons’ problems in the context of cereal trading—farmers knowingly supply ‘low quality’ because local bulking and mixing implies they are not incentivised to produce a high-quality crop. This is consistent with earlier evidence in the context of perishables (notably dairy and horticulture), but we are the first to document ‘lemons problems’ in the market for staples. Second, we are the first to probe the financial viability of running a grading and certification booth in remote areas with low farmer densities. The finding that booth owners may not be able to break even has important implications for policy, given the pivotal role of addressing information asymmetries at the local level for agricultural development.

This paper is organised as follows. In Section 2, we introduce wheat markets in Ethiopia and discuss the market failure that we seek to address. In Section 3, we discuss our (experimental) approach and introduce our main hypotheses. Section 4 introduces and discusses patterns in our aggregate data. In Section 5, we analyse individual data using regression analysis. Section 6 contains the outcomes of a simple cost–benefit analysis. The conclusions and discussion ensue.

2. Background and study region

Wheat is among the most important crops grown in Ethiopia, both as a source of food for consumers (representing 14 per cent of total caloric intake) and as a source of income for farmers. There are approximately 5 million wheat producers, of which some 88 per cent are smallholders—accounting for 55 per cent of marketed wheat supply (Shiferaw et al., 2014; Minot et al., 2015). Despite quadrupling of production in the last two decades, the rapidly increasing demand fuelled by a growing population, urbanisation and rising income is increasingly met by imports which now exceed one-third of domestic consumption. High transaction costs and low quality associated with local production contribute to the displacement of domestic producers from national markets (Gebreselassie, Haile and Kalkhul, 2017). Recognising the importance of agriculture for development and food security, the Ethiopian government has focused its efforts on accelerating the commercialisation of smallholder farming and on promoting agricultural transformation through value chain development (e.g. Abate et al., 2018). A key ambition is to promote the production of high-quality wheat.

Government authorities have developed a grading system where quality is based on the rate of impurities (% foreign matter, but also the percentage of other cereals or varieties), the flour extraction rate and a measure of cereal ‘hardness’. Flour companies often use a combination of hard and soft wheat for bread and mainly soft wheat for preparation of cakes. Unit prices vary accordingly. However, key attributes such as the flour extraction rate, moisture content and hardness are not readily observable with the naked eye. Instead, the flour extraction rate is based on a standard test-weight measure. Moisture content and hardness are measured with a specialised moisture meter and pliers, respectively. Each of these dimensions is graded on a three-point scale, and an overall grade is assigned based on the lowest factor approach (Abate and Bernard, 2017). Large millers use the grading system to assess the quality of output using their own measurement instruments, and the Ethiopia Commodity Exchange (ECX) certifies large shipments of wheat. However, certification services are not available for local value chain actors.

How is the wheat value chain organised? Smallholder wheat producers are linked to processors along multiple channels. Three important institutional arrangements are contract farming, farmer cooperatives and spot market trading. Contract farming is an agreement over the production and sale of wheat, often at pre-agreed quality, quantity and price. It enables smallholders to access high-value markets and may help them to obtain quality-enhancing inputs, advice or credit as part of the contract. Marketing cooperatives help smallholders to address quantity, quality and frequency of supply constraints to access high-value markets (and contract farming schemes). Cooperatives enable farmers to aggregate their crop, which lowers transaction costs and diseconomies of scale and enhances bargaining power (but introduces the same dilution of incentives for producing above-average quality as occurs with bulking and mixing on spot markets). The empirical evidence of the impact of contract farming and cooperatives on smallholder production and welfare is discussed by Biggeri et al. (2018).

In this paper, we focus on the third major value chain modality—spot market trading. Farmers make their own production plan and individually sell their produce to traders on weekly markets where farmers and traders come together. In our study area, weekly wheat markets draw hundreds of farmers and dozens of traders from the nearby region. Traders weigh output on the spot, using their own scale, and negotiate prices with farmers. They bulk and mix the supply of individual farmers and sell the aggregated quantity either to a larger trader or to a miller or pasta factory (depending on quantity and wheat type). As mentioned, large processors and millers have technologies to assess wheat quality and vary the price they offer based on the quality of wheat that is supplied. Our own miller level transaction data suggest that millers are willing to pay a premium for high-quality wheat ranging between 6 and 7 per cent (see below). To local traders, however, wheat quality is imperfectly observable, so farmers are not rewarded for producing quality. If individual supply is not graded, farmers treat prices as exogenous and can only increase their revenues by supplying larger volumes (Abate and Bernard, 2017). To overcome information asymmetries, third-party quality assessment could play an important role—creating conditions under which quality investments are properly rewarded.

We distinguish two approaches to improve wheat quality. First, farmers can improve wheat quality through on-farm crop management—using quality-enhancing inputs (choice of variety, fertilisers and agrochemicals). Adopting quality-enhancing inputs and practices increases the flour extraction rate and kernel hardness (e.g. Tayebeh, Abbas and Seyed, 2011; Zörb, Ludewig and Hawkesford, 2018). For instance, nitrogen application affects grain quality if applied in the right quantities at the right time (during the production season). Second, post-harvest management activities (e.g. drying, cleaning and sorting) reduce the rate of impurities and help to further increase the flour extraction rate. Sorting produces output that is more homogeneous, which can therefore be tailored to the needs of clients as millers and bakers demand wheat of uniform quality. Post-harvest quality-enhancing measures are the main focus of the current analysis.

3. The grading and certification intervention

Four pilot booths were introduced during the 2018–2019 wheat cropping season, in June 2019, in collaboration with the Ethiopian Standard Agency, the Ethiopian Agricultural Transformation Agency and the Ethiopian Millers Association. Grading was based on the government’s official grading system also used by millers and the ECX.⁴ We randomly picked two woredas from a sample of high-potential wheat areas adjacent to the main wheat belt. Each certification booth had a unique ID, reported on each graded bag, allowing for random quality controls by the Ethiopian Conformity Assessment Agency. We selected four representative markets at the kebele level and opened our booths next to the main market gate for maximum exposure. The booth was accessible to all farmers and traders visiting the market. As personnel in the booths, we recruited employees of a large Non-Governmental Organisation (NGO), who received special training for this purpose.⁵

During the harvest season, smallholders supply small quantities of wheat to the market multiple times, depending on crop maturation and cash needs. Our initial sample consists of 375 wheat-producing smallholders that visited the market on Day 1 of the experiment. These subjects were locally recruited by a random process as they passed the booth. We organised one session per market, per day. In the experimental session, farmers first received information about certification. We explained the relevant dimensions of wheat quality and explained that (certified) higher-quality wheat was likely to yield a higher price. We informed farmers about the availability of a grading and certification service, based on existing grading standards, which could be accessed depending on choices in the BDM auction. Next, we introduced the BDM mechanism, explained that bidding one’s true WTP is a weakly dominant strategy, provided extensive instructions for the participants and organised a (hypothetical) practice round in which we auctioned off an umbrella.⁶ The BDM design is incentive-compatible, and farmers have no reason to under-state or over-state their true valuation. The reason is that they do not pay their own bid but the (unknown) strike price, in case their own bid exceeds the strike price. Bidding more than one’s value increases the probability of winning the auction while you really would like to lose. Bidding less than one’s value lowers the probability of wining while you really would be willing to pay the strike price and obtain the good or service that is auctioned off. We did not commence with the actual auction until all subjects realised this. For more information about BDM auctions, refer to Bohm, Linden and Sonnegard (1997).

Details of the BDM were as follows. If a farmer’s stated bid in the auction exceeded the so-called strike price, they were entitled to have one 50-kg bag of output graded and certified, paying the strike price as the certification fee.⁷ We used a lottery to randomly pick the strike price from a range of possible prices that ranged from 9 birr to 22.5 birr (in June 2019, USD 1 ≈ 29 birr). For this purpose, we used 10 cards with ascending prices, using steps of 1.5 birr (a uniform distribution). This procedure resulted in a strike price of 15 birr. The strike price was selected in advance and uniformly applied across the four pilot sites for comparability of data across markets. The strike price was written on a piece of paper and put in a closed and sealed envelope, which was placed in full view of the subjects during the bidding stage for maximum integrity. Farmers made their bid in private, and information about bidding was not made public to attenuate concerns about peer effects or herding behaviour. Subjects were informed that the strike price was selected using a lottery but were not told the distribution of possible prices from which the strike price was drawn.

As part of the experiment, we graded the output of all participating farmers—those bidding in excess of the strike price and those bidding less. Of course certificates were only provided to the ‘winners’ of the auction. In addition, we informally informed a random sub-sample of auction ‘losers’ about the quality of their wheat (but their output was not certified). This provided this sub-group with a signal of the quality of their crop. The remaining auction ‘losers’ did not receive any feedback.

Grains were tested according to official standards using standard tools and transferred in standardised 50-kg bags—sealed with a special thread to avoid tampering. Overall grade and dimension-specific quality levels were reported on the label of each certified bag, along with the date of certification and woreda name (indicating origin). Mimicking the label is possible but complex, and among other things involves finding the right bag used for packaging, which carried a special print. Once certified, wheat bags were returned to producers who subsequently sold them to traders, producer organisations or millers on the market. The premium they subsequently earn depends on the quality, the market competition and their own ability to bargain. For this study, we rely on bid amounts by smallholders; we did not collect transaction-level data to gauge the realised premium for high-quality output.⁸

The four test booths were open for 2 days, and we measured WTP for grading and certification on both occasions. On Day 1, the presence of the booth on the market was a surprise to all market participants. The quality of the wheat supplied that day was therefore the ‘normal or regular quality’ that farmers deliver—lacking any incentives for producing quality stemming from grading—and exogenous to the grading service. We refer to the bid for on-the-spot grading of farmer i in the first BDM auction as |$WTP_i^1$| and the grade of his output as |$Q_i^1$|⁠.

After the first BDM experiment, we informed participating smallholders that the booth would again be open 1 week later and that they could again participate in another auction (with a new strike price) to have their output graded. With 1-week notice, smallholders could allocate extra effort to simple post-harvest management to improve the quality of their crop, such as sorting and the removal of impurities (quality cannot be affected by farm practice within the time frame of the experiment). Importantly, we did not provide any information to farmers about what they could do to improve quality, in order to avoid confounding the incentive effect of certification with an information effect. On Day 2, 336 participants returned for the second BDM auction experiment. Denote the bid of farmer i during the second BDM auction as |$WTP_i^2$| and the quality of his wheat as |$Q_i^2$|⁠.⁹

In addition to participating in two BDM auctions, our subjects also participated in two hypothetical auctions, both organised on the first day of the experiment. Specifically, after extracting WTP¹, we asked smallholders about their hypothetical WTP to have their output graded 2 weeks later and their hypothetical WTP to have their output graded after the next cropping season—in Spring 2020. It is evident that the hypothetical nature of these bids may introduce bias (Miller et al., 2011; Schmidt and Bijmolt, 2020), so we refrain from comparing incentivised and hypothetical bids. However, we do compare the two hypothetical bids with each other, which allows us to obtain a rough sense of the expected (discounted) value of the quality improvement that can be obtained by adjusting on-farm management. Denote by H-WTP¹ and H-WTP² the hypothetical WTP for certification with a 2-week or 1-year delay, respectively. The design of our wheat certification study is summarised in Figure 1.

Fig.1.

Outline of the wheat grading intervention.

Open in new tabDownload slide

Comparing average outcomes across BDM auctions, we hypothesise that, with individual grading, farmers are incentivised to improve the quality of the wheat they supply to the market through post-harvest management. Both wheat quality and WTP for certification should be higher on Day 2 (with 1-week notice) than on Day 1 (surprise booth). Farmers can (further) improve the quality of the wheat they bring to the market through on-farm management, so hypothetical WTP for certification 1 year in the future should be greater than hypothetical WTP for certification 2 weeks in the future.

Moreover, we can examine the correlation between the quality of wheat supplied by farmers and their WTP for current grading in the BDM auction. This allows testing whether farmers ‘know’ the quality of the wheat they supply and whether those farmers supplying higher-quality wheat are willing to pay more for grading.

To further examine the extent to which farmers know the quality of their wheat, we also look at Day 2 bidding behaviour by the sub-sample of smallholders who lost the first BDM auction. We hypothesise that farmers are informed about the quality of wheat they bring to the market, so the WTP for grading and certification is positively correlated with the quality of their wheat. As mentioned, a random sub-sample of these farmers received an informal signal about the quality of their output. If smallholders have a perfect understanding of the quality of their crop, then receiving a signal about the quality of their crop after the first auction should not affect the bidding behaviour during the second BDM auction or the quality delivered.

On Day 1, we also collected household data that we include as covariates in the regression models below. This included data on the stated level of trust in local wheat collectors (traders). We asked farmers: ‘Do you trust that the weighing scale that is used by traders to weigh the quantity of wheat that you supply provides an accurate measurement of the quantity of wheat that you are selling?’. We used a 4-point Likert scale (very low trust, low trust, high trust and very high trust) and combined the two top categories into a single binary variable denoted by Trust.¹⁰ Low-trust farmers should be willing to pay more for a certified 50-kg bag than high-trust farmers if they expect that the price they receive for their crop will be determined by the weight on the label (as opposed to the trader’s biased scale). This is an empirical matter.

4. Analysis of aggregate data

On Day 1, we interviewed 375 farmers of which 336 also participated in the Day 2 session. This implies a 10 per cent attrition rate, which is low considering that we do not visit smallholders at their house but tried to meet up at the local market. In Appendix Table A1, we estimate two probit models to show that attrition is not correlated with our baseline socio-economic variables.¹¹ In other words, in terms of observables, the two samples are rather similar. However, we do observe a positive correlation between attrition and the quality of wheat supplied at Day 1. Perhaps high-quality producers visit the market less often because they trade smaller quantities. Non-random attrition implies the sample of participants on Day 2 is likely less representative of the target population than the sample on Day 1, as inclusion in the sample on Day 2 was conditional on participating on Day 1. The fact that high-quality producers are under-represented on Day 2 affects the interpretation of cross-day comparisons of bids and quality (see below).

Our dependent variables and covariates are summarised in Table 1. In our sample, the average smallholder is 40 years old (median age is 38) and some 95 per cent of the smallholders in our sample is male—reflecting the dominant role of men in the marketing of cereals in Ethiopia. Most of them (80.3 per cent) are married, and households are composed of up to 11 individuals. Average land size under wheat cultivation is relatively small. Farm size for half of these farmers is less than 1.75 ha. We note that 70.1 per cent had attained some formal education. In total, 34, 192, 77 and 72 farmers participated in the bidding sessions at booths 1, 2, 3 and 4, respectively.

We observe considerable heterogeneity in wheat quality offered during the first day. This heterogeneity may be due to the fact that some variation in quality can be realised at virtually no cost for particular farmers—those with better soils or access to some level of mechanisation at harvest. In addition, it may be that a small price premium exists on the spot market based on observables (impurities, varietal mix, etc.) that are correlated with other relevant characteristics. Uncertainty about one’s ability to recover this premium may explain why only some farmers engage in quality-enhancing activities. We find the following.

Result 1a: If farmers are informed ex ante about the possibility to grade and certify their individual wheat supply, they use post-harvest management to supply wheat of higher qualityand are willing to pay more for grading (compared to when grading and certification is available ‘by surprise’).

Result 1bFarmers are willing to pay more for individual grading and certification1year in the future (so that post-harvest management and on-farm management can be adjusted) than2weeks in the future (when only post-harvest management can be adjusted).

While 34.1 per cent supplied high-quality wheat on Day 1, this proportion increased to 61.0 per cent on Day 2. Nearly half of the farmer in the panel (48.8 per cent) producing low or medium quality on Day 1 supplied wheat of a higher-quality category on Day 2. T-tests reveal that Q² > Q¹ (p = 0.00). Moreover, WTP² > WTP¹ (WTP² = 26.1 birr and WTP¹ = 20.3 birr; p = 0.00), suggesting that information about the availability of grading crowds in post-harvest management effort, which in turn increases the value of certification. Hence, farmers can improve wheat quality, if they are properly incentivised. Finally, H-WTP² > H-WTP¹ (H-WTP¹ = 22.9 birr and H-WTP² = 30.6; p = 0.00), so the prospect of being able to adjust on-farm management by adopting quality-enhancing inputs and practices further increases the value of certification.

Result 1 suggests that in the absence of grading and certification, farmers bring unsorted and unclean wheat to the market—compromising the average value of the product that is traded and contributing to low prices (based on expectations about average quality). However, providing incentives for quality production by grading at the level of the individual supplier attenuates this lemons problem.

Observe that hypothetical bids for delayed grading are in the same ballpark as the incentivised bids, suggesting hypothetical bias may be limited. Also recall that, due to non-random attrition, there are fewer high-quality wheat producers on Day 2 than on Day 1. This adverse selection effect implies that the estimated incentive effect of grading and certification on product quality is likely a lower bound of the true incentive effect (assuming that there is a positive correlation between the quality of wheat supplied by individual farmers over time).

Consider bidding behaviour in more detail. During the surprise visit on Day 1, some 30 per cent of the subjects bid 0 birr for the service. On Day 2, when the booth was pre-announced, the proportion of zero bids dropped to 8.0 per cent. In Table 2, we summarise bid levels for farmers sorted in different quality categories—distinguishing between Day 1 and Day 2. Farmers who supply higher-quality wheat on average bid more for grading and, compared to Day 1, many more farmers moved from Quality-low or Quality-medium to the Quality-high category. It is also clear that, on Day 2, the average WTP of high-quality farmers is lower than before (35.81 birr vs 43.55 birr). This may reflect learning about realised price premiums on the local spot market on Day 1. It could also be due to the selection effect (different farmers supply Quality-high on Day 2 than on Day 1).

Full demand schedules for the surprise and announced grading and certification service are provided in Figure 2a, based on counting the number of smallholders willing to pay the various fee levels in the incentivised BDM auctions. Figure 2a reveals that most of the effect of announcing the booths occurs for a relatively small group of producers. Approximately 90 farmers are willing to pay 25 birr or more to purchase grading and certification, and especially these high-bidding farmers are willing to pay extra for flexibility provided by the 1-week delay in grading. For most farmers with low WTP, the difference between grading and certifying today versus next week is small. These farmers do not seem to care much about quality—not now and not next week—or do not trust that the certification service will yield them any net benefits.

Fig. 2.

Demand curves for grading and certification. (a) Demand curves from actual WTP valuations. (b) Demand curves from hypothetical WTP valuations.

Open in new tabDownload slide

We repeat this exercise for hypothetical bids. Upon comparing WTP for a short (2-week) and long (1-year) delay in grading, we obtain the demand schedules in Figure 2b. Again, as expected, WTP for the ‘later option’ on average exceeds WTP for the earlier option. We now find that the great majority of farmers is willing to pay more for the delay, as the entire demand curve has shifted to the right. This is consistent with the interpretation that improving wheat quality by adjusting on-farm management (Figure 2b) is within reach for a larger sample of smallholders than increasing quality through (tedious) post-harvest management (Figure 2a).

Finally, consider how bids in the auctions compare to expected price premiums. Many farmers are willing to pay nothing, or very small amounts, for grading. This may reflect the poor quality they supply or an expectation that they will not be able to share in the extra value that is created anyway. However, other farmers are bidding high amounts for grading, even up to 100 birr. According to the Ethiopian Grade Trade Enterprise, the average wheat wholesale price (paid by millers to traders) during the time of the auctions (June 2019) was 1,528 birr for 100 kg of wheat (min 1200 birr, max 1800 birr). Millers transaction data in our study region reveal that millers are willing to pay a premium of 6–7 per cent for high-quality wheat. Hence, the value created through certifying a 50-kg bag of high-quality wheat is approximately 0.07 × 1,528/2 = 54 birr. This sum would have to be divided between the trader and the farmer (and possibly an intermediary broker), depending on their bargaining power.

It therefore appears as if some farmers have overly optimistic expectations with respect to the short-term price premium they can seize—some 20 per cent of the farmers expect to take the full certification rent (or even more than that). This is clearly unrealistic, and we expect farmers to learn about price premiums quickly. WTP should soon converge towards more realistic values, depending on the intensity of competition between traders on local markets. As is evident from Figure 2, while average WTP goes down, such learning and convergence did not fully occur for all farmers during the 1-week interval between Day 1 and Day 2. Perhaps, this reflects that many farmers improved the quality of their produce during this week and optimistically expected to qualify for a higher grade and larger premium on Day 2.¹²

5. Regression results

We estimate model (1) for the two (market) day sessions separately and summarise regression results in Table 3. Columns (1) and (2) explain variation in bids during Day 1 and columns (3)–(5) explain variation in bids during Day 2. Columns (5) and (6) are based on the sub-sample of farmers bidding below the strike price on Day 1—some of which received the informal quality signal, and others did not. In column (5), we study WTP. In column (6), we explain variation in the quality of wheat offered on Day 2 and ask whether receiving the quality signal affects the quality of wheat brought to the market.

Result 2a: WTP for grading and certification is positively correlated with output quality, suggesting that farmers ‘know’ the quality of wheat they supply. This applies both for wheat supply during the surprise visit of the booth (WTP¹) and during the announced grading service (WTP²).

Smallholders supplying higher-quality wheat are willing to pay more for certification. On Day 1, farmers supplying medium-quality wheat are willing to pay 17 birr more than farmers supplying the lowest quality, and smallholders supplying the highest grade are willing to pay 45 birr more than farmers supplying low quality. The latter bid is significantly greater than the bid of farmers supplying medium-quality wheat.

From Result 1, we know that wheat quality on Day 2 is affected by the prospect of grading and therefore endogenous. Coefficients β₁ and β₂ in columns (3)–(5) therefore do not necessarily capture a causal effect of quality on bidding. Nevertheless, the significant association between high-quality wheat and stated bids are consistent with the idea that farmers have a sense of the quality of wheat they are supplying. This is true for the full samples present on Days 1 and 2, but also for the sub-sample of farmers who were not certified on Day 1 (bidding less than the strike price; see column 5).

Table 4 reveals that similar conclusions can be drawn using the hypothetical bids. Farmers supplying higher-quality wheat are willing to pay more for certification.

One comparison warrants additional discussion. First, comparing the coefficients of the quality variables in columns (1) and (2) to those in columns (3) and (4) of Table 3, it appears as if the nature of the relationship between wheat quality and WTP ‘evolves’ over time. Specifically, the additional WTP for grading, relative to low-quality suppliers, is lower on Day 2 for medium- and high-quality wheat suppliers. Indeed, WTP for medium-quality wheat certification on Day 2 does not significantly differ from WTP for low-quality wheat certification (even if the coefficients are positive). As noted earlier, this could reflect learning about price premiums paid by traders on the local wheat market, or a selection effect, as the sample of farmers supplying Quality-medium wheat consists of different farmers across the weeks. It also reflects that low-quality suppliers—the omitted category—are willing to pay more for grading on Day 2 than on Day 1. According to column (1) of Table 2, WTP for these farmers increases from 0.16 to 7.55 birr.¹³ The gap between WTP of Quality-low and Quality-medium suppliers therefore becomes more narrow.

Next, turn to the effect of the informal quality signal. A random sub-sample of auction losers (farmers bidding below the strike price on Day 1) were informed about the grade of their wheat by the booth manager, but their produce was not certified or packaged. Column (5) establishes that such a signal has no effect on bids made on Day 2. However, column (6) reveals that this signal improved the average quality of wheat supplied on Day 2. This suggests many farmers overestimated the quality of their wheat on Day 1 and that the signal carried information value—inducing farmers to invest additional effort in post-harvest care.

Result 2b: Providing feedback to farmers about the quality of wheat supplied on Day 1 improves the quality of wheat supplied on Day 2. The signal has information value.

Taking Results 2a and 2b together suggests that farmers have good but imperfect understanding of the quality of wheat they bring to the market.

Finally, we consider the effect of trust in traders’ weighing scales and WTP for grading and quality production. While more than 40 per cent of the farmers have low trust in fair weighing by traders (Table 1), we find no evidence that this affects either bidding for certification or the production of quality. Column (2) shows that stated trust is not correlated with bids on Day 1, columns (4) and (5) tell us that it also does not matter for WTP on Day 2 and we also fail to document evidence that trust matters for quality (column 6).¹⁴ We expected low-trust farmers to bid more for the grading and certification service as the independent weighting service would make them less reliant on the (biased) scales of traders, but perhaps low-trust farmers are not confident that they are able to prevent cheating by ill-intending traders anyway. We obtain the same result when eliciting hypothetical WTP for future certification (columns 2 and 4, Table 4).

Result 3Trust in fair weighing by traders is not significantly correlated with willingness to pay for grading and certification or the quality of wheat supplied to the market.

6. Costs, benefits and optimal pricing

To what extent can grading and certification emerge as a market-mediated solution to improve the quality of smallholder supply? Using the demand functions estimated above, it is possible to gauge the (financial) viability of ‘local-level’ certification kiosks by computing predicted booth profits defined as the difference between predicted revenues and the sum of fixed, variable and flow costs associated with running a grading and certification booth.

With respect to revenues, we assume booths are local ‘certification monopolists’ that can charge the profit-maximising price, given the demand function they face. We use raw WTP data to numerically compute the profit-maximising price, reading adoption of certification for different price level of the four demand curves provided in Figure 2. We pick the price that maximises profits for the booth and find that a price of (approximately) 50 birr maximises profits for booth owners for all four demand scenarios. Based on reconnaissance studies in the region, we make the following assumptions. The ‘catchment area’ of a local market comprises 1,500 farmers, and each farmer visits the market five times during the 6-month trading season to sell 1 quintal (or 100 kg) of wheat per visit. Each market is organised 25 times per trading season. On an average market day, approximately 1,500 × 5 ÷ 25 = 300 farmers visit the market, selling 30 tons of wheat in total.¹⁵

We consider two cases. First, we assume the booth is immobile and can only serve local customers. Second, we consider the case that booths, like local traders, are mobile. Employees and equipment ‘travel’ from one market to the next. Specifically, wheat markets in the study region are weekly markets, and we assume the booth can service multiple local markets per week in a fixed rotation to increase revenues.¹⁶ We assume that the one-way costs associated with moving between two locations (from one market to the next) amount to USD 15. Servicing two markets involves two trips, back-and-forth, and costs USD 30 per week in terms of transaction costs. Servicing three markets involves three trips or USD 45 per week.

Details about other costs are provided in Appendix Table A2. Variable costs are the costs of the bag, stamps and closing thread provided with every graded bag of wheat. These amount to USD 0.50 (15 birr). Fixed costs are the costs of hardware and equipment (chondrometer, sieve, hardness meter and so on), which we estimate to be 65,250 birr (USD 2,175). Assuming a 3-year depreciation period, annual costs of the ‘hardware’ amounts to 21,750 birr (USD 725). Fixed flow costs are the costs associated with wages for certification agent and assistants, rent and utilities and so on, predicted to equal 120,000 birr (USD 4,000) per year.

We consider all four different demand curves discussed earlier—two based on incentive-compatible WTP and two based on hypothetical WTP. The results based on WTP¹ (the surprise booth) and WTP² likely underestimate true profitability for the booth because farmers were unable to fully improve the quality of their supply (using only post-harvest management). In contrast, the results based on H-WTP² (the hypothetical bid for grading next year) may overestimate true profitability as bidding behaviour by farmers may be inflated due to hypothetical bias. We therefore consider the profit estimates based on our incentive-compatible data as lower bounds of true profitability and the estimate based on hypothetical data for delayed grading as an upper bound. Given the above-mentioned assumptions, we compute profit-maximising prices, profit levels and adoption shares as summarised in Table 5.

As mentioned, profit-maximising prices are relatively high, amounting to 50 birr per bag. The price may represent overly optimistic expectations of farmers about potential price premiums. Given our estimate of the available premium, a price of 50 birr can only be sustained in case of (nearly) complete pass-through of the quality premium to the farmers or in case of competitive markets for wheat. At baseline, we measured 30 active traders per market, which may suggest a fairly competitive market. Dillon and Dambro (2017) reviewed the available evidence about competition on African (staples) markets and also conclude they are fairly competitive. Nevertheless, it is an open question whether this price level can be supported over time, which gives rise to the idea that our estimate may be an overestimate of the financial returns for real kiosks in market equilibrium.

The share of adopters varies between 17 and 27 per cent, depending on the demand curve chosen. The great majority of farmers therefore opts to forego the certification opportunity when the kiosk charges the monopoly price and continues to supply ungraded (low-quality) wheat instead. This important finding suggests considerable heterogeneity in costs associated with raising quality across smallholders. It is evident that low adoption rates reduce booth profitability and threaten viability. Specifically, if 20 per cent of farmers adopt, each certifying his full supply (10 bags), then revenues amount to only 150,000 birr. This sum approximately covers the sum of fixed and flow costs but is insufficient to recover variable costs. In other words, and as also evident from Table 5, certification shops should be mobile, follow traders and service multiple markets in order to recoup the sum of fixed and flow costs.

Result 4: Smallholder demand for grading and certification falls short of covering the full cost of running a grading booth for a monopolistic entrepreneur, unless certification shops are mobile and service multiple markets.

7. Discussion and conclusions

Most African smallholders produce small marketable surpluses, if any. Transaction costs preclude many smallholders from engaging in trade or curb the profitability of entering high-value chains. To reduce transaction costs (and sometimes to improve smallholders’ bargaining position vis-a-vis trade partners), smallholder supply is often ‘bulked and mixed’ at the local level. But bulking and mixing implies basing payments on average quality, rather than individual quality, which invites the well-known lemons problem. An equilibrium eventuates where smallholders supply low-quality output, receive low prices and are unable to invest in productivity- and quality-enhancing practices and inputs. This contributes to the perpetuation of rural poverty. The first contribution of this paper is to document that the lemons problem is not only relevant for high-value perishables from the dairy and horticulture sector, where price premia for high-quality produce are much higher, but also relevant for staples (cereals).

One approach to break this deadlock is the grading and certification of individual supply (and subsequent bulking, distinguishing between different quality levels). The idea of individual grading as a market-mediated solution to promoting farm output quality and encouraging agricultural modernisation is now gaining traction in the dairy and horticulture sector. However, much remains unknown about the design and implementation of such strategies and about the effects on production and marketing. In this paper, we focus on grading and certification of individual farmers trading wheat on the local spot market.

We find that farmers knowingly supply a low-quality crop but rapidly improve the quality of their crop when incentivised to do so. Individual grading and certification affects quality in the context of (African) cereal production. Farmers are heterogeneous in terms of the quality of wheat they supply. In the context we study, key production decisions are ‘sunk’ and farmers can only improve quality by investing in post-harvest management. However, our data also suggest that informing farmers about grading opportunities before production decisions are made will likely invite further quality improvements. We further document that farmers who supply output of higher quality are willing to pay more for certification. This is consistent with theoretical work (e.g. Jovanovic, 1982) and reveals that farmers have a sense of the quality level they are offering. The auxiliary finding that informal quality signals affect subsequent quality, however, suggests that their level of understanding is incomplete.

While individual grading improves the quality and value of smallholder supply, it is not evident that grading schemes will survive as a market-based solution—even if certifiers can operate as profit-maximising monopolists on local wheat markets. This is our second contribution. Margins for certification shops are small and possibly negative. This may be due to the fact that price premiums for graded cereals tend to be smaller than for other commodities such as dairy and horticultural products. We find that demand for certification is elastic over a range of the demand curve, and many smallholders are unwilling (or unable) to pay the profit-maximising fee. This is the case for nearly three quarters of the target population. Since local markets are small and farmers supply small quantities, the certification shop will have to be mobile and serve multiple markets in a rotational fashion.

Unless certification booths are mobile (or grading can be taken up as a new activity in existing shops), market-based grading and certification opportunities are unlikely to emerge for farmers in remote areas. However, this does not imply that there is no future for individual grading in Ethiopian wheat value chains. For example, independent grading may facilitate contract farming by broadening the scope of contracts that can be written (and broadening the range of farmers that can be included in such contracts). Grading can also be taken up as a cooperative activity, enabling cooperatives to reward members based on the quality of their supply. Cooperatives currently bulk and mix individual supply, much like local traders do, presenting equally dulled incentives for quality production to farmers. Finally, individual grading can be offered as a public service or be supported by subsidies for booth owners. The case for public involvement can perhaps be argued if there are strong (intertemporal) externalities, for example, if production of high-quality output sets in motion an agricultural transformation process. In addition, policy support to improve the quality of wheat production may enable Ethiopia to economise on the import of high-quality wheat—saving scarce and valuable foreign exchange.

Acknowledgements

We thank three anonymous referees and seminar participants in Wageningen and at the EUDN PhD student workshop in Passau for their helpful comments and suggestions. Remaining errors are our own. We thank the Policies, Institutions and Markets (PIM) research program run by IFPRI for financial support.

Funding

The authors gratefully acknowledge financial support from the International Food Policy Research Institute (IFPRI) through the Consultative Group for International Agricultural Research (CGIAR)’s Research Program on Policies, Institutions, and Markets (PIM).

Footnotes

References

Appendix