Introduction
The food supply chain connects actors of heterogeneous sizes, ranging from small farms to large multinational retailers. This market structure of the food sector, with its large imbalances in size, has long been a focus for both academics and policy makers. In recent years, policy makers, especially in Europe, but also in the US and Australia, have become increasingly concerned about the competitive conditions and power asymmetries in the food chain, and their negative welfare effects and distributional implications (Deconinck, Reference Deconinck2021; Sexton, Reference Sexton2013; Swinnen, Olper and Vandevelde, Reference Swinnen, Olper and Vandevelde2019). The most meaningful new regulatory intervention to address these concerns is the EU unfair trading practices (UTPs) directive (referred to as the “UTP directive”), which aims to address power imbalances in the agricultural and food supply chain by protecting small producers against UTPs in their relations with larger buyers. With this paper, we aim (1) to offer an updated assessment of the competitive conditions within the EU food sector by presenting and comparing various indicators across Member States and subsectors and (2) to critically examine the conceptual and empirical arguments supporting, as well as the practical implications of, utilizing turnover size as an indicator for supplier-buyer relations deserving legal protection under the UTP directive.
A perfectly competitive market consists of many buyers and sellers and has homogeneous products and perfect information among sellers and buyers about product characteristics and price. Sexton (Reference Sexton2013) argues that very few agricultural markets fulfill these requirements. Market power refers to the negative welfare effect resulting from having only a few buyers (or sellers) in the market, and is described in Bonanno, Russo and Menapace (Reference Bonanno, Russo and Menapace2018) as “the ability of a firm to set and maintain a price above (if selling) or below (if buying) the level that would prevail under perfect competition.” This generates profits for the seller (buyer), but shrinks demand (or supply) in the product (or procurement) markets, leading to welfare losses. Many food and agricultural markets, however, are more complicated than the typical case of one-sided market power, and welfare implications may be less straightforward. Many transactions in the food chain involve some degree of horizontal or vertical coordination – usually implemented through contracting – among firms in the food supply chain (Bonanno, Russo and Menapace, Reference Bonanno, Russo and Menapace2018; Ménard and Valceschini, Reference Ménard and Valceschini2005; Sheldon, Reference Sheldon2017; Swinnen et al., Reference Swinnen2007). Such contractual arrangements may increase overall economic surplus, but how this surplus is shared between the different agents of the supply chain will depend on their relative bargaining power. The concept of bargaining power refers to the power of obtaining concessions from another party by threatening to impose a cost, withdraw a benefit, or by the threat of disagreement, if the party does not grant the concession (Bonanno, Russo and Menapace, Reference Bonanno, Russo and Menapace2018).Footnote 1 For instance, when retail alliances and large food processing firms negotiate contracts, typically both parties in the negotiation have some level of market power, resulting in tough price negotiations (Colen et al., Reference Colen, Bouamra-Mechemache, Daskalova and Nes2020). Moreover, negotiations may extend beyond the aspects of price and quantity of traded goods. They may also shape other contractual terms governing the relationship between actors in the supply chain, including quality standards, payment conditions, or handling of unsold products. While bargaining power does not necessarily lead to a reduction in demand or supply and the resulting overall welfare losses, both market power and bargaining power can have distributional consequences, by redirecting transaction surplus towards the ’stronger’ party or imposing risks on the ’weaker’ party.Footnote 2
The distribution of turnover size along the food chain has often been described by an “hourglass” shape, with numerous small farmers supplying a limited number of large processors, wholesales, and retailers, who, in turn, sell to a large number of consumers (Deconinck, Reference Deconinck2021). Vertical buyer-supplier relationships in the food and agricultural sector thus often feature actors of varying sizes, raising concerns about imbalances in bargaining power that may lead to UTPs. The UTP directive, which was adopted by the European Parliament and the Council in 2019 (European Commission, 2019), bans 16 UTPsFootnote 3 in business-to-business relationships between suppliers and buyers of asymmetric size. The directive applies a stepwise approach, using the relative turnover size of seller and buyer as a proxy for potential power imbalances. In particular, a buyer-supplier relationship is subject to the directive if the buyer is in a greater turnover size class than its supplierFootnote 4 , as illustrated in Figure 1.
Figure 1. UTP directive thresholds for turnover. Source: European Commission (2019).
While larger firms are often intuitively associated with more bargaining power – both in the input and the output market – (relative) turnover size is not typically employed to identify problematic market structures or power asymmetries. We begin by examining the economic justifications for the more commonly used indicators (market concentration indicators and markups), before exploring the arguments for using turnover size as an indicator of competitive conditions and imbalances of power, as done in the UTP directive.
Two main market concentration indicators are used to measure the extent to which markets are concentrated in the hands of a few firms to proxy for the competition intensity in a given market (Cavalleri et al., Reference Cavalleri, Eliet, McAdam, Petroulakis, Cristina Soares and Vansteenkiste2019): the concentration ratio (CR
${_q}$
) and the Herfindahl-Hirschman index (HHI). The CR
${_q}$
is the combined market share of the q largest firms in the industry – often considered as either the 4 or 5 largest firms (CR4 or CR5). The HHI sums the squared market shares of all the firms within the same sector.
When a firm has a substantial share of total market turnover, it may be able to restrict total supply in the market and set the sales price above the competitive level, i.e., the firm has market power. More generally, concentration may lead to higher prices, but also entry barriers, less competition, and lower productivity (Covarrubias, Gutiérrez and Philippon, Reference Covarrubias, Gutiérrez and Philippon2020). However, in certain situations, concentration indicators may not provide a good prediction of competitive conditions. Larger market concentration may be a response to higher fixed costs, without implications for market power (e.g., Dong, Balagtas and Byrne, Reference Dong, Balagtas and Byrne2023; Watson and Winfree, Reference Watson and Winfree2022). Also, in markets where an entering firm challenges an incumbent firm, or where technological advances lead to scale advantages, the competition may be fierce, even though the CR
${_q}$
within the market is high (Sheldon, Reference Sheldon2017). An example is the highly concentrated German food retail market, where the entrance of discounters has led to strong competitive pressure and lower consumer prices (Cleeren et al., Reference Cleeren, Verboven, Dekimpe and Gielens2010).
Moreover, a concentration index depends on a correctly specified relevant market, which is not obvious in the case of agricultural markets (Sexton, Reference Sexton2013): for example, agricultural products tend to be bulky and perishable, which imposes geographical restrictions on farmers’ ability to sell their products, resulting in a relevant market that is considerably smaller than what is implied by the country-level concentration indicators. On the other hand, some products may be sold to foreign buyers as well, resulting in the relevant market going beyond the domestic market. Such a misspecification of the relevant market may thus result in concentration indicators that underestimate or overestimate effective market power.
Although concentration levels do not directly measure bargaining power, one could argue that firms with a larger market share are more difficult for the trading partners to be replaced by an alternative buyer/supplier in case negotiations break down. This may result in a stronger position to negotiate prices or other contractual elements, even though many other, and likely more important, factors will also influence its bargaining position. Thus, in more concentrated sectors, bargaining power vis-a-vis other trading partners could be, but is not necessarily, larger.
In addition to concentration indicators, markups are commonly used to capture potential market power in the supply chain. Instead of reflecting the market structure, markup estimations provide an estimate of a firm’s competitive conduct. Markups estimate the wedge between a firm’s output price and marginal cost, defined as
${\mu _i} = {{{P_i}} \over {M{C_i}}}$
. As a perfectly competitive firm is assumed to set the price equal to marginal cost, any deviation from this practice by the firm may be considered a sign of market power.
Although markups are widely used (Cavalleri et al., Reference Cavalleri, Eliet, McAdam, Petroulakis, Cristina Soares and Vansteenkiste2019), the suitability of markups as a measure of market power can also be challenged. Differences in markups across firms or sectors can be sensitive to differences in technology, demand, production functions, or cost structures. For instance, an increase in markups may result from productivity gains rather than from increased market power (Peltzman, Reference Peltzman2022) or may be due to a higher markup being needed to cover larger fixed or sunk costs (De Loecker, Eeckhout and Unger, Reference De Loecker, Eeckhout and Unger2020). Yet, higher sunk costs may also deter new entrants, potentially reducing competition and allowing markups to remain high. In cases of vertical contracting, higher markups may also reflect the outcome of a negotiation process in which the firm was able to capture a larger share of the surplus, thereby gaining bargaining power. Moreover, markups are not directly obtained, and the possible role of non-random measurement error in the data used to estimate markups must be taken into account. Finally, it should be noted that markups aim to measure competitive conduct of suppliers in the output market. To assess the power of buyers in the procurement market, i.e., vis-a-vis upstream suppliers, the corresponding indicator would be the input markdown, which measures the ability to set the input price below the marginal value product of that input. Despite recent advances in the estimation of markdowns (Avignon and Guigue, Reference Avignon and Guigue2022), our data do not provide the necessary input data to calculate markdowns, which is why we focus on markups only and refrain from conclusions on buyer power in the upstream food supply chain.
Turnover size is not frequently used to measure competitive conditions, and the economic arguments for absolute firm size being connected to larger market power or stronger bargaining positions are limited. First, whether a large firm has market power depends not on its absolute size, but instead on its size relative to the overall market, i.e., on the firm’s market share. Several large competitors in a market with a large overall turnover will likely provide a more competitive environment compared to just two small companies covering the majority of an overall small market. Additionally, average turnover size may be higher in sectors with larger-scale economies, which would naturally lead to higher levels of market concentration, and possibly – but not necessarily – to lower competition, as argued earlier. In sectors with bulky and perishable products, such as dairy, larger buyers may be able to cover a larger catchment area, engage in stronger spatial price discrimination, and obtain larger markups than competing firms (Koppenberg and Hirsch, Reference Koppenberg and Hirsch2022). Possibly, firms with a larger turnover have a stronger bargaining position. When large processors and large retailers face each other, size may indeed be important in countervailing the market power of the trading partner (Sexton and Xia, Reference Sexton and Xia2018), possibly resulting in larger markups for larger companies. However, the role of size likely depends on the overall market size, making market share the more relevant measure, and other characteristics (e.g., the quality of supplied goods, specificity of investments, location of alternative trading partners) may play an equally or more important role. Small firms can very well exert bargaining power through a hold-up effect, without having an impact on the overall market. For example, the threat of side-selling or technology diversion by small farmers is real (Kuijpers and Swinnen, Reference Kuijpers and Swinnen2016), and small firms focusing on niche markets or local products may have a stronger bargaining position and high markups compared to their larger competitors (Bonnet and Bouamra-Mechemache, Reference Bonnet and Bouamra-Mechemache2016; Jensen et al., Reference Jensen, Christensen, Denver, Ditlevsen, Lassen and Teuber2019; Koppenberg and Hirsch, Reference Koppenberg and Hirsch2022). Thus, while concerns around market concentration or abuse of power in the supply chain are often targeted at large firms, the theoretical arguments for using turnover size as an indicator are not straightforward. This obviously does not exclude that other considerations may have played a role in using the turnover classes upon shaping the policyFootnote 5 , as we will briefly reflect on in the conclusions section.
In this paper, we provide a comprehensive and updated overview of the competitive conditions in a large number of EU food processing and retailing subsectors. By comparing different indicators, we critically evaluate the use of turnover size classes to identify potential uncompetitive behaviors or abuse of power, as stipulated in the UTP directive. Our contribution is threefold.
First, we contribute to the literature by providing updated numbers for the HHI and markups across various markets and product categories. Due to differences in data availability, we focus on 7 European Member States for the markups and on 10 European Member States for the concentration index. Studies suggest that CR
${_q}$
in European food processing were already high in the mid-1990s (McCorriston, Reference McCorriston2014) and are unlikely to have fallen since then (Sheldon, Reference Sheldon2017). A study from the European Commission (2014) provides an overview of retail and supplier concentration in the food sector of 14 EU Member States in 2012, using sales data from Planet Retail and Euromonitor. The report finds that, only for Germany, the average HHI over the 23 product categories considered indicates a competitive procurement market. Several papers have calculated markups for a specific country or agri-food sector (e.g., Grau and Hockmann, Reference Grau and Hockmann2018; Kaditi, Reference Kaditi2013; Koppenberg and Hirsch, Reference Koppenberg and Hirsch2022; Richards, Bonnet and Bouamra-Mechemache, Reference Richards, Bonnet and Bouamra-Mechemache2018). The study that comes closest to our paper is the report by Čechura, Hockmann and Kroupová (2014), who use Orbis data to calculate markups and markdowns for four food manufacturing industries in 24 EU countries for the years 2003–2012.Footnote
6
They find EU average markups of 12% for the dairy sector, 11% for the fruit and vegetable sector, 10% for the milling sector, and 9% for the slaughtering sector.Footnote
7
Second, our study contributes by comparing different indicators, each capturing different aspects of the competitive conditions or the power relations within the chain (Dong, Balagtas and Byrne, Reference Dong, Balagtas and Byrne2023; Sheldon, Reference Sheldon2017). Given the broader sector-level focus of our analysis and the type of data we have available, Berry, Gaynor and Morton (Reference Berry, Gaynor and Morton2019) warn against a causal analysis, given that the complex relations between indicators cannot be convincingly identified. Instead, in line with Covarrubias, Gutiérrez and Philippon (Reference Covarrubias, Gutiérrez and Philippon2020) and Berry, Gaynor and Morton (Reference Berry, Gaynor and Morton2019), we compare the arguments for the different indicators and their shortcomings. We then evaluate the resulting expected correlations between indicators empirically. In this way, we provide a comprehensive picture of the different indicators and shed light on their interpretation and meaning for policy makers.
Finally, we contribute by analyzing the implications of these results for the UTP directive. Despite the rather intense policy focus on UTPs (Swinnen, Olper and Vandevelde, Reference Swinnen, Olper and Vandevelde2019), the related academic literature is relatively scarce and focuses on the occurrence and determinants of UTPs (Di Marcantonio et al., Reference Di Marcantonio, Havari, Colen and Ciaian2022; Di Marcantonio, Ciaian and Castellanos, Reference Di Marcantonio, Ciaian and Castellanos2018; Fałkowski et al., Reference Fakowski, Ménard, Sexton, Swinnen, Vandevelde and Ciaian2017; Russo et al., Reference Russo, Menapace, Pokrivcak, Sorrentino and Swinnen2020). We examine whether the UTP directive, by using turnover size, uses a suitable proxy (compared to more commonly used indicators of competitive conditions) and captures those buyer-seller relationships where farmers may face power imbalances and UTPs are more likely to occur.
The remainder of the paper is structured as follows: The section “Data and methods” discusses the database used for the calculations, followed by a description of the econometric methods used to estimate the markups and the correlation between the indicators. The “Results” section presents the results and the “Conclusions” section summarizes the main findings.
Data and methods
In this section, we first discuss the database used in the analysis, including the data selection, and representativeness. We then discuss the methodology used to calculate size distributions, concentration indicators, and to estimate markups.
The Orbis database
The Orbis database is a commercial database provided by the Bureau van Dijk Electronic Publishing (BvD) covering administrative data from companies (Kalemli-Ozcan et al., Reference Kalemli-Ozcan, Sorensen, Villegas-Sanchez, Volosovych and Yesiltas2015). The database contains firm-level information on company type, sectors they operate in, and financial and balance sheet information collected by local authorities to meetlegal requirements. For the purposes of this study, only firms that reported operating in the food sector (in either retail, wholesale, or manufacturing) are used.Footnote 8 The coverage of the database depends on the reporting rules of the local authority. Due to variations in reporting requirements among the EU Member States, the coverage of firms by the Orbis database varies, and – in a majority of the Member States – it only contains information on a subset of the firms within the economy.Footnote 9 As a result, we restrict the analysis to Member States that have sufficient coverage in terms of number of firms and total turnover covered in Eurostat. Different information is needed to calculate various indicators of market power. Only the turnover for each firm is needed to calculate the HHI. For estimating markups, however, information on firm-level expenditures on capital, employment, and material costs is needed as well, resulting in a smaller number of Member States included in the markup estimations.Footnote 10 The appendix includes details on the criteria for selecting Member States. To further examine the representativeness of the Orbis database, Table A1 in the Appendix compares the coverage of data on the number of firms and total turnover in a sector in the Orbis database with the coverage in Eurostat and shows the number of years available per country.
Turnover size and calculation of concentration indicators
The turnover indicator is taken directly from the Orbis database. We use the HHI as an indicator for concentration, which is given by:
$$HHI = \mathop \sum \limits_{i = 1}^N MSh_i^2$$
In Equation (1),
$MS{h_i}$
is the market share of firm i and is defined as
$MS{h_i} = {{Turnove{r_i}} \over {\mathop \sum \nolimits_i^N Turnove{r_i}}}$
. In our calculation, both the turnover of the specific firm and the industry turnover used to calculate the market share are from the Orbis database.Footnote
11
The index is commonly used by economists and competition authorities (Deconinck, Reference Deconinck2021). The U.S. Department of Justice and the Federal Trade Commission (2010) defines a market with an HHI below 1,500 as an unconcentrated market, a market with an HHI between 1,500 and 2,500 as moderately concentrated, and a market with an HHI above 2,500 as highly concentrated.Footnote
12
Markups estimation
The calculation of markups follows the method developed by De Loecker and Warzynski (Reference De Loecker and Warzynski2012) and estimates production functions using the approach by Ackerberg, Caves and Frazer (Reference Ackerberg, Caves and Frazer2015). De Loecker and Warzynski (Reference De Loecker and Warzynski2012) derive the relationship between markups (
$\mu $
), share of expenditures on input
$x$
in total sales (
${\alpha ^x}$
), and the output elasticity with respect to input
$x$
(
${\theta ^x}$
) from a standard cost minimization problem, as follows:
$$\mu = {\theta ^x}{({\alpha ^x})^{ - 1}}$$
The expenditure shares can be calculated from the accounting data using turnover and input cost, but the firm’s input elasticity needs to be estimated. In particular, we estimate a translog production function for each subsector and Member State with labor, capital, and material as inputs, as defined by:
$$\begin{gathered}
{y_{it}} = {\beta _l}{l_{it}} + {\beta _k}{k_{it}} + {\beta _m}{m_{it}} + {\beta _{ll}}l_{it}^2 + {\beta _{kk}}k_{it}^2 + {\beta _{mm}}m_{it}^2 + {\beta _{lk}}{k_{it}}{l_{it}} + {\beta _{km}}{m_{it}}{k_{it}} + {\beta _{ml}}{l_{it}}{m_{it}} \\ \!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!+ {\omega _{it}} + {_{it}} \\
\end{gathered}$$
where
${y_{it}}$
is the log of output,
${k_{it}}$
is the log of capital input,
${m_{it}}$
is the log of material input, and
${l_{it}}$
is the log of labor input. The error term in this production function consists of two parts:
${\omega _{it}}$
and
${\epsilon _{it}}$
. The first part of the error term refers to the part of the variation that is observed by the firm – for instance a productivity shock – but unobserved in the data. Such a productivity shock is likely to impact the firm’s input choices, leading to
${\omega _{it}}$
being correlated with the input choices. Various methods use alternative identification strategies to overcome the endogeneity related to the asymmetry between the information available to the firm and the information observed by the data. Notable, Olley and Pakes (Reference Olley and Pakes1996) used the timing of investment decisions to estimate the production function. Levinsohn and Petrin (Reference Levinsohn and Petrin2003) argue that the investment decision may be missing (or non-existing) or bulky for a share of the firms. Instead, they suggest using material input decisions rather than investments, as it is available for a larger proportion of the data set. Our preferred method, developed by Ackerberg, Caves and Frazer (Reference Ackerberg, Caves and Frazer2015), referred to as ACF, builds on Levinsohn and Petrin (Reference Levinsohn and Petrin2003) to use the material input decision and the law of motion of productivity to account for this endogeneity. However, ACF also argues that the method by Levinsohn and Petrin (Reference Levinsohn and Petrin2003) could suffer from collinearity and therefore lack of independent variation to estimate the coefficients and propose a correction to this identification problem. Another advantage of the ACF method is that it allows us to estimate the translog production function instead of a Cobb-Douglas production function. Compared to the Cobb-Douglas production function, the translog production function does not require the assumption of smooth substitution between production factors, making it a more flexible functional form (Rovigatti and Mollisi, Reference Rovigatti and Mollisi2018). In line with recent studies,Footnote
13
we use the ACF with translog production function as our preferred specification. In the appendix, we have included a Figure comparing our preferred specification to two alternative specifications: i) ACF approach with a Cobb-Douglas production function and ii) the approach suggested by Levinsohn and Petrin (Reference Levinsohn and Petrin2003). A detailed explanation of the estimation can be found in the appendix, together with a description of the estimated coefficients.
The coefficients estimated using ACF are used to calculate the output elasticity with respect to material, as defined by:
$$\theta _{it}^m = {\hat \beta _m} + 2{\hat \beta _{mm}}{m_{it}} + {\hat \beta _{lm}}{l_{it}} + {\hat \beta _{km}}{k_{it}}$$
Combining the resulting output elasticity with the expenditure shares (computed according to De Loecker and Warzynski (Reference De Loecker and Warzynski2012)), the final markups are calculated as:
$${\hat \mu _{it}} = \hat \theta _{it}^m{\left( {{{{m_i}t} \over {{{{{\hat y}_{it}}} \over {exp\left( {{\epsilon _{it}}} \right)}}}}} \right)^{ - 1}}$$
Since various subsectors are expected to differ in their production function – for instance, due to differences in technology – the production function is estimated at the most detailed level of disaggregation possible while taking into consideration data limitations. Subsectors corresponding to four-digit NACE codes for the manufacturing of beverages were considered, with some adjustments.Footnote 14 For all other manufacturing subsectors, and for the retail and wholesale sectors, the production functions are estimated at the level of three-digit NACE codes. Markups are not calculated for the manufacturing of tobacco (code 12.0) due to an insufficient number of data points.
Similar to the HHI index, the coverage of the Orbis database may bias the results. Compared to the HHI indicator, the coverage of the firms for the markup estimation is lower and may exclude a higher portion of smaller firms. Although it is not possible to assert the direction of this bias, if we can expect the larger firms to have higher and more stable markups, excluding some of the smaller firms from the estimation may skew the average markup upwards and the standard deviation downwards.Footnote 15
Correlation between indicators
To derive correlations between the different indicators, we estimate the following reduced-form equations:
$$lnHH{I_{ktj}} = {\alpha _1} + {\alpha _2}Ln{(Turnove{r_{top4}})_{ktj}} + {\eta _{kt}} + {\epsilon _{ktj}}$$
$$ln{({\mu _{top4}})_{ktj}} = {\beta _1} + {\beta _2}Ln{(Turnove{r_{top4}})_{ktj}} + {\beta _3}Ln{(TF{P_{top4}})_{ktj}} + {\eta _{kt}} + {\epsilon _{ktj}}$$
$$ln{\mu _{iktj}} = {\gamma _1} + {\gamma _2}Ln{(Turnover)_{iktj}} + {\gamma _3}Ln{(TFP)_{iktj}} + {\eta _{jk}} + {\phi _t} + {\epsilon _{iktj}}$$
$$ln{({\mu _{top4}})_{ktj}} = {\gamma _1} + {\gamma _2}Ln{(HHI)_{ktj}} + {\gamma _3}Ln{(TFP)_{top4}} + {\eta _{kt}} + {\epsilon _{ktj}}$$
In the equations, i is the firm, j is subsector, k is country and t is year. The equations are estimated using ordinary least squares – with country-year fixed effects (
${\eta _{jk}}$
) in Equations 6, 7, and 9, and subsector-country (
${\eta _{jk}}$
) and year (
${\phi _t}$
) fixed effects in Equation 8 – using data for the available years between 2006 and 2017. First, consider the relationship between HHI and turnover (Equation 6). Since the dependent variable, ln(HHI), is defined at subsector level, the turnover level should reflect the representative firm within the subsector. A handful of firms generate a significant portion of the turnover within the subsectors. Accordingly, it is expected that the firms that represent most of the transactions within the subsectors are the largest firms. Therefore, the turnover variable used in the regressions – Ln(turnover
${_{top4}}$
) – is the average turnover of the largest four firms in the subsector.
To illustrate this point further, consider, for example, the retail sector in Spain. In Spain, the four largest retail firms account for more than half of the total turnover generated in this subsector, and these firms are, therefore, expected to carry out most of the procurement in this subsector (i.e., chains like Carrefour, Lidl, or Mercadona). In contrast, the median retail firm in this subsector is most likely a smaller retail store (in 2016, the median retail firm in Spain had a turnover of EUR 422 575), which may not be representative of the firm size of most of the transactions within this subsector, and therefore, we did not choose the median firm as the representative buyer in the subsector. The results are robust by using median turnover and top 10 firms, which are provided in the appendix.
The relation between markups and turnover can be evaluated at firm level or subsector level. To compare markups and turnover across subsectors, we use Equation 7. In line with the equation for HHI, we use, for each subsector, the top-4 turnover and markups. Since both turnover and markups are identified at the firm level, we also assess their correlation at firm level in Equation 8 (i.e., within subsectors). To analyze the within-subsector correlation, we use year-fixed and subsector-Member State fixed effects. The key difference between across-subsector correlation and within-subsector correlation is that the former examines correlations between firms from different subsectors, while the latter explores correlations between firms belonging to the same subsector. As a result, the across-subsector correlation corresponds more closely to the use of relative turnover size classes of buyers vs. sellers in the UTP directive: buyers and sellers are typically firms from different subsectors (e.g., between a grain manufacturing firm and a grain wholesale firm) rather than between firms operating as competitors within the same subsector (e.g., between grain manufacturing firm 1 and grain manufacturing firm 2). The regressions for markups also include the variable, log(TFP), which stands for the total factor productivity of the firm (in Equation 8) or the total factor productivity of the top-4 firms in across-subsector Equation 7. We introduce the TFP variable to control for the possibility that higher markups not only result from higher prices due to market power but also from reduced costs due to higher productivity (Peltzman, Reference Peltzman2022). This variable is estimated along with the markups estimations.Footnote 16
Equation 9 examines the correlation between HHI and markups. As described in the previous estimations in Equations 6 and 7, the estimation is conducted at subsector level and we control for TFP.
Finally, these regressions establish correlation between the indicators, and not causation. Establishing the causal relationship between concentration and markups is complex. For instance, Covarrubias, Gutiérrez and Philippon (Reference Covarrubias, Gutiérrez and Philippon2020) summarize the previous literature and divide concentration into “good” and “bad.” Good concentration occurs when there are increases in the elasticity of substitution or technological change, leading to increasing returns to scale, and lower costs and prices. Bad concentration reflects barrier to entry and reduced competition, and increased prices and markups. The relation with turnover is even less straightforward. Due to the complexity of these relationships, previous literature tends to derive either stylistic facts or derives hypotheses from structural models (Berry, Gaynor and Morton, Reference Berry, Gaynor and Morton2019; Covarrubias, Gutiérrez and Philippon, Reference Covarrubias, Gutiérrez and Philippon2020) and support these facts or hypotheses through estimated correlation. In line with previous literature, we do not attempt to claim causality across our indicators.
Coverage of UTP directive
Finally, we assess to what extent the defined turnover size classes in the UTP directive cover supplier-buyer relations in the EU food sector, focusing on those relations faced by agricultural producers. We examine to what extent the trading relations that farmers engage in are covered by the directive and to what extent these correspond to those that would be considered as possibly problematic as indicated by the commonly used measure of market concentration, the HHI.
First, we narrow our analysis to examine the buyers of agricultural products only. Therefore, we exclude NACE codes that are unlikely to buy directly from farmers.Footnote 17 Second, we assume that the majority of farms have a turnover that is less than the lower cut-off for the UTP directive.Footnote 18
The coverage of the agricultural subsector by the UTP directive is then calculated as follows: since all the farmers are assumed to have a turnover below the lower threshold of the directive, all buyers with a turnover exceeding EUR 2 million must comply with the directive, and, consequently, they are prohibited from engaging in a set of trading practices defined as unfair. The coverage of farmers facing each buyer-subsector is subsequently calculated as the percentage of turnover in this buyer-subsector that is generated by firms exceeding this threshold. For example, for dairy farmers in Spain, the coverage is calculated as the percentage of turnover generated by all Spanish dairy processing firms with a turnover exceeding the lower threshold of EUR 2 million.
For each buyer-subsector and Member State, the coverage of transactions with farmers is then compared to the HHI. This allows us to identify whether some buyer-subsectors exhibit a high level of concentration while having a low percentage of turnover covered by the UTP directive. If so, this could indicate a subsector where potential power imbalances would remain uncovered by the directive. Alternatively, in a reverse case, the results would suggest over-coverage by the directive of a subsector.
When using the HHI to identify concentrated markets, we calculate an industry’s HHI based on each firm’s share in total sales in the output market, instead of each firm’s share in the input market where it is procuring agricultural products, which would be the more relevant measure. Due to lack of information on raw agricultural inputs, we cannot construct HHI on the input market. In this exercise on the coverage of the UTP directive, we thus assume that the HHI on the output market is reflective of the concentration on the input market. We realize that this is a rather strong assumption. Farmers themselves may be able to choose among buyers within different subsectors (e.g., grain processors and animal feed processors), or among domestic and foreign buyers, which would lead the HHI to overestimate concentration among buyers. On the other hand, especially for perishable products, the number of buyers that effectively operate in a farmers’ geographical area may be much smaller, reflecting a much lower procurement market concentration than the output market concentration index may suggest. Analogously to this exercise, we could examine to what extent the UTP directive covers relationships that are indicated as possibly problematic from looking at the markup estimations. Yet, also our markup estimations reflect competitive conditions in the output market, which cannot be assumed to proxy for competitive conduct on their procurement market. The relevant indicator to use would be the markdown on agricultural inputs, which our data do not allow us to estimate. Moreover, the stronger data requirements of the markup estimator would limit the number of Member States included in this analysis, further reducing our coverage of the EU food sector.
Results
Size distribution of firms by food sector and country
Table 1 shows the distribution of the number of firms and the percentage of sector turnover by firm turnover size class (i.e., those related to the UTP directive thresholds as shown in Figure 1), Member State, and sector. For instance, for the retail sector in Bulgaria, the table shows that around 43% of the sector’s turnover is generated by firms with less than EUR 2 million in turnover and these firms account for around 99% of the total number of retail firms in Bulgaria.
Table 1. Size distribution of firms by food sector and country
Source: Authors’ calculations, based on Orbis database (BvD, 2019). The data are for 2016 for all Member States except France, for which the data are from 2014 due to availability.
Regarding the size distribution of firms, there are some similarities across Member States and sectors. For each sector and selected Member State, more than 60% of firms have a turnover of less than EUR 2 million. Furthermore, a low proportion of firms have a turnover greater than EUR 50 million in all Member States and sectors. The wholesale sector is typically composed of larger firms than the retail and manufacturing sectors. Note that Spain and France – and to a lesser extent Sweden – tend to have somewhat larger firms than the rest of the Member States.
While the distribution of the number of firms is highly skewed to the left, towards the smaller size classes, the distribution of the sector turnover over size categories is skewed to the right, towards larger firms. This is the case in most Member States – Bulgaria being an exception – and sectors, especially in the retail sector.
Overall, we can state that most firms in the European food sector are small, but most of the turnover is generated by a few, larger firms.
Concentration indicators
The HHI for each subsector is reported in Table 2 and shows considerable variation across Member States and subsectors.Footnote 19 For the retail and wholesale sectors, only around 11% of the 140 (= 14 subsectors * 10 Member States) subsector–Member State pairs are considered to be moderately concentrated a further 11% are considered to be highly concentrated markets. The remaining 76% are not concentrated. No retail or wholesale subsectors is concentrated (or highly concentrated) in all selected Member States. Tobacco wholesale is considered to be concentrated in many Member States (i.e., in 8 of the 10 selected Member States), followed by coffee and tea (in 5 Member States). The remaining subsectors are concentrated in fewer than 5 selected Member States. Three (out of 14) subsectors – flowers, retail in non-specialized food stores and other/general food – are not concentrated in any of the selected Member States.
Table 2. HHI in the wholesale and retail sectors, by subsector and Member State
Source: Authors’ calculations, based on Orbis database (BvD, 2019). Note: Shaded cells indicate subsectors that are considered concentrated (HHI larger than 1 500). The data are for 2016 for all Member States except France, for which the data are from 2014 due to availability. (*) Wholesale of unmanufactured tobacco; (**) Includes ornamental flowers and bulbs; (***) Ready-to-drink beverages; (****) Wholesale of manufactured tobacco.
According to the HHI for the manufacturing sector, 35% of the 170 subsector–Member State combinations are considered to be highly concentrated markets, 14% are moderately concentrated and the remaining 51% are unconcentrated. Four of the manufacturing subsectors are considered to be concentrated (or highly concentrated) in all of the selected Member States. These subsectors include the manufacturing of sugar, tobacco, beer and malt, and other fermented beverages. At the other end of the spectrum, four of the manufacturing subsectors – animal feed, bakery, other food products, and fruit and vegetables – are not considered to be concentrated in any of the selected Member States. The remaining subsectors – meat, fish, oils and fats, dairy, grain and starch, spirits, wine, cider/other fruit wines and soft drinks/mineral water – are only considered concentrated in some of the selected Member States.
These figures suggest that overall, concentration is higher in the manufacturing sector, although in a few countries, also certain wholesale and retail subsectors are considered to be concentrated.
Markup results
Figure 2 shows the firm-level distribution of the estimated markups for retail, wholesale, and manufacturing at the country level. The distributions are centred around 1 and are slightly skewed to the right, implying that the median markup is positive, but close to break-even. With the exception of Romania, the distribution for manufacturing is considerably flatter than that for the retail and wholesale subsectors, implying greater variation in markups between manufacturing firms than between firms in the retail and wholesale subsectors. This result is further shown in Table 3, which reports the results at the sector and Member State level.
Figure 2. Markups distribution. Source: Authors’ calculations, based on Orbis database (BvD, 2019). Note: The year used in the Figure is 2015 for all countries, except for France and Romania, where the distributions are calculated for the years 2014 and 2013, respectively.
Table 3. Characteristics of markup distributions, by sector and Member State
Source: Authors’ calculations, based on Orbis database (BvD, 2019). Note: The year used in the Figure is 2015 for all countries, except for France and Romania, where the estimates are calculated for the years 2014 and 2013, respectively.
The median retail or wholesale firm earns a markup ranging from 5% to 16%, depending on the Member State, with retail markups being somewhat higher than wholesale markups in most Member States. The median markup in manufacturing is slightly higher, ranging from 10% in Romania to 42% in France. The share of firms with a markup below 1 is in most cases higher in the wholesale sector than in the other sectors. With a few exceptions, we find that the markups of the four largest firms (in terms of turnover) are higher than those of the median firm. Consistent with Figure 2, the mean markups in Table 3 are larger than the median markups, indicating a distribution that is skewed to the right.
A narrower distribution of markups (i.e., a smaller standard deviation) would be expected in settings with more homogeneous firms both between and within subsectors in a given Member State. As shown in Table 3, the largest variation in markups is found among manufacturing firms, which can be attributed to the greater heterogeneity compared to the wholesale and retail sectors. This heterogeneity is further reflected in the range of markups, with the maximum being higher in the manufacturing sector than in the other two sectors.Footnote 20
Empirical correlations between the indicators
The results for the across-subsector correlation analyses are shown in Table 4. First, consider the HHI index and turnover. The across-subsector correlation is negative and statistically significant between the HHI and the turnover (column 1). This implies that even if a subsector has firms with larger turnovers than another subsector, it does not necessarily imply a greater HHI. Rather, it implies the reverse: subsectors with larger firms tend to have a smaller HHI index. In addition, correlating turnover with markups across subsectors provides no evidence of a positive relationship (column 2).Footnote 21 Even if the average turnover of the four largest firms is greater in one subsector than another, it does not mean that these firms have greater markups on average. That is, what constitutes a large firm size (i.e., a large turnover) for one subsector may be small in the context of other subsectors.
Table 4. Correlation between market power indicators
Standard errors are clustered at MS-subsector level.
*** p < 0.01, ** p < 0.05, * p < 0.1.
Figure 3. Relationship between the coverage of UTP directive and HHI. The year used in the Figure is 2016 for all countries, except for France, where the indicators are calculated for the year 2014. This is due to data availability. The red line is set at HHI = 1500. HHI values above 1500 indicate concentrated subsectors.
Column 3 shows the correlation between HHI index and markups. A full equivalency between the three indicators is not expected because they each capture different aspects of market power. While markup reflects more the profitability definition of market power (i.e., firms’ ability to affect market prices), the HHI values (and potentially turnover) measure the relative market shares of firms or the competition intensity, which also captures its ability to influence non-price elements of a contract.
The result of the across-subsector estimation shows a significant positive correlation between the markups of the four largest firms in a subsector and the HHI in that subsector. This implies that more concentrated subsectors also tend to have higher markups for the largest firms, confirming that these two more commonly used indicators cover an at least partially overlapping concept of market power.
The results for the correlation between markups and turnover at firm level are shown in column 4. The findings indicate that within the same subsector, if a firm has a 1% larger turnover than another firm in that subsector, it tends to have a 2.4% greater markup. In other words, when comparing firms within the same Member State and subsector, a higher turnover tends to be associated with a larger markup. As discussed earlier, the within-subsector correlation is less relevant for the UTP directive because it is less likely to capture buyer-seller transactions targeted by the directive. Instead, the across-subsector correlations correspond more closely to the transactions focused on by the UTP directive, which were shown above to be statistically insignificant.Footnote 22
In columns (2)–(4), the variable ln(TFP) is included in the regressions as a measure of productivity. In the regression that uses within-subsector variation – column 4 – it is positive, significant correlation between a firm’s productivity and markups. This is in line with previous literature as it implies that more productive firms tend to have higher markups (Melitz, Reference Melitz2003). In contrast, in the across-sector correlation between markups and productivity is negative. Cavalleri et al. (Reference Cavalleri, Eliet, McAdam, Petroulakis, Cristina Soares and Vansteenkiste2019) discuss across-sector relationship between market power and TFP growth and argue two upside effects: i) market power leads to higher barrier to entry and therefore hamper innovation and productivity and ii) higher markups are consistent with the superstar firms in which highly productive firms draw market share.
Finally, it is worth noting that the R-squared value is relatively low. A low R-squared implies that only a small portion of the variation in the data is explained by the variables included in the regressions, including turnover. This means that most of the variation in Table 4 is attributed to factors other than those included in the variables.
Coverage of the UTPs directive
The previously estimated correlations illustrate that the different indicators are not perfectly overlapping and may capture different aspects of power in the food chain. Moreover, the turnover size categories used in the UTP directive seem not to correlate well with the more commonly used indicators, suggesting that its use may lead to over- and/or under-regulation.
Figure 3 shows the level of coverage of the subsector turnover faced by farmers; i.e., it shows the share of total turnover generated by buyers in that subsector that are above the lowest thresholds set in the UTP directive (i.e., above EUR 2 million). Panel (a) of the Figure plots the coverage level and the level of HHI index. The figure shows that the majority of subsectors that are considered concentrated according to the HHI index have more than 80% of the total turnover generated by companies that are covered by the directive. Only a few subsector-Member State pairs have less than 80% of turnover covered. Looking closer, several of these pairs concern sectors that are small in their respective countries, or that may source their produce largely from farmers abroad (e.g., wine and tobacco manufacturing in Finland). The issue may be more relevant for fish manufacturing in Finland and live animal wholesale in Bulgaria – two pairs with a coverage of around 65% of turnover, while considered concentrated according to the HHI. In these sectors and countries, the UTP directive may leave possible power imbalances faced by fishermen or livestock producers uncovered.
The results in panel (a) are based on the assumption that farmers only sell to one subsector. However, farmers in certain subsectors may sell their products to various subsectors. We therefore further aggregate the subsectors (for instance, fruit and vegetable manufacturing is aggregated with fruit and vegetable wholesale) as a robustness test in panel (b). Unsurprisingly, a higher level of aggregation results in a lower level of concentration, with most of the concentrated buyer-subsectors having more than 90% of turnover created by companies above the lower turnover threshold, and thus 90% of turnover covered by the UTP directive.
Overall, we can conclude that, with a few exceptions, the UTP directive covers the highly concentrated sectors for the agricultural sector quite well, also in smaller Member States. It is important to note that this does not necessarily imply the occurrence of abuse of power or a definitive need for increased protection against unfair practices in the sector.
Figure 3 also reveals that a significant proportion of the subsector-Member State pairs, with which farmers are expected to trade, is covered by the UTP directive, while the concentration index does not indicate problematic levels of concentration (they are below the red line). For example, in 44% of the subsector-country combinations in Figure 3a, more than 80% of total turnover is covered by the UTP directive while the HHI indicates an unconcentrated subsector where unfair practices are expected to be unlikely. Again, this does not mean that problematic conduct is impossible in those cases, as the concentration index does not cover all aspects of power in the supply chain, but the likelihood of it occurring is likely lower. Nevertheless, when the UTP directive restricts practices in unproblematic sectors, this could result in possible over-regulation and related costs.
Finally, it is worthwhile to note that the UTP directive applies to suppliers in the downstream sector as well as the agricultural producers. Ideally, we would conduct the same analysis for downstream suppliers. However, as mentioned earlier, this would necessitate establishing vertical linkages between suppliers and buyers both within and across subsectors – e.g., a flour mill may sell its flour to bakeries, supermarkets, other food manufacturers, etc. These vertical linkages become increasingly more complex in downstream sectors. Therefore, we are not able to perform a similar analysis for these sectors.Footnote 23
Conclusions
This paper has provided an overview of the competitive conditions within the EU food industry using firm-level accounting data from the Orbis database. The paper reviews and calculates possible indicators for market structure and power in the food supply chain. In particular, it presents the distribution of turnover, market concentration (HHI), and markups for several food chain sectors for 10 and 7 Member States, respectively, and analyzes the relationship between these indicators. The focus is specifically on the relation between turnover size – used as a proxy for power imbalances in the EU UTP directive – and more commonly employed indicators of power in the chain (market concentration and markups).
Our findings can be summarized as follows: first, although most of the firms are small, most of the turnover across different food sectors is generated by larger firms. Second, the results reveal a considerable level of concentration in the EU food supply chain, with approximately a quarter of retail and wholesale subsectors and about half of manufacturing subsectors considered concentrated. Third, there is a positive correlation between market concentration and markups at the subsector level, indicating that sectors with higher concentration tend to also have larger markups. Interestingly, no correlation is observed between a subsector’s turnover size – the proxy used in the EU UTP directive – and either of these two commonly used indicators. Fourth, we observe a positive correlation between markups and turnover within subsectors, suggesting that these two indicators capture some similar elements of power variation among firms operating in the same subsector. These results suggest that, to a certain extent, the considered indicators (at least turnover and markups) can be used interchangeably as proxies for power within subsectors but not across subsectors. However, in the context of the UTP directive, across-subsector correlation seems to be more relevant than within-subsector correlation, because buyer-seller transactions targeted by the directive are more likely to occur between firms from different subsectors (e.g., between a grain manufacturing firm and a grain wholesale firm) rather than between firms operating within the same subsector (e.g., between grain manufacturing firm 1 and grain manufacturing firm 2).
Finally, we evaluate whether – by using turnover size classes to categorize supplier-buyer relationships – the EU UTP directive provides adequate coverage to farmers in their relations with downstream buyers. Despite limited theoretical and empirical support for the use of turnover as an indicator of power, we find that – in its practical implementation – most buyers in the highly concentrated food subsectors, would indeed fall under the directive. While being a very imprecise indicator, it has the advantage of being easy and provide predictability in the legal rights and obligations for operators engaging in transactional relationships. For the large majority of EU farmers, the UTP directive can indeed be expected to provide legal protection against UTPs. Nevertheless, there may be a risk of over-regulation and unnecessary (and efficiency-reducing) restrictions of transactional practices, as there are several subsectors that are not considered problematic in terms of market concentration, while most buyers do fall under the UTP directive in their relations with farmers.
It is necessary to be aware that our findings and implications obviously reflect methodological assumptions and data limitations. First, the market power indicators calculated in this paper rely on the Orbis database, which under-represents certain firms (particularly small ones), which may lead to a slight overestimation of the concentration indices and likely also of the median values of the markups. A second caveat of our analysis is that the Orbis database does not provide information on vertical (trade) linkages between sectors and has limited information on procurement and input costs, limiting our ability to calculate markdowns and to analyze in more detail the supplier-buyer relations targeted by the UTP directive. For that reason, we narrow our analysis on the coverage of the UTP directive to buyers of agricultural products only and use HHI in the output market as a proxy for market structure, although this is not the best measure of the vertical relationships and competitive structure relevant to the directive. This approach may have ambiguous implications for the true coverage of the directive because the concentration at the subsector level considered in the analyses may underrepresent power structures faced by some farmers (e.g., those specialized in bulky and perishable products), while potentially overrepresenting power faced by other farmers (e.g., those specialized in tradable products like grains). Even more complicated is tracing the impact of concentration or market power over several stages of the supply chain. For instance, as highlighted by Colen et al. (Reference Colen, Bouamra-Mechemache, Daskalova and Nes2020), there is a limited understanding of the impact of increased concentration in the retail sector on prices and welfare effects in the farming sector.
Our findings represent general trends in the EU food chain regarding concentration and markups and support the notion that market power likely exists in several of the food chains in Europe, potentially creating negative welfare and distributional effects. Markups and concentration indicators show correlations across subsectors, and markups and turnover exhibit correlations within subsectors, suggesting that these indicators may identify where these negative effects may potentially occur. However, it is important to note that these correlations are not perfect, and other firm or market characteristics beyond power imbalances may be captured. Therefore, these indicators should be considered in conjecture with other market and firm characteristics to induce a more effective identification of the abuse of market power or superior bargaining positions. Overall, our results underscore the need for further research aimed at identifying power imbalances among vertically linked supply chain actors and formulating policies to increase efficiencies, welfare, and fairness within the European food sector.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S1068280524000169
Data availability statement
The data that support the findings of this study are available from BvD. Restrictions apply to the availability of these data, which were used under licence for this study.
Funding statement
This research received no specific grant from any funding agency, commercial, or not-for-profit sectors.
Competing interests
None.
Open access
