Cost pass-through in the Swedish coffee market

Cost pass-through in the Swedish coffee market Abstract Cost pass-through to retail prices shows how changes in marginal costs are allocated between producers and consumers, and it is, therefore, closely related to market structure and competition. This paper uses Swedish data on coffee products at the barcode level to evaluate pass-through from the cost of green coffee beans, the main marginal cost, to the retail price of roasted and ground coffee. First long-run cost pass-through is estimated for each product, and then regression is used to analyse how pass-through varies across market shares, retailer-owned brands and other product characteristics. A general result is that pass-through is roughly complete for products with large market shares, while those with small market shares have low pass-through rates. There is no evidence that retailer-owned brands have higher pass-through than brand-name products with similar market shares, which would be the case if retailer-owned brands avoided double marginalisation through vertical integration. Thus, although there is not perfect competition in the Swedish coffee market, a large part of it appears to be highly competitive. 1. Introduction Many consumer markets for coffee are characterised by a few dominant multinational or national roasters, with a combined market share often exceeding 60 per cent, and a large number of small roasters and importers (Sutton, 2007). Not surprisingly, the large roasters are regularly alleged to charge high prices for processed coffee relative to the price of their main input, i.e. green coffee beans.1 The rapid growth of Fairtrade coffee sales, which increased by about 1,400 per cent between 2000 and 2015, is probably a result of these allegations (Fairtrade Foundation, 2012; Fairtrade International, 2017). Nevertheless, there is little empirical evidence of market power in consumer coffee markets for roasted and ground coffee. In fact, most studies find no or a tiny markup of price over marginal cost in Sweden and other developed countries (Bettendorf and Verboven, 2000; Feuerstein, 2002; Durevall, 2007a, 2007b; Gibbon, 2007; Gilbert, 2007, 2008). However, these studies treat coffee as a homogeneous product, due to paucity of data, and therefore fail to capture important market features (Sexton, 2013). A few recent studies analyse pass-through from marginal costs to retail prices using product-level data and find cost pass-through to be incomplete, i.e. less than 1 in the USA (Leibtag et al., 2007; Nakamura and Zerom, 2010), and far below 1 in Germany and France (Bonnet et al., 2013; Bonnet and Villas-Boas, 2016). The low pass-through indicates that there is imperfect competition and that prices are higher than marginal costs. This paper analyses how changes in marginal costs, measured by fluctuations in world market green coffee bean prices, are passed on to Swedish retail coffee prices and how pass-through rates are related to market structure. As in previous studies, I focus on roasted and ground coffee since green beans is the major input in the production process. Moreover, ground coffee has a market share of 80 per cent in the Swedish retail market. The Swedish coffee retail market is likely to be representative of many other markets in developed countries (Durevall, 2003; Sutton, 2007). In the ground coffee segment, the four largest roasters have a combined market share of about 85 per cent. There are a few retailer-owned brands (private labels), several small, mostly local, roasters and some foreign brands. Moreover, as in most developed countries, the retail sector is very concentrated (McCorriston, 2013). For example, the largest food chain has about 50 per cent of the market and the four largest account for well over 90 per cent of all sales (Swedish Competition Authority, 2011). The first part of the study uses unrestricted error correction models to estimate pass-through of industry-wide shocks to marginal costs for product/regional market combinations of each barcode-specific coffee product. The choice of model is based on the high persistence of the green bean price series, which can be characterised as unit root non-stationary. The unit root implies that bean prices should be cointegrated with the retail price for costs to have long-run impact, given other input costs. Since I also have data on quantities, in contrast to most other studies, the second part uses regression analysis to analyse the relationship between market structure and pass-through. Horizontal market structure is measured by coffee product, blend and brand market shares, and vertical market structure is measured by the presence of private labels. It is a common assumption that products with a large market share have a lower pass-through than products with small market shares, due to higher markup and more market power. This is what the Dornbusch (1987) model of monopolistic competition predicts, a model widely used in international economics (Burstein and Gopinath, 2014; Feenstra, 2016). Vertical integration, in the form of private labels, sidesteps the market power of manufacturers of branded products and should prevent double marginalisation, i.e. when a downstream firm (retailer) applies a markup on a price set by an upstream firm (roaster) that has market power and, therefore, also applies a markup. Avoidance of double marginalisation should lead to higher pass-through, given market share (Hong and Li, 2017). Therefore, by controlling for market shares, I can indirectly evaluate whether there is double marginalisation. Finally, as a robustness check, I use the cointegrated VAR model of Johansen (1991) to analyse cost pass-through and adjustment to cost shocks in one regional market, eastern Sweden.2 I focus on four dominant blends, which together have a market share of close to 50 per cent. The Johansen approach allows me to formally test whether there is perfect pass-through and evaluate the speed of adjustment to cost shocks. The paper is organised as follows: Section 2 gives a brief review of why pass-through rates might vary across firms and of the findings of earlier studies. Section 3 describes the data and Section 4 outlines the empirical approach. Section 5 estimates the pass-through rates and analyses the role of market structure, while Section 6 uses the cointegrated VAR model to analyse prices and costs of large blends in eastern Sweden. Section 7 concludes the paper. 2. Pass-through and market structure: theory and empirical evidence Pass-through of costs to prices has been a research topic for a long time, particularly in the fields of marketing, agricultural and public economics (Bulow and Pfleiderer, 1983; Tyagi, 1999; Weyl and Fabinger, 2013; Lloyd, 2017). Numerous studies on the topic have recently been published in the field of international economics as well, and have mainly focused on the effect of changes in exchange rates on consumer prices (Burstein and Gopinath, 2014). Several factors may affect pass-through, and no model captures all of them. Thus, for expository purposes, I use a Bertrand model from Anderson, De Palma and Kreider (2001) to illustrate the key ones when there are industry-wide changes in costs, conveniently measured by a unit tax.3 I then informally discuss other factors of relevance to the retail coffee market.4 The model has N firms that each produces one differentiated product, q, at constant marginal cost c. The profit function of firm i is πi=[pi−(ci+t)]q(pi,p−i), (1) where pi is the firm i’s price, p−i is the set of all other firms’ prices, q(pi,p−i) is the demand for the output of firm i and t is the unit tax. Profit maximisation gives the first-order condition [pi−(c'i+t)]δqi(pi,p−i)δpi+qi(pi,p−i)=0. (2) Pass-through is obtained by total differentiation with respect to t. By assuming symmetry between the firms, we can derive an expression for pass-through based on the diversion ratio, D, which measures the share of firm i’s sales that is captured by other firms when firm i raises its price, the elasticity of the slope of firm i’s demand curve with respect to the common price εm=[p/(∂qi/∂pi)][∂(∂qi/∂pi)/∂p], and firm i’s price elasticity with respect to a shift in its own price, given all other prices εdd=[p/q][∂qi/∂pi] (see RBB, 2014 for details), dpidt=12−D−εmεdd. (3) Equation (3) shows that in the case of complete product differentiation, that is, firm i is a monopolist, then D = 0 and the value of the ratio between the slope elasticity of the demand curve and the price elasticity determines pass-through. For instance, when demand is linear, εm=0, we have the textbook case of a pass-through of ½. It is reasonable to assume that D is generally higher the more firms there are in the market, since then there is more competition. Thus, as competition (the number of firms) increases, D approaches 1 and pass-through approaches 1. Consequently, even when demand is linear, pass-through can be 1. However, competition will also increase firm i’s price elasticity, and when there is perfect competition, the ratio εm/εdd will be zero irrespective of the slope elasticity of the demand curve. With monopolistic competition and a given D, the curvature of the demand function determines pass-through. When demand is concave, the slope of the demand curve becomes flatter as the price increases, i.e. the firm loses sales rapidly when raising prices. As a result, pass-through is low since εm increases with price. When demand is convex, εm declines when the price increases, and it is possible that pass-through is over 1. There are several multiproduct roasters in the Swedish coffee market, which is not captured by the model. A few studies analyse pass-through in theoretical models where firms produce more than one good, and the results depend on assumptions about functional form and the size of second-order effects (Besanko, Dubé and Gupta, 2005; Moorthy, 2005). However, by developing a structural model for the Swedish beer market, Rohman and Friberg (2016) provide evidence that multiproduct firms are likely to have lower pass-through than single-product firms, both when there are industry-wide and firm-specific cost shocks. They also show that the difference in pass-through between a single and a multiproduct firm may be small unless the multiproduct firm sells very many products, making it challenging to detect differences in an observational study. The shape of the cost function may also affect pass-through. An upward-sloping marginal cost function decreases pass-through both in perfectly competitive and oligopolistic markets. Nonetheless, the assumption of constant marginal costs is standard in the literature on coffee since processing of ground coffee is highly mechanised, but simple, and involves only a few workers (Bettendorf and Verboven, 2000; Sutton, 2007; Nakamura and Zerom, 2010). Moreover, as coffee consumption in Sweden is small relative to the world supply of beans, demand does not affect bean prices. Therefore, upward-sloping marginal costs are unlikely to have a major impact on pass-through. Some models allow for heterogeneous firms (Dornbusch, 1987; Feenstra, 2016, Ch. 6). In these models, there is usually a negative relationship between market share and pass-through: firms with large market shares have large markups and adjust the markup instead of the price after a change in marginal costs. Several studies also find a link between the degree of product pass-through and horizontal market structure (Atkeson and Burstein, 2008; Amiti, Itskhoki and Konings, 2014; Auer and Schoenle, 2016; Hong and Li, 2017).5 Most ground coffees are brand-name products processed and distributed to retailers by roasters. When upstream and downstream firms, i.e. roasters and retailers, have market power, there might be double marginalisation.6 As a result, there is markup on markup, and pass-through is lower than when a monopolist owns both the upstream and downstream firms. The number of private labels has been increasing for several years in many countries, including Sweden. This can be seen as vertical integration, since products without the manufacturers’ label are introduced. Private labels are expected to have a higher pass-through than brand-name products since they sidestep the roasters’ markup, though a large market share would have a counterbalancing effect. However, by controlling for market shares, Hong and Li (2017) show that private labels in the USA have substantially higher pass-through rates than branded products. In Sweden, large roasters and retail chains meet regularly to negotiate wholesale prices, so relative bargaining power and pricing contracts could influence pass-through.7 Thus, double marginalisation could be prevented by outright collusion, though this is illegal and unlikely given the number of actors. However, there are many types of vertical contracts. Based on simulation of structural models, Bonnet et al. (2013) find evidence of resale price maintenance in the German ground coffee market, which increases pass-through relative to double marginalisation. Finally, the model is static and pass-through is treated as the equilibrium price response to a change in costs, ignoring dynamics. Nakamura and Zerom (2010) examine short-run adjustments in the US coffee market using a structural model. They find that menu costs matter in the short run but not after a year or so.8 Thus, to conclude, studies of market power in coffee retail markets, which assume that coffee is a homogenous product and use average market prices, fail to find evidence of market power (Gibbon, 2007), while those that use product-level data find pass-through rates that are below one and conclude that there is imperfect competition. Assuming highly elastic supply, a pass-through rate below or above 1 is evidence against perfect competition. However, as evident from theory, the presence of a pass-through rate of 1 is neither a necessary nor a sufficient condition for perfect competition. Yet, by analysing how pass-through of individual products is related to horizontal and vertical integration, an informed judgement can be made. 3. The data The retail data are from the Nielsen company, which collects weekly scanner data the EAN (barcode) level of daily sales from 3,088 Swedish stores. Coffee is grouped into seven market segments, but I focus on the by far largest segment, roasted and ground coffee. Nielsen also collects information on a number of product characteristics, such as brand, manufacturer and whether the product is organic. I use this information to distinguish between pass-through rates for the major brands, three types of private labels (discount, premium and standard) and conventional and organic coffee; as all Fairtrade blends are organic,9 it is not modelled separately. Unfortunately, Nielsen does not make store-level information available in Sweden, so prices are averages for six regional markets. I use monthly averages to reduce the number of missing observations (all products are not sold every week) and the noise-to-information ratio; it clearly takes more than a week for a change in coffee bean prices to affect retail prices. Thus prices, obtained by dividing value by volume for a specific package size, are in Swedish kronor (SEK) per kilo at the barcode level by region and month. For simplicity, I refer to product/regional market combinations as products, and coffee products, such as Gevalia 0.500 kg Mellanrost Brygg, as blends. One feature of retail prices is that they sometimes decrease due to price promotions, but there is no information in the database on promotions. Some studies filter the price series to remove temporary decreases and analyse normal prices or reference prices (Eichenbaum, Jaimovich and Rebelo, 2011; Lloyd et al., 2014). However, my prices are averages by region, not individual store prices, so it is not clear what would be filtered out. Moreover, I analyse a relatively long period, almost 6 years, and sales are not synchronised across food chains, so occasional sales probably only have a minor impact on pass-through rates. I, therefore, use actual average prices in the analysis. Since I analyse individual time series, products with missing observations have been dropped. They make up 12 per cent of the sample in volume terms. Most of the products belong to blends that were introduced in, or removed from, the market during the study period. As Table 1 shows, the sample thus consists of 378 products and 68 blends, belonging to 20 brands (plus unspecified Other Brands), and runs from March 2009 to November 2014. Almost all roasters produce one brand. The noteworthy exception to this is Gevalia, which produces three brands: Gevalia, Blå Mocca and Maxwell House. Table 1. Brands, coffees, products, market shares and average prices Brand Blends Products Market share (%) Mean real price Gevalia 5 30 39.37 59.04 Zoega 10 60 26.80 70.63 Classic 9 54 14.02 65.63 Löfbergs Lila 7 38 6.41 57.67 ICA 6 36 5.05 48.43 Blå Mocca 3 18 3.18 61.75 Coop 6 36 1.61 49.19 Änglamark 1 6 1.10 62.74 Maxwell House 1 6 0.59 68.56 Coop X-tra 1 6 0.48 36.44 ICA I Love Eco 2 12 0.36 59.72 Lindvall 4 23 0.34 49.72 Euroshopper 1 6 0.28 33.91 Sackeus 2 7 0.11 79.74 Ideekaffe 1 6 0.09 85.13 Folke Bergman 1 1 0.08 47.52 Servtrade 3 16 0.05 39.22 Najjar 2 9 0.04 109.40 Other Brands 2 7 0.03 66.60 Lyxkaffe 1 1 0.00 38.68 Sum 68 378 100 Brand Blends Products Market share (%) Mean real price Gevalia 5 30 39.37 59.04 Zoega 10 60 26.80 70.63 Classic 9 54 14.02 65.63 Löfbergs Lila 7 38 6.41 57.67 ICA 6 36 5.05 48.43 Blå Mocca 3 18 3.18 61.75 Coop 6 36 1.61 49.19 Änglamark 1 6 1.10 62.74 Maxwell House 1 6 0.59 68.56 Coop X-tra 1 6 0.48 36.44 ICA I Love Eco 2 12 0.36 59.72 Lindvall 4 23 0.34 49.72 Euroshopper 1 6 0.28 33.91 Sackeus 2 7 0.11 79.74 Ideekaffe 1 6 0.09 85.13 Folke Bergman 1 1 0.08 47.52 Servtrade 3 16 0.05 39.22 Najjar 2 9 0.04 109.40 Other Brands 2 7 0.03 66.60 Lyxkaffe 1 1 0.00 38.68 Sum 68 378 100 Note: Based on the period March 2009–November 2014. The average retail price is in constant 2010 prices net of 12 per cent value added tax. Products are geographic region/coffee combinations. The market shares are measured as volume of sales. View Large Table 1. Brands, coffees, products, market shares and average prices Brand Blends Products Market share (%) Mean real price Gevalia 5 30 39.37 59.04 Zoega 10 60 26.80 70.63 Classic 9 54 14.02 65.63 Löfbergs Lila 7 38 6.41 57.67 ICA 6 36 5.05 48.43 Blå Mocca 3 18 3.18 61.75 Coop 6 36 1.61 49.19 Änglamark 1 6 1.10 62.74 Maxwell House 1 6 0.59 68.56 Coop X-tra 1 6 0.48 36.44 ICA I Love Eco 2 12 0.36 59.72 Lindvall 4 23 0.34 49.72 Euroshopper 1 6 0.28 33.91 Sackeus 2 7 0.11 79.74 Ideekaffe 1 6 0.09 85.13 Folke Bergman 1 1 0.08 47.52 Servtrade 3 16 0.05 39.22 Najjar 2 9 0.04 109.40 Other Brands 2 7 0.03 66.60 Lyxkaffe 1 1 0.00 38.68 Sum 68 378 100 Brand Blends Products Market share (%) Mean real price Gevalia 5 30 39.37 59.04 Zoega 10 60 26.80 70.63 Classic 9 54 14.02 65.63 Löfbergs Lila 7 38 6.41 57.67 ICA 6 36 5.05 48.43 Blå Mocca 3 18 3.18 61.75 Coop 6 36 1.61 49.19 Änglamark 1 6 1.10 62.74 Maxwell House 1 6 0.59 68.56 Coop X-tra 1 6 0.48 36.44 ICA I Love Eco 2 12 0.36 59.72 Lindvall 4 23 0.34 49.72 Euroshopper 1 6 0.28 33.91 Sackeus 2 7 0.11 79.74 Ideekaffe 1 6 0.09 85.13 Folke Bergman 1 1 0.08 47.52 Servtrade 3 16 0.05 39.22 Najjar 2 9 0.04 109.40 Other Brands 2 7 0.03 66.60 Lyxkaffe 1 1 0.00 38.68 Sum 68 378 100 Note: Based on the period March 2009–November 2014. The average retail price is in constant 2010 prices net of 12 per cent value added tax. Products are geographic region/coffee combinations. The market shares are measured as volume of sales. View Large The market is highly concentrated; the two largest brands, Gevalia and Zoegas have 39 per cent and 27 per cent of the ground coffee market (in volume terms) and are owned by Mondelez and Nestlé, respectively. The market shares of the other two large brands, Classic and Löfbergs Lila, are 14 per cent and 6 per cent, respectively. They are produced by family-owned roasters. The brands ICA, Coop, Änglamark, ICA I Love Eco and Euroshopper are private labels, the other brands are produced by small Swedish roasters or imported. There is no information on commodity costs for manufacturers or brands, so green coffee bean prices in SEK are used to measure cost shocks. There are two sources for bean prices: the International Coffee Organization (ICO) collects price data from terminal markets in Europe and the USA, and Statistics Sweden publishes value and volume of monthly imports of green coffee beans. I use prices based on import data from Statistics Sweden, but the choice does not matter much for the results, except that ICO prices take longer to affect retail prices. The reason for the similarity between of import prices and prices set at the commodity exchanges is that the latter function as price guides for physical coffee trade (ITA, 2011). The use of the same cost, that is, the weighted average cost of beans, for all blends introduces a bias in the estimation of pass-through rates. The estimates will be too low for blends with lower than average costs and too high for blends with higher than average costs. However, the effect of biases on the relative sizes of the pass-through rates can be checked by taking the log of the variables, estimating pass-through elasticities. This works because only Arabica beans are used and their prices have the same stochastic trends since they are close substitutes, i.e. they are cointegrated (see Fousekis and Grigoriadis, 2017).10 Table 1 also shows that average real retail prices (net of VAT) vary from 34 SEK to 109 SEK per kg, which can be compared with the average cost of green coffee beans of 32.60 SEK. To highlight the evolution of prices over time, Figure 1 shows the average price of unweighted ground coffee and cost of green coffee beans, measured by the bean price times 1.19 to account for weight lost from roasting.11 Both series are measured in constant 2010 prices using the consumer price index. There were large changes in prices and costs during the study period. The average per kg price of ground coffee fluctuated between 60 SEK and 80 SEK, and the cost of green coffee beans fluctuated between 20 SEK and 45 SEK per kg. Figure 1 also shows that the series are highly persistent; when the augmented Dickey–Fuller (ADF) test is applied on the individual retail prices, only about 10–15 tests reject the null hypothesis of a unit root (results available on request). The bean price series also appears to have a unit root: the ADF test with one lag (longer lags are insignificant), a constant, and a time trend gives a t-value of −1.4 (5 per cent significance value is −3.47) and an estimated root of 0.97. Fig. 1. View largeDownload slide Average price per kg of ground coffee and cost of imported green coffee beans in net of VAT 2010 SEK, 2009:3–2014:11. Fig. 1. View largeDownload slide Average price per kg of ground coffee and cost of imported green coffee beans in net of VAT 2010 SEK, 2009:3–2014:11. 4. Empirical approach The most commonly used model to estimate pass-through is the distributed lags model in rates of change, i.e. the first difference of the log-level of prices and costs (Nakamura and Zerom, 2010; Aron, Macdonald and Muellbauer, 2014),12 though Bonnet and Villas-Boas (2016) and Bonnet et al. (2013) estimate static fixed effects models in log levels. One reason for using rates of change is probably that often only price indices are analysed in international economics and the focus is on changes in the exchange rate. Another reason is the risk of spurious correlation, since some series might be unit root non-stationary. A drawback of using first differences is that important information about the long-run pass-through is ignored if the series have unit roots and are cointegrated, which, as evident from Figure 1, seems to be the case for ground coffee. Thus, I estimate models in levels. Spurious regression is unlikely to be a major issue for ground coffee, since the cost of green coffee beans is by far the largest component of marginal cost, estimated to equal 50–90 per cent (Bettendorf and Verboven, 2000; Nakamura and Zerom, 2010; Bonnet et al., 2013). Moreover, bean costs usually fluctuate much more than other production-related costs, and this was true for Sweden during the study period, when compared with wages and the consumer price index.13 Bean costs and retail coffee prices are, therefore, expected to have the same stochastic trends. However, some retail prices are stationary, as mentioned, and some could be non-stationary but unrelated to costs, implying that long-run pass-through is zero. This could occur because real retail prices are kept more or less fixed over the study period, or because of large abrupt price changes due to revised price policies. In these cases, the pass-through estimates are small or in some cases even negative. Static models in levels are easy to estimate and might provide adequate long-run pass-through rates when the variables are integrated of order one and cointegrated. However, a static model with monthly data is not likely to capture the dynamics adequately; i.e. the common factor restriction (implicitly) imposed is often invalid (Ericsson and MacKinnon, 2002). It might also fail to provide consistent estimates of pass-through when the dependent variable is stationary and lags are needed to specify the model correctly. I therefore estimate pass-through coefficients for each product using an unrestricted error correction model specified as Δpt=α+α1Δpt−1+α2Δcostt−1+β1pt−1+β2costt−1+trend+seas+εt, (4) where pt is the net-of-VAT price level of a product at month t, cost is the price of one kg of green coffee beans adjusted for weight lost from roasting, Δ is the difference operator, β1 and β2 are the parameters of interest, β2/−β1 is the estimate of long-run pass-through and εt is the error term. The lags in first differences, Δpt−1andΔcostt−1, potentially improve the estimates of β1 and β2 by capturing short-run dynamics. Adding more lags only affects the estimates of β2/−β1 marginally and does not change any of the results. Trend is a deterministic trend that potentially captures other marginal costs and slow-moving changes in technologies, and seas are seasonal dummies. I do not base the subsequent analysis on formal tests for cointegration of each of the 378 equations, but very low or negative pass-through rates indicate a lack of a long-run relationship. However, I also estimate distributed lag models in first differences and static models in levels, as well as the error correction model with variables in log-levels. Moreover, as robustness check, in Section 6, I formally test for cointegration in a multivariate framework in one of the six regional markets using the battery developed by Johansen (1991). 5. Pass-through rates and the role of market structure Figure 2 reports the estimates of the pass-through rates.14 The vast majority are larger than zero, as expected. Those with negative pass-through are mostly odd products with almost constant prices and very small market shares. One product with a pass-through rate of 2.1 is excluded from the figure to improve readability; a detailed analysis revealed that the high value is due to a large structural break at the end of the sample. Fig. 2. View largeDownload slide Estimates of pass-through rates for 378 products. Note: Estimates are based on unrestricted error correction models with one lag. One observation with a pass-through of 2.1 has been dropped. Fig. 2. View largeDownload slide Estimates of pass-through rates for 378 products. Note: Estimates are based on unrestricted error correction models with one lag. One observation with a pass-through of 2.1 has been dropped. Most pass-through rates fall in the 0.5–1 range. This is also the case when long-run pass-through is estimated with a static model in levels, though the pass-through rates are somewhat lower on average (see Figure A1 in the Appendix). The correlation between the two measures is 0.92. When pass-through rates are estimated with distributed lags models in first differences, the pass-through rates are clearly lower, as shown in Figure A2 in the Appendix. The models have 24 lags, but adding more lags does not increase pass-through rates. Thus, distributed lags models seem to underestimate the size of the pass-through rates since they do not use information contained in the levels of the variables. Lastly, Figure A3 in the Appendix shows that there is a strong correlation between pass-through rates and pass-through elasticities, 0.93, implying that the relative ranking of the pass-through rates is only marginally affected by the use of weighted average costs. Some estimates are larger than 1, but this is probably due to estimation uncertainty; almost no 95 per cent confidence interval excludes the value of 1. Thus, there is no evidence of highly convex demand functions, i.e. functional forms that generate pass-through larger than 1. Figure 3 depicts the estimated pass-through coefficients and market shares, in logs for visibility, for the 68 blends. It shows that the larger the market share, the higher the pass-through. This relationship is contrary to the prediction of models in which the size of the market share is related to market power, indicating that blends with large market shares compete for customers. The relationship is also non-linear, as a pass-through rate of 1 seems to constitute a (fuzzy) upper limit. Fig. 3. View largeDownload slide Pass-through rates for products and log of blend market shares. Fig. 3. View largeDownload slide Pass-through rates for products and log of blend market shares. Table 2 reports regressions that test for the association between pass-through rates and market shares. Since it is not obvious how to measure market shares, three measures are used: product, blend and brand market shares. To capture the non-linear relationship, both level and squared market share terms are included. Table 2. Pass-through and market shares, ground coffee (1) (2) (3) (4) (5) (6) Product share 1.039 3.268 (0.333)** (1.171)** Product share sq −5.761 (2.255)* Blend share 1.728 7.596 (0.943) (2.095)** Blend share sq −24.372 (7.336)** Brand share 0.468 2.253 (0.245) (1.260) Brand share sq −3.641 (2.241) Najjar/Bosnia −0.911 −0.889 −0.900 −0.844 −0.876 −0.791 (0.083)** (0.085)** (0.084)** (0.087)** (0.110)** (0.116)** Constant 0.711 0.689 0.700 0.642 0.674 0.579 (0.039)** (0.042)** (0.041)** (0.046)** (0.074)** (0.093)** R2 0.30 0.32 0.31 0.37 0.31 0.35 N 378 378 378 378 378 378 (1) (2) (3) (4) (5) (6) Product share 1.039 3.268 (0.333)** (1.171)** Product share sq −5.761 (2.255)* Blend share 1.728 7.596 (0.943) (2.095)** Blend share sq −24.372 (7.336)** Brand share 0.468 2.253 (0.245) (1.260) Brand share sq −3.641 (2.241) Najjar/Bosnia −0.911 −0.889 −0.900 −0.844 −0.876 −0.791 (0.083)** (0.085)** (0.084)** (0.087)** (0.110)** (0.116)** Constant 0.711 0.689 0.700 0.642 0.674 0.579 (0.039)** (0.042)** (0.041)** (0.046)** (0.074)** (0.093)** R2 0.30 0.32 0.31 0.37 0.31 0.35 N 378 378 378 378 378 378 Note: Najjar/Bosnia is a dummy for Najjar and Bosnia (Zlatna Dzezva) products, out of which all but one have negative pass-through rates. Standard errors are clustered at the product level except when brand shares are included; in which case clustering is at the brand level. *p < 0.05; **p < 0.01. View Large Table 2. Pass-through and market shares, ground coffee (1) (2) (3) (4) (5) (6) Product share 1.039 3.268 (0.333)** (1.171)** Product share sq −5.761 (2.255)* Blend share 1.728 7.596 (0.943) (2.095)** Blend share sq −24.372 (7.336)** Brand share 0.468 2.253 (0.245) (1.260) Brand share sq −3.641 (2.241) Najjar/Bosnia −0.911 −0.889 −0.900 −0.844 −0.876 −0.791 (0.083)** (0.085)** (0.084)** (0.087)** (0.110)** (0.116)** Constant 0.711 0.689 0.700 0.642 0.674 0.579 (0.039)** (0.042)** (0.041)** (0.046)** (0.074)** (0.093)** R2 0.30 0.32 0.31 0.37 0.31 0.35 N 378 378 378 378 378 378 (1) (2) (3) (4) (5) (6) Product share 1.039 3.268 (0.333)** (1.171)** Product share sq −5.761 (2.255)* Blend share 1.728 7.596 (0.943) (2.095)** Blend share sq −24.372 (7.336)** Brand share 0.468 2.253 (0.245) (1.260) Brand share sq −3.641 (2.241) Najjar/Bosnia −0.911 −0.889 −0.900 −0.844 −0.876 −0.791 (0.083)** (0.085)** (0.084)** (0.087)** (0.110)** (0.116)** Constant 0.711 0.689 0.700 0.642 0.674 0.579 (0.039)** (0.042)** (0.041)** (0.046)** (0.074)** (0.093)** R2 0.30 0.32 0.31 0.37 0.31 0.35 N 378 378 378 378 378 378 Note: Najjar/Bosnia is a dummy for Najjar and Bosnia (Zlatna Dzezva) products, out of which all but one have negative pass-through rates. Standard errors are clustered at the product level except when brand shares are included; in which case clustering is at the brand level. *p < 0.05; **p < 0.01. View Large Nor is it obvious how to deal with very low and negative pass-through rates, which presumably are zero. Hong and Li (2017) simply exclude products with negative pass-through in their analysis. However, two types of products have negative pass-through rates, private labels and imported blends from brands primarily sold to immigrant groups. Thus, instead of removing all of them, I include a dummy for Najjar and Bosnia (Zlatna Dzezva) products, which make up about half of those with negative pass-through rates. The others are primarily private labels, which are of interest for the study. I also estimated models with all pass-through rates below 0.1 set to zero, but the results only changed marginally (the results are available on request). Table 2 shows that there is a strong positive, but non-linear, association between pass-through and market share. For example, at the median blend market share, 0.005, the pass-through is 0.70, and at a blend market share of 0.05, it is 0.84. Market shares measured at the blend level are somewhat more correlated with pass-through than the others, while the lack of significance for brand shares might be because of the relatively small number of brands, but most likely it is because the product and blend levels matter more for pricing decisions than the brand level (as shown below). Table A1 in the appendix reports the same specifications but with the pass-through rate replaced by the pass-through elasticity. The results are practically the same, showing that the measurement errors are of minor importance. Table 3 reports specifications that aim to further describe differences in pass-through. Column (1) adds three types of private labels (discount premium, and standard), type of roast (dark and unspecified; the base is medium roast) and organic coffee to Table 2’s column (2). Private labels should have a higher pass-through in monopolistic markets, given market share, since they avoid potential double marginalisation. However, both discount and premium private labels have a lower pass-through rate than other blends, while the standard one has the same rate, given market shares. The dummies for dark roast and organic coffee are insignificant, while unknown roast is significant at the 5 per cent level. The coefficients of product market share and squared product market share are unaffected by the inclusion of the other variables. Table 3. Pass-through, private labels and large brands, ground coffee (1) (2) (3) (4) (5) (6) Private label Brand dummies Brand dummies Four large brands Gevalia Classic Blend share 7.147 1.631 2.369 19.069 (2.128)** (1.834) (0.442)** (5.393)** Blend share sq −23.840 −5.789 −7.085 −252.179 (7.480)** (6.548) (1.605)** (72.375)** Classic 0.354 0.320 (0.086)** (0.061)** Löfbergs lila 0.413 0.378 (0.087)** (0.063)** Gevalia 0.311 0.267 (0.086)** (0.057)** Zoega 0.334 0.325 (0.122)** (0.104)** Private Disc −0.752 −0.602 (0.071)** (0.091)** Private Prem −0.283 −0.013 (0.097)** (0.111) Private Stand 0.003 0.191 (0.058) (0.085)* Dark roast 0.093 0.053 (0.060) (0.054) Unknown roast 0.253 0.248 (0.105)* (0.081)** Organic 0.102 −0.016 (0.081) (0.054) Odd size −0.996 −0.855 −0.765 (0.066)** (0.078)** (0.090)** Constant 0.632 0.517 0.565 0.863 0.750 0.731 (0.072)** (0.087)** (0.052)** (0.045)** (0.012)** (0.057)** R2 0.57 0.65 0.49 0.01 0.64 0.49 N 378 378 378 182 30 54 (1) (2) (3) (4) (5) (6) Private label Brand dummies Brand dummies Four large brands Gevalia Classic Blend share 7.147 1.631 2.369 19.069 (2.128)** (1.834) (0.442)** (5.393)** Blend share sq −23.840 −5.789 −7.085 −252.179 (7.480)** (6.548) (1.605)** (72.375)** Classic 0.354 0.320 (0.086)** (0.061)** Löfbergs lila 0.413 0.378 (0.087)** (0.063)** Gevalia 0.311 0.267 (0.086)** (0.057)** Zoega 0.334 0.325 (0.122)** (0.104)** Private Disc −0.752 −0.602 (0.071)** (0.091)** Private Prem −0.283 −0.013 (0.097)** (0.111) Private Stand 0.003 0.191 (0.058) (0.085)* Dark roast 0.093 0.053 (0.060) (0.054) Unknown roast 0.253 0.248 (0.105)* (0.081)** Organic 0.102 −0.016 (0.081) (0.054) Odd size −0.996 −0.855 −0.765 (0.066)** (0.078)** (0.090)** Constant 0.632 0.517 0.565 0.863 0.750 0.731 (0.072)** (0.087)** (0.052)** (0.045)** (0.012)** (0.057)** R2 0.57 0.65 0.49 0.01 0.64 0.49 N 378 378 378 182 30 54 Note: Najjar/Bosnia is a dummy for Najjar and Bosnia (Zlatna Dzezva) products, out of which all but one have negative pass-through rates. Standard errors are clustered at the product level except when brand shares are included; in which case clustering is at the brand level. *p < 0.05; **p < 0.01. View Large Table 3. Pass-through, private labels and large brands, ground coffee (1) (2) (3) (4) (5) (6) Private label Brand dummies Brand dummies Four large brands Gevalia Classic Blend share 7.147 1.631 2.369 19.069 (2.128)** (1.834) (0.442)** (5.393)** Blend share sq −23.840 −5.789 −7.085 −252.179 (7.480)** (6.548) (1.605)** (72.375)** Classic 0.354 0.320 (0.086)** (0.061)** Löfbergs lila 0.413 0.378 (0.087)** (0.063)** Gevalia 0.311 0.267 (0.086)** (0.057)** Zoega 0.334 0.325 (0.122)** (0.104)** Private Disc −0.752 −0.602 (0.071)** (0.091)** Private Prem −0.283 −0.013 (0.097)** (0.111) Private Stand 0.003 0.191 (0.058) (0.085)* Dark roast 0.093 0.053 (0.060) (0.054) Unknown roast 0.253 0.248 (0.105)* (0.081)** Organic 0.102 −0.016 (0.081) (0.054) Odd size −0.996 −0.855 −0.765 (0.066)** (0.078)** (0.090)** Constant 0.632 0.517 0.565 0.863 0.750 0.731 (0.072)** (0.087)** (0.052)** (0.045)** (0.012)** (0.057)** R2 0.57 0.65 0.49 0.01 0.64 0.49 N 378 378 378 182 30 54 (1) (2) (3) (4) (5) (6) Private label Brand dummies Brand dummies Four large brands Gevalia Classic Blend share 7.147 1.631 2.369 19.069 (2.128)** (1.834) (0.442)** (5.393)** Blend share sq −23.840 −5.789 −7.085 −252.179 (7.480)** (6.548) (1.605)** (72.375)** Classic 0.354 0.320 (0.086)** (0.061)** Löfbergs lila 0.413 0.378 (0.087)** (0.063)** Gevalia 0.311 0.267 (0.086)** (0.057)** Zoega 0.334 0.325 (0.122)** (0.104)** Private Disc −0.752 −0.602 (0.071)** (0.091)** Private Prem −0.283 −0.013 (0.097)** (0.111) Private Stand 0.003 0.191 (0.058) (0.085)* Dark roast 0.093 0.053 (0.060) (0.054) Unknown roast 0.253 0.248 (0.105)* (0.081)** Organic 0.102 −0.016 (0.081) (0.054) Odd size −0.996 −0.855 −0.765 (0.066)** (0.078)** (0.090)** Constant 0.632 0.517 0.565 0.863 0.750 0.731 (0.072)** (0.087)** (0.052)** (0.045)** (0.012)** (0.057)** R2 0.57 0.65 0.49 0.01 0.64 0.49 N 378 378 378 182 30 54 Note: Najjar/Bosnia is a dummy for Najjar and Bosnia (Zlatna Dzezva) products, out of which all but one have negative pass-through rates. Standard errors are clustered at the product level except when brand shares are included; in which case clustering is at the brand level. *p < 0.05; **p < 0.01. View Large Column (2) replaces market shares with dummies for the four large brands. They are all significant at the 1 per cent level and show that the large brands have pass-through rates that on average are about 0.3–0.4 SEK higher than the other brands. Column (3) shows that these results are not due to the other variables included in the model. Column (4) restricts the sample to the four large brands to test whether pass-through rates differ across their products. Although the market share coefficients have the same signs as in the other regressions, they are much smaller and insignificant. However, two of the large brands, Gevalia and Classic, have pass-through rates that vary with market shares, as reported in columns (5) and (6). This is not the case for the other two, Löfbergs Lila and Zoega (not reported). Figure 4 sheds further light on the relationship between pass-through and the 20 brand market shares. Average pass-through rates are similar for the four largest brands, but there is a fairly large variance among the products of each brand. Moreover, it is clear that only very small brands have pass-through rates close to, or below, zero. Fig. 4. View largeDownload slide Pass-through rates for products and brand market shares. Note: Estimates are based on unrestricted error correction models with one lag. One observation with a pass-through of 2.1 has been dropped. Fig. 4. View largeDownload slide Pass-through rates for products and brand market shares. Note: Estimates are based on unrestricted error correction models with one lag. One observation with a pass-through of 2.1 has been dropped. 6. Analysis of large coffees in the Eastern Swedish Market To provide additional support to the findings that a substantial part of the Swedish coffee market is competitive, I use the cointegrated VAR model and the Johansen approach to analyse pass-through and dynamic price interaction among large blends. I focus on one market, eastern Sweden, which covers the most densely populated areas of the country, including the Stockholm region, Uppsala, Nyköping and Norrtälje. The advantage of the Johansen approach is that it allows me to formally test hypotheses about long- and short-run parameters in a multivariate setup, i.e. the size of pass-through rates and the direction of adjustment to changes in costs (Juselius, 2006; Hoover, Johansen and Juselius, 2008). Figures 5 and 6 depict real retail prices net of VAT for the four largest blends in terms of volume market shares nationally: Gevalia Mellanrost Brygg, Gevalia Mellanrost E-brygg, Classic Mellan Brygg and Zoegas Skånerost. By far the largest blend is Gevalia Mellanrost Brygg, which has a market share of about 20 per cent in eastern Sweden. Classic Mellan Brygg has a market share of 14 per cent, while Gevalia Mellanrost E-brygg and Zoegas Skånerost have about 5 per cent each. Fig. 5. View largeDownload slide Real prices of Gevalia Mellanrost Brygg, Gevalia Mellanrost E-brygg and the cost of beans. Fig. 5. View largeDownload slide Real prices of Gevalia Mellanrost Brygg, Gevalia Mellanrost E-brygg and the cost of beans. Fig. 6. View largeDownload slide Real prices of Classic Mellan Brygg and Zoegas Skånerost and the cost of beans. Fig. 6. View largeDownload slide Real prices of Classic Mellan Brygg and Zoegas Skånerost and the cost of beans. As evident from the figures, all prices seem to be non-stationary and follow a pattern similar to the cost of coffee beans. Moreover, the prices of Gevalia Mellanrost Brygg and Gevalia Mellanrost E-brygg are practically identical, while Classic Mellan Brygg is a bit more expensive than the Gevalia coffees and Zoegas Skånerost is clearly more expensive. There is a notable increase in the difference between Zoegas Skånerost and the others in 2012, as the price of Zoegas declines more slowly than the other prices, but in 2014, the difference has returned to the pre-2012 level. The data are not informative of whether this deviation was a one-time event. Since the two Gevalia prices provide the same information, I drop Gevalia Mellanrost E-brygg from the analysis. The Johansen approach is based on the VAR model specified in error correction form: Δxt=αβ'xt+Γ1Δxt−1+ΦDt+εt (5) where xt is a vector of potentially endogenous variables, Δ is the first difference operator, D is a vector of deterministic terms, such as constant and seasonal dummies and εt is white noise (see Juselius, 2006). The lag length of the corresponding VAR in levels is set to two for illustrative purposes. The coefficients and hypotheses tested are easiest to understand if the model is (partly) written in matrix form. I assume (and later show) that there are three long-run relations (cointegrating vectors). We then have (pGevpClapZoecbea)=(α11α12α13α21α22α23α31α32α33α41α42α43)(β1'xt−1β2'xt−1β3'xt−1)t+Γ1Δxt−1+ΦDt+(ε1tε2tε3tε4t), (6) where pGev,pCla,pZoe and cbea are product retail prices and bean cost, αii is an adjustment coefficient, βi'xt is an error correction term for a long-run relationship and Γ1 is a matrix of short-run coefficients. With four variables (each of which has a unit root) and three long-run relationships, there is one common trend, presumably resulting from the stochastic trend of the cost of coffee beans. This implies that there is a long-run relationship, βi'xt, between each retail price and the cost of beans, though other combinations are possible, in principle. When there is complete pass-though, the coefficients of the three vectors β1',β2' and β3' are (1, 0, 0, −1), (0, 1, 0, −1) and (0, 0, 1, −1), which is a testable hypothesis. I can also test whether cbea is weakly exogenous, i.e. whether α41=α42=α43=0, which would indicate that the cost of beans does not adjust to maintain any long-run relationship, a reasonable assumption given the small size of the Swedish coffee market. I can further test whether all adjustments are due to coffee bean costs or there is interaction between retail prices such that for example the price of Gevalia Mellanrost Brygg influences the adjustment of the other two prices. In this case, α21≠0,α31≠0. Finally, the model provides estimates of the speed of the adjustment to the long-run relationship through the values of α11,α22andα33, and of whether there is short-run adjustment, Γ1, due to lagged changes in Δxt. A key step in the analysis is to determine the cointegration rank (assumed to be three above), i.e. the number of long-run relationships. The Johansen approach uses the trace test, which is based on the maximum likelihood procedure. Given that I only have 66 observations, additional information should also be used to determine the rank (Juselius, 2001). I use the point estimates of the eigenvalues, which should be clearly larger than zero, the size of the adjustment coefficients, and economic reasoning. Table 4 reports ADF unit root tests, trace tests, estimated coefficients and likelihood ratio test of restrictions on the system. The ADF tests (Panel 1) indicate that all four variables have a unit root; none of the tests statistics are significant and all estimated roots are above 0.9. Panel 2 reports the trace test statistics for the VAR model with two and three lags; the Hannan-Quinn and Schwartz information criteria clearly favoured a model with two lags, while the Akaike criterion selected a model with three lags. Adding up to five lags to the model does not change any findings. The trace test shows that there are three cointegrating vectors in the two-lag model, i.e. the rank = 2 is rejected, while there are four cointegrating vectors in the three-lag model, as the rank = 3 is rejected. If there were four cointegrating vectors, all variables would be stationary, contradicting the ADF unit root tests. Since the eigenvalues for rank = 4 are close to zero (0.04 and 0.08) and the adjustment coefficients for the product retail prices α11,α22andα33 are large in absolute terms and the one for the cost of beans, α44, is close to zero, −0.04 (Panel 3), I proceed under the assumptions that the rank is 3 and there are three cointegrating vectors.15 Table 4. Cointegration tests and hypotheses tests on the three main coffees in eastern Sweden Panel ADF unit root tests: two lags and a trend. 5% = −3.48, 1% = −4.10 pGev pCla pZoe cbea 1 t-adf −1.38 −1.65 −1.24 −1.83 Root 0.92 0.91 0.93 0.95 Johansen approach 2 Trace test of number of cointegrating vectors: H0: rank≤ Two lags Three lags Eigenvalue p-value Eigenvalue p-value Rank 0 – [0.000]** – [0.000]** Rank 1 0.54 [0.000]** 0.41 [0.005]** Rank 2 0.46 [0.027]* 0.22 [0.008]** Rank 3 0.18 [0.092] 0.19 [0.018]* Rank 4 0.04 – 0.08 – 3 Adjustment coefficients α11,α22,α33,α44, unrestricted 3-lag model pGev pZoe pCla cbea −0.48 −0.16 −0.36 −0.04 4 Long-run normalised coefficients, βii, for three vectors pGev 1.00 0.00 0.00 pZoe 0.00 1.00 0.00 pCla 0.00 0.00 1.00 cbea −0.90 −0.98 −0.98 (0.05) (0.19) (0.06) Adjustment coefficients and standard errors pGev −0.79 0.07 −0.11 (0.17) (0.07) (0.18) pZoe −0.08 −0.23 −0.23 (0.21) (0.08) (0.22) pCla −0.06 0.01 −0.63 (0.16) (0.06) (0.16) cbea −0.03 −0.06 −0.05 (0.11) (0.04) (0.11) 5 LR test of complete pass-through: χ2(3)=3.43[0.33] β21=β31=β12=β32=β13=β23=0;β11=β22=β33=1;β41=β42=β43=−1; 6 LR test χ2(8)=8.13[0.42] β21=β31=β12=β32=β13=β23=0;β11=β22=β33=1;β41=β42=β43=−1; α12=α13=α21=α23=α31=α32=α41=α42=α43=α41=0 Panel ADF unit root tests: two lags and a trend. 5% = −3.48, 1% = −4.10 pGev pCla pZoe cbea 1 t-adf −1.38 −1.65 −1.24 −1.83 Root 0.92 0.91 0.93 0.95 Johansen approach 2 Trace test of number of cointegrating vectors: H0: rank≤ Two lags Three lags Eigenvalue p-value Eigenvalue p-value Rank 0 – [0.000]** – [0.000]** Rank 1 0.54 [0.000]** 0.41 [0.005]** Rank 2 0.46 [0.027]* 0.22 [0.008]** Rank 3 0.18 [0.092] 0.19 [0.018]* Rank 4 0.04 – 0.08 – 3 Adjustment coefficients α11,α22,α33,α44, unrestricted 3-lag model pGev pZoe pCla cbea −0.48 −0.16 −0.36 −0.04 4 Long-run normalised coefficients, βii, for three vectors pGev 1.00 0.00 0.00 pZoe 0.00 1.00 0.00 pCla 0.00 0.00 1.00 cbea −0.90 −0.98 −0.98 (0.05) (0.19) (0.06) Adjustment coefficients and standard errors pGev −0.79 0.07 −0.11 (0.17) (0.07) (0.18) pZoe −0.08 −0.23 −0.23 (0.21) (0.08) (0.22) pCla −0.06 0.01 −0.63 (0.16) (0.06) (0.16) cbea −0.03 −0.06 −0.05 (0.11) (0.04) (0.11) 5 LR test of complete pass-through: χ2(3)=3.43[0.33] β21=β31=β12=β32=β13=β23=0;β11=β22=β33=1;β41=β42=β43=−1; 6 LR test χ2(8)=8.13[0.42] β21=β31=β12=β32=β13=β23=0;β11=β22=β33=1;β41=β42=β43=−1; α12=α13=α21=α23=α31=α32=α41=α42=α43=α41=0 Note: Panel 1 reports Augmented Dickey–Fuller unit root tests, t-values and the estimated roots. Panel 2 reports the trace tests and eigenvalues. The null is that the rank is equal to or less than a certain number when the eigenvalue is zero. Models with two and three lags are tested. Panel 3 reports the adjustment coefficients for each variable when the model is estimated with three lags. Panel 4 reports the long-run and the adjustment coefficients when there are assumed to be three cointegrating vectors. Panel 5 reports the likelihood ratio test that pass-through is 1, and Panel 6 reports the test that pass-through is I and all the adjustment is due to costs. View Large Table 4. Cointegration tests and hypotheses tests on the three main coffees in eastern Sweden Panel ADF unit root tests: two lags and a trend. 5% = −3.48, 1% = −4.10 pGev pCla pZoe cbea 1 t-adf −1.38 −1.65 −1.24 −1.83 Root 0.92 0.91 0.93 0.95 Johansen approach 2 Trace test of number of cointegrating vectors: H0: rank≤ Two lags Three lags Eigenvalue p-value Eigenvalue p-value Rank 0 – [0.000]** – [0.000]** Rank 1 0.54 [0.000]** 0.41 [0.005]** Rank 2 0.46 [0.027]* 0.22 [0.008]** Rank 3 0.18 [0.092] 0.19 [0.018]* Rank 4 0.04 – 0.08 – 3 Adjustment coefficients α11,α22,α33,α44, unrestricted 3-lag model pGev pZoe pCla cbea −0.48 −0.16 −0.36 −0.04 4 Long-run normalised coefficients, βii, for three vectors pGev 1.00 0.00 0.00 pZoe 0.00 1.00 0.00 pCla 0.00 0.00 1.00 cbea −0.90 −0.98 −0.98 (0.05) (0.19) (0.06) Adjustment coefficients and standard errors pGev −0.79 0.07 −0.11 (0.17) (0.07) (0.18) pZoe −0.08 −0.23 −0.23 (0.21) (0.08) (0.22) pCla −0.06 0.01 −0.63 (0.16) (0.06) (0.16) cbea −0.03 −0.06 −0.05 (0.11) (0.04) (0.11) 5 LR test of complete pass-through: χ2(3)=3.43[0.33] β21=β31=β12=β32=β13=β23=0;β11=β22=β33=1;β41=β42=β43=−1; 6 LR test χ2(8)=8.13[0.42] β21=β31=β12=β32=β13=β23=0;β11=β22=β33=1;β41=β42=β43=−1; α12=α13=α21=α23=α31=α32=α41=α42=α43=α41=0 Panel ADF unit root tests: two lags and a trend. 5% = −3.48, 1% = −4.10 pGev pCla pZoe cbea 1 t-adf −1.38 −1.65 −1.24 −1.83 Root 0.92 0.91 0.93 0.95 Johansen approach 2 Trace test of number of cointegrating vectors: H0: rank≤ Two lags Three lags Eigenvalue p-value Eigenvalue p-value Rank 0 – [0.000]** – [0.000]** Rank 1 0.54 [0.000]** 0.41 [0.005]** Rank 2 0.46 [0.027]* 0.22 [0.008]** Rank 3 0.18 [0.092] 0.19 [0.018]* Rank 4 0.04 – 0.08 – 3 Adjustment coefficients α11,α22,α33,α44, unrestricted 3-lag model pGev pZoe pCla cbea −0.48 −0.16 −0.36 −0.04 4 Long-run normalised coefficients, βii, for three vectors pGev 1.00 0.00 0.00 pZoe 0.00 1.00 0.00 pCla 0.00 0.00 1.00 cbea −0.90 −0.98 −0.98 (0.05) (0.19) (0.06) Adjustment coefficients and standard errors pGev −0.79 0.07 −0.11 (0.17) (0.07) (0.18) pZoe −0.08 −0.23 −0.23 (0.21) (0.08) (0.22) pCla −0.06 0.01 −0.63 (0.16) (0.06) (0.16) cbea −0.03 −0.06 −0.05 (0.11) (0.04) (0.11) 5 LR test of complete pass-through: χ2(3)=3.43[0.33] β21=β31=β12=β32=β13=β23=0;β11=β22=β33=1;β41=β42=β43=−1; 6 LR test χ2(8)=8.13[0.42] β21=β31=β12=β32=β13=β23=0;β11=β22=β33=1;β41=β42=β43=−1; α12=α13=α21=α23=α31=α32=α41=α42=α43=α41=0 Note: Panel 1 reports Augmented Dickey–Fuller unit root tests, t-values and the estimated roots. Panel 2 reports the trace tests and eigenvalues. The null is that the rank is equal to or less than a certain number when the eigenvalue is zero. Models with two and three lags are tested. Panel 3 reports the adjustment coefficients for each variable when the model is estimated with three lags. Panel 4 reports the long-run and the adjustment coefficients when there are assumed to be three cointegrating vectors. Panel 5 reports the likelihood ratio test that pass-through is 1, and Panel 6 reports the test that pass-through is I and all the adjustment is due to costs. View Large Panel 4 reports the long-run coefficients for product prices set to either 1 or 0 in each equation. Note that no restrictions have been imposed on the system, except for the cointegration rank, even though five long-run coefficients are zero. The estimates of the coefficients of the bean costs are all negative and close to 1. The adjustment coefficients, also reported in Panel 4, show that all adjustments are probably due to changes in the individual product retail prices; there is no evidence of one product retail price affecting the others since all αij are close to zero and insignificant. Next, I test the restrictions that all the bean cost coefficients are −1, i.e. that the pass-through is 1 (Panel 5). The likelihood ratio test static has a p-value of 0.33, so the hypothesis is not rejected. Then I set adjustment coefficients to zero to test whether the retail prices affect each other (Panel 6). The p-value is 0.42, so there is no evidence that adjustment to the long-run equilibrium in one vector affects other prices. Finally, to highlight the short-run dynamics, I report the vector error correction model in Table 5, where clearly insignificant lags have been excluded.16 The error correction terms are lagged 2 months, since it takes several weeks for the imported beans to be processed and reach the market, though lagging them one month does not alter the results much. There is fairly rapid adjustment after a change in costs for the Gevalia and Classic coffees: About 60 per cent of a deviation from the long-run equilibrium is eliminated within a month. Moreover, lagged changes in bean costs affect the prices of Gevalia and Classic coffees, and changes in the price of Gevalia speed up changes in the price of Classic coffee in the following month. Table 5. The error correction model (ΔpGevΔpZoeΔpCla)t=(−0.58000−0.26000−0.60) (pGev−cbeapZoe−cbeapCla−cbea )t−2+ (−0.76000.500−0.5000.260−0.730.60 ) (ΔpGevΔpZoeΔpClaΔcbea )t−1+D (ΔpGevΔpZoeΔpCla)t=(−0.58000−0.26000−0.60) (pGev−cbeapZoe−cbeapCla−cbea )t−2+ (−0.76000.500−0.5000.260−0.730.60 ) (ΔpGevΔpZoeΔpClaΔcbea )t−1+D Note: All coefficients are significant at the 5 per cent level, at least. See Tables A1 and A2 in the Appendix for the general and parsimonious models. View Large Table 5. The error correction model (ΔpGevΔpZoeΔpCla)t=(−0.58000−0.26000−0.60) (pGev−cbeapZoe−cbeapCla−cbea )t−2+ (−0.76000.500−0.5000.260−0.730.60 ) (ΔpGevΔpZoeΔpClaΔcbea )t−1+D (ΔpGevΔpZoeΔpCla)t=(−0.58000−0.26000−0.60) (pGev−cbeapZoe−cbeapCla−cbea )t−2+ (−0.76000.500−0.5000.260−0.730.60 ) (ΔpGevΔpZoeΔpClaΔcbea )t−1+D Note: All coefficients are significant at the 5 per cent level, at least. See Tables A1 and A2 in the Appendix for the general and parsimonious models. View Large The equations in Table 5 look unintuitive because of the negative coefficients. Thus, to see how prices evolve over time, it is convenient to re-write the error correction model into levels. Equation (7) shows models for the prices of Gevalia and Zoegas Skåne; Classic Mellanrost is similar to Gevalia, pGevt=c+0.24pGev,t−1+0.18pGev,t−2+0.50cbea,t−1+0.08cbea,t−2pZoet=c+0.50pZoe,t−1+0.24pZoe,t−2+0.26cbea,t−2. (7) Both prices respond to an increase in bean costs with a lag. Then there is adjustment over the following months: after a 1 SEK increase in bean cost, the price of Gevalia increases by 0.50 SEK in the following month, and reaches 1 SEK after about 12 months. The price of Zoegas adjusts somewhat more slowly after a cost increase, rising to 1 SEK over 18 months. This might be because of the sluggish price decline during period 2012–2014, shown in Figure 6. 7. Concluding remarks The purpose of this paper is to shed light on the functioning of the Swedish retail coffee market by estimating long-run pass-through rates from the cost of green coffee beans to retail prices and analysing how they are related to horizontal and vertical market structure. The focus is on roasted and ground coffee, the market segment analysed in almost all previous studies on coffee markets. The Swedish market, which is similar to other Northern European markets, has four large brands and a very concentrated food retail sector with three dominating food chains (Swedish Competition Authority, 2011). I estimate that a 1 SEK cost increase raises prices by 0.69 SEK. However, pass-through rates are highly disbursed and vary from about 0 to slightly above 1. As these estimates include both the direct effect of the change in costs and the indirect effects due to strategic interaction, market structure may play a key role in explaining the dispersion. In fact, the four large brands, each of which sells several blends, have average pass-through rates that are 0.30–0.40 SEK higher than other brands. Another key finding is that the association between pass-through and market share is positive, though non-linear as a pass-through of about 1 seems to be an upper limit. This relationship seems to primarily be due to the market share of the blend and not the brand, i.e. pass-through rates are low for blends with small market shares irrespective of whether they are of a large or small brand. This finding of a positive association between pass-through and market share is the opposite of the prediction of Dornbusch’s (1987) model and the findings of Atkeson and Burstein (2008) and Hong and Li (2017), among others. Yet another finding is that vertically integrated blends do not have high pass-through rates, which is contrary to the results of Hong and Lee (2017). Discount, and to some extent premium, private labels have low pass-through rates relative to their market shares, while standard private labels have pass-through rates consistent with their market shares. Thus, there is no evidence of double marginalisation. The cointegration analysis of the dominant blends in eastern Sweden strengthens earlier findings. It is not possible to reject a pass-through rate of 1 for any of the four products. The adjustment to a cost shock takes about a year to be completed, probably due to the combined effect of the time it takes to process coffee beans, the existing stocks in food stores and the bargaining between roasters and retail chains. Since the four largest blends have a volume market share of about 50 per cent and the 10 largest a market share of nearly 70 per cent, the finding of pass-through rates of about 1 for large blends indicates that a substantial part of the Swedish coffee market is highly competitive. One reason earlier studies on the Swedish coffee market have not found evidence of market power is most likely their use of average retail prices collected by Statistics Sweden, which were based on a small selection of prices of popular blends.17 The question is why products with small market shares have low pass-through rates and vice versa. One hypothesis is that it is due to the functional form of consumer demand function, which would have to be concave and vary systematically across market shares, such that the smaller the market share the larger the response of the price elasticity to a price change. However, it is not obvious why the demand for small blends should respond more to price changes than the demand for large blends. Another hypothesis is that the regularly occurring bargaining between roasters and retail chains generates the relationship. Roasters might face strong resistance from retail chains when they wish to raise the wholesale prices of blends with small market shares, since retailers prefer to fill their shelves with the most popular blends or their own brands. And retail chains might be reluctant to lower the retail prices of small blends when wholesale prices go down, since they wish to increase their markups on blends that sell little. Prices on large blends, on the other hand, are changed when costs change, because both roasters and retailers compete on market shares. To conclude, the Swedish ground coffee market seems to be far from a typical monopolistic or oligopolistic market; pass-through rates are high, particularly for popular blends. In fact, a large part of the market seems to be quite competitive, although there are deviations from perfect competition. Nevertheless, more research is needed to explain the systematic positive relationship between pass-through rates and market shares. Supplementary data Supplementary data are available at European Review of Agricultural Economics online. Acknowledgements The author would like to thank Sven-Olof Daunfeldt, Niklas Rudholm, three anonymous reviewers and the editor for helpful comments. Financial support from the Swedish Competition Authority is gratefully acknowledged. Footnotes 1 See e.g. Morisset (1998), Dicum and Luttinger (1999), Fitter and Kaplinsky (2001), Oxfam (2002), Moore (2003), McCorriston, Sexton and Sheldon (2004), Talbot (2004, 2011), Consumers International (2005), Daviron and Ponte (2005), Green (2005), Gibbon (2007), Levy (2008), Fairtrade Foundation (2012) and World Vision (2014). 2 The behaviour of large coffee retail prices is similar in the other regional markets. 3 Since my measure of marginal cost is an industry-wide cost shock, the study is related to studies on tax incidence and on studies in international economics that analyse the effect of exchange rate changes on retail prices. 4 See RBB (2014) for a general review of pass-through. 5 The Dixit–Stiglitz model of monopolistic competition has identical firms and constant markup, and thus a pass-through rate of 1. Dornbusch (1987) introduces the assumption that firms take the effect of its own price on the overall price index into account, which makes pass-through lower the larger the impact a firm’s price has on the price index. The Dornbusch (1987) model has become a workhorse in international economics (Burstein and Gopinath, 2014). 6 Adachi and Ebina (2014a, 2014b) discuss the case of vertical relationships with homogeneous manufacturers and retailers in detail. 7 See Gaudin (2016) for the role of relative bargaining power between wholesalers and retailers in determining pass-through. 8 A related question is whether prices rise faster than they fall after a change in marginal costs (Peltzman, 2000). Asymmetry is usually considered a short-run phenomenon, which I ignore in the present study. In any case, it is difficult to identify asymmetry with 6 years of data when costs are highly persistent (see Figure 1). 9 There are a few non-organic Fairtrade blends in the Swedish market but their sales are tiny and they are not included in the sample. Including a dummy for organic Fairtrade blends in the regression does not affect the results. 10 It is also possible to check the effect of biases by re-scaling costs, up or down, by reasonable amounts. For example, when costs are 20 per cent higher than average costs, the actual pass-through rate is roughly 20 per cent lower than the estimate, and vice versa. A 20 per cent deviation from the average cost is large given that the quality of Arabica beans used is mainstream (International Trade Centre, 2011, chap. 11) and that most of them are substitutes. As reported below by Figure 2, the estimates of the pass-through rates vary much more than can be explained by the measurement errors. 11 The weight lost from roasting is common knowledge. See e.g. European Coffee Federation (2011). 12 Hassouneh et al. (2012) provide a review of approaches used when analysing fairly long-time series of market data. 13 If we assume that other marginal costs followed the consumer price index, as Bettendorf and Verboven (2000) do, they would have increased, roughly monotonically, by about 5 per cent in total over the study period. Wage costs probably increased a bit more, but they make up small components of costs (Durevall, 2007b). 14 Since I estimate many regressions, some of the key results are reported in Figures 2 and 4 and Figures A1 and A2 in the Appendix. The rest of the results are available on request. 15 In principle, there may be more than one common trend in the model because of other real marginal costs. If these contained a stochastic trend, product retail prices would not form a cointegrating vector with coffee bean costs only. However, as mentioned, other real marginal costs evolved slowly during the study period (relative to bean prices) and they do not seem to matter for the long-run relationships. In fact, other marginal costs are captured by the intercept, as deterministic trends added to the VAR model were insignificant (not reported). 16 No contemporaneous retail prices enter the model. This is partly because I do not have any instruments, but it is primarily because there is very little correlation between the prices when measured in rates of change. I condition on contemporaneous coffee bean costs in the general model since they are weakly exogenous. However, all the coefficients are insignificant, as one would expect (see Table A1 in the Appendix) . 17 Since 2012, Statistics Sweden uses cash register data to construct indices of food retail prices. Prior to that year, small samples of products were selected randomly with a probability based on their market share (Statistics Sweden, 2007, 2014). As a result, the series of average prices reported by Statistics Sweden for ground coffee primarily comprised coffees with large market shares. References Adachi , T. and Ebina , T. ( 2014 a). Cost pass-through and inverse demand curvature in vertical relationships with upstream and downstream competition . Economics Letters 124 ( 3 ): 465 – 468 . Google Scholar CrossRef Search ADS Adachi , T. and Ebina , T. ( 2014 b). Double marginalization and cost pass-through: Weyl–Fabinger and Cowan meet Spengler and Bresnahan–Reiss . Economics Letters 122 ( 2 ): 170 – 175 . Google Scholar CrossRef Search ADS Amiti , M. , Itskhoki , O. and Konings , J. ( 2014 ). Importers, exporters, and exchange rate disconnect . American Economic Review 104 ( 7 ): 1942 – 1978 . Google Scholar CrossRef Search ADS Anderson , S. P. , De Palma , A. and Kreider , B. ( 2001 ). Tax incidence in differentiated product oligopoly . Journal of Public Economics 81 ( 2 ): 173 – 192 . Google Scholar CrossRef Search ADS Aron , J. , Macdonald , R. and Muellbauer , J. ( 2014 ). Exchange rate pass-through in developing and emerging markets: a survey of conceptual, methodological and policy issues, and selected empirical findings . Journal of Development Studies 50 ( 1 ): 101 – 143 . Google Scholar CrossRef Search ADS Atkeson , A. and Burstein , A. ( 2008 ). Pricing-to-market, trade costs, and international relative prices . American Economic Review 98 ( 5 ): 1998 – 2031 . Google Scholar CrossRef Search ADS Auer , R. A. and Schoenle , R. S. ( 2016 ). Market structure and exchange rate pass-through . Journal of International Economics 98 : 60 – 77 . Google Scholar CrossRef Search ADS Besanko , D. , Dubé , J.-P. and Gupta , S. ( 2005 ). Own-brand and cross-brand retail pass-through . Marketing Science 24 ( 1 ): 123 – 137 . Google Scholar CrossRef Search ADS Bettendorf , L. and Verboven , F. ( 2000 ). Incomplete transmission of coffee bean prices: evidence from the Netherlands . European Review of Agricultural Economics 27 ( 1 ): 1 – 16 . Google Scholar CrossRef Search ADS Bonnet , C. , Dubois , P. , Villas-Boas , S. B. and Klapper , D. ( 2013 ). Empirical evidence on the role of nonlinear wholesale pricing and vertical restraints on cost pass-through . Review of Economics and Statistics 95 ( 2 ): 500 – 515 . Google Scholar CrossRef Search ADS Bonnet , C. and Villas-Boas , S. B. ( 2016 ). An analysis of asymmetric consumer price responses and asymmetric cost pass-through in the French coffee market . European Review of Agricultural Economics 43 ( 5 ): 781 – 804 . Google Scholar CrossRef Search ADS Bulow , J. I. and Pfleiderer , P. ( 1983 ). A note on the effect of cost changes on prices . Journal of Political Economy 91 ( 1 ): 182 – 185 . Google Scholar CrossRef Search ADS Burstein , A. and Gopinath , G. ( 2014 ). International prices and exchange rates. In: Gopinath, G., Helpman, E. and Rogoff, K. (eds), Handbook of International Economics , Vol. 4 . Amsterdam : Elsevier , 391 – 451 . Consumers International . ( 2005 ). From Bean to Cup: How Consumer Choice Impacts upon Coffee Producers and the Environment . London : Consumers International . Daviron , B. and Ponte , S. ( 2005 ). The Coffee Paradox: Global Markets, Commodity Trade and the Elusive Promise of Development . London : ZED Books . Dicum , G. and Luttinger , N. ( 1999 ). The Coffee Book: Anatomy of an Industry From Crop to the Last Drop . New York : New Press . Dornbusch , R. ( 1987 ). Real exchange rate and prices . American Economic Review 77 ( 1 ): 93 – 106 . Durevall , D. ( 2003 ). Competition and pricing: an analysis of the market for roasted coffee. Chap. 5 in High Prices in Sweden—A Result of Poor Competition? Swedish Competition Authority Report. Durevall , D. ( 2007 a). Demand for coffee: the role of prices, preferences and market power . Food Policy 32 ( 5–6 ): 566 – 584 . Google Scholar CrossRef Search ADS Durevall , D. ( 2007 b). Competition in the Swedish coffee market 1978–2002 . International Journal of Industrial Organization 25 ( 4 ): 721 – 739 . Google Scholar CrossRef Search ADS Eichenbaum , M. , Jaimovich , N. and Rebelo , S. ( 2011 ). Reference prices, costs, and nominal rigidities . American Economic Review 101 ( 1 ): 234 – 262 . Google Scholar CrossRef Search ADS Ericsson , N. R. and MacKinnon , J. G. ( 2002 ). Distributions of error correction tests for cointegration . Econometrics Journal 5 ( 2 ): 285 – 318 . Google Scholar CrossRef Search ADS European Coffee Federation . ( 2011 ). European Coffee Report 2010/2011. Rijswijk, The Netherlands. Fairtrade Foundation . ( 2012 ). Fairtrade and Coffee, Commodity Briefing, London. Fairtrade International . ( 2016 ). Driving Sales, Deepening Impact: Annual Report 2015–2016. https://annualreport15-16.fairtrade.net/en/. Accessed 9 February 2018. Feenstra , R. C. ( 2016 ). Advanced International Trade: Theory and Evidence . Princeton: Princeton University Press . Feuerstein , S. ( 2002 ). Do coffee roasters benefit from high prices of green coffee? International Journal of Industrial Organization 20 ( 19 ): 89 – 118 . Google Scholar CrossRef Search ADS Fitter , R. and Kaplinsky , R. ( 2001 ). Who gains from product rents as the coffee market becomes more differentiated? A value chain analysis . IDS Bulletin 32 ( 3 ): 69 – 82 . Google Scholar CrossRef Search ADS Fousekis , P. and Grigoriadis , V. ( 2017 ). Joint price dynamics of quality differentiated commodities: copula evidence from coffee varieties . European Review of Agricultural Economics 44 ( 2 ): 337 – 358 . Gaudin , G. ( 2016 ). Pass-through, vertical contracts, and bargains . Economics Letters 139 : 1 – 4 . Google Scholar CrossRef Search ADS Gibbon , P. ( 2007 ). Africa, tropical commodity policy and the WTO Doha Round . Development Policy Review 25 ( 1 ): 43 – 70 . Google Scholar CrossRef Search ADS Gilbert , C. ( 2007 ). Have we been mugged? Market Power in the Coffee Industry. Working Paper, CIFERM and Department of Economics, Università degli studi di Trento. Gilbert , C. ( 2008 ). Value chain analysis and market power in commodity processing with application to the cocoa and coffee sectors. Commodity Market Review 2007–08, 5–38, Food and Agriculture Organization of the United Nations, Rome. Green , D. ( 2005 ). Conspiracy of silence: old and new directions on commodities. Report, Oxfam, London. Hassouneh , I. , von Cramon-Taubadel , S. , Serra , T. and Gil , J. M. ( 2012 ). Recent developments in the econometric analysis of price transmission. Working Paper (2) , Transparency of Food Pricing (TRANSFOP) . Hong , G. H. and Li , N. ( 2017 ). Market structure and cost pass-through in retail . Review of Economics and Statistics 99 ( 1 ): 151 – 166 . Google Scholar CrossRef Search ADS Hoover , K. D. , Johansen , S. and Juselius , K. ( 2008 ). Allowing the data to speak freely: the macroeconometrics of the cointegrated vector autoregression . The American Economic Review 98 ( 2 ): 251 – 255 . Google Scholar CrossRef Search ADS ITA ( 2011 ). The Coffee Exporter’s Guide . International Trade Centre, Geneva . Johansen , S. ( 1991 ). Estimation and hypothesis testing of cointegration vectors in Gaussian vector autoregressive models . Econometrica 59 ( 6 ): 1551 – 1580 . Google Scholar CrossRef Search ADS Juselius , K. ( 2001 ). Unit roots and the demand for cigarettes in Turkey: Pitfalls and possibilities. Discussion Paper at the Economics Department, University of Copenhagen. Juselius , K. ( 2006 ). The Cointegrated VAR Model: Econometric Methodology and Empirical Applications . Oxford: Oxford University Press . Leibtag , E. , Nakamura , A. , Nakamura , E. and Zerom , D. ( 2007 ). Cost pass-through in the U.S. coffee industry. USDA Economic Research Report No. (ERR-38). Levy , D. L. ( 2008 ). Political contestation in global production networks . Academy of Management Review 33 ( 4 ): 943 – 963 . Google Scholar CrossRef Search ADS Lloyd , T. ( 2017 ). Forty years of price transmission research in the food industry: insights, challenges and prospects . Journal of Agricultural Economics 68 ( 1 ): 3 – 21 . Google Scholar CrossRef Search ADS Lloyd , T. A. , McCorriston , S. , Morgan , C. W. , Poen , E. and Zgovu , E. ( 2014 ). Retail price dynamics and retailer heterogeneity: UK evidence . Economics Letters 124 ( 3 ): 434 – 438 . Google Scholar CrossRef Search ADS McCorriston , S. ( 2013 ). Competition in the Food Chain. Working Paper (11), Transparency for Food Pricing (TRANSFOP). McCorriston , S. , Sexton , R. J. and Sheldon , I. M. ( 2004 ). Vertical Market Structure, Commodity Exports and Trade Reform. Paper presented at 7th Annual Conference on Global Economic Analysis, World Bank, June 2004, Washington, DC, USA. Moore , M. ( 2003 ). A World without Walls: Freedom, Development, Free Trade and Global Governance . Cambridge : Cambridge University Press . Moorthy , S. ( 2005 ). A general theory of pass-through in channels with category management and retail competition . Marketing Science 24 ( 1 ): 110 – 122 . Google Scholar CrossRef Search ADS Morisset , J. ( 1998 ). Unfair trade? The increasing gap between world and domestic prices in commodity markets during the past 25 years . World Bank Economic Review 12 ( 3 ): 503 – 526 . Google Scholar CrossRef Search ADS Nakamura , E. and Zerom , D. ( 2010 ). Accounting for incomplete pass-through . Review of Economic Studies 77 ( 3 ): 1192 – 1230 . Google Scholar CrossRef Search ADS Oxfam ( 2002 ). Mugged: Poverty in Your Coffee Cup . London : Oxfam International and Make Fair Trade . Peltzman , S. ( 2000 ). Prices rise faster than they fall . Journal of Political Economy 108 ( 3 ): 466 – 502 . Google Scholar CrossRef Search ADS RBB Economics . ( 2014 ). Cost pass-through: theory, measurement, and potential policy implications. Report prepared for the Office of Fair Trading, London. Rohman , A. and Friberg , R. ( 2016 ). Price rigidity in the Swedish beer market. Beiträge zur Jahrestagung des Vereins für Socialpolitik 2016: Demographischer Wandel–Session: Price-Setting Behavior of Firms, No. G11–V3. Sexton , R. J. ( 2013 ). Market power, misconceptions, and modern agricultural markets . American Journal of Agricultural Economics 95 ( 2 ): 209 – 219 . Google Scholar CrossRef Search ADS Statistics Sweden . ( 2007 ). Konsumentprisindex, Nettoprisindex, Harmoniserat index för konsumentpriser 2007, PR0101. Stockholm, Statistiska centralbyrån. Statistics Sweden . ( 2014 ). Konsumentprisindex (KPI) 2014 SCBDOK 3.2 PR0101. Stockholm, Statistiska centralbyrån. Sutton , J. ( 2007 ). Sunk Cost and Market Structure . Cambridge, MA : MIT Press . Swedish Competition Authority . ( 2011 ). Konkurrens och makt I den svenska livsmedelskedjan. Report prepared by Agrifood Economics Centre, Lund. Talbot , J. M. ( 2004 ). Grounds for Agreement: The Political Economy of the Coffee Commodity Chain . Lanham : Rowman and Littlefield . Talbot , J. M. ( 2011 ). The coffee commodity chain in the world-economy: Arrighi’s systemic cycles and Braudel’s layers of analysis . Journal of World-Systems Research 17 ( 1 ): 58 – 88 . Google Scholar CrossRef Search ADS Tyagi , R. K. ( 1999 ). A characterization of retailer response to manufacturer trade deals . Journal of Marketing Research 36 ( 4 ): 510 – 516 . Google Scholar CrossRef Search ADS Weyl , E. G. and Fabinger , M. ( 2013 ). Pass-through as an economic tool: principles of incidence under imperfect competition . Journal of Political Economy 121 ( 3 ): 528 – 583 . Google Scholar CrossRef Search ADS World Vision . ( 2014 ). Coffee’s Hidden Kick: Labour Exploitation in the Global Coffee Industry. DTL Fact Sheet Coffee, World Vision Australia. Author notes Review coordinated by Iain Fraser © Oxford University Press and Foundation for the European Review of Agricultural Economics 2018; all rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png European Review of Agricultural Economics Oxford University Press

Cost pass-through in the Swedish coffee market

Loading next page...
 
/lp/ou_press/cost-pass-through-in-the-swedish-coffee-market-xm8d9iY3GF
Publisher
Oxford University Press
Copyright
© Oxford University Press and Foundation for the European Review of Agricultural Economics 2018; all rights reserved. For permissions, please e-mail: journals.permissions@oup.com
ISSN
0165-1587
eISSN
1464-3618
D.O.I.
10.1093/erae/jby002
Publisher site
See Article on Publisher Site

Abstract

Abstract Cost pass-through to retail prices shows how changes in marginal costs are allocated between producers and consumers, and it is, therefore, closely related to market structure and competition. This paper uses Swedish data on coffee products at the barcode level to evaluate pass-through from the cost of green coffee beans, the main marginal cost, to the retail price of roasted and ground coffee. First long-run cost pass-through is estimated for each product, and then regression is used to analyse how pass-through varies across market shares, retailer-owned brands and other product characteristics. A general result is that pass-through is roughly complete for products with large market shares, while those with small market shares have low pass-through rates. There is no evidence that retailer-owned brands have higher pass-through than brand-name products with similar market shares, which would be the case if retailer-owned brands avoided double marginalisation through vertical integration. Thus, although there is not perfect competition in the Swedish coffee market, a large part of it appears to be highly competitive. 1. Introduction Many consumer markets for coffee are characterised by a few dominant multinational or national roasters, with a combined market share often exceeding 60 per cent, and a large number of small roasters and importers (Sutton, 2007). Not surprisingly, the large roasters are regularly alleged to charge high prices for processed coffee relative to the price of their main input, i.e. green coffee beans.1 The rapid growth of Fairtrade coffee sales, which increased by about 1,400 per cent between 2000 and 2015, is probably a result of these allegations (Fairtrade Foundation, 2012; Fairtrade International, 2017). Nevertheless, there is little empirical evidence of market power in consumer coffee markets for roasted and ground coffee. In fact, most studies find no or a tiny markup of price over marginal cost in Sweden and other developed countries (Bettendorf and Verboven, 2000; Feuerstein, 2002; Durevall, 2007a, 2007b; Gibbon, 2007; Gilbert, 2007, 2008). However, these studies treat coffee as a homogeneous product, due to paucity of data, and therefore fail to capture important market features (Sexton, 2013). A few recent studies analyse pass-through from marginal costs to retail prices using product-level data and find cost pass-through to be incomplete, i.e. less than 1 in the USA (Leibtag et al., 2007; Nakamura and Zerom, 2010), and far below 1 in Germany and France (Bonnet et al., 2013; Bonnet and Villas-Boas, 2016). The low pass-through indicates that there is imperfect competition and that prices are higher than marginal costs. This paper analyses how changes in marginal costs, measured by fluctuations in world market green coffee bean prices, are passed on to Swedish retail coffee prices and how pass-through rates are related to market structure. As in previous studies, I focus on roasted and ground coffee since green beans is the major input in the production process. Moreover, ground coffee has a market share of 80 per cent in the Swedish retail market. The Swedish coffee retail market is likely to be representative of many other markets in developed countries (Durevall, 2003; Sutton, 2007). In the ground coffee segment, the four largest roasters have a combined market share of about 85 per cent. There are a few retailer-owned brands (private labels), several small, mostly local, roasters and some foreign brands. Moreover, as in most developed countries, the retail sector is very concentrated (McCorriston, 2013). For example, the largest food chain has about 50 per cent of the market and the four largest account for well over 90 per cent of all sales (Swedish Competition Authority, 2011). The first part of the study uses unrestricted error correction models to estimate pass-through of industry-wide shocks to marginal costs for product/regional market combinations of each barcode-specific coffee product. The choice of model is based on the high persistence of the green bean price series, which can be characterised as unit root non-stationary. The unit root implies that bean prices should be cointegrated with the retail price for costs to have long-run impact, given other input costs. Since I also have data on quantities, in contrast to most other studies, the second part uses regression analysis to analyse the relationship between market structure and pass-through. Horizontal market structure is measured by coffee product, blend and brand market shares, and vertical market structure is measured by the presence of private labels. It is a common assumption that products with a large market share have a lower pass-through than products with small market shares, due to higher markup and more market power. This is what the Dornbusch (1987) model of monopolistic competition predicts, a model widely used in international economics (Burstein and Gopinath, 2014; Feenstra, 2016). Vertical integration, in the form of private labels, sidesteps the market power of manufacturers of branded products and should prevent double marginalisation, i.e. when a downstream firm (retailer) applies a markup on a price set by an upstream firm (roaster) that has market power and, therefore, also applies a markup. Avoidance of double marginalisation should lead to higher pass-through, given market share (Hong and Li, 2017). Therefore, by controlling for market shares, I can indirectly evaluate whether there is double marginalisation. Finally, as a robustness check, I use the cointegrated VAR model of Johansen (1991) to analyse cost pass-through and adjustment to cost shocks in one regional market, eastern Sweden.2 I focus on four dominant blends, which together have a market share of close to 50 per cent. The Johansen approach allows me to formally test whether there is perfect pass-through and evaluate the speed of adjustment to cost shocks. The paper is organised as follows: Section 2 gives a brief review of why pass-through rates might vary across firms and of the findings of earlier studies. Section 3 describes the data and Section 4 outlines the empirical approach. Section 5 estimates the pass-through rates and analyses the role of market structure, while Section 6 uses the cointegrated VAR model to analyse prices and costs of large blends in eastern Sweden. Section 7 concludes the paper. 2. Pass-through and market structure: theory and empirical evidence Pass-through of costs to prices has been a research topic for a long time, particularly in the fields of marketing, agricultural and public economics (Bulow and Pfleiderer, 1983; Tyagi, 1999; Weyl and Fabinger, 2013; Lloyd, 2017). Numerous studies on the topic have recently been published in the field of international economics as well, and have mainly focused on the effect of changes in exchange rates on consumer prices (Burstein and Gopinath, 2014). Several factors may affect pass-through, and no model captures all of them. Thus, for expository purposes, I use a Bertrand model from Anderson, De Palma and Kreider (2001) to illustrate the key ones when there are industry-wide changes in costs, conveniently measured by a unit tax.3 I then informally discuss other factors of relevance to the retail coffee market.4 The model has N firms that each produces one differentiated product, q, at constant marginal cost c. The profit function of firm i is πi=[pi−(ci+t)]q(pi,p−i), (1) where pi is the firm i’s price, p−i is the set of all other firms’ prices, q(pi,p−i) is the demand for the output of firm i and t is the unit tax. Profit maximisation gives the first-order condition [pi−(c'i+t)]δqi(pi,p−i)δpi+qi(pi,p−i)=0. (2) Pass-through is obtained by total differentiation with respect to t. By assuming symmetry between the firms, we can derive an expression for pass-through based on the diversion ratio, D, which measures the share of firm i’s sales that is captured by other firms when firm i raises its price, the elasticity of the slope of firm i’s demand curve with respect to the common price εm=[p/(∂qi/∂pi)][∂(∂qi/∂pi)/∂p], and firm i’s price elasticity with respect to a shift in its own price, given all other prices εdd=[p/q][∂qi/∂pi] (see RBB, 2014 for details), dpidt=12−D−εmεdd. (3) Equation (3) shows that in the case of complete product differentiation, that is, firm i is a monopolist, then D = 0 and the value of the ratio between the slope elasticity of the demand curve and the price elasticity determines pass-through. For instance, when demand is linear, εm=0, we have the textbook case of a pass-through of ½. It is reasonable to assume that D is generally higher the more firms there are in the market, since then there is more competition. Thus, as competition (the number of firms) increases, D approaches 1 and pass-through approaches 1. Consequently, even when demand is linear, pass-through can be 1. However, competition will also increase firm i’s price elasticity, and when there is perfect competition, the ratio εm/εdd will be zero irrespective of the slope elasticity of the demand curve. With monopolistic competition and a given D, the curvature of the demand function determines pass-through. When demand is concave, the slope of the demand curve becomes flatter as the price increases, i.e. the firm loses sales rapidly when raising prices. As a result, pass-through is low since εm increases with price. When demand is convex, εm declines when the price increases, and it is possible that pass-through is over 1. There are several multiproduct roasters in the Swedish coffee market, which is not captured by the model. A few studies analyse pass-through in theoretical models where firms produce more than one good, and the results depend on assumptions about functional form and the size of second-order effects (Besanko, Dubé and Gupta, 2005; Moorthy, 2005). However, by developing a structural model for the Swedish beer market, Rohman and Friberg (2016) provide evidence that multiproduct firms are likely to have lower pass-through than single-product firms, both when there are industry-wide and firm-specific cost shocks. They also show that the difference in pass-through between a single and a multiproduct firm may be small unless the multiproduct firm sells very many products, making it challenging to detect differences in an observational study. The shape of the cost function may also affect pass-through. An upward-sloping marginal cost function decreases pass-through both in perfectly competitive and oligopolistic markets. Nonetheless, the assumption of constant marginal costs is standard in the literature on coffee since processing of ground coffee is highly mechanised, but simple, and involves only a few workers (Bettendorf and Verboven, 2000; Sutton, 2007; Nakamura and Zerom, 2010). Moreover, as coffee consumption in Sweden is small relative to the world supply of beans, demand does not affect bean prices. Therefore, upward-sloping marginal costs are unlikely to have a major impact on pass-through. Some models allow for heterogeneous firms (Dornbusch, 1987; Feenstra, 2016, Ch. 6). In these models, there is usually a negative relationship between market share and pass-through: firms with large market shares have large markups and adjust the markup instead of the price after a change in marginal costs. Several studies also find a link between the degree of product pass-through and horizontal market structure (Atkeson and Burstein, 2008; Amiti, Itskhoki and Konings, 2014; Auer and Schoenle, 2016; Hong and Li, 2017).5 Most ground coffees are brand-name products processed and distributed to retailers by roasters. When upstream and downstream firms, i.e. roasters and retailers, have market power, there might be double marginalisation.6 As a result, there is markup on markup, and pass-through is lower than when a monopolist owns both the upstream and downstream firms. The number of private labels has been increasing for several years in many countries, including Sweden. This can be seen as vertical integration, since products without the manufacturers’ label are introduced. Private labels are expected to have a higher pass-through than brand-name products since they sidestep the roasters’ markup, though a large market share would have a counterbalancing effect. However, by controlling for market shares, Hong and Li (2017) show that private labels in the USA have substantially higher pass-through rates than branded products. In Sweden, large roasters and retail chains meet regularly to negotiate wholesale prices, so relative bargaining power and pricing contracts could influence pass-through.7 Thus, double marginalisation could be prevented by outright collusion, though this is illegal and unlikely given the number of actors. However, there are many types of vertical contracts. Based on simulation of structural models, Bonnet et al. (2013) find evidence of resale price maintenance in the German ground coffee market, which increases pass-through relative to double marginalisation. Finally, the model is static and pass-through is treated as the equilibrium price response to a change in costs, ignoring dynamics. Nakamura and Zerom (2010) examine short-run adjustments in the US coffee market using a structural model. They find that menu costs matter in the short run but not after a year or so.8 Thus, to conclude, studies of market power in coffee retail markets, which assume that coffee is a homogenous product and use average market prices, fail to find evidence of market power (Gibbon, 2007), while those that use product-level data find pass-through rates that are below one and conclude that there is imperfect competition. Assuming highly elastic supply, a pass-through rate below or above 1 is evidence against perfect competition. However, as evident from theory, the presence of a pass-through rate of 1 is neither a necessary nor a sufficient condition for perfect competition. Yet, by analysing how pass-through of individual products is related to horizontal and vertical integration, an informed judgement can be made. 3. The data The retail data are from the Nielsen company, which collects weekly scanner data the EAN (barcode) level of daily sales from 3,088 Swedish stores. Coffee is grouped into seven market segments, but I focus on the by far largest segment, roasted and ground coffee. Nielsen also collects information on a number of product characteristics, such as brand, manufacturer and whether the product is organic. I use this information to distinguish between pass-through rates for the major brands, three types of private labels (discount, premium and standard) and conventional and organic coffee; as all Fairtrade blends are organic,9 it is not modelled separately. Unfortunately, Nielsen does not make store-level information available in Sweden, so prices are averages for six regional markets. I use monthly averages to reduce the number of missing observations (all products are not sold every week) and the noise-to-information ratio; it clearly takes more than a week for a change in coffee bean prices to affect retail prices. Thus prices, obtained by dividing value by volume for a specific package size, are in Swedish kronor (SEK) per kilo at the barcode level by region and month. For simplicity, I refer to product/regional market combinations as products, and coffee products, such as Gevalia 0.500 kg Mellanrost Brygg, as blends. One feature of retail prices is that they sometimes decrease due to price promotions, but there is no information in the database on promotions. Some studies filter the price series to remove temporary decreases and analyse normal prices or reference prices (Eichenbaum, Jaimovich and Rebelo, 2011; Lloyd et al., 2014). However, my prices are averages by region, not individual store prices, so it is not clear what would be filtered out. Moreover, I analyse a relatively long period, almost 6 years, and sales are not synchronised across food chains, so occasional sales probably only have a minor impact on pass-through rates. I, therefore, use actual average prices in the analysis. Since I analyse individual time series, products with missing observations have been dropped. They make up 12 per cent of the sample in volume terms. Most of the products belong to blends that were introduced in, or removed from, the market during the study period. As Table 1 shows, the sample thus consists of 378 products and 68 blends, belonging to 20 brands (plus unspecified Other Brands), and runs from March 2009 to November 2014. Almost all roasters produce one brand. The noteworthy exception to this is Gevalia, which produces three brands: Gevalia, Blå Mocca and Maxwell House. Table 1. Brands, coffees, products, market shares and average prices Brand Blends Products Market share (%) Mean real price Gevalia 5 30 39.37 59.04 Zoega 10 60 26.80 70.63 Classic 9 54 14.02 65.63 Löfbergs Lila 7 38 6.41 57.67 ICA 6 36 5.05 48.43 Blå Mocca 3 18 3.18 61.75 Coop 6 36 1.61 49.19 Änglamark 1 6 1.10 62.74 Maxwell House 1 6 0.59 68.56 Coop X-tra 1 6 0.48 36.44 ICA I Love Eco 2 12 0.36 59.72 Lindvall 4 23 0.34 49.72 Euroshopper 1 6 0.28 33.91 Sackeus 2 7 0.11 79.74 Ideekaffe 1 6 0.09 85.13 Folke Bergman 1 1 0.08 47.52 Servtrade 3 16 0.05 39.22 Najjar 2 9 0.04 109.40 Other Brands 2 7 0.03 66.60 Lyxkaffe 1 1 0.00 38.68 Sum 68 378 100 Brand Blends Products Market share (%) Mean real price Gevalia 5 30 39.37 59.04 Zoega 10 60 26.80 70.63 Classic 9 54 14.02 65.63 Löfbergs Lila 7 38 6.41 57.67 ICA 6 36 5.05 48.43 Blå Mocca 3 18 3.18 61.75 Coop 6 36 1.61 49.19 Änglamark 1 6 1.10 62.74 Maxwell House 1 6 0.59 68.56 Coop X-tra 1 6 0.48 36.44 ICA I Love Eco 2 12 0.36 59.72 Lindvall 4 23 0.34 49.72 Euroshopper 1 6 0.28 33.91 Sackeus 2 7 0.11 79.74 Ideekaffe 1 6 0.09 85.13 Folke Bergman 1 1 0.08 47.52 Servtrade 3 16 0.05 39.22 Najjar 2 9 0.04 109.40 Other Brands 2 7 0.03 66.60 Lyxkaffe 1 1 0.00 38.68 Sum 68 378 100 Note: Based on the period March 2009–November 2014. The average retail price is in constant 2010 prices net of 12 per cent value added tax. Products are geographic region/coffee combinations. The market shares are measured as volume of sales. View Large Table 1. Brands, coffees, products, market shares and average prices Brand Blends Products Market share (%) Mean real price Gevalia 5 30 39.37 59.04 Zoega 10 60 26.80 70.63 Classic 9 54 14.02 65.63 Löfbergs Lila 7 38 6.41 57.67 ICA 6 36 5.05 48.43 Blå Mocca 3 18 3.18 61.75 Coop 6 36 1.61 49.19 Änglamark 1 6 1.10 62.74 Maxwell House 1 6 0.59 68.56 Coop X-tra 1 6 0.48 36.44 ICA I Love Eco 2 12 0.36 59.72 Lindvall 4 23 0.34 49.72 Euroshopper 1 6 0.28 33.91 Sackeus 2 7 0.11 79.74 Ideekaffe 1 6 0.09 85.13 Folke Bergman 1 1 0.08 47.52 Servtrade 3 16 0.05 39.22 Najjar 2 9 0.04 109.40 Other Brands 2 7 0.03 66.60 Lyxkaffe 1 1 0.00 38.68 Sum 68 378 100 Brand Blends Products Market share (%) Mean real price Gevalia 5 30 39.37 59.04 Zoega 10 60 26.80 70.63 Classic 9 54 14.02 65.63 Löfbergs Lila 7 38 6.41 57.67 ICA 6 36 5.05 48.43 Blå Mocca 3 18 3.18 61.75 Coop 6 36 1.61 49.19 Änglamark 1 6 1.10 62.74 Maxwell House 1 6 0.59 68.56 Coop X-tra 1 6 0.48 36.44 ICA I Love Eco 2 12 0.36 59.72 Lindvall 4 23 0.34 49.72 Euroshopper 1 6 0.28 33.91 Sackeus 2 7 0.11 79.74 Ideekaffe 1 6 0.09 85.13 Folke Bergman 1 1 0.08 47.52 Servtrade 3 16 0.05 39.22 Najjar 2 9 0.04 109.40 Other Brands 2 7 0.03 66.60 Lyxkaffe 1 1 0.00 38.68 Sum 68 378 100 Note: Based on the period March 2009–November 2014. The average retail price is in constant 2010 prices net of 12 per cent value added tax. Products are geographic region/coffee combinations. The market shares are measured as volume of sales. View Large The market is highly concentrated; the two largest brands, Gevalia and Zoegas have 39 per cent and 27 per cent of the ground coffee market (in volume terms) and are owned by Mondelez and Nestlé, respectively. The market shares of the other two large brands, Classic and Löfbergs Lila, are 14 per cent and 6 per cent, respectively. They are produced by family-owned roasters. The brands ICA, Coop, Änglamark, ICA I Love Eco and Euroshopper are private labels, the other brands are produced by small Swedish roasters or imported. There is no information on commodity costs for manufacturers or brands, so green coffee bean prices in SEK are used to measure cost shocks. There are two sources for bean prices: the International Coffee Organization (ICO) collects price data from terminal markets in Europe and the USA, and Statistics Sweden publishes value and volume of monthly imports of green coffee beans. I use prices based on import data from Statistics Sweden, but the choice does not matter much for the results, except that ICO prices take longer to affect retail prices. The reason for the similarity between of import prices and prices set at the commodity exchanges is that the latter function as price guides for physical coffee trade (ITA, 2011). The use of the same cost, that is, the weighted average cost of beans, for all blends introduces a bias in the estimation of pass-through rates. The estimates will be too low for blends with lower than average costs and too high for blends with higher than average costs. However, the effect of biases on the relative sizes of the pass-through rates can be checked by taking the log of the variables, estimating pass-through elasticities. This works because only Arabica beans are used and their prices have the same stochastic trends since they are close substitutes, i.e. they are cointegrated (see Fousekis and Grigoriadis, 2017).10 Table 1 also shows that average real retail prices (net of VAT) vary from 34 SEK to 109 SEK per kg, which can be compared with the average cost of green coffee beans of 32.60 SEK. To highlight the evolution of prices over time, Figure 1 shows the average price of unweighted ground coffee and cost of green coffee beans, measured by the bean price times 1.19 to account for weight lost from roasting.11 Both series are measured in constant 2010 prices using the consumer price index. There were large changes in prices and costs during the study period. The average per kg price of ground coffee fluctuated between 60 SEK and 80 SEK, and the cost of green coffee beans fluctuated between 20 SEK and 45 SEK per kg. Figure 1 also shows that the series are highly persistent; when the augmented Dickey–Fuller (ADF) test is applied on the individual retail prices, only about 10–15 tests reject the null hypothesis of a unit root (results available on request). The bean price series also appears to have a unit root: the ADF test with one lag (longer lags are insignificant), a constant, and a time trend gives a t-value of −1.4 (5 per cent significance value is −3.47) and an estimated root of 0.97. Fig. 1. View largeDownload slide Average price per kg of ground coffee and cost of imported green coffee beans in net of VAT 2010 SEK, 2009:3–2014:11. Fig. 1. View largeDownload slide Average price per kg of ground coffee and cost of imported green coffee beans in net of VAT 2010 SEK, 2009:3–2014:11. 4. Empirical approach The most commonly used model to estimate pass-through is the distributed lags model in rates of change, i.e. the first difference of the log-level of prices and costs (Nakamura and Zerom, 2010; Aron, Macdonald and Muellbauer, 2014),12 though Bonnet and Villas-Boas (2016) and Bonnet et al. (2013) estimate static fixed effects models in log levels. One reason for using rates of change is probably that often only price indices are analysed in international economics and the focus is on changes in the exchange rate. Another reason is the risk of spurious correlation, since some series might be unit root non-stationary. A drawback of using first differences is that important information about the long-run pass-through is ignored if the series have unit roots and are cointegrated, which, as evident from Figure 1, seems to be the case for ground coffee. Thus, I estimate models in levels. Spurious regression is unlikely to be a major issue for ground coffee, since the cost of green coffee beans is by far the largest component of marginal cost, estimated to equal 50–90 per cent (Bettendorf and Verboven, 2000; Nakamura and Zerom, 2010; Bonnet et al., 2013). Moreover, bean costs usually fluctuate much more than other production-related costs, and this was true for Sweden during the study period, when compared with wages and the consumer price index.13 Bean costs and retail coffee prices are, therefore, expected to have the same stochastic trends. However, some retail prices are stationary, as mentioned, and some could be non-stationary but unrelated to costs, implying that long-run pass-through is zero. This could occur because real retail prices are kept more or less fixed over the study period, or because of large abrupt price changes due to revised price policies. In these cases, the pass-through estimates are small or in some cases even negative. Static models in levels are easy to estimate and might provide adequate long-run pass-through rates when the variables are integrated of order one and cointegrated. However, a static model with monthly data is not likely to capture the dynamics adequately; i.e. the common factor restriction (implicitly) imposed is often invalid (Ericsson and MacKinnon, 2002). It might also fail to provide consistent estimates of pass-through when the dependent variable is stationary and lags are needed to specify the model correctly. I therefore estimate pass-through coefficients for each product using an unrestricted error correction model specified as Δpt=α+α1Δpt−1+α2Δcostt−1+β1pt−1+β2costt−1+trend+seas+εt, (4) where pt is the net-of-VAT price level of a product at month t, cost is the price of one kg of green coffee beans adjusted for weight lost from roasting, Δ is the difference operator, β1 and β2 are the parameters of interest, β2/−β1 is the estimate of long-run pass-through and εt is the error term. The lags in first differences, Δpt−1andΔcostt−1, potentially improve the estimates of β1 and β2 by capturing short-run dynamics. Adding more lags only affects the estimates of β2/−β1 marginally and does not change any of the results. Trend is a deterministic trend that potentially captures other marginal costs and slow-moving changes in technologies, and seas are seasonal dummies. I do not base the subsequent analysis on formal tests for cointegration of each of the 378 equations, but very low or negative pass-through rates indicate a lack of a long-run relationship. However, I also estimate distributed lag models in first differences and static models in levels, as well as the error correction model with variables in log-levels. Moreover, as robustness check, in Section 6, I formally test for cointegration in a multivariate framework in one of the six regional markets using the battery developed by Johansen (1991). 5. Pass-through rates and the role of market structure Figure 2 reports the estimates of the pass-through rates.14 The vast majority are larger than zero, as expected. Those with negative pass-through are mostly odd products with almost constant prices and very small market shares. One product with a pass-through rate of 2.1 is excluded from the figure to improve readability; a detailed analysis revealed that the high value is due to a large structural break at the end of the sample. Fig. 2. View largeDownload slide Estimates of pass-through rates for 378 products. Note: Estimates are based on unrestricted error correction models with one lag. One observation with a pass-through of 2.1 has been dropped. Fig. 2. View largeDownload slide Estimates of pass-through rates for 378 products. Note: Estimates are based on unrestricted error correction models with one lag. One observation with a pass-through of 2.1 has been dropped. Most pass-through rates fall in the 0.5–1 range. This is also the case when long-run pass-through is estimated with a static model in levels, though the pass-through rates are somewhat lower on average (see Figure A1 in the Appendix). The correlation between the two measures is 0.92. When pass-through rates are estimated with distributed lags models in first differences, the pass-through rates are clearly lower, as shown in Figure A2 in the Appendix. The models have 24 lags, but adding more lags does not increase pass-through rates. Thus, distributed lags models seem to underestimate the size of the pass-through rates since they do not use information contained in the levels of the variables. Lastly, Figure A3 in the Appendix shows that there is a strong correlation between pass-through rates and pass-through elasticities, 0.93, implying that the relative ranking of the pass-through rates is only marginally affected by the use of weighted average costs. Some estimates are larger than 1, but this is probably due to estimation uncertainty; almost no 95 per cent confidence interval excludes the value of 1. Thus, there is no evidence of highly convex demand functions, i.e. functional forms that generate pass-through larger than 1. Figure 3 depicts the estimated pass-through coefficients and market shares, in logs for visibility, for the 68 blends. It shows that the larger the market share, the higher the pass-through. This relationship is contrary to the prediction of models in which the size of the market share is related to market power, indicating that blends with large market shares compete for customers. The relationship is also non-linear, as a pass-through rate of 1 seems to constitute a (fuzzy) upper limit. Fig. 3. View largeDownload slide Pass-through rates for products and log of blend market shares. Fig. 3. View largeDownload slide Pass-through rates for products and log of blend market shares. Table 2 reports regressions that test for the association between pass-through rates and market shares. Since it is not obvious how to measure market shares, three measures are used: product, blend and brand market shares. To capture the non-linear relationship, both level and squared market share terms are included. Table 2. Pass-through and market shares, ground coffee (1) (2) (3) (4) (5) (6) Product share 1.039 3.268 (0.333)** (1.171)** Product share sq −5.761 (2.255)* Blend share 1.728 7.596 (0.943) (2.095)** Blend share sq −24.372 (7.336)** Brand share 0.468 2.253 (0.245) (1.260) Brand share sq −3.641 (2.241) Najjar/Bosnia −0.911 −0.889 −0.900 −0.844 −0.876 −0.791 (0.083)** (0.085)** (0.084)** (0.087)** (0.110)** (0.116)** Constant 0.711 0.689 0.700 0.642 0.674 0.579 (0.039)** (0.042)** (0.041)** (0.046)** (0.074)** (0.093)** R2 0.30 0.32 0.31 0.37 0.31 0.35 N 378 378 378 378 378 378 (1) (2) (3) (4) (5) (6) Product share 1.039 3.268 (0.333)** (1.171)** Product share sq −5.761 (2.255)* Blend share 1.728 7.596 (0.943) (2.095)** Blend share sq −24.372 (7.336)** Brand share 0.468 2.253 (0.245) (1.260) Brand share sq −3.641 (2.241) Najjar/Bosnia −0.911 −0.889 −0.900 −0.844 −0.876 −0.791 (0.083)** (0.085)** (0.084)** (0.087)** (0.110)** (0.116)** Constant 0.711 0.689 0.700 0.642 0.674 0.579 (0.039)** (0.042)** (0.041)** (0.046)** (0.074)** (0.093)** R2 0.30 0.32 0.31 0.37 0.31 0.35 N 378 378 378 378 378 378 Note: Najjar/Bosnia is a dummy for Najjar and Bosnia (Zlatna Dzezva) products, out of which all but one have negative pass-through rates. Standard errors are clustered at the product level except when brand shares are included; in which case clustering is at the brand level. *p < 0.05; **p < 0.01. View Large Table 2. Pass-through and market shares, ground coffee (1) (2) (3) (4) (5) (6) Product share 1.039 3.268 (0.333)** (1.171)** Product share sq −5.761 (2.255)* Blend share 1.728 7.596 (0.943) (2.095)** Blend share sq −24.372 (7.336)** Brand share 0.468 2.253 (0.245) (1.260) Brand share sq −3.641 (2.241) Najjar/Bosnia −0.911 −0.889 −0.900 −0.844 −0.876 −0.791 (0.083)** (0.085)** (0.084)** (0.087)** (0.110)** (0.116)** Constant 0.711 0.689 0.700 0.642 0.674 0.579 (0.039)** (0.042)** (0.041)** (0.046)** (0.074)** (0.093)** R2 0.30 0.32 0.31 0.37 0.31 0.35 N 378 378 378 378 378 378 (1) (2) (3) (4) (5) (6) Product share 1.039 3.268 (0.333)** (1.171)** Product share sq −5.761 (2.255)* Blend share 1.728 7.596 (0.943) (2.095)** Blend share sq −24.372 (7.336)** Brand share 0.468 2.253 (0.245) (1.260) Brand share sq −3.641 (2.241) Najjar/Bosnia −0.911 −0.889 −0.900 −0.844 −0.876 −0.791 (0.083)** (0.085)** (0.084)** (0.087)** (0.110)** (0.116)** Constant 0.711 0.689 0.700 0.642 0.674 0.579 (0.039)** (0.042)** (0.041)** (0.046)** (0.074)** (0.093)** R2 0.30 0.32 0.31 0.37 0.31 0.35 N 378 378 378 378 378 378 Note: Najjar/Bosnia is a dummy for Najjar and Bosnia (Zlatna Dzezva) products, out of which all but one have negative pass-through rates. Standard errors are clustered at the product level except when brand shares are included; in which case clustering is at the brand level. *p < 0.05; **p < 0.01. View Large Nor is it obvious how to deal with very low and negative pass-through rates, which presumably are zero. Hong and Li (2017) simply exclude products with negative pass-through in their analysis. However, two types of products have negative pass-through rates, private labels and imported blends from brands primarily sold to immigrant groups. Thus, instead of removing all of them, I include a dummy for Najjar and Bosnia (Zlatna Dzezva) products, which make up about half of those with negative pass-through rates. The others are primarily private labels, which are of interest for the study. I also estimated models with all pass-through rates below 0.1 set to zero, but the results only changed marginally (the results are available on request). Table 2 shows that there is a strong positive, but non-linear, association between pass-through and market share. For example, at the median blend market share, 0.005, the pass-through is 0.70, and at a blend market share of 0.05, it is 0.84. Market shares measured at the blend level are somewhat more correlated with pass-through than the others, while the lack of significance for brand shares might be because of the relatively small number of brands, but most likely it is because the product and blend levels matter more for pricing decisions than the brand level (as shown below). Table A1 in the appendix reports the same specifications but with the pass-through rate replaced by the pass-through elasticity. The results are practically the same, showing that the measurement errors are of minor importance. Table 3 reports specifications that aim to further describe differences in pass-through. Column (1) adds three types of private labels (discount premium, and standard), type of roast (dark and unspecified; the base is medium roast) and organic coffee to Table 2’s column (2). Private labels should have a higher pass-through in monopolistic markets, given market share, since they avoid potential double marginalisation. However, both discount and premium private labels have a lower pass-through rate than other blends, while the standard one has the same rate, given market shares. The dummies for dark roast and organic coffee are insignificant, while unknown roast is significant at the 5 per cent level. The coefficients of product market share and squared product market share are unaffected by the inclusion of the other variables. Table 3. Pass-through, private labels and large brands, ground coffee (1) (2) (3) (4) (5) (6) Private label Brand dummies Brand dummies Four large brands Gevalia Classic Blend share 7.147 1.631 2.369 19.069 (2.128)** (1.834) (0.442)** (5.393)** Blend share sq −23.840 −5.789 −7.085 −252.179 (7.480)** (6.548) (1.605)** (72.375)** Classic 0.354 0.320 (0.086)** (0.061)** Löfbergs lila 0.413 0.378 (0.087)** (0.063)** Gevalia 0.311 0.267 (0.086)** (0.057)** Zoega 0.334 0.325 (0.122)** (0.104)** Private Disc −0.752 −0.602 (0.071)** (0.091)** Private Prem −0.283 −0.013 (0.097)** (0.111) Private Stand 0.003 0.191 (0.058) (0.085)* Dark roast 0.093 0.053 (0.060) (0.054) Unknown roast 0.253 0.248 (0.105)* (0.081)** Organic 0.102 −0.016 (0.081) (0.054) Odd size −0.996 −0.855 −0.765 (0.066)** (0.078)** (0.090)** Constant 0.632 0.517 0.565 0.863 0.750 0.731 (0.072)** (0.087)** (0.052)** (0.045)** (0.012)** (0.057)** R2 0.57 0.65 0.49 0.01 0.64 0.49 N 378 378 378 182 30 54 (1) (2) (3) (4) (5) (6) Private label Brand dummies Brand dummies Four large brands Gevalia Classic Blend share 7.147 1.631 2.369 19.069 (2.128)** (1.834) (0.442)** (5.393)** Blend share sq −23.840 −5.789 −7.085 −252.179 (7.480)** (6.548) (1.605)** (72.375)** Classic 0.354 0.320 (0.086)** (0.061)** Löfbergs lila 0.413 0.378 (0.087)** (0.063)** Gevalia 0.311 0.267 (0.086)** (0.057)** Zoega 0.334 0.325 (0.122)** (0.104)** Private Disc −0.752 −0.602 (0.071)** (0.091)** Private Prem −0.283 −0.013 (0.097)** (0.111) Private Stand 0.003 0.191 (0.058) (0.085)* Dark roast 0.093 0.053 (0.060) (0.054) Unknown roast 0.253 0.248 (0.105)* (0.081)** Organic 0.102 −0.016 (0.081) (0.054) Odd size −0.996 −0.855 −0.765 (0.066)** (0.078)** (0.090)** Constant 0.632 0.517 0.565 0.863 0.750 0.731 (0.072)** (0.087)** (0.052)** (0.045)** (0.012)** (0.057)** R2 0.57 0.65 0.49 0.01 0.64 0.49 N 378 378 378 182 30 54 Note: Najjar/Bosnia is a dummy for Najjar and Bosnia (Zlatna Dzezva) products, out of which all but one have negative pass-through rates. Standard errors are clustered at the product level except when brand shares are included; in which case clustering is at the brand level. *p < 0.05; **p < 0.01. View Large Table 3. Pass-through, private labels and large brands, ground coffee (1) (2) (3) (4) (5) (6) Private label Brand dummies Brand dummies Four large brands Gevalia Classic Blend share 7.147 1.631 2.369 19.069 (2.128)** (1.834) (0.442)** (5.393)** Blend share sq −23.840 −5.789 −7.085 −252.179 (7.480)** (6.548) (1.605)** (72.375)** Classic 0.354 0.320 (0.086)** (0.061)** Löfbergs lila 0.413 0.378 (0.087)** (0.063)** Gevalia 0.311 0.267 (0.086)** (0.057)** Zoega 0.334 0.325 (0.122)** (0.104)** Private Disc −0.752 −0.602 (0.071)** (0.091)** Private Prem −0.283 −0.013 (0.097)** (0.111) Private Stand 0.003 0.191 (0.058) (0.085)* Dark roast 0.093 0.053 (0.060) (0.054) Unknown roast 0.253 0.248 (0.105)* (0.081)** Organic 0.102 −0.016 (0.081) (0.054) Odd size −0.996 −0.855 −0.765 (0.066)** (0.078)** (0.090)** Constant 0.632 0.517 0.565 0.863 0.750 0.731 (0.072)** (0.087)** (0.052)** (0.045)** (0.012)** (0.057)** R2 0.57 0.65 0.49 0.01 0.64 0.49 N 378 378 378 182 30 54 (1) (2) (3) (4) (5) (6) Private label Brand dummies Brand dummies Four large brands Gevalia Classic Blend share 7.147 1.631 2.369 19.069 (2.128)** (1.834) (0.442)** (5.393)** Blend share sq −23.840 −5.789 −7.085 −252.179 (7.480)** (6.548) (1.605)** (72.375)** Classic 0.354 0.320 (0.086)** (0.061)** Löfbergs lila 0.413 0.378 (0.087)** (0.063)** Gevalia 0.311 0.267 (0.086)** (0.057)** Zoega 0.334 0.325 (0.122)** (0.104)** Private Disc −0.752 −0.602 (0.071)** (0.091)** Private Prem −0.283 −0.013 (0.097)** (0.111) Private Stand 0.003 0.191 (0.058) (0.085)* Dark roast 0.093 0.053 (0.060) (0.054) Unknown roast 0.253 0.248 (0.105)* (0.081)** Organic 0.102 −0.016 (0.081) (0.054) Odd size −0.996 −0.855 −0.765 (0.066)** (0.078)** (0.090)** Constant 0.632 0.517 0.565 0.863 0.750 0.731 (0.072)** (0.087)** (0.052)** (0.045)** (0.012)** (0.057)** R2 0.57 0.65 0.49 0.01 0.64 0.49 N 378 378 378 182 30 54 Note: Najjar/Bosnia is a dummy for Najjar and Bosnia (Zlatna Dzezva) products, out of which all but one have negative pass-through rates. Standard errors are clustered at the product level except when brand shares are included; in which case clustering is at the brand level. *p < 0.05; **p < 0.01. View Large Column (2) replaces market shares with dummies for the four large brands. They are all significant at the 1 per cent level and show that the large brands have pass-through rates that on average are about 0.3–0.4 SEK higher than the other brands. Column (3) shows that these results are not due to the other variables included in the model. Column (4) restricts the sample to the four large brands to test whether pass-through rates differ across their products. Although the market share coefficients have the same signs as in the other regressions, they are much smaller and insignificant. However, two of the large brands, Gevalia and Classic, have pass-through rates that vary with market shares, as reported in columns (5) and (6). This is not the case for the other two, Löfbergs Lila and Zoega (not reported). Figure 4 sheds further light on the relationship between pass-through and the 20 brand market shares. Average pass-through rates are similar for the four largest brands, but there is a fairly large variance among the products of each brand. Moreover, it is clear that only very small brands have pass-through rates close to, or below, zero. Fig. 4. View largeDownload slide Pass-through rates for products and brand market shares. Note: Estimates are based on unrestricted error correction models with one lag. One observation with a pass-through of 2.1 has been dropped. Fig. 4. View largeDownload slide Pass-through rates for products and brand market shares. Note: Estimates are based on unrestricted error correction models with one lag. One observation with a pass-through of 2.1 has been dropped. 6. Analysis of large coffees in the Eastern Swedish Market To provide additional support to the findings that a substantial part of the Swedish coffee market is competitive, I use the cointegrated VAR model and the Johansen approach to analyse pass-through and dynamic price interaction among large blends. I focus on one market, eastern Sweden, which covers the most densely populated areas of the country, including the Stockholm region, Uppsala, Nyköping and Norrtälje. The advantage of the Johansen approach is that it allows me to formally test hypotheses about long- and short-run parameters in a multivariate setup, i.e. the size of pass-through rates and the direction of adjustment to changes in costs (Juselius, 2006; Hoover, Johansen and Juselius, 2008). Figures 5 and 6 depict real retail prices net of VAT for the four largest blends in terms of volume market shares nationally: Gevalia Mellanrost Brygg, Gevalia Mellanrost E-brygg, Classic Mellan Brygg and Zoegas Skånerost. By far the largest blend is Gevalia Mellanrost Brygg, which has a market share of about 20 per cent in eastern Sweden. Classic Mellan Brygg has a market share of 14 per cent, while Gevalia Mellanrost E-brygg and Zoegas Skånerost have about 5 per cent each. Fig. 5. View largeDownload slide Real prices of Gevalia Mellanrost Brygg, Gevalia Mellanrost E-brygg and the cost of beans. Fig. 5. View largeDownload slide Real prices of Gevalia Mellanrost Brygg, Gevalia Mellanrost E-brygg and the cost of beans. Fig. 6. View largeDownload slide Real prices of Classic Mellan Brygg and Zoegas Skånerost and the cost of beans. Fig. 6. View largeDownload slide Real prices of Classic Mellan Brygg and Zoegas Skånerost and the cost of beans. As evident from the figures, all prices seem to be non-stationary and follow a pattern similar to the cost of coffee beans. Moreover, the prices of Gevalia Mellanrost Brygg and Gevalia Mellanrost E-brygg are practically identical, while Classic Mellan Brygg is a bit more expensive than the Gevalia coffees and Zoegas Skånerost is clearly more expensive. There is a notable increase in the difference between Zoegas Skånerost and the others in 2012, as the price of Zoegas declines more slowly than the other prices, but in 2014, the difference has returned to the pre-2012 level. The data are not informative of whether this deviation was a one-time event. Since the two Gevalia prices provide the same information, I drop Gevalia Mellanrost E-brygg from the analysis. The Johansen approach is based on the VAR model specified in error correction form: Δxt=αβ'xt+Γ1Δxt−1+ΦDt+εt (5) where xt is a vector of potentially endogenous variables, Δ is the first difference operator, D is a vector of deterministic terms, such as constant and seasonal dummies and εt is white noise (see Juselius, 2006). The lag length of the corresponding VAR in levels is set to two for illustrative purposes. The coefficients and hypotheses tested are easiest to understand if the model is (partly) written in matrix form. I assume (and later show) that there are three long-run relations (cointegrating vectors). We then have (pGevpClapZoecbea)=(α11α12α13α21α22α23α31α32α33α41α42α43)(β1'xt−1β2'xt−1β3'xt−1)t+Γ1Δxt−1+ΦDt+(ε1tε2tε3tε4t), (6) where pGev,pCla,pZoe and cbea are product retail prices and bean cost, αii is an adjustment coefficient, βi'xt is an error correction term for a long-run relationship and Γ1 is a matrix of short-run coefficients. With four variables (each of which has a unit root) and three long-run relationships, there is one common trend, presumably resulting from the stochastic trend of the cost of coffee beans. This implies that there is a long-run relationship, βi'xt, between each retail price and the cost of beans, though other combinations are possible, in principle. When there is complete pass-though, the coefficients of the three vectors β1',β2' and β3' are (1, 0, 0, −1), (0, 1, 0, −1) and (0, 0, 1, −1), which is a testable hypothesis. I can also test whether cbea is weakly exogenous, i.e. whether α41=α42=α43=0, which would indicate that the cost of beans does not adjust to maintain any long-run relationship, a reasonable assumption given the small size of the Swedish coffee market. I can further test whether all adjustments are due to coffee bean costs or there is interaction between retail prices such that for example the price of Gevalia Mellanrost Brygg influences the adjustment of the other two prices. In this case, α21≠0,α31≠0. Finally, the model provides estimates of the speed of the adjustment to the long-run relationship through the values of α11,α22andα33, and of whether there is short-run adjustment, Γ1, due to lagged changes in Δxt. A key step in the analysis is to determine the cointegration rank (assumed to be three above), i.e. the number of long-run relationships. The Johansen approach uses the trace test, which is based on the maximum likelihood procedure. Given that I only have 66 observations, additional information should also be used to determine the rank (Juselius, 2001). I use the point estimates of the eigenvalues, which should be clearly larger than zero, the size of the adjustment coefficients, and economic reasoning. Table 4 reports ADF unit root tests, trace tests, estimated coefficients and likelihood ratio test of restrictions on the system. The ADF tests (Panel 1) indicate that all four variables have a unit root; none of the tests statistics are significant and all estimated roots are above 0.9. Panel 2 reports the trace test statistics for the VAR model with two and three lags; the Hannan-Quinn and Schwartz information criteria clearly favoured a model with two lags, while the Akaike criterion selected a model with three lags. Adding up to five lags to the model does not change any findings. The trace test shows that there are three cointegrating vectors in the two-lag model, i.e. the rank = 2 is rejected, while there are four cointegrating vectors in the three-lag model, as the rank = 3 is rejected. If there were four cointegrating vectors, all variables would be stationary, contradicting the ADF unit root tests. Since the eigenvalues for rank = 4 are close to zero (0.04 and 0.08) and the adjustment coefficients for the product retail prices α11,α22andα33 are large in absolute terms and the one for the cost of beans, α44, is close to zero, −0.04 (Panel 3), I proceed under the assumptions that the rank is 3 and there are three cointegrating vectors.15 Table 4. Cointegration tests and hypotheses tests on the three main coffees in eastern Sweden Panel ADF unit root tests: two lags and a trend. 5% = −3.48, 1% = −4.10 pGev pCla pZoe cbea 1 t-adf −1.38 −1.65 −1.24 −1.83 Root 0.92 0.91 0.93 0.95 Johansen approach 2 Trace test of number of cointegrating vectors: H0: rank≤ Two lags Three lags Eigenvalue p-value Eigenvalue p-value Rank 0 – [0.000]** – [0.000]** Rank 1 0.54 [0.000]** 0.41 [0.005]** Rank 2 0.46 [0.027]* 0.22 [0.008]** Rank 3 0.18 [0.092] 0.19 [0.018]* Rank 4 0.04 – 0.08 – 3 Adjustment coefficients α11,α22,α33,α44, unrestricted 3-lag model pGev pZoe pCla cbea −0.48 −0.16 −0.36 −0.04 4 Long-run normalised coefficients, βii, for three vectors pGev 1.00 0.00 0.00 pZoe 0.00 1.00 0.00 pCla 0.00 0.00 1.00 cbea −0.90 −0.98 −0.98 (0.05) (0.19) (0.06) Adjustment coefficients and standard errors pGev −0.79 0.07 −0.11 (0.17) (0.07) (0.18) pZoe −0.08 −0.23 −0.23 (0.21) (0.08) (0.22) pCla −0.06 0.01 −0.63 (0.16) (0.06) (0.16) cbea −0.03 −0.06 −0.05 (0.11) (0.04) (0.11) 5 LR test of complete pass-through: χ2(3)=3.43[0.33] β21=β31=β12=β32=β13=β23=0;β11=β22=β33=1;β41=β42=β43=−1; 6 LR test χ2(8)=8.13[0.42] β21=β31=β12=β32=β13=β23=0;β11=β22=β33=1;β41=β42=β43=−1; α12=α13=α21=α23=α31=α32=α41=α42=α43=α41=0 Panel ADF unit root tests: two lags and a trend. 5% = −3.48, 1% = −4.10 pGev pCla pZoe cbea 1 t-adf −1.38 −1.65 −1.24 −1.83 Root 0.92 0.91 0.93 0.95 Johansen approach 2 Trace test of number of cointegrating vectors: H0: rank≤ Two lags Three lags Eigenvalue p-value Eigenvalue p-value Rank 0 – [0.000]** – [0.000]** Rank 1 0.54 [0.000]** 0.41 [0.005]** Rank 2 0.46 [0.027]* 0.22 [0.008]** Rank 3 0.18 [0.092] 0.19 [0.018]* Rank 4 0.04 – 0.08 – 3 Adjustment coefficients α11,α22,α33,α44, unrestricted 3-lag model pGev pZoe pCla cbea −0.48 −0.16 −0.36 −0.04 4 Long-run normalised coefficients, βii, for three vectors pGev 1.00 0.00 0.00 pZoe 0.00 1.00 0.00 pCla 0.00 0.00 1.00 cbea −0.90 −0.98 −0.98 (0.05) (0.19) (0.06) Adjustment coefficients and standard errors pGev −0.79 0.07 −0.11 (0.17) (0.07) (0.18) pZoe −0.08 −0.23 −0.23 (0.21) (0.08) (0.22) pCla −0.06 0.01 −0.63 (0.16) (0.06) (0.16) cbea −0.03 −0.06 −0.05 (0.11) (0.04) (0.11) 5 LR test of complete pass-through: χ2(3)=3.43[0.33] β21=β31=β12=β32=β13=β23=0;β11=β22=β33=1;β41=β42=β43=−1; 6 LR test χ2(8)=8.13[0.42] β21=β31=β12=β32=β13=β23=0;β11=β22=β33=1;β41=β42=β43=−1; α12=α13=α21=α23=α31=α32=α41=α42=α43=α41=0 Note: Panel 1 reports Augmented Dickey–Fuller unit root tests, t-values and the estimated roots. Panel 2 reports the trace tests and eigenvalues. The null is that the rank is equal to or less than a certain number when the eigenvalue is zero. Models with two and three lags are tested. Panel 3 reports the adjustment coefficients for each variable when the model is estimated with three lags. Panel 4 reports the long-run and the adjustment coefficients when there are assumed to be three cointegrating vectors. Panel 5 reports the likelihood ratio test that pass-through is 1, and Panel 6 reports the test that pass-through is I and all the adjustment is due to costs. View Large Table 4. Cointegration tests and hypotheses tests on the three main coffees in eastern Sweden Panel ADF unit root tests: two lags and a trend. 5% = −3.48, 1% = −4.10 pGev pCla pZoe cbea 1 t-adf −1.38 −1.65 −1.24 −1.83 Root 0.92 0.91 0.93 0.95 Johansen approach 2 Trace test of number of cointegrating vectors: H0: rank≤ Two lags Three lags Eigenvalue p-value Eigenvalue p-value Rank 0 – [0.000]** – [0.000]** Rank 1 0.54 [0.000]** 0.41 [0.005]** Rank 2 0.46 [0.027]* 0.22 [0.008]** Rank 3 0.18 [0.092] 0.19 [0.018]* Rank 4 0.04 – 0.08 – 3 Adjustment coefficients α11,α22,α33,α44, unrestricted 3-lag model pGev pZoe pCla cbea −0.48 −0.16 −0.36 −0.04 4 Long-run normalised coefficients, βii, for three vectors pGev 1.00 0.00 0.00 pZoe 0.00 1.00 0.00 pCla 0.00 0.00 1.00 cbea −0.90 −0.98 −0.98 (0.05) (0.19) (0.06) Adjustment coefficients and standard errors pGev −0.79 0.07 −0.11 (0.17) (0.07) (0.18) pZoe −0.08 −0.23 −0.23 (0.21) (0.08) (0.22) pCla −0.06 0.01 −0.63 (0.16) (0.06) (0.16) cbea −0.03 −0.06 −0.05 (0.11) (0.04) (0.11) 5 LR test of complete pass-through: χ2(3)=3.43[0.33] β21=β31=β12=β32=β13=β23=0;β11=β22=β33=1;β41=β42=β43=−1; 6 LR test χ2(8)=8.13[0.42] β21=β31=β12=β32=β13=β23=0;β11=β22=β33=1;β41=β42=β43=−1; α12=α13=α21=α23=α31=α32=α41=α42=α43=α41=0 Panel ADF unit root tests: two lags and a trend. 5% = −3.48, 1% = −4.10 pGev pCla pZoe cbea 1 t-adf −1.38 −1.65 −1.24 −1.83 Root 0.92 0.91 0.93 0.95 Johansen approach 2 Trace test of number of cointegrating vectors: H0: rank≤ Two lags Three lags Eigenvalue p-value Eigenvalue p-value Rank 0 – [0.000]** – [0.000]** Rank 1 0.54 [0.000]** 0.41 [0.005]** Rank 2 0.46 [0.027]* 0.22 [0.008]** Rank 3 0.18 [0.092] 0.19 [0.018]* Rank 4 0.04 – 0.08 – 3 Adjustment coefficients α11,α22,α33,α44, unrestricted 3-lag model pGev pZoe pCla cbea −0.48 −0.16 −0.36 −0.04 4 Long-run normalised coefficients, βii, for three vectors pGev 1.00 0.00 0.00 pZoe 0.00 1.00 0.00 pCla 0.00 0.00 1.00 cbea −0.90 −0.98 −0.98 (0.05) (0.19) (0.06) Adjustment coefficients and standard errors pGev −0.79 0.07 −0.11 (0.17) (0.07) (0.18) pZoe −0.08 −0.23 −0.23 (0.21) (0.08) (0.22) pCla −0.06 0.01 −0.63 (0.16) (0.06) (0.16) cbea −0.03 −0.06 −0.05 (0.11) (0.04) (0.11) 5 LR test of complete pass-through: χ2(3)=3.43[0.33] β21=β31=β12=β32=β13=β23=0;β11=β22=β33=1;β41=β42=β43=−1; 6 LR test χ2(8)=8.13[0.42] β21=β31=β12=β32=β13=β23=0;β11=β22=β33=1;β41=β42=β43=−1; α12=α13=α21=α23=α31=α32=α41=α42=α43=α41=0 Note: Panel 1 reports Augmented Dickey–Fuller unit root tests, t-values and the estimated roots. Panel 2 reports the trace tests and eigenvalues. The null is that the rank is equal to or less than a certain number when the eigenvalue is zero. Models with two and three lags are tested. Panel 3 reports the adjustment coefficients for each variable when the model is estimated with three lags. Panel 4 reports the long-run and the adjustment coefficients when there are assumed to be three cointegrating vectors. Panel 5 reports the likelihood ratio test that pass-through is 1, and Panel 6 reports the test that pass-through is I and all the adjustment is due to costs. View Large Panel 4 reports the long-run coefficients for product prices set to either 1 or 0 in each equation. Note that no restrictions have been imposed on the system, except for the cointegration rank, even though five long-run coefficients are zero. The estimates of the coefficients of the bean costs are all negative and close to 1. The adjustment coefficients, also reported in Panel 4, show that all adjustments are probably due to changes in the individual product retail prices; there is no evidence of one product retail price affecting the others since all αij are close to zero and insignificant. Next, I test the restrictions that all the bean cost coefficients are −1, i.e. that the pass-through is 1 (Panel 5). The likelihood ratio test static has a p-value of 0.33, so the hypothesis is not rejected. Then I set adjustment coefficients to zero to test whether the retail prices affect each other (Panel 6). The p-value is 0.42, so there is no evidence that adjustment to the long-run equilibrium in one vector affects other prices. Finally, to highlight the short-run dynamics, I report the vector error correction model in Table 5, where clearly insignificant lags have been excluded.16 The error correction terms are lagged 2 months, since it takes several weeks for the imported beans to be processed and reach the market, though lagging them one month does not alter the results much. There is fairly rapid adjustment after a change in costs for the Gevalia and Classic coffees: About 60 per cent of a deviation from the long-run equilibrium is eliminated within a month. Moreover, lagged changes in bean costs affect the prices of Gevalia and Classic coffees, and changes in the price of Gevalia speed up changes in the price of Classic coffee in the following month. Table 5. The error correction model (ΔpGevΔpZoeΔpCla)t=(−0.58000−0.26000−0.60) (pGev−cbeapZoe−cbeapCla−cbea )t−2+ (−0.76000.500−0.5000.260−0.730.60 ) (ΔpGevΔpZoeΔpClaΔcbea )t−1+D (ΔpGevΔpZoeΔpCla)t=(−0.58000−0.26000−0.60) (pGev−cbeapZoe−cbeapCla−cbea )t−2+ (−0.76000.500−0.5000.260−0.730.60 ) (ΔpGevΔpZoeΔpClaΔcbea )t−1+D Note: All coefficients are significant at the 5 per cent level, at least. See Tables A1 and A2 in the Appendix for the general and parsimonious models. View Large Table 5. The error correction model (ΔpGevΔpZoeΔpCla)t=(−0.58000−0.26000−0.60) (pGev−cbeapZoe−cbeapCla−cbea )t−2+ (−0.76000.500−0.5000.260−0.730.60 ) (ΔpGevΔpZoeΔpClaΔcbea )t−1+D (ΔpGevΔpZoeΔpCla)t=(−0.58000−0.26000−0.60) (pGev−cbeapZoe−cbeapCla−cbea )t−2+ (−0.76000.500−0.5000.260−0.730.60 ) (ΔpGevΔpZoeΔpClaΔcbea )t−1+D Note: All coefficients are significant at the 5 per cent level, at least. See Tables A1 and A2 in the Appendix for the general and parsimonious models. View Large The equations in Table 5 look unintuitive because of the negative coefficients. Thus, to see how prices evolve over time, it is convenient to re-write the error correction model into levels. Equation (7) shows models for the prices of Gevalia and Zoegas Skåne; Classic Mellanrost is similar to Gevalia, pGevt=c+0.24pGev,t−1+0.18pGev,t−2+0.50cbea,t−1+0.08cbea,t−2pZoet=c+0.50pZoe,t−1+0.24pZoe,t−2+0.26cbea,t−2. (7) Both prices respond to an increase in bean costs with a lag. Then there is adjustment over the following months: after a 1 SEK increase in bean cost, the price of Gevalia increases by 0.50 SEK in the following month, and reaches 1 SEK after about 12 months. The price of Zoegas adjusts somewhat more slowly after a cost increase, rising to 1 SEK over 18 months. This might be because of the sluggish price decline during period 2012–2014, shown in Figure 6. 7. Concluding remarks The purpose of this paper is to shed light on the functioning of the Swedish retail coffee market by estimating long-run pass-through rates from the cost of green coffee beans to retail prices and analysing how they are related to horizontal and vertical market structure. The focus is on roasted and ground coffee, the market segment analysed in almost all previous studies on coffee markets. The Swedish market, which is similar to other Northern European markets, has four large brands and a very concentrated food retail sector with three dominating food chains (Swedish Competition Authority, 2011). I estimate that a 1 SEK cost increase raises prices by 0.69 SEK. However, pass-through rates are highly disbursed and vary from about 0 to slightly above 1. As these estimates include both the direct effect of the change in costs and the indirect effects due to strategic interaction, market structure may play a key role in explaining the dispersion. In fact, the four large brands, each of which sells several blends, have average pass-through rates that are 0.30–0.40 SEK higher than other brands. Another key finding is that the association between pass-through and market share is positive, though non-linear as a pass-through of about 1 seems to be an upper limit. This relationship seems to primarily be due to the market share of the blend and not the brand, i.e. pass-through rates are low for blends with small market shares irrespective of whether they are of a large or small brand. This finding of a positive association between pass-through and market share is the opposite of the prediction of Dornbusch’s (1987) model and the findings of Atkeson and Burstein (2008) and Hong and Li (2017), among others. Yet another finding is that vertically integrated blends do not have high pass-through rates, which is contrary to the results of Hong and Lee (2017). Discount, and to some extent premium, private labels have low pass-through rates relative to their market shares, while standard private labels have pass-through rates consistent with their market shares. Thus, there is no evidence of double marginalisation. The cointegration analysis of the dominant blends in eastern Sweden strengthens earlier findings. It is not possible to reject a pass-through rate of 1 for any of the four products. The adjustment to a cost shock takes about a year to be completed, probably due to the combined effect of the time it takes to process coffee beans, the existing stocks in food stores and the bargaining between roasters and retail chains. Since the four largest blends have a volume market share of about 50 per cent and the 10 largest a market share of nearly 70 per cent, the finding of pass-through rates of about 1 for large blends indicates that a substantial part of the Swedish coffee market is highly competitive. One reason earlier studies on the Swedish coffee market have not found evidence of market power is most likely their use of average retail prices collected by Statistics Sweden, which were based on a small selection of prices of popular blends.17 The question is why products with small market shares have low pass-through rates and vice versa. One hypothesis is that it is due to the functional form of consumer demand function, which would have to be concave and vary systematically across market shares, such that the smaller the market share the larger the response of the price elasticity to a price change. However, it is not obvious why the demand for small blends should respond more to price changes than the demand for large blends. Another hypothesis is that the regularly occurring bargaining between roasters and retail chains generates the relationship. Roasters might face strong resistance from retail chains when they wish to raise the wholesale prices of blends with small market shares, since retailers prefer to fill their shelves with the most popular blends or their own brands. And retail chains might be reluctant to lower the retail prices of small blends when wholesale prices go down, since they wish to increase their markups on blends that sell little. Prices on large blends, on the other hand, are changed when costs change, because both roasters and retailers compete on market shares. To conclude, the Swedish ground coffee market seems to be far from a typical monopolistic or oligopolistic market; pass-through rates are high, particularly for popular blends. In fact, a large part of the market seems to be quite competitive, although there are deviations from perfect competition. Nevertheless, more research is needed to explain the systematic positive relationship between pass-through rates and market shares. Supplementary data Supplementary data are available at European Review of Agricultural Economics online. Acknowledgements The author would like to thank Sven-Olof Daunfeldt, Niklas Rudholm, three anonymous reviewers and the editor for helpful comments. Financial support from the Swedish Competition Authority is gratefully acknowledged. Footnotes 1 See e.g. Morisset (1998), Dicum and Luttinger (1999), Fitter and Kaplinsky (2001), Oxfam (2002), Moore (2003), McCorriston, Sexton and Sheldon (2004), Talbot (2004, 2011), Consumers International (2005), Daviron and Ponte (2005), Green (2005), Gibbon (2007), Levy (2008), Fairtrade Foundation (2012) and World Vision (2014). 2 The behaviour of large coffee retail prices is similar in the other regional markets. 3 Since my measure of marginal cost is an industry-wide cost shock, the study is related to studies on tax incidence and on studies in international economics that analyse the effect of exchange rate changes on retail prices. 4 See RBB (2014) for a general review of pass-through. 5 The Dixit–Stiglitz model of monopolistic competition has identical firms and constant markup, and thus a pass-through rate of 1. Dornbusch (1987) introduces the assumption that firms take the effect of its own price on the overall price index into account, which makes pass-through lower the larger the impact a firm’s price has on the price index. The Dornbusch (1987) model has become a workhorse in international economics (Burstein and Gopinath, 2014). 6 Adachi and Ebina (2014a, 2014b) discuss the case of vertical relationships with homogeneous manufacturers and retailers in detail. 7 See Gaudin (2016) for the role of relative bargaining power between wholesalers and retailers in determining pass-through. 8 A related question is whether prices rise faster than they fall after a change in marginal costs (Peltzman, 2000). Asymmetry is usually considered a short-run phenomenon, which I ignore in the present study. In any case, it is difficult to identify asymmetry with 6 years of data when costs are highly persistent (see Figure 1). 9 There are a few non-organic Fairtrade blends in the Swedish market but their sales are tiny and they are not included in the sample. Including a dummy for organic Fairtrade blends in the regression does not affect the results. 10 It is also possible to check the effect of biases by re-scaling costs, up or down, by reasonable amounts. For example, when costs are 20 per cent higher than average costs, the actual pass-through rate is roughly 20 per cent lower than the estimate, and vice versa. A 20 per cent deviation from the average cost is large given that the quality of Arabica beans used is mainstream (International Trade Centre, 2011, chap. 11) and that most of them are substitutes. As reported below by Figure 2, the estimates of the pass-through rates vary much more than can be explained by the measurement errors. 11 The weight lost from roasting is common knowledge. See e.g. European Coffee Federation (2011). 12 Hassouneh et al. (2012) provide a review of approaches used when analysing fairly long-time series of market data. 13 If we assume that other marginal costs followed the consumer price index, as Bettendorf and Verboven (2000) do, they would have increased, roughly monotonically, by about 5 per cent in total over the study period. Wage costs probably increased a bit more, but they make up small components of costs (Durevall, 2007b). 14 Since I estimate many regressions, some of the key results are reported in Figures 2 and 4 and Figures A1 and A2 in the Appendix. The rest of the results are available on request. 15 In principle, there may be more than one common trend in the model because of other real marginal costs. If these contained a stochastic trend, product retail prices would not form a cointegrating vector with coffee bean costs only. However, as mentioned, other real marginal costs evolved slowly during the study period (relative to bean prices) and they do not seem to matter for the long-run relationships. In fact, other marginal costs are captured by the intercept, as deterministic trends added to the VAR model were insignificant (not reported). 16 No contemporaneous retail prices enter the model. This is partly because I do not have any instruments, but it is primarily because there is very little correlation between the prices when measured in rates of change. I condition on contemporaneous coffee bean costs in the general model since they are weakly exogenous. However, all the coefficients are insignificant, as one would expect (see Table A1 in the Appendix) . 17 Since 2012, Statistics Sweden uses cash register data to construct indices of food retail prices. Prior to that year, small samples of products were selected randomly with a probability based on their market share (Statistics Sweden, 2007, 2014). As a result, the series of average prices reported by Statistics Sweden for ground coffee primarily comprised coffees with large market shares. References Adachi , T. and Ebina , T. ( 2014 a). Cost pass-through and inverse demand curvature in vertical relationships with upstream and downstream competition . Economics Letters 124 ( 3 ): 465 – 468 . Google Scholar CrossRef Search ADS Adachi , T. and Ebina , T. ( 2014 b). Double marginalization and cost pass-through: Weyl–Fabinger and Cowan meet Spengler and Bresnahan–Reiss . Economics Letters 122 ( 2 ): 170 – 175 . Google Scholar CrossRef Search ADS Amiti , M. , Itskhoki , O. and Konings , J. ( 2014 ). Importers, exporters, and exchange rate disconnect . American Economic Review 104 ( 7 ): 1942 – 1978 . Google Scholar CrossRef Search ADS Anderson , S. P. , De Palma , A. and Kreider , B. ( 2001 ). Tax incidence in differentiated product oligopoly . Journal of Public Economics 81 ( 2 ): 173 – 192 . Google Scholar CrossRef Search ADS Aron , J. , Macdonald , R. and Muellbauer , J. ( 2014 ). Exchange rate pass-through in developing and emerging markets: a survey of conceptual, methodological and policy issues, and selected empirical findings . Journal of Development Studies 50 ( 1 ): 101 – 143 . Google Scholar CrossRef Search ADS Atkeson , A. and Burstein , A. ( 2008 ). Pricing-to-market, trade costs, and international relative prices . American Economic Review 98 ( 5 ): 1998 – 2031 . Google Scholar CrossRef Search ADS Auer , R. A. and Schoenle , R. S. ( 2016 ). Market structure and exchange rate pass-through . Journal of International Economics 98 : 60 – 77 . Google Scholar CrossRef Search ADS Besanko , D. , Dubé , J.-P. and Gupta , S. ( 2005 ). Own-brand and cross-brand retail pass-through . Marketing Science 24 ( 1 ): 123 – 137 . Google Scholar CrossRef Search ADS Bettendorf , L. and Verboven , F. ( 2000 ). Incomplete transmission of coffee bean prices: evidence from the Netherlands . European Review of Agricultural Economics 27 ( 1 ): 1 – 16 . Google Scholar CrossRef Search ADS Bonnet , C. , Dubois , P. , Villas-Boas , S. B. and Klapper , D. ( 2013 ). Empirical evidence on the role of nonlinear wholesale pricing and vertical restraints on cost pass-through . Review of Economics and Statistics 95 ( 2 ): 500 – 515 . Google Scholar CrossRef Search ADS Bonnet , C. and Villas-Boas , S. B. ( 2016 ). An analysis of asymmetric consumer price responses and asymmetric cost pass-through in the French coffee market . European Review of Agricultural Economics 43 ( 5 ): 781 – 804 . Google Scholar CrossRef Search ADS Bulow , J. I. and Pfleiderer , P. ( 1983 ). A note on the effect of cost changes on prices . Journal of Political Economy 91 ( 1 ): 182 – 185 . Google Scholar CrossRef Search ADS Burstein , A. and Gopinath , G. ( 2014 ). International prices and exchange rates. In: Gopinath, G., Helpman, E. and Rogoff, K. (eds), Handbook of International Economics , Vol. 4 . Amsterdam : Elsevier , 391 – 451 . Consumers International . ( 2005 ). From Bean to Cup: How Consumer Choice Impacts upon Coffee Producers and the Environment . London : Consumers International . Daviron , B. and Ponte , S. ( 2005 ). The Coffee Paradox: Global Markets, Commodity Trade and the Elusive Promise of Development . London : ZED Books . Dicum , G. and Luttinger , N. ( 1999 ). The Coffee Book: Anatomy of an Industry From Crop to the Last Drop . New York : New Press . Dornbusch , R. ( 1987 ). Real exchange rate and prices . American Economic Review 77 ( 1 ): 93 – 106 . Durevall , D. ( 2003 ). Competition and pricing: an analysis of the market for roasted coffee. Chap. 5 in High Prices in Sweden—A Result of Poor Competition? Swedish Competition Authority Report. Durevall , D. ( 2007 a). Demand for coffee: the role of prices, preferences and market power . Food Policy 32 ( 5–6 ): 566 – 584 . Google Scholar CrossRef Search ADS Durevall , D. ( 2007 b). Competition in the Swedish coffee market 1978–2002 . International Journal of Industrial Organization 25 ( 4 ): 721 – 739 . Google Scholar CrossRef Search ADS Eichenbaum , M. , Jaimovich , N. and Rebelo , S. ( 2011 ). Reference prices, costs, and nominal rigidities . American Economic Review 101 ( 1 ): 234 – 262 . Google Scholar CrossRef Search ADS Ericsson , N. R. and MacKinnon , J. G. ( 2002 ). Distributions of error correction tests for cointegration . Econometrics Journal 5 ( 2 ): 285 – 318 . Google Scholar CrossRef Search ADS European Coffee Federation . ( 2011 ). European Coffee Report 2010/2011. Rijswijk, The Netherlands. Fairtrade Foundation . ( 2012 ). Fairtrade and Coffee, Commodity Briefing, London. Fairtrade International . ( 2016 ). Driving Sales, Deepening Impact: Annual Report 2015–2016. https://annualreport15-16.fairtrade.net/en/. Accessed 9 February 2018. Feenstra , R. C. ( 2016 ). Advanced International Trade: Theory and Evidence . Princeton: Princeton University Press . Feuerstein , S. ( 2002 ). Do coffee roasters benefit from high prices of green coffee? International Journal of Industrial Organization 20 ( 19 ): 89 – 118 . Google Scholar CrossRef Search ADS Fitter , R. and Kaplinsky , R. ( 2001 ). Who gains from product rents as the coffee market becomes more differentiated? A value chain analysis . IDS Bulletin 32 ( 3 ): 69 – 82 . Google Scholar CrossRef Search ADS Fousekis , P. and Grigoriadis , V. ( 2017 ). Joint price dynamics of quality differentiated commodities: copula evidence from coffee varieties . European Review of Agricultural Economics 44 ( 2 ): 337 – 358 . Gaudin , G. ( 2016 ). Pass-through, vertical contracts, and bargains . Economics Letters 139 : 1 – 4 . Google Scholar CrossRef Search ADS Gibbon , P. ( 2007 ). Africa, tropical commodity policy and the WTO Doha Round . Development Policy Review 25 ( 1 ): 43 – 70 . Google Scholar CrossRef Search ADS Gilbert , C. ( 2007 ). Have we been mugged? Market Power in the Coffee Industry. Working Paper, CIFERM and Department of Economics, Università degli studi di Trento. Gilbert , C. ( 2008 ). Value chain analysis and market power in commodity processing with application to the cocoa and coffee sectors. Commodity Market Review 2007–08, 5–38, Food and Agriculture Organization of the United Nations, Rome. Green , D. ( 2005 ). Conspiracy of silence: old and new directions on commodities. Report, Oxfam, London. Hassouneh , I. , von Cramon-Taubadel , S. , Serra , T. and Gil , J. M. ( 2012 ). Recent developments in the econometric analysis of price transmission. Working Paper (2) , Transparency of Food Pricing (TRANSFOP) . Hong , G. H. and Li , N. ( 2017 ). Market structure and cost pass-through in retail . Review of Economics and Statistics 99 ( 1 ): 151 – 166 . Google Scholar CrossRef Search ADS Hoover , K. D. , Johansen , S. and Juselius , K. ( 2008 ). Allowing the data to speak freely: the macroeconometrics of the cointegrated vector autoregression . The American Economic Review 98 ( 2 ): 251 – 255 . Google Scholar CrossRef Search ADS ITA ( 2011 ). The Coffee Exporter’s Guide . International Trade Centre, Geneva . Johansen , S. ( 1991 ). Estimation and hypothesis testing of cointegration vectors in Gaussian vector autoregressive models . Econometrica 59 ( 6 ): 1551 – 1580 . Google Scholar CrossRef Search ADS Juselius , K. ( 2001 ). Unit roots and the demand for cigarettes in Turkey: Pitfalls and possibilities. Discussion Paper at the Economics Department, University of Copenhagen. Juselius , K. ( 2006 ). The Cointegrated VAR Model: Econometric Methodology and Empirical Applications . Oxford: Oxford University Press . Leibtag , E. , Nakamura , A. , Nakamura , E. and Zerom , D. ( 2007 ). Cost pass-through in the U.S. coffee industry. USDA Economic Research Report No. (ERR-38). Levy , D. L. ( 2008 ). Political contestation in global production networks . Academy of Management Review 33 ( 4 ): 943 – 963 . Google Scholar CrossRef Search ADS Lloyd , T. ( 2017 ). Forty years of price transmission research in the food industry: insights, challenges and prospects . Journal of Agricultural Economics 68 ( 1 ): 3 – 21 . Google Scholar CrossRef Search ADS Lloyd , T. A. , McCorriston , S. , Morgan , C. W. , Poen , E. and Zgovu , E. ( 2014 ). Retail price dynamics and retailer heterogeneity: UK evidence . Economics Letters 124 ( 3 ): 434 – 438 . Google Scholar CrossRef Search ADS McCorriston , S. ( 2013 ). Competition in the Food Chain. Working Paper (11), Transparency for Food Pricing (TRANSFOP). McCorriston , S. , Sexton , R. J. and Sheldon , I. M. ( 2004 ). Vertical Market Structure, Commodity Exports and Trade Reform. Paper presented at 7th Annual Conference on Global Economic Analysis, World Bank, June 2004, Washington, DC, USA. Moore , M. ( 2003 ). A World without Walls: Freedom, Development, Free Trade and Global Governance . Cambridge : Cambridge University Press . Moorthy , S. ( 2005 ). A general theory of pass-through in channels with category management and retail competition . Marketing Science 24 ( 1 ): 110 – 122 . Google Scholar CrossRef Search ADS Morisset , J. ( 1998 ). Unfair trade? The increasing gap between world and domestic prices in commodity markets during the past 25 years . World Bank Economic Review 12 ( 3 ): 503 – 526 . Google Scholar CrossRef Search ADS Nakamura , E. and Zerom , D. ( 2010 ). Accounting for incomplete pass-through . Review of Economic Studies 77 ( 3 ): 1192 – 1230 . Google Scholar CrossRef Search ADS Oxfam ( 2002 ). Mugged: Poverty in Your Coffee Cup . London : Oxfam International and Make Fair Trade . Peltzman , S. ( 2000 ). Prices rise faster than they fall . Journal of Political Economy 108 ( 3 ): 466 – 502 . Google Scholar CrossRef Search ADS RBB Economics . ( 2014 ). Cost pass-through: theory, measurement, and potential policy implications. Report prepared for the Office of Fair Trading, London. Rohman , A. and Friberg , R. ( 2016 ). Price rigidity in the Swedish beer market. Beiträge zur Jahrestagung des Vereins für Socialpolitik 2016: Demographischer Wandel–Session: Price-Setting Behavior of Firms, No. G11–V3. Sexton , R. J. ( 2013 ). Market power, misconceptions, and modern agricultural markets . American Journal of Agricultural Economics 95 ( 2 ): 209 – 219 . Google Scholar CrossRef Search ADS Statistics Sweden . ( 2007 ). Konsumentprisindex, Nettoprisindex, Harmoniserat index för konsumentpriser 2007, PR0101. Stockholm, Statistiska centralbyrån. Statistics Sweden . ( 2014 ). Konsumentprisindex (KPI) 2014 SCBDOK 3.2 PR0101. Stockholm, Statistiska centralbyrån. Sutton , J. ( 2007 ). Sunk Cost and Market Structure . Cambridge, MA : MIT Press . Swedish Competition Authority . ( 2011 ). Konkurrens och makt I den svenska livsmedelskedjan. Report prepared by Agrifood Economics Centre, Lund. Talbot , J. M. ( 2004 ). Grounds for Agreement: The Political Economy of the Coffee Commodity Chain . Lanham : Rowman and Littlefield . Talbot , J. M. ( 2011 ). The coffee commodity chain in the world-economy: Arrighi’s systemic cycles and Braudel’s layers of analysis . Journal of World-Systems Research 17 ( 1 ): 58 – 88 . Google Scholar CrossRef Search ADS Tyagi , R. K. ( 1999 ). A characterization of retailer response to manufacturer trade deals . Journal of Marketing Research 36 ( 4 ): 510 – 516 . Google Scholar CrossRef Search ADS Weyl , E. G. and Fabinger , M. ( 2013 ). Pass-through as an economic tool: principles of incidence under imperfect competition . Journal of Political Economy 121 ( 3 ): 528 – 583 . Google Scholar CrossRef Search ADS World Vision . ( 2014 ). Coffee’s Hidden Kick: Labour Exploitation in the Global Coffee Industry. DTL Fact Sheet Coffee, World Vision Australia. Author notes Review coordinated by Iain Fraser © Oxford University Press and Foundation for the European Review of Agricultural Economics 2018; all rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

European Review of Agricultural EconomicsOxford University Press

Published: Sep 1, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off