Buyer Power and Non-Binding Retail Price Recommendation

Transcription

1 Working Paper Series Economics No November 2017 Buyer Power and Non-Binding Retail Price Recommendation In Kyung Kim and Vladyslav Nora 1 1. Department of Economics, Nazarbayev University, Astana, Kazakhstan (c) copyright 2017, In Kyung Kim & Vladyslav Nora. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including notice, is given to the source.

2 Buyer Power and Non-Binding Retail Price Recommendations In Kyung Kim Vladyslav Nora November 2017 Abstract We study how the use of non-binding retail price recommendations (RPRs) is affected by buyer power in grocery retail market. Adopting the idea that RPRs serve as information sharing device between manufacturers and retailers, we show that increasing buyer power discourages the use of RPRs. Using the hand collected data set on the presence of RPRs for grocery products in Korea, we find that the more the sales of a product rely on powerful retailers, the less likely the manufacturer is to recommend a price, and hence share the information. Keywords: buyer power, manufacturer suggested retail prices, and grocery retailing JEL Classification Numbers: L11, L13, L81 Department of Economics, Nazarbayev University, in.kim@nu.edu.kz. Department of Economics, Nazarbayev University, vladyslav.nora@nu.edu.kz. 1

3 1 Introduction A salient feature of the retail industry in the last few decades is the increasing dominance of large firms. Chain retailers have gained substantial power in negotiating with suppliers. This might affect retail prices (Inderst and Valletti, 2011; Chen, 2003), producer incentives to invest in quality (Battigalli, Fumagalli, and Polo, 2007), or alter product variety (Chen et al., 2004; Inderst and Shaffer, 2007). Another possible effect is on the information sharing between firms. When facing tougher bargains from large retail chains, suppliers may be better off keeping to themselves. Indeed, giving away cost or demand information might result in powerful retailers capturing the suppliers information rents. Although product information can be shared through various channels, we focus on retail price recommendations (RPRs), whose role as information sharing device was recently highlighted in the literature (Buehler and Gärtner, 2013; Lubensky, 2017; Faber and Janssen, 2017). There are numerous examples when manufacturers provide non-binding price recommendations to retailers. For instance, in many countries recommended book prices are printed on covers, while gas prices are posted on websites of oil companies. In Australia recommended prices of tobacco products are published in a quarterly magazine, and RPRs for some grocery products in Korea appear directly on packages. Anecdotal evidence suggests that small independent retailers such as mom-and-pop stores often rely on this information for their pricing decisions. This paper offers theoretical and empirical investigation of the effect of buyer power on manufacturers use of RPRs. In our model, a manufacturer sells its product through two types of retailers, strong and weak. Strong retailers make take-it-or-leave it offers to the manufacturer, while weak retailers have to accept or reject the manufacturer s offer. Then buyer power is captured by a market share of strong retailers. We compare ex ante profit of the manufacturer in two cases, when it truthfully shares demand information with retailers, and when it keeps it private. A key tradeoff faced by the manufacturer is that sharing information increases downstream profit which can be captured from weak retailers, but destroys the information rent from trade with strong ones. We show that incentives to share information are decreasing in buyer power. This is the main idea underlying our empirical analysis. We examine how the variation in the use of RPRs across products can be explained by buyer 2

4 power in the product markets. The analysis exploits the hand-collected RPR information for a sample of products that covers more than 80 percent of sales in four processed food categories in Korea. We measure buyer power in each product market by the proportion of sales through chain retailers. Because they are larger firms with nation-wide presence and store brands, chain stores have higher leverege in negotiating with suppliers comparing to other retailers. Using the above data, we find that the increase in buyer power discourages the use of RPRs. Specifically, 10 percent point increase in the share of sales through chain stores induces 6 percent point decrease in the probability of using RPRs. Moreover, we find that the probability to use RPRs increases with the size of a manufacturer. This is consistent with the buyer power explanation because, other things being equal, the larger the manufacturer, the lower buyer power it faces. The estimation results are robust across different specifications and remain significant after addressing the potential endogeneity of the use of RPRs. This paper contributes to the literature on the effects of buyer power. Battigalli, Fumagalli, and Polo (2007) show that the increase in buyer power transfers the surplus from a producer to a retailer, and hence reduces the incentive of the producer to invest in the product quality. Similarly, Inderst and Shaffer (2007) show that increase in the concentration in the retail industry induces suppliers to produce less differentiated products, reducing product variety. Amongst papers that study the effect of buyer power on retail prices are Chen (2003) and Inderst and Valletti (2011). Empirical literature on the effect of the horizontal merger, and hence buyer power, on retail prices in grocery retail markets found evidence that either prices increase after the merger (Smith (2004) and Allain, Chambolle, Turolla, and Villas-Boas (2017)) or the effect depends on the market structure (Hosken, Olson, and Smith (2017)). 1 This paper is the first one to emphasize the effect of buyer power on information sharing between suppliers and retailers. Another strand of closely related literature is on the motivation and effects of RPRs. Buehler and Gärtner (2013) were the first to suggest that RPRs transmit a manufacturer s cost and demand information to a retailer. They show that under repeated interactions such communication can be credible and helps to achieve efficiency in the supply chain. Lubensky (2017) also treats RPRs as a communication device, but this time between a manufacturer and consumers, who infer industry 1 Also, there is a growing empirical literature on estimating a structural model of demand and supply that incorporates bargaining over prices in various markets (e.g. Crawford and Yurukoglu (2012) for multichannel television markets, Grennan (2013) for medical device markets). 3

5 costs from RPRs to guide their search decisions. RPRs are also used to transmit information in Faber and Janssen (2017), who empirically investigate the practice in gasoline markets. However, instead of conveying product information, RPRs help to sustain collusion between retailers by coordinating their prices. De los Santos, Kim, and Lubensky (2013) find empirical evidence that RPRs reduce retail prices in packaged food industry. Behavioral mechanism behind RPRs is examined by Puppe and Rosenkranz (2011) and Bruttel (2016) who consider RPRs as reference points that affect consumers valuations. Adopting the idea of Buehler and Gärtner (2013) that RPRs transmit information from manufacturers to retailers, we show how the use of RPRs is affected by the rise in buyer power. The rest of the paper is organized as follows. Section 2 provides institutional details on RPR and grocery retail industry in Korea. Section 3 develops a theoretical framework behind the effect of buyer power on information sharing incentives. Section 4 explains the sample data and provides descriptive statistics. Section 5 describes our empirical models and findings. Finally, Section 6 concludes. 2 Background Buyer power in retail industry Manufacturers distribute their products through a variety of retail channels including hypermarkets, supermarket chains, convenience store chains, corner shops etc. Retailers differ with respect to their size, assortment, geographical coverage, the way they procure the products, availability of store brands etc. Chain retailers may be present in many cities throughout the country, whereas independently owned local shops operate in a single location. 2 Retail chains have centralized purchasing departments that directly negotiate with producers, whereas independently owned shops buy the products from local wholesale dealers for predesignated prices. Also, retail chains often have variety of store brands and can easily substitute away from the branded products. These give rise to the differences in bargaining power between retailers, when negotiating with manufacturers. 3 2 For example, an average convenience store chain in Korea operated 1,750 stores in 2009 (Retail Magazine Vol.11, 2011). 3 Draganska, Klapper, and Villas-Boas (2010) empirically assess the role of the retailer size, the presence of store brand, and assortment as determinants of buyer power. 4

6 Independent store owners claim high costs, the lack of products, and no delivery as reasons of their low participation. It is clear that chains have higher bargaining power compared to independent stores. The main features of Korean grocery retailing are summarized in Table A-1, that shows the number of stores and their sizes (as measured by the number of employees) by retailer category. Although the number of independent stores is higher than the number of chain stores, the average chain store is three times larger. 4 Retail price recommendations Historically, in Korea manufacturers were suggesting retail prices for their products. These recommended prices were typically printed on packages and updated from time to time by manufacturers. Interestingly, however, in parallel with the rapid spread of the chain stores in recent decades, the use of RPRs by manufacturers has declined. This trend can be partially attributed to the more stringent regulation towards RPRs. For instance, in 1999 price recommendations were banned on certain products including clothes and some consumer electronics such as TVs and VTRs, and the ban has been extended to cameras in 2000 and PCs in However, RPRs also have disappeared on some products not affected by the policy changes. For example, toys and sport equipments that traditionally had RPRs, no longer have them. During the use of RPRs was briefly banned on some processed food categories: snacks, biscuit & pie, ice cream, and instant noodle. Interestingly, when the ban has been revoked one year later, manufacturers reintroduced RPRs on some but not all of their products. 5 Our empirical analysis is a case study of the logic behind the reintroduction of RPRs for products in these four processed food categories after the lift of the ban. The main advantage of using this product group is that the lift of the ban has provided us with a natural experiment that allows to observe manufacturers decisions to start using RPRs. Because RPRs are essentially cheap talk, a mere fact that a product has RPR does not imply that it is informative. 6 However, if a 4 Recently, there was an effort to increase the bargaining power of independent stores by establishing regional distribution centers that purchase products in bulk and distribute them to independent retailers. However, only 17 percent of independent stores are transacting with them in 2013 (Kim, Kim, and Cho (2013)). 5 The main reason behind the lift of the ban is that contrary to expectation of the authority, the retail price went up. See press release, Ministry of Trade, Industry, and Energy, 8 July 2009 and 30 June In general it is hard to identify if the retail prices respond to the information delivered by RPRs, or to the changes in costs. One exception is Faber and Janssen (2017) who examine the gasoline markets where the spot prices are observable, and show that RPRs indeed deliver additional information. 5

7 manufacturer reintroduces RPR, the manufacturer must rationally expect that it transmits some information, and increases profit. Moreover, the sales during the ban period provide a natural instrumental variable for our empirical analysis. 3 Theoretical framework There is a single manufacturer M and two independent downstream markets m {S, W }. In each market m there is a single retailer R m. Retailers have different bargaining power in relation to the manufacturer. The party with higher bargaining power has an advantage of making take-it-orleave-it offer to its trading partner. In particular, R S is a strong retailer who makes take-it-or-leave it offer to M, whereas R W is weak retailer who has to accept or reject M s offer. 7 M produces an intermediate good at zero cost, which then can be transformed by retailers into the final good using a costless one-to-one technology. We suppose that M has an alternative to integrate forward in each of the markets by producing and distributing the final good on its own. If M integrates forward in a market, then it becomes a sole supplier of the final good in this market. However, integration is not efficient because M pays marginal cost c > 0 of transforming the intermediate good into the final good, whereas the corresponding cost is zero for the existing retailer. Consumer demand in markets S and W is, correspondingly, λd(p, θ) and (1 λ)d(p, θ), where p is a retail price, λ is the relative size of market S, and θ {H, L} is the demand parameter. We let D(p, θ) be decreasing in p and D(p, L) < D(p, H) for each p. The demand is low (θ = L) with probability α (0, 1). We suppose that before learning the demand, M can commit to truthfully share this information with retailers using RPR. 8 Moreover, M cannot share information exclusively with one of the retailers and either must reveal the demand to both, or to no one. 9 We study M s incentives to use RPRs by comparing its profits with information sharing between M and retailers, and without information sharing. Formally we have the sequential Bayesian game which proceeds 7 There are many cases when big retail chains are known to give non-negotiable offers to branded good producers. For instance, in September 2015, the Australian Competition and Consumer Commission flagged concerns with supermarkets Woolworths and Aldi that they were offering suppliers an agreement which gives the impression the terms cannot be negotiated. See ACCC press release MR 152/16 from 25 August Demand fluctuates over time due to changes in consumer tastes, introduction of new products, rebranding etc. Naturally, manufacturer cannot foresee all changes, but can learn their impact, whereas retailers remain unaware. Buehler and Gärtner (2013) show that under repeated interactions RPRs can credibly transmit this new information. 9 Information leakage within a supply chain is emphasized in the literature. In the extreme case a manufacturer cannot prevent the information shared with one retailer from reaching the others. See Ha and Tang (2017). 6

8 in the following stages. 1. Quality θ is drawn from the prior distribution and is observed by M. 2. Given θ, public signal s {H, L} is chosen. In information sharing regime the signal reveals the state, i.e. P (s = θ θ) = 1, whereas without information sharing it is uninformative, i.e. P (s L) = P (s H) for s {H, L}. 3. Simultaneously, R S makes an offer to M, and M makes an offer to R W. An offer is a vector (w, t) R 2 specifying wholesale price w and transfer t from a retailer to M. 4. M and R W simultaneously decide to accept or reject offers. If an offer in market m is rejected, then M integrates forward in this market. 5. Retailers choose retail prices and profits are realized. We investigate Perfect Bayesian equilibrium outcomes of the game. Given θ, denote M s profit from both markets when integrating forward by Π θ = arg max p (p c)d(p, θ), the maximum profit from both markets achievable by an efficient retailer under complete information by Π θ = arg max p pd(p, θ), and the maximum expected surplus from both markets achievable by an efficient retailer under incomplete information by Π 0 = max p pe[d(p, θ)]. In order to guarantee positive information rent we assume that Π 0 Π H α ( Π L Π L). (1) First, consider information sharing regime. Clearly, R S captures the surplus of M by offering contract (0, λ Π θ ). Moreover, M captures the surplus of R W by offering (0, (1 λ)π θ ), So, under information sharing M obtains ex ante expected profit λe[ Π θ ] + (1 λ)e [ Π θ]. Now consider the game without information sharing. Note that any contract offered by R S and accepted by H type of M is also accepted by L type. Hence, R S can trade either with both types by offering (0, Π H ), or only with L type by offering (0, Π L ). Comparing R S s profit and using (1) it follows that R S trades with both types of M. Finally, M captures the entire expected surplus of R W by offering (0, U 0 ). So, under no information sharing M obtains ex ante expected profit equal to λ Π H + (1 λ)π 0. The following proposition summarizes our results. 7

9 0 1 Figure 1: Comparison of the manufacturer s ex ante profit with and without information sharing. Proposition. There exists λ such that equilibrium ex ante expected profit of M is strictly higher with information sharing if and only if λ < λ. In particular the profits are as follows: (i) With information sharing: λe[ Π θ ] + (1 λ)e [ Π θ]. (ii) Without information sharing: λ Π H + (1 λ)π 0. Figure 1 illustrates the results. The ex ante expected profit of M with and without information sharing is represented, correspondingly, with a solid and a dotted line. When the size of the market with powerful retailer R S is relatively small, λ < λ, we see that M prefers sharing information, and keeps information private otherwise. This is the main idea underlying our empirical analysis: sharing information increases downstream profit which can be captured from low power retailers, but destroys the information rent from trade with high power retailers. So, if RPRs transmit information from manufacturers to retailers, we should observe a negative relation between the use of RPRs and the sales share of powerful retailers. 8

10 4 Data We use sales data posted at Food Information Statistics System (FIS) maintained by Korea Agro- Fisheries & Food Trade Corporation, a public institution in Korea. 10 It provides quarterly sales information for all retailers aggregated across the following categories: department store, hypermarket, chain supermarket, convenience store, (independent) supermarket, and corner shop. Retailers in the first four categories own multiple stores nationwide and directly negotiate with manufacturers, whereas retailers in the remaining two categories are independently owned and supplied by wholesalers. Hence, we call stores in the first four categories chain stores and stores in the last two categories independent stores. 11 Using a web crawler, we download sales information for all products in the four processed food categories (biscuit & pie, ice cream, instant noodle, and snack) for the year To prevent the seasonality of sales affecting our results, we aggregate quarterly sales into yearly sales for each product. Table A-2 presents sales of all producers and eight major producers only. 12 The combined market share of the major producers ranges from 78 to 96 percent across four product categories. Also, several firms appear to be major producers in multiple categories. For instance, the market share of Lotte is 29 percent, 51 percent, and 12 percent in biscuit & pie, ice cream, and snack category, respectively. Given that the market is dominated by the major producers, we focus our attention on their RPR decisions. To collect RPR information we visited 18 stores in three cities in Korea: Seoul, Sungnam, and Donghae. These stores cover all retailer categories in our data. Out of 606 products produced by major producers, we found and recorded RPR information for 318 products. The products not offered in any of the visited stores mainly have low sales. For instance, Table A-2 shows that the combined sales of the 84 products in snack category produced by major producers that we did find is 782 million dollars, whereas the combined sales of the remaining 75 products that we could not find is only 43 million dollars. Table 1 reports the descriptive statistics of the sample data. Among 318 products, 42 percent (134 products) have RPRs and the rest 58 percent (184 products) do not. For each product we Some of the convenience stores are independently owned. For example, Table A-1 shows that there are 12,549 chain convenience stores (77%) and 3,783 independent ones (23%). We consider convenience stores as chain stores. 12 They are the largest producers in terms of sales in the four categories we consider in the analysis. 9

11 Table 1: Descriptive Statistics Variable Avg. Std. Dev. Min. Max. Obs. Product Level Use of RPR Percentage of sales through chain stores Sample period (2016) Ban period (July June 2011) Producer Level Number of products (Thousand) Total sales (Bil. USD) We apply the average exchange rate during 2016 which is 1,161 Korean Won per dollar. compute the proportion of chain store sales in the total sales. It measures the market size of chain stores relative to the entire market size, and hence, captures buyer power. In 2016 the average proportion of chain store sales is 60 percent. The left panel of Figure 2 shows that the percentage of products with RPRs ranges from 26 for instant noodles to 52 percent for ice cream. Also, the right panel shows that each major producer is using RPRs on some of its product. A striking feature of the data is that the retail channel structure is different for products with and without RPRs. In Figure 3 we compare distributions of the proportion of sales through chain stores between the two product groups. Proportion of sales through chain stores tends to be higher for products without RPRs. Specifically, the average proportion of chain stores for products that do not have RPRs is 63.5 percent, whereas it is only 54.2 percent for products with RPRs. Breaking up the comparison by product category, Figure A-1 shows the similar pattern. In the next section we confirm this observation by the regression analysis controlling for other factors that may affect the use of RPRs. 5 Empirical Analysis Using the probit model we examine how buyer power affects the manufacturer s decision to use RPR. Let π denote the change in profits due to the introduction of RPR given by π i = δ ChainStore i + z i λ + u i, (2) 10

12 By Product Category By Producer Biscuit & Pie Bingrae Crown/Haitai Ice cream Instant Noodle Lotte Nongsim Orion Ottogi Paldo Snack Samyang Percentage of Products with RPR Percentage of Products with RPR Figure 2: Percentage of Products with RPR. Average Percentage Density No RPR: 63.5 RPR: Percentage of sales through chain stores No RPR RPR Figure 3: Distribution of the Chain Store Proportion for Products with and without RPR. 11

13 Table 2: Marginal Effects of Buyer Power on RPR Use Variable (1) (2) (3) (4) (5) Chain Store (0.001)*** (0.001)*** (0.002)*** (0.001)*** (0.002)*** Total Sales (0.010)*** (0.011)*** Fixed Effects Product Category N N Y N Y Producer N Y Y N N Pseudo R-squared Observations The table presents marginal effects at sample means. The dependent variable RP R is equal to one when RPR is used for the product and zero when it is not. Robust standard errors are in parentheses. The notation *** indicates significance at 1%, ** at 5%, and * at 10% level. where ChainStore i is the percentage of sales through chain stores for product i, and z i includes a constant as well as a full set of controls for producer and product category. 13 We assume that the error term u i has a standard normal distribution. Let the variable RP R i be equal to one if RPR is used for product i and be equal to zero otherwise. Assuming that a producer would use RPR if and only if it increases profit, we obtain the probit model P r[rp R i = 1 x i ] = Φ(x i β), (3) where x i = [ChainStore i, z i ], β = [δ, λ ], and Φ( ) is the cdf of the standard normal distribution. The first three columns in Table 2 presents the marginal effects at sample means. We start from the simplest specification without any fixed effect in (1) and arrive at the full model in (3). We find that the higher the percentage of sales through chain stores, the less likely the product is to have RPR. For instance, estimates in column (3) show that the likelihood that the producer recommends retail prices decreases by 6 percent point when the proportion of chain store sales rises by 10 percent point. The result is consistent with our idea that a producer who faces a lot of powerful retailers has lower incentives to use RPR. 14 Producer characteristics also affect its bargaining power in relation to retailers. One possibility is that the larger the producer, the more leverage it has in negotiating the contract with retailers. 13 As noted earlier, we could not collect RPR information for some products with low sales. However, not all products with low sales are unavailable for our analysis, relieving the concern for potential sample selection problem. The minimum sales of a product in the sample data is 9 thousand dollars. 14 We also consider linear probability model (LPM) by replacing π i with RP R i in equation (2). LPM estimate of the marginal effect is approximately the same. 12

14 Estimate of the Producer Fixed Effect Total Sales in the Professed Food Category (Billion USD) Figure 4: Firm Size and the Likelihood of Using RPR. Therefore, larger producers are more likely to use RPRs. Indeed, the estimated producer fixed effect seem to be increasing in the total sales of the producer in the processed food category as illustrated in Figure 4. We confirm this observation by replacing the producer fixed effect with the total sales of the producer of product i in equation (2). 15 The last two columns in Table 2 report the results under this specification. Products of manufacturers with higher total sales are more likely to have RPRs: one billion dollar increase in the total sales lead to 3.8 percent point increase in the probability of using RPR. We also estimate the probit model by product category. The marginal effects of the chain store proportion reported in Table A-3 are negative for all product categories, and significant at 5 percent for two categories (biscuit & pie and snack). The effect is the largest for snack category: given 10 percent point increase in the proportion of chain store sales, the probability of using RPR decreases by 23 percent point. Robustness In this section, we address two empirical issues: the potential endogeneity of the chain store proportion, and the presence of a vertically integrated producer. The use of RPR may affect the distribution of sales of a product across retail channels. One 15 When instead of the total sales we use the number of products as a proxy for the manufacturer s size, we find similar results. 13

15 possibility is that without RPRs retail chains still have more demand information than small independent stores, and so their retail prices are closer to the optimum. For instance, suppose that a manufacturer learns that demand for its instant noodle brand has declined due to the overall shift of consumer tastes away from packaged food. Whereas sophisticated retail chains might have already incorporated this information in their pricing, the mom-and-pop stores might be unaware of it. Hence, when the manufacturer adjusts the RPR to reflect this information, independent retailers decrease their prices while chain stores do not. Hence, the proportion of sales through independent stores grows in response to the change in RPR. Clearly, this effect can go both ways. Hence, the above analysis may suffer from endogeneity problem. To alleviate this concern, we exploit the ban of RPR in Korea from July 2010 to June Because the proportion of chain store sales is likely to be correlated across years, we use the average proportion of chain store sales during the ban period, ChainStore 2010, as an instrumental variable in the model below, ChainStore i = α ChainStore 2010 i + z i γ + ε i, (4) π i = δ ChainStore i + z i λ + u i. (5) The sales information during the ban period comes from Nielsen Korea Retail Measurement Service and covers 152 out 318 products in our sample. The products for which the data is unavailable were introduced after the lift of the ban in According to Table 1, the average proportion of chain store sales during the ban period is 43 percent, and is lower than in Using the sub-sample of 152 products, we employ the following estimation strategies. First, under the assumption that (ε i, u i ) is independent and identically distributed bivariate normal, we jointly estimate equations (4) and (5). 16 Second, we use the linear two-stage least squared (2SLS) approach. That is, we first estimate equation (4) and then regress RP R i on the predicted value of ChainStore i along with z i. The advantage of this approach is that it depends on fewer distributional assumptions. Table 3 presents the estimation results of both strategies. The upper panel shows that the chain store proportions are strongly correlated, validating our instrument. The estimation results 16 Note that instead of π i, we observe RP R i that is equal to one if π i > 0 and zero otherwise. 14

16 Table 3: Estimates using Instrumental Variables MLE 2SLS Variable (1) (2) (3) (4) Dependent Variable: Chain Store Chain Store (0.115)*** (0.112)*** (0.119)*** (0.113)*** Total Sales (0.477)*** (0.485)*** Constant (7.813)*** (8.193)*** (8.099)*** (8.332)*** Dependent Variable: RPR Chain Store (0.012)*** (0.011)*** (0.005)** (0.005)** Total Sales (0.041)*** (0.017)*** Constant (0.776)*** (0.667)** (0.314)*** (0.308)*** Fixed Effects Product Category Yes Yes Yes Yes Producer Yes No Yes No Endogeneity Test p-value Observations The table presents estimation results of equations (4) and (5). None of the 7 products produced by Ottogi have RPR and hence, they are dropped in (1) and (3). Robust standard errors are in parentheses. The notation *** indicates significance at 1%, ** at 5%, and * at 10% level. of equation (5) in the bottom panel suggest that the marginal effect of chain store proportion is approximately the same under the two approaches. Under the first approach the estimates suggest that the likelihood of using RPR decreases by 13 percent points when the proportion of chain store sales increases by 10 percent points. 17 Similarly, under 2SLS approach 10 percent point increase in the proportion of chain store sales leads to 12 percent point decrease in the likelihood of using RPR. Test of the null hypothesis of exogeneity is rejected at the 10 percent level in 2 out of 4 specifications, justifying our use of the instrumental variable. Lotte, the largest producer in the sample, is also running its own hypermarket and supermarket chains. As there is no bargaining between them, the old channel proportion for these products may be misleading. However, when we estimate the probit model without 83 products produced by Lotte, the estimation results are qualitatively the same as before as reported in Table 4: the 17 We calculate the marginal effect at sample means as the estimate of the coefficient of ChainStore (-0.032) times the value of the standard normal pdf at x ˆβ. 15

17 Table 4: Marginal Effects without Products of the Vertically Integrated Manufacturer Variables (1) (2) (3) (4) (5) Chain Store (0.002)*** (0.002)*** (0.002)*** (0.002)*** (0.002)*** Total Sales (0.026)*** (0.028)*** Fixed Effects Product Category No No Yes No Yes Producer No Yes Yes No No Pseudo R-squared Observations The table presents marginal effects at sample means. Products produced by Lotte are dropped from the analysis. The dependent variable RP R is equal to one when RPR is used for the product and zero when it is not. Robust standard errors are in parentheses. The notation *** indicates significance at 1%, ** at 5%, and * at 10% level. probability of recommending retail prices goes down by 6 percent point in response to 10 percent point increase in the proportion of chain store sales, and goes up by 8.8 percent point in response to 1 billion dollar increase in total sales. 6 Conclusion In this paper, we have proposed a simple of model that explains how buyer power affects the manufacturer s incentive to share the demand information with retailers. The model emphasizes the tradeoff between sharing information to increase downstream profit which can be captured from retailers with low buyer power, and keeping information private which creates information rent from trade with high power retailers. Using the unique hand-collected data, we show that the likelihood that a manufacturer is using RPR is decreasing in the market share of chain stores. Moreover, the larger the manufacturer, the more likely it recommends retail prices for its products. In light of recent literature arguing that RPR is a vehicle for the information transmission from the supplier to retailers, our results suggest that increase in buyer power is detrimental to the information sharing in the supply chain. 16

18 References Allain, M.-L., C. Chambolle, S. Turolla, and S. B. Villas-Boas (2017): Retail Mergers and Food Prices: Evidence from France, The Journal of Industrial Economics, 65(3), Battigalli, P., C. Fumagalli, and M. Polo (2007): Buyer power and quality improvements, Research in Economics, 61(2), Bruttel, L. (2016): The Effects of Recommended Retail Prices on Consumer and Retailer Behaviour, Economica. Buehler, S., and D. L. Gärtner (2013): Making sense of nonbinding retail-price recommendations, The American Economic Review, 103(1), Chen, Z. (2003): Dominant Retailers and the Countervailing-Power Hypothesis, The RAND Journal of Economics, 34(4), Chen, Z., et al. (2004): Monopoly and product diversity: The role of retailer countervailing power, Discussion paper, Carleton University, Department of Economics. Crawford, G. S., and A. Yurukoglu (2012): The welfare effects of bundling in multichannel television markets, The American Economic Review, 102(2), De los Santos, B., I. K. Kim, and D. Lubensky (2013): Do MSRPs decrease prices?, Working Papers , Indiana University, Kelley School of Business, Department of Business Economics and Public Policy. Draganska, M., D. Klapper, and S. B. Villas-Boas (2010): A larger slice or a larger pie? An empirical investigation of bargaining power in the distribution channel, Marketing Science, 29(1), Faber, R. P., and M. Janssen (2017): On the effects of suggested prices in gasoline markets, Forthcoming in Scandinavian Journal of Economics. Grennan, M. (2013): Price discrimination and bargaining: Empirical evidence from medical devices, The American Economic Review, 103(1), Ha, A. Y., and C. S. Tang (2017): Handbook of Information Exchange in Supply Chain Management. Springer. Hosken, D. S., L. M. Olson, and L. K. Smith (2017): Do retail mergers affect competition? Evidence from grocery retailing, forthcoming in Journal of Economics & Management Strategy. Inderst, R., and G. Shaffer (2007): Retail mergers, buyer power and product variety, The Economic Journal, 117(516), Inderst, R., and T. M. Valletti (2011): Buyer power and the waterbed effect, The Journal of Industrial Economics, 59(1), Kim, S.-K., C.-K. Kim, and C.-H. Cho (2013): A Study on improving wholesale logistics system for strengthennig competitiveness of Korean small-medium sized retailing, Working Papers , Korea Institute for Industrial Economics & Trade. 17

19 Lubensky, D. (2017): A model of recommended retail prices, The RAND Journal of Economics, 48(2), Puppe, C., and S. Rosenkranz (2011): Why Suggest Non-Binding Retail Prices?, Economica, 78(310), Smith, H. (2004): Supermarket choice and supermarket competition in market equilibrium, The Review of Economic Studies, 71(1),

20 Appendix Table A-1: Distribution Channel Number Avg. Number Retailer Category of Stores of Employees Chain Stores Chain Convenience Store 12, Chain Supermarket Department Store Hypermarket Total 13, Independent Stores Supermarket 3, Corner Shop 40, Convenience Store 3, Total 47, Source: Retail Magazine Vol.11, 2011 and Korea Statistical Information Service. The number of stores is as of September 2010 except department stores that is as of the end of The number of employees is also as of the end of An independent store equipped with two or more cash registers is defined as supermarket, whereas a store with only one cash register is defined as a corner shop. A store that runs 24 hours a day is defined as convenience store. 19

21 Table A-2: Number of Products and Sales of Major Producers Number of Products Sales Producers # % Mil. USD % Biscuit & Pie All Producers Major Producers Only Crown/Haitai Lotte Orion Major Total Included in the Sample Ice cream All Producers Major Producers Only Bingrae Crown/Haitai Lotte Major Total Included in the Sample Instant Noodle All Producers , Major Producers Only Nongsim Ottogi Paldo Samyang Major Total , Included in the Sample , Snack All Producers 1, , Major Producers Only Crown/Haitai Lotte Nongsim Orion Samyang Major Total Included in the Sample The average exchange rate during 2016, 1,161 Korean Won per dollar, is applied. 20

22 Table A-3: Marginal Effects by Product Category Biscuit & Pie Ice cream Instant Noodle Snack Variables (1) (2) (3) (4) (5) (6) (7) (8) Chain Store (0.003)** (0.003)** (0.002) (0.002) (0.004) (0.004) (0.007)*** (0.006)*** Total Sales (0.021)*** (0.016) (0.036)*** (0.031)*** Fixed Effects Product Category Yes Yes Yes Yes Yes Yes Yes Yes Producer Yes No Yes No Yes No Yes No Pseudo R-squared Observations The table presents marginal effects at sample means. Estimation is performed using products in each of the 4 product categories one by one. The dependent variable RP R is equal to one when RPR is used for the product and zero when it is not. Seven instant noodles produced by a producer that do not use RPR at all (Samyang) are dropped. Robust standard errors are in parentheses. The notation *** indicates significance at 1%, ** at 5%, and * at 10% level. 21

23 Biscuit & Pie Ice cream Density Average Percentage No RPR: 65.3 RPR: 57.5 Density Average Percentage No RPR: 50.4 RPR: Percentage of sales through chain stores Percentage of sales through chain stores Instant Noodle Snack Density Average Percentage No RPR: 69.0 RPR: 69.1 Density Average Percentage No RPR: 70.3 RPR: Percentage of sales through chain stores Percentage of sales through chain stores No RPR RPR Figure A-1: Distribution of the Chain Store Proportion by Product Category. 22

24 (a) Instant Noodle with RPR (b) Instant Noodle without RPR Figure A-2: RPR Example 23