Gatekeepers to consumer markets: the role of retailers in international trade

Abstract

This study explores how recent developments in the retail sector affect trade in consumer goods. It focuses on three developments in the retail sector: (i) internationalization; (ii) market structure; and (iii) the growing market share of retailers' private labels. Using gravity model estimation techniques it is found that the foreign operations of a retailer are positively related to imports from the host to the home country of the retailer. Imports are negatively related to ownership concentration, while the market share of private labels is negatively related to imports of food and positively related to imports of non-food consumer goods. For both product categories private labels shift sourcing towards poor countries. The trade response to trade liberalization is higher both at the extensive and intensive margin in countries with lower retail concentration.

Keywords:

• international trade
• private labels
• market structure

Acknowledgements

The author would like to thank Anna Jankowski, Hans-Petter Hanson, Samuel Hill and Sajedul Hoq for excellent research assistance, Enrico Pinali and Massimo Geloso-Grosso for providing case studies to the OECD Trade Policy Working Paper on which this paper is built and Julian Arkell, Steve Burt, John Dawson, Linda Fulponi, Tommy Stahl Gabrielsen, Thomas Reardon, Frank von Tongeren, participants at the Groningen workshop on the gravity model 18–19 October 2007, and three anonymous referees for useful comments and suggestions. The usual disclaimer applies.

Notes

1. So-called slot allowances are widespread in the industry and have been analyzed extensively in the industrial organization literature, focusing on the competitive impact of such practices. This literature uses theoretical models to explain the competitive outcome of observed practices, often in the United States. However, see also Foros and Kind (2008) for a recent study based on the Norwegian experience and Dueñas-Caparas and del Prado (2006) for a case study on the Philippines.

2. Dell, Alticor, Avon and AAFES have global coverage and are not included in the average.

3. Wal-Mart actually operates the largest private satellite communication network in the world (Basker 2007).

4. Consumers do not have much opportunity to store food in their homes and therefore are willing to pay for nearby shopping.

5. The 2005 data include 38 countries and 80 product categories. US data do not cover Wal-Mart. A separate study of this retailer found that private labels accounted for 17% of this company's sale – not too far from the US average.

6. TRAINS is a database provided by UN Statistics in cooperation with the World Bank and WTO and is available online to subscribers. EU is treated as one nation in TRAINS and it is assumed that all intra-EU tariffs are zero. Bilateral tariffs are collinear with regional trade agreements, which is why these are not included in the regressions.

7. CEPII is a French research institute, http://www.cepii.fr/francgraph/bdd/distances.htm

8. Since the FDI dummy is a bilateral variable, both exporter and importer country fixed effects are applied in this section. The Vuong test suggests that zero inflated Poisson is better than PPML. Zero inflated Poisson regressions typically yield coefficients with opposite signs for the probit and the count model. This is the case in the regressions presented in Table 2 as well, except for the coefficient on tariffs on non-food consumer goods, which is negative in both regressions. This implies that countries facing a relatively high tariff have a lower probability of zero trade than countries facing a lower tariff. The results are consistent with our model that predicts that the retail margin, ceteris paribus, increases with the tariff rate.

9. Retailers do establish sourcing/procurement offices in a number of developing countries in order to access and monitor suppliers. The FDI dummy does not comprise sourcing/procurement offices, however.

10. Mintel kindly provided these data. The data were, however, difficult to replicate as it is not clear what the denominator in the reported market share is. The share used here is the sales of the five largest retailers as reported by Mintel divided by private consumption expenditure. A possible problem with this is that private consumption expenditure includes expenditure on a host of services, including housing that may take a different share of consumer expenditure in different countries. However, all reporters are OECD countries with similar expenditure patters, so it should not be a serious problem.

11. The correlation coefficient is –0.24, significant at a 1% level.

12. http://epp.eurostat.ec.europa.eu/portal/page?_pageid=1996,45323734&_dad=portal&_schema= PORTAL&screen=welcomeref&open=/&product=EU_MASTER_distributive_trade&depth= 2 ; http://www.census.gov/epcd/www/concentration.html; http://www.stat.go.jp/english/data/ service/2004/index.htm (Eurostat 2007).

13. The Vuong test for the food imports regressions showed that zero inflated Poisson is preferred, while for non-food imports this was not the case (there are far fewer zeros for non-food imports). A result that only appears in the regressions in Table 3 is a positive relation between the probability of zero and the common border dummy. In other words, when controlling for retail density the probability of importing food from an adjacent country is lower than from a more distant country. One can only speculate what may drive this result. One possibility is a spurious correlation between a geographically fragmented retail sector and high non-tariff barriers on food that competes with local production, and that neighboring countries have similar food production patterns, e.g. due to similar climate conditions. More research is needed before any firm conclusions can be reached on this issue.

14. The number of zeros is smaller for this category and the Vuong test indicates that PPML is the preferred estimator.

15. These regressions are included in order to check robustness, but it is difficult to interpret the economic meaning of the estimated parameter.

16. The Vuong test suggested that PPML is the preferred estimation technique here.

17. A high level of regulation is negatively correlated with market concentration and significantly so, but the correlation is far from perfect. The Spearman rank correlation is –0.37 and significant at a 1% level.

18. Product differentiation within countries and product categories are not considered in this paper.

19. See Hummels, Lugovskyy, and Skiba (2007) for details.

Additional information

Notes on contributors

Hildegunn Kyvik Nordås

The views expressed in this article should be attributed to the author and do not represent the position or opinions of the OECD Secretariat or OECD members.