Antitrust analysis for the internet upstream market: A border gateway protocol approach

Journal of Competition Law and Economics

Volumn 2, Issue 1, 2006, Pages 43-69

  D'Ignazio, Alessio ^a   Giovannetti, Emanuele ^a,b,c  

^a UNIVERSITY OF CAMBRIDGE   (United Kingdom)

^b UNIVERSITY OF CAPE TOWN   (South Africa)

^c SAPIENZA UNIVERSITY OF ROME   (Italy)

Author keywords

[No Author keywords available]

Indexed keywords

EID: 33645572062     PISSN: 17446414     EISSN: 17446422     Source Type: Journal    
DOI: 10.1093/joclec/nhl003     Document Type: Article

References (45)

1. However its impact on the dynamic and productive efficiency is controversial, because of the role of market power as an incentive to innovate.
2. For a review of this debate see Emanuele Giovannetti and Cristiano Andrea Ristuccia, "Estimating Market Power in the Internet backbone using the I.P. transit Band-X database," 29 Telecommunications Policy 269 (2005).
3. Many traditional markets are characterized by the fact that, while market shares and characteristics of the firms involved in an antitrust investigation can be obtained, their potential rivals' data is normally concealed and hard to reach.We show that, when dealing with the Internet, there are ways to overcome these difficulties that may not exist for traditional markets.
4. IXPs are independent organizations composed by Internet service providers, where they can route their traffic in a cost-effective and technically efficient way.
5. These data were obtained for research purposes from the London Internet exchange point under confidentiality and a non-disclosure agreement.
6. U.S. Department of Justice, "The Horizontal Merger Guideline," issued 2 April 1992 and revised 8 April 1997; originally issued as part of U.S. Department of Justice Merger Guidelines, 14 June 1984.
7. Martin Cave, "Interconnection and the Internet: Competition and Regulation Issues at Local Access and Backbone Levels," OECD, Working Party on Telecommunication and Information Services (1999).
8. For a more recent analysis see Nicholas Economides, "The Economics of the Internet Backbone," in S. Majumdar, I. Vogelsang and M. Cave (eds), Handbook of Telecommunications Economics, Volume 2. Amsterdam, North Holland, 2006.
9. On 11 January 2000, the European Commission received a notification that MCI-WorldCom would merge with Sprint by an exchange of shares. After an extensive investigation into the merger proposal, on 28 June 2000 the Commission adopted the decision that "The notified concentration consisting of the merger between MCI-WorldCom and Sprint is declared incompatible with the common market and the functioning of the EEA Agreement." See Official Journal Of the European Commission (2000), Regulation (EEC) N 4046/89, Merger Procedure, Bruxelles, European Commission, DGXIII.
10. Paolo Buccirossi, Laura Ferrari Bravo, and Paolo Siciliani, "Competition in the Internet Backbone Market," 28(2) World Competition 235 (2005).
11. The dominant position of WorldCom had been attained through a very active acquisition policy. In the Civil Action brought by Department of Justice of the United States against the acquisition of Intermedia Communications by WorldCom are described some of the more than 60 acquisitions operated by this company: In 1995 WorldCom acquired the network service operations of Williams Telecommunications; in 1996, through the acquisition of MFS Communications Company, WorldCom obtained the control of UUNET, the world's largest Internet backbone provider. In 1998 WorldCom acquired Compuserve a leading Internet provider and ANS, AOL's primary Internet backbone network. Other acquired backbones were GridNet, Unicom-Pipex, InNet, NL Net and Metrix Interlink. As a result of the leadership position reached in these years, the WorldCom acquisition of MCI in September 1998 has been accompanied by the imposition, by the U.S. Department of Justice and the EU Commission, that MCI divest its Internet assets to Cable & Wireless. See U.S. Department of Justice, Civil Action no. 00-CV-2789 against WorldCom and Intermedia, www.usdoj.gov/atr/cases/f7000/7042.pdf, 2000.
12. In detail, the FTC uses the Hirschman-Herfindahl Index (HHI) of market concentration. The HHI is calculated by summing the squares of the individual market shares of all the participants and multiplying by ten thousand. The guidelines focus on two figures: Pre-merger HHI concentration index level, and post-merger HHI concentration increments. When the post-merger HHI is below 1000, the guidelines consider the market as being non-concentrated, and no further analysis is required. When the post-merger HHI ranges between 1000 and 1800 the guidelines consider these markets as moderately concentrated. In this case further scrutiny is required only if the merger would increase the HHI by more than 100 points. Finally, when post-merger HHI is above 1800, the markets are considered highly concentrated. In this case an increase in the HHI of more than 50 points will start raising concern, while an increase of more than 100 points is considered as potentially dangerous due to the increase in market power it entails.
13. See U.S. Department of Justice Civil Action no. 01-CV-04080 against WorldCom and Sprint, www.usdoj.gov/atr/cases/f5000/5051.pdf, 2000.
14. The final test was published on 4 of February 2002 (see Official Journal 2002 (http://europa.eu.int/information_society/topics/telecoms/ regulatory/new_rf/documents/03672en1.pdf).
15. G.J. Werden and L.M. Froeb, "The Effects of Mergers in Differentiated Products Industries: Logit Demand and Merger" (1994)
16. David T. Scheffman and Pablo T. Spiller, "Geographic Market Definition under the U.S. Department of Justice Merger Guidelines," 30(1) Journal of Law & Economics, 123 (1987).
17. Clearly substitution of a potential competitor will take longer than that of an existing competitors.
18. In economics, it is usually assumed that the exact combination of inputs, in our case peering and transit agreements, is derived by a cost-minimizing choice conditional on a given level of output. This depends, of course, upon the functional form of the production function describing the technology, which uses these inputs (transit and peering agreements) to produce the ISP output: Universal connectivity.
19. This is the definition that the Court of Justice case-law ascribes to the concept of dominant position in Article 82 of the Treaty of Rome.
20. According to established case-law, very large market shares - in excess of 50 percent - are in themselves evidence of the existence of a dominant position.
21. In particular, in the case of bulk products, preference is given to volume, whereas in the case of differentiated products (i.e. branded products), sales in value and their associated market share will often be considered to reflect better the relative position and strength of each provider.
22. Kimberly Claffy, Tracie, E. Monk, and Daniel McRobb, "Internet Tomography," Nature, 7 January 1999.
23. Beatrice Cossa, "Economic and policy analysis of mergers among Internet backbone providers" (MSc Thesis, MIT, http://itel.mit.edu/itel/publications.html, 2000).
24. The computer science literature also considers a fourth type of relation, among siblings, where both ASs belong to the same ISP. The ISP partitions its network to ease the technical management by hiding internal information for each AS.
25. Lixin Gao, "On Inferring Autonomous System Relationships in the Internet," 9(6) IEEE/ACM Trans. Netw. 681 (2001).
26. Xenofontas Dimitropoulos, Dmitri Krioukov, Bradley Huffaker, Kimberly C. Claffy and George Riley, "Inferring AS Relationships: Dead End or Lively Beginning?" www.caida.org/outreach/papers/2005/ asrelationships/as_relationships.pdf, access date: July 2005.
27. For technical reasons, large ISPs operate using multiple ASs (e.g. UUNET uses 13), and it would be a misrepresentation to rank each of these ASs separately, since they belong to the same company. The CAIDA ranking aggregates ASs with company names that are similar in the ARIN database (two names are considered similar if they are identical except for the last several characters). We rely on this grouping to consider rankings of ISPs, instead of rankings of ASs only, since it improves the representation of the ranking, despite some shortcomings.
28. IXPs play a crucial role in Internet traffic routing. As an example, it is sufficient to think that more than 90 percent of the Internet UK traffic is routed through the LINX-IX in London.
29. AIX, AMS-IX, BCIX, BIX, BNIX, CATNIX, CIXP, DE-CIX, Equinix*+, ESPANIX, FICIX, GIGAPIX, GN-IX, INEX, JPNAP*+, LINX, LIPEX, LIX, LoNAP,MaNAP,MIX, MSK-IX, NaMeX, NDIX, Netnod*, NIX, NIX.CZ, NOTA+, PARIX, RoNIX, SIX, TIX, TOPIX, VIX, and XchangePoint*, where * indicates IXPs with multiple unconnected locations and + indicates IXPs located outside Europe.
30. The customer cone can be defined using three different precision levels: the AS cone, the AS prefix customer cone, and the AS/24 prefix customer cone. The AS cone indicates the size of the customer cone in terms of number of ASs; this is a rough measure, since individual AS sizes can be very different. Since each AS advertises a different number of prefixes, and the smallest bit of a prefix is the /24, the other two measures provide greater accuracy in assessing the size of the customer cone. For full details see the CAIDA web page at www.caida.org/analysis/topology/ rank_as/index.xml, access date: May 2005.
31. Beatrice Huber, Simon Leinen, Regina O'Dell and Roger Wattenhofer "Inferring AS Relationships Beyond Counting Edges," 446 D-INFK Tech. Report, ftp://ftp.inf.ethz.ch/pub/publications/tech-reports/4xx/446.pdf, access date: June 2005.
32. For the full details, see the University of Oregon routeviews project webpage at www.routeviews.org/, access date: May 2005.
33. A. Shimbel, "Structural parameters of communication networks," 15 Bulletin of Mathematical Biophysics, 501 (1953).
34. Ulrik Brandes, "A Faster Algorithm for Betweenness Centrality," 25 Journal of Mathematical Sociology 163 (2001).
35. Ideally, market shares within the IXP are derived by looking at the traffic flows. Since these data are confidential we use the metrics introduced in the previous sections.
36. The effective HHI is 0.021, while the indexes calculated with the CAIDA customer cone and betweenness are 0.024 and 0.05 respectively. Moreover, the market share squares calculated with the CAIDA customer cone showed the highest correlation (0.74) with the squares of effectives market shares (the correlation is instead 0.33 for the betweenness-based proxies). This is a positive result, indicating that it is sensible to calculate the HHI index using the CAIDA rankbased proxy market shares, and also that these proxies are probably a good approximation of the effective market shares.
37. This class actually contains about 18 known ISPs, but in the CAIDA ranking several of these ISPs are ranked significantly below number 12 (e.g. British Telecom = rank 36 and AOL Transit Data Network = rank 48).
38. The #/24-number in the customer cone drops by over one order of a magnitude within this class.
39. The #/24-number in the customer cone drops by over one order of a magnitude within this class.
40. The #/24-number in the customer cone drops by a further 63 percent within this class.
41. For this application we have subdivided the original Core class introduced above into two classes: Inner Core (composed by ASs ranking between 13 and 49) and Outer Core (composed by ASs ranking between 50 and 250), leaving the other classes unchanged.
42. ASs with insufficient information in terms of rank or interconnection agreements were deleted.
43. U.S. Department of Justice, supra n 6.
44. See U.S. Department of Justice, supra n 6.
45. There is a clear tendency showing an improvement in the reliability of both the Internet routing databases and the inference techniques, reducing the possibility of errors associated with the proposed method. See, for example, the source of BGP data at University of Oregon Route Views Project (www.routeviews.org/).