Hit and Run or Sit and Wait? Contestability Revisited in a Price-Comparison Site-Mediated Market

Abstract

The price-comparison site, with its (near-)zero sunk costs of entry, would appear to approximate the “almost perfectly contestable market” envisaged by the contestability theorists where “hit-and-run” entry was conjectured to constrain sellers to zero-profit outcomes. We investigate hit and run using a unique unbalanced panel of 295 digital-camera markets mediated by NexTag.com. We find, however, in line with Farrell (1986a)’s prediction, a bifurcation of strategies with low reputation/smaller participants favouring a hit-and-run strategy involving lower entry prices and shorter forays into the market than their high reputation/larger rivals. Furthermore, the former entrants induce a much larger price response from low-reputation incumbents, reflecting the more intense rivalry for the price-sensitive consumers willing to eschew retailer reputations.

Key Words:

• Internet
• Contestability
• Entry.

JEL classifications:

• D4
• L1.

The authors would like to acknowledge the considerable assistance of Eleanor Morgan, Editor of the Journal, and two anonymous referees in revising this paper. Many helpful comments on the paper were also provided by Pascuale Schiraldi and other participants at the ZEW Digital Economy Conference, Mannheim, Germany, October 2011, and the IOS Conference in Boston, Massachusetts, May 2013, especially Felipe Zurita.

Notes

1. The robustness issue is debated by inter alia Weitzman (1983), Shepherd (1984), and Farrell (1986b), and the empirical predictions are examined in Morrison and Winston (1987), Hurdle et al. (1989), and references therein.

2. The term employed by both Baumol, Panzar, and Willig (1982) and Farrell (1986b).

3. Some PCS platforms charge a monthly listing fee. Clearly, on those sites – unlike NexTag.com – there is an obvious sunk cost of market entry.

4. Since the shopbot does not publish its ranking algorithm, the weight given to factors other than the bid cannot be determined

5. For example, Baye et al. (2009) report a ceteris paribus decline in clicks of 17% per ranking position, with a 40% discontinuity between positions one and two. Their results are sensitive to the number of sellers with further discontinuities at the ends of pages.

6. It has been suggested that attracting interest in this way inflates clicks for the top-ranked sellers in the listing but correspondingly lowers their conversion rates: see http://www.mobile-o.com/docs/Top-Vertical-Search-Sites.html.

7. Baye and Morgan (2001) pose an insider–outsider model in which entry reduces the proportion of uninformed buyers, thus encouraging sellers to pursue the more price sensitive consumers and so generating a predicted negative relationship between price and n. This is achieved by introducing entry costs which, in reality, appear trivial in many e-markets.

8. This type of low-quality hit-and-run entry in e-markets may be favoured by the ease of exit and subsequent name change which reduces the incentive to build reputations (Ellison and Ellison 2009).

9. Empirical evidence across e-commerce – reviewed in OFT (2007) – suggests both a much more frequent and smaller price adjustments than occurs in traditional markets.

10. Specific software to generate and transfer product feed data via FTP is available for as little as $25.

11. Some shopbots, such as Shopper.com, obviate this requirement by providing small sellers with storefront services which provide them with a selling site in exchange for commission.

12. Although collection was automated, screenshot data do require some cleaning before use, and time costs prohibited more frequent visits.

13. The upc originally appeared on NexTag.com’s screen display but is currently not available.

14. We used a cut-off of 100 leads, since we were interested in studying behaviour in active markets.

15. We chose a tax-free zip code in New Hampshire.

16. We also repeated all of the analysis using final prices including shipping costs. This did not materially affect our results.

17. This is the number of exits for which we have a record of their entry.

18. This is calculated by dividing the average number of entrants/exits by day by the average number of seller–product observations and then multiplying by seven to arrive at a weekly figure.

19. Our duration analysis necessarily takes product, seller, and market characteristics as exogenous or at least predetermined. No doubt at a finer – but unobservable – level of disaggregation, potential entrants and incumbents are exploring their conjectures about one another’s behaviour. Thus, our duration analysis of the determinants of exit is strictly reduced form estimates. More complex interactions are explored on the pricing side below.

20. Some entrants survive beyond the end of our sample period.

21. The probability of discounted entry is determined by seller-specific characteristics, which themselves determine the duration of a seller’s visit, resulting in a simultaneity problem. In an earlier paper (see Haynes and Thompson 2013), we found that seller-specific characteristics dominate market factors in the entry decision. The coefficients in the probit regression are all individually significant at the 1% level.

22. Vella and Verbeek (1999) show that this type of IV estimator generates results similar to Heckman’s (1978, 1979) endogeneity bias corrected OLS estimator. All of the seller reputation variables used as regressors in the discounted entry probit estimation were significant at the 1% level or above.

23. If we look at sellers who enter at or below the previous period’s minimum price, then that hazard is even higher. The proportion of entrants entering at or below the previous period minimum price is approximately 27% of the total number of entries taking place. The results from this estimation are included in Table A1 in Appendix A.

24. A similar result was obtained when membership of the top-three sellers in the default ranking was used instead.

25. A chi-square test on the difference between the high- and low-reputation coefficients is highly significant.

26. These decisions have been modelled as a joint decision process in earlier work by Haynes and Thompson (2013).

27. Frank and Salkever (1997) report that entry by generic pharmaceuticals stimulates price cuts among other generic sellers but price increases among branded sellers, while Simon (2005) finds entry into a magazine segment triggers price cuts among the more recently founded titles. McCann and Vroom (2010) report broadly similar findings for hotels.

28. Since entry is potentially endogenous, we re-estimate equation (3) after instrumenting entry. As in the duration model, we use predicted values from a probit regression of entry on seller-reputation variables as an instrument. Since the instrumented variables do not alter the pattern of our results, for space reasons, we have not reported the instrumented results for Tables 4–9. These results are available from the authors.

29. These results are available from the authors.