1. Do All Markets Ultimately Tip? Experimental Evidence Tanjim Hossain Hong Kong University of Science & Technology And John Morgan University of California, Berkeley

2. Two-sided Markets Two-sided market platforms that bring two populations together are very common — auction markets, stock exchanges, credit cards, dating sites, video-gaming consoles, search engines Participation of both populations in the same platform is necessary for trade Opposite-typed players beneficial (market size effect) Competition for business (market impact effect) Overall positive network externalities (scale effect)

3. Competition in Two-sided Markets A market may tip to one platform or many platforms may coexist in equilibrium US online auction market has tipped to eBay while Yahoo is the dominant player in Asia VCR market tipped to VHS HD-DVD left the market while Blu-ray stayed on Two major players in credit cards—Visa/Master and American Express; Three major players in video-gaming consoles market A market may end up at a less efficient equilibrium QWERTY vs. Dvorak keyboards, VHS vs. Betamax, MS- Dos vs. Apple OS, Windows vs. Linux

4. When Do Markets Tip? Price competition between platforms that charge different prices to the two types lead to tipping; however coexistence can be sustained using specific payment structures Caillaud and Julien (2001 and 2003) Two platforms of different sizes can sustain if platforms do not compete in prices when the market-impact effect is large enough Ellison and Fudenberg (2003), Ellison, Fudenberg and Mobi ü s (2004)

5. Objective Designing laboratory experiments to test platform choice in two sided markets Creating a setup where a static analysis is possible while studying dynamic interactions Test the impact of the market-impact effect (Two platforms can coexist if the market-impact effect is large enough) Track evolution of platform competition in a dynamic setting Creating a realistic framework of two-sided markets using heterogeneous platforms Connecting experimental results to the real world

6. Experimental Design Sixteen subjects form a session of 40 (60) periods with four separate 4-player markets in each period Four subjects in a market—two males and two females (denoted by “square” and “triangle”) Two platforms A and B (“%” and “#”) charge the same entry fee to both types and their gross payoff functions are presented as payoff matrices Subjects simultaneously decide to join one of the platforms (single-home) in each period They learn last period’s outcome in each period

7. Experimental Design Before period 10k +1, a subject is randomly assigned a type and to one of four 4-player markets Subjects participate in the same 4-player market in periods 1 to 10 with given gross payoff matrices In periods 11 to 20, they participate in new markets with new gross payoff matrices for the platforms Gross payoffs in periods 21 to 30 are the same as those in periods 1 to 10 Gross payoffs in periods 31 to 40 are the same as those in periods 11 to 20

8. Experimental Design Subjects play each of two games N and T twice Two treatments for each setting — NTNT in odd numbered sessions and TNTN in even sessions Used zTree to program the experiment In total, 352 HKUST students participated in 5 settings (10 treatments) in 20 sessions

9. Screenshot of a Payoff Matrix

10. Homogenous Platforms: Payoff Matrix of Game N Number of players of the player's own type (including herself) in the platform she joined 12 Number of players of the opposite type in the platform the player joined 055 196 21211 The subscription fees are, p A = 4 and p B = 2

11. Payoff Matrix of Homogeneous N Number of players of the player's own type (including herself) in the platform she joined 1 2 Number of players of the opposite type in the platform the player joined 055 19 6 21211 The subscription fees are, p A = 4 and p B = 2

12. Payoff Matrix of Homogeneous T Number of players of the player's own type (including herself) in the platform she joined 12 Number of players of the opposite type in the platform the player joined 055 198 21211 The subscription fees are, p A = 4 and p B = 2

13. Payoff Matrix of Homogeneous T Number of players of the player's own type (including herself) in the platform she joined 12 Number of players of the opposite type in the platform the player joined 055 198 21211 The subscription fees are, p A = 4 and p B = 2

14. Equilibria of the N and T Games Game N has a non-tipped equilibrium where a pair of square and triangle players go to each market in addition to the two tipped equilibria Game T only has the two tipped equilibria The market tipping to the cheaper platform B is the unique equilibrium predicted by both Pareto- dominance and risk-dominance criteria in both games

15. Results from the Homogeneous Experiments

17. Results from the Homogeneous- Large Experiments

18. Summary of Homogeneous Platform Results Sessions 7 and 8: 8-player games—4 square and 4 triangle players—32 subjects in a session Markets converged quickly to the cheaper platform within the first set of 10 (15) periods and then stayed that way This equilibrium Pareto dominates other equilibria and is also the risk-dominant equilibrium No QWERTY kind of outcome Choosing the cheap platform explains the result

19. Vertically Differentiated Platforms “Choose the cheaper platform” explains the results It is more realistic that platforms vary in matching efficiency as well as entry fees—Google vs. yahoo or Alta Vista, eHarmony relationship questionnaire Platforms have different gross payoff matrices Platforms are different in two dimensions Platform A charges a higher entry fee but offers a higher gross expected payoff for most allocations Risk and Pareto dominance may not be equivalent

20. Summary of Differentiated Platforms Treatments Risk-Dominance Pareto Dominance DifferentiatedTip to Platform B Differentiated ExpensiveTip to Platform A Differentiated RDTip to Platform BTip to Platform A

21. Differentiated-Expensive Setting: Payoff Matrices in Games N and T The subscription fees are, p A = 3 and p B = 2 Number of players of the player's own type (including herself) 12 Number of players of the opposite type 0(4, 4) 1(11, 8)[(8, 6)] {(10, 6)} 2(13, 11)(12, 10)

22. Differentiated-Expensive: Results

23. Differentiated – Expensive Setting: Results

24. Differentiated RD: Payoff Matrices in Games N and T The subscription fees are, p A = 3 and p B = 2 Number of players of the player's own type (including herself) 12 Number of players of the opposite type 0(4, 4) 1(11, 8)[(8, 6)] {(10, 6)} 2(13, 22)(12, 10)

25. Differentiated RD: Results

26. Determinants of Individual Platform Choice OR CoefficientHomogeneous Differentiated Expensive Differentiated RD Last Choice1.38 (.45)5.14*** (1.82)8.62*** (.89) Initial Mkt Shr47.25 *** (34.54)337.65*** (348.29)12.52*** (3.18) Best Response2.25*** (.63)5.28*** (1.68)5.90*** (.60) Tip Session.74 (.19)1.71 (.60)1.22** (.12) Sub Period1.46*** (.09)1.16*** (.05)1.04*** (.01) Set1.67*** (.30)2.27*** (.68)1.47*** (.07) N37443776 Dependent Variable: Choosing the Pareto Dominant Platform

27. Conclusion First set of laboratory experiments on platform competition in two-sided markets The market-impact effect seems irrelevant Perfect coordination can easily be achieved with homogeneous platforms Differentiated platforms experiments provide a more realistic description of the world Subjects sometimes reach a less efficient tipped equilibrium with differentiated platforms, but reach the Pareto dominant outcome most often

28. Conclusion Do markets always tip? A non-tipped outcome, if reached, is unstable Apparent overall non-tipped outcome may result from slow convergence or segmented markets tipping to different platforms Geographic segmentation in dating and auction markets Theory models should allow differentiated platforms Possible sources of coexistence in the real world may be multi-homing and agent heterogeneity