the competition complexity of auctions
play

The Competition Complexity of Auctions: Bulow-Klemperer Results for - PowerPoint PPT Presentation

The Competition Complexity of Auctions: Bulow-Klemperer Results for Multidimensional Bidders Oxford, Spring 2017 Alon Eden, Michal Feldman, Ophir Friedler @ Tel-Aviv University Inbal Talgam-Cohen, Marie Curie Postdoc @ Hebrew University Matt


  1. Multidimensional B-K theorems Competition complexity: Fix an environment with ๐‘œ i.i.d. bidders. What is ๐’š such that the revenue of VCG with ๐’ + ๐’š bidders is โ‰ฅ OPT with ๐’ bidders. Thm. [EFFTW] Let ๐‘ซ be the competition complexity of ๐’ additive bidders over ๐‘› items. The competition complexity of ๐’ additive bidders with identical downward closed constraints over ๐‘› items is โ‰ค ๐‘ซ + ๐’ โˆ’ ๐Ÿ . Competition Complexity of Auctions 31 Eden et al. EC'17 Inbal Talgam-Cohen

  2. Multidimensional B-K theorems Competition complexity: Fix an environment with ๐‘œ i.i.d. bidders. What is ๐’š such that the revenue of VCG with ๐’ + ๐’š bidders is โ‰ฅ OPT with ๐’ bidders. Thm. [EFFTW] Let ๐‘ซ be the competition complexity of ๐’ additive bidders over ๐‘› items. The competition complexity of ๐’ additive bidders with randomly drawn downward closed constraints over ๐‘› items is โ‰ค ๐‘ซ + ๐Ÿ‘(๐’ โˆ’ ๐Ÿ) . Competition Complexity of Auctions 32 Eden et al. EC'17 Inbal Talgam-Cohen

  3. Additive with constraints โ€ข Constraints = set system over the items โ€“ Specifies which item sets are feasible โ€ข Bidder โ€™ s value for an item set = her value for best feasible subset โ€ข If all sets are feasible, bidder is additive Competition Complexity of Auctions 33 Eden et al. EC'17 Inbal Talgam-Cohen

  4. Example of constraints โ€ข No constraints Total value = $21 $6 $5 $10 Competition Complexity of Auctions 34 Eden et al. EC'17 Inbal Talgam-Cohen

  5. Example of constraints โ€ข Example of โ€œ matroid โ€ constraints: Only sets of size ๐‘™ = 1 are feasible Total value = $21 $10 Substitutes $6 $5 $10 Competition Complexity of Auctions 35 Eden et al. EC'17 Inbal Talgam-Cohen

  6. Example of constraints โ€ข Example of โ€œ downward closed โ€ constraints: Sets of size 1 and { } are feasible Total value = $10 $16 Substitutes $6 $5 Complements $10 Competition Complexity of Auctions 36 Eden et al. EC'17 Inbal Talgam-Cohen

  7. Complements in what sense? โ€ข No complements = gross substitutes: โ€“ ิฆ ๐‘ž โ‰ค ิฆ ๐‘Ÿ item prices โ€“ ๐‘‡ in demand( ิฆ ๐‘ž) if maximizes utility ๐‘ค ๐‘— ๐‘‡ โˆ’ ๐‘ž(๐‘‡) โ€“ โˆ€๐‘‡ in demand( ิฆ ๐‘ž) , there is ๐‘ˆ in demand(ิฆ ๐‘Ÿ) with every item in ๐‘‡ whose price didn โ€™ t increase ๐‘ป $ $ $ $ ๐‘ผ Competition Complexity of Auctions 37 Eden et al. EC'17 Inbal Talgam-Cohen

  8. Complements in what sense? โ€ข No complements = gross substitutes: โ€“ ิฆ ๐‘ž โ‰ค ิฆ ๐‘Ÿ item prices โ€“ ๐‘‡ in demand( ิฆ ๐‘ž) if maximizes utility ๐‘ค ๐‘— ๐‘‡ โˆ’ ๐‘ž(๐‘‡) โ€“ โˆ€๐‘‡ in demand( ิฆ ๐‘ž) , there is ๐‘ˆ in demand(ิฆ ๐‘Ÿ) with every item in ๐‘‡ whose price didn โ€™ t increase $ ๐‘ป 10 ๐‘ž ิฆ = (5, ๐œ—, ๐œ—) 5 6 ๐‘ผ Competition Complexity of Auctions 38 Eden et al. EC'17 Inbal Talgam-Cohen

  9. Competition complexity โ€“ summary Upper bound Valuation ๐‘œ + 2 ๐‘› โˆ’ 1 Additive Additive s.t. identical downward ๐‘œ + 3 ๐‘› โˆ’ 1 closed constraints Additive s.t. random downward closed ๐‘œ + 4 ๐‘› โˆ’ 1 constraints Additive s.t. identical matroid ๐‘œ + 2 ๐‘› โˆ’ 1 + ๐œ constraints ๐‘œ Lower bounds of ฮฉ ๐‘œ โ‹… log ๐‘› + 1 for additive bidders and ฮฉ ๐‘› for unit demand bidders are due to ongoing work by [Feldman-Friedler-Rubinstein] and to [Bulow-Klemperer โ€™ 96]

  10. Related work Multidimensional B-K theorems [Roughgarden T. Yan โ€˜ 12] : for unit demand bidders, revenue of VCG with ๐’ extra bidders โ‰ฅ revenue of the optimal deterministic DSIC mechanism. [Feldman Friedler Rubinstein โ€“ ongoing] : tradeoffs between enhanced competition and revenue. Prior-independent multidimensional mechanisms [Devanur Hartline Karlin Nguyen โ€˜ 11] : unit demand bidders. [Roughgarden T. Yan โ€˜ 12] : unit demand bidders. [Goldner Karlin โ€˜ 16] : additive bidders. Sample complexity [Morgenstern Roughgarden โ€˜ 16] : how many samples needed to approximate the optimal mechanism?

  11. MULTIDIMENSIONAL B-K THEOREM PROOF SKETCH Competition Complexity of Auctions 41 Eden et al. EC'17 Inbal Talgam-Cohen

  12. Bulow-Klemperer theorem Thm. Revenue of the 2 nd price auction with n+1 bidders โ‰ฅ Revenue of the optimal auction with n bidders. Competition Complexity of Auctions 42 Eden et al. EC'17 Inbal Talgam-Cohen

  13. Bulow-Klemperer theorem Thm. Revenue of the 2 nd price auction with n+1 bidders โ‰ฅ Revenue of the optimal auction with n bidders. Proof. (in 3 steps of [Kirkegaard โ€™ 06]) I. Upper-bound the optimal revenue. Competition Complexity of Auctions 43 Eden et al. EC'17 Inbal Talgam-Cohen

  14. Bulow-Klemperer theorem Thm. Revenue of the 2 nd price auction with n+1 bidders โ‰ฅ Revenue of the optimal auction with n bidders. Proof. (in 3 steps of [Kirkegaard โ€™ 06]) I. Upper-bound the optimal revenue. II. Find an auction ๐ต with more bidders and revenue โ‰ฅ the upper bound. Competition Complexity of Auctions 44 Eden et al. EC'17 Inbal Talgam-Cohen

  15. Bulow-Klemperer theorem Thm. Revenue of the 2 nd price auction with n+1 bidders โ‰ฅ Revenue of the optimal auction with n bidders. Proof. (in 3 steps of [Kirkegaard โ€™ 06]) I. Upper-bound the optimal revenue. II. Find an auction ๐ต with more bidders and revenue โ‰ฅ the upper bound. III. Show that the 2 nd price auction โ€œ beats โ€ ๐ต . Competition Complexity of Auctions 45 Eden et al. EC'17 Inbal Talgam-Cohen

  16. Proof: Myerson โ€™ s optimal mechanism Step I. Upper-bound Price โ‰ฅ ๐‘ž the optimal revenue. ๐‘ค 1 โˆผ ๐บ โ‰ฅ ๐‘ค 2 โˆผ ๐บ โ‰ฅ . . . . . . โ‰ฅ ๐‘ค ๐‘œ โˆผ ๐บ 46

  17. Proof: Step II. Find an auction ๐ต with more bidders and ๐‘ค 1 โˆผ ๐บ revenue โ‰ฅ the upper bound. ๐‘ค 2 โˆผ ๐บ . . . . . . ๐‘ค ๐‘œ โˆผ ๐บ ๐‘ค ๐‘œ+1 โˆผ ๐บ 47

  18. Proof: Step II. Find an auction ๐ต with more bidders and ๐‘ค 1 โˆผ ๐บ revenue โ‰ฅ the upper bound. ๐‘ค 2 โˆผ ๐บ Run Myerson โ€™ s mechanism on . . . . ๐’ bidders . . ๐‘ค ๐‘œ โˆผ ๐บ ๐‘ค ๐‘œ+1 โˆผ ๐บ 48

  19. Proof: Step II. Find an auction ๐ต with more bidders and ๐‘ค 1 โˆผ ๐บ revenue โ‰ฅ the upper bound. ๐‘ค 2 โˆผ ๐บ Run Myerson โ€™ s mechanism on . . . . ๐’ bidders . . ๐‘ค ๐‘œ โˆผ ๐บ If Myerson does not allocate, give item to the additional ๐‘ค ๐‘œ+1 โˆผ ๐บ bidder 49

  20. Proof: Step III. Show that the 2 nd price auction โ€œ beats โ€ ๐ต . ๐‘ค 1 โˆผ ๐บ Observation. 2 nd price ๐‘ค 2 โˆผ ๐บ auction is the optimal . . mechanism out of the . . . . mechanisms that always ๐‘ค ๐‘œ โˆผ ๐บ sell. ๐‘ค ๐‘œ+1 โˆผ ๐บ 50

  21. Competition complexity of a single additive bidder Plan: Follow the 3 steps of the B-K proof. I. Upper-bound the optimal revenue. II. Find an auction ๐ต with more bidders and revenue โ‰ฅ the upper bound. III. Show that VCG โ€œ beats โ€ ๐ต . Competition Complexity of Auctions 51 Eden et al. EC'17 Inbal Talgam-Cohen

  22. Competition complexity of a single additive bidder and i.i.d. items Plan: Follow the 3 steps of the B-K proof. I. Upper-bound the optimal revenue. II. Find an auction ๐ต with more bidders and revenue โ‰ฅ the upper bound. III. Show that VCG โ€œ beats โ€ ๐ต . Competition Complexity of Auctions 52 Eden et al. EC'17 Inbal Talgam-Cohen

  23. I. Upper-bound the optimal revenue โ€ข Single additive bidder and i.i.d. items ๐‘ค 1 โˆผ ๐บ ๐‘ค 2 โˆผ ๐บ . . . . . . ๐‘ค ๐‘› โˆผ ๐บ Competition Complexity of Auctions 53 Eden et al. EC'17 Inbal Talgam-Cohen

  24. I. Upper-bound the optimal revenue Use the duality framework from [Cai Devanur Weinberg โ€˜ 16] . OPT โ‰ค ๐œ’ + ๐‘ค ๐‘˜ โ‹… 1 โˆ€๐‘˜ โ€ฒ ๐‘ค ๐‘˜ >๐‘ค ๐‘˜โ€ฒ + ๐‘ค ๐‘˜ โ‹… 1 โˆƒ๐‘˜ โ€ฒ ๐‘ค ๐‘˜ <๐‘ค ๐‘˜โ€ฒ E ๐‘คโˆผ๐บ ๐‘› เท ๐‘˜ 1โˆ’๐บ ๐‘ค ๐œ’ ๐‘ค = ๐‘ค โˆ’ ๐‘”(๐‘ค) is the virtual valuation function. 54

  25. I. Upper-bound the optimal revenue Use the duality framework from [Cai Devanur Weinberg โ€˜ 16] . OPT โ‰ค ๐œ’ + ๐‘ค ๐‘˜ โ‹… 1 โˆ€๐‘˜ โ€ฒ ๐‘ค ๐‘˜ >๐‘ค ๐‘˜โ€ฒ + ๐‘ค ๐‘˜ โ‹… 1 โˆƒ๐‘˜ โ€ฒ ๐‘ค ๐‘˜ <๐‘ค ๐‘˜โ€ฒ E ๐‘คโˆผ๐บ ๐‘› เท ๐‘˜ Distribution appears in proof only! 1โˆ’๐บ ๐‘ค ๐œ’ ๐‘ค = ๐‘ค โˆ’ ๐‘”(๐‘ค) is the virtual valuation function. 55

  26. I. Upper-bound the optimal revenue Use the duality framework from [Cai Devanur Weinberg โ€˜ 16] . OPT โ‰ค ๐œ’ + ๐‘ค ๐‘˜ โ‹… 1 โˆ€๐‘˜ โ€ฒ ๐‘ค ๐‘˜ >๐‘ค ๐‘˜โ€ฒ + ๐‘ค ๐‘˜ โ‹… 1 โˆƒ๐‘˜ โ€ฒ ๐‘ค ๐‘˜ <๐‘ค ๐‘˜โ€ฒ E ๐‘คโˆผ๐บ ๐‘› เท ๐‘˜ Take item ๐‘˜ โ€™ s virtual value if it โ€™ s the most attractive item 1โˆ’๐บ ๐‘ค ๐œ’ ๐‘ค = ๐‘ค โˆ’ ๐‘”(๐‘ค) is the virtual valuation function. 56

  27. I. Upper-bound the optimal revenue Use the duality framework from [Cai Devanur Weinberg โ€˜ 16] . OPT โ‰ค ๐œ’ + ๐‘ค ๐‘˜ โ‹… 1 โˆ€๐‘˜ โ€ฒ ๐‘ค ๐‘˜ >๐‘ค ๐‘˜โ€ฒ + ๐‘ค ๐‘˜ โ‹… 1 โˆƒ๐‘˜ โ€ฒ ๐‘ค ๐‘˜ <๐‘ค ๐‘˜โ€ฒ E ๐‘คโˆผ๐บ ๐‘› เท ๐‘˜ Take item ๐‘˜ โ€™ s value if there โ€™ s a more attractive item 1โˆ’๐บ ๐‘ค ๐œ’ ๐‘ค = ๐‘ค โˆ’ ๐‘”(๐‘ค) is the virtual valuation function. 57

  28. II. Find an auction ๐ต with more bidders and revenue โ‰ฅ upper bound 58

  29. II. Find an auction ๐ต with ๐‘› bidders and revenue โ‰ฅ upper bound 59

  30. II. Find an auction ๐ต with ๐‘› bidders and revenue โ‰ฅ upper bound VCG for additive bidders โ‰ก 2 nd price auction for each item separately. Therefore, we devise a single parameter mechanism that covers item ๐’Œ โ€™ s contribution to the benchmark. E ๐‘คโˆผ๐บ ๐‘› ๐œ’ + ๐‘ค ๐‘˜ โ‹… 1 โˆ€๐‘˜ โ€ฒ ๐‘ค ๐‘˜ >๐‘ค ๐‘˜ โ€ฒ + ๐‘ค ๐‘˜ โ‹… 1 โˆƒ๐‘˜ โ€ฒ ๐‘ค ๐‘˜ <๐‘ค ๐‘˜ โ€ฒ 60

  31. II. Find an auction ๐ต ๐‘˜ with ๐‘› bidders and revenue โ‰ฅ upper bound for item ๐‘˜ E ๐‘คโˆผ๐บ ๐‘› ๐œ’ + ๐‘ค ๐‘˜ โ‹… 1 โˆ€๐‘˜ โ€ฒ ๐‘ค ๐‘˜ >๐‘ค ๐‘˜โ€ฒ + ๐‘ค ๐‘˜ โ‹… 1 โˆƒ๐‘˜ โ€ฒ ๐‘ค ๐‘˜ <๐‘ค ๐‘˜โ€ฒ Run 2 nd price auction Item ๐‘˜ ๐‘ค ๐‘˜ โˆผ ๐บ with โ€œ lazy โ€ reserve price = ๐œ’ โˆ’1 0 for agent ๐‘˜ ๐‘ค 1 โˆผ ๐บ 0 for agents ๐‘˜ โ€ฒ โ‰  ๐‘˜ . . . . . . ๐‘ค ๐‘› โˆผ ๐บ Competition Complexity of Auctions 61 Eden et al. EC'17 Inbal Talgam-Cohen

  32. II. Find an auction ๐ต ๐‘˜ with ๐‘› bidders and revenue โ‰ฅ upper bound for item ๐‘˜ E ๐‘คโˆผ๐บ ๐‘› ๐œ’ + ๐‘ค ๐‘˜ โ‹… 1 โˆ€๐‘˜ โ€ฒ ๐‘ค ๐‘˜ >๐‘ค ๐‘˜โ€ฒ + ๐‘ค ๐‘˜ โ‹… 1 โˆƒ๐‘˜ โ€ฒ ๐‘ค ๐‘˜ <๐‘ค ๐‘˜โ€ฒ Case I: ๐‘ค ๐‘˜ > ๐‘ค ๐‘˜ โ€ฒ for all ๐‘˜ โ€ฒ : Item ๐‘˜ ๐‘ค ๐‘˜ โˆผ ๐บ ๐‘˜ wins if his virtual value is non-negative. ๐‘ค 1 โˆผ ๐บ Expected revenue = . . . . . . Expected virtual value ๐‘ค ๐‘› โˆผ ๐บ [Myerson โ€™ 81] Competition Complexity of Auctions 62 Eden et al. EC'17 Inbal Talgam-Cohen

  33. II. Find an auction ๐ต ๐‘˜ with ๐‘› bidders and revenue โ‰ฅ upper bound for item ๐‘˜ E ๐‘คโˆผ๐บ ๐‘› ๐œ’ + ๐‘ค ๐‘˜ โ‹… 1 โˆ€๐‘˜ โ€ฒ ๐‘ค ๐‘˜ >๐‘ค ๐‘˜โ€ฒ + ๐‘ค ๐‘˜ โ‹… 1 โˆƒ๐‘˜ โ€ฒ ๐‘ค ๐‘˜ <๐‘ค ๐‘˜โ€ฒ Case II : ๐‘ค ๐‘˜ < ๐‘ค ๐‘˜ โ€ฒ for some ๐‘˜ โ€ฒ : Item ๐‘˜ ๐‘ค ๐‘˜ โˆผ ๐บ The second price is at least ๐‘ค 1 โˆผ ๐บ the value of agent ๐‘˜ . . . . . . . ๐‘ค ๐‘› โˆผ ๐บ Competition Complexity of Auctions 63 Eden et al. EC'17 Inbal Talgam-Cohen

  34. III. Show that VCG โ€œ beats โ€ ๐ต Competition Complexity of Auctions 64 Eden et al. EC'17 Inbal Talgam-Cohen

  35. III. Show that 2 nd price โ€œ beats โ€ ๐ต(๐‘˜) Competition Complexity of Auctions 65 Eden et al. EC'17 Inbal Talgam-Cohen

  36. III. Show that 2 nd price โ€œ beats โ€ ๐ต(๐‘˜) โ‰ค 2 nd price with ๐‘ฉ(๐’Œ) with ๐’ bidders โ‰ค Myerson with ๐’ bidders ๐’ + ๐Ÿ bidders Competition Complexity of Auctions 66 Eden et al. EC'17 Inbal Talgam-Cohen

  37. III. Show that 2 nd price โ€œ beats โ€ ๐ต(๐‘˜) โ‰ค 2 nd price with ๐‘ฉ(๐’Œ) with ๐’ bidders โ‰ค Myerson with ๐’ bidders ๐’ + ๐Ÿ bidders ๏ƒ  The competition complexity of a single additive bidder and ๐‘› i.i.d. items is โ‰ค ๐‘› . FF Competition Complexity of Auctions 67 Eden et al. EC'17 Inbal Talgam-Cohen

  38. Going beyond i.i.d items โ€ข Single additive bidder and i.i.d. items ๐‘ค 1 โˆผ ๐บ 1 ๐‘ค 2 โˆผ ๐บ 2 . . . . . . ๐‘ค ๐‘› โˆผ ๐บ ๐‘› Competition Complexity of Auctions 68 Eden et al. EC'17 Inbal Talgam-Cohen

  39. Going beyond i.i.d items + ๐‘ค ๐‘˜ โ‹… 1 โˆ€๐‘˜ โ€ฒ ๐‘ค ๐‘˜ >๐‘ค ๐‘˜โ€ฒ + ๐‘ค ๐‘˜ โ‹… 1 โˆƒ๐‘˜ โ€ฒ ๐‘ค ๐‘˜ < ๐‘ค ๐‘˜โ€ฒ E ๐‘ค 1 โˆผ๐บ ๐œ’ ๐‘˜ 1 ๐‘ค 2 โˆผ๐บ 2 โ€ฆ ๐‘ค ๐‘› โˆผ๐บ ๐‘› Item ๐‘˜ ๐‘ค ๐‘˜ โˆผ ๐บ ๐‘˜ Run 2 nd price auction ๐‘ค 1 โˆผ ๐บ with โ€œ lazy โ€ reserve price = ๐‘˜ ๐œ’ โˆ’1 0 for agent ๐‘˜ . . . . . . 0 for agents ๐‘˜ โ€ฒ โ‰  ๐‘˜ ๐‘ค ๐‘› โˆผ ๐บ ๐‘˜ Competition Complexity of Auctions 69 Eden et al. EC'17 Inbal Talgam-Cohen

  40. Going beyond i.i.d items + ๐‘ค ๐‘˜ โ‹… 1 โˆ€๐‘˜ โ€ฒ ๐‘ค ๐‘˜ >๐‘ค ๐‘˜โ€ฒ + ๐‘ค ๐‘˜ โ‹… 1 โˆƒ๐‘˜ โ€ฒ ๐‘ค ๐‘˜ < ๐‘ค ๐‘˜โ€ฒ E ๐‘ค 1 โˆผ๐บ ๐œ’ ๐‘˜ 1 ๐‘ค 2 โˆผ๐บ 2 โ€ฆ ๐‘ค ๐‘› โˆผ๐บ ๐‘› Run 2 nd price auction Item ๐‘˜ ๐‘ค ๐‘˜ โˆผ ๐บ ๐‘˜ with โ€œ lazy โ€ reserve price = ๐œ’ โˆ’1 0 for agent ๐‘˜ ๐‘ค 1 โˆผ ๐บ ๐‘˜ 0 for agents ๐‘˜ โ€ฒ โ‰  ๐‘˜ . . . . Cannot couple the event . . โ€œ bidder ๐‘˜ wins โ€ and ๐‘ค ๐‘› โˆผ ๐บ ๐‘˜ โ€œ item ๐‘˜ has the highest value โ€ Competition Complexity of Auctions 70 Eden et al. EC'17 Inbal Talgam-Cohen

  41. Use a different benchmark + ๐‘ค ๐‘˜ โ‹… 1 โˆ€๐‘˜ โ€ฒ ๐บ ๐‘˜ (๐‘ค ๐‘˜ )>๐บ ๐‘˜โ€ฒ (๐‘ค ๐‘˜โ€ฒ ) + ๐‘ค ๐‘˜ โ‹… 1 โˆƒ๐‘˜ โ€ฒ ๐บ ๐‘˜ (๐‘ค ๐‘˜ )<๐บ ๐‘˜โ€ฒ (๐‘ค ๐‘˜โ€ฒ ) E ๐‘ค 1 โˆผ๐บ ๐œ’ ๐‘˜ 1 ๐‘ค 2 โˆผ๐บ 2 โ€ฆ ๐‘ค ๐‘› โˆผ๐บ ๐‘› Item ๐‘˜ ๐‘ค ๐‘˜ โˆผ ๐บ ๐‘˜ ๐‘ค 1 โˆผ ๐บ ๐‘˜ . . . . . . ๐‘ค ๐‘› โˆผ ๐บ ๐‘˜ Competition Complexity of Auctions 71 Eden et al. EC'17 Inbal Talgam-Cohen

  42. Use a different benchmark + ๐‘ค ๐‘˜ โ‹… 1 โˆ€๐‘˜ โ€ฒ ๐บ ๐‘˜ (๐‘ค ๐‘˜ )>๐บ ๐‘˜โ€ฒ (๐‘ค ๐‘˜โ€ฒ ) + ๐‘ค ๐‘˜ โ‹… 1 โˆƒ๐‘˜ โ€ฒ ๐บ ๐‘˜ (๐‘ค ๐‘˜ )<๐บ ๐‘˜โ€ฒ (๐‘ค ๐‘˜โ€ฒ ) E ๐‘ค 1 โˆผ๐บ ๐œ’ ๐‘˜ 1 ๐‘ค 2 โˆผ๐บ 2 โ€ฆ ๐‘ค ๐‘› โˆผ๐บ ๐‘› Item ๐‘˜ ๐‘ค ๐‘˜ โˆผ ๐บ ๐‘˜ ๐‘ค 1 โˆผ ๐บ ๐‘˜ . . . . . . ๐‘ค ๐‘› โˆผ ๐บ ๐‘˜ ๏ƒ  The competition complexity of a single additive bidder and ๐‘› items is โ‰ค ๐‘› .

  43. Going beyond a single bidder โ€ข Step I: โ€“ Benchmark more involved โ€ข Step II: โ€“ Devise a more complex single parameter auction A(j) (involves a max) โ€“ Proving A(j) is greater than item j โ€™ s contribution to the benchmark is more involved and requires subtle coupling and probabilistic claims BB Competition Complexity of Auctions 73 Eden et al. EC'17 Inbal Talgam-Cohen

  44. EXTENSION TO CONSTRAINTS Competition Complexity of Auctions 74 Eden et al. EC'17 Inbal Talgam-Cohen

  45. Recall โ€ข Example of โ€œ downward closed โ€ constraints: Sets of size 1 and { } are feasible Total value = $16 Substitutes $6 $5 Complements $10 Competition Complexity of Auctions 75 Eden et al. EC'17 Inbal Talgam-Cohen

  46. Extension to downward closed constraints Add โ‰ค VCG ๐‘œ+๐ท Add OPT ๐‘œ Competition complexity โ‰ค ๐ท Competition Complexity of Auctions 76 Eden et al. EC'17 Inbal Talgam-Cohen

  47. Extension to downward closed constraints DC โ‰ค Add โ‰ค VCG ๐‘œ+๐ท Add OPT ๐‘œ OPT ๐‘œ Competition Larger complexity outcome โ‰ค ๐ท space Competition Complexity of Auctions 77 Eden et al. EC'17 Inbal Talgam-Cohen

  48. Extension to downward closed constraints DC โ‰ค Add โ‰ค VCG ๐‘œ+๐ท DC Add OPT ๐‘œ โ‰ค VCG ๐‘œ+๐ท+๐‘›โˆ’1 OPT ๐‘œ Competition Larger complexity outcome โ‰ค ๐ท space Competition Complexity of Auctions 78 Eden et al. EC'17 Inbal Talgam-Cohen

  49. Extension to downward closed constraints DC โ‰ค Add โ‰ค VCG ๐‘œ+๐ท DC Add OPT ๐‘œ โ‰ค VCG ๐‘œ+๐ท+๐‘›โˆ’1 OPT ๐‘œ Competition Larger complexity outcome โ‰ค ๐ท space The competition complexity of ๐‘œ additive bidders with identical downward closed constraints over ๐‘› items is โ‰ค ๐ท + ๐‘› โˆ’ 1 .

  50. Extension to downward closed constraints Main technical challenge DC โ‰ค Add โ‰ค VCG ๐‘œ+๐ท DC Add OPT ๐‘œ โ‰ค VCG ๐‘œ+๐ท+๐‘›โˆ’1 OPT ๐‘œ Competition Larger complexity outcome โ‰ค ๐ท space The competition complexity of ๐‘œ additive bidders with identical downward closed constraints over ๐‘› items is โ‰ค ๐ท + ๐‘› โˆ’ 1 .

  51. Claim. VCG revenue from selling ๐’ items to ๐’€ = ๐’ + ๐‘ซ additive bidders whose values are i.i.d. draws from ๐บ โ‰ค VCG revenue from selling them to ๐’€ + ๐’ โˆ’ ๐Ÿ bidders with i.i.d. values drawn from ๐บ , whose valuations are additive s.t. identical downward-closed constraints. Competition Complexity of Auctions 81 Eden et al. EC'17 Inbal Talgam-Cohen

  52. Add โ‰ค VCG ๐‘Œ+๐‘›โˆ’1 DC VCG ๐‘Œ VCG for additive bidders โ‰ก 2 nd price auction for each item separately. Add = Therefore, the revenue from item ๐’Œ in VCG ๐‘Œ 2 nd highest value out of ๐’€ i.i.d. samples from ๐‘ฎ ๐’Œ . Competition Complexity of Auctions 82 Eden et al. EC'17 Inbal Talgam-Cohen

  53. Add โ‰ค VCG ๐‘Œ+๐‘›โˆ’1 DC VCG ๐‘Œ 83

  54. Add โ‰ค VCG ๐‘Œ+๐‘›โˆ’1 DC VCG ๐‘Œ 84

  55. Add โ‰ค VCG ๐‘Œ+๐‘›โˆ’1 DC VCG ๐‘Œ 5 โˆผ ๐บ 7 โˆผ ๐บ 2 โˆผ ๐บ 85

  56. Add โ‰ค VCG ๐‘Œ+๐‘›โˆ’1 DC VCG ๐‘Œ 5 โˆผ ๐บ 7 โˆผ ๐บ 2 โˆผ ๐บ 3 6 4 4 5 1 3 4 2 3 2 1 86

  57. Add โ‰ค VCG ๐‘Œ+๐‘›โˆ’1 DC VCG ๐‘Œ 5 โˆผ ๐บ 7 โˆผ ๐บ 2 โˆผ ๐บ 3 6 4 4 5 1 3 4 2 3 2 1 87

  58. Add โ‰ค VCG ๐‘Œ+๐‘›โˆ’1 DC VCG ๐‘Œ 5 โˆผ ๐บ 7 โˆผ ๐บ 2 โˆผ ๐บ 3 6 4 Claim. Revenue for item ๐’Œ in DC 4 5 1 VCG ๐‘Œ+๐‘›โˆ’1 โ‰ฅ value of the highest unallocated bidder for 3 4 2 item ๐‘˜ . 3 2 1 88

  59. Add โ‰ค VCG ๐‘Œ+๐‘›โˆ’1 DC VCG ๐‘Œ 5 โˆผ ๐บ 7 โˆผ ๐บ 2 โˆผ ๐บ 3 6 4 4 5 1 3 4 2 3 2 1 89

  60. Add โ‰ค VCG ๐‘Œ+๐‘›โˆ’1 DC VCG ๐‘Œ 5 โˆผ ๐บ 7 โˆผ ๐บ 2 โˆผ ๐บ 3 6 4 4 5 1 3 4 2 3 2 1 90

  61. Add โ‰ค VCG ๐‘Œ+๐‘›โˆ’1 DC VCG ๐‘Œ 5 โˆผ ๐บ 7 โˆผ ๐บ 2 โˆผ ๐บ 3 6 4 4 5 1 3 4 2 3 2 1 91

  62. Add โ‰ค VCG ๐‘Œ+๐‘›โˆ’1 DC VCG ๐‘Œ 5 โˆผ ๐บ 7 โˆผ ๐บ 2 โˆผ ๐บ 3 6 4 4 5 1 3 4 2 Externality at least 9 3 2 1 92

  63. Add โ‰ค VCG ๐‘Œ+๐‘›โˆ’1 DC VCG ๐‘Œ 5 โˆผ ๐บ 7 โˆผ ๐บ 2 โˆผ ๐บ 3 6 4 4 5 1 3 4 2 3 2 1 93

  64. Add โ‰ค VCG ๐‘Œ+๐‘›โˆ’1 DC VCG ๐‘Œ 5 โˆผ ๐บ 7 โˆผ ๐บ 2 โˆผ ๐บ 3 6 4 4 5 1 3 4 2 3 2 1 94

  65. Add โ‰ค VCG ๐‘Œ+๐‘›โˆ’1 DC VCG ๐‘Œ 5 โˆผ ๐บ 7 โˆผ ๐บ 2 โˆผ ๐บ 3 6 4 4 5 1 3 4 2 3 2 1 95

  66. Add โ‰ค VCG ๐‘Œ+๐‘›โˆ’1 DC VCG ๐‘Œ 5 โˆผ ๐บ 7 โˆผ ๐บ 2 โˆผ ๐บ 3 6 4 4 5 1 3 4 2 Externality at least 2 3 2 1 96

  67. Add โ‰ค VCG ๐‘Œ+๐‘›โˆ’1 DC VCG ๐‘Œ Competition Complexity of Auctions 97 Eden et al. EC'17 Inbal Talgam-Cohen

  68. Add โ‰ค VCG ๐‘Œ+๐‘›โˆ’1 DC VCG ๐‘Œ 2 nd highest Highest value Add (๐‘˜) = DC of ๐‘Œ samples of unallocated โ‰ค VCG ๐‘Œ VCG ๐‘Œ+๐‘›โˆ’1 (๐‘˜) from ๐บ bidder for ๐‘˜ ๐‘˜ Competition Complexity of Auctions 98 Eden et al. EC'17 Inbal Talgam-Cohen

  69. Add โ‰ค VCG ๐‘Œ+๐‘›โˆ’1 DC VCG ๐‘Œ 2 nd highest Highest value Add (๐‘˜) = DC of ๐‘Œ samples โ‰ค of unallocated โ‰ค VCG ๐‘Œ VCG ๐‘Œ+๐‘›โˆ’1 (๐‘˜) from ๐บ bidder for ๐‘˜ ๐‘˜ Competition Complexity of Auctions 99 Eden et al. EC'17 Inbal Talgam-Cohen

  70. Add โ‰ค VCG ๐‘Œ+๐‘›โˆ’1 DC VCG ๐‘Œ 2 nd highest Highest value Add (๐‘˜) = DC of ๐‘Œ samples โ‰ค of unallocated โ‰ค VCG ๐‘Œ VCG ๐‘Œ+๐‘›โˆ’1 (๐‘˜) from ๐บ bidder for ๐‘˜ ๐‘˜ DC Identify ๐‘Œ bidders in VCG ๐‘Œ+๐‘›โˆ’1 before sampling their value for item ๐‘˜ out of which at most one will be allocated anything Competition Complexity of Auctions 100 Eden et al. EC'17 Inbal Talgam-Cohen

Recommend


More recommend