Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Daniel Doubrovkine on Auctions and bidding: A guide for computer scientists

Daniel Doubrovkine on Auctions and bidding: A guide for computer scientists

There is a veritable menagerie of auctions — single dimensional, multi-dimensional, single sided, double sided, first price, second price, English, Dutch, Japanese, sealed bid — and these have been extensively discussed and analyzed in the economics literature. We will survey this literature from a computer science perspective, primarily from the viewpoint of computer scientists who are interested in learning about auction theory, and to provide pointers into the economics literature for those who want a deeper technical understanding. We'll also take a peek into how the paper was used to build a new online auction platform Artsy.

Papers_We_Love

May 24, 2017
Tweet

More Decks by Papers_We_Love

Other Decks in Science

Transcript

  1. MANOLO VALDÉS, Ivy, 2012
    https://www.artsy.net/artwork/manolo-valdes-ivy-5
    2017
    Daniel Doubrovkine | [email protected] | @dblockdotorg
    Auctions and bidding:
    A guide for computer
    scientists.

    View Slide

  2. A Paper I Love

    http://curatorsintl.org

    View Slide

  3. What’s in an Auction?

    View Slide

  4. A Louis XVI Sofa

    View Slide

  5. Auctions Everywhere!

    View Slide

  6. A Menagerie of Auctions

    - Single Dimensional
    - Multi Dimensional
    - Single Sided
    - Double Sided
    - First Price
    - Second Price
    - Kth Price
    - English
    - Dutch
    - Japanese
    - Open-Cry
    - Sealed Bid
    - Combinatorial

    View Slide

  7. TOC

    - Auction Families
    - Analyzing Auctions
    - Abstract Process View
    - Computer Science
    - Actual Implementation

    View Slide

  8. Auction Families

    View Slide

  9. When were auctions invented?

    - Auction = Exchange of Money for Goods

    Money ~ 700 B.C. 

    Rome and Greece ~ 5th Century B.C.

    View Slide

  10. Independent Features (Zoology)

    - Single dimensional, or 

    multi-dimensional.
    - One sided or two sided.
    - Open-cry or sealed bid.
    - First price or kth price.
    - Single-unit or multi-unit.
    - Single-item or multi-item 

    (combinatorial).

    View Slide

  11. Single vs. Multi-Dimensional

    - Single Dimensional: price only
    - Multi Dimensional: other aspects

    View Slide

  12. One vs. Two-Sided

    - One-Sided: either buyer or seller bids
    - Two-Sided: both sides submit bids

    - The auctioneer decides

    View Slide

  13. Open Cry vs. Sealed Bid

    - Open Cry: every bidder “hears” bids
    - Sealed Bid: only auctioneer sees bids

    View Slide

  14. First Price vs. K-th Price

    - First Price: winning bidder pays the price
    - Second Price: winning bidder pays the

    underbidder price
    - …
    - 16th price? Risk seekers!

    View Slide

  15. Single Unit vs. Multi Unit

    - Single Unit: bid for a single good
    - Multi Unit: bid for a set of goods

    View Slide

  16. Combinatorial Auction

    - Multiple, heterogeneous goods
    - Bids on arbitrary combinations

    View Slide

  17. “English” (or “Silent”) Auction

    - A single dimensional, one sided, single good, open-cry auction
    - May have reserve price
    - Bids are made in ascending order
    - Auctioneer calls the next bid
    - Ends when nobody accepts a higher bid or at a given time

    View Slide

  18. “Dutch” Auction

    - A single dimensional, one sided, 

    single good, open-cry auction
    - May have reserve price
    - Bids are made in descending order
    - Auctioneer calls the next bid
    - Ends when someone accepts a bid 

    or at reserve
    18

    View Slide

  19. “Japanese” Auction

    - A single dimensional, one sided, single good, open-cry auction
    - May have reserve price
    - Bids are made in ascending order
    - Auctioneer calls increasing bids
    - Bidders indicate that they drop out
    - Auction ends when there’s a single bidder

    View Slide

  20. “Vickrey” Auction

    - Second price, silent
    - Bidder is free to bid max
    - Possible trouble maximizing seller revenue

    View Slide

  21. “Buy Side” Auctions

    - Sellers offering various terms
    - Credit cards, mortgages, procurement

    View Slide

  22. “Double Auction”, “Call Market” or “Clearing House”

    - Ask = a request from someone to buy a good
    from the asker at a price
    - Bid = a request from a buyer to buy a good
    from the seller at a price
    - Offer = a general term for a “bid” or an “ask”
    trade price = k · bid price + (1 − k) · ask price

    View Slide

  23. “Continuous Double Auction”, Stock Market

    View Slide

  24. Multi-Dimensional & Multi-Attribute Auctions

    … although auctions with side constraints on
    overall assignments can be cleared in
    polynomial time using network flow algorithms,
    when the demand is indivisible, assessing an
    optimal allocation becomes computationally
    intractable and requires solving np-hard
    optimization problems such as the generalized
    assignment problem, the multiple knapsack
    problem and the bin packing problem…

    View Slide

  25. Analyzing Auctions

    View Slide

  26. Concerns

    - Efficiency: achieve allocation of money and goods that
    maximizes the total value of all participants

    - Optimality: maximizing the revenue of the bid taker

    View Slide

  27. Auction Models

    - Independent private values model
    - Correlated values model
    - Almost common values model

    View Slide

  28. Independent Private Values Model

    - Many buyers bid for a single, invisible object
    - Each buyer is risk neutral
    - Each buyer bids up to the private value of the object
    Dutch and First Price Sealed Bid Auctions are equivalent: a bidder only has one choice, pick a price to bid
    English and Second Price Sealed Bid Auctions are equivalent: bid up to the value to the bidder
    this model is Pareto optimal
    this model gives almost identical revenue to the seller
    leads to winner’s curse where the winner overpays

    View Slide

  29. Revenue Equivalence Theorem (RET)

    - All major types of auctions generate the same expected revenue
    “Assume each of a given number of risk neutral potential buyers has a privately known valuation independently
    drawn from a strictly increasing atomless distribution, and that no buyer wants more than one of the k identical
    indivisible objects. Then any mechanism in which (i) the objects always go to the k buyers with the highest
    valuations and (ii) any bidder with the lowest feasible valuation expects zero surplus, yields the same expected
    revenue (and results in each bidder making the same expected payment as a function of her valuation).”

    View Slide

  30. Correlated Values Model

    - Generalization of the independent private values model
    - Not all buyers have an accurate value of the goods sold
    - English auction will lead to higher prices than a Second Bid sealed auction
    - Revealing information will increase prices overall, even if the information is bad

    View Slide

  31. Almost Common Values Model

    - Bidders have almost common values, but different estimates
    - A bidder with a higher value will bid more aggressively
    - Competing bidders with lower value will bid more conservatively, increase winner’s curse
    - A small information advantage is magnified
    - There’re advantages to appear aggressive

    View Slide

  32. Generalized Analysis w/ Mechanisms

    - Agents
    - Outcomes
    - Decision rules (efficient or dictatorial)
    - Transfers
    - Social choice functions
    (skip …)

    View Slide

  33. Collusion, lying, and other sharp practice

    - Rings of bidders, also called “pies” and “kippers”
    - Bidding “clubs”
    - Sniping
    - Bid shielding
    - Retaliation

    View Slide

  34. Auctioneer (seller) fraud

    - Inserting fake bids to raise the price of a second bid auction
    - Fixing commissions
    - “Shills” or “puffers”
    - “Book” bids

    View Slide

  35. Abstract Models

    View Slide

  36. Parametric Model

    - Common auction characteristics
    - Auction parameter space
    - Matching functions

    View Slide

  37. Abstract Process Model

    - Bid call
    - Ask call
    - Bid collection
    - Ask collection
    - Bid retraction
    - Ask retraction
    - Winner determination
    - Clearing
    - Information revelation
    - Tie breaking
    - Stage switch
    - Closing

    View Slide

  38. Practical Implementations

    - Building an English Auction w/ MongoDB

    http://artsy.github.io/blog/2014/04/17/building-an-english-auction-with-mongodb/

    - Build a Real-Time Auction Engine in Scala using Event Sourcing

    https://www.youtube.com/watch?v=szgmgBfvDrs&index=19

    View Slide

  39. Big Questions

    - Can automated trading outperform humans?
    - Can a market-based method compute the outcome of a
    distributed problem?

    View Slide

  40. http://www.sci.brooklyn.cuny.edu/~parsons/projects/mech-design/publications/bluffers-final.pdf

    View Slide

  41. Thank you.
    Daniel Doubrovkine | [email protected] | @dblockdotorg | code.dblock.org

    View Slide