1 / 53

A Combinatorial Prediction Market for the U.S. Elections

A Combinatorial Prediction Market for the U.S. Elections. MIROSLAV DUDÍK, SEBASTIEN LAHAIE, DAVID M. PENNOCK, DAVID M. ROTHSCHILD Microsoft Research. Thanks: Dan Osherson , Arvid Wang, Jake Abernethy, Rafael Frongillo , Rajiv S ethi , Rob Schapire , Jenn Wortman Vaughn.

reece
Download Presentation

A Combinatorial Prediction Market for the U.S. Elections

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. A Combinatorial Prediction Market for the U.S. Elections MIROSLAV DUDÍK, SEBASTIEN LAHAIE,DAVID M. PENNOCK, DAVID M. ROTHSCHILD Microsoft Research Thanks: Dan Osherson, Arvid Wang, Jake Abernethy, Rafael Frongillo, Rajiv Sethi, Rob Schapire, JennWortman Vaughn

  2. An ExamplePredictionMarket • A random variable, e.g. • Turned into a financial instrument payoff = realized value of variable Will Obama win 2012 Presidential election?(Y/N) I am entitled to: Obamawins Romneywins $100 if $0 if

  3. Trade elections like stocks

  4. Trade elections like stocks

  5. Super Tuesday • Predictions as of2:52 p.m. ETMonday, March 5, 2012

  6. Super Tuesday • Predictions as of2:52 p.m. ETMonday, March 5, 2012

  7. Super Tuesday • Predictions as of2:52 p.m. ETMonday, March 5, 2012

  8. Conditional probability “logistic regression” “combinatorial prediction markets”

  9. Super Tuesday: Polls vs Markets • Prior to MI primary, Santorum favored in Ohio: leading polls by wide margin • Within hours of Romney winning MI, our OH prediction swung to Romney • New York Times Nate Silver’s poll-based forecast a couple days behind at least

  10. Real-Time! • Real-time reaction to big events. • Romney dominating when others led the polls.

  11. GOP Primary: Polls vsMarkets

  12. Answering ?-mark headlines with data [Santorum’s] odds of winning the nomination rose slightly, to a non-negligible 11.3 percent likelihood to Romney's 79.5 Reuters: “The dominant narrative, the day after the presidential election, is the triumph of the quants.” Mashable: “here is the absolute, undoubted winner of this election: Nate Silver and his running mate, big data.” ReadWrite: “This is about the triumph of machines and software over gut instinct”

  13. Intrade = Predictions x100 • Presidential – National • Presidential – State-by-state • Presidential – Electoral College • Senate • House • Gubernatorial

  14. PredictWiseQ = Predictions 100 • 22% chance Romney will win in Iowa but Obama will win the national election • 75.7% chance the same party will win both Michigan and Ohio • 48.3% chance Obama gets 300 or more Electoral College votes • 12.3% chance Obama will win between 6 and 8 states that begin with the letter M • Path of blue from Canada to Mexico

  15. Some Counting 54 “states”: 48 + DC + Maine (2), Nebraska (3) 254= 18 quadrillion possible outcomes 22541018008915383333485 distinct predictionsMore than a googol, less than a googolplex NOT independent

  16. A combinatorial question:How pivotal was Ohio? • Day after 1st debate (October 4) • If Obama wins Ohio, 90% he’ll win the election • If Obama wins Florida, 95%; Virginia, 92% • If Romney wins Ohio, 41% he’ll win election; Florida, 32%; Virginia 31% • If October jobs report bleak, will Romney win?

  17. PredictWiseQ • In theory, combo PM yields billions more predictions “for free”. What about in practice? • A combinatorial prediction market for the election: http://PredictWiseQ.com • Fully working beta & field test • Sep 16 – Nov 6, 2012 • 680 users; 437 made at least one trade • 3137 trades in all 23 possible prediction categories • 514 distinct security types were bought261 of these were traded by a unique user • Empirical paper to appear in ACM EC’13 • Market maker algorithm details in ACM EC’12

  18. http://PredictWiseQ.com

  19. http://PredictWiseQ.com

  20. Why Expressiveness? • Dem Pres, Dem Senate, Dem HouseDem Pres, Dem Senate, GOP HouseDem Pres, GOP Senate, Dem HouseDem Pres, GOP Senate, GOP House... • Dem PresDem HouseDem wins >=270 electoral votesDem wins >=280 electoral votes...

  21. Standard vs Combinatorial • Industry standard: Ignore relationshipsTreat them as independent markets • Las Vegas sports bettingKentucky horseracingWall Street stock optionsHigh Street spread betting • Combinatorial • Automatic information propagation: reward traders for information, not computational horsepower

  22. Consistent pricing 1 Independent markets A&B&C 0 0 1 A&B’&C

  23. Consistent pricing 1 Independent markets Prices p A&B&C 0 0 1 A&B’&C

  24. Consistent pricing 1 Independent markets A&B&C 0 0 1 A&B’&C

  25. Consistent pricing 1 Independent markets C = 0.9 A = 0.8 B = 0.6 A&B&C 0 0 1 A&B’&C

  26. Consistent pricing 1 0.9 C = 0.9 0.8 A = 0.8 0.6 B = 0.6 A&B&C 0 0 1 A&B’&C

  27. Consistent pricing 1 0.9 C = 0.9 0.8 A = 0.8 0.6 B = 0.6 A&B&C 0 0.9 0 0.4 1 0.8 A&B’&C

  28. Consistent pricing 1 0.9 C = 0.9 0.8 A = 0.8 0.6 B = 0.6 A&B&C 0 0.9 0 0.4 1 0.8 A&B’&C

  29. Approximate pricing 1 0.9 C = 0.9 0.8 A = 0.8 0.6 B = 0.6 A&B&C 0 0.9 0 0.4 1 0.8 A&B’&C

  30. Approximate pricing 1 0.9 C = 0.9 0.8 A = 0.8 Prices p 0.6 B = 0.6 A&B&C 0 0.9 0 0.4 1 0.8 A&B’&C

  31. Approximate pricing 1 0.9 C = 0.9 0.8 A = 0.8 Prices p BuyNotB A&B&C 0.5 B = 0.5 0 0.9 0 0.5 1 0.8 A&B’&C

  32. Approximate pricing 1 0.9 C = 0.9 0.8 A = 0.8 Prices p A&B&C 0.5 B = 0.5 0 0.9 0 0.5 1 0.8 A&B’&C

  33. Approximate pricing 1 0.9 C = 0.9 0.8 A = 0.8 Prices p B = 0.55 A&B&C 0.5 0 0.9 0 0.5 1 0.8 A&B’&C

  34. Does it work? 10 States Tested on over 300K complex predictions from Princeton study Budget

  35. Does it work? 50 States Tested on over 300K complex predictions from Princeton study Log Score Budget

  36. Does it work? Revenue Tested on over 300K complex predictions from Princeton study

  37. No really, does it work? • http://PredictWiseQ.com

  38. User activity, Mean daily profit

  39. Focal predictions Top singleton securities Top combinatorial securities VA, WI = Dem CO, IN, NH, OH, VA = Dem Fed, OH = Rep OH, VA = Dem FL=Rep, Fed=Dem FL, NC = Rep FL, Fed, OH = Dem NH, NV, OH, VA, WI = Dem • Federal, 43K • FL, 26K • OH, 17K • VA, 15K • CO, 12K • Electoral votes, 10K • NC, 10K • Senate, 8K

  40. Return

  41. Types of securitiesby user return

  42. “Diversification” by user return

  43. Accuracy

  44. Accuracy

  45. Full distributions

  46. Counterfactuals

  47. Further reading Our paper in ACM EC’13coming soon: email me Our paper in ACM EC’12http://research.microsoft.com/apps/pubs/default.aspx?id=167977 Blog post on PredictWiseQhttp://blog.oddhead.com/2012/10/06/predictwiseq/ Blog on gory details: What is (and what good is) a combinatorial prediction market?http://bit.ly/combopm Guest post on Freakonomicshttp://bit.ly/combopmfreak

  48. NYSE 1926 http://online.wsj.com/article/SB10001424052748704858404576134372454343538.html

  49. NYSE 1987 http://online.wsj.com/article/SB10001424052748704858404576134372454343538.html

  50. NYSE 2006-2011 • 2011 Deutsche BörseAG • 2007 Euronext • 2006 Archipelago, ipo

More Related