Transcript Market Games for Mining Customer Information
Prediction Markets and the Wisdom of Crowds
David Pennock Yahoo! Research NYC
Research Outline
•
Prediction Markets Survey
• • •
What is a prediction market?
Examples Some research findings
•
The Wisdom of Crowds: A Story
Research Bet = Credible Opinion
Hillary Clinton will win the election “I bet $100 Hillary will win at 1 to 2 odds” • •
Which is more believable?
More Informative?
Betting intermediaries
• • •
Las Vegas, Wall Street, Betfair, Intrade,...
Prices: stable consensus of a large number of quantitative, credible opinions Excellent empirical track record
Research Example: March Madness
Research More Socially Redeemable Example
QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
http://intrade.com
Screen capture 2007/05/18 QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
Research A Prediction Market
•
Take a random variable, e.g.
Bird Flu Outbreak US 2007?
(Y/N) •
Turn it into a financial instrument payoff = realized value of variable
I am entitled to: $1 if Bird Flu US ’07 $0 if Bird Flu US ’07
Research Why?
• • Get information
price
probability of uncertain event (in theory, in the lab, in the field, ...more later) Is there some future event you’d like to forecast?
A prediction market can probably help
Research A Prediction Market
•
Take a random variable, e.g.
2008 CA Earthquake?
US’08Pres = Dem?
•
Turn it into a financial instrument payoff = realized value of variable
I am entitled to: $1 if = 6 $0 if 6 = 6 ?
Research
• • •
Aside: Terminology
Key aspect: payout is uncertain Called variously: asset, security, contingent claim, derivative (future, option), stock, prediction market, information market, gamble, bet, wager, lottery Historically mixed reputation
• •
Esp. gambling aspect A time when options were frowned upon
•
But when regulated serve important social roles...
Research Getting Information
• Non-market approach:
ask an expert How much would you pay for this?
$1 if = 6 $0 if 6 •
A: $5/36
•
$0.1389
caveat: expert is knowledgeable
• • •
caveat: expert is truthful caveat: expert is risk neutral, or ~ RN for $1 caveat: expert has no significant outside stakes
Research Getting Information
• Non-market approach: • pay an expert
Ask the expert for his report r of the probability
• • • • • •
Offer to pay the expert
• $100 + log r if = 6 “logarithmic scoring rule”, a “proper” scoring rule • $100 + log (1-r) if 6
It so happens that the expert maximizes expected profit by reporting r truthfully caveat: expert is knowledgeable caveat: expert is truthful caveat: expert is risk neutral, or ~ RN caveat: expert has no significant outside stakes
Research Getting Information
• Market approach:
“ask” the public—experts & non experts alike —by opening a market:
I am entitled to: $1 if = 6 $0 if 6 • • • Let any person i submit a bid order:
an offer to buy q i units at price p i
Let any person j submit an ask order:
an offer to sell q j units at price p j
= 6
Match up agreeable trades (many poss. mechs...)
Non-Market Alternatives vs. Markets
Opinion poll Sampling No incentive to be truthful Equally weighted information Hard to be real-time Ask Experts Identifying experts can be hard Incentives Combining opinions can be difficult Prediction Markets Self-selection Monetary incentive and more Money-weighted information Real-time Self-organizing
Non-Market Alternatives vs. Markets
Machine learning/Statistics Historical data Past and future are related Hard to incorporate recent new information Prediction Markets No need for data No assumption on past and future Immediately incorporate new information
Function of Markets
Risk Management 2:
If is bad for me, I buy a bunch of $1 if $0 otherwise If my house is struck by lightening, I am compensated.
Risk Management Examples
Insurance I buy car insurance to hedge the risk of accident Futures Farmers sell soybean futures to hedge the risk of price drop Options Investors buy options to hedge the risk of stock price changes
Financial Markets vs. Prediction Markets
Primary Secondary Financial Markets Social welfare (trade) Hedging risk Information aggregation Prediction Markets Information aggregation Social welfare (trade) Hedging risk
Giving/Getting Information
• What you can say/learn % chance that – Hillary wins – GOP wins Texas – YHOO stock > 30 – Duke wins tourney – Oil prices fall – Heat index rises – Hurricane hits Florida – Rains at place/time • Where – IEM, Intrade.com
– Intrade.com
– Stock options market – Las Vegas, Betfair – Futures market – Weather derivatives – Insurance company – Weatherbill.com
QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
http://intrade.com
http://tradesports.com
Screen capture 2007/05/18 QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
Research Intrade Election Coverage
QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
Research
http://www.biz.uiowa.edu/iem
QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
http://www.hedgestreet.com/
Screen capture 2007/05/18 QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
http://www.wsex.com/
Screen capture 2007/05/18 QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
Play money; Real predictions
http://www.hsx.com/
http://www.ideosphere.com
Cancer cured by 2010
http://us.newsfutures.com/
Machine Go champion by 2020
Research Yahoo!/O’Reilly Tech Buzz Game
http://buzz.research.yahoo.com/
More Prediction Market Games
• BizPredict.com
• CasualObserver.net
• FTPredict.com
• InklingMarkets.com
• ProTrade.com
• StorageMarkets.com
• TheSimExchange.com
• TheWSX.com
• • Alexadex, Celebdaq, Cenimar, BetBubble, Betocracy, CrowdIQ, MediaMammon,Owise, PublicGyan, RIMDEX, Smarkets, Trendio, TwoCrowds http://www.chrisfmasse.com/3/3/markets/#Play-Money_Prediction_Markets
Catalysts
• Markets have long history of predictive accuracy: why catching on now as tool?
• No press is bad press: Policy Analysis Market (“terror futures”) • Surowiecki's “Wisdom of Crowds” • Companies: – Google, Microsoft, Yahoo!; CrowdIQ, HSX, InklingMarkets, NewsFutures • Press: BusinessWeek, CBS News, Economist, NYTimes, Time, WSJ, ...
http://us.newsfutures.com/home/articles.html
Does it work?
Yes, evidence from real markets, laboratory experiments, and theory Racetrack odds beat track experts [Figlewski 1979] Orange Juice futures improve weather forecast [Roll 1984] I.E.M. beat political polls 451/596 [Forsythe 1992, 1999][Oliven 1995][Rietz 1998][Berg 2001][Pennock 2002] HP market beat sales forecast 6/8 [Plott 2000] Sports betting markets provide accurate forecasts of game outcomes [Gandar 1998][Thaler 1988][Debnath EC’03][Schmidt 2002] Market games work [Servan-Schreiber 2004][Pennock 2001] Laboratory experiments confirm information aggregation [Plott 1982;1988;1997][Forsythe 1990][Chen, EC’01] Theory: “rational expectations” [Grossman 1981][Lucas 1972] More later …
Example: IEM 1992
Example: IEM
Example: IEM
[Source: Berg, DARPA Workshop, 2002]
Example: IEM
AAAI’04 July 2004 MP1-35
[Source: Berg, DARPA Workshop, 2002]
Example: IEM
AAAI’04 July 2004 MP1-36
[Source: Wolfers 2004]
Speed: TradeSports
Contract: Pays $100 if Cubs win game 6 (NLCS) Price of contract (=Probability that Cubs win) Cubs are winning 3-0 top of the 8 th 1 out.
Fan reaches over and spoils Alou’s catch. Still 1 out.
The Marlins proceed to hit 8 runs in the 8 th inning Time (in Ireland) MP1-37 AAAI’04 July 2004
Research Does money matter? Play vs real, head to head
• •
Experiment 2003 NFL Season ProbabilitySports.com Online football forecasting competition
• • • • •
Contestants assess probabilities for each game Quadratic scoring rule ~2,000 “experts”, plus: NewsFutures (play $) Tradesports (real $)
• Used “last trade” prices • •
Results: Play money and real money performed similarly
•
6 th and 8 th respectively Markets beat most of the ~2,000 contestants
•
Average of experts came 39 th (caveat)
Electronic Markets
, Emile Servan Schreiber, Justin Wolfers, David Pennock and Brian Galebach
100 90 30 20 10 0 80 70 60 50 40 Prediction Accuracy
Research
100 Market Forecast Winning Probability and Actual Winning Probability TradeSports: Correlation=0.96
NewsFutures: Correlation=0.94
0 10 20 30 40 50 60 Trading Price Prior to Game 70 80 90 100 Data are grouped so that prices are rounded to the nearest ten percentage points; n=416 teams in 208 games 75 50 25 Prices: TradeSports and NewsFutures Fitted Value: Linear regression 45 degree line 0 0 20 40 60 NewsFutures Prices n=416 over 208 NFL games.
Correlation between TradeSports and NewsFutures prices = 0.97
80
Prediction Performance of Markets Relative to Individual Experts
0 20 40 60 80 100 120 140 160 180 200 220 240 260 280 300 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Week into the NFL season
NewsFutures Tradesports 100
Research Does money matter? Play vs real, head to head
Probability Football Avg TradeSports (real-money) NewsFutures (play-money) Difference TS - NF Mean Absolute Error
= lose_price
[lower is better]
Root Mean Squared Error
= ?Average( lose_price 2 )
[lower is better] 0.443 (0.012) 0.476 (0.025) 0.439 (0.011) 0.468 (0.023)
0.436
(0.012)
0.467
(0.024) 0.003 (0.016) 0.001 (0.033)
Average Quadratic Score
= 100 - 400*( lose_price 2 )
[higher is better]
Average Logarithmic Score
= Log(win_price)
[higher (less negative) is better] 9.323 (4.75) -0.649 (0.027) 12.410 (4.37)
-0.631
(0.024)
12.427
(4.57)
-0.631
(0.025) -0.017 (6.32) 0.000 (0.035) Statistically: TS ~ NF NF >> Avg TS > Avg
average log score arbitrage closure AAAI’04 July 2004
Real markets vs. market games
HSX IEM MP1-41
Real markets vs. market games
HSX FX, F1P6 probabilistic forecasts expected value forecasts 100 50 20 10 5 489 movies 2 1 actual AAAI’04 July 2004 1 2 5 10 20 forecast source
F1P6 linear scoring F1P6 F1-style scoring
betting odds F1P6 flat scoring F1P6 winner scoring 50 estimate 100 MP1-42 avg log score
-1.84
-1.82
-1.86
-2.03
-2.32
Research A Wisdom of Crowds Story
• • •
ProbabilitySports.com
Thousands of probability judgments for sporting events
• • •
Alice: Jets 67% chance to beat Patriots Bob: Jets 48% chance to beat Patriots Carol, Don, Ellen, Frank, ...
Reward: Quadratic scoring rule: Best probability judgments maximize expected score 1
/7 Story Survey Research Opinion
Individuals Research
QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture.
• •
Most individuals are poor predictors 2005 NFL Season
• • •
Best: 3747 points Average: -944 Median: -275 1,298 out of 2,231 scored below zero (takes work!)
Research Individuals
•
Poorly calibrated (too extreme)
•
Teams given < 20% chance actually won 30% of the time
•
Teams given > 80% chance actually won 60% of the time
Research The Crowd
• •
Create a crowd predictor by simply averaging everyone’s probabilities
• •
Crowd = 1/n(Alice + Bob + Carol + ... ) 2005: Crowd scored 3371 points (7th out of 2231) !
Wisdom of fools: Create a predictor by averaging everyone who scored below zero
• •
2717 points (62nd place) !
(the best “fool” finished in 934th place)
Research The Crowd: How Big?
More: http://blog.oddhead.com/2007/01/04/the-wisdom-of-the-probabilitysports-crowd/ http://www.overcomingbias.com/2007/02/how_and_when_to.html
Research
• •
Can We Do Better?: ML/Stats
[Dani et al. UAI 2006]
Maybe Not
• •
CS “experts algorithms” Other expert weights
• •
Calibrated experts Other averaging fn’s (geo mean, RMS, power means, mean of odds, ...)
•
Machine learning (NB, SVM, LR, DT, ...) Maybe So
• •
Bayesian modeling + EM Nearest neighbor (multi-year)
Research Can we do better?: Markets
Prediction Performance of Markets Relative to Individual Experts
0 20 40 60 80 100 120 140 160 180 200 220 240 260 280 300 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Week into the NFL season
NewsFutures Tradesports