Five Problems in Information Market Design

Five problems in information market design l.jpg
1 / 38
0
0
898 days ago, 317 views
PowerPoint PPT Presentation
I Had A Dream. 1989, at Xanadu, predicted web,

Presentation Transcript

Slide 1

Five Problems in Information Market Design Robin Hanson George Mason University Talk at Microsoft Research, 9Feb04

Slide 2

I Had A Dream 1989, at Xanadu, anticipated web, & that insufficient "Data" tech discovers bits , but rather information still subtle What impact firearm control on wrongdoing, Iraq war on dread Honest & sound specialists require couple of bits to concur 100 bits can give  90% possibility sentiment diff  10% 10 6 bits can give  99% shot conclusion diff  1% I envisioned "Thought Futures" Betting chances as agreement, store examine by sponsor Disputants anticipated that would "put where is"

Slide 3

"Pays $1 if Bush wins" Will value rise or fall? offer E[ value change | ?? ] purchase value offer Lots of ?? get attempted, cost incorporates all! purchase Buy Low, Sell High (All theory is "betting"!)

Slide 4

Today's Prices 63-64% President Bush re-chose 2004 91-93% Kerry is Democratic chosen one 2004 82-83%   LOTR Oscar best picture of 2003 44-47% Bin Laden caught by 2005 25-30% Palestinian State by 2006 33-40%   Michael Jackson liable of salacious acts TradeSports.com

Slide 5

Information Markets Most theoretical markets total data well Very elusive data to beat normal give back Some make markets for data total reason In direct examinations, beat different establishments Racetrack chances beat track specialists (Figlewski 1979) OJ fates enhance climate gauge (Roll 1984) I.E.M. beat president surveys 451/596 (Berg et al 2001) HP advertise beat deals estimate 6/8 (Plott 2000) Stocks beat Challenger board (Maloney & Mulherin 2003)

Slide 6

Info Markets Benefits People self-select as specialists – we require not pick Incentive to be straightforward with yourself, and to remain out in the event that you don't have a clue. If not legit, in the long run lose cash & take off. Exact and constantly refreshed Consistent crosswise over differing settings Can practice rectify any inclination that you see

Slide 7

Corporate Applications Past Trials: Product Delivery: Xanadu, Seimens Sales: HP, Tradesports Current Trials: Drug, Insurance, Bank, … Key issue: ask high esteem questions Cost fluctuates nearly nothing, advantage shifts much HP dropped printer deals markets

Slide 9

Potential Problems Self-satisfying predictions Decision determination predisposition Price control Thin markets Combinatorial blasts Moral peril Regulation Secrecy Bozos Today's discussion Reduce data sharing Rich more "votes" Risk bending Bubbles

Slide 10

1. Self-Fulfilling Prophecies Problem: self-satisfying/crushing gauges Expect high deals, so raise promoting spending plan Expect not make due date, so quit making a decent attempt Fear dread assault on flight, so cross out it Solution: estimate result given exertion Predict deals given advertising spending plan Predict fruition date given hours/week work Predict dread assault given permit flight

Slide 11

$1 if Bad Virus & Cut Feature P(Cut) $1 if Cut Feature P(Virus | Cut) $1 Compare ! P(Virus | not Cut) $1 if Not Cut Feature $1 if Bad Virus & Not Cut Feature P(not Cut) If Cut Feature, Avoid Bad Virus?

Slide 12

Decision Market Applications E[ stock cost | fire CEO? ] E[ stock cost | get organization? ] E[ item deals | employ promotion office X? ] E[ wrongdoing rate | firearm control charge pass? ] E[Democrat win | Nominate Dean? ] E[ GDP | Bush re-chose? ] E[ SA common war | US moves troops out? ]

Slide 13

2. Choice Selection Bias If merchants think deciders will utilize data brokers don't have Market guidance may negate dealer information Related to "Newcomb's Problem"

Slide 14

Best to keep for this situation Stock if keep CEO Better to dump Best to dump for this situation Expected an incentive over dissemination is focal point of mass Stock if dump CEO A Decision State Space Imagine a uniform circulation over this range

Slide 15

If Deciders Have Same Info Market costs here if choice not corresponded with state Stock if keep CEO Better to dump Stock if dump CEO

Slide 16

Well-Informed Deciders Keep Stock if keep CEO Apparent focus Dump True focus Stock if dump CEO

Slide 17

Problem Seems Uncommon

Slide 18

Avoiding Selection Bias Problem situations irregular, however ... Let chiefs & their consultants exchange Make choice time clear to brokers Focus on costs just before choice time

Slide 19

3. Control Fears (e.g. PAM) Bad folks pick up $ by giving data, changing acts More conceivable if wager on particular points of interest, thick market PAM not on specifics, max pick up/exchange < $100 A decent arrangement for us if give few $, increase much information Terror & corporate harm now impacts enormous markets Bad folks lose $ to cloud showcase data If ease back acclimate to reputation, most pessimistic scenario is get no information $1M PAM justified, despite all the trouble if 0.1% shot increase 0.1% of $400B/yr We see little impact in lab, field tests If little, "clamor dealers" pull in others, net include data

Slide 20

Kyle Style Market Microstructure Price Manipulation Model Market producer Manipulator Informed merchant Noise broker Equilibrium

Slide 21

Bias Model Implications Manipulators are basically commotion dealers If educated dealers pockets excessively shallow, making it impossible to counter, clamor exchanging harms precision Else , commotion exchanging incites data exertion, helps ! Powerful to standard silliness demonstrate (QRE) Average outcome – "your mileage may shift"

Slide 22

4. Thin Market Problem Trade requires arrange in Assets and Time : holding up offers endure antagonistic choice Call markets, combo match, can help a few, however Most conceivable data markets don't exist Most are unlawful, and for the vast majority of the rest Expect couple of merchants, so don't make offer If realized that just a single individual has feeling on a subject, cost of basic market not uncover it!

Slide 23

Simple Info Markets Market Scoring Rules Scoring Rules conclusion pool issue thin market issue 100 .001 .01 .1 10 Best of Both Accuracy Estimates per broker

Slide 24

$ e i in the event that i $ s (1)- s (0) Market Scoring Rules Scoring standard: report r , state is i, remunerate s i ( r ) p = argmax r S i p i s i ( r ), S i p i s i ( p )  0 MSR: User t faces $ control: D s i = s i ( p t ) - s i ( p t-1 ) "Anybody can utilize scoring guideline if pay off last client" Is auto showcase creator, cost from net deals s Tiny deal expense:  p i ( s ) e i ( s i  s i + e i ) Big deal charge:  0 1 S i p i ( s (t)) s i '(t) dt Log run measured, cost  entropy  # darken

Slide 25

Every nation*quarter: Political strength Military movement Economic development US $ help US military action & worldwide, uncommon & all mixes

Slide 26

Return to Focus ? Exchange IQcs4 < 85 03 SAum3 105-125 03 Update Payoffs: If & Ave. pay Select New Price 65% Max Up 95.13% +$34.74 - $85.18 - $19.72 Buy 10% Up 68.72% +$2.74 - $3.28 - $1.07 You Pick 65 % +1.43 - 2.04 +0.34 Saudi Arabian Economic Health No Trade 62.47% $0.00 125 30 15 10% Dn 56.79% - $2.61 +$2.74 - $1.12 65 70 Sell Exit Issue 48.54% - $15.34 +$26.02 - $6.31 35 40 100 94 100 Max Dn 22.98% - $120.74 +$96.61 - $22.22 < 85 25 35 30 10 75 1 2 3 4 1 2 > 03 04 ? Come back to Form Execute a Trade If US military association in Saudi Arabia in 3 rd Quarter 2003 is not in the vicinity of 105 and 125, this exchange is invalid and void. Something else, if Iraq common soundness in 4 th Quarter 2003 is underneath 85, then I will get $1.43, yet in the event that it is not beneath 85, I will pay $2.04. Prematurely end exchange if cost has changed? Execute Trade Scenario

Slide 27

Laboratory Tests Joint work with John Ledyard (Caltech), Takashi Ishida (Net Exchange) Caltech understudies, get ~$30/session 6 periods/session, 12-15 minutes each Trained in 3var session, return for 8var Metric: Kulback-Leibler  i q i log (p i/q i ) separate from market costs to Bayesian convictions given all gathering information

Slide 28

Environments: Goals, Training (Actually: X Z Y ) Want in Environment: Many factors, few specifically related Few individuals, each not see all factors Can register sound gathering gauges Explainable, quick, impartial Training Environment: 3 twofold factors X,Y,Z, 2 3 = 8 combos P(X=0) = .3, P(X=Y) = .2, P(Z=1)= .5 3 individuals, see 10 instances of: AB, BC, AC Random guide XYZ to ABC Case A B C 1 1 - 1 2 1 - 0 3 1 - 0 4 1 - 0 5 1 - 0 6 1 - 1 7 1 - 1 8 1 - 0 9 1 - 0 10 0 - 0 Sum: 9 - 3 Same A B C A - - - - 4 B - - - - - - C - - - - - -

Slide 29

Experiment Environment ( Really: W V X S U Z Y T ) 8 paired vars: STUVWXYZ 2 8 = 256 mixes 20% = P(S=0) = P(S=T) = P(T=U) = P(U=V) = … = P(X=Y) = P(Y=Z) 6 individuals, each observe 10 cases: ABCD, EFGH, ABEF, CDGH, ACEG, BDFH irregular guide STUVWXYZ to ABCDEFGH Case A B C D E F G H 1 0 1 0 1 - - - - 2 1 0 0 1 - - - - 3 0 0 1 1 - - - - 4 1 0 1 1 - - - - 5 0 1 1 1 - - - - 6 1 0 0 1 - - - - 7 0 1 1 1 - - - - 8 1 0 0 1 - - - - 9 1 0 0 1 - - - - 10 1 0 0 1 - - - - Sum 6 3 4 10 - - - - Same A B C D E F G H A - - 1 2 6 - - - - - - - - B - - - - 7 3 - - - - - - - - C - - - - - - 4 - - - - - - - - D - - - - - - - - - - - - - - - - …

Slide 30

Mechanisms Compared Survey Mechanisms (# cases: 3var, 8var) Individual Scoring Rule (72,144) Log Opinion Pool (384,144) Market Mechanisms Simple Double Auction (24,18) Combined Value Call Market (24,18) MSR Market Maker (36,17)

Slide 31

Accuracy (95% C.L.)

Slide 32

KL(prices,group) 1-KL(uniform,group) MSR Info versus Time –

SPONSORS