Information Markets II: Theory, Outputs, Inputs, Foul Play, Combinatorics, Applications

1
Information Markets II Theory, Outputs, Inputs,
Foul Play, Combinatorics, Applications

• Robin Hanson
• Economics
• George Mason University

2
Theory I - Old

• No info - Supply and Demand
• Assume beliefs not respond to prices
• Price is weighted average of beliefs
• More influence risk takers, rich
• Info, Static - Rational Expectations
• Price clears, but beliefs depend on price
• No trade if not expect noise traders
• Price not reveal all info
• More influence info holders

3
Theory II - Market Microstruture

• Info, Dynamic Game Theory
• Example Kyle 85
• X - Informed trader(s) risk averse
• Y - Noise trader fool or liquidity pref
• Market makers no info, deep pockets
• If many compete, Price Evaluexy
• Info markets use risk-neutral limit
• If Y larger, X larger to compensate more info
  gathered, so more accuracy!

4
Theory III Behavioral Finance

• Humans are overconfident
• Far more speculative trade than need
• Mere fact of disagreement shows
• Overconfidence varies with person, experience,
  consequence severity
• Implications
• Price in part an ave of beliefs?
• Adds noise to price aggregates?
• Prices more honest than talk, polls,

5
Outputs

• What price is best estimate?
• last? median? an average? Reweight trades?
• If not last, auto-trader to fix makes !
• This good discipline re if really can fix
• Imagine Govt agency fixing stock prices!
• Require post comment with each trade?
• Use trade record in performance review?
• Reward contribution vs. infer other abilities
• Crunch trade data to see who thinks what
• Give more a feeling of participation?
• Dont let these issues distract you from

6
Ask the Right Questions

• High value to more accurate estimates!
• Relevant standard beat existing institutions
• Where suspect more accuracy is possible
• Suspect info is withheld, or not sure who has it
• Prefer fun, easy to explain and judge
• Can let many know best estimates
• Not fear estimates reveal secrets
• Not using uncertainty, biases to motivate
• Avoid inducing foul play

7
Conditional Estimates

• Can avoid self-defeating predictions
• Condition on decision, advises it
• Dont confuse correlation and cause
• Bias if decision makers will know more
• Clear decision time and use prices then
• Choose instrumental variables
• E.g., condition on random decision

8
Inputs I

• Final Judging using prices risks gaming!
• Audit lotteries reduce ave cost, but more risk
• Refine claim central vs. decentralized
• Credentialing as compromise?
• Participants
• Mainly want people can get relevant info
• Diversity helps, but only of info
• Trading experience a plus, but not the key
• Standard trading needs min traders/claim
• Fools are fine, up to a point

9
Inputs II

• Cash, play money, or prizes to best traders?
• Recent paper on football, real vs. play-prizes
  same
• Note prizes risk inducing large random trades!
• Real Choice stuff vs. brag rights vs. fun
• Fun risks them not caring enough to be honest
• Scale economies of bragging rights?
• Info concept brag of value of info add to
  org
• How much must pay?
• If many have info, just need induce them to tell
• If traders must do research, must be paid more
• Bigger trader pool helps find low cost providers
• When pay cash upfront, per trade, market maker
• Subsidized market maker pays only for new info

10
Foul Play I

• Generic fix limit who/when trade
• Lying
• If advisors can bet, may talk less
• Fix? Let advisors show bet stake
• Manipulation
• Idea lose on trades, gain in decisions
• Field Effect rare, short-lived
• Lab no net effect? (see conf talks)
• Theory trading on any consideration other than
  asset value is noise trading

11
Foul Play II

• Sabotage (Moral Hazard)
• Rare (Not 9/11, 82 Tylenol, 02 PaineWebber)
• Hard match willing capital skilled labor
• Fix Avoid thick market on small events
• Fix Bound individual stakes (eg finish project)
• Embezzlement
• Stat insiders windfall? Keep info from team?
• Fix Special accounts trade first
• Fix? new color of , subsidy at info value est.
• Retribution anonymity helps at a cost
• Can still brag re overall record

12
Combinatorics I The problem

• Each trader wants to trade on his info, be
  insured against all other issues
• Ex what weather can we forecast?
• Per hour per zip code?
• Distribution over wind, rain amount?
• Conditional on recent, nearby weather?
• Old story
• Vast possible Arrow-Debreu assets
• But fixed costs, traders avoid thin
• But regulation is biggest cost by far
• Many computing tricks not tried

13
Combinatorics II - Approaches

• All decompose trades into state assets
• Example Win, place, show overlaps
• Call markets
• Compute to find matches in offer pool
• Related markets thicken each other
• Recent computational complexity results
• Market makers
• Stands ready to trade all assets
• Requires subsidy per base claim, but not for
  adding all combos of base
• Open issues re combinatorial explosion

14
Pushing the Limit
Simple Info Markets
Accuracy
thin market problem
Estimates per trader
15
Accuracy (95 C.L.)
16
Applications

• Private Policy
• Sales (own and others)
• Project completion, quality (bug rate)
• Decisions mergers, subcontractor choice,
  regional expansions,
• Public Policy
• Epidemics,
• monetary policy, health policy, ...
• School job applicants