Predictocracy: Market Mechanisms for Public and Private Decisionmaking

1
Predictocracy Market Mechanisms for Public and
Private Decisionmaking
                                                  
                                               

• by Michael Abramowicz

2
Goals of project

• Describe different ways of structuring prediction
  markets to address potential problems.
• Explain how prediction markets can be used for
• supporting decisions
• assessing decisions
• making decisions

3
Prediction markets The traditional design

• Traders can buy and sell shares whose payoff
  depends on some future event.
• Iowa Electronic Markets
• Winner Take All market Winning candidates share
  pays off 1
• Vote share market Republican Democratic shares
  pay off in proportion to fraction of two-party
  vote

4
2004 Winner-take-all
5
Continuous double auction

• Bid queue contains offers to buy shares
• Ask queue contains offers to sell shares
• If trader enters an order to buy, it will either
  be
• matched based on most attractive offer on ask
  queue
• or added to the end of the bid queue
• Midpoint of bid and ask queues generally
  considered to be the instantaneous market forecast

6
Continuous double auction
7
Market accuracy

• Cant judge by any single case must look in
  aggregate
• Can we interpret prices as probabilities?

8
2005 MLB Tradesports
9
Market manipulation

• If markets affect policy, people will try to
  manipulate them.
• Literature indicates that attempts at
  manipulation
• have short-run effects
• but in the long term improve overall market
  accuracy by providing more profit potential.
• Manipulation can be successful on a particular
  contract only to the extent that someones
  bluffing actual succeeds in fooling others.
• If I offer to bet 100 that the temperature will
  rise 5 degrees in 10 minutes, I might succeed in
  misleading others but only because it is
  rational to place some epistemic weight on such
  offers.

10
Subsidized markets

• Existing markets are mostly unsubsidized, making
  them zero sum games.
• For some applications (esp. boring ones),
  subsidies may be necessary, and there are a
  variety of approaches to providing them.

11
Overcoming low liquidity

• Many say that only some prediction markets will
  have sufficient liquidity to allow meaningful
  probability generation.
• But subsidies can attract liquidity.
• Moreover, automated market maker mechanisms
  provide another tool, allowing individuals to
  trade without finding someone to trade against
• In effect, the house will enter into trades based
  on current prediction.
• In many real applications, we want a single
  expert or a handful of people to do a very
  careful analysis, but with strong financial
  incentives to get it right. Markets with
  automated market makers can find wisdom in
  crowds, not just wisdom of crowds.

12
Deliberative prediction market

• Reward depends on degree to which ones movement
  of the market sticks
• If previous prediction is 6, and I announce 8,
  but about a week later it is 7, then I break
  even.
• Last predictor is rewarded based on a scoring
  rule
• Gives incentive to convince other market
  participants

13
Why use prediction markets to support decisions?

• Objectivity
• Eliminates cheap talk
• Incentives
• May encourage search for or production of
  information
• Draws those most confident of ability to make
  accurate predictions
• Can allow organization to make credible forecasts
  to outsiders without revealing underlying
  information (e.g., CIA WMD market)

14
Why use prediction markets to predict decisions?

• If pool of decisionmakers is expected to be
  relatively good, then prediction itself will be
  of a good decision.
• Decisionmaker benefits from
• Numeric prediction
• Arguments (in deliberative market)
• Public benefits by reputationally constraining
  decisionmaker, identifying arbitrary and
  idiosyncratic decisionmaking.

15
Predictive peer review

• Editor still makes eventual decision
• But market provides a numeric assessment and
  gives incentives for public to make arguments and
  expose the flaws in others arguments.
• Currently considering experiment involving SSRN
  submissions.

16
Assessing decisions Conditional markets

• Predict B given A (e.g., corporate stock price if
  we open or dont open a factory)
• Simplest approach Unwinding (cancel contracts
  corresponding to contingency that doesnt occur).

17
Normative prediction markets

• The event being predicted is the announcement of
  someones opinion or evaluation. This can be used
  not only to assess decisions, but also
  potentially to make decisions.
• Example Predictive cost-benefit analysis
• Market predicts a cost-benefit analysis of a
  proposal to be performed in ten years.
• The actual cost-benefit analysis is used only to
  discipline the market. The prediction is used to
  determine whether the policy is enacted.

18
Normative prediction markets advantage over
voting regimes in general

• More representative decisions
• More informed decisions
• More consistent decisions
• More principled decisions
• More particularistic decisions
• More insulated decisions
• More scalable decisions
• More inexpensive decisions
• More decisions

19
Normative prediction markets advantage over
administrative agencies

• More representative decisions
• Prediction will weigh Democratic and Republican
  views roughly equally
• More principled decisions
• Ex post decisionmaker is not directly affecting
  policy, which has already been determined
• Thus, that decisionmaker will have less incentive
  to cheat off high-level principles, such as
  faithful statutory interpretation

20
Normative prediction markets advantage over
popular referenda

• More informed decisions
• The advantage of referenda is that they are very
  representative. But the cost of greater inclusion
  is less information per decisionmaker.
• A normative prediction market can continue to be
  representative, but traders and ex post assessors
  can have comprehensive information.
• More inexpensive decisions
• Markets must be subsidized, but fewer traders and
  assessors probably means less cost overall

21
Normative prediction markets advantage over
legislative bodies

• More insulated decisions
• Special interests wont know who ex post assessor
  will be
• Once time for ex post assessment comes, special
  interests have no reason to attempt to exercise
  influence
• Ex post assessors could be chosen by legislators
  and thus given more independence
• More decisions
• Reduced strategic bargaining and thus impasse

22
Normative prediction markets advantage over
courts

• More consistent decisions
• Decisions depend on what most judges would do,
  not on ideology or preferences of particular
  judges.
• More particularistic decisions
• Because decisionmaking is more consistent, there
  is less need for rules to constrain
  decisionmakers. As a result, there can be more
  standards and more particularistic
  decisionmaking.
• More scalable decisions
• Can choose how great a subsidy to allocate to a
  particular case.
• For example, could create legislative adjustment
  assistance regime to distribute a large fund to
  each of millions of people arguably affected by a
  gas tax, allocating just a few dollars in subsidy
  per case. With existing approaches, such cheap
  adjudication would either depend on inflexible
  rules or caprices of individual officials.