Felix

Document Sample
Felix Powered By Docstoc
					Prediction Markets

   Justin Wolfers and
     Eric Zitzewitz
                  Quality of Predictions

     Political domain
       Markets predict better than polls



     Economic derivatives
       Markets do no better than the “consensus” forecast




2
           Why Would Markets Do Better?

     Incentives to produce new information
       Unlikely given the low incentives
       Not even quite clear to what extent traders are
        motivated by monetary gains (NFL example)

     Truthful revelation of available information
       Experimental markets in firms

     Superior aggregation of information


3
                 Superior Aggregation

     Gallup Poll Question: Next, we’d like you to think
      about the general election for President to be held
      in November. If Vice-President Al Gore were the
      Democratic Party’s candidate and Texas Governor
      George W. Bush were the Republican Party’s
      candidate, who would you be more likely to vote
      for?
     A “Gore” answer implies that there is a probability
      p>.5 that the person will vote for Gore
     If .6 say they vote for Gore, the fraction who will
      vote for Gore lies in the interval [.3,.8]
4
                Probabilistic Information

     How likely is it that you will vote in the 2000 election
      for President [definitely vote, probably vote,
      probably not vote, definitely not vote]?
     What do you think is the percent chance that you
      will cast a vote for President?
        72% say they will “definitely vote”
        54% are 100% certain, 10% vote with probability [.91-
         .99], 8% with probabilty [.51-.9]




5
                        Consequences

     Market design
        TradeSports.com offers contract for 2004 Year End
         Unemployment Rate to be ON or ABOVE 4.5%


     Probabilistic elicitation
        Much more optimism about the possibility to elicit
         probabilistic information
        “early” results indicate good predictive power of
         probabilistic information (Nyarko and Schotter, 2002)



6
                            Questions

     To what extent is the superiority of prediction
      markets the result of information that is particularly
      difficult to aggregate?

     Are prices market probabilities? (part II)
        Prices do not equal the mean beliefs of traders (Manski,
         2004)
        Are (betting) budgets statistically independent of beliefs?




7
                        Questions II
     What happens if we take prediction markets
      seriously?
       We could use markets to predict counterfactuals
           How many soldiers will die in a war
           Which division has the greater chance of success?
       Payoff now depends on the likelihood of predicting the
        right outcome plus the likelihood that trading changes
        economic consequences
     Manipulation
       Little historic evidence (Rhode and Strumpf, 2004)
     Strategic provision of information
       Unexpected accruals and incentive compensation
        (Bergstresser&Phillipon, 2003; Gao&Shrieves, 2003)
8

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:1
posted:2/14/2012
language:
pages:8