Understanding pricing anomalies in prediction and betting markets with informed traders Peter Norman Sørensen Økonomi, KU GetFIT, February 2012 Peter Norman Sørensen (Økonomi, KU) Prediction Markets GetFIT, February 2012 1 / 18
Prediction markets Definition of prediction markets: Trading mechanisms that produce probability forecasts by aggregating expectations of traders Applications to business/public policy decision making Iowa Political Stock Markets Economic Derivatives at Chicago Mercantile Exchange In a complete asset market, can interpret price as expected (discounted) cash flow Can invert and extract market s risk-neutral belief Prediction markets are special: not intended for risk-sharing trade based mostly on difference in beliefs/information verifiable outcome realization Peter Norman Sørensen (Økonomi, KU) Prediction Markets GetFIT, February 2012 2 / 18
Prediction and betting markets Particularly simple financial markets Laboratory for testing economic theories: of decision making under uncertainty of price formation & market micro-structure Economics research may help: to interpret market outcomes/prices to improve market design and trading rules (not today) Peter Norman Sørensen (Økonomi, KU) Prediction Markets GetFIT, February 2012 3 / 18
Outline 1 Some considerations of prediction markets 2 Theories of mispricing in prediction markets 3 Discussion Draws on joint work with Marco Ottaviani Peter Norman Sørensen (Økonomi, KU) Prediction Markets GetFIT, February 2012 4 / 18
Why markets? We could ask individuals to express individual recommendations or estimates, and then combine those Herding may arise if we do so openly Experiments suggest that sequential information aggregation is hard How to combine? How do we get individuals to express their uncertainty? Experiments suggest markets aggregate information better Forsythe and Lundholm (1990), Palfrey and Wang (2009) Why do people participate? Recreation, gambling, showing expertise Peter Norman Sørensen (Økonomi, KU) Prediction Markets GetFIT, February 2012 5 / 18
General considerations (open) Benefits of market in an organization: Potential to aggregate dispersed information Directly delivers a consensus probability estimate Alternative to conversations Involves all, anonymous, incentivised When repeated, greater weight to better forecasters Some potential challenges: Lack of liquidity, continued interest (subsidies) Market price may affect outcome; clear definition of assets Reasons behind probability may not be communicated Incentive to manipulate decision Rewarding insiders Do traders trust anonymity? Peter Norman Sørensen (Økonomi, KU) Prediction Markets GetFIT, February 2012 6 / 18
Special consideration: Favorite-longshot bias Special case: predict probabilities of sports outcomes Lots of data available Empirical literature on racetrack betting Group horses by market probability Compute empirical probability of winning in races for group Basic empirical finding market prob > empirical prob for longshots market prob < empirical prob for favorites Market risk-neutral belief needs not be calibrated to empirical chance Peter Norman Sørensen (Økonomi, KU) Prediction Markets GetFIT, February 2012 7 / 18
Snowberg and Wolfers (JPE, 2010) Peter Norman Sørensen (Økonomi, KU) Prediction Markets GetFIT, February 2012 8 / 18
Outline 1 Some considerations of prediction markets 2 Theories of mispricing in prediction markets 3 Discussion Peter Norman Sørensen (Økonomi, KU) Prediction Markets GetFIT, February 2012 9 / 18
Theory 1: Wealth effect 1 Price reaction to information this is the information designers are interested in extracting 2 When traders have heterogeneous prior beliefs natural given their limited experience with unique underlying event 3 In the presence of wealth effects due to limit on amount invested or due to risk aversion Question: How does market price relate to traders posterior beliefs? Main result: Market price underreacts to information (in the short run) Peter Norman Sørensen (Økonomi, KU) Prediction Markets GetFIT, February 2012 10 / 18
Intuition Favorable information increases asset price Affects terms of trade, implying a wealth effect: Optimistic traders can buy fewer units of asset Pessimistic traders sell more units of asset To equilibrate, the market price adjusts to move pessimists to optimistic side The indifferent trader (that determines equilibrium price) is then someone with more pessimistic prior belief Sum: favorable information raises asset price, but less fast than a posterior probability Longer run: traders with margin constraints may be forced to unwind positions in same direction as new information Initial under-reaction followed by over-reaction Peter Norman Sørensen (Økonomi, KU) Prediction Markets GetFIT, February 2012 11 / 18
Under-reaction: Implications Possible (and simple) explanation for: 1 Favorite-longshot bias observed in betting markets Favorable information leads to a high market price that nevertheless understates posterior probability Opposite for the longshot the market overbets the longshot 2 Post-event drift and momentum observed in financial markets Peter Norman Sørensen (Økonomi, KU) Prediction Markets GetFIT, February 2012 12 / 18
Theory 2: Surprise effect Imagine bets are placed simultaneously in pool Winners share the pool If bets have no information content (as in Lotto), underbet events ( longshots ) give higher payoff: reverse bias results Information contained in bets, revealed after betting ends, can generate favorite-longshot bias Final note: there are many more theories of the FL bias in sports betting Miscalibrated probability beliefs, risk loving bettors, risk-averse bookmakers, transactions costs, adverse selection,... Peter Norman Sørensen (Økonomi, KU) Prediction Markets GetFIT, February 2012 13 / 18
Outline 1 Some considerations of prediction markets 2 Theories of mispricing in prediction markets 3 Discussion Peter Norman Sørensen (Økonomi, KU) Prediction Markets GetFIT, February 2012 14 / 18
Discussion Extent of the bias likely to be affected by factors in the market Consultants on prediction markets may gain experience on translating the market forecast Institutions for trade may matter We presented an optimistic case for information aggregation Hard to test outside laboratory More work is needed on game-theoretic modeling of information aggregation in markets Active finance literature studies price reactions to information Peter Norman Sørensen (Økonomi, KU) Prediction Markets GetFIT, February 2012 15 / 18
Speculation An asset may be over-demanded early on with the view to trade it later Harrison and Kreps (1978) suggest that the greatest optimist buys the asset Expects that in a later period, someone more optimistic will come along Note, there is no underreaction to information once this trader s belief is taken as market belief Palfrey and Wang (2009) imagine that some traders put greater weight on information Who is the greatest optimist changes endogenously when information changes between favoring-disfavoring an asset Asset price overreacts to good news, underreacts to bad news We focused on prior belief heterogeneity, but heterogeneous information processing is natural alternative Peter Norman Sørensen (Økonomi, KU) Prediction Markets GetFIT, February 2012 16 / 18
Conclusion Prediction markets produce an aggregate forecast We uncovered theoretical mechanisms by which the forecast under-reacts to information Empirically observed mispricing potentially compatible with competitive equilibrium! Peter Norman Sørensen (Økonomi, KU) Prediction Markets GetFIT, February 2012 17 / 18
References Works by Marco Ottaviani and me Surprised by the Parimutuel Odds?, AER 2009, 99(5), 2129 2134 "Noise, Information and the Favorite-Longshot Bias in Parimutuel Predictions," AEJ: Microeconomics 2010, 2(1), 58 85 Aggregation of Information and Beliefs: Asset Pricing Lessons from Prediction Markets, 2010, on our webpages "Outcome Manipulation in Corporate Prediction Markets," JEEA 2007, 5(2-3), 554 563 "The Favorite-Longshot Bias: An Overview of the Main Explanations," Handbook of Sports and Lottery Markets 2008, 83 101. Peter Norman Sørensen (Økonomi, KU) Prediction Markets GetFIT, February 2012 18 / 18