Monday, September 21, 2015

Prediction Markets (Rating: 4 out of 5 stars)

Note: This post represents the synthesis of the thoughts, procedures and experiences of others as represented in the  articles read in advance (see previous posts) and the discussion among the students and instructor during the Advanced Analytic Techniques class at Mercyhurst University in September 2015 regarding Prediction Markets as an Analytic Technique specifically. This technique was evaluated based on its overall validity, simplicity, flexibility and its ability to effectively use unstructured data.

Prediction markets are designed and conducted for the primary purpose of aggregating information so that market prices forecast future events. These markets differ from typical, naturally occurring markets in their primary role as a forecasting tool instead of a resource allocation mechanism. For example, hedge fund prediction markets can allow people to stake bets, both real or imaginary, in predicting an outcome.

  • This method allows the development of a consensus estimate/prediction.
  • This method can give participants an incentive to provide accurate and truthful information.
  • This method provides anonymity for its participants.
  • This method accounts for more nuance than random opinion polling.
  • This method allows a percentage bet to occur, thus giving the bet some value instead of only a straight yes/no. You can thus quantify something like “highly likely”.
  • This method provides fairly accurate answers compared to other analytical techniques predicting the same event with a similar participant pool unless they are trying to answer a question in a format other than ‘Yes or No’ format.

  • This method requires a high volume of trading.
  • This method relies on how good the information is that is given.
  • This method works well if the preconditions are set properly.
  • This method is highly accurate only for very narrow questions. It lacks to answer DM’s broad requirements.
  • This method works best with a large group of people.
  • This method does leave room for bias and manipulation amongst participants.

How-To: (Note: there are many ways of conducting a prediction market, the exercise conducted is just one of many and was based on’s method)
  1. Create a market for the buying and selling of predictions on a specific yes/no question.
  2. Over time, as participants buy shares of a particular outcome they become more highly valued and will result in less of a gain if the prediction is correct.
  3. The predictions can be bought and sold amongst participants so that the market accurately reflects the current perception of the overall population.
  4. As the market continues it will close in on the end result with greater accuracy than the best estimate of an individual.
  5. The question must be resolved in a yes or no. The winnings are provided to those who estimated correctly while those who were incorrect lose what they staked.

Personal Application of Technique:
The prediction market exercise consisted of providing estimates on the 2003 NCAA Men’s Basketball tournament.  However, the year and outcomes were concealed from the class until the end of the exercise.

In conducting the exercise, we utilized Google Sheets to allow the analysts to submit their estimates and the percentage which they thought was likely.  They then were given the opportunity to examine the “market,” what other students had estimated, and alter their own estimate.  They submitted estimates on three games, the two semi-final games (Final Four), and the championship game.

After minor changes in the first round, analysts were only allowed to make estimates of 50% or greater, and if correct, would be compensated. We used M&Ms as betting currency and granted compensation with additional candies (Correct estimates: 50% = 0 additional candies; 60% = 1 additional candy; 70% = 2 additional candies; 80% = 3 additional candies; and 90% = 4 additional candies).  Incorrect estimates would forfeit the candies they had put at stake.

The exercise was limited through by a small sample size, an inability to trade/buy/sell other shares, and a simplistic overall scheme which did not reflect a real market.

When employing this version of prediction market, the question asked must be resolvable. In order to use the technique as a teaching tool or as an example as we didn, historically resolved information must be used (like the 2003 NCAA Men’s Basketball tournament). The participants must also not have any knowledge on the data, or means of researching the information.

For additional information:
For the exercise PowerPoint, please email the author.

No comments:

Post a Comment