Andreas Weigend
Data Mining and E-Business: The Social Data Revolution
Spring 2009
Stanford University

Homework 3: Design a Prediction Market

(Note: Homework 3 is to be done on an individual basis)
Assigned: Tues Apr 28, 2009
Due: 2pm, Mon May 4, 2009
Email pdf to:, If possible, please also bring a hard copy to the class (May 4) for all non-SC{D students.

THANKS: Three great people helped significantly with this homework:
  • Bo Cowgill (took Stats 252 in 2008 and ran the predction markets at Google),
  • Weichong Khor (took Stats 252 in 2007 and implemented the prediciton market then), and
  • Ron Chung (who brought in his experience of participating of how INSEAD taught prediction markets, and oversaw the assignment)

A prediction market is a system in which participants can buy and sell contracts regarding predictions of uncertain events in the future (e.g., who will win the 2009 Berkeley - Stanford Facebook Page contest). Prediction markets are based on the "wisdom of the crowds;" so a successful prediction market is one in which many players actively participate. The transactions in the market should reveal the crowd's expectation of some uncertain event (e.g., probability of that event occurring). As mentioned in class, there are a lot of resources available to post here but we explicitly want you to do some research on your own.

The Assignment:

1. Identify four well-defined/landmark events that a Facebook Page may experience in the future. The goal of this assignment is two-fold: (1) to create a well-defined contract for each of the identified events and (2) to discuss how a prediction market for each event can be successfully implemented. In your homework submission, please address the following bullet points.
  • For each of the events identified, write a clear contract detailing the conditions of the payoff. That is, what determines the payoff? What are the time intervals involved? You will be writing four contracts (1 for each event).
  • How would you set up a prediction market for each event?
    • How do individuals first learn about your prediction market?
    • How do people enter your market?
    • How can you incentivize individuals to actively participate in your market?
    • What information do people have to participate in your market?
    • How do people play? When individuals enter your market, does everyone start off equally with the same number of points or "money"?
    • What barriers would you set up to make sure that your market is tamper resistant?
    • What types of variables should be collected? Note what variables need to be measurable.
    • What tools do you use to setup your market and why?

2. Now, identify four well-defined/landmark decisions that Facebook Page managers may experience in the future. Similar to the process you went through for your prediction markets above, (1) create a well-defined contract for each of the identified decisions and (2) discuss how the prediction market can implemented. [EDIT 5/3] Your decisions should be from the perspective of a Facebook Page manager (eg you have become a Facebook Page manager from HW 1) and should propose decisions you can implement now.
  • For each of the decisions identified, write a clear contract detailing the conditions of the payoff. That is, what determines the payoff? What are the time intervals involved? You will be writing four contracts (1 for each decision).
  • How would you set up a conditional prediction market for Facebook Page manager decisions?
  • More clarifications:
    A "decision market" or "conditional prediction market" is a prediction market whose purpose is to predict what will happen if a certain decision is adopted.

    For example: Suppose Steve Jobs was thinking about lowering the price of iPhones to $100. Jobs calculates that to make up for a smaller profit margin, Apple would have to sell 20 million new iPhones in the first month.

    His dilemma: Would the price drop result in 20 million new sales in the first month? If so, the drop is worth it. If not, it isn't. How can he tell the answer?

    Jobs might consider opening a conditional prediction market, or "decision market." The central question in this market is: "If Apple lowers the price of iPhones to $100, will iPhone sales go above 20 million in the month following the price change?"

    Traders in this market will disagree. Some will bet that the 20M figure is too large of an increase to expect. Others will feel that it is too small to expect, and that the actual increase will be even larger. Some traders will use gut instincts, and others will use statistical methods. Some will poll their friends or use other methods. Eventually the market will settle on a price that will reflect the probability of getting 20M sales if the price is lowered to $100. It's important to recognize that the method of decision making is not important. All sources of information are applicable and can be applied by the individual buying or selling the contract component.

    Lets suppose that probability is 80%. Jobs can move forward with a good amount of confidence that lowering the price of the iPhone will result in more than enough sales to justify the price decrease. This might help him make up his mind. Alternatively, suppose that the probability is 10% -- then he knows that the plan probably isn't worth it, and he should find an alternative besides lowering the price. Or, perhaps the probability is 50% -- at which case, the price change is a gamble.

    Armed with information about the expected outcome, Jobs can now decide if the $100 price drop is worth it. This is how a decision maker can design a market to test the outcomes of hypothetical changes in course.

A Few References:

A few Remarks:
  • This assignment does not involve programming but focuses on design of the market.
  • You will not actually implement your prediction markets. Rather, we want you to focus your energy on developing well thought-out submissions