Share with your friends










Submit

Analytics Magazine

Predictive Markets: Predicting the rise of prediction markets

March/April 2011

CLICK HERE TO GO TO THE DIGITAL VERSION OF THIS ARTICLE

prediction markets

Social analytic tool taps into the “wisdom of crowds.”

Ajit KambilBy Ajit Kambil

In the coming year, executives will once again hear a lot about the value of analytics and its ability to convert corporate data and information from hindsight to insight and then, ideally, to valuable and actionable foresight. Generally, the two most common approaches to generating foresights are: 1. the use of statistics and mathematical models, and 2. the use of machine learning tools such as neural nets and genetic algorithms software to mine existing data and generate predictions. While we may explore these approaches and their business applications to finance in future articles, this one features an emerging third approach to generating predictions: the use of “social analytics” through prediction markets. Instead of fancy statistics and software, prediction markets tap into the “wisdom of crowds” and the capacity of markets to aggregate beliefs to generate predictions. This article explores how organizations can use prediction markets as part of their portfolio of analytics tools.

Prediction Markets

Online prediction markets are built on the principle that markets serve to aggregate the beliefs of multiple traders to generate a forecast. For example, at any given time, a stock price is the aggregate collective belief of the traders of the company’s expected future earnings allocated to the share. Similar to the stock market, which serves to assign a price to the future estimated earnings of a stock, “prediction markets” assign a value to a belief about the future (a prediction).

Intrade.com is an online prediction market that creates markets based on various real-world events such as what movie will win the Oscar, what candidate will be nominated by Republicans for president and where will the DJIA finish the year. Stock futures in February 2010 for the DJIA trading above 10,000 by the end of the year showed a closing price of about $50. With a maximum payout of $100, this suggests that traders believed there was approximately a 50 percent chance, as of Feb. 12, 2010, that the DJIA would end 2010 above 10,000. Similarly, there was a 17 percent probability expectation that the Dow would close below 6,500. Looking at the historical trade data, it is clear that market fears of a meltdown below 6,500 had progressively declined from the previous year.

These simple examples illustrate three valued characteristics of prediction markets versus many other forecasting approaches. First, the trading price of stocks on future events, which pay out $100 if the event occurs and nothing if the event doesn’t occur, provides a reasonable estimate of what the traders in aggregate believe to be the probability of the event. Second, prediction markets allow dispersed sources of information and insight to be aggregated through the market process to generate a composite market forecast. In essence, one can tap into a vast pool of intelligence and information dispersed among many diverse traders to generate a forecast. Third, market traders can continuously adjust the price and hence the probability of the event conditioned to the new market information. This creates a dynamic forecasting and near real-time application that attempts to incorporate all information among traders. Thus, dramatic shifts in price and volume can trigger decision-makers to further investigate or respond to specific issues that seem to underlie price changes.

For example, if the price and trading in a stock “New China Plant Begins Production in June” suddenly drops in value in April, it may be prudent to further investigate whether there are problems in meeting the scheduled start date of the plant. This may be followed by policy actions to remedy delays or to alert investors and analysts about expected delays. How accurate are these markets?

Academic studies show that the performance of prediction markets versus polling on election forecasts is as good as or better than polls [1]. Reviews of various prediction markets by Justin Wolfers and Eric Zitzewitz also show that prediction markets perform as well as or better than other forecasting techniques [2]. Prediction markets have also been used in diverse corporate settings and generate useful forecasts to motivate their continued use, but assessments compared to other methods are rarely available to the public.

Nevertheless, the press to date suggests they are useful for sales forecasting and to forecast the completion times of various projects. Indeed these tools have been deployed at companies such as Google, Best Buy, Motorola and Corning for various forecasting applications. While we want to tap into the “wisdom of crowds,” events such as the Tulip mania of 1637 and the South Seas Bubble of 1711 sometimes illustrate the “madness of mobs” in markets. Therefore, market information and forecasts should always be critically reviewed and evaluated.

How Organizations Can Use Prediction Markets

Organizations can use prediction markets to reduce uncertainty. Begin by considering the greatest areas of uncertainty that affect your organization. Is it the sales forecasts in a particular business segment? Is it the differences in sales across regions? Is it the cost of a critical resource such as oil? Is it the timely completion of a particular project? Is it uncertainly about whether a project will be within budget and the variance if it is not?

For each of these types of questions we can generate prediction markets within an organization where the employees trade on different “prediction stocks.” The value of this trading is that it creates valuable signals for management that include the probability of different outcomes. It can alert management to take action to remedy potential problems such as a project potentially going off track from a timing and budget perspective.

Typically, the most important forecasts to financial analysts and investors are expected earnings per share. However, forecasting the next quarter or year’s earnings per share may not necessarily be a good application of these markets. A predictive market on that indicator could risk a large group of insider participants having access to privileged information and beliefs before analysts and public investors, creating a risk of insider trading or possibly worse.

However, organizations may use these markets to generate forecasts of variables that can potentially be utilized in other forecasting models. Thus, a prediction market generating sales forecasts in a region or business segment may provide useful input into broader earnings per share forecasting models. The other benefit executives can gain from these markets is a sense of uncertainty and volatility in specific variables that may motivate them to take steps to better manage or reduce the uncertainty. These may include entering into long-term sales contracts to reduce sales or price variability in a specific region.

Issues to Consider

The first issue prediction market adopters should consider is the value of resolving a particular uncertainty. Can, for example, knowing the delay of a project enable cost savings or other benefits? Can having better sales forecasts enable the company to confidently pay down debt? Rank-ordering uncertainties that need to be resolved, based on the value of resolution, identify a priority list of potential prediction market applications.

The second issue to consider is whether prediction markets are the right tool to address a particular uncertainty. If the item to be forecasted has, or known drivers that impact the forecast have, a time series of historical data available, it may be easier to apply other forecasting techniques such as the use of statistics or machine learning. These techniques do not require the participation of many individuals in the process – unlike a prediction market that requires many traders. Prediction markets are especially suited to situations where sparse data is otherwise available that may be used to define a forecasting model.

The third issue to consider is the “cost” of prediction markets, as it is a social analytic approach. Prediction markets require multiple participants, and the individuals using them require multiple traders be recruited and incentivized to participate from within the firm. Fortunately, the traders do not have to trade with real money, and the trading of points that may then be converted to a reward, such as a vacation or an iPod, seems to generate forecasts on par with trading money. The true cost or benefit, in some cases, is the transparency of such a market forecasting system. It makes areas of uncertainty, and issues such as projects going off track, more visible to a wide variety of participants. Again, organizations using these markets need to decide if the organization desires and benefits from the transparency that these markets drive.

The fourth issue to consider is the availability of the right expertise to design and deploy these markets. The technology is easy and widely available. The design of a program with incentives, recruitment of participants, good forecasting questions and alignment to a company’s culture may not be easy to achieve. While many prediction market vendors are probably too ready to sell their market and technology as a solution, what is really important is their capacity to support the organizational acceptance of the technology.

Conclusions

Prediction markets are emerging as a valuable forecasting tool in diverse application areas from sales forecasts to project success. This social analytics strategy could potentially help resolve a number of business uncertainties, especially where prior data may be sparse or the situation is so unique that other forecasting tools are less useful. While there are a number of design considerations in these markets, the most critical is whether you want to engage in having multiple participants share in a market process to resolve the uncertainty. While the technology for markets is easy to implement, aggregating a number of traders and involving them in a market process to resolve uncertainty in an organization is more difficult. Thus, the application of these methods to forecasting requires broader leadership and an effective process for organizational buy-in [3].

Ajit Kambil (akambil@deloitte.com) is the global research director of Deloitte CFO Services program and has advised numerous companies on technology, eCommerce and strategy issues. His management research focuses on harnessing the value of emerging technologies to improve business performance. His technical research focuses on combining human and machine intelligence to improve information retrieval and knowledge management in organizations. Kambil holds a Ph.D. in information systems and three other degrees from the Massachusetts Institute of Technology. For more information about Deloitte Analytics, visit www.deloitte.com/analytics.

REFERENCES

  1. J.E. Berg, F.D. Nelson and T.A. Rietz,  “Prediction Market Accuracy in the Long Run,” International Journal of Forecasting, Vol. 24, No. 2, pp. 285-300, April-June 2008.
  2. Justin Wolfers and Eric Zitzewitz, “Prediction Markets,” Journal of Economic Perspectives, Vol. 18, No. 2, Spring 2004; and Wolfers and Zitzewitz, “Interpreting Prediction Market Prices as Probabilities,” working paper, January 2007.
  3. Ajit Kambil and Eric van Heck, “Making Markets: How Firms Can Design and Profit from Online Markets and Exchanges,” chapters 3 and 6, Harvard Business School Press, June 2002.

CLICK HERE TO GO TO THE DIGITAL VERSION OF THIS ARTICLE

Related Articles

Headlines

ITIF: Expand computer science education to keep up with demand

Computer skills are in high demand among employers in a wide range of industries, not just tech-related fields, yet despite growing interest in the subject, a new report from the Information Technology and Innovation Foundation (ITIF) finds that too few U.S. students are taking quality computer science classes at the high school and university levels. ITIF makes the case for public action to support and maintain the groundswell of interest in computer science and capture the economic and social benefits that will come from fostering a more highly skilled workforce. Read more →

Report outlines benefits of data-driven customer experiences

According to a new report by Forbes Insights and SAS, “Data Elevates the Customer Experience: New Ways of Discovering and Applying Customer Insights,” the benefits of brands evolving to data-driven customer experiences (data-driven CX) are wide-ranging. The study notes data-driven methods enhance revenue generation and enable cost reduction, as well as accelerate process efficiencies and quality improvements. Read more →

Debunking myths about harm from artificial intelligence

Artificial intelligence (AI) holds great promise for economic growth and social progress, but pervasive, inaccurate myths about hypothetical harms could encourage policymakers to retard further innovation in the technology, according to a new report from the Information Technology and Innovation Foundation (ITIF). ITIF, a leading tech policy think tank, makes the case that policymakers should actively support further development and use of artificial intelligence if we want society to reap the myriad benefits it has to offer. Read more →

Report: Technology rarely used to detect fraud

Technology is an important tool to help companies fight fraud, but many are not succeeding in using data analytics as a primary tool for fraud detection. Meanwhile, fraudsters are leveraging technology to perpetrate fraud, according to a new report by KPMG International, “Global Profiles of the Fraudster.” Read more →

CAP® EXAM SCHEDULE

CAP® Exam computer-based testing sites are available in 700 locations worldwide. Take the exam close to home and on your schedule:


 
For more information, go to 
https://www.certifiedanalytics.org.