Googles Bo Cowgill takes a swipe at the prediction market software vendors.

No Gravatar

Using Prediction Markets to Track Information Flows: Evidence from Google – (PDF file – PDF file) – by Bo Cowgill, Justin Wolfers, and Eric Zitwewitz – 2008-01-06

Bo Cowgill:

[&#8230-] Trade-by-trade data can reveal characteristics of specific working groups: What they know, how they feel, how they process and share information and how all of that changes over time. I didn&#8217-t try to put any of this in the paper because the conclusions would be sensitive, and I thought this application was pretty obvious to anybody who understood our methodology. [&#8230-]

Bo Cowgill:

I&#8217-ve also heard that other companies would find it impossible to analyze the interaction between their market and the organization. Why? Lack of data. [&#8230-]

Bo Cowgill:

Some more remarks about applications that combine prediction markets and organizational data (org charts, social networks, seating locations). The obstacle to these applications is not a lack of data. Jed mentions privacy concerns &#8212- and if he thinks this is a big obstacle then I&#8217-d be interested in discussing his thoughts.

A bigger problem is that that current prediction market vendors and consultants cannot support these applications. At heart, these vendors are software engineers and salespeople at heart, not statisticians or data miners. They want to write one system that can support lots of clients. At conferences, one hears PM vendors complain about having to do &#8220-customization&#8221- work for clients.

This approach would not work for the applications I describe for two reasons:

  1. The inputs for different clients won&#8217-t be the same. Each client&#8217-s organizational data will likely take a different structure. This makes it difficult for prediction market vendors to architect a single system that can served many clients (yet another challenge with integrating markets with other corporate IT services).
  2. The outputs for different clients won&#8217-t be the same. The business relevance and statistical power of each analysis will differ with each client&#8217-s data.

Prediction market vendors may also need to familiarize themselves with the statistical learning methods necessary to fully utilize these rich datasets. So what&#8217-s the solution? First, move to a software-and-consulting model. By &#8216-consulting,&#8217- I don&#8217-t mean &#8216-consulting on how to implement the market.&#8217- I&#8217-m talking about helping the client solve its problem using a variety of data, including prediction market data.

Second, the vendors also need to pitch prediction markets as more than a forecasting tool. People in the business world commonly identify as data junkies &#8212- probably more so than they identify with the &#8216-wisdom of crowds&#8217- ethos. It is unclear how much companies really care about accurate forecasting anyway.

On a related note, there is something that only the prediction market software vendors could do, at this time, for those who are in capacity to do so: setting up inter-industry prediction markets &#8212-or at least, handing over (with everybody&#8217-s agreement) anonymized prediction market data on industry topics to anyone else in the industry who is a client of that PM firm. I don&#8217-t know about NewsFutures or Inkling Markets, but if you look at Consensus Point&#8217-s list of clients, you&#8217-ll see that David Perry&#8217-s firm is strong in the (consumer) electronic industry &#8212-Motorola, Qualcomm, Siemens, Nokia. Use your imagination, or ask David Perry directly, for more&#8230- (I can&#8217-t talk- otherwise, next thing, I&#8217-m a dead blogger.)&#8230-

Previous blog posts by Chris F. Masse:

  • Last year’s best April Fool’s Day Joke had something to do with the Wisdom Of Crowds.
  • Will HedgeStreet USA, the hypothetical InTrade USA, and the hypothetical TradeFair USA, be regulated in the future by a merged SEC+CFTC regulatory structure?
  • WORST THAN ELIOT SPITZER (if it were possible): Formula One boss, Max Mosley, had sado-masochist sex with 5 prostitutes, for 5 hours (!!), reenacting a concentration camp scene (!!) in which he played the role of both Nazi guard and inmate.
  • Is BetFair Poker a booby trap for the gullible novices? Does The Sporting Exchange (the operator of the BetFair brands) help gangs plucking down innocent recreational poker players?? To get an inkling, don’t read The Guardian, seeded by the BetFair spin doctor- read Midas Oracle.
  • The video that the technologically retarded BetFair spin doctor should watch.

Googles Bo Cowgill @ 2008 DIG Conference

No Gravatar2008 DIG Conference: Leveraging Information for Breakthrough Business Performance – (featuring Google&#8217-s Bo &#8220-Grandizer&#8221- Cowgill) – @ Las Vegas, Nevada, U.S.A. – 2008-05-13~15


RELATED NOTE: I&#8217-m preparing a long post about the Cowgill/Wolfers/Zitzewitz paper&#8230- If you have positive or negative tips, contact me today&#8230-

Using Prediction Markets to Track Information Flows: Evidence from Google – (PDF file – PDF file) – by Bo Cowgill, Justin Wolfers, and Eric Zitwewitz – 2008-01-06

Previous blog posts by Chris F. Masse:

  • How Decision Markets Work (with emphasis on InTrade) – by Robin Hanson
  • MIDAS ORACLE POWER: People googling about BetFair’s new bet-matching logic are automatically directed to our group blog. BetFair’s SEO can return to the locker room.
  • David Pennock, a respected expert in prediction markets and market design, discusses some aspects of BetFair’s new bet-matching logic.
  • Inkling Markets has automated the making of the contractual agreement with its prospects. So, now, if you don’t want to, you can avoid talking to Adam “The Shark” Siegel.
  • Topic: “prediction markets” – MyBlogLog

Enterprise Prediction Markets … Without Office Politics

No Gravatar

Google&#8217-s Bo Cowgill, as reported by IT Business:

[&#8230-] If you let people bet on things anonymously, they will tell you what they really believe because they have money at stake. This is a conversation that&#8217-s happening without politics. Nobody knows who each other is, and nobody has any incentive to kiss up. [&#8230-]

[In one example,] the market was predicting that [a project] was behind. A manager says what&#8217-s going on here, &#8230-starts investigating and finds some glitches. [&#8230-]

Read the previous blog posts by Chris F. Masse:

  • Ratted by his bank, sex-addict New York governor Eliot Spitzer (alias “Client 9”) resigns.
  • BBC’s coverage of politics is dull like taxes, death and German sausages.
  • Never talk when you can nod, and never nod when you can wink, and never write an e-mail because it’s death. You’re giving prosecutors all the evidence we need.
  • Is Justin Wolfers a libertarian? Probably not.
  • The information technology that caught Eliot Spitzer
  • Eric Zitzewitz’s 10 minutes of fame
  • Fun with conditional probabilities


No Gravatar

Our testosteroned Grandizer [JOKE] is trapped in communist China:

Sadly, I&#8217-m not able to read the announcement on the Google Analytics blog thanks to the Great Firewall. I&#8217-m in Shanghai until the end of the month with Google.

Great Wall


Google Analytics blog post:

Benchmarking now available plus additional opt-in settings
Wednesday, March 05, 2008

We&#8217-re happy to announce the launch of two related new Google Analytics features: a beta version of industry benchmarking and a data-sharing settings page. Both are designed to give our customers more choice and better control over their data. We are also launching an integration with Audio Ads today, which we&#8217-ll discuss in more detail in tomorrow&#8217-s post. All of these features will begin appearing in customer accounts today, though benchmarking reports may take up to a couple weeks to show data.

Industry benchmarking is a commonly requested new service that enables customers to see how their site data compares to sites in any available industry vertical. We believe this data will provide actionable insights by providing context for users to understand how their site is doing. For example, if you have a travel website and you get a spike in traffic on Mondays, you may want to know whether other travel sites get that same spike on Mondays.

You can also compare your site against an industry vertical different than your own. For example, you might see that your industry&#8217-s traffic dips at certain times of the year while another industry&#8217-s traffic increases. Based on that information, you may wish to explore cross promotional opportunities to drive traffic back and forth.

[Link to their screen shot.]

For more information, take a look at the benchmarking FAQs in the Google Analytics help center.

Of course, benchmarking only works if people can opt to share their data into the system, so we&#8217-re also introducing a new data-sharing settings page. On this page, customers can choose whether to opt in or opt out of sharing their Analytics data. To be clear, we are not sharing individual data with competitors- we bucket data into industry verticals and then anonymize and aggregate the data. Once you opt in, it may take a couple weeks for the reports to populate.

You can also elect to share your data with other Google services. This setting will allow us to provide you with additional advanced new features. For example, many of you have asked us to integrate Conversion Optimizer (which is currently only available to AdWords Conversion Tracking users) into Google Analytics. By opting to share your data with Google, you&#8217-ll be able to take advantage of these related new features as they become available. For more information, take a look at the data-sharing options FAQs in the help center.

Look for tomorrow&#8217-s post on the Audio Ads integration.

Posted by Brett Crosby, Group Manager, Google Analytics

Read the previous blog posts by Chris F. Masse:

  • Is Justin Wolfers a libertarian? Probably not.
  • The information technology that caught Eliot Spitzer
  • Eric Zitzewitz’s 10 minutes of fame
  • Fun with conditional probabilities
  • Wrongly Crafted Headlines Of The Day
  • an American, petite, very pretty brunette, 5 feet 5 inches, and 105 pounds
  • Mississippi: Is it a primary or a caucus?

Prediction market sessions of the OReilly Money-Tech Conference suffer fatally from the absence of the worlds most knowledgeable, most innovative and most trustworthy prediction market expert.

No Gravatar

O&#8217-Reilly Money-Tech Conference – 2008-02-06~07

Predicting the Future of Prediction Markets + Google as Prediction Market

Wharton&#8217-s Justin Wolfers, Google&#8217-s Bo Cowgill, Inkling&#8217-s Adam Siegel, and Sean Park (representing Himself).

No more Robin Hanson. :(

Better to stay home watching a re-play of the December 2006&#8217-s Yahoo! Confab, where Robin Hanson does appear.

Confab Yahoo! on prediction markets – Streaming Video: 100k300k – 2006-12-13


UPDATE: Robin Hanson comments&#8230-

I was invited, but the date conflicted with a SETI conference I&#8217-ll be speaking at.

Predictocracy: Market Mechanisms for Public and Private Decisionmaking – THE MARKET WEB

Predictocracy: Market Mechanisms for Public and Private Decisionmaking &#8211- by Michael Abramowicz &#8211- 2007-xx-xx &#8211- (fall)

Chapter: The Market Web &#8211- (towards the end of the book)


Michael Abramowicz:

If prediction markets should become commonplace, decisionmakers might link to them in their own analyses.

Will trading play-money and/or real-money event derivative contracts become commonplace? It&#8217-s likely, at the contrary, that trading will remain an elite occupation and that prediction markets with appropriate liquidity will remain scarce. Unless Google, Yahoo! (with Yootopia) and/or MicroSoft has/have a secret plan to popularize betting exchanges &#8212-which could well be since Bo Cowgill, David Pennock and Todd Proebsting are ambitious guys.


Michael Abramowicz:

For example, suppose that a corporation is deciding whether to build a new factory in a particular area. That decision might depend on variables like future interest rates and geographic patterns. And so, a decisionmaker might build a spreadsheet containing live links to prediction markets assessing these issues.

Interest rate prediction markets would help, for sure. As for geographic forecasting, maybe non-trading mechanisms could help &#8212-for real estate, I&#8217-m thinking of Zillow, or some improved mechanisms derived on Zillow.


Michael Abramowicz:

The Market Web

If prediction markets should become commonplace, decisionmakers might link to them in their own analyses. For example, suppose that a corporation is deciding whether to build a new factory in a particular area. That decision might depend on variables like future interest rates and geographic patterns. And so, a decisionmaker might build a spreadsheet containing live links to prediction markets assessing these issues. That way, as the market predictions change, the spreadsheet&#8217-s bottom line would change as well. Predictions in many prediction markets may be interrelated, and so market participants in one prediction market will often have incentives to take into account developments in other prediction markets. Prediction markets thus can affect one another indirectly, as participants in one update their models based on developments in another.

Sometimes, however, it might be desirable to construct links among prediction markets so that changes in one automatically lead to changes in another. Consider, for example, the possibility of a market-based alternative to class action litigation. In Chapter 8, each adjudicated case represented a separate prediction market, but often there will be issues in common across cases. Many thousands of cases may depend in part on some common factual issues, as well as on some distinct issues. Legal issues also may be the same or different across cases. Someone who improves the analysis of any common factual or legal issue can thus profit on that only by changing predictions in a very large number of cases. A better system might allow someone to make a change across a single market and have that change propagate automatically to individual cases.

The critical step needed to facilitate creation of the market web is to allow a market participant to propose a mathematical formula to be used for some particular prediction market. Some of the variables in that formula could be references to other, sometimes new, prediction markets. For example, a market participant might propose in a market determining how much amages the plaintiff should receive a formula dependent on variables such as the probability that the plaintiff states a cause of action, the probability that the plaintiff was in fact injured, the probability given injury that the defendant caused the injury, the probability given a cause of action that the defendant is subject to strict liability, the probability given no strict liability that the defendant was negligent, and the damages that the plaintiff should be awarded if liability is proved. This formula, for example, presumably would allow for no damages where the plaintiff probably does not state a cause of action. Each of the components of this formula might be assessed with a separate prediction market. We can easily build the market web by combining three existing tools. The first tool is a text-authoring market. The relevant text would be the formula itself, including specifications of other prediction markets that would be used to calculate specific variables. As with any text-authoring market, a timing market would determine when a proposal to change the text should be resolved. Other markets might become live only once proposals to take them into account were approved. Ex post decisionmakers would assess the wisdom of these markets&#8217- recommendations in some fraction of cases to discipline the market&#8217-s functioning.

The second tool would be a simple normative prediction market corresponding to the text-authoring market. It might also be possible to have computer software that automatically parses the formula and consults various sources, but the market sponsor need not build this tool. Rather, ex post decisionmakers will assess the appropriate value for the normative prediction market based on the formula. An advantage of this approach is that it would make it easy to use complicated formulas, as well as formulas that depend in part on numbers from sources other than prediction markets, or from prediction markets of other types. In addition, this approach makes it easy to collapse a formula into a single prediction market, if that should prove desirable. The formula text simply would be changed to a description of the market to be created, such as &#8220-adjudication of plaintiff&#8217-s liability in a particular case.&#8221-

Finally, the third tool necessary is a mechanism for determining the market subsidy. A separate subsidy would be needed for the text-authoring market and the normative prediction market. Each of these subsidies could be determined by additional normative prediction markets, perhaps with fixed subsidies. The subsidy for the text-authoring market in turn would be distributed by the text-authoring market to individuals who have proposed particular amendments, and individuals who have participated in the assessment of particular amendments. The text-authoring market also could allocate a subsidy to the first individual who creates the market and proposes some text for it. When the text-authoring market produces a new formula reflecting additional prediction markets, the subsidy for the main prediction market would fall (since calculating a formula based on other prediction markets will often be relatively easy).

A single node in the market web would thus consist of a text-authoring market describing the node and providing a formula for calculating it, a normative prediction market, and a set of additional prediction markets for determining how to distribute a subsidy to the different components of the node. The nodes collectively create a web because the formulas link to other nodes- software, of course, could easily make these links clickable. At the same time, a mechanism is needed to determine what portion of the market subsidy each node should receive. A simple approach would be for a prediction market to be used for every link, to determine the portion of the subsidy for each node that should be allocated to each node linked to it. The total should add up to less than 1, leaving some portion of the subsidy for the node itself.

With these markets established, software could easily distribute a single subsidy for the market as a whole to market participants who have traded on individual nodes when the market closes. Market participants working on one portion of the web, meanwhile, would not have to assess the relative importance of one node to nodes that are only distantly related. It would also be straightforward to have a continuously open market, periodically collecting and distributing money in accordance with individual participants&#8217- success on the market.

This assumes that the market web would be arranged on a single server. It is possible, though, that a node on one market web might link to a node on another market web. If market sponsors allowed such links, it could promote competition among prediction market providers. It also partially answers one potential criticism of using prediction markets for decisionmaking, that a software engineer might hijack the government by faking some prediction market results. Market participants at least will have incentives to identify fake prediction markets and not link to them. In principle, it is possible to have government decisions based entirely on decentralized prediction markets. A caveat is that the government might want to subsidized individual market web providers, and it might use centralized prediction markets to accomplish that.

Whether or not the markets themselves are decentralized, they would allow market participants to make it easier to assess the basis for market predictions. Indeed, the market web is in some ways a substitute for deliberative prediction markets, because both provide means of helping observers understand the basis for the market&#8217-s predictions. An observer could look at any individual node of the market web and understand how it has been calculated, though inevitably there must be some &#8220-leaf&#8221- nodes that themselves do not contain any formulas. At the same time, software might allow an observer to find all of the nodes that link to a particular node. So a market participant addressing a factual issue relevant to many cases could link to all of the cases represented by that factual issue. As a particular issue becomes increasingly important, the subsidy for that node should rise, and market participants can profit on their analysis of the issues relevant to that node without worrying about details of individual cases.


Brainy stuff. I&#8217-ll mind this for a while. I&#8217-m sure that the Midas Oracle readers will find this idea original &#8212-and maybe, interesting.