Prediction market beats midterm-election statistical models

[Emile Servan-Schreiber]

Hi,

With Intrade gone and the rise of sophisticated statistical models à la 538 operated by various U.S. media, we haven't heard much about prediction markets during this US midterms election cycle. It is as if the allure of big data and statistical rock stars like Nate Silver had eclipsed the robust and well-documented success of collective human intelligence. Are prediction markets doomed to be road kill on the big-data super highway?

Not so fast.

There is evidence that prediction markets still have a lot to offer above and beyond the much-hyped statistical machines. A few months ago, Lumenogic launched a new type of prediction market at Hypermind.com. It is an invitation-only venue focusing on politics, geopolitics and business. It is populated by several hundred "elite" prediction traders recruited and rewarded on the basis of merit and performance. Although it operates with play money, it distributes significant cash rewards.

Since June, Hypermind has listed several stocks on the midterm elections in the U.S., focusing on control of the Senate and 5 of the most undecided individual races in Kansas, Iowa, North Carolina, Colorado, and Georgia. The full trading history can be viewed here:

https://hypermind.lumenogic.com/hypermind/app.html?gtp=vitrine&selection=midterms
Password: « data »

This allows comparisons between the market's predictions and those of the 7 major statistical models: Daily Kos, Washington Post, New York Times, FiveThirtyEight, Huffington Post, Princeton Election Consortium, and PredictWise. Without easy access to the full history of predictions from these models, I could only compare the "ultimate" predictions of each model on election day with the latest election-eve trading price on Hypermind. The results are striking: in brier-score comparisons, our  market outperformed all the models but one. The results are detailed in the attached pdf. 

Isn't it neat to realize that even in the age of big data and cold statistics, a modest gathering of human brains can still outperform the machines?

By the way, the full history of Hypermind trading data on those midterm markets is available on request. Just ask.

Cheers!

--Emile

[Emile Servan-Schreiber]

Updated results and analysis of Hypermind vs big-data models are published at the link below. The page also features a link to download the Hypermind data for those interested. 

Hypermind prediction market outpredicts big-data models in the 2014 U.S. Midterm elections

https://hypermind.lumenogic.com/hypermind/app.html?gtp=midterms#midterms