Comments for Prediction Markets vs. Survey Data: Fight!

Prediction Markets vs. Survey Data: Fight!

2008-06-11T17:29:59Z

Prediction markets are speculative markets in which investors put money into ideas that are tied to a future event -- such as "Will the Lakers cover the spread?" The idea is that the more people who buy into an idea, the more likely it is to occur. This is an application of crowd wisdom. One such prediction market, Intrade, which has been in operation since 2001 and has 73,000 traders, is running a number of prediction contracts on the 2008 US presidential election. Are their predictions more or less accurate then good old fashioned phone surveys?

]]> Prediction markets, according to Wikipedia, have generally been found to be "at least as accurate as other institutions predicting the same events with a similar pool of participants." However, prediction markets are also subject to speculative bubbles and potential manipulation. Phone surveys were shown to be all over the map in terms of their predictions in the last presidential election -- so they have potential problems as well. How do these two predictive methods stack up?

Intrade has contracts open for the electoral vote in each member of electoral college in the US. Based on that data, ElectoralMap.net has come up with a predicted map that's updated every couple of days. With 270 electoral votes needed to claim the presidency, the Intrade prediction market currently has Obama winning 289 to 227, with 18 votes up for grabs.

The map at Electoral-Vote.com, which relies on the latest phone survey data, has Obama winning 287 to 227 with 24 votes in a dead heat. Both maps show similar states in play, though the phone surveys tend to show a more nuanced picture of how the country may be planning to vote (note: that could be due to how each site's author is reading the respective data sets).

That said, it is still very early in election season, so you can expect both maps to change between now and November 4th. It will be interesting to see which of these methods is more accurate at ultimately predicting the outcome of the election, though it seems likely that the prediction markets will mirror the phone surveys because that's one of the major sources of information about how each state may vote for the investors.

Comment from Paul Irish on 2008-06-11

2008-06-11T18:56:09Z

FiveThirtyEight.com looks to be the champion of aggregated polling info this time around. His methodology is insanely complicated and documented. "The guy (Nate Silver) who runs it got some notice after being the only source to accurately predict the North Carolina & Indiana Democratic primary results (everyone else all predicted Hillary would do a lot better)."

Comment from Mark Harrison on 2008-06-11

2008-06-11T21:03:27Z

In the whole polls vs. markets debates, there are two separate things going on, which I think are getting conflated here.

1: The more people participate in EITHER type of "forum", the more likely the results are to be representative. ie, big samples beat small samples.

2: The more people have vested in making a correct prediction, the more likely they are to make the prediction correctly.

Because most people don't have any vested interest in polls, they have no incentive to try to work out the answer to the question "Who do you think will win?", so the question pollsters are reduced to asking is "Who will YOU vote for".

With markets, however, because people are putting their money up, they are spending more time trying to work out the answer to the question "Who will win?"... The more people think that, say, Obama, will be the next President, the more money you have to put up to buy an Obama contract / place an Obama bet compared to a McCain (or even Clinton) contract. Because markets are (we hope) generating reasonable "odds" in real time, based on large numbers of "gamblers", the markets should work better.

However, the whole point about markets is that they're only right "long term, on average". Market theory suggests that markets will get SOME predictions DEAD WRONG, and focussing on the ability of markets to predict a single event, even one as important as the US Presidential Election, as many commentators are doing, misses that likelihood...

Comment from Bob Boynton on 2008-06-11

2008-06-12T02:23:07Z

Some economists at the University of Iowa have been running election prediction markets for more than a decade -- they have covered Europe as well as the US. They say the market always beats the survey results -- sometimes more than other times. However, partisans could easily upset this if they wanted to.

Phone surveys are lucky to get a 30% response rate. Luck is about the only thing that keeps them from disaster -- or precision is not very important.