For the curious

At about 8:30 a.m., is giving the Republicans a 13 to 15 percent chance of Senate control.  The Webb-Allen race, described by newspapers as "too close to call," is being called for Webb; Allen’s chance is about five percent.

Addendum: Chris Masse reports: if Virginia and Montana go Democratic, as of morning yesterday, the prediction markets called every race correctly.


a win for libertarians! of course they are not taken actual leadership positons, but as much as could reasonably be hoped. The international community has also been sent a message that the US no longer whole heartedly backs torture. Perhaps not as lofty an issue as we may have wished, but we have to start somewhere.

How is a victory for lockstep Democrats a win for libertarians, theCoach?

Tyler, the prediction markets were way off. They only started giving the Senate to the Democrats after the Dems had basically won. It's like saying I was correct in calling the game for the Redskins but doing so only after that crazy play with no time left that setup a second game winning field goal.

The volatility before closing is a feature, not a bug; it reflects the prediction markets assimilating new information as it comes in from exit polls, partial returns, etc. IMHO, the prediction markets were much better and faster than the major media at calling various races last night.

But yes, calling this accurate prediction is highly misleading. it's like saying that a rain gauge on your roof "predicts" weather better than a meteorologist. You need to judge predictions on a clear time frame shorter than "after polls close".

The relevant question is whether the prediction markets got it right before people voted, not after the exit polls, partial returns, etc. started coming out.

I'm surprised Chris stopped masterbating about BetFair long enough to type out a post for you. (Speaking of arbitrage, Betfair still gives nearly even odds. With a rake of 5% and a good 5k on the table, that's a hefty profit.)

The prediction markets weren't crazy. Remember that for the Democrats to take the Senate, they had to win a whole bunch of individual races. Even if the Dems were favored by, say, 70% in each of the five races, the odds that they'd win all five was under 17%, assuming that the outcome of each race was an independent event.

Interesting observation Anonymous. Did I mention I got a C in stats/probability? Now I remember why.

Its easy to be right when you pick both sides. Yesterday, tradesports showed the GOP keeping control of the senate, but the democrats winning the individual races necessary to take away control.

if Virginia and Montana go Democratic, as of morning yesterday, the prediction markets called every race correctly -- which is to say, the lower-value contracts were probably overpriced and the higher-value ones underpriced. If predication markets are right all of the time, then they are underpricing variance.

All the explanations of how the Dems can be favored in every race but not to take the overall Senate ignore the fact that there is a very high correlation in the results. In some sense, the big unknown before the election was the degree to which people would vote locally (uncorrelated) or vote nationally (a few national issues on which opinion swings are highly correlated across districts). Tradesports et al clearly got the correlation structure wrong.

I wonder if the prediction markets factor in possible voter fraud into their models. I think they must. It would also be interesting to see how much of a difference it makes. I think it would make a pretty big difference since we have elections that are being run so tight. The race in Montana had a difference of how many votes between the democrats and the republics? I think it was about 1,200. So just goes to show that small amounts of voter fraud can swing elections. On a side note, The state of our election process is pretty dire as I see it. The mechanisms that we use to vote can't really be trusted. This problem needs to be solved. Don't get me wrong though, I'm happy with control being distributed amongst the parties. However, for future elections, it would be nice if failsafe voting systems were installed so that there won't be a question regarding accuracy and count.

Hello Professor Tyler Cowen and all the commenters,

#1. Professor Lance Fortnow made a specific point: the TradeSports's prediction markets for the individual 2006 US Senate races were accurate (provided that Virginia and Montana go democratic).

#2. Professor Lance Fortnow DID NOT SAY that the TradeSports's prediction market for the control of the US Senate was accurate. Please, don't put words in his mouth.

#3. Analysis reports from economists and statisticians are coming, but, please give them time to digest the data... once the dust has settled.

Thanks for your attention,

Chris Masse

Also, re: time horizon. While it's true that markets react to new information, shouldn't the price reflect how close something is? I don't know stats or options pricing well, but to me a 70% probability suggests a very strong belief compared to, say, 53%. There was no major news yeterday other than the Dems were winning - no new scandlas, not even a hint that Rumsfeld would resign. So why would the prices be so out of whack in the first place?

I do not understand the justifications for getting the overall senate race incorrect while predicting the individual races correctly.

Shouldn't any correlation and probability summing of independent events be factored into the price of the overall contract. Shouldn't ALL information, even that which is difficult to obtain, be factored into the overall senate control contract?

It appears to me that the markets just plain old got it wrong.

Some libertarians understand probability. You are no more numerate, to generalize, than conservatives or big government American liberals.

The assertion that the "markets" made an earlier call then the medeia is not a fair comparision. The media years ago agreed to not announce their predictions before the polls closed. Too many West Coast people complained that the predictions being released before West Coast polls closed discouraged some people from voting. Thus you have one entity bound to not predict before a certain time and the other able to predict whenever they want. Not otherwise a fair comparison.

Comments for this post are closed