This is quite brilliant... I've always felt the issues described in the article w/ regards to political correctness and turning a blind eye to certain legal but immoral practices in the voting system affect poll accuracies, sometimes rather heavily; but never thought to consider the gambling market as an alternative....
Really great read, especially the history of gambling and the elections.
Currenly betfairpredicts.com has Obama's likeliness of a win 92% to McCain's 08% (don't ask how :P). Interestingly enough, fivethirtyeight.com shows similar figures - that's another really good and accurate resource.
538 had a post earlier this year, during the primaries I think, about how In-trade was lagging behind their projections. That seems to make sense to me. Smart betters use all the info at their disposal. 538 is one of the smartest sources around. 538 was also able to highlight curious movement at In-trade well before it became obvious that someone was manipulating the McCain margin with heavy betting.
If I was placing any bet, I'd bet on 538's numbers - even at the state-level, before any betting market. Better to trust many aggregates of randomish samples (polls) over one highly-motivated, and non-random, sample. 538 goes the step further to weight polls based on their historical accuracy and weight samples based on predictions about party ID, demographics, etc. Just a highly tuned model with lots of data to support the tweaks.
It will be interesting to see how close 538, versus the betting markets, came based on electoral votes and percentage win. There's no empirical test of likeliness.
I just had a look at 538 and I'd defintely put my money with betfair or intrade. There is no accounting for things like the 'Bradley effect' or other highly correlated events on 538. I'm sure many bettors do use it as a resource but their judgement is likely to be a lot more subtle and incorporate more factors than the statistical analyses provided by the site. Kind of like the way that humans are still smarter than artificial intelligence (which is currently much like statistical techniques).
> If I was placing any bet
Well you should make a fair bit of money if 538 is accurate. Betfair has had 50K available right now for Obama at 1.08 implying a 92.5% chance which looks pretty good considering 538 says it is 98.1%. If you bet 50K you'd have a probabilistic profit of nearly 3 grand (50K * 1.08 * 98.1%). But when you actually went to put the cold hard cash down I'm sure you would have some second thoughts about 538's models. The whole point of betting markets is that talk is cheap.
> There's no empirical test of likeliness
Yes that is a problem. If betfair was (accurately) pricing in a 10% chance of a significant, accoss-the-board Bradley effect then there would be a 90% chance it would not occur and 538's results would look more accurate (across the electorates). However that 10% chance (if accurate) would be enough to make betting according to betfair's odds profitable vs 538's odds.
In general this is reflective of the 'Black Swan' problem of uncommon events such as a market crashes. Those that underestimate the risk can often make money for long periods of time by betting against it. But when the risk event strikes all those profits and more can be wiped out.
The reason that there is not accounting for the 'Bradley effect' is that it is highly questionable that it ever existed in the first place and its impact has diminished to the point of being statistical noise at this point. Nate has had several great articles on the site showing how the Bradley effect is BS.
The 'Bradley effect' is only one of any number of possible phenomena that may occur and distort the results. Perhaps a new phenomena will occur and it will be named the 'Obama effect'. In 2004 the exit polls showed that Bush had very likely lost but this was wrong. Only after the event did they they came up with a explanation.
I remember checking InTrade sometime in March, and it showed something like a 70%-chance that Hillary Clinton will be president. Guess that the gamblers become more accurate as time to election approaches zero.
Could you provide further explanation, or will you leave this hanging mysteriously in the air, shrouded like a punter knowingly tapping his nose?
It seems cut and dry the way the article is written: Intrade/Betfair correctly predicted 50 states. Are you going to try to make some money tomorrow night, or is there another reason why you won't give up your closely-held betting expertise? :-)
I don't think I can do it tomorrow because the election isn't close enough. But yeah, I think it's pretty exploitable when they are.
People say that stock markets are efficient, but then the last couple months show that the money there is just as prone to groupthink as anywhere else. The same is true of Intrade. More true probably due to betting/deposit limitations.
You are using the techical description of "efficient markets" right? (i.e. perfectly accurate odds). Of course by this standard you are bound to be correct. However I'm sure you realized that by participating in and 'beating' the market you are increasing its accuracy.
The important question is 'how useful or accurate are these markets compared to anything else?'. I'd challenge you to find a more accurate predictive indicator that is publicly available (even if not for free).
fivethirtyeight.com may have a more reliable method. Time will tell. An aggregate of polls with solid math applied to it might just win by having more data.
Now, if there were a US-legal prediction market, that would be something.
I think it is unlikely that fivethirtyeight.com would be more reliable. The human minds are still more subtle than statistical methods alone. However even if 538 did prove more accurate its results would soon be incorporated into the market by betters. The reverse is not so likely to occur.
> Now, if there were a US-legal prediction market, that would be something.
You would have to think that involving US citizens would improve the market's accuracy. It's ironic that US-citizens can't bet.
Interestingly, a friend a couple months ago was telling me about ultra ETFs. They use leverage to ensure (quite reliably so far, though they're new) a daily return of double the index they're based on. He specifically recommended one that is short the S&P.
I haven't talked to him since, but I suspect he's buying a beach house ATM.
Yeah, over time an ultra etf should make 2x what the s&p gains (and a short one would lose 2x). In the short term though that's a hell of a profit (on the ultra shorts) when things go the way they have.
Really great read, especially the history of gambling and the elections.
Currenly betfairpredicts.com has Obama's likeliness of a win 92% to McCain's 08% (don't ask how :P). Interestingly enough, fivethirtyeight.com shows similar figures - that's another really good and accurate resource.