November 1, 2004
WHICH SEEMS ABOUT RIGHT:
The United States 2004 Election Prediction (Stanford Predicts: The 2004 Presidential Election)
Probability of winning presidency based on electoral college only (not popular vote)
Posted by Orrin Judd at November 1, 2004 8:01 PM
Last Updated: Nov 1, 2004, 2:40 pm
Next Updates Expected at: 6:00 pm, 9:00 pm
All the market based indicators, from Fair to tradesports to the bookies, seem to back the incumbent much more than the polls, although it's not correct to equate probability (odds or future prices) with vote %.
The polls indicate W will lose as a tie means the dem ground game in key states will prevail.
It'll be an interesting test.
The cascade of polls showing W an average of 1.5% up on Kerry but failing to break 50% most of the time. I envy anybody with the willpower to ignore them. I cannot.
For amusement, I just did a numerical analysis of the tradesports' presidential contracts by state and get:
Tie (269 to 269 which should go to Bush as well): 1.5%
The polls don't take into account things like fraud, litigation, etc., all of which favor the democrats. The future's contracts take such things into account (theorectically).
If I applied the numerical methods to the Stanford polls, I'd get the same results.
JAB: The last few days have seen multiple polls (Zogby and CNN/Gallup, I think) admitting that they are fiddling with the numbers to get the "right" answer. Because they "know" that the race is tied, they can use this to figure out how the undecideds will break, what partisan breakdown to use, etc. It's nicely circular, and makes their results utterly meaningless.
Bret: if your numbers are good there's a 4.5 point arbitrage (currently 53 ASK) to be had between the 'GWB to win' contract and what you show for all the states. Problem is you'd burn up 3x that much just trying to get it put on. Darn those efficient markets anyway. The burning question for me is how many 300+ EV's contracts to buy. That may require some beer consultation time.
W ticked down today on tradesports, but I think it's volatile. I only recently began checking it compulsively.
John: Bush was at 55% when I did the analysis so it is even more efficient. Even worse, the spread in the States' Bid/Asks are pretty big, so there's no way to make money (I used the average of the Bid/Ask in my analysis as opposed to "Last").
Bret, I agree on the spread and your numbers seem about right. Bush winning OH just took a hit, but who knows.
As of this writing (10:16 Eastern) the Tradesports for Bush has dropped to 52.3, but the IEM Vote Share (share of Dem + Republican, ignoring third parties) really tanked for Bush in what is probably manipulation.
How the heck does one calculate the probability of a single trial to the hundredth of a percent?