Another Monday, another weekly Intrade data round up and it’s pretty much game over.
To start with we’ll run through the daily trackers:
The Obama surge on the daily timescale seems to have peaked – understandably so and we’ll get to the why’s of that in a minute, but on the weekly timescale it’s been a strong push in the Democrat’s direction with the probability change in the State by State markets over the last week running almost universally in the one direction.
If we look at the weekly probability trackers of both the headline market and the State simulation, we can see how this consolidation of win probabilities has really strengthened up over the last few weeks:
Our State market simulation tells the story with the Democrats a whisker below a 90% probability of victory simply as a result of the polarisation of the probabilities of the State markets.
The reason for the State simulation probability surge is pretty simple – decreasing numbers of the States the Democrats are predicted to win on Intrade are staying below the 70% win probability level, while the new States to enter that 30-70 spectrum of contestability are seats once considered to be safely held by the Republicans. As a result, the Democrat Electoral College vote numbers become, literally, pushed upwards.
The projected Electoral College Vote numbers from the simulation tell the story:
Interesting is that the modal value of the simulations seems to have been a bit of a long term leading indicator.
To highlight the “game over” nature of the election campaign, with a party needing 270 Electoral College votes to win the Presidency there are 158 Electoral College votes worth of safe Republican seats, 291 EV’s worth of safe Democrat seats and only 89 EV’s worth of contested seats. If we look at all contestable States where win probabilities are between 30 and 70% and their recent activity we get:
Which shows why Obama is sitting solidly above 350 projected Electoral College Votes. Of those 89 contested EV’s, Obama holds 47 of them above a 65% win probability (making them barely contestable), while being favourite in another 15 EVs (North Carolina).
We can break down the entire State Intrade market results into an Electoral Pendulum similar to the type we’re used to in Australia -although we require an additional axis of information.
The blue columns show each State and the number of Electoral College Votes each state is worth on the left hand axis. Those columns are sorted on the bottom axis from highest Democrat win probability to lowest going from left to right. The numbers below those columns show the cumulative number of electoral college votes for the Democrats after they win each state in order from left to right and finally, the red line read from the right hand side is the Democrat win probability of each state.
What this let’s us do is pick any number of electoral college votes from the bottom, trace it up to the value of the red line and see what the probability is of the Democrats winning the state that would cumulatively represent that many EVs. It’s a bit information dense, but with 3 dimensions of data needing to be represented it’s unfortunately unavoidable.
What’s worth noting about this chart is the slope of the red line across the entire chart. We have two gentle slopes on the far left and far right of the chart literally packed with states, but a steep yet sparsely populated curve in the middle – the middle represents the contested states and shows just how few there are, and just how few there are likely to be between now and election day.
For those of you not fussed by these markets, thinking that Intrade believers need to be visited by Sarah Palin’s pastor to be cured of their witchcraft, here’s a couple of charts you might find interesting. If we take the Pollster.com data used by the Princeton Election Consortium and the Intrade State data here to run a scatterplot of Intrade probability margins vs. State polling margins we get something pretty spiffy.
This chart has every State electoral contest on it except the District of Columbia as Obama has a poll margin there so big it stuffs the graph up – so we’ll do what most Americans do and just ignore Washington. What you might notice is that there is only one single State at the moment where the Democrats being ahead in the polls is not reflected in the Intrade market as a Democrat win probability above 50% – it’s Missouri sitting there in the bottom right hand quadrant. Missouri has regularly been showing a 50%+ Intrade win probability for the Democrats over the last week so it’s not particularly out of kilter here.
If we break the top right (Democrat States) and bottom left (Republican States) quadrants up into individual charts and label them, you can better see where all of the States respectively sit.
The Intrade markets and the polls are becoming very very correlated.
UPDATE:
A few folks have wondered what a line of best fit through that Intrade margins vs. polling margins chart would look like. A local regression using an Epanechnikov kernel seems to work pretty well.
21 Comments
Possum, the ‘Intrade Contested States – 89 EV’s’ page has dates ranging from 01-Jan-00 to 17-Jan-00.
“- so we’ll do what most Americans do and just ignore Washington.”
Oops – lost some digits! Ta – all fixed up.
Stark.
I can’t wait for the post-election analysis.
Thanks for that Poss, v good reading, even if I don’t have a clue who Epanechnikov is! I was just having an idle fantasy, what is the least numer of states one could win and yet win the required ECVs (ie the least number of biggest ECV states)? I Know BarryO will never get Texas but I’m just curious. (I spose I could get a calulator out and do this myself, but thought this might be info you’d have to hand).
I noticed that there are states who are above (in some cases significantly) our 70% intrade win probability who are on the same poll margin as those on this line or slightly below. This is making me a trifle suspicious (although there’s probably a logical explanation), though the line of best fit on the face of it seems to make an undeniable case for the polls tracking intrade. Poss?
An Cu,
11 States would get you 271. From largest to smallest:
CA TX NY FL IL PA OH MI GA NJ NC
55 34 31 27 21 21 20 17 15 15 15
Dan,
That is interesting in that there’s a fair bit of probability variance for any given polling margin. There would be a good chunk of local knowledge responsible for at least some of that – for instance, if the Dems have the working class and Hispanic vote pretty rock solid in the polling cross tabs in a place like New Mexico, then it would be almost impossible for them to lose it even though they might only be 5 ahead in the polling.
So a five point lead in New Mexico might be more likely to lead to a Democrat Victory than a five point lead in, say, Florida where the demographic coalition for the Democrats might be close to being equally large, but not as strong on partisan identification or voting determination (hence greater uncertainty over the Florida result).
Also interesting is that most (not all, but most) States were out in front of the polls until recently, suggesting that local knowledge on the ground knew a whole lot more about the actual nature of the race on the ground in a given state than the polls were picking up earlier. There might still be a bit of that in action.
@Dan – pardon for the ignorance as I’m new to to this but how does the line of best fit indicate that the polls seem to be tracking intrade?
I agree with your analysis. Our market is currently forecasting that Obama is 83% likely to win
( http://www.hubdub.com/m17795/Who_will_win_the_2008_US_Presidential_Election ) and our EV map is forecasting he picks up 364 electoral votes ( http://www.hubdub.com/election_map ). However I would be very careful about declaring game over.
A very unpopular Conservatives won an election in the UK in 1992 despite being long odds as did Howard in Australia in 2004. At the current forecast there is a 1 in 6 chance that McCain wins.
Aha! Nigel of hubdub fame – like your work.
In 1992, the UK pollsters used really poorly specced, unweighted quota sampling which was at the heart of the 4 pollsters getting it so wrong on election eve (making the prediction markets behave badly). That sort of unsophisticated sampling frame isnt used anymore by any pollster worth their weight anywhere… so to speak.
For the polls to be as wrong as the 92 UK experience, but wrong across multiple types of sampling frames and multiple sampling methodologies (from landline phones, landlines+cell phones, internet panels and face to face – right through to multiple weightings from age, partisan identification and socio-economic stratification weights etc) would require a type of systemic bias that we’ve never really witnessed before in 50 years of political polling, across the dozen or so nations where it’s been operating for that time.
Intrade has been leading the polls for quite a while in the US, and the polls are really catching up to market sentiment rather than the other way round – so I’m pretty confident they have it right (I’m normally a polling analyst but in US political contests, prediction markets have the edge hence my change in focus for this election)
On the Howard 2004 polling, his opponent Latham was ahead at the very beginning of the campaign, but after the first 2 weeks of that campaign the 4 pollsters here had him behind. Howard remained in front for the last 3-4 weeks of the campaign using the four pollster average weighted by sample size. The betting markets never had Howard behind in the campaign.
Of all the political contests I’ve looked at on Intrade (and Tradesports in its previous incarnation) – no political contestant has ever come from this far back to even nearly win, let alone actually win. On the polling, again, no candidate has ever come from this far behind to win – which is only bolstered by the enormity of polling data this year.
Throw into that mix the likelihood that the polling is more likely to have underestimated the Obama vote (because of the cellphone effect, the turnout effect and the demographic voter bulge in the under 30’s) than it is to have have artificially inflated it – through issues like the Bradley Effect (which I’m not really an empirical fan of) through to what we call the Shy Tory effect (where conservatives don’t tell pollsters what they think) – then it makes it even more unlikely that McCain will claw back.
The other problem being of course the geographic issues – where would McCain’s EV’s come from? He’d need to not only be underestimated by 8 points in the national trackers, but to win the Electoral College votes he would have to be 10 points underestimated in key swing states like Colorado and PA – states that he cannot win the Presidency without.
5 point turnarounds – sure
7 point turnarounds – maybe, but not likely.
10 point turnarounds? Holy psephological phenomenon Batman!
Add to that the turnarounds required on the prediction markets where the magical 70-30 rule collapses not once, not twice – but half a dozen times?
If I had to write someone off on the basis of evidence, It would be hard to find a more compelling case than McCain.
Possum,
Here’s an interesting chart.
http://www.clubforgrowth.org/2008/10/obama_vs_sp_500.php
That’s good Greeny!
It’s not too dissimilar to Bush and the S&P last election:
http://possumcomitatus.files.wordpress.com/2008/06/wolfersetal.jpg
Wall Street knows that its bread is distinctly Republican buttered.
So, in conclusion you agree that electing a Republican is good for business and will solve the financial market meltdown?
Is this the answer to your “World Peace” question? Just asking?
Electing a Republican is good for some businessmen on Wall Street certainly – but looking back over Bush Mk1, Clinton and Bush Mk 2, well one of them is not like the others in terms of running a successful economic policy!
Surely we’ve realised by now that what’s good for Wall Street pay packets isn’t the same as what is good for the real economy, a real economy that equity markets were once far more anchored to than they are today – a situation that will probably return quicker than most of us think.
There’s pretty solid debunking of the Bradley effect over at 538:
http://www.fivethirtyeight.com/2008/10/if-bradley-effect-is-gone-what-happened.html
Particularly interesting when you look at the understatement of the obama numbers in the dem primary. Then again the dem demographic (ha ha) is probably not as conducive to producing it than a general population sample. Don’t have any numbers for that, but if it’s not an issue I suppose it doesn’t matter.
Pollyanna – I’m not an expert here by any means, but ordinarily if there was a direct relationship between the x and y axis (the markets and the polls) you’d see a straight diagonal line across the graph (as one goes up the other does equally). The reason it plateaus at the top and bottom, producing that S shape, is because once the polling numbers reach a certain level (you can see this around the 15-point mark) people betting on the outcome of the state are acknowledging that there is basically no chance of one party or the other winning. So knowing that punters know that, you’d expect it to look kind of like that as a line of best fit. I was just surprised that there was such a big variation of the markets for states on the same polling margin, but poss has shed some light @6.
Poss, are you still contactable via the email addy on your old site? Have info re the wi-fi for Nov. 5 if you want to suss signal strenth etc. prior to E-Day?
Yep – it’s over on this site on the Contact page too.
Possum,
There’s an old Chinese saying, “When the fat man gets thinnner, the thin man disappears”.
The snake oil salesmen and hucksters have disappeared for the time being. But, they’ll be back when the times suit. In the meantime a lot of people are going to be scratching around for a living.
If you think the new commercial paradigm is going to be much different to the past, history says you will be disappointed
Speaking of Chinese philosophies and the like, I got out my copy of Sun Tzu’s The Art of War.
I can’t help but feel that the following applied to the US is appropriate:
“When General McCain, unable to estimate the enemy’s strength, allows an inferior GOP force to engage the Democratic one, or hurls a weak detachment against a powerful one, and neglects to place picked soldiers in the front rank, the result must be rout.”
There are plenty of other signs that could be applied from Sun Tzu, just as I am sure Lu Kewen Rudd did with his campaign.
Poss, some time back you wrote a lengthy piece in the comments of Nate Silver’s 538.com suggesting that looking at individual State Intrade market activity was a more accurate way of viewing the Big Picture. Furthur, you pointed to evidence of someone playing “silly buggers” with the Intrade BHO V JMc market.
Don’t believe Mr. Silver replied (he’s a busy man ‘n’ all) but it turns out that all the “silly buggers” was coming from one li’l ol’ source not tangential to the centre of McCain’s campaign.
Wonder if Nate, who made his name analyzing baseball stats, would have let something like this go through to the catcher?
http://www.cqpolitics.com/wmspage.cfm?parm1=5&docID=news-000002976265
I finished reading that very artcile just 5 minutes ago Ecky! The trades were from some “institutional investor” – read; foundation or company. Read; Some outfit with Republican links making a narrative play.
I’ll be writing something about this tomorrow or Monday if I can just find where I stored all the screengrabs of this trader pissing hundreds of thousands of dollars up the wall.