The following Intrade Tracker charts will be updated daily.
Comments for this page can be directed over to the latest US Intrade data article here.
All charts are thumbnails – click on them to expand.
Intrade/Gallup Daily Trackers
The following charts will be updated once a week, every Monday for week ending Sunday data . [Last Update - Nov 3rd]. All data is derived from the Intrade US political markets.
Projected Electoral College Votes by State + Expected Electoral College Votes
This graphic charts the number of Electoral College votes given to the Democrats by the Intrade State markets as well as the Expected Value of the ECVs. The Expected Value is derived by multiplying the Intrade probability of the Democrats winning each state by the Electoral College Votes of each state, then summing those totals up.
States holding Democrat:
MA HI DC IL NY CA DE RI VT MD CT WA MI OR NJ ME IA WI MN PA NM NH CO NV VA OH FL NC
Current Democrat win probabilities by State
If we redo this chart using a fixed probability scale on the bottom axis, it hammers home the massive probability gap between state clusters. This Chart was done using data from October 20
Win Probability Change Over Past Week – By State
This shows how the Democrat win probability for each state changed over the past week – an increase in the probability of the Democrats taking a State is shown as a positive on the left hand axis, a reduced chance as a negative and where States are sorted left to right on the bottom axis according to electoral votes (so the small population States start from the left and go through to the large population States on the right)
Intrade Election Map
Now as a spiffy Intrade widget:
And while we’re at it, this is the real time updated polling map from Pollster.com
Simulation of State by State Intrade probability data – 100,000 trials.
This simulation adjusts for the non-independence of State markets as well as the dodgy behaviour of the probability tails called Longshot Bias, and the fact that given Intrade probabilities are only approximately their true value at any given time. To see the methodology of the simulation – it’s described at the very bottom of the page.
That last chart lets us see the probability (on the left) of the Democrats gaining at least any number of Electoral College Votes (the bottom axis) according to the current state of the Intrade state markets.
Intrade Electoral College Simulations Over Time
If we look at how the mean, median and mode of those weekly simulations has changed over time for the projected number of Democrat Electoral College votes we get:
Simulated State by State Democrat win probability over time.
If we use those simulations to look at how the probability of the Democrats getting at least 270 Electoral College votes has changed through time compared to the headline market of “Democrat as President”, we get:
The reason we are running State market based simulations is because of my suspicion that the collective information held by the State markets is superior to that held by the national market. The underlying reasoning is simple – the US election is effectively the combined result of 50 separate but interdependent, smaller, local elections.
The information that market participants in those state markets know about a given state election (as a proportion of the total amount of information in existence about that State election) is greater than the information that market participants in the national market know as a proportion of the total amount of information in existence about the entire country. The hypothesis is effectively that the knowledge gap in each state is less than the knowledge gap of all states combined into a national headline market.
Although it’s early days yet and we don’t have enough observations to empirically test the hypothesis at the moment, the early indications are good and the State by State market simulation appears to be a leading indicator of the national headline market.
Simulations are easier than they sound.For instance, let’s take some hypothetical US State – we’ll call it Canada – and let’s say that Canada had an Intrade probability of a Democrat victory of 70%. If we generate a random number between 0 and 100 and then compare that random number to the 70% probability for Canada, if the random number is less than or equal to 70, we give the State’s Electoral College votes to the Democrats. If the random number is greater than 70 we give that State’s Electoral College Votes to the Republicans. If we did this 100 times, we’d expect to get results showing that around 70% of the time, Canada goes Democrat.
But we don’t just do that for any single State, we do it for every State, and then sum the total of the Electoral College Votes that the Dems could be expected to win. We then do it another 99,999 times, we get those histogram and CDF charts which tell us the probabilities of the Democrats getting any number of Electoral College Votes.
However, there are three key problems with Intrade probabilities.The first is that the further away from 50% a probability gets, the less likely is the chance that the implied Intrade probability is true. For instance, if we had 5 states all with a 20% probability of a Democrat victory – if an election were held tomorrow we’d expect one of those States to show up as a Democrat win. If we had 50 states with a 20% probability of a Democrat victory we could be highly certain that one of those States would show up as a Democrat victory. If we had 200 such States (yes, I know – there’s not 200 States in the Union but stick with me here), we could not only be virtually guaranteed that at least one of those States would be a Democrat victory, but we would expect that somewhere around 40 of those States would go Democrat – simply on the implied probabilities.
However, in all the Intrade political results I’ve looked at, literally hundreds including all of the Intrade Congressional, Senate and State Presidential winners, I’ve never seen a single example of a political party winning anything with a probability of less than 20%. So the Intrade tails aren’t a true representation of the real probability.
We can adjust for this by saying that any probability lower than 10% is actually 0%, and any probability higher than 90% is actually 100% (we’ll use a 10% figure rather than the 20% figure simply to stay conservative with the numbers, and to give the rest of the probability spectrum here enough rope to hang itself… so to speak).
Hence, when we draw our random numbers, we don’t draw them between 0 and 100, we draw them between 10 and 90. As we approach the election, we expect the Intrade markets to gain more certainty, so we will reduce that draw to 20/80, then 30/70 as we approach E-Day.
UPDATE: As of Election Day minus 40, each day the draw width gets reduced by 1% – or one half of one percent from each end. Hence on Election Day minus 39 the draw width was 10.5/89.5, on Election Day minus 38 it was 11/89 etc etc so that on E-Day itself the draw will be 30/70.
Another problem is that the Intrade probabilities arent true (even at 50%), but are only approximately true at any given point because markets like Intrade are always in a state of flux, attempting to reach a hypothetical equilibrium that can never be achieved (because political events that drive these political markets are constantly occurring). This relates to another problem, a much larger problem, in that State results aren’t independent from each other.
If the Democrats lose New York for instance, they aren’t going to be winning Virginia or Colorado any time soon. In fact, if they lose New York they will probably have lost nearly every State in the process- so we need to adjust our simulation to accommodate the fact that nearly all States are dependent on the broader result happening across the entire country.
To solve these final two problems, and thanks to our readers here, and Caf particularly, for their exceptional assistance with this, we do a number of things.
For each iteration of the simulation we first generate a random number between 10 and 90 (to solve the first problem). Let’s call this number M.
Then we generate another random number for every State, but a random number from a normal probability distribution that has a mean of M and standard deviation of Q -we’ll call this number S.
If S is less than or equal to the State Intrade probability of a Democrat victory, we give that State and its Electoral College votes to the Dems, if the number S is greater than the Intrade probability we give the State’s Electoral College votes to the Republicans.
This solves the second problem on the one hand (as it simulates intrade probabilities being only approximately true) and allows us to solve the third problem on the other hand by making the random numbers that are drawn to call each State dependent on each other by way of their sharing a common mean value and standard deviation of the probability distributions they are drawn from.
Using the Intrade election data from 2004 to calibrate our system, having a standard deviation of 1 (which is a lovely and convenient round number by any yardstick) gave us the most accurate results – so we’ll use 1 as the value of Q for our simulation this year.
After each State has been called by their own probability distributions based on one value of S (with mean M and standard deviation Q), we sum up the total number of Electoral College votes for the Democrats and that gives us one simulated election result. We then run another iteration, where another value of M is drawn and where each State then draws another random number from their own probability distributions (of new mean value M and standard distribution Q), to determine whether each State goes Democrat or Republican. We add the ECV’s up and this gives us another simulated election result.
We do this 100 000 times and arrive at our final weekly simulation results.