Possums Pollytics

Politics, elections and piffle plinking

US Election –Introducing Intrade

Posted by Possum Comitatus on May 27, 2008

As the US elections inch closer, it’s probably about time we pulled our finger out and started to have a look at it.

There are two big problems with the US election when it comes to getting a psephological grip on it. Firstly, the Republicans and the Democrats have yet to appoint an official nominee – meaning that the system hasn’t properly been switched “ON” yet. As a result the polling and betting market data is polluted – however small – by multiple candidates being the focus. The head to clash hasn’t precipitated out fully yet and wont until each party holds their convention.

The second big problem is US polling. Due to the US having voluntary voting, polling results have a big extra dollop of uncertainty added to them. People that say they will vote a certain way not only produce the uncertainty that they might not actually vote that way in practice (the universal problem political pollsters have), but that they might not end up voting at all. Many firms use lists of registered voters in an attempt to overcome this problem, but with a candidate like Obama that is driving massive new turnout, and with parts of the Republican base getting apathetic and not bothering to turn up – this all gets a little problematic.

The other part of the polling problem in the US is that there are far too many shithouse polling organisations that couldn’t successfully poll what a family of 4 wants for dinner, let alone what the political intentions of 300 odd million people are at any given time. If you remember back to last years election, we poured big buckets of the smelly stuff over pollsters that used small sample sizes.

In the US, sample sizes of less than 500 are a regular feature.

At a later date we’ll whittle down the polling organisations to those that can be trusted and we’ll ignore these small sample tossers that give polling a bad name – that way we might be able to get ourselves some credible US polling goodness.

But in the mean time, I’ll introduce you to the data that we’ll be using most for the upcoming US election – a betting market called Intrade. Betting markets in the US have outperformed polls at every election since 1988 when the Iowa Electronic Markets came on-line. One of the reasons is pretty simple – polls aren’t predictive, they simply measure a snapshot of opinion at any given time. To make them predictive we have to take all those snapshots and build a forecasting model to project ahead into the future. Prediction markets like Intrade on the other hand are predictive by design – people bet on who they think the eventual winner will be.

Intrade isn’t quite like the betting markets we have in Australia which more closely represent laying a bet with a bookie – at Intrade, people buy and sell contracts based on what they believe will happen, so it removes any house manipulation and lets the market find it’s own natural equilibrium based on nothing more the price people offer to sell a contract and the price people are willing to pay to buy such a contract.

The beauty of this type of market, especially the way Intrade is designed in terms of the price limits they place on any single contract, is that implied probabilities of an event occurring like a Democrat becoming president for instance, is simply the Democrat “last price” divided by 100.

I used a start date of April 1st 2008 as the beginning of my Intrade data set, so to give an example of what the probabilities currently look like for the Democrats winning the presidential election, well use the contract value at the end of each day since April 1st right through to yesterday.

As you can see, at the moment, the Democrats have a probability of winning the US presidential election of 62%. Worth mentioning is that this result also has a built in implied probability of Obama becoming the Democrat nominee of 92%, as that’s what his current probability to get the Dem nomination is currently running at over at Intrade.

What’s noticeable here is the small surge that happened after the 6th May. Maybe someone could fill me in on the why’s of that?

Intrade have a whole lot of things they run contract markets on, Senate races, VP nominations, House election etc, including a market for the presidential winner of each US State. Because the US election winner is ultimately decided by who can get the magical 270 of the 538 Electoral College votes that are available, where each State has a set number of Electoral College votes to give, it’s worthwhile looking at the presidential election as a collective of 51 separate elections.

When we total up the current projected Intrade winner in each of those 50 State elections (plus the District of Columbia) and tally up the electoral votes associated with those States, we get a measure of how many electoral votes the Intrade market currently has each Party on.

We’ll call this figure “Intrade EV Win Tally” for it tells us who is actually ahead in the electoral vote contest and by how much – we’ll graph it in a minute.

The other thing we’ll regularly use here is a measure of how the broader support levels of each political party are changing through time . To do this we simply multiply the electoral votes that each state has to give with the current probability that a party will win that State.

So if, for instance, we take Colorado as an example – Colorado has 9 Electoral Votes to give. The current probability of the Democrats winning Colorado is running at 62.4% or 0.624. So we multiply the 9 Electoral college votes by 0.624 to get a figure of 5.616. When we do that for all the States and add them up, what we get is a probabilistically weighted estimate of the broad level of support for the Democrat party based in terms of hypothetical Electoral College Votes. We’ll call this “Total EV probability Sum” – shitty name I know, but I’m stuck for something better at the moment.

This might not seem that important today – but it will come in extraordinarily handy as the election progresses for all sorts of reasons that we’ll get to as the need arises.

So we can now graph these two things – the projected Electoral College votes using the State based markets which tells us who is currently leading and by how much, as well as the Total EV Probability sum which for now gives us a measure of broad support change.

We’ll just do this for the Democrats rather than for both parties as the Republican chart is just a mirror image of the Democrat chart, and we’ll use weeks rather than days just to knock out some of the volatility – again starting from the 1st April.

So what we can see is that currently, Intrade has the projected Electoral College Votes for the Democrats standing at 293, down 5 from early April when the probability of the Democrats winning Nevada and their 5 Electoral College votes fell from 51.5% to the 48.5% probability that the Dems have today of winning that State.

The magical number is 270 – so the Democrats are projected by Intrade at the moment to win the election by 23 Electoral College Votes.

The EV Probability Sum shows us that the broad level of support for the Dems in the combined State markets is slowly increasing. Doesnt mean much now, but it certainly will a little later on.

If you want to have a browse around Intrade, you can pop over HERE and have a squiz. Just follow the politics links.

UPDATE: There’s a new button at the top of the site called “Intrade Data”. “US Election”

It has time series charts of the three Intrade measurements that I’ll mainly be using for the US election, among a few other things and a spiffy map. They’ll be updated every Tuesday.

UPDATE 2:

Two new bits added to the Intrade page – a chart that shows where each state stands on the probability spectrum along with the Electoral College votes they provide:

And probabilities done as a map – with blue for Dem and red for Republican (the darker the colour the higher the probability that the respective party will win it, 0.5 is white)

AddThis Social Bookmark Button

add to kwoff

17 Responses to “US Election –Introducing Intrade”

  1. Cat said

    Possum both Mumble and Jackman have expressed doubts about Obama being elected (no matter how on the nose the Republicans are) because of race. I personally though cannot see any one could seriously predict the outcome given factors such as the strongly unpopular Republican party putting up a traditional old white guy up against a “fresh” (or inexperienced depending on your perspective) candidate who happens to be bi-racial. Will fresh trump old cancer survivor or will race become a bigger problem in a general election. In other words do you think the polls have any real chance of predicting the outcome in such an original election?

  2. Andrew B said

    May 6th matches up with the Democratic Primaries in North Carolina and Indiana. That’s pretty much when Clinton’s chances finished.

  3. George said

    Andrew B is right – 6th of May is when it all came to an end of Clinton and a surge in superdelegate pledges began for Obama and very quickly surpases Clinton for the first time in the campaign; take a look at this graph at the New York Times:

    http://www.nytimes.com/interactive/2008/05/09/us/politics/20080509_SUPERS_GRAPHIC.html

    Poss, I have a question for you with respects to the betting markets, and I have to admit I’m a total ignoramus when it comes to how they work. But let’s say you have a million people each betting $1 for candidate A and 500,000 people betting $5 each for candidate B, what would the odds look like? Would such a scenario show odds that would indicate a landslide for candidate A?

    [Sorry George – you were hiding in the spam bin… Poss ]

  4. Adrian said

    The problem is that both of these analyses are flawed. The style of EV Probability Sum works fine when you have hundreds of regions to even out the errors, but there are really only 10 states or so that can be considered “in play” in the US, and getting a proportion of the vote in most of these will net the Democrats nothing. Likewise, the Intrade weighting is only as good as the punter’s understanding of the US electoral system, which is not great. The numbers for Democrats in the House look fantastic, because here a nation-wide 5% boost nets a lot of seats, but in the Presidential race it doesn’t and my bet is that people generally are not taking electoral system into account. The only way to predict the election is by state-by-state polling. Granted, all the problems with poor polling and the primaries are still going on, but Electoral Vote has compiled the state-by-state polls and gives Obama 266, McCain 248 and 24 ties. To me, that says close race, which is not reflected in your numbers.

    http://www.electoral-vote.com/evp2008/Obama/Maps/May27.html

  5. Possum Comitatus said

    Adrian, The EV Probability Sum just measures broad movements in support across the 51 contests – not electoral college votes. The usefulness of this metric will become apparent the further into the contest we get – you’ll just have to trust me on that one for now.

    The punters understanding of the electoral system at Intrade is exceptional – they wouldnt be betting on who wins each State if that were not the case, nor the Senate races for that matter – they’d be playing in the headline contest markets instead.Intrade isnt a bookie – its a contract market, it’s works in an entirely different way to betting markets like Centrebet, and has a different demographic altogether. With a bookie you simply roll up and put money down on the prevailing odds, at Intrade you have to buy a contract on an outcome by offering a price; if the price is too low, no one will put their money in to match it. That’s how prices, and as a result probabilities are set – on the matched contracts.

    Electoral Vote and places like the RCP polling area and whatnot are great sites – but the polling averages they use, for my mind, need to be cleaned up first to remove the shonky pollsters and polls (of which there are a great many in the US) and reduce the uncertainty contained in the aggregated polling series before they even get near the ballpark of being accurate.

    We’re not only going to follow Intrade here for the US Election – we’ll do polls as well, but not until the conventions or at some period when the nominations are 100% settled. Then we can set about building a decent poll aggregate that excludes the firms that spend more on their media relations than they do on actual polling – and knock out those silly little small sample polls to boot.

  6. Possum Comitatus said

    George – at the bottom of the Intrade site front page under “Intrade News”, there’s a heading called “Intrade videos”.
    The first link under that is “Trading 101 Video ” – that explains the ins and outs far better than I can.

  7. caf said

    Nebrasksa and Maine don’t use a winner-takes-all method to allocate their electoral college votes – does the “Intrade EV win tally” measure take this into account?

  8. Possum Comitatus said

    Caf – even though Maine and Nebraska have the potential to split their electoral college votes, in practice they’ve never done it (the ultimate winner has won all districts since it was introduced in 1972 for Maine and 1996 for Nebraska).

  9. David Gould said

    Very interesting stuff, Possum. I have to say that reading what you have been saying over at pollbludger and now this has reduced my doubts re Obama’s electability considerably.

  10. Don said

    Poss, Hawaii and Alaska are listed on the graph but I can’t find them on the map. Is there a reason for that?

  11. Possum Comitatus said

    There sure is Don – I haven’t figured out how to move Hawaii and Alaska next to the continental US yet!:mrgreen:

    I could always open GIMP and do they old cut and paste, but I’m trying to figure a more automated way (which isn’t looking good so far). It’s looking like I might have to treat those Eskimos and Hula dancers like Tasmanians and leave them off the map.

  12. robertbe said

    What did it on May the 6th was that Obama ran Hillary so close in Indiana. That grew tighter as the night wore on and had a big effect on the punditry. Culturally speaking, Indiana sure ain’t San Francisco (a friend who lived there tells some horror stories). Which meant the pundits were primed for the polls to have overestimated Obama’s vote. When he exceeded expectations that pulled them up short and they started saying, hang on he may have this.

    Which brings me to Mumble and Jackman’s negative thoughts on Obama. Quite surprising given their normally balanced approach to things. Whilst there are a lot of unknowns, and this really is new terriotorty, for them to take such a definitive stand this far out is, well, a little weird. Makes you wonder if there is not some emotional baggage clouding the statistical judgement there. Which is not to imply racism or anything. More likely a not-daring-to-hope double reverse psych-out or something.

  13. Possum Comitatus said

    It was really only Mr Mumbles that made the point on Obamas electability, Simon just crunched some really really interesting numbers.

  14. robertbe said

    You’re right. On rereading Simon’s piece, even Mumble’s revised mention of him is a stretch

  15. Andrew B said

    The problem that I see with all the established punditry is that no one has any idea, at least on the democratic side, for support levels in Michigan or Florida. And given the importance of those states they entire not campaigning there thing leaves them untested to a certain extent.

  16. Stephen Lloyd said

    Not allowing Michigan and FLorida votes to count may cause a backlash in the general, too.

    Also, really McCain is polling very favorably against Obama in the states that count, and as long as good news keeps coming out of Iraq, as it has been of late, I think it will be a lot closer than people think.

  17. jimmykun said

    Hi Poss, I’ve been lurking on your blog since election time last year, and enjoyed every minute of it. As I live in Tokyo, it was great to have the real action deftly laid out in the easily digestible chunks you threw up every day. Keep up the great work.

    Readers with an interest in news prediction markets may wish to have a squiz at http://www.hubdub.com This site is a playground version of Intrade, and users are given $1000 imaginary dollars on joining to punt as they wish. The users make the questions, and as might be guessed, many are news junkies (like myself). It’s aimed at the American market, so the politics section is chok’ full of election 2008 questions. Markets predicting poll outcomes are often put up, and many questions are insightful and prescient. My participation has lead me to this great overview of US polling:
    http://www.realclearpolitics.com/ This superb site lists all major polls and is evidence of some of the insanely variable results that some poll companies spit out, presumably with a straight face.

    [Sorry Jimmy – you were in the spam bin….. Poss ]

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

 
%d bloggers like this: