PredictIt.org Electoral College Predictions
So I’ve been following some PredictIt.org markets lately. Most notably, the Trump vs Clinton market. Clinton’s shares are currently trading in the $0.65-$0.68 range which means traders believe there is a roughly 65%-68% that Clinton wins in November. PredictIt.org also has state by state markets for the electoral college for around 30 states. So I wondered what would happen if I took these individual state markets and used them to predict the outcome of the electoral college in November.
PredictIt.org has a nice API platform that allows for easy access to their data and I wrote an R script to get all of the closing prices for the Trump vs Clinton state by state markets (Code is on GitHub. If you have a better way to parse the information that come in from the predictit.org API, I would like to know a better way. I hate how I did it, but I it seems to work fine). I then used the individual market prices as approximate probabilities the Clinton or Trump would win each state. When a market wasn’t available, I used a probability of 0 or 1 depending on who won the state in the 2012 election. (These are states like Idaho and Connecticut, we know what going to happen there.)
Based on these state by state markets (as of July 29, 2016), I simulated the election 100,000 times and estimate that Clinton has a 83.16% chance of winning, Trump has a 16.21% chance, and there is a 0.64% chance of a tie. (I will try to update this as often as possible this fall, though with a baby on the way in Septemeber (!!!) we’ll see how often than actually happens). 16% sure is a scary percentage though if you are absolutely petrified (as I am) of a Trump presidency. In Texas Hold ’em, 16% is about the probability of a pair showing up on the flop, making a full house or better by the river after flopping two pair, or flopping a guy shot and hitting it by the river. I’ve seen that happen way too many times to sleep comfortably at 16%.
Clinton’s probability here is interesting because the state by state markets have her at over 80%, but the overall market (Trump vs Clinton) has Clinton at under 70%. I personally believe the 80+% number (for the purposes of keeping my sanity) and think Clinton vs Trump the market is way under priced for Clinton. (I have invested accordingly). This 80+% number is also inline with the Bayesian model from Princeton Election Consortium, which currently has Clinton at 85%. (Their random drift model has Clinton at 65%).
FiveThirtyEight currently has their polls only model at 50.1% for Trump.
And their polls + forecast model has Clinton at 60.1%
I think these numbers are way too low for Clinton, and I think in general, FiveThirtyEight’s predictions are much too conservative in that they tend to be too close to 50%-50%. I wrote a blog post before the 2012 election about how many states we expect Silver to get wrong and there was only about a 5% chance that Silver would get every state correct. Based on his state by state probabilities, there was a very large chance he would get at least 1 state wrong. But Silver actually got all the states correct. Based on his results in 2012 and his current election predictions in 2016, I think Silver is too conservative in his election forecasts and he’s actually even better at it than his probabilities show. (Princeton Election Consortium had the probability of an Obama win at 99% on November 1, 2012).
Next I wanted to look at win probabilities given that a candidate wins a specific state. In order to do this, I took the state by state win probability vector and one at a time flipped each state to a probability of 1 for a given candidate and left all the other probabilities alone. I then simulated the election 50,000 times to look at Clinton and Trump win probabilities and the tables are below ordered by the single state that gives the candidate the highest probability of winning the election. For Clinton this state is Texas. With it’s 38 votes, if Clinton can win Texas (extremely unlikely, but possible; Houston Chronicle just endorsed her) she has about a 97% chance to win the election. The next four states for Clinton are Florida, Georgia, Tennessee, and Indiana, which if she can win any of these states give her at least an 89% chance to win the election.
|Given Clinton Wins:||Clinton Win||Tie||Trump Wins|
For Trump, the state that would raise his win probabilities the most are California and New York, both very blue states that Trump, for some reason, believes he has a shot at in November. If Trump could somehow manage to win California, his win probability jumps to 70% with a massive haul of 55 electoral votes. This is a very unlikely scenario, but nothing is impossible in this election. After New York, the next states for Trump are Illinois, Florida, Pennsylvania, New Jersey and Michigan. If Trump can win any of these states, his chances of winning go up to at least 23%, still a big underdog.
|Given Trump Wins:||Trump Win||Tie||Clinton Win|
Finally, I looked at what would happen if Clinton and Trump won different combinations of the three swing states Ohio, Florida, and Pennsylvania.
If Clinton wins all three of these states, she has about a 97% chance to win the election. If she wins Florida and Ohio or Florida and Pennsylvania her chances are about 95% and if she wins Ohio and Pennsylvania she would be at about 93%.
If Trump wins Ohio and Pennsylvania he has about a 35% chance to win, is he wins Florida and Ohio he has about a 37% chance to win, and if he wins Florida and Pennsylvania he has about a 43% chance to win. So even if he were to win 2 out of 3 of these states, he would still be an underdog. What happens if he wins all three? If Trump wins Florida, Ohio, and Pennsylvania, he has about a 60% chance to win the election. Think about that. For Trump to win Florida, Ohio AND Pennsylvania, things would have to go absolutely perfectly for him and there would still be a 40% chance that he loses the election. Trump has a YUGE (sorry) electoral college problem.