Wednesday, September 10, 2008

This week's presidential electoral vote simulation

I made an error in last week's simulation using the Intrade Prediction Market data by using the bid value for the probabilities instead of the last value. The last value seems to be an average of the bid value and ask value and better represents the actual payout at the expiration of the option.

Here is last week's chart showing 1000 simulations using the correct value data from Sept 2nd

The difference is only 70% of simulations showing the democrats winning as apposed to 75% on the incorrect chart from Thursday. I suppose I developed the formula for the concept but erred in its application. The above chart uses a more correct value for the probability built from the last value rather than bid or ask.

Here is this week's chart using showing 1000 simulations the value data from Sept 10th.

The probability of a given state being won by the Democrats is still the Democratic win value divided by the Democratic win value plus the Republican win value. Each state is still winner take all, and the rand() function takes care of the simulation part. This week shows the results of the Palin bounce (I don;t think it should be called the Republican convention or McCain bounce), but still shows the Democrats with more than the 270 electoral votes needed to win in 60% of the simulations. The median result is 278 electoral votes for the Democrats. We can monitor the situation over the next few weeks to see if the intrade data shows the republican bounce going away or not. Even with the poles where they are the simulations from the Intrade data still seem to predict a Democratic victory.

Intrade also allows trading in how many electoral votes will go to the Democrats or Republicans and gives many trading options at different bands of electoral votes, such as Democrats win more than 270 electoral votes. If we assume that the value for these trades represents the community's aggregate probability estimate we can compare the data to each other for consistency (democrats plus republicans should equal 538) and to the simulation above. I will build these charts for later.

3 comments:

TruthIsAll said...

The use of Intrade to calculate state win probabilities is mathematically incorrect.

The probability of Obama winning the election, assuming it is held today, is based on the state win probabilities which are a function of the latest polling spreads.

For example, if Obama holds a 51-49% lead in Florida, he has a 69% probability of winning the state. This is the value (0.69) which is compared to a random number (from 0-1) in a 5000 election trial Monte Carlo simulation.

Obama won over 99% of the election trials.

The Election Model assumes that
1) the election is held today, 2)it is fraud-free, and 3) the state polls fairly represent the preferences of the electorate.

TruthIsAll said...

I forgot the Election Model link:

http://www.geocities.com/electionmodel/2008ElectionModel.htm

TruthIsAll said...

This is the full link:

http://www.geocities.com/
electionmodel/2008 ElectionModel.htm