If you're so smart, why ain't you rich?

First, daily update: Obama 58.5% chance of winning, mean of 276 EV. One change in how I count the results: I am going to count 269-269 ties as a win for Obama, assuming Democratic House and partisan voting. I have it as a 1.3% chance of a tie today.

Obama's currently winning 293 EV worth of states in the intrade markets, so we still have the situation of fluctuations being to his detriment.

Overall, Obama has moved up dramatically in the last week or so. Purely using the latest polls in each state, you might not be aware of this. In fact, it might take weeks to be sure of what happened, and then you would be confused about whether to interpret the polls of two weeks from now in light of today's events or in light of two weeks from now's events. I got very interested in doing all this because the Democratic convention bounce could only be seen a week later, when the Republican convention was actually going on - which made it look like the Republican convention was producing a negative bounce. But a week after it was over, the Republican convention bounce did show up in the polls.

Now, why you might ask is the Obama for president contract selling for only 51.5 when the chance of his winning is possibly 57% according to these other 51 markets? Doesn't that represent an arbitrage opportunity? Yes, there is something there. I would think that the state-market bettors have a better idea of what they are betting on. It is really quite difficult to assign a probability on any given day to Obama winning in November (hence this whole effort of mine).

As I do some backtesting, I can see if for example, this metric is a leading indicator of the overall "Obama wins" contract price.

There's a lot to work out though, before I would lay my own money on the line.

• Are price movements really normally distributed?
• Is the correlation matrix stable?
• Is there a stable variance in each state, or does that need to be stochastically modeled? Some have suggested that variance ought to decrease as the election nears. That is part of the reason I use the 90-day window, although 90-days may or may not be the ideal length. Perhaps a decay function on the input to the variance calculation to weigh older data less would be a good way to go.
• Intrade prices probably cannot literally be interpreted as probabilities, especially far from 50.
• As a subpoint to that, investors require a risk premium, so at best we are observing risk-neutral probabilities.