I can produce a model that will correctly predict every president going back to 1789, and predicts that George Clooney will be President in 2016. What do I win?
Unless he is a time traveler, his model did not correctly predict all of those elections. He is creating a model from the elections.
Mine came up with a prediction of Rosemary Clooney.
Ladies and gentlemen, I present the party of Lincoln, 2016.
Trump’s rally in Huntsville on Sunday has been moved from a local building to a High School Football stadium which will hold 3-4,000 in the stands and maybe a couple thousand more on the field, which is more than the indoor site could hold. So maybe his folks are anticipating a large crowd on Sunday.
They need room for the Photshopped crowds. 
It’s about 5 miles from my house, maybe I’ll have a peek.
Great. Another political dynasty.
While that is shameful behavior, I have to admit it was rather clever of them.
Huh?
It’s an indirect enough insult that I’m surprised teenaged kids could come up with it.
President Romney will be upset at this fact. It’s called back-fitting your data.
I don’t believe that’s the same model–I think like every college’s political science department has built out a presidential prediction model. This one from the Stony Brook professor did not predict Romney as the 2012 winner, now or then. But yes, most of these models accurately predict most past elections because they tweak the models continually until it mostly matches prior results.
This link has a PDF of the prediction he posted during the 2012 primaries, and it correctly predicted Obama would win reelection.
His model is called the “primary model.”
No, it isn’t the same model. But my point was exactly what you’re saying in the first paragraph - fitting the data this strongly has little value and means virtually nothing. You can be right for the wrong reasons, and wrong for the right reasons.
I’d be much more interested in a model that predicts Senate elections reliably, even if retrospectively. At least there’s a statistically meaningful amount of information there, unlike presidential elections.
According to the link, it also says Nixon would beat JFK in 1960.
Shouldn’t a good model fail to predict lots of elections that clearly turned on exogenous events? Or does it anticipate presidential burglary and electoral college/popular vote splits, etc?
Why would it need to anticipate electoral college/popular vote splits? Electoral college is all that matters. It would of course have to use state level popular support to estimate the electoral college votes.
The article says they think the 1960 election was rigged.