A number of the people I'm reading are batting around potential models for election predicting. As the title shows, I think this is mostly a waste of time. As much as economists in particular seem to be into this stuff, the field of economics and social sciences generally are, in my impression anyway, exceptionally bad at forecasting.
It isn't a knock against those fields. It's just that human behavior, despite seeming like it's so predictable, just isn't once someone has to start trying to predict it. There are too many variables. Kevin Drum notes, for instance, that despite the attempts of various people to come up with a model for predicting presidential election winners, one can get just as much accuracy by simply predicting that if the incumbent's party has occupied the White House for less than 8 years, they win, and if more, they lose.
You see another weakness, though, in the fact that these models always start at 1948. Why? Because election behavior was very different with FDR, who beat just about every electoral adversity imaginable. And before FDR, there was a very easy, but starkly different, way to predict elections: if the Republican candidate was running against a candidate not named Grover Cleveland or Woodrow Wilson, he won. Before 1860, it was the exact opposite: if one candidate belongs to a party with "democrat" in the name, he's probably going to win.
And how long will any given model last? Until it stops being predictive, which he won't know until it fails to predict several elections.