Predicting the Election with Macro Data

There’s a lot of blog posts, a big academic literature, and even popular books that use macroeconomic data to forecast whether Obama will win or lose in 2 months.

Reader KL remarked that I said in class years ago that no incumbent has won if the rate of growth of real GDP was too low.

Well, back before voluntaryXchange I had another spot called MacroBlog.* And on Macroblog I called the 2004 election 12 months before it took place.

How did I do that? Well, in class my students built an election forecasting model too.† And it showed that no incumbent has ever won reelection if the average growth rate of real GDP was 1.4% or lower, and on incumbent has lost if the it was 3.1% or higher.

So, where does Obama sit in all this? His average growth rate is 2.1%. So he’s right in the no man’s land where our model predicted a toss-up.

For my money I’d look at Intrade or the Iowa Electronic Market, where the current odds are about 2:3 that Obama will win. That seems about right to me.

One thing’s for sure … Obama seems certain to get less electoral votes now than in 2008, and the only incumbent to do that in our lifetime is Jimmy Carter. But, Carter did not edge Ford by much in 1976, so he had a smaller cushion to work with than Obama does.

* My Macroblog (hosted on Blogger) was more class oriented than general interest. Part of the reason I started vX was that I was shifted out of the macro class I’d been doing every semester, so continuing its blog didn’t make much sense. And then, when blogging took off, David Altig and his coauthors at the Federal Reserve Banks of Cleveland and Atlanta didn’t really look around before naming their new offering on Typepad Macroblog.

† Forecasting elections is fraught with two problems that are common in macroeconomics.

One is high unexplained variation: in short, voters are funky. The major candidates always get 50% ± 10%, but tightening up that range is the tough part.

The second is a statistical problem called multicollinearity. This means that you have a large set of variables you could use, but they all cover just about the same ground. So, you can use one or the other and get just about the same results, but different combinations probably don’t help you out much. This is why there everyone can have their own election forecasting model that’s a little different from their peers: it’s easy to make it different, and hard to make it better. If someone could make a better model, everyone would switch to it.

Technorati Tags: predict,forecast,2012,presidential,election,obama,romney

Comments

[Kit Lloyd]

Yep, that was it. After making this prediction, you mentioned this after-the-fact in your Macro-Econ class that I took. This would have been Fall '05 or maybe Spring '06. Thanks for the post.