Predicting the Outcome of Tomorrow's Midterm Election

The first academic paper I ever published was an empirical analysis of the “midterm gap” in American politics.? (I couldn’t find an ungated version, but it’s not really worth reading anyway!)

In almost every midterm election in the 20^th century, the party of the president has lost seats in the House of Representatives.? I wrote the paper as an undergraduate for a class taught by one of my favorite professors, Alberto Alesina.? He liked the paper and suggested I try to publish it.? I waited a few years to try while I dabbled in management consulting, but when I decided to go back to get a Ph.D., I dug out the paper, revised it and sent it off to an academic journal.

The paper ended up having a very typical life for an academic paper.? It got rejected from the first journal I sent it to.? Then it got rejected from a second journal.? Finally, I got a letter back from the editor at a third journal that was so cryptic I wasn’t sure whether the paper had been accepted or rejected.? I showed the letter to my MIT advisor Jim Poterba, and he gave me some sound advice: when in doubt, act like the paper has been accepted.? So I revised it and sent it back, and indeed it was accepted at long last.? That was in 1994.? In the last 16 years, it has managed to collect a whopping four citations according to Google Scholar. ?That’s probably about the median number of cites for an economics paper.

With all the buzz about this year’s midterm election, I pulled that paper out to have a look at what it predicts this year.? I found that a handful of (obvious) factors were important predictors of how many seats the party of the sitting president would lose.? The decline tends to be worse when the presidency switched parties two years earlier (i.e., it looks like the incumbent president’s party gets punished in both presidential election years and at the midterm, so the congressional vote was abnormally high in 2008 for the Democrats).? It helps the president’s party if GDP is growing quickly, if the president has a high approval rating, and if the president’s party has a lot of incumbent congress members.

When I crunch through the numbers, the prediction of the model is that the Democrats will end up with 199 seats, or a loss of 56 seats.? I would put essentially zero faith in this estimate on its own, but I find it interesting that the prediction it yields matches almost perfectly with the market prediction at Intrade.com.? As I write this, the market thinks there is a 47 percent chance that the Democrats will lose 60 seats or more.? Interestingly, back in August the market only put a 15 percent chance on a loss that big.? If you had believed my model (which would have made the same prediction in August), you could have made some money betting against the Democrats at that time.

The model did come through with flying colors the only other time I asked it to make a prediction.? That was back in 1994.? My paper was published in March 1994.? In the original conclusion to the paper, I estimated that in November of that year the GOP would take control of the House for the first time in decades.? My advisor Jim Poterba, when he read that prediction, told me I had to take it out of the paper. The first rule of paper writing, he told me, is to never make predictions about the future.? Also, he said, that prediction is just plain dumb.? Everyone knows that will never happen.? So he made me take the prediction out of the paper, and he told me he would buy me a pizza if I was right.

Even up to the day of the election, my prediction looked crazy.? I was at a political economy seminar at Harvard on election day, and when Poterba announced to the assembled crowd that I predicted the Democrats would lose the House, they guffawed uproariously.? Shockingly, the Democrats were routed, just as the model predicted.

Oh how I wished I had left that prediction in there.

If I had, I might have gotten five cites.? Or maybe even six.