May 14, 2004,
While polls show Bush and Kerry neck-and-neck, a sophisticated econometric model operated at Yale University the same kind of model used for simulating the entire U.S. economy is calling Bush the winner by a wide margin, with almost 58 percent of a two-party vote.
The model is the brainchild of Professor Ray Fair, a fellow at the International Center for Finance at Yale, and one of the world’s most respected authorities on econometrics. He came up with the model in 1978, and published it in a book called Predicting Presidential Elections and Other Things.
The “other things” include the quality of wine, the outcome of sporting events, whether your spouse will be faithful to you, and classroom grades. Fair told me that the book was all about getting college students interested in econometric methods: booze, sports, sex, and grades seemed like the right topics (in approximately that order).
The idea of econometric modeling is to try to explain a complex phenomenon in terms of a small number of objective variables. Then, if you can predict those variables, you can predict the complex phenomenon.
Fair has found that presidential elections can be explained by just six factors. And his track record is spectacular. His model explains elections since 1960 with an average forecast error of only 2.4 percent far better than any conventional predictions derived from polling.
The only election he really blew was in 1992, when Bill Clinton beat George H. W. Bush. If you take that one out, his record shows only a 1.6 percent average error. He even correctly called the squeaker in 2000 with Gore marginally winning the popular vote, when all the traditional polls were calling for George W. Bush to win the popular vote handily.
Let's look at the factors.
Fair’s found that there’s been an historical advantage for candidates who are incumbent presidents, Republicans, and a member of whichever party has not been in power for more than two terms. As an incumbent Republican finishing his first term, following eight years of a Democratic presidency, George W. Bush has all three of those factors working for him.
According to Fair’s model, those three factors alone start Bush out with a structural advantage that puts him at 55.6 percent in a two-party race.
From there, it’s the economy, stupid. The remaining three factors are the per capita GDP growth rate (Fair calls this factor GROWTH), the number of quarters before the election in which per capita GDP growth exceeded 3.2% (GOODNEWS), and inflation (INFLATION). Obviously, GROWTH and GOODNEWS are positive factors (Bush wants bigger numbers going into the model) and INFLATION is negative (Bush wants a smaller number).
Now nobody knows exactly what those three factors are going to be on election day, but it’s not hard to make educated guesses. You can go to a page on professor Fair’s website and input your own guesses, run his model, and see who wins. If you use Fair’s guesses, which are based on simple extrapolations of today’s numbers, Bush wins with 58.7 percent of a two-way vote.
Fair told me, “There’s no way that Bush could lose if this thing is correctly specified.”
So what are the chances that the model is not correctly specified?
First, the inputs could change between now and Election Day. They could change for the worse if the economy deteriorates, or they could change for the better if it improves.
I’m in the camp that the economy is going to keep getting better. But be that as it may, there’s a good case to be made that the GOODNEWS factor is already underestimated. Fair notes that so far there have been three quarters during the Bush administration in which per capita GDP growth has exceeded 3.2 percent but there were two other quarters where it came within a hair’s breadth of beating the threshold. Change the GOODNEWS factor from three to five, and click the “submit” button Bush wins with 60.4 percent of a two-way vote.
But what about the model itself?
One possible risk element is Ralph Nader. Fair’s model just looks at the votes received by the two major parties, ignoring third parties or independents. Traditional polling and common sense confirms that Nader is more likely to draw votes from Kerry than Bush, so this particular blind spot in the model probably means that Bush’s chances are even better than they already seem.
The most profound risk to the model is that voters will be swayed by issues that have nothing to do with the model’s inputs, which are all oriented around the economy. It only makes sense that passions about the war on terrorism and the occupation of Iraq could crowd out the economy in many voters’ minds.
That said, there have always been important issues that have nothing to do with the economy, yet over many years the model has proven to be quite successful nevertheless. Will this time be different? Could be, but there’s no particular reason to think so.
Fair thinks the most relevant risk is the potential gap between perception and reality about the economy. Fair says that his model’s prediction of a Bush victory in 1992 was thwarted despite a recovering economy, because public perceptions of recovery lagged the reality.
This time around polling data suggest that the same thing might be happening, with voters’ negative perceptions of the economy strikingly at variance with its true health. But Fair notes that the current economic recovery has already been underway longer than the one in 1992, and the next election is still six months away.
Fair’s model is no political deus ex machina, but it has the virtue of grounding our subjective appraisals of a very emotional matter in solid historical reality. With the beating that George W. Bush is taking every day in the liberal media over real and imagined problems in Iraq, Fair’s model may go a long way toward explaining why Bush’s poll numbers are staying surprisingly strong, and Kerry’s surprisingly weak.
Think about it this way. If you were going to bet on the election, whose opinion would you trust? The rigorously proven econometric models of Prof. Ray Fair, or the editorial page of the New York Times?