Courtesy of: NATE SILVER
November 22, 2010
Saturday’s Times had an article by Jeff Sommer that highlighted the work of Ray C. Fair, a Yale economist who has designed a model to predict the outcomes of presidential elections based on a series of economic variables.
According to Mr. Fair’s model, President Obama could be in surprisingly good shape in 2012. Based on Mr. Fair’s assumptions about economic growth (he expects gross domestic product to improve at an annualized rate of about 3.7 points in the first three quarters of 2012) and inflation (which he expects to remain fairly low), Mr. Obama would be expected to receive about 56 percent of the vote in 2012 and the Republican candidate 44 percent (excluding any votes for third parties). By modern standards, that would constitute a landslide, with Mr. Obama probably winning in the neighborhood of 400 electoral votes.
Needless to say, there are a couple of issues with this, many of which Mr. Fair is quite happy to acknowledge, once one reads the fine print behind his forecasts.
First, there is a lot of uncertainty about what the economy will look like two years from now. Forecasting the economy even one year in advance is quite tricky. A series of expert forecasters surveyed by The Wall Street Journal, trying to anticipate the state of the economy in December 2011, predicted that the unemployment rate would be from 7.6 percent to 10 percent, that G.D.P. would grow from 1.8 percent to 5.5 percent, and that inflation would run 0.6 percent to 3.8 percent. And that only gets us through next year, to say nothing of what could happen in 2012.
Obviously, the upper and lower bounds of these ranges could make the choice fairly clear for swing voters. It is hard to imagine Mr. Obama failing to be re-elected if the economy is humming along at a 5 percent growth rate by this time next year, and 600,000 jobs are being created every month. Likewise, if unemployment is still at 10 percent or if there is a second period of recession brought on by, for instance, the European debt crisis — something extraordinarily might have to occur for him to win a second term.
But there is a lot of middle ground, and here’s where models like these might not be much help.
Mr. Fair’s model predicts, for instance, that if G.D.P. growth were just 1 percent in 2012, and there was just one more “good news” quarter between now and then (one in which G.D.P. growth proceeded at an annualized rate of 3.2 percent or higher), Mr. Obama would nevertheless receive about 50 percent of the vote — that is, he would have an even shot at re-election, depending on the vagaries of the electoral college. That intuitively does not seem right to me: while the economy would not technically be in recession, it would certainly be failing to meet voter expectations. While Mr. Obama might have some shot at re-election if he had had a major foreign policy success or two, or if a controversial candidate like Sarah Palin were the Republican nominee, it is likely he would be a clear underdog under such circumstances.
It could be that my intuition is wrong — the incumbency advantage has often been quite powerful in presidential elections.
But there is a potential missing link in the form of unemployment, which is not included in Mr. Fair’s model. Unemployment, after all, has been the central frame for discussion about the economy in recent months. The word “Obama” has appeared in conjunction with “unemployment” in about 26,000 news articles since the start of the year, compared with about 8,000 articles that mention both “Obama” and “inflation”, and 6,000 that mention both “Obama” and “G.D.P.”
Why isn’t unemployment included in the model? The literal answer is because, in designing his model originally — way back in 1978 — Mr. Fair found that it did not much improve the accuracy of his predictions once other variables like G.D.P. were accounted for.
But here, things get a little bit more complicated.
Historically, employment and G.D.P. are quite strongly correlated: when the economy grows over all, it also creates jobs, and vice versa. This contrasts with something like, for instance, real (inflation-adjusted) G.D.P. and the inflation rate, which are weakly correlated: there have been plenty of periods of high growth and high inflation, of high growth and low inflation, of low growth and high inflation, and so forth. So, if you knew what G.D.P. growth was going to be, it would be redundant to account for employment: by knowing G.D.P., you already had a pretty good idea of what employment was.
During the most recent recession, however, the strength of this relationship broke down. Clearly, the economy was not in good shape by any measure throughout most of 2008 and 2009. But it lost quite a few more jobs than would ordinarily be predicted based on the downturn in G.D.P. alone.
The graph below presents a version of Okun’s law, which is an empirically observed relationship between employment and G.D.P. growth. On the horizontal axis is the change in real G.D.P. from one quarter to the next (the values are not annualized, which is why they might seem small), and on the vertical axis, the change in nonfarm payroll employment, which is probably the most reliable of the several statistics that measure the employment situation. The relationship is not perfect, but it is generally fairly strong, particularly given that economic data can be quite noisy.
The next chart presents exactly the same data, except that it highlights the four quarters in the year 2009. In each of these quarters, the decline in employment was quite a bit larger than what Okun’s law would hypothesize. For instance, in the fourth quarter of 2009, when the economy grew by a decent clip, Okun’s law would predict about 250,000 jobs to be created; instead, about 450,000 were lost.
Over all, the economy wound up having about 4 million fewer jobs by the end of 2009 than would have been expected based on Okun’s law. (It has not yet begun to make up for the ground it lost; the three data points thus far in 2010 have been fairly consistent with Okun’s law, but job creation has not run ahead of pace, as it would need to do for employment to “catch up.”) If the economy now had those jobs in place, the unemployment rate would be somewhere on the order of 7 percent rather than 9.6 percent right now, and we might have a very different economic — and political — picture.
Why was this relationship violated? A number of theories abound, ranging from longer-term structural trends to poor definitions of G.D.P. to the questionable design of the stimulus package to the particular nature of the economic crisis that the country faced.
It is not necessary for our purposes to decide upon an answer: the point is simply that G.D.P. — as it is usually measured — did not mirror employment the way it normally does.
That is problematic for models like Mr. Fair’s. What we’d like to see is what happened in the past when there had been a similar discrepancy: when the economy performed all right according to G.D.P., but very poorly according to employment. However, since violations of Okun’s law are fairly rare, and since presidential elections are also fairly rare, there are really no good data points to work with: an econometrician would say the solution lies “out of sample.”
To the extent that there is a relevant example, it’s what took place this month — and it isn’t an auspicious one for Mr. Obama. A version of Mr. Fair’s model can also be applied to Congressional elections. It had predicted that Republicans would win about 51 percent of the two-party vote on Nov. 2 (excluding any votes for minor party candidates), when in fact, they won closer to 54 percent. While that isn’t a huge discrepancy, it does suggest that today’s voter — faced with lukewarm G.D.P. numbers, but awful employment numbers — will tend to focus on the latter variable, and punish the party in power more rather than less.
While one could probably publish an alternate version of Mr. Fair’s model that used employment in lieu of G.D.P., it would still face some problems. For instance, how to measure unemployment? Is it based on the change from the previous period? (E.g., unemployment has increased by 4 percentage points since President Obama took office?) Or is it based on the absolute level of unemployment? (E.g., the unemployment rate is 9.6 percent?) In 2010, this might not have made much difference: the unemployment level was terrible both in absolute terms, and relative to where it had been previously. But in 2012, the distinction could matter a lot more. If the unemployment rate is, say, 7.6 percent when voters go to the polls in November 2012, will President Obama get credit for having reduced it significantly from its peak? Or will he continue to receive blame because it is still fairly high relative to where it is ordinarily? (And higher than when he took office.)
There simply aren’t good answers to these questions in the historical record. Mr. Fair’s studies cover 24 presidential elections, since 1916. Even assuming that all of these data points are relevant (and they might not be, when one considers — for instance — the difference between wartime and peacetime elections, between incumbent and nonincumbent elections, and that the behavior of the electorate has changed in some significant ways since 1916), that’s not really enough data to permit much in the way of subtlety when deciding among several different versions of several different economic variables. This is particularly so when many of the variables are strongly correlated with one another, which creates both theoretical and practical problems for any sort of regression analysis, which is the technique that these models use.
I don’t mean to suggest that these econometric models are of no value. When Mr. Fair introduced his model in 1978, it was quite revolutionary in identifying a relationship between economic performance and voting behavior; some of the scholarship at the time had (wrongly) concluded that there wasn’t. Mr. Fair, moreover, is an outstanding economist, and has been careful to declaim exactly some of the limitations that I have identified here.
But given all of the different ways to measure the relationship between economic performance and voting behavior, there are too many ambiguities to come to any terribly specific conclusions. Mr. Obama, of course, will do better if the economy is performing better rather than worse. But exactly how well he will do — enough to win re-election and, if so, by a comfortable margin? — is not something the data can tell us right now.
No comments:
Post a Comment