Economies and elections
So I guess I depressed Peter in my comments on his class warfare piece, and enjoyed the conversation thereafter. As a separate post, I thought I'd give you a link to one of the veterans in using economics to predict elections. Prof. Ray Fair of Yale was one of those I read intently writing my dissertation in the 1980s, and he still maintains a model for predicting presidential and Congressional elections. He predicts a very close election, with Obama winning 50.3% of the two-party vote:
For a moderately growing economy, which the US model is now forecasting, the election is predicted to be close. The current US model forecast is probably somewhat more optimistic than consensus, but with slightly slower growth in 2012, the election would still be predicted to be close. If the economy suddenly starts to boom---say 5 or 6 percent growth---Obama would be predicted to win by a comfortable amount. If the economy suddenly goes into another recession---say minus 5 or 6 percent growth---the Republicans would be predicted to win by a comfortable amount. As of this writing the economy in 2012 looks like it will be ok, but not great, which means a close election---essentially too close to call.
The economic variables in Fair's model are growth in per capita GDP and inflation, along with the number of quarters of blow-out growth. That blow-out growth quarter is a forecasted on for the third quarter of 2012, consistent with the political business cycle story I told in comments on Peter's post. (Fair's model of elections is tied to a broader model for the economy that he has maintained for decades.) Fair notes that his forecast assumes extension of the payroll tax holiday to year-end 2012. If it had not happened, the predicted two-party vote share for Obama would have fallen to 49%.
At least there's one area for smiles: Democrats are expected in Fair's model to take 46.7% of the two-party vote share in Congress in 2012. But not extending it would have increased that share, which may be why the House Republicans agreed to it.
- Comment (7)
- · Quote
- · UnfollowFollow (0)
Comments:
Mar '11
Re: Economies and elections
The underlying fault of this analysis, and all others that rely on historical data, is that they don't see the "black swans" approaching. These happen about once a generation - 1932, 1952, 1980, 1994, 2010. They search for causation in correlations, a fundamental logical error.
Edited on February 19, 2012 at 10:15pmMay '10
Re: Economies and elections
I don't question his work - he is obviously vastly more learned than I. I do have some questions however:
1. Inflation calculations have changed over the years and today have little relevance when compared to the challenges faced by people trying to square their household budget. Does he use a constant inflation calculation, or does he assume no variance in perception?
2. Very high structural unemployment, oil over $100 barrel and looking to go higher, high commodity prices in spite of lower demand, Baltic Dry in a "challenged" state, Europe heading into recession, China slowing, Japan in recession, regulatory tsunami continuing unabated, mountains of money parked on the sideline, $trillion+ deficits as far as the eye can see - where exactly is the growth spurt going to come from? I don't see how QE3 in the form of target NGDP is going to produce it.
I'm not questioning the validity of his study, I just don't understand it.
Re: Economies and elections
Roy, within the link is a link to his latest research paper which shows the results of the equations he uses. You can evaluate how well he did later there. And to Steve, I acknowledge that equations that explain past elections don't necessarily predict future ones. That's true of all forecasting -- many of my models that I use to forecast are not the ones with the tightest fit to past data. As to Fair's model, here's the latest memo. His model shows very strong growth in the second half of the year (4th quarter is higher than 3rd.) I teach students how to tinker with his assumptions to make their own forecasts; his software permits you to do that if you wish.
Dec '10
Re: Economies and elections
Roy Lofquist: The underlying fault of this analysis, and all others that rely on historical data, is that they don't see the "black swans" approaching. These happen about once a generation - 1932, 1952, 1980, 1994, 2010. They search for causation in correlations, a fundamental logical error. ยท 1 hour ago
Edited 1 hour ago
Even when a model accurately reflects causation rather than mere correlation, the Black Swan error says that the real world can and will introduce factors that haven't been modeled because the modelers couldn't imagine them ever occurring. It's not wrong to build models, but it's wrong to assume that any model will be completely comprehensive and predict all possible outcomes.
Feb '12
Re: Economies and elections
I think the housing situation will play a big part in this election. Foreclsures are expected to increase this year and so Obama will concentrate on advocating taxing the evil banks to help out the honorable poor victims of underwater mortgages. The other will be fuel prices, every voter will be angry in that poll booth if gas is 5.00 dollars a gallon. I'll be interested to see Obama's response. He will at least be able to release reserves in order to bring down prices a bit. There are many in the 'middle' who will not vote for him again no matter what. It is definitely the republicans race to lose .
Jan '11
Re: Economies and elections
I took an Econometrics course in college; we use Fair's book to understand the concepts of prediction models and then attempted to use regression models to find other variables that might predict presidential election outcomes. We could choose popular vote outcome for our "Y=" or we could choose electoral-college outcome for our "Y=". I did the later. I turns out that from 1928 onward [most of the NIPA tables from the BEA only go back that far], the electoral college outcome from 28 years earlier was a statistically significant indicator of what the outcome would be in an upcoming presidential election and the magnitude of the effect was strong. 28 years ago Ronald Reagan took all but one state!
Edited on February 20, 2012 at 12:27pmMar '11
Re: Economies and elections
Models are useful for well behaved systems. They pretty well suck for non-linear systems with multiple network-like interdependencies.
For example, A determines B in some manner. It also affects C, D and E. C affects B and E, usw. The test of a model is accomplished by varying one parameter through its expected range of values while holding the others constant. The number of iterations required very quickly (factorialy) exceeds the computing capacity available.
Additionally, the uncertainty of the measurements (error bars) and the uncertainties of the posited dependencies quickly render the results extremely unreliable - at best.
Real world, useful, models are only deployed after extensive comparison with actual results. As we used to say, back in the days of steam powered computers, "our new computer can make a million mistakes a second".