Economic modeling is far more Art than Science. I know, cause I do it for a living.
You set up your parameters (which builds in your own biases), and then back test it against history. Make some minor adjustments. Run it again. You then tweak it, by de-emphasizing the data points where results are weak or ambiguous. You emphasize what does work.
Rinse, lather, repeat.
You may never get a perfect fit, but you can get close. Even if you do manage a perfect historical match — the problem is, you are looking at history. There are no guarantees that the future will be close, less identical, to the past. As Twain so eloquently put it, “History doesn’t repeat, but it rhymes.”
That’s why economic models are so good at predicting the past, and less good at predicting the future. Of course, predicting the past is much easier.
Imagine it via this analogy: The dinosaurs had an excellent model of survival predictions, hardwired (via Darwin) into their DNA. It worked great — right up to the moment that 10 mile asteroid slammed into the planet. After that, their “model” was a steaming pile of Dinosaur shit.
Economic models are no different. Ray Fair’s election forecasting is a perfect example — not only did it fail to predict Bush 41’s loss, it did so during a somewhat similar economic period of a jobless recovery. Note however, that there are too many dissimilarities today to 1992 to list — 9/11, major 3rd party candidate then, major post bubble environment at present. That 1992 failure may not be significant. Or, it might be preiscient about Fair’s model failing in 2004. You cannot tell for certain in advance.
Welcome to the wonderful world of modeling.
This is true whether we are discussing stock market models, consumer purchasing patterns or presidential elections. History can provide fascinating insights, patterns and guidelines, but it does not foretell the future. If it did, acting on that knowledge would eventually impact prices — or voting patterns — and alter future patterns. Hence, any advantage the provides would disappear.
In our prior discussion “Presidential Job Worries,” we essentially created two simple single variable models. We could create more complex ones, with 100s of variables — but why bother? They are mere academic exercises. Its amazing that the media trots out these “seers” as if they have special powers. The modellers themselves have drunk their own cool aid. You rarely hear sober caveats about why this is not a hard science or the limitations of these forecasts. If the President loses, the best part of his defeat would be never having to read any of Ray Fair’s cocky ravings again.
Billmon at the Whiskey Bar points to several articles (see below) which rely on various flavors of modeling to predict Presidential victors. He gives the whole subject a solid fisking, and if you are at all interested in econ modelling, his take is well worth reading.
UPDATE: March 7, 2004 6:51 am
I just came across an MSNBC article, “Why 2004 election will defy history.” The author points out 3 specific historical “patterns that have been shattered already, in the nominating season just ended.”
Patterns such as these form the basis for economic modelling. Looking at 3 patterns for th present election:
1) In modern times — since the advent of contested primaries — the major parties always had nominated the guy who collected the most cash and who led in the Gallup Poll by the end of the year before the voting began.
This time around, of course, that guy was the unstoppable Gov. Howard Dean. He had raised an unheard of $40 million and led in all the national polls — not to mention the local polls in key “early” states, such as Iowa and New Hampshire
2) Iowa caucuses aren’t that important or predictive. Winners there — people such as George H.W. Bush in 1980, Dick Gephardt and Bob Dole in 1988 — tended, on balance, not to go on to win the nomination. This time, in retrospect, it is clear that Iowa was the ballgame. Kerry and his strategists bet it all on Iowa, and they were right.
3) The 50 percent approval factor: Incumbent presidents lose if their job-performance numbers dip below 50 percent as the campaign season begins. Bush’s is hovering at that level, depending on which polling survey you watch. The corollary: No incumbent president can win re-election if the Dow Jones and the employment numbers are lower than they were when he was sworn in.
Bush still wins prediction
By Martin Gottlieb
Dayton Daily News
March 5, 2004 11:48 PM
Why 2004 election will defy history
Patterns of past wouldn’t allow Bush or Kerry to win
By Howard Fineman
MSNBC/Newsweek, March 05, 2004