How can one be certain about outcomes in a complex system that we’re not really all that good at modeling? Anyone who’s familiar with the history of macroeconomic modeling in the 1960s and 1970s will be tempted to answer “Umm, we can’t.” Economists thought that the explosion of data and increasingly sophisticated theory was going to allow them to produce reasonably precise forecasts of what would happen in the economy. Enormous mental effort and not a few careers were invested in building out these models. And then the whole effort was basically abandoned, because the models failed to outperform mindless trend extrapolation — or as Kevin Hassett once put it, “a ruler and a pencil.”
Computers are better now, but the problem was not really the computers; it was that the variables were too many, and the underlying processes not understood nearly as well as economists had hoped. Economists can’t run experiments in which they change one variable at a time. Indeed, they don’t even know what all the variables are.
This meant that they were stuck guessing from observational data of a system that was constantly changing. They could make some pretty good guesses from that data, but when you built a model based on those guesses, it didn’t work. So economists tweaked the models, and they still didn’t work. More tweaking, more not working.
Eventually it became clear that there was no way to make them work given the current state of knowledge. In some sense the “data” being modeled was not pure economic data, but rather the opinions of the tweaking economists about what was going to happen in the future. It was more efficient just to ask them what they thought was going to happen. People still use models, of course, but only the unflappable true believers place great weight on their predictive ability.
Megan McArdle, “Global-Warming Alarmists, You’re Doing It Wrong”, Bloomberg View, 2016-06-01.
February 15, 2018
QotD: Computer models
Comments Off on QotD: Computer models
No Comments
No comments yet.
RSS feed for comments on this post.
Sorry, the comment form is closed at this time.
