(On Monday, I started a conversation concerning brand-new-Nobel-Laureate Tom Sargent’s admonition that, to model situations that are dynamic, uncertain, and ambiguous, the proper course of action is to roll up one’s sleeves and learn and use some mathematics. In starting the conversation, I said that I agreed with Sargent, but I reserved my reasons until others had commented. My reasons–…

## Want to comment on stories like these? Become a member today!

### You'll have access to:

- All Ricochet articles, posts and podcasts.
- The conversation amongst our members.
- The opportunity share your Ricochet experiences.

I worry that the main ‘feature’ of a model – mathematical precision – is also its biggest potential liability because it is a false precision if the model itself is incomplete or wrong. Garbage in, garbage out.

Friedrich Hayek’s criticism of the over-reliance on mathematics in economics comes to mind. An economy is a complex adaptive system full of very smart actors who do not necessarily behave in predictable ways, and whose responses will change as they assimilate information. The actual structure of the economy and the billions of inter-related factors is not knowable to people looking from the top down. The information required to understand it is locked up in the heads of economic actors.

There are ways to model these kinds of systems, but such models are more along the lines of what you described as models that are not used to predict specific outcomes, but to help understand the behavior of the system in order to learn more about it. I do not believe this includes the simple Keynesian models currently being used to predict things like the number of jobs saved/created when X dollars of stimulus are injected into the economy.

Just by way of background – I spent 23 years doing math modeling – 6 in electroquasistatics, the rest in torpedo defense and underwater acoustics. Although the math in those is technically more difficult than economics, the problems are infinitely simpler. That’s because there are few, if any, feedback mechanisms, and the underlying statistical structures are generally static and well known. Because the problems are “well-behaved”, one could, by combining classical science, statistical methods, and frequent recalibration, achieve useful results in reality.

Now my experience in mathematical economics is limited – I used to pick up grad level textbooks and read through them for interest and the occasional mathematical challenge, and I enjoyed adapting various Bayesian algorithms to trading systems – but I did get familiar enough to realize that at least at the grad level, there’s huge holes in the mathematical approach, particularly with respect to assumptions regarding the underlying distributions. Nassim Taleb is a wonderful, non-technical read on this (for those who are interested). Those fat tailed distributions (the unknown unknowns) always get you in the end.

(continued)

Because of this, I would say that the real orthodoxy to be addressed is the assumption that mathematics is, in fact, a suitable tool at all. My own sense is that it is not (at least above the simplest, illustrative level). Even rule-of-thumb folk philosophy seems to work better.

Don’t put your eggs in one basket would have considerably helped the Nobel Prize winners at LTCM, for example.Don’t borrow above your means would have helped many immensely. And of course,TANSTAAFL ought to be tattooed backward on the forehead of every politician, so they can read it as they gaze in the mirror.But math is sexy these days, and it tends to attract the same way Latin Mass did. (Who knows what they are saying? But it sure sounds good…).

.Now maybe someday economics will become a real science, and the problems mathematically tractable – say along the lines psychology is just beginning to become. But in the meantime I know I would be a lot happier if the math was kept to the undergrad level, and the real arguments allowed to take place along philosophical, historical, and moral lines.

Wow. That is the most fascinating, informative, and lucid discussion of mathematical modeling I’ve ever read.

Thanks, David. But be careful. Tuition at Stanford business school these days is a

lot. You’re going to undercut your own institution if you start giving education away for free.To Blue Ant: What do you mean by “prediction?” I think it means something less precise than you do. To give an example: When I teach microecon to MBAs or execs in our Exec Ed programs, my favorite case concerns GE and Westinghouse struggling to restrain themselves from engaging in cutthroat competition. What is stopping them? Halfway through the case discussion, we stop and play/analyze a pretty simple game (repeated prisioner’s dilemma with noise). This, to me, is a (mathematical!) model of the situation facing GE and Westinghouse. It is very, very far from the situation they face, but it captures beautifully some critical aspects of their situation, and it allows the students to predict: IF GE can find a way to strip out the noise, GE and W can probably do a lot better (and consumers a lot worse—important insights for the Antitrust Division of the DoJ). (And, btw, GE found a stunning way to do just that.) If that’s a model (and it sure is mathematical) and a prediction (NOT mathematical), we’re using math to make predictions in a very fruitful way. Which, I contend, is precisely what Tom Sargent has in mind.

You’re right that non-precise mathematical models are useful for conveying qualitative information… but that already moves us away from the kind of precise models which engineers distrust, which Hayek critiqued, and which central bankers borrow from modern macroeconomics.

It may be more accurate to say economic relationships can be expressed using mathematical

notation, not models, to convey useful information. Examples include the IS/LM equations, the Laffer curve, and the geekiest shirt ever.One of my earliest “aha!” moments was in my high school economics class**. The teacher asked me to draw “the supply and demand curves” on the board. But when I picked up the chalk, he announced it was a trick question: there’s no such thing as

the curves, they differ with each good produced. The important thing is how the two relate to each other; the rest is just simple graphing exercises with arbitrary numbers.If you’re saying use models to

representtheories, I think we’re in agreement. If you’re saying use mathematical models for prediction… isn’t that copying Arrow’s weather unit?**Do they still teach economics in high school? If not, the OWS protests start to make sense.

The models you’re describing are very different than the climate models or the macroeconomic models that HalifaxCB and myself are talking about, though.

I don’t think you’ll find anyone here arguing against the general concept of mathematical modeling or the utility of using mathematical models to test certain scenarios where boundaries and behaviors are well known, or where you are searching to understand the general behavior of a system in a testable way.

So I don’t have a problem with mathematical models, I’ve used them many times. What I have a problem with is the use of them where they are not applicable or where there are too many unknown variables, feedbacks, or potentially chaotic responses to inputs, and then to use the apparent precision of the models to pretend that you have a deep understanding of what’s going on. This is especially problematic when those models are applied to systems for which there are no controls to test against or no way to determine what was wrong if reality deviates from the model.

Another problem I have with models of complex systems is that they necessarily require the system to be simplified in order to make it tractable to modeling. In macroeconomics, we have the use of aggregate measures of supply, demand, labor, and money. Climate modeling also involves numerous simplifying assumptions.

Friedrich Hayek argued against the use of aggregates in macroeconomics, because he believed the microeconomic factors were critically important, and by aggregating them you hide the information you need to actually understand the economy. Furthermore, by injecting fiscal stimulus or other changes into the economy by a central authority, you would damage the underlying fine structure of the economy but this damage would never show up in a model that reduces economic data to aggregate values.

More than one wag has likened the use of aggregates in economic modeling to a drunk looking for his car keys under a streetlight half a mile from the path he took from the bar to his car, under the theory that it’s the only place lit well enough to find them.

If we are going to model a system as complex as a national economy, agent-based modeling seems more appropriate to me.

“…Tom Sargent’s admonition that, to model situations that are dynamic, uncertain, and ambiguous, the proper course of action is to roll up one’s sleeves and learn and use some mathematics.”

As in chaos theory? Seems appropriate considering the social, political, and economic state of the world.