« Introducing the Author | Main | John Taylor, and is Spending Really Better than Tax Cuts and Transfers? »

August 24, 2011



There's a whole class of deterministic chaotic systems which are also too complex to predict, the most famous example being weather. There's a great many true things we can say about weather without being able to accurately predict it past a few days. The equations that we use to describe particle interactions in an accelerator's collision chamber formally include randomness such that experimental results aren't strictly predictable even in theory. So, I am not sure that lack determinism in the problem domain sets economics apart.

But I think that Nathan puts his finger right on it when he says that making a prediction changes things in economics, as this is true in a way that is not so in any of the natural sciences*. Even if we could come up with rigorous, testable hypotheses about a certain dynamical scenario that actually exists in the world, it's bound to shift over time. Worse, even if status quo bias was strong enough to keep economic systems steady-state as long as there was no major pressure, the very act of proving the predictive ability of a model would surely overcome any such bias as market actors moved to take advantage of the predictive tool and so changed everything. Alternately, a self-reinforcing claim that's actually based on mistaken reasoning or evidence may still create the effect it predicts by changing expectations to match its prediction.

While I wouldn't want to make all the claims Nathan does about the virtues of introspection, I do think it's a critical piece, because a straight natural-science approach would never be able to keep up with an object of study that changes in response to study. I'm not sure we'll ever be able to hang the word 'Truth' on any of economics' pronouncements in the same way, but as a tool for humans to manage our constantly-mutating affairs it is just as indispensable.

*There's been much made of the effects of observers in physics, but most of this is confusion, and anyway we're talking about prediction, not observation.


If your standard is 100% accuracy and 100% precision with zero variance, then you're right, we can't predict anything let alone human behavior. But normally when we talk about prediction it's with the expectation that the outcome will fall within some range of possible outcomes. If you cast the net wide enough, you can make predictions about human behavior that are correct 100% of the time (but if the net is too wide, the prediction will be of little value). The other thing to note is that most things that seem random can almost always be characterized by a non-uniform probability distribution. This means that even though, say, there's a small chance that I could walk through a wall due to quantum tunneling, the chance is so small as to be of measure zero, and so we can say almost surely that I will never be able to walk through a wall. Certain market behaviors can similarly be shown to have a measure zero chance of occurring. I recommend sitting in on a rigorous graduate class on random processes to understand these things better. Your claims about non-deterministic systems seem very naive to me.

Nathan Smith

Um, I *have* taken rigorous graduate classes about random processes. Aced them. :) It's true that I was speaking loosely in treating natural sciences as dealing with deterministic systems: they are better understood as random, although for many purposes they can be regarded as virtually deterministic, as you say. The difference is that one can define precise, permanent laws about the probability distribution of events. That cannot be done with economic systems. Some statistical descriptions have some degree of persistent, but there are fundamental reasons why the certainty and precision that is routinely acheived in the natural sciences is usually unattainable in the social sciences.


"there are fundamental reasons why the certainty and precision that is routinely acheived in the natural sciences is usually unattainable in the social sciences"

Agreed, but it seems strange to say this is because humans 'aren't deterministic'. That doesn't really seem to have much to do with the matter at all.


In any case the overall argument carries quite as well without the whole 'determinism' bit, which the original post already acknowledges.

Nathan Smith

Yes, the distinctive position on the demarcation of different types of scientists, though inspired in my case by a belief in free will, is consistent with philosophical determinism too.


Pardon me for being a bit of a badger on this, because I am a bit curious. It is not merely that we *need not* reference determinism as a distinction between the natural sciences and economics. The reason determinism does not seem to me to offer a distinction is because the natural sciences don't assume determinism qua determinism any more than economics does. Perhaps 'determinism' here is shorthand for 'operation under the idea of probabilistic functions'?

Nathan Smith

Well, it seems to me that the natural sciences *do* assume determinism, most of the time. The whole idea of running an experiment is that if you hold everything else constant and do *x*, and *y* occurs, you've proved that *x* causes *y.* But of course, you can say, and scientists nowadays typically would, that what is really going is that there is a probabilistic distribution of results, only they are very tightly clustered around *y,* and the chances that anything not recognizable as *y* will occur is small enough to be ignored. For other purposes, scientists would hold that there is a mathematically precise probabilistic distribution, and seek to describe it. The distribution could in principle be known to an arbitrary degree of precision. If that's all you mean by 'determinism is a shorthand for operation under the idea of probabilistic functions,' then yes. I don't think a typical scientist gives much thought to quantum uncertainty in his everyday work, though.


Any science studying complex dynamical systems will have to deal with probabilistic functions, however, including meteorology, biology, medicine... even the large astrophysical bodies get modeled without reference to particular entities and instead get represented by random elements that fit the distribution curves of the body to be modeled. Which is to say that regardless of how supposedly deterministic a natural science supposedly is, it's rare that the matter under study is deterministic in a first-order way. "Galaxies in domain X will tend to do Y because of dynamic Z" go the findings, or "populations of size A tend to have allele drift at rate B." I think working natural scientists spend a lot of time worrying about whether results are outside the error bar. The results are supposed to take a clear causal form or they're hardly interesting, but the idea that they have to be deterministic in the philosophical sense hardly has any bearing at all.


The fact that there are highly non-uniform and tightly clustered distributions in nature should give us quite a bit of confidence in inductive reasoning. Not 100% confidence, but enough to allow us to do pretty amazing things with Mathematics, Science, and Engineering (heck, even Economics). Deterministic natural laws are the result of the realization that even though we can't really predict the future, we can make pretty good guesses due to the nature of the randomness that we observe.

abercrombie & fitch outlet

i like your website very much,well done

The comments to this entry are closed.

My Photo

Only use a payday cash advance as a last resort.


Blog powered by Typepad