Philosophical Problems with Prediction

Alex Rosenberg and Tyler Curtain recently wrote an essay criticizing the idea that economics is a science. While I am skeptical of any joint undertaking between Duke and UNC – where Rosenberg and Curtain teach, respectively – it’s important to dismiss, or severely injure, the notion that economics can – at a philosophical level – predict the business cycle.

An oft-lodged complaint of neoclassical macroeconomics was its inability to foresee the recent financial crisis and recession that followed. Many critics – left, right; idiotic, and lauded – complain that if economics wasn’t so tied up in senseless assumptions and adopted a “more scientific” approach to its methods its predictive capacity could be expanded.

Of course, critics admonish economists to become “more scientific” without explaining what exactly that means, but the vacuity of this claim is far deeper. The business cycle cannot be predicted by a credible model.

First, let’s distinguish logical predictions, from random frets of fury that are eventually vindicated. The former derive legitimacy from logical assumptions that are substantiated by reality, and hence are expected to hold true for the future. The latter comes from one man’s “intuition” (perhaps that credit booms will cause a crash) without  an accompanying model – and describes most “correct” predictions about the recent crash.

The only point of predicting a recession from a correct logical model is for it to remain credible among society. There’s no point in having a right model that no body believes (and if it’s right for long enough, people by definition of Bayesian revision, will come to believe this is true – so this distinction is an artifact).

Let’s say I’m a brilliant economist that finally got rid of those dumb assumptions. Agents are no longer rational, and I’ve even measured how irrational each of Earth’s seven billion inhabitants are so that I can fine tune heterogeneous tastes and preferences perfectly. And because, let’s say, P = NP I can tractably compute even the most complicated world’s to predict the economy. I have a perfect model founded on the most scientific and empirical of observations. Everyone has complete faith in my brilliance.

And let’s say one morning I forebodingly write in the Wall Street Journal that in six months it is 75% likely the United States economy will contract (YoY) by ten whopping percent. But until then it will continue to grow, or stagnate. Even my beautiful model is not invincible against what happens next.

Since my model can logically predict the recession, all firms expect consumer demand to crash in six months. To maximize expected profit, firms lay off workers and curtail all but the longest term of investments. (If I’m building something that is expected to generate profits in two hundred years, the near term horizon will not affect this decision).

Suddenly, regardless of what my model predicted would happen today, the economy tricks itself into a recession. This has nothing to do with the kind of model I write, the assumptions I make, the math I use, the abstractions I allow, the realities I ignore, or the ideology to which I subscribe. Any and every model, will definitionally fall prey to fickle human expectations. No amount of devotion to the scientific method can fix this.

At this point, maybe the critics retreat and say, but then why can’t economics tell us that fiscal expansion works (does not work). For one, there’s no conclusive proof that it doesn’t (does). But more importantly, let’s say a completely trustable model tells us that deficit spending in recessions is magical and brings the economy roaring back. When the government, following economists’ advice, expands its budget drastically, expectations will drive the economy into a boom. The model, its merits aside, becomes a self-fulfilling prophecy. 

A lot of this is common sense to anyone whose taken more than five seconds to think about what’s wrong with the economics profession. It’s foolish to expect models to be founded on the same, rigorously true assumptions – or adhere to the same scientific method – of physics and economics.

It is easy to appeal to human nature – to which economics is ultimately connected – as a cop out of the scientific method. But even in the most mundane of senses, an economist does not have the luxury of a model that is forever and always correct. Self observation produces confusing – even paradoxical results – which economists can never overcome. Sure, they can add frictions representing momentum of human expectations, but once we learn (formally) of this addition there will be yet another meta level expectational conflict.

This logic can be taken to an extreme. Not everyone believes a model and business decisions aren’t formed on economic theory. But ultimately it illustrates an important distinction between economics and every natural science.

Asking economists to follow the stylistic and epistemological guidelines of physicists and chemists is not the best way to improve the profession. In fact, it is impossible.

  1. Bryan Willman said:

    This can be summed up with a shorter observation.

    In the physical sciences, the entities under study do not have volition, and of course no incentive of any kind. The moon doesn’t change it’s mass to gain competitive advantage after the announcement of a theory of gravity.

    In economics, the entities under study by definition do have volition, agency, incentives, goals, and in fact win prizes of various sorts by responding to economic predictions (or regulatory changes.) You give several examples above.

    In the physical sciences, the entities under study are not constrained by cognitive limits – oxygen never forgets to bind with hydrogen because oxygen is tired or hungry.

    In economics the entities under study are constrained by cognitive limits – and so even a perfectly rational actor will miss many opportunities to make better decisions due to simple exhaustion of one sort or another.

    • This is true, but I think very different from what I’m trying to say. Even if there was no special volition, agency, or irrationality or we had a model that adjusted for every possible such problem possible economic prediction would be impossible because of metalevel expectations.

  2. OK, I get what you’re saying here, but doesn’t that just mean that your example 75% prediction of contraction is a poor model? I don’t understand why metaexpectations can’t be included in a model. Basically, why would your prediction not also be a function of the expected reaction of markets to the prediction…?

    • Because it’s turtles all the way down after that. Humans endogenize information from any predictive models into all action, to the extent they believe it is true.

      • But so is any infinite series, and Xeno’s paradox still has a limit…

        Point being, I don’t understand (from this) why in the infinite limit there would *necessarily* be no convergence to some sort of stable attractor/equilibrium.

  3. Great post, but I have a few qualms:
    + You have a lot of faith in rapidly-adjusting expectations. Add lags and your critique may not hold nearly as well.
    + There’s no inherent reason why a model can’t be kept secret –or perhaps shared with a few senior policy-makers (see the CIA for precedent of this principle).
    + Your critique isn’t generalisable to non-business cycle economics, which construes (a) the majority and (b) the most scientific (most data) econ. I feel like people get away with ignoring 90% of econ in these ‘is econ a science’ debates, which is a little of unfair. I have literally worked on strategic models of externality pricing which embodies your sort of critique.

    My problem with the ‘economics isn’t a real science because it doesn’t look like physics’ crowd is much simpler — it’s absolutely bizarre to claim that a science need look like physics. That excludes medicine (based almost entirely on empirics), psychology (the same), evolutionary biology (genetic drift is almost by definition impossible to model), chemistry (ask your chemist friends how often a computational or theoretical result corresponds to an actual result in the lab and you’ll get a real lesson on the limits of models) and every other non-physics science. I’d go as far as to say that physics is the anomaly, not general case, except I’m not even sure physics meets this sort of criterion (see: the n-body problem).

  4. There’s a difference between “there will be a recession on Thursday” and “this mechanic is closely associated with recessions so we should guide policy to deal with it appropriately”. It is perfectly reasonable to expect economists to understand the latter to at least some degree.

    • I have nothing against that. Certainly if some form of financial instability mess causes recessions, and a model predicts that, but because it does the process is self fulfilling, the instability is still the proximal reason and should be stymied.

      But that doesn’t mean failure to predict crisis is a good criticism of neoclassical economics or, similarly, that seeing the crisis coming is a good example of a successful model.

      • But that doesn’t mean failure to predict crisis is a good criticism of neoclassical economics or, similarly, that seeing the crisis coming is a good example of a successful model.

        I think that a failure to foresee and prepare for the key mechanics of the crisis is a failure to understand how the economy works and therefore a failure of economics.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s