If you don't understand, don't tinker

Russ Roberts writes:

How can any economist today argue for say, a stimulus package, with any confidence? Or a further lowering of interest rates by the Fed? ... Doesn't the current situation and the inability of macroeconomists to predict it (or to have any certainty about whether we are going to have a mild recession or a serious Depression) suggest some humility?

Someone could reply that this works both ways. If we really can't predict whether an intervention will help or hurt, then by the same token we can't predict whether failure to intervene will help or hurt. This symmetry can then be used as a license by those who are inclined to intervene.

But the real situation is not symmetrical. A blind intervention is (on average) harmful. We can clearly see this in familiar cases. Blindly hitting keys while editing a document will harm the document. Blindly cutting into a patient will harm the patient. Blindly operating on a car engine will harm the engine. Blindly drinking random chemicals will kill you. Blind activity is harmful.

Blind action is more harmful than no action at all. Russ Roberts is pointing out that interventions are blind. It follows that the expected outcome of these actions is to harm the economy.

Share this

20/100 vision?

Blind activity is harmful.

One could argue that having "some" economic knowledge is more akin to having 20/100 vision rather than being completely blind. Activity in this state may be less damaging and possibly successful. This isn't my argument, but I haven't a great counter for it.

It is sheer speculation

One could argue that having "some" economic knowledge is more akin to having 20/100 vision rather than being completely blind. Activity in this state may be less damaging and possibly successful.

"May be" and "possibly". That's speculation. There isn't any knowledge here. If we don't know whether we know but are forced to speculate that we might know, then we don't know. If someone comes at you with such speculations, that's one answer. If they tell you that they do know - well, that's what Russ Roberts was answering, the hubris of the economic expertise.

And there is a threshold of knowledge below which it remains preferable to do nothing even though the knowledge is not zero. If you took a course in physiology I would still prefer that you not perform surgery on me, because mere book knowledge of physiology does not (in my estimation) rise to the level at which surgery will on average do more good than harm. I want you have practiced on corpses so you can have familiarized yourself with the layout.

So, what is the threshold for knowledge of the economy? We don't know. We are reduced to speculating that somebody might have passed that threshold. Actually I am being too nice. The failure of economists to predict events that Roberts points out suggests that they have not reached that threshold.

Russ Roberts's point is that the experts know less than they think they know, and ought to correct their self-assessment. They ought to be correspondingly more reluctant to consider themselves to be above the threshold. And the result, if they do drop themselves below the threshold, is not to then assign a 50/50 probability that an intervention will work as an expression of their humble uncertainty. I've argued that a 50/50 assignment as an expression of total uncertainty is in this case way off the mark. A blind intervention is almost certain to make things worse.

"The law of unintended consequences" virtually guarantees that there is a great deal of blindness to any change, even a change about which one "knows" many things. It makes it hard to be sufficiently expert, even if one is "expert" relative to others, to make a change which is not almost certain to make things worse. "Expertise", remember, is a relative statement: it means only that one person is more expert than the layman. But relative expertise is not enough to make a person sufficiently expert to intervene non-blindly into the market. Whether that intervention is blind is not a function of a person's relative expertise - and it is his relative expertise which earns him the label "expert". So there is no guarantee that the experts can make a significant change that is not blind. And if I were betting on this, based on their record so far I would say that any significant market intervention proposed by our best economists is virtually guaranteed to make things worse.

Don't all your examples have

Don't all your examples have to do with blind interference with a designed system?

The document has assumedly been prepared by someone else.

Natural selection has effectively designed the body to survive.

And of course the engine has been created by someone else.

With all these we have preexisting systems engineered towards some purpose: the body to live, the document to communicate, the engine to combust. Is it your position that the macroeconomy has been similarly designed? To what end? By whom? The invisible hand?

I suppose one might argue that its mere continued existence is testament to its fitness--as one would with the human body.

Though I prefer the previous commenter's counterargument.

disaster ensues

The market economic order is designed - it is designed by each individual. The idea that some overall planner working from the greater good will know better than the "blind anarchy of the market" - otherwise known as the invisible hand - has been tried many times, and has a pretty uniform record of utter disaster.

The bailout plan changes abruptly every day - a process reminiscent of the early Soviet efforts to implement socialism. It is manifest that these people have no idea what they are doing, and if one reads economists, economists who support the bailout, they have no idea either.

The disaster will of course be used as evidence that the intervention was vitally needed, and more intervention is required.

Is designed

James is right. Thanks, James. The first sentence answers Scott's objection.

If you're willing to talk about an economy in terms of efficiency, then a blind change to the economy will, with probability close to 1, reduce the efficiency. If, like me, you are not all that happy with the concept of efficiency, then you can phrase the same essential point as follows: a blind change to the economy has the expected result of harming a given participant in the economy.

There is a general reason why this is so - why random changes to things which have already gone through some process of optimization cause harm. In the space of all possible states of a thing, only a tiny minority of states are noticeably improved over the vast majority, and only a tiny minority of that tiny minority is significantly improved. That's not a necessary truth but it's true in real case after real case. A random change to a significantly improved thing, which shifts the thing randomly to another nearby state, will tend to undo the improvement simply because the vast majority of possible states are worse than the significantly improved state. This general point is applicable to both documents and economies.

To see this for the economy: The economy as a whole is too complex to even begin to imagine, but imagine that just a sliver of the economy has been randomized - a state has been selected at random from the space of all possible states for that sliver. For example, imagine that there was a major mixup in a cafeteria, and people got other people's orders, selected at random. Somebody who wanted black coffee got orange juice instead, someone who ordered a hamburger got pancakes, and so on. The probability that any given person's lunch has been improved is low. The probability that a given person's lunch is less satisfactory to him after the random reassignment of lunches is high. The probability that the average level of satisfaction, however you might calculate it, has gone down is very close to 1. Realistically, a random reassignment of food orders is not going to make things better, is not even going to keep things the same. It's going to make things worse. But to say this is nothing more or less than to say that in the space of possible lunch assignments, the vast majority of possible assignments are worse than the actual assignment which was produced within the market by people buying lunch for themselves.

You can look at other slivers. Just look at one person who ordered pancakes. Now suppose that the ingredients of his order have been randomized. Instead of containing flour, baking powder, sugar, salt, oil, eggs, and milk, the "pancakes" now contain 6 random ingredients selected from the kitchen shelves. Include the dishwashing soap among the potential ingredients and you have a recipe for likely disaster. (The pancake, which is part of the economy, is a designed thing in the sense Scott brought up and illustrates the point that the economy itself is a tightly interconnected web of individual designs. Blindly act on the economy, and you have blindly acted on the web of individual designs and therefore you have blindly acted on the individual designs. A market is a collection of transactions, and each transaction is part of the designs of each of the two participants. Interrupt the transaction - an example of an intervention - and you have intervened into two designs. And since people's designs in the market are largely about assisting other people with their designs, if you intervene in one person's designs, you have thereby intervened in the designs of all the people he's working to serve.)

We are precariously balanced on a lonely spot of optimization in a sea of mostly chaos. Almost any change is going to make things worse. This conclusion is robust across different ways of defining "better" and "worse", because all the definitions need to satisfy is something the likely candidates do satisfy: they are significantly positive only over a minuscule fraction of the state space (by which I mean phase space, the space of all possible states of a thing). It's not a necessary truth from purely formal considerations (that I have found), but once you look at the specifics it's true.

There is a tendency, in the face of ignorance, to assign 50/50 probabilities to things. If you don't know whether a coin is biased, then you might reasonably assign it a 50/50 chance of coming up heads - not as a claim that the coin is fair but as an expression of your own subjective degree of certainty. 50/50 expresses "complete uncertainty".

What I am arguing is that in the case of blind mutations to already-optimized things, this is way, way off. The chance that a random mutation is going to improve you is very small; the chance that a series of random mutations (without selection) is going to improve you is vanishingly small. If you express your own total lack of knowledge about a blind intervention into something by giving it a 50/50 chance of making things better (by whatever measure you have chosen), then that's a major mistake.