Supply and demand can explain anything. That's not great.

But the alternatives are worse!

You are reading Economic Forces, a free weekly newsletter on economics, especially price theory, without the politics. Economic Forces arrives weekly in the inboxes of over 10,000 subscribers. You can support our newsletter by sharing this free post or becoming a paid subscriber:

---

I like supply and demand. That may be a shocker. I know.

But one of the weaknesses of using supply and demand in practice is that if you’re halfway clever, you can always reverse engineer an explanation for what you observe in the world. Did you see prices go up? Go and look for something that causes demand to increase or supply to decrease, and you have your explanation. Whether the price went up, down, or sideways, some explanation out there fits the data.

For example, after every inflation reading, I could search through industries and “explain” what happened using my theory. I could probably get lots of TV appearances, too.

That’s obviously not great for a theory that one may want to be able to falsify. It leaves us no better off than pointing to the will of the gods as the cause of crop yields. This flexibility risks turning the core of economics into pure sophistry. In data analysis, there is the idea of overfitting. You concocted a model that fits the data perfectly but will never predict out of sample. This is easy to do with supply and demand.

Yet, I keep coming back to supply and demand. Why?

There are a few ways to avoid the problems of overfitting with supply and demand if we focus on predictions when we can. For example, if I read a news story about inflation over the past year, I may conjecture a story about how supply chain problems and the War in Ukraine caused the inflation. The theory I came up with is “supply shocks caused the inflation I see,” and that theory generates a prediction about what happened to output.

If it’s a negative supply shock, quantity must go down. I can then go and test that theory against what we observed with output. For some time periods, that theory makes sense. For some time periods, that theory is contradicted by the output data. That’s not the most ideal test of the theory, but it allows some sort of external validation of the theory.

While the constraints on theorizing with supply and demand are not perfect, we must always ask, “Compared to what?” It takes a model to beat a model.

The problem is that other frameworks or models often have even more degrees of freedom. For example, I’ve explained before how the standard monopoly model has all the flexibility (problems) of basic, competitive supply and demand but also another degree of freedom: market power. Demand, supply, or increased market power could generate increased prices. With more degrees of freedom, it’s more susceptible to overfitting.

An extreme example of flexibility that generates overfitting is when a theory has multiple equilibria. For example, in my Comparative Economic Systems class, I used to teach John Roemer’s theory of Marxian exploitation of workers. The opening chapters try to provide a formal model of how wages could be below what the workers produce. If you translate the words into math, you recover a model with a backward-bending labor supply curve.

If supply curves go out and back, they can cross the demand curve multiple times, implying multiple equilibria or predictions of the model. In the high wage equilibrium, the workers fully receive what they produce; in the low wage equilibrium, they are exploited.

The problem is that nothing in the theory explains why wages are high or low. No comparative static tells us when wages move from high to low. Wages just fall from the sky that way. Sometimes, the economy is at a good/efficient outcome, and sometimes, there is a coordination failure, which is bad. It’s a just-so theory, but they show up a lot in economics.

A way that we can see the problems with this type of overfitting is that coordination failure models are often highly sensitive to the exact timing of the model. If workers can find a way to commit and demand high wages, the model predicts high wages. If employers do the opposite, we get low wages. Which should we predict “out of sample”? We may have an intuition, but it’s not grounded in the multiple equilibrium world but in knowledge of the bargaining process.

I hope it is clear how these sorts of theories are completely overfitting. The theory can justify anything and predict nothing. But it can get worse. Multiple equilibria doesn’t stop at two.

In economics, there are many “folk theorems,” but The Folk Theorem is an “anything goes” result for repeated games. Good outcomes, bad outcomes, meh outcomes? They are all equilibria if people are sufficiently patient.

Folk theorems are important and show how something could possibly happen. For example, someone may say, “Collusion between won’t work because each firm has an incentive to undercut the other firms and win the whole market.” In response, someone could respond and show how there are equilibria with collusion in a dynamic game. The promise of future shared-collusion profits may outweigh the short-run gains from undercutting. The folk theorem says this is possible, so we need to be careful about overanalyzing the static game.

Again, it says what is possible. It doesn’t say which outcome is likely or which outcome was what actually happened. It shows collusion is possible. In a repeated game, the no-collusion outcome is also possible. No collusion is still an equilibrium of the repeated game. In that way, the theorem is not a theory in that it doesn’t give any way to predict collusion or not. Lots of things can happen.

This may seem like an arcane aspect of economic theory, but it matters for popular debates about inflation. If there is one topic we circle back to almost as much as supply and demand, it’s inflation.

For example, Isabella Weber and Evan Wasner have a paper that has received much attention. They argue that supply shocks allow firms to coordinate on higher prices. As I’ve said before, that’s exactly what supply and demand says. In that sense, their theory shares some attributes of supply and demand.

However, Weber and Wasner want it to say more than supply and demand. Prices rose more than predicted by supply and demand because there is “implicit agreement which can be coordinated by sector-wide cost shocks.” In that way, the price rise was more than we would have expected with competitive markets or standard market power models because this additional collusion kicked on.

Their model of implicit collusion relies on applying the folk theorem, which they don’t explicitly flesh out but point readers to a working paper by Anton Korinek and Joseph Stiglitz.

Again, appealing to a folk theorem is, in effect, saying anything could happen. It says collusion is a possibility. But collusion was a possibility before the supply shock, too. It is still a possibility right now while inflation fall.

The important part is that while supply and demand is overly flexible, the issue is worse with other models. Any story that relies on implicit agreements, folk theorems, and coordination failure arguments is likely unfalsifiable. There is no way to tell when firms move prices together due to collusion or due to normal supply and demand. The two outcomes are “observationally equivalent.”

Why did collusion start when it did? There was a supply shock. But do you know what other theory predicts supply shocks lead to higher prices? Supply and demand.

These models do not show us how a market switches from no collusion (low prices) to collusion (high prices). The folk theorem offers no means for the prediction of future prices. Will prices come back down? That takes a different model.

May I suggest—you got it—supply and demand?