Arnold Kling  

Of Models and PSST

PRINT
The Julian Simon Club... Today on the Eurozone Crisis...

Some comments on an earlier post raised a number of issues.

1. Shouldn't the economy just have stayed on its production possibility curve in 1929, rather than go through the Depression? Well, the problem is if it stays on the 1929 PPF, it never gets to the 1950 PPF. In any case, there really is no way to freeze a pattern of specialization and trade in place, given that demography is changing, capital is aging, and so on.

2. Can we model uncertainty using a stochastic simulation or somesuch? I don't think that solves the problem that concerns me, which is the "unknown unknown." In 1931, as tractors continue to replace farm laborers, the problem is to find something for those farm laborers to do. There is a lot of random exploration involved, but I don't think that specification of a probability distribution is sufficient to characterize the problem.

3. What about agent-based modeling?

In some sense, one ought to be able, by playing around with assumptions, to come up with an agent-based model that can produce just about any outcome you want. If I am correct about that conjecture, then this represents both a strength and a weakness. It is a strength in that properties that get ruled out of other models to make them tractable do not have to be ruled out in agent-based models. It is a weakness in that it can be difficult for an outsider (or even the model-builder) to evaluate the reasonableness of assumptions as well as the validity of the interpretation of the result.

I still think that the best model of equilibrium is the 2x2x2 model of international trade. But the PSST story is about an economy groping from one equilibrium to the next (or, if you want to make things really difficult to formalize, an economy never in equilibrium). Even more important, not all production technologies are known. So it is not simply a matter of searching for a price vector. The groping process includes trial-and-error of new methods within firms and creation of new businesses.

Maybe you could set up a simulation with a more complex pattern of international trade (nxnxn), where there are two equilibria. An initial equilibrium, with full employment but that does not make the best use of recently-discovered technologies; and a final equilibrium, which does make the best use, but which involves production arrangements that are not yet known and that differ considerably from those employed in the initial equilibrium. Every experiment with new production technologies takes time. Many experiments fail. etc.


Comments and Sharing


CATEGORIES: Economic Methods



COMMENTS (6 to date)
Alex Godofsky writes:

Shouldn't the economy just have stayed on its production possibility curve in 1929, rather than go through the Depression? Well, the problem is if it stays on the 1929 PPF, it never gets to the 1950 PPF.

Oh? The economy would have been simply frozen in time, forever?

Also, I'd rather have the 1950 PPF in the "no Great Depression" alternate history than the 1950 PPF in reality. I bet it would involve a lot fewer tanks...

Randy writes:

"...if you want to make things really difficult to formalize, an economy never in equilibrium..."

I think this is the best point. What "equilibrium"? And why would we assume that any given "equilibrium" is good? If, for example, we choose to believe that "full employment" is a good equilibrium, then why do we have child labor laws? - and why don't we require everyone to work?

Alex Godofsky writes:

Which required a bigger change in patterns of sustainable specialization and trade, Arnold: when we realized we were making too many consumer goods in 1929, or when we released millions of men from the Army and stopped devoting a huge fraction of our industrial capacity to building military equipment?

Lord writes:

One could as well expect the 1950 ppf to be higher. I doubt it since we were maxed out by 1950, but we wouldn't have to have WWII to get there.

Dave writes:

The nice thing about agent based modeling is that you can show under what assumptions certain results emerge, and then those micro assumptions can be debated/tested.

It's easy for modelers to share and discuss assumptions: they can just share their code. It might be harder for traditionally trained economists to follow, but the models tend to use relatively simple heuristics for behavior.

In the Origin of Wealth, the author shows a simple models where wide wealth inequality emerges, and another where there are sudden and drastic changes in patterns after long relatively stable equilibriums due to endogenous innovations (using evolutionary algorithms).

Nathan Smith writes:

re: "In some sense, one ought to be able, by playing around with assumptions, to come up with an agent-based model that can produce just about any outcome you want."

As someone with some experience in agent-based modeling, I don't think this is quite right. I definitely ran into some fiascos, plans that made sense in my brain but were impossible to implement in practice. But I think you're right that agent-based modeling lacks a corpus of established results and methods to provide the type of discipline that the Walrasian equilibrium tradition provides.

Also, while in principle it's possible to share code, it's much harder to decipher someone else's code than to decipher someone else's equations. If agent-based model results don't make a direct appeal to intuition, I'd say ignore them. But some agent-based model results do pass that test handily. Once you see the result, it's intuitively obvious; but without the model, you might never have seen it.

Comments for this entry have been closed
Return to top