Permanent disequilibrium models, plus agent based simulations, will eventually alter the way we see history
Simple equilibrium models are often popular because they tell a simple story about the world, but these models are inaccurate. After all, the world is never truly at equilibrium.
Possibly an unpopular opinion:
Permanent disequilibrium models, plus agent based simulations, will eventually alter the way we see the past.
Equilibrium models got going in the 1600s because the math is simple enough that people can do it in their heads. You don't need a computer for the kinds of equilibrium models that brought the West so much intellectual success in the 1700s and 1800s and early 1900s. But over-reliance on equilibrium models has probably damaged the way we understand human societies, as well as damage our understanding of certain biological processes. In particular, such models as:
supply and demand
Darwin's natural selection
The Prisoner's Dilemma
Simple equilibrium models are often popular because they tell a simple story about the world -- often a story so simple that most people can understand the model without actually doing any math. For instance, think of the complicated math that is involved if you take a course in population genetics, and compare that to the much simpler story that Darwin told about Natural Selection. Most people go through life with an understanding of Darwins's theory of Natural Selection, but they lack any understanding of the places where modern population genetics modify Darwin's model. Something like "promoter genes" have now been well documented, but would have been a shock to Darwin. And their implications for rates of mutation, or rates of successful adaptations, or needed population size to support a radiation into a new environment, all have some nuances that would have been counter-intuitive to biologists 100 years ago.
On some level, we all understand that permanent disequilibrium models are more accurate than equilibrium models. After all, the real world never actually reaches equilibrium, but rather, permanent disequilibrium is what we see when we look around at the real world.
The Nobel Prize winning economist Paul Krugman was asked why permanent disequilibrium models haven't become more popular, and he responded that the math was extremely difficult, whereas so far the insights were not that much more profound than older, simpler models. But I think that will change with time. As permanent disequilibrium models become more standardized, the math will also become more standardized, and so the application will become more simplified.
Paul Krugman is a champion of equilibrium models, in particular the ISLM framework. In the aftermath of the crisis of 2008, Krugman seemed to do excellent work explaining the difficulties of the zero interest lower bound, with reference to the ISLM framework. Some have noted that other economists, also using the ISLM framework, didn’t see things the way Krugman saw things, and so Krugman’s insights were really a product of the ISLM framework plus a long study of Japan’s stagnation. (I recall reading Krugman’s first essay about Japan’s stagnation, back in the 1990s, and thinking that Krugman was wrong, but events eventually proved him correct.) I think this reveals how much economics remains an art and not a science, it’s basic ideas are mere tools that need to be used by a good craftsman to be assembled into something useful. Indeed, some of the informal “wedges” and assumptions that economists make to try to make their models fit the data are the kinds of informal and ad-hoc improvisations that are not allowed in other sciences. Consider what John Ferguson wrote:
Coincidentally, the ISLM also came up at the LSE lecture that I mentioned last week. Olivier Blanchard was asked about the teaching of macroeconomics and while he said that post-graduates should be looking more into credit, leverage and financial stability, he said that for undergraduates he would look into extending the IS LM framework. Specifically, he said he would add a wedge between the interest rate set by the central bank and that faced by the real economy. This wedge, or spread, would be a function of the financial sector’s ability to function as an efficient transmission mechanism of central bank policy. So, can the two issues of IS LM disequilibrium and two distinct interest rates be brought together? What might this extended framework look like?
That treats disequilibrium informally, as something to be approached with a bunch of ad-hoc tweaks and bandaids. This reminds me of the way aerospace and jet engineers had to handle feedback and faster-than-sound chaotic air flow, before the emergence of chaos theory in the 1960s — treating the dynamic aspects of the system as a kind error margin that had to be fudged. Apparently Larry Summers suggested something more rigorous was needed:
In the question time of the LSE lecture, Larry Summers made some very interesting points. He spoke of his wariness of DSGE models (Dynamic Stochastic General Equilibrium models) and his concern that just adding more frictions to these models misses the main lesson from the crisis. I interpreted this to suggest that the Economics profession should also be exploring ‘disequilibrium’ models and frameworks. We need to understand things like the build-up of credit and leverage and how this might lead to a large crash like that of 2007 and 2008. Overall, it probably makes sense to pursue both paths, that is, put more effort into adding financial market frictions to existing models, while also developing theories that are prone to disequilibrium.
I am sympathetic to the argument that disequilibrium models make the math complicated, relative to the older and simpler models. Consider one of the oldest ideas in economics, the Law Of Supply And Demand. We've seen models where supply and demand were scalars, then vectors, then vectors that respond to each other in a way that takes us into the world of matrix calculus. Do we gain some new and important insight in exchange for the extra work we are doing? The point is subtle, and the case is hard to make. Possibly we will find it easier to model where the opportunities for arbitrage exist, since the whole system is constantly in motion. Anecdotally, my friends in fintech, working on HFT models, tell me they've worked on models such as this, in their effort to find new alpha signals. (Given that HFT arguably does harm to society, I realize this may not be much of an endorsement.)
It's also possible that people, and researchers, don't like permanent disequilibrium models exactly because they don't tell the kind of clean, simple, straightforward stories that equilibrium models do. But that would be a cultural factor that will need to change with time. We cannot remain loyal to inaccurate models simply because we like the stories they tell.
The point I’d like to make is that economics has influenced all of the other social sciences, so a revolution in economics will also trigger a revolution in the other social sciences. Ideas about rational actors, utility, game theory, competition, incentives and money flows grew out of economics and influenced how historians think about history and influenced how sociologists think about society. And this will change if economics changes. There might come a moment when we re-interpret large aspects of history, based on new models from economics, in particular around the issues of information and ignorance. Ignorance, in particular, is something that every historian has to consider and explain, but without really having formal models for thinking about how ignorance, at all levels, has shaped history. I believe there is some exciting research that is waiting to happen.
There is also the research that can be done with agent based simulations (basically video games built for research purposes). These offer a different way of exploring something like permanent disequilibrium. Again, these are only possible in the era of computers. In the 1600s, 1700s, 1800s, you could not trace a simulation in which a million agents each pursue their own goals, but now I can easily write the code for such a simulation and I can run it on my laptop. So it is a whole new avenue of research waiting to be explored (in this case the word “new” refers to the last 50 years).
It's important to realize that there are areas where an agent based simulation actually allows an easier approach to old problems. For instance, imperfect information, or asymmetric information. Early equilibrium models assumed everyone knew everything all the time (everyone had perfect information in simple supply and demand models). Imperfect information was actually something that had to be bolted on later on, and it made the math more complicated. But with an agent based simulation, the fact that each agent has limited information is something that arises organically from the nature of a simulation. We all know this from playing video games -- we don't know everything that is happening everywhere in the world we are exploring. Ignorance is naturally and organically built into the system.
These models will change the way we understand economic history. In particular, with an agent based simulation, it becomes easier to model how merchants and military leaders lacked information, and how much the lack of information shaped human history. You could say “All historians do this already” and of course I mean in the sense of testing some formal model. You can definitely ask whether formal models will offer us deeper insights than the artful insights that we already get from a great historian. This is still unknown, but here, as with all other aspects of human knowledge, the possible advantage of developing a formal model is that it makes it easier to transfer knowledge from one situation to another — a possible transformation of an art to a science.
When it comes to agent based simulations, the resistance to these models, I think, is that such simulations don't necessarily tell a story. I don't have a rigorous proof, but presumably any simulation also faces the consequences of the Entscheidungsproblem (the halting problem). In 1937, Alan Turing was able to show there is no rigorous way to prove any arbitrary piece of software will end at some point. Of course, there are many short programs where a computer programmer can look at the code and say with certainty, "Yes, this software terminates". But given all the possible software that might exist, there is no foolproof way of knowing if the software will stop at some point.
Also, with agent based simulations, many of them do fall into an equilibrium, which seems to allow a simple story to be told. The great biologist, Richard Dawkins, worked with simple computer simulations in the 1970s. In his 1976 book, The Selfish Gene, he mentions a simulation he did to discover the ideal ratio of males to females, in a species with 2 sexes. He ran the simulation many times, and over and over again it fell into an equilibrium of 50% males, 50% females. So, sometimes, an agent based simulation appears to recreate the kinds of simple stories that older equilibrium models tell. But agent based simulations, in the abstract, face the Entscheidungsproblem, that is, you can never know, for certainty, if the equilibrium you think you see is a permanent one. You might run a simulation for a month, or a year, or a million years, and it seems to have fallen into a steady equilibrium, and yet if you run it for one more minute the whole situation might change, and therefore the conclusion that you might draw from the agent based simulation might also change. Anyone who has played long running, immersive, first person video games probably has some intuitions about the overall uncertainty of some new situation -- how many millions of times would we need to confront a situation to have scientific levels of certainty about the meaning of that situation?
Does that mean that agent based simulations are useless? I think they are useful, but in a new way. We should think about these the same way that we ask pilots to use flight simulators — playing a simulation over and over again might allow regulators to develop both ingrained habits of safety as well as intuitions about the kinds of ambushes and reversals that they might face. That's a very different kind of teaching and learning than what people get from the study of mathematical equilibrium models, but it is a type of teaching and learning that we should want to see in those who will have to make important decisions, sometimes under crisis conditions. Airplane pilots often supplment actual flying time with time in a simulator -- why shouldn't we also want Fed regulators to spend time in an economic simulator that models how people act during an economic crisis?
To be clear, I'm not saying that we abandon all equilibrium models. Given how much of the Western intellectual tradition is wrapped up in such models, that would not be possible, nor is it even desirable. But I do think we've spent so much of the last 400 years pursuing equilibrium models that there is something of a law of diminishing returns affecting how much more progress we can make with that kind of research. By contrast, we can generate new insights by pursuing those models that can only really be investigated thanks to computers. This is a wide open field, so this is where we should focus our efforts over the next decade or two.
I am not a professional economist nor a professional historian, I’m only an amateur with an interest in these topics. Having followed both subjects with some real interest for 25 years, I am both excited by these areas, but also frustrated with some of the blind spots that exist in these fields. For instance, there remains a wide gap between those historians who focus on the irrationality and foolishness of leaders (think Barbara W. Tuchman) using purely descriptive and anecdotal methods, versus those who have tried to use insights from economic models to explain aspects of history, especially economic history. I’d like to see the gap between these two groups close, but I think that can only happen when economists do a better job of explaining the messier aspects of real life. For historians, static equilibrium models will forever be too limiting, since history does not offer us a single moment when any aspect of human life was at equilibrium. Since historians are looking at a world in permanent disequilibrium, having better models for permanent disequilibrium would help them.
For my part, I hope to make a small contribution. As this Substack is focused on democracy, and especially on voting, I am working on an agent based simulation to model voting, which I hope will eventually make it easier to model all of the many different voting systems that theorists have suggested (rank, approval, MMP, STV, etc
), and how agents in the systems (both voters and politicians) will react to incentives of each system of voting.