Harnessing Innovation for Climate Change Solutions

Farmer is the Baillie Gifford Professor of Complex Systems at Oxford's Institute for New Economic Thinking. Before joining Oxford in 2012, he worked at Los Alamos National Laboratory and the Santa Fe Institute, where he studied complex systems and economic dynamics. During the 1990s, he took a break from academia to run a successful quantitative trading firm using statistical arbitrage strategies.

Farmer has been a pioneer in chaos theory and complexity economics, including the development of agent-based models to understand economic phenomena. His work spans from housing markets to climate change, and he recently authored Making Sense of Chaos exploring complexity science and economic modeling.

In This Episode

* What is complexity economics? (1:23)

* Compliment or replacement for traditional economics (6:55)

* Modeling Covid-19 (11:12)

* The state of the science (15:06)

* How to approach economic growth (20:44)

Below is a lightly edited transcript of our conversation.

What is complexity economics? (1:23)

We really can model the economy as something dynamic that can have its own business cycles that come from within the economy, rather than having the economy just settle down to doing something static unless it's hit by shocks all the time, as is the case in mainstream models.

Pethokoukis: What does the sort of economics that people would learn, let's say, in the first year of college, they might learn about labor and capital, supply-demand equilibrium, rational expectations, maybe the importance of ideas. How does that differ from the kind of economics you are talking about? Are you looking at different factors?

Farmer: We’re really looking at a completely different way of doing economics. Rather than maximizing utility, which is really the central conceptual piece of any standard economic model, and writing down equations, and deducing the decision that does that, we simulate the economy.

We assume that we identify who the agents in the and economy are, who's making the decisions, what information do they have available, we give them methods of making the decisions — decision-making rules or learning algorithms — and then they make decisions, those decisions have economic impact, that generates new information, other information may enter from the outside, they make decisions, and we just go around and around that loop in a computer simulation that tries to simulate what the economy does and how it works.

You've been writing about this for some time. I would guess — perhaps I'm wrong — that just having more data and more computer power has been super helpful over the past 10 years, 20 years.

It's been super helpful for us. We take much more advantage of that than the mainstream does. But yes, computers are a billion times more powerful now than they were when Herb Simon first suggested this way of doing things, and that means the time is ripe now because that's not a limiting factor anymore, as it was in the past.

So if you're not looking at capital and labor per se, then what are the factors you're looking at?

Well, we do look at capital and labor, we just look at them in a different way. Our models are concerned about how much capital is there to invest, what labor is available. We do have to assign firms production functions that tells, given an amount of capital and labor and all their other inputs, how much can the firms produce? That part of the idea is similar. It's a question of the way the decision about how much to produce is made, or the way consumers decide how much to consume, or laborers decide at what price to provide their labor. All those parts are different.

Another difference — if I'm understanding it correctly — is, rather than thinking about economies that tend toward equilibrium and focusing how outside shocks may put an economy in disequilibrium, you're looking a lot more at what happens internally. Am I correct?

We don't assume equilibrium. Equilibrium, it has two senses in economics: One is supply equals demand. We might or might not run a model where we assume that. In many models we don't, and if that happens, that's great, but it's an outcome of the model rather than an assumption we put in at the beginning.

There's another sense of equilibrium, which is that everybody's strategy is lined up. You've had time to think about what you're doing, I've had time to think about what I'm doing, we’ve both come to the optimal decision for each of us to make, taking the other one into account. We don't assume that, as standard models typically do. We really can model the economy as something dynamic that can have its own business cycles that come from within the economy, rather than having the economy just settle down to doing something static unless it's hit by shocks all the time, as is the case in mainstream models. We still allow shocks to hit our models, but the economy can generate dynamics even without those shocks.

This just popped in my head: To whom would this model make more intuitive sense, Karl Marx or Adam Smith?

Adam Smith would like these models because they really allow for emergent behavior. That is, Smith's whole point was that the economy is more than the sum of its parts, that we get far more out of specializing than we do out of each acting like Robinson Crusoes. Our way of thinking about this gets at that very directly.

Marx might actually like it too, perhaps for a different reason. Marx was insightful in understanding the economy as being like, what I call in the book, the “metabolism of civilization.” That is, he really did recognize the analogy between the economy and the metabolism, and viewed labor as what we put together with natural resources to make goods and services. So those aspects of the economy are also embodied in the kind of models we're making.

I think they both like it, but for different reasons.

Compliment or replacement for traditional economics (6:55)

There are many problems where we can answer questions traditional methods can't even really ask.

The way I may have framed my questions so far is that you are suggesting a replacement or alternative. Is what you're suggesting, is it one of those things, or is it a compliment, or is it just a way of looking at the world that's better at answering certain kinds of questions?

I think the jury is out to find the answer to that. I think it is certainly a compliment, and that we're doing things very differently, and there are some problems where this method is particularly well-suited. There are many problems where we can answer questions traditional methods can't even really ask.

That said, I think time will tell to what extent this replaces the traditional way of doing economics. I don't think it's going to replace everything that's done in traditional economics. I think it could replace 75 percent of it — but let me put an asterisk by that and say 75 percent of theory. Economists do many different things. One thing economists do is called econometrics, where they take data and they build models just based on the data to infer things that the data is telling them. We're not talking about that here. We're talking about theories where economists attempt to derive the decisions and economic outcomes from first principles based on utility maximization. That's what we're talking about providing an alternative to. The extent to which it replaces that will be seen as time will tell.

When a big Wall Street bank wants to make a forecast, they're constantly incorporating the latest jobless claims numbers, industrial production numbers, and as those numbers get updated, they change their forecasts. You're not using any of that stuff?

Well, no. We can potentially could ingest any kind of data about what's going on.

But they're looking at big, top-down data while you're bottom-up, you're sort of trying to duplicate the actual actors in the economy.

That is true, but we can adjust what's at the bottom to make sure we're matching initial conditions. So if somebody tells us, “This is the current value of unemployment,” we want to make sure that we're starting our model out, as we go forward, with the right level of unemployment. So we will unemploy some of the households in our model in order to make sure we're matching the state of unemployment right now and then we start our simulation running forward to see where the economy goes from here.

I would think that the advent of these large language models would really take this kind of modeling to another level, because already I'm seeing lots of papers on their ability to . . . where people are trying to run experiments and, rather than using real people, they're just trying to use AI people, and the ability to create AI consumers, and AI in businesses — it would have to be a huge advance.

Yes. This is starting to be experimented with for what we do. People are trying to use large language models to model how people actually make decisions, or let's say, to simulate the way people make decisions, as opposed to an idealized person that makes perfect decisions. That's a very promising line of attack to doing this kind of modeling.

Large language models also can tell us about other things that allow us to match data. For example, if we want to use patents as an input in our modeling — not something we're doing yet, but we've done a lot of studies with patents — one can use large language models to match patents to firms to understand which firms will benefit from the patents and which firms won't. So there are many different ways that large language models are likely to enter going forward, and we're quite keen to take advantage of those.

Modeling Covid-19 (11:12)

We predicted a 21.5 percent hit to UK GDP in the second quarter of 2020. When the dust settled a year later, the right answer was 22.1. So we got very close.

Tell me, briefly, about your work with the Covid outbreak back in 2020 and what your modeling said back then and how well it worked.

When the pandemic broke out, we realized right away that this was a great opportunity to show the power of the kind of economic modeling that we do, because Covid was a very strong and very sudden shock. So it drove the economy far out of equilibrium. We were able to predict what Covid would do to the UK economy using two basic ideas: One is, we predicted the shock. We did that based on things like understanding a lot about occupational labor. The Bureau of Labor Statistics compiles tables about things like, in a given occupation, how close together do people typically work? And so we assumed if they worked closer together than two meters, they weren't going to be able to go to their job. That combined with several other things allowed us to predict how big the shock would be.

Our model predicted how that shock would be amplified through time by the action of the economy. So in the model we built, we put a representative firm in every sector of the economy and we assumed that if that firm didn't have the labor it needed, or if it didn't have the demand for its product, or if it didn't have the inputs it needed, it wouldn't be able to produce its product and the output would be reduced proportional to any of those three limiting factors.

And so we started the model off on Day One with an inventory of inputs that we read out of a table that government statistical agencies had prepared for each sector of the economy. And we then just looked, “Well, does it have the labor? Does it have demand? Does it have the goods?” If yes, it can produce at its normal level. If it's lacking any of those, it's going to produce at a lower level. And our model knew the map of the economy, so it knew which industries are inputs to which other industries. So as the pandemic evolved day by day, we saw that some industries started to run out of inputs and that would reduce their output, which, in turn, could cause other industries to run out of their inputs, and so on.

That produced quite a good prediction. We predicted a 21.5 percent hit to UK GDP in the second quarter of 2020. When the dust settled a year later, the right answer was 22.1. So we got very close. We predicted things pretty well, industry by industry. We didn't get them all exactly right, but the mistakes we made averaged out so that we got the overall output right, and we got it right through time.

We ran the model on several different scenarios. At the time, this was in April of 2020, the United Kingdom was in a lockdown and they were trying to decide what to do next, and we tested several different scenarios for what they might do when they emerged from the full lockdown. The one that we thought was the least bad was keeping all the upstream industries like mining, and forestry, and so on open, but closing the downstream, customer-facing industries like retail businesses that have customers coming into their shop, or making them operate remotely. That was the one they picked. Already when they picked it, we predicted what would happen, and things unfolded roughly as we suggested they would.

The state of the science (15:06)

Mainstream models can only model shocks that come from outside the economy and how the economy responds to those shocks. But if you just let the model sit there and nothing changes, it will just settle down and the economy will never change.

I'm old enough to remember the 1990s and remember a lot of talk about chaos and complexity, some of which even made it into the mainstream, and Jurassic Park, which may be the way most people heard a little bit about it. It's been 30 years. To what extent has it made inroads into economic modeling at central banks or Wall Street banks? Where's the state of the science? Though it sounds like you're really taking another step forward here with the book and some of your latest research.

Maybe I could first begin just by saying that before Jurassic Park was made, I got a phone call and picked up the phone, and the other end of the line said, “Hi, this is Jeff Goldblum, have you ever heard of me?” I said, “Yeah.” And he said, “Well, we're making this movie about dinosaurs and stuff, and I'm going to play a chaos scientist, and I'm calling up some chaos scientists to see how they talk.” And so I talked to Jeff Goldblum for about a half an hour. A few of my other friends did too. So anyway, I like to think I had a tiny little bit of impact on the way he behaved in the movie. There were some parallels that it seemed like he had lifted.

Chaos, it's an important underlying concept in explaining why the weather is hard to predict, it can explain some forms of heart arrhythmias, we use it to explain some of the irregular behavior of ice ages. In economics, it was tossed around in the ’90s as something that might be important and rejected. As I described in the book, I think it was rejected for the wrong reasons.

I'm proposing chaos, the role it plays in here is that, there's a debate about business cycles. Do they come from outside? The Covid pandemic was clearly a business cycle that came from outside. Or do they come from inside the economy? The 2008 financial crisis, I would say, is clearly one that came from inside the economy. Mainstream models can only model shocks that come from outside the economy and how the economy responds to those shocks. But if you just let the model sit there and nothing changes, it will just settle down and the economy will never change.

In contrast, the kinds of models we build often show what we call endogenous business cycles, meaning business cycles that the model generates all on its own. Now then, you can ask, “Well, how could it do that?” Well, basically the only plausible way it can do that is through chaos. Because chaos has two properties: One is called sensitive dependence on initial conditions, meaning tiny changes in the present can cause large changes in the future; but the other is endogenous motion, meaning motion that comes from within the system itself, that happens spontaneously, even in very simple systems of equations.

Would something like consumer pessimism, would that be an external shock or would something more internal where everybody, they're worried about the futures, then they stop spending as much money? How would that fit in?

If the consumer pessimism is due to the fear of a nuclear war, I would say it's outside the economy, and so that's an external shock. But if it's caused by the fact that the economy just took a big nose dive for an internal reason, then it's part of the endogenous dynamics

I spent many years as a journalist writing about why the market's going up, the market's going down, and by the end of the day, I had to come up with a reason why the market moved, and I could — I wasn't always quite confident, because sometimes it wasn't because of a new piece of data, or an earnings report, they just kind of moved, and I had no real reason why, even though I had to come up . . . and of course it was when I was doing that was when people started talking about chaos, and it made a lot of intuitive sense to me that things seem to happen internally in ways that, at least at the time, were utterly unpredictable.

Yeah, and in fact, one of the studies I discuss in the book is by Cutler, Poterba, and Summers — the Summers would be Larry Summers — where they did something very simple, they just got the 100 largest moves of the S&P index, they looked up what the news was the next day about why they occurred in the New York Times, and they subjectively marked the ones that they thought were internally driven, versus the ones that were real news, and they concluded they could only find news causes for about a third of them.

There is always an explanation in the paper; actually, there is one day on the top 12 list where the New York Times simply said, “There appears to be no cause.” That was back in the ’40s, I don't think journalists ever say that anymore. I don't think their paper allows them to do it, but that's probably the right answer about two-thirds of the time, unless you count things like “investors are worried,” and, as I point out in the book, if the person who invests your money isn't worried all the time, then you should fire them because investors should worry.

There are internal dynamics to markets, I actually show some examples in the book of simple models that generate that kind of internal dynamics so that things change spontaneously.

How to approach economic growth (20:44)

I'm not saying something controversial when I say that technological change is the dominant driver of economic growth, at least for the economy as a whole.

You recently founded a company, Macrocosm, trying to put some of these ideas to work to address climate change, which would seem to be a very natural use for this kind of thinking. What do you hope to achieve there?

We hope to provide better guidance through the transition. We're trying to take the kind of things we've been doing as academics, but scale them up and reduce them to practice so they can be used day-in and day-out to make the decisions that policymakers and businesspeople need to make as the transition is unfolding. We hope to be able to guide policymakers about how effective their policies will be in reducing emissions, but also in keeping the economy going and in good shape. We hope to be able to advise businesses and investors about what investments to make to make a profit while we reduce emissions. And we think that things have changed so that climate change has really become an opportunity rather than a liability.

I write a lot about economic growth and try to figure out how it works, what are the key factors. . . What insights can you give me, either on how you think about growth and, since I work at a think tank, the kind of policies you think policy makers should be thinking about, or how should they think about economic growth, since that seems to be on top-of-mind in every rich country in the world right now?

I'm not saying something controversial when I say that technological change is the dominant driver of economic growth, at least for the economy as a whole. And we've spent a lot of time studying technological change by just collecting data and looking for the patterns in that data: What does the technology cost through time and how rapidly is it deployed? We've done this for 50 or 60 technologies where we look at past technological transitions, because typically, as a technology is coming in, it's replacing something else that's going out, and what we've seen are a couple of striking things:

One is, many technologies don't really improve very much over time, at least in terms of cost. Fossil fuels cost about the same as they did 140 years ago once you adjust for inflation. In fact, anything we mine out of the ground costs about the same as it did a hundred years ago.

In contrast, solar energy from solar photovoltaic panels costs 1/10,000th what it did when it was introduced in the Vanguard satellite in 1958. Transistors have been going down at 40 percent per year, so they cost about a billionth of what they did back in 1960. So some technologies really make rapid progress, and the economy evolves by reorganizing itself around the technologies that are making progress. So for example, photography used to be about chemistry and film. Photography now is about solid-state physics because it just unhitched from one wagon and hitched itself to another wagon, and that's what's happening through the energy transition. We're in the process of hitching our wagon to the technologies that have been making rapid progress, like solar energy, and wind energy, and lithium ion batteries, and hydrogen catalyzers based on green energy.

I think we can learn a lot about the past, and I think that when we look at what the ride should be like, based on what we understand, we think the transition is going to happen faster than most people think, and we think it will be a net saving of money

So then how do you deal with a wild card, which I think if you look at the past, nuclear power seems like it’s super expensive, no progress being made, but, theoretically, there could be — at least in the United States — there could be lots of regulatory changes that make it easier to build. You have all these venture capital firms pouring money into these nuclear startups with small reactors, or even nuclear fusion. So a technology that seems like it's a mature technology, it might be easy to chart its future, all of a sudden maybe it's very different.

I'm not arguing we should get rid of nuclear reactors until they run their normal lifetime and need to be gotten rid of, but I think we will see that that is not going to be the winning technology in the long run, just because it's going to remain expensive while solar energy is going to become dirt cheap.

In the early days, nuclear power had faced a very favorable regulatory environment. The first nuclear reactors were built in the ’50s. Until Three Mile Island and Chernobyl happened, it was a very regulatorily friendly environment and they didn't come down in cost. Other countries like France have been very pro-nuclear. They have very expensive electricity and will continue to do so.

I think the key thing we need to do is focus on storage technologies like green hydrogen. Long-term storage batteries have already come down to a point where they're beginning to be competitive; they will continue to do so. And in the future, I think we'll get solid-state storage that will make things quite cheap and efficient, but I don't think small modular reactors are going to ever be able to catch up with solar and wind at this point.

On sale everywhere The Conservative Futurist: How To Create the Sci-Fi World We Were Promised

Micro Reads

▶ Economics

* United States Economic Forecast - Deloitte

* The Hidden Threat to National Security Is Not Enough Workers - WSJ

▶ Business

* DOGE Can’t Do It All. Here’s What It Can Do. - Politico

* AI Startup Perplexity Closes Funding Round at $9 Billion Value - Bberg

▶ Policy/Politics

* US Homeland Security chief attacks EU effort to police AI - FT

* The Trump Bump: The Republican Fertility Advantage in 2024 - IFS

* House unveils AI ‘road map’ but punts on setting priorities - Wapo

* Did Tariffs Make American Manufacturing Great? - Cato

▶ AI/Digital

* Call ChatGPT from any phone with OpenAI’s new 1-800 voice service - Ars

* Homo-Silicus: Not (Yet) a Good Imitator of Homo Sapiens or Homo Economicus - SSRN

* Is AI finally ready to replace your doctor? - NS

* The Age of Quantum Software Has Already Started - WSJ

* This is where the data to build AI comes from - MIT

* The New AI Stock Pickers Are Destined to Disappoint - Bberg Opinion

▶ Clean Energy/Climate

* Fusion Start-Up Plans to Build Its First Power Plant in Virginia - NYT

* Will the World's First Nuclear Fusion Power Plant Be Built in Virginia? Here's Why We're Skeptical - SciAm

* The deepest hole on Earth: Inside the race to harness unlimited power from our planet's core - SF

* Dubai transforms into walkable city with air-conditioned paths - New Atlas

* Oklo inks record deal for using nuclear to power data centers - E&E

▶ Robotics/AVs

* AI Robots Are Coming, and They’ll Be Made in Asia - Bberg Opinion

▶ Space/Transportation

* Boeing Starliner crew’s long awaited return delayed to March - Wapo

▶ Up Wing/Down Wing

* What Could Go Right? The Best News of 2024 - The Progress Network

▶ Substacks/Newsletters

* Why Don’t EU Firms Innovate? The Hidden Costs of Failure - Conversable Economist

* Why Did the Industrial Revolution Happen? - Oliver Kim

* One Down, Many To Go - Hyperdimensional

* The Experience Curve - Risk & Progress

* The case for clinical trial abundance - Slow Borin

* Nuclear Waste: Yes, In (or Under) My Backyard - Breakthrough Journal

* Answer Time: Can We Imagine Pluralistic Futures? - Virginia’s Newsletter

* What just happened - One Useful Thing

Faster, Please! is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit fasterplease.substack.com/subscribe