Can We Predict The Unpredictable? with J. Doyne Farmer
33:14 Share

Have you ever seen the film Minority Report? I'm placing you under arrest for the future murder of Sarah Marks. Give the man his hat. The future can be seen. It imagines a world in which we can collect enough data about the present to predict the future. Even predict something as chaotic as human actions. There hasn't been a murder in six years. There's nothing wrong with the system. It is perfect. I agree. Murder.

Using data to prevent crime is one thing, but what if we applied that same idea to economics?

What if we could accurately forecast the twists and turns of something as chaotic as the stock market? It was one of the most profound events in generations, a crash that most experts did not foresee. Its effects and of the recession that followed on income, wealth, inequality, and our politics are still with us. Or predict how a tax hike would widen or shrink the gap between

between the wealthy and the poor, or exactly what policy could reverse inflation. The simple question, which is unanswerable, is what does the Fed do in reaction to these sorts of numbers? Do they find this as good news or bad news as they try to fight inflation? Of course, this is all science fiction, right? Well, maybe not. That's right. In a sense, it's what science is about.

We look at things that look like they're random or we assign superstitious causes to what makes them happen. And then as we begin to understand them, we realize that actually there are things there we can predict. That's Doan Farmer, one of the world's leading scholars in the science of complex systems and a professor at Oxford University. Complex systems is the study of emergent phenomena.

Anything where the units that make up a system behave qualitatively differently than the behavior when you put them all together. Take a neuron. A neuron's a cell. You put a signal in it. What happens when I hook them up? Not at all obvious, but we know that if you take 80 billion of them and you hook them up in just the right way, we have the human brain. So we have an emergent phenomena that's really qualitatively different than the building blocks that compose it.

Similarly, the economy is another good example. The economy can seem inscrutable, with so many actors making so many choices that it can appear to be just chaos. But Farmer doesn't see randomness. He sees patterns and hidden order. In his latest book, Making Sense of Chaos, Farmer reveals how the most cutting-edge computers combined with oceans of data are making it possible to predict the unpredictable.

from the smallest individual decisions to the largest market shifts. Complex science is giving us new ways to forecast and potentially change what's coming next. We can make predictions about things that traditional economic models can't even ask. We think as our models get better,

Central banks will start using them. Companies, commercial enterprises will start using them. And eventually that'll put pressure on academic economic departments to start developing them. Welcome to Big Brains, where we translate the biggest ideas and complex discoveries into digestible brain food. Big Brains, Little Bites from the University of Chicago Podcast Network. I'm your host, Paul Rand. On today's episode, predicting the economy for a better future.

The University of Chicago Leadership and Society Initiative guides accomplished executive leaders in transitioning from their long-standing careers into purposeful, on-court chapters of leadership for society.

The initiative is currently accepting candidates for its second cohort of fellows. Your next chapter matters for you and for society. Learn more about this unique fellowship experience at leadforsociety.uchicago.edu.

Farmer is one of the leading experts on complex systems and chaos in the world. Now, chaos is an example of an emergent phenomenon. And the surprising thing about chaos, there's two surprising things. One is sensitive dependence on initial conditions.

meaning you can take two states of the world that are essentially the same, and as you follow them forward in time, they exponentially move apart. You start out with what looks like something you should be able to predict, but it becomes unpredictable through time. The weather is a great example. You can predict the weather pretty well a minute ahead, but you can't predict it at all really two or three weeks ahead. There's one other property of chaos, what we technically call endogenous motion, meaning motion from within,

So, for example, if you sit next to a mountain stream,

Even though the rocks in the stream are all fixed, the way the water is coming in is the same through time, the stream will turn around, splash, have turbulence, even though there's nothing, no obvious source of that turbulence. It's really inherent to the motion of the water. And it requires ongoing computation to track what that chaos is doing. In contrast, say, you know, if I want to predict chaos,

the motion of a planet, I can just write down a formula, well, to some approximation, write down a formula, plug in my measurements, and the prediction's good for a very long time. Farmer started his career as a physicist, trying to measure, understand, and ultimately predict chaos. His first project revolved around what may seem like the most unpredictable system imaginable, roulette.

When you see a roulette wheel, you see total randomness. But if you look at it as a physicist, you go, wait a minute, this is just a rolling ball on a circular track with a rotor in the middle that's rotating the other way. These are simple physical systems. So as physicists, we should be able to predict what they're going to do. And in the 1970s, that's exactly what they tried to do. While most people placed their bets on luck, Farmer and his friends turned to science to try to beat the house. ♪

And here we were aided by the fact that there's typically 10 to 15 seconds that elapse between the point where the croupier spins the ball and where the croupier closes the bats

So you have that 15 seconds to gather information about how fast the ball is going and where it is. According to Newton's laws, if you know the forces, you know the position and velocity, you can predict the future. So we spent a lot of time understanding the forces. Turns out it's wind resistance is the dominant thing that slows the ball down.

We solved the equations of motion for a roulette ball. We then built actually the first wearable digital computer. It was even a concealable digital computer that we tucked under our armpit in the early version. The later version was in a shoe. You sound like Ocean Zate back in the 1970s here. It was. It was. And we had switches in our shoes that we operated with our big toes. And when the ball rolled by a reference point on the wheel, we would make a click.

When it would roll past it again, we'd make another click. We knew how long that took because the computers are good at counting. So we knew the velocity. And then we plugged that into our program, made a prediction, sent it to a second person who would then make bets in roughly the area where we thought the ball was likely to come to rest. And did you get wildly rich? We beat the house by a good margin, by about a 20% margin. We did not get wildly rich. Why? Well, we were concerned about our kneecaps.

and we'd read reputable reports that the casinos might take us in the back room and beat us up. So every time we started to make a pile of money, we would really feel the casino heat.

And we left. We did well, but I can't say we got wildly rich. But it planted a seed for Farmer that would later become one of his most prominent projects. If they could use complex system science to predict the chaos of a roulette wheel, could they do the same thing with the stock market? Well, Prediction Company, we reasoned that, unlike in the casino,

They don't kick you out for winning. Or take your knees out. They don't break your kneecaps if you win. So we like that part. Okay. So with Prediction Company, we got a business partner. We got funding. We partnered with what eventually became UBS. We acted like a proprietary trading group within UBS. We traded UBS's money. So were they able to use complex systems models to beat the market? Our return to risk ratio was much better.

better than a standard bet in the stock market, like six times as good. Wow. We did well, made quite a lot of money for UBS, made some money for ourselves. I left after eight years. The company lasted for 28 years, and they made profits 27 out of the 28 years. But Farmer still wasn't satisfied. He wanted to see if the new models he was developing with complexity science could predict something even larger than the stock market.

What if they could predict entire economies? In some of our models, for example, we can have millions of agents scattered around the world, each making their decisions within the constraints of the country they live in, where the rules may be different.

looking at different demographic groups, different ages, different income brackets. And we can capture the heterogeneity and interactivity of real world agents much better than standard or mainstream methods.

This was the beginning of a whole new branch of economics. Spearheaded by Farmer and others, it would come to be known as Complexity Economics. You've got a new book out. It's called Making Sense of Chaos, A Better Economics for a Better World.

And the thrust of this is this idea of complexity economics. And I wonder if you can, again, give us a little more context of what that is and compare it against what may be thought of as traditional economics. So my book is presenting an alternative to economic theory. And how does economic theory work? Well, you begin by writing down what's called a utility function for everybody. It's just a scorecard.

that says what they prefer. I'd rather have this than that. I'd rather have money than not have money. I'd rather consume than not consume. You assign a model of the world to everybody.

The classic model of the world is called rational expectations. So they have their utility function. They're going to take all the information they have available, figure out exactly what it means, calculate the decision that will give them the most utility. And then finally, you assume equilibrium. That is, you assume supply equals demand. Or in some cases, it means that everybody's strategy is the best strategy they can have without everybody else changing their strategies.

And you do this, you assume everybody takes everybody else into account. You write all this down in equations. You solve the equations where everybody's making the best possible decision. You assume everybody makes that decision and you look at its implications for the economy. Economic modeling makes aggregate assumptions and so models things. It would be as if we tried to model the weather based on the global average temperature and pressure and wind velocity.

So part of what I argue in the book is we really need to model things in a fine-grained way, and agent-based modeling gives us a way to do that. Agent-based modeling involves using advanced computers to create simulations of the real world.

You see, while traditional economics assumes everyone acts like a robot, making the best decisions based on perfect information, agent-based modeling actually assumes people aren't making perfect decisions. It bakes in the flaws and messiness of the real world. It's an idea that actually goes all the way back to the 60s.

Herb Simon, famous guy, even won a Nobel Prize in economics. One of the key ideas Herb Simon had back in the 60s was bounded rationality. He said, look, there are problems. We don't have the computing power in our brains to solve them. And, you know, we're familiar with some of those chess, for example.

The greatest chess players in the world aren't perfect chess players. They can't calculate the optimum strategy. They make use of heuristics, you know, capture the center of the board. It's better to swap a rook for a queen than vice versa. So they make use of these heuristics so that over the long term, they do a decent job. And that's what we do as human beings. We use heuristics all the time without even realizing it. For instance, when trying to choose a restaurant, you may go for the one that looks busier.

assuming that must mean it's good, even though that doesn't necessarily prove anything. And that's the approach that we use in our agent-based models. We try and find the strategies real people use, the good enough strategies real people use in complicated situations. We put them in our models and we let our computer show what happens when everybody makes those kind of decisions.

Farmer argues that while agents in these models may be simpler or less intelligent, you may say, than the rational agents in traditional economics, they're actually better at predicting reality in aggregate because that's closer to how we really move through the world. We look at the information that's flowing to the agents. We write down in computer code the decision rules that the agents will use.

Those decisions could be, I mean, in an extreme case, random. They just, I'll just choose something at random. Or it could be a heuristic. If I'm writing a model about value investors, I'll say, well, value investors buy undervalued assets.

Or it could be imitate your neighbors, pick your neighbor who's doing the best, do what that neighbor does. Or it could be trial and error. Try something, doesn't work, try something else. So you write those things down. So you have a decision rule for every agent. Then the agents just make their decisions.

and those decisions have impact on the economy, that generates new information. In addition, you may have other information flowing in from outside and you just repeat the loop. But it's a totally different way of doing things than the way the economists construct theories. And it's an alternative that we think has advantages because for one thing, since we're not restricted to writing down equations, we can let things be pretty complicated.

and in particular make models even with millions of agents. Traditional economic models, because you're trying to calculate the optimum, as soon as things start to get at all complicated, you can't have very many agents or put very many effects in before the equations can't be solved anymore. And so you can't even compute what the optimum would be, which I think is a symptom that maybe you're on the wrong track because after all, we make decisions all the time.

If we can't compute the optimum, what we do the best we can. These theories were groundbreaking when Herbert Simon first proposed them in the 60s. But the technology to test them wasn't there. Now we have the computing power and we have the data. Yeah, and I argue that the time is ripe because these ideas that I'm talking about have been around since the 60s, but it was hard to do it back in his day. So economics took a different path.

I'm arguing it's now time to revisit that choice because computers are a billion times more powerful. We have vastly more data. We have a much better understanding of human psychology. And we understand things about how to build agent-based models now. So it's time to really push, realize its potential. For example, we built an agent-based model of housing markets.

where the individual agents are households who might decide to rent or buy or sell. We just simulated housing markets the way things really work. A household decides it wants to buy a house, it goes to a real estate agent, get a loan from a bank. So we simulated everything verbatim. Traditional models can't do that. Traditional models can't deal with the fact that housing markets don't clear, meaning supply is generally not equal to demand.

When you're in a housing boom, there are many more buyers than sellers. When you're in a housing bust, there are many more sellers than buyers. And things aren't getting clear. So we can do that. We made a model for the UK government of the economic impact of the COVID pandemic.

And our model successfully predicted before things happened, what the economic hit would be to the GDP economy, how it would hit the different sectors of the economy. I mean, second quarter of 2021 or 2020, we predicted a 21.5% hit.

to GDP, when the dust settled, it was 22.1%. My goodness. And so that's the key selling factor is people want better predictions. Predictions aren't any good unless they're good. How do traditional economists at a place like UChicago, among others, think about this idea of complexity economics? Well, they're not very keen on it. Okay. You know, that's just not the way they do things.

And, you know, graduate students who want to do things the way I described will be advised that's dangerous. Don't don't go there. Now, is that it sounds like what I used to hear and maybe in some cases still do about behavioral economics. Behavioral economics is not well integrated into economic theory because economic theory till still typically requires on rational expectations. This homo economicus, Mr. Spock idea about the agents. Now,

Many economists are trying to bring behavioral economics into economic theory. There's nothing that's well accepted. And the workhorse models that the Federal Reserve or the U.S. Treasury or the

economic forecasters use still has not really incorporated behavioral economics. So we argue that the way we're doing things is a better way to incorporate behavioral economics because it's very easy to put it in there. In a computer simulation, you give me a model for how people behave, I program it up on the computer, and I use that model to make decisions. No problem. And this isn't just the technology of the future.

Farmer is using complexity economics today, even working with some national banks, to make prediction-informed economic decisions to tackle everything from inequality to climate change. That is after the break.

The University of Chicago Leadership and Society Initiative guides accomplished executive leaders in transitioning from their longstanding careers into purposeful on-court chapters of leadership for society. The initiative is currently accepting candidates for its second cohort of fellows. Your next chapter matters for you and for society.

Learn more about this unique fellowship experience at leadforsociety.uchicago.edu.

If you're getting a lot out of the important research shared on Big Brains, there's another University of Chicago podcast network show you should check out. It's called Not Another Politics Podcast. Not Another Politics Podcast provides a fresh perspective on the biggest political stories, not through opinions and anecdotes, but through rigorous scholarship, massive data sets, and a deep knowledge of theory. If you're getting a lot out of the important research shared on Big Brains, there's another University of Chicago podcast network show you should check out.

If you want to understand the political science behind the political headlines, then listen to Not Another Politics Podcast, part of the University of Chicago Podcast Network. Here we are even further along with more powerful computers, additional data points, and

What are we trying to do next? What's the next iteration of this and what are we looking for as an outcome? Well, I always try to learn by my mistakes. And the thing I didn't like about Prediction Company was, first of all, we had to keep everything secret. So we weren't doing any good for the world. I mean, we made some Swiss bankers richer and, you know, I have a sailboat as a result. But there was no...

public welfare. The other thing is we didn't really understand what made it all tick. To us, the stock market was a stream of numbers. We fit, we use things like neural nets to fit, find patterns in that stream of numbers and use that to make bets. But I really wanted to understand why does this work? So with my new company, Macrocosm, we are modeling the economy from the ground up

in a causal way. So again, we have agents who make decisions. Those decisions have economic impact. The economic impact modifies the decisions they're making. So it's a setup where we can understand why what's happening is happening. We're doing it in a way that's much more open. We're trying to make as much of our software open source as we can. We save the secret sauce or the really good stuff

for our clients because I want to use the commercial world to fund this. The academic world isn't going to fund it. Foundations have provided us with some funding, thanks to them, but not at the scale that we really need to make this realize its potential. So we want to use

a commercial venture. We want to have something we can sell to people, they'll pay for, that we can fund to make it better and better. And we want to use it to do good for the world. We want to make it available to central banks and treasury departments and NGOs and people who want to understand how the economy ticks. And in particular, we want to use it to guide us better through the

to the energy transition driven by climate change. - All right, so two big questions or thoughts out of this. Let's talk about from the private sector point of view, what are you trying to do and what expected outcome are you banking on? - Well, we're trying to give them advice

That will help them make more money and maybe do so more sustainably. I mean, my long-term dream is to do for economic decision-making what Google Maps did for traffic planning. You know, Google Maps is a great thing. You've got it on your phone. You go, I want to go from A to B. It tells you the best way to get there. It screws up once in a while, but mostly it's pretty darn good.

And where Dunkin' Donuts is. Yeah, exactly. And how to go around the turn right to get in the entryway. We want to build something at Macrocosm that has every business in the world in it. You can see your business. You can see how your business is fitting into the rest of the economy. You can ask, what happens if I change my strategy? You can ask it what if questions. Got it. Maybe if I get this new thing, it will make me more sustainable. What's it going to do to my profits?

And have a projection about what's going to happen, have a prediction that people can use to run their businesses better and hopefully more sustainably. On the private side of the business, it could sound a little bit like you're trying to judge the roulette ball and predict the roulette ball in the casino and somebody's going to get rich and somebody's going to lose off of this. Are there ethical considerations you need to think about? Sure.

You know, technologies are always scary because they can be used for good or they can be used for bad. And unfortunately, it's an aspect of a capitalist economy that some people get rich and some people become poor.

One of the things that my colleagues have shown is that just by chance, this is something we should expect is going to happen. If you take a laissez faire economy, you turn it loose just a chance. You have a good luck in a few of your business dealings. Somebody else has bad luck in a few of their business dealings. You end up richer than they do. And if you keep running that long enough, the differences become extreme. So in fact,

In order to prevent that from happening, we have to understand how to guide the economy to damp that effect, to make inequality less bad than it is. That's why we really need to understand, well, what's the best way to do that? We'd like to find the right policies. Is it better to do a minimum wage, implement a tax, universal basic income?

tariffs? I mean, what do we do? What is the right way to do this? Where's the public good take you in this? The central bankers can be using and they can be running simulations to say, well, what if we change our interest rates? What if we change policies about collateral? What if we provide a stimulus to this sector of the economy? What if we change the tax rates? What if we put in a minimum wage? Ask all those what if questions.

and get a better understanding of what the likely outcomes would be. I think if the central bank knew how to run the economy better, that's a public good. The depression that followed the 2008 financial crisis

That hit a lot of people very hard. And, you know, in general, I think business cycles could be smoother. Recessions could be smaller. We just need to better understand how the economy works so that we can steer the economy in a more sensible way. We're already seeing central banks using some of these tools. The agent based model of housing markets I mentioned is now used by about seven or eight central banks.

In Europe, the Bank of Canada is now using one of these models as part of its regular decision making. The Bank of Italy is developing a model like this. The Bank of Hungary is developing one like that. I think we're going to see more and more central banks using it, and this wave will build through time. Another area that you've been looking at the implications of is thinking about technology and innovation. How do you think about technology innovation, technological innovation?

Well, I mean, first of all, technological innovation is what drives economic growth. So it's really important to understand it. The surprising thing about it is we can't predict what the new innovations will be. But what we've seen is you can often predict the rate at which things will improve. So Moore's law being a great example, you know, in 1965, Gordon Moore said the density of electronic circuits doubles every two years.

That prediction has held since then quite well. Now, it turns out lots of other technologies satisfy some version of that. Solar panels, for example, have improved at a rate of about 10% per year.

solar panels now are a factor of 10,000 cheaper than they were when they were first commercially used in the Vanguard satellite in 1958. Now, not all technologies are equal. 40% per year improvement under Moore's law, 10% improvement for solar panels. Fossil fuels over the last 140 years have not gotten noticeably cheaper once you adjust for inflation. So even though there's a lot of technological progress in that sector,

Things have just worked out for reasons, frankly, nobody really understands. We're trying to create a roadmap for how to bring about change as fast as possible, but with a minimum amount of suffering and disorder and chaos. And we've done a couple of things. One is gathering lots of data on technologies so that we know which technologies make good bets.

But the other is mapping out how industries behave. And in particular, we're in the process of building a model of energy investment, an agent-based model, where the agents are energy companies.

We're literally doing it at one-to-one scale in a literal way. So you can look at all the energy companies in the world, the assets they own, be they solar farms or oil wells. And we have a 25-year data set showing us historically how that system has changed that we can use to calibrate our model and then use the model to predict what's likely to happen going forward, depending on what policies the governments of the world enact.

Going back to my earlier Google Maps for Economics idea, companies can see themselves in that map and say, oh, OK, what happens if I change my strategy? How can I do that sustainably and still make a profit? The energy transition to deal with climate change is going to happen faster than most people think. Our models are telling us that, you know, it looks like not much is happening, but that's because the process is exponential.

And exponential change can be misleading because it's small, it's small, and then suddenly it's big. And it goes from being small to big quite quickly. So it's going to happen quickly. Now, are you talking EVs, carbon capture, everything included? Well, carbon capture, I don't think, is going to be a big player. Okay. And our models view carbon capture...

You know, we'll need it for things like cement and the hard to decarbonize sectors. But I don't think carbon capture and storage on a gas plant or a coal-fired electricity plant is ever going to be cost competitive. And I think we're going to get there mainly by using solar energy and wind energy, long-range transmission, long-term storage using hydrogen-based fuels. Geothermal is an interesting term.

dark horse that could end up playing a significant role. What about nuclear? I don't think nuclear is going to play a big role either, just because it's expensive. Okay. Nuclear is one of these technologies that hasn't dropped in cost. It started in 1958, about the same year as solar energy, and it now costs three times what it did then. Hmm.

Same period of time, solar energy is dropped by a factor of 10,000. The key bottleneck is better storage technologies. We need to be able to store energy at a low cost. Battery technologies are getting cheaper and cheaper very quickly. So overnight storage is not a big problem, but batteries can't really provide us with long-term storage to deal with cloudy days, cloudy still days when we don't have solar and wind energy.

So, and I think within 20 years, the transition will be mostly done. There will still be competition from natural gas because as the demand for fossil fuels drops, the cost is going to drop too. So they'll get more and more competitive. Data is becoming more accessible, easier to collect, easier to transmit. Computing power is getting stronger by the second.

Do we expect that you will start finding complexity economics being quickly adopted at some point, much in the same way where you talked about energy evolution? It builds little by little until it's just here. Yeah, I think that I mean, I've seen this in the past. I've been involved in.

a couple of scientific revolutions. Chaos, for example, when we first started doing it as graduate students was an unheard of thing and it seemed a little weird and flaky, but then we showed that you could see it in experiments. It's now a well-accepted thing. Complex systems, similarly, there's an increasing recognition that

many problems really are complex systems and we can learn by looking across different disciplines, that's become more well accepted. I think complexity economics is going to undergo a similar transition and I think it's going to have huge practical impact. And once the success stories

really become clear. And once we have models that can be used day in and day out to answer the kind of questions you were imagining the world leaders are going to pose to us, and once we show our track records better than the other guys, there's going to be a massive switchover. Big Brains is a production of the University of Chicago Podcast Network.

We're sponsored by the Graham School. Are you a lifelong learner with an insatiable curiosity? Access more than 50 open enrollment courses every quarter. Learn more at graham.uchicago.edu slash bigbrains. If you like what you heard on our podcast, please leave us a rating and review. The show is hosted by Paul M. Rand and produced by Leah Cesarine and me, Matt Hodap. Thanks for listening.