W. Brian Arthur
Professor
Santa Fe Institute

"Nonequilibrium Economics"

Michael called me up a week or two ago and I asked what I should talk about here. He told me I should talk about economics and I said, "what time during the day." He said it would be a little before lunch. I thought, "Well, this is not necessarily the subject I would have chosen for that time." [laughter]

I'm a great aficionado of the history of science and have been my entire career. When I used to read about science I would get this hankering to have been there at the time. Wouldn't it be marvelous to be in Germany in the 1920s when quantum theory was happening? Or maybe to be in England in the 1780s when the big breakthroughs were being made in chemistry, or in the 1940s and 1950s in the field of computation. I thought, well, it isn't going to happen in this lifetime-maybe next lifetime. If you look at any of these major changes where science shifts, the shift is in some sense remarkably fast. It takes place over five to ten years, which in the history of thought is really, really fast. The other thing you can notice is that barring half a dozen people in the middle of the change, nobody knows it's happening.

I was at the Santa Fe Institute in 1988-1990. I was in charge of a small group there, but it was a powerful group; the first research program at the Institute. We were still in the convent on Canyon Road. I had Stuart Kauffman, Chris Langton, John Holland and other really smart people. I was there for two years and well into our program when I went back to Stanford. Two or three years later I had one of these hankerings: "I wish I had been in some field when it was changing rapidly." And it struck me just out of the blue that maybe I have been, unbeknownst to me. During those years, things had been changing very rapidly and maybe I was lucky enough to have been there. I wasn't even sure at the time, but now I'm sure.

Economics is changing. When I thought of the subtitle to this presentation (Or, it's not your father's economics anymore) I thought it hadn't changed terribly since the time of Adam Smith, so it should read "it's not your great, great, great grandfather's economics anymore." [laughter]

I got into economics almost by mistake. I was working as a graduate student in applied mathematics for operations research. I was working for a small firm, McKinsey & Company in Düsseldorf, Germany. We used to fly to our client in a private plane and then we'd rent a car-it had to be a rather cheap looking car because you couldn't arrive in a private plane. We used to fly around Germany and some of my fellow team members would point out the window and say, "see that company down there, it's 50% owned by Shell Oil." And then they'd bank the plane and say, "that one down there was bought last year by BASF and BASF doesn't even know that they own this one at the moment but it will become important for them."

And suddenly people are telling me the history of the Ruhrgebeit, or industrial area of the Rheinland and how Krupp had been the steel company there when coal was important, how Germany had built this area up. This was in the later stages of Germany's economic miracle. I was enormously fascinated by the economy. It was enormously alive and organic and had path dependence (what was happening depended on what had happened in the past). It was full of all these players who were doing all these things and it was enormously alive for me.

So I went back to Berkeley to finish my PhD and decided to switch to economics. But something happened in Berkeley that greatly disappointed me. I was studying with the best people in the world. Three of my teachers have since received Nobel Prizes. They were incredibly smart. But to them, economics was a kind of museum of theorems. In a "glass case" was a theorem on some part of trade theory, and in another was one on intergenerational economies, and in another something on general equilibrium theory. There was something very deadly about it. I could do the mathematics, but there was nothing alive about the economics we were studying. It was a great disappointment. That experience at Berkeley motivated everything I did afterwards.

Let me start now by discussing complex systems. You probably know a great deal about them so I'll run through it very fast. The kinds of complex systems we talk about all have some features in common. There are many individual elements and they somehow do things because of the pattern they collectively create. This is an idea we evolved at the Santa Fe Institute. Chris Langton has a diagram that has all these elements doing things that react to the pattern they co-create. It's as if these elements-ions in their spin bias, or ants in a colony, or stars in a galaxy-collectively create something that they're reacting to. Each element may only react to the local field of the system.

When complexity started running around the scientific community in the mid 80s, people could start to ask questions and answer questions with this different approach. "What happens if you collide two galaxies?" There are billions of stars, and these stars collide. In that collision, each element, which might be a star of a different size, is reacting in a pretty simple way to the gravitational field caused by all the stars and if you wanted to sit down and work that by pencil and paper, you'd have a very difficult problem unless you made gross simplifications. Even working out what happens with three bodies is deterministic but very difficult to get any solution.

Why did this approach come along and is it a fad? It came along because in the mid 80s we all got desktop workstations. This question had been along for a long time but with the advent of computers, we could throw away the yellow pads. We didn't have to reduce everything to equations. Each of these elements became a program or object in the computer. We could simplify the problem and have 10,000 elements to represent the billions of stars in the colliding galaxy problem I mentioned. We might see the formation of spiral arms, certain pulsations and so on. We started to look at how elements reacted to the patterns they created.

This became a fad by the early 90s. We were all built up in the magazine Scientific American. Then there was the reaction, "isn't this all going to go away?" My answer is absolutely no. We're still in early days and discovering new ways to look at complex systems. One of the most recent ways is in networks, as you just heard in the previous presentation. If you add links to a network, those links may be added depending on how that network already looks and what behaviors it exhibits. It won't go away because complex systems thinking is the difference between looking at things statically and looking at things as they unfold. It's working its way through just about every science.

How is it working its way through economics? There's a curious thing about economics itself. It was one of the very first complex systems to be studied. Go back to Adam Smith's time. Smith was very aware that the economy consisted of what we call agents-people who do things. They could be companies or individuals and they would be interacting and doing something together. And in that interaction they would create a market. As it showed various prices, the agents would respond to those prices. Patterns of prices would appear.

Smith was aware that individual behaviors created something larger but obviously wanted to get analytical solutions. Economics started to get mathematized around the 1820s and 1840s (the time of John Stuart Mill) and then in the 1900s it became much more sophisticated. One of my teachers at Berkeley was very interested in economics being a collection of axioms from which everything about the economy could be deduced. The economy was supposed to be an axiomatic system. This deadened the whole field. You could compare it to Isaac Newton. It is said that one of the measures of the greatness of Isaac Newton was that his theory of dynamics was so brilliant that it held up all other progress for 100 years and so far outshone everything else that no one could see past it.

What economists did from the 1770s to the mid 1980s was to reduce the question. Instead of asking how the behavior of economic agents co-creates something that the agents respond to, they asked a simpler question, "What behavior is consistent with the pattern it creates?" That stops the motion and led to equilibrium economics where things are balanced and there are no incentives to change the behavior.

You may have studied general equilibrium theory. If you have many people consuming and firms producing, people taking part in the labor force and using their wage budget to buy things, what quantity of goods would be sold or produced in that economy and at what prices? The entire thing was in equilibrium. It was not an easy question to answer and was not satisfactorily answered until the 1940s.

Another branch of economics, game theory, asked a similar question. If people are interacting in some situation, what strategy could one of them choose given their ideas of what strategies other people might be using so that he or she could do best with their options. What kind of strategic behavior is mutually consistent?

There is a different approach with complex systems. If we don't mind studying more complicated systems we can go back to the original question and ask, "How does individual behavior adapt to the pattern it creates?" Or, "what larger patterns might you see, given that individuals might adapt and readapt and learn and change the behavior over time-what things might we see forming?" That creates a field of out-of-equilibrium economics.

Von Neumann made a very important point in the 1930s and 1940s. Most of mathematics up until that time had been based upon linearity or straight line functions. At that stage, mathematicians were starting to turn their attention to nonlinear functions. Von Neumann said that getting the theory of non-linear mathematics is like getting the theory of non-elephants. Linear problems are like elephants. They're well defined and we know what they look like. But what does a non-elephant look like? Non-equilibrium economics is larger, more general and encompasses equilibrium economics as a special case. It may be that as individual behaviors are adapting, whatever pattern they create, it may settle into something that doesn't change much. It's not that the new economics denies the old, but it contains the old.

When I studied at Berkeley, I was given this impression that economics was like a large table and over in one corner, measuring one millimeter on a side was the field of non-equilibrium economics-the exceptional case. We're finding now that it's the converse. The entire table is non-equilibrium economics, and here and there are pockets or patches that might be in equilibrium. But everything is always changing. It's like a kaleidoscope. The individual elements or agents might be the same, but they're creating new patterns. There's a book I ran across by Denis Diderot from the enlightenment, in which he says, "everything changes, everything passes on-there is only that which stays the same."

Once you take the out-of-equilibrium approach and you ask how elements adapt and patterns change in the economy, you get a very different set of themes than you did in equilibrium economics. You move from stasis to process and from equilibrium to a progressive ecology, or evolution. You move from rationality to cognitive understanding. Let me explain that.

If I'm looking at a problem and I'm assuming everyone is using all the information in the best way that they can, so that they have no incentive to change their behavior, they're trying to solve a deductive problem. And they're using everything in their brains to solve it. That leads to a lot of paradoxes in economics. It might take an economist six months to solve a problem that they assume everyone can solve instantaneously in their heads. But there has to be a lot of emphasis on rationality if you assume that no one has any incentive to change. If you assume that people adapt, on the other hand, you can also assume that the problem isn't that well defined and that they're just slapping a vague understanding on top of it and doing their best.

For example, ten years ago, what should we have done in Bosnia? That problem doesn't have a deductively rational solution waiting to be solved. People might try to put frameworks on top of the problem: "it's a little like this case I studied in the 1840s," or "it's more like something in the 1890's…" People are trying to put cognitive frames on top of the problem. A cognitive understanding means that the problem may not be well-defined. We have in Northern Ireland a situation that is very similar to the one in Bosnia or in Iraq and it's lasted for decades. The saying goes, "if you're not confused, you don't know anything." We need an economics that allows people to be confused and lay frames on top of it. As they get more information, they might add to the understanding and create a different frame.

Non-equilibrium economics also represents a shift to heterogeneity from homogeneity. You don't have to assume that all of your agents are like the identical, robot-like storm troopers in the Death Star. Instead, the agents can be individual and human. Also there's a very strong feeling that years ago, economics was supposed to be a soft science. If you ran into a physicist, he would not hesitate to tell you that economics wasn't really a science. That gave economists a massive inferiority complex. We know that one way to get around a complex like that is to react by going overboard by mathematizing your field and another way to react is to become arrogant.

We had a famous meeting at the Santa Fe Institute in 1987-the kickoff meeting for the whole Institute. Ten economists and ten physicists were present. At the end of the meeting, Richard Palmer came up to me shaking his head. He said, "I thought that physicists were arrogant, but…" [laughter]

So, economics has become a high complexity science. To understand why, lets look at physics for a moment. Consider a question about "spin glass." It's a bunch of ions. Randomly you stick different ions like copper into it, and the copper ion has to "decide" whether it will point up or down. It makes that decision very dumbly on the basis of its neighbors. When physicists asked me what the basic difference was between physics and economics, I would say that economics is more difficult because our "copper ions" can think and they might think along these lines: "I might point up, but if I point up, my neighboring ions might point down and they'll point down, knowing that their neighboring ions will be doing something else. So if I point up based on what I think they might do, then my decision might be such and such…" Our ions in economics have strategy and expectations. They forecast and have to strategize. This makes economics at least one order of magnitude more difficult than standard physics. In turn, it's brought an enormous number of physicists into economics-Per Bak was one of them.

The various parts of economics include things like trade theory, economic geography, the economics of high tech, microeconomics and macroeconomics. I want to concentrate more on finance for the remainder of the talk and show you how things are playing out in that area.

Finance is driven by theories of how forecasts are made. This is one of the most difficult areas of economics-how do people make forecasts or predictions? Every action in the economy is based upon some prediction, or in economic jargon, an expectation. If you buy a candy bar, you're expecting that the candy bar will be fresh, that what's inside the wrapper is what you bought last time, and so on. This sounds trivial, but every economic action is taken based upon some consequence you're expecting. If you buy a house-a pretty simple transaction-it's bought partly based upon your expectation that you won't find termites in the house, what the house might be worth in ten or twenty years time, what's happening in the neighborhood, and so on. Even trivial economic actions are based upon expectations.

Expectations theory became very popular in the 1980s. It was the high tide of the standard approach in 1985 or 1990. To create a theory about expectations in the equilibrium or static way, you have to assume that people have no incentive to change. You also assume that they're coming upon forecasts exhausting all possible information to them. In equity markets they would be looking at past earnings and prices, the behavior of other investors, and so on. They would use all that information. Then economists took that one step deeper and said that people are not just looking at earnings or cash flow but also looking at the way other people are betting. So we got into the cycle of hyperrationality.

When we're making a forecast, we want to find out all information. But everyone else is gathering information. The way they're playing the market is information and there's information about that information and it creates deeper and deeper hyper-cycles. It led to paradoxes. If everybody gathers information and makes their bets on that information and if I think they're good information gatherers, then they'd be making smart decisions. Then why should I gather any information myself? And to add further difficulties, the theory assumes that everybody is the same-they're all Death Star robots. That would then mean that nobody should gather information if that's the conclusion that I came to. But then we were assuming that they were acting on information. And where does the information come from?

So economics started to chase its tail. The theories from the 1980s were very beautiful but suffering from all this hyper-rationality. I began to realize about that time that there was a race going on to see if any of the people working on these theories would get their Nobel Prizes before the theories were done in.

So on to the theory of rational expectations. I often think about teaching undergraduates about rational expectations. I had a friend who came up for his PhD orals in Berkeley. The orals were in thermodynamics. The first question he got asked by the august committee was, "Mr. Patterson, how would you explain entropy to the man in the street?" Patterson thought deeply about it for about five minutes and then said, "I would tell him to forget it." [laughter] But I won't take that route today!

Suppose that all of us are playing some sort of investment game and we're all in our rooms in the hotel and we're wired together. You don't know what everybody else is doing but we're playing some sort of game that involves forecasting or expectations. You don't know who's in the next room. There's no collusion or Wall Street Journal. You're trying to be in some game that involves some sort of asset that might be appreciating and you're wondering whether to invest in it or not. Suppose you have some idea of what it might mean for dividends to be paid or earnings out of this asset. You think, "maybe I should buy on the basis of earnings and that would be rational." But then you say, "what if the people three doors down are not rational and I can make money out of that." And what are those people thinking about other people? Well perhaps my forecast ought to change if I start to assume that not everybody is on board with this so-called rational solution. All I have to assume is that some people are not that smart or don't see it my way or have a different theory of what's going on or read a different newspaper. Then my expectations change. If you start to look at that argument, you begin to see that the actual problem of expectations formation is deductively indeterminate. There is no correct solution. Most economists who teach finance courses will not tell you that. There's no correct solution because I can't rely upon everybody to have exactly the same information and come to exactly the same conclusions that would support the price that actually comes out.

So there isn't any more of a correct solution than there is to the question, "who is the most beautiful actress in Hollywood?" In a sense-and Keynes pointed this out in 1936-if someone asked that question, I'm basically trying to figure out what others might figure out and so on. Suppose you're in this situation where you win in this game if you choose the most beautiful actress in Hollywood and that's the most popular choice. Maybe you think it's Julia Roberts but can you rely on everybody else to have the same taste? You get into these loops of trying to second guess.

Rational expectations theory is a way out of this sort of thinking. It's an academic dodge. It's brilliant and it doesn't work. You're asking whether in any market there is a magical forecasting machine where at one end you feed in the Wall Street Journal and snippets you've heard about Bill Miller's portfolio and out the other end comes a price or a set of prices for the next period for the stocks you're interested in. And everybody supposedly has the same machine. So all the information goes in and forecasts come out the other end. What would that machine look like if it gave forecasts that led people to have buy sell behavior that resulted in prices in that market that on average would be consistent with the machine's forecast? That's the question we're asking. We don't ask that the machine is perfect, only on average that it shows no bias-that on average it's correct. It seems a very reasonable think to ask and this is what the economists asked in the 1970s and 1980s. It's quite difficult to solve these kinds of problems.

Around 1985, economists split into two groups. There was one group who said that the theory was wonderful and they supported it. They didn't necessarily feel comfortable with it, but acknowledged that it was the best that could be done. The other group, including myself, said, "this theory doesn't ring true." This bothered me for several years. There was a French general who was shown the troops in formation, beautifully dressed in scarlet and doing all their drills. He commented, "this is magnificent, but this isn't war." Rational expectations theory is a beautiful creation but I don't think it's realistic.

Let me walk you through a problem. I was trained in engineering. Very often in engineering you take a structure, like a bridge, and you stress it until it breaks. So I figured I'd come up with a problem to stress rational expectations until it broke.

It's called the El Farol problem, inspired by a bar in Santa Fe along Canyon Road. Imagine that one hundred people decide independently whether to show up at the bar or not. Here is the rule. It's a small bar but has a great atmosphere. If it's too crowded, it's just horrible, but if it's not crowded, you can have a really good time and hang out with your friends. So the rule is that if you think more than sixty people will show up at El Farol, you'll stay home, and if you think that fewer than sixty will show up, you'll go. What happens to attendance? You'd think that this would be simple until you start to think about it. There's no collusion-nobody's phoning anybody else. It's everybody's private decision and they're all perfectly rational.

Someone might go through the thought process like this. Three weeks ago, there were 35 people there and two weeks ago there were 43 and last week it wasn't that crowded, only 23 people showed up. If I average that, it's 32 or so. So I assume that based on the average, that next week only 32 people will be there. But I also know that everybody is equally rational. So everybody will do a similar calculation and everybody will decide to go. There will be 100 people there. Well, if 100 people are going to go… Even if everyone doesn't think the same, a large number of people will go because it hasn't been crowded. If I'm hyper-rational I can go one step deeper. If 100 people are going to go, then I'd better not go. But everybody else can come to the same conclusion so nobody's going to go. But if nobody's going to go, then I should go. The problem can be diabolical.

Let me do the same exercise using rational expectations. Everybody is issued a rational expectations machine from the Santa Fe Institute. Past attendance is fed into this machine. The 100 people each press the button on their machine to see the attendance next Thursday and it says 84 people will turn up. Nobody goes, and that nullifies that forecast. The next week it says that 23 people will turn up and everybody goes, which nullifies that forecast once again. If you do the math, you find out it's a really bad predictor and is wrong because everybody believes it and their behavior nullifies the prediction. It's the expectations version of prisoner's dilemma. I published the problem in 1994 and it's been fairly well neglected by economists. There have been 200 papers in physics published on the El Farol problem and two or three books are coming out. It started an industry in physics.

So how do I think about a question like this? My way out was to imagine that agents form subjective hypotheses about the problem. That is, they might have very different predictors. Some agents may average the last three weeks, others the last four weeks and then add or subtract some amount based on the probability that others will be looking at averages as well. I gave the agents several predictors and then let them learn which ones were currently most accurate. One hundred bargoers each have a different hypothesis about what rules the market and they're each making their own subjective predictors.

I put it into the computer and hit the return button to see what would happen. I thought it would go chaotic and subsequent work proved that was the case. But an ecology of predictors emerged. My forecasting method is only good in this situation, in an ecology, or a sea, of forecasting methods that the others are using. My method is only going to be accurate if it's accurate against the actions taken by other people. And their actions depend upon their predictors. So there's a kind of fitness and evolution going on. If my predictor isn't much good, I'll switch to another one.

Strangely enough, the model arrives at the efficient solution-an attendance of 60-and the average doesn't deviate from that. Why does the efficient solution come out? One person might be using a stupid predictor like saying that there will only be 24 people at the bar each week regardless. Imagine that the attendance averaged 45. Then anything that predicted 45 would start to get very accurate. Then more people would use the predictor, and more people would show up, disturbing the accuracy of the predictor, and causing it to fall in its fitness. So there's a natural attraction to the equilibrium point but the predictors themselves never settle down. There was no stage at which you could say that everyone was using the same prediction method. That's because if one person changed, that could ripple through the model and change attendance and nullify other predictors. What happened was that market psychology had arisen and there was an ecology of forecasting devices.

The economists who looked at this were really rattled. I'd like to apply the same sort of thinking to stock markets or asset markets. The problem in economics is called the standard theory of asset pricing. There are many models of financial markets, but I'll use the standard theoretical model. Everybody has the same forecasting machine, which is correct on average. Information comes into the machine. Based on that information, the machine spits out a forecast of the next period's price. Based on that, people buy or sell to adjust their portfolio. Those orders go through a market maker who adjusts the price and the new price is declared. What forecasting machine is validated on average by the time sequence of prices? That's the question we want to answer. It's not easy to work out. We're looking for a machine that will give accurate predictions, not just the predictions itself. What forecasting machine could validate itself?

Several results come out. One is that the market is efficient. The prices are validated. There's no way to beat this market because it's efficient and everyone is doing the same thing. If everybody forecasts the same, then everybody would be doing the same thing, so there would be nobody on the other side of the transaction. So there would be no trading. Everybody would have an identical portfolio. There's no incentive to gather information. There's no speculative or technical trading possible. If past prices had any useful information the machine would already have discounted that information into the current prices. It gives outputs that don't have fat tails, prices are uncorrelated, volatility is stationary, the market is statistically stationary. To first order, this model works. Economists have gathered reams of data, tried it against the model, and roughly it's correct. But why do people trade in the market at all if the model works? There should be no investing beyond people simply wanting dividends.

Here's how we approached the problem by creating an artificial stock market. I met John Holland in 1987. He invented the genetic algorithm-a way to make computer programs smarter. When I heard him talk, somehow I knew he had the answer to a problem, and all I had to do was cook up the problem. What if I had a real economy in vitro in the corner of my office? Obviously it would be in a computer and I'd start with a bunch of agents. They'd maybe start trading sheep fleeces for obsidian and then they'd be allowed to discover things and the model would bootstrap itself up to more complications. I might come back in a couple of days and find that they had joint stock companies. Maybe labor unions and central banks would emerge. That was my dream but I didn't see how to carry it out. I collared John in 1988 and hit him up with the idea of creating a financial market in my computer. So we did.

We created a number of artificially intelligent investors. There were maybe 1,000 of them. We assumed they all differed. They act inductively, coming to their own conclusions as to what moves the market. When I hit the return button, the computer issued a single stock and there followed a sequence of earnings. Each of my agents had many forecasting models or hypotheses about what moved the market. We gave them sixty models. Each agent had many rules so it could recognize the patterns that might tell it that the market had gone up the last three periods, dividends have been such and such, the stock is selling at x times earnings and so on. Each agent used a different set of these conditions and from the set it used a function to come up with a decision. Individual agents had different hypotheses. They were learning to recognize patterns and to do forecasting.

We made these patterns totally random to start with. Investors get smart in two ways. First, they start to notice which hypotheses or functions work well and they act according to those. Second, poorly performing models were dropped out through a "trap door" and discarded. New ones were generated in their place. We had one of the University of Chicago people advise us. He was one of the Death Star economists. We asked him what he thought would happen and he said, "nothing-it will converge to equilibrium quite fast and won't budge after that." John Holland and I thought we would get really rich behavior out of it including market psychology and technical trading.

What did we find? No one won the bet. Both cases came out. We had a parameter, lambda, called the learning rate. It determined how fast agents would drop old ideas. If something isn't accurate, how fast do you switch to using something that's better or how fast do you pull the lever on the trap door and discard it? If that rate is low and you start the market with everybody having similar forecasting machines, then it all converges and accumulates around rational expectations with some people occasionally making deviations to see if they can extract any more value. It duplicates the standard model.

But if you switch up the learning rate a little higher, then you get something much more like El Farol and it quickly becomes complex. We call this the complex regime. A market psychology emerges. People's predictions keep readjusting depending on other people's predictions. Technical trading emerges. We found that some predictors might say, "if the market goes up, buy." And that would put the market up further and validate that predictor temporarily. There would be a little run up in prices. Overall it was clustered around rational expectations and it was hard to tell the time sequence apart from the rational expectations one. To first order, the prices were efficient but we found out that some traders could exploit differences and if they got in early on a run up and got out, they could make a lot of money.

We also found that avalanches of change would run through the system. I'd call up the person who was watching the program and ask, "Paul, what's our market doing today?" He would look at my machine and say, "it's been very dull." Two days later he'd say "wow, this thing's gone crazy; you wouldn't believe what's going on." It's a little bit like earthquake faults. Some of the agents might discover a better method to forecast and they'd be taking heavier positions. That would shift the behavior of prices and in turn cause other agents to shift their forecasters differently. That's like an earthquake that triggers other earthquakes that trigger yet other earthquakes. We'd get power law behavior. Change triggered further change. Then it would be followed by a period of quiescence because everybody was using reasonable hypotheses. During those periods of volatility there were huge moves in prices and that gave us fat tails-not normal distribution of prices but sometimes quite extreme distribution. All of this behavior you see in real markets.

We also did something we came to call the Jurassic Park experiment. We were doing statistical measurements on this model. To first order, we couldn't distinguish our stock market from the rational expectations one. It was in the second or third decimal place. The whole thing looked to be statistically stationary, meaning that the behavior was the same except for the high volatility, low volatility patterns. We had no evidence that the market ever changed.

We let the machine run for 100,000 periods and then checked to see who the smartest investment programs were. We took them out and froze them. Then we let the machine run another million periods. And then we re-injected the smart programs back in. But they weren't allowed to catch up. They were adapted to the market as it was one million periods ago. They underperformed the average by about 10%. Therefore indirectly we had proved that market psychology had changed.

We called it the Jurassic Park experiment because these guys were the velociraptors or ace predators of their particular day. We froze their DNA and stuck them back in but the world had changed. I ran into Stephen Jay Gould one time and explained it to him. He thought that my outcome would more likely match the fate of velociraptors if we could actually bring them back today. They'd be terribly badly adapted to today's atmosphere, prey, diet, and so on.

If everybody started more or less identical, and explored very slowly, then efficient market behavior resulted and it served as an attractor. You'd lose money if you deviated from it. However, if you let people do a little more exploring, the market does support efficiency but a psychology emerges and there are periods of rapid change where people affect one another's behavior and adapt. And in real markets we see fat tails, market psychology, technical trading and Jurassic Park phenomenon as well. If you back off from the purely deductive model, you create a more realistic looking market, but it's still a pretty efficient one.

Where does all of this get us? So what? We start to see novel phenomena. Economics is going through the biggest shift that it's gone through since Adam Smith's day. If you went to the American Economic Association meetings and buttonholed the average economist to ask them if economics was changing, they'd probably reply, "I haven't heard that." But if you grab the top theorists, they'd say that it's changing and changing rapidly. It's going from being deductive to being inductive. It's going from thinking about a sort of mechanistic equilibrium and well-oiled machine with every part in balance with every other part to something more organic, an ecology, where my behaviors only fit in an ecology of actions, strategies and forecasts that's created by your behavior. So markets are ecologies of beliefs and, almost equivalently, of strategies.

Novel phenomena and cascades of change create perpetual novelty. I was talking with John Holland once about chess. Chess has a solution-by the way-it's just really hard to compute. Every two-person, zero-sum game has a solution. In principle, it could be solved mathematically. My caricature of economics is a bit cruel. Imagine that two University of Chicago Rational Expectations economists sit down and play chess. No moves are made for three hours. Finally one stands up and says, "I resign; there's no way I could have beaten the 167th move you would have made." When I asked John about the approach, he noted that chess grand masters of today could handily beat the grand masters of 100 years ago. I asked why. He said that the grand masters of today understand what the grand masters did 100 years ago but the envelope of strategies has pushed out. It doesn't mean that they're smarter. They have more of a repertoire to think about. There's perpetual novelty. It builds on what was there and continues. That is shocking to equilibrium economists who believe that aside from technology, not much is changing.

The new model suits the 21st Century economy better. Equilibrium economics came out of the 19th Century-a static one. Running a coal mine from year to year is not very different and you could assume stasis or equilibrium. I live in Silicon Valley and there's no way you can say that anything is in equilibrium there. The game keeps changing almost month by month. If you disappeared for two months and then came back, you'd have no idea what was happening. You'd be like one of the Jurassic Park players. The economy keeps changing. The high tech economy introduces novel business models faster and new strategies, ideas and hypotheses. You have to keep readjusting your strategy.

The comments, opinions and any forward predictions presented about any particular security, the economy and "the market" are based on the analysis of the speaker. These are not necessarily the opinion of, and should not be construed as a recommendation on the part of Legg Mason Capital Management or any of its affiliates.