SOCIAL AND INFORMATION SCIENCES LABORATORY: Markets and Other Noisy Human Artifacts—Can Computation Bring Them Out of the Bronze Age?
A Conversation with Yaser S. Abu-Mostafa, K. Mani Chandy, and John O. Ledyard
Winter 2003

PDF | Table of Contents | Go to Bottom of this Page

From left to right: Yaser Abu-Mostafa, Mani Chandy, and John Ledyard.

ABU-MOSTAFA: There is an abundance of data and an abundance of computational resources in the world, yet our ability to manage these resources, to be able to look at data and efficiently extract the correct information, is limited. Highly distributed, data-rich, and generally unstructured, the world's financial markets seem to work well—remarkably well given the loose structures and lack of supervision—but they can be improved. The players in the markets are individuals, institutions, sometimes simply computer programs. They are looking at pieces of information that may be different from one source to another. They're all interpreting information differently. They have their own ideas and preferences regarding risk, value, volume, etc. Eventually, all of this is aggregated in global quantities like price, volatility—things of that sort. So a basic understanding of how such a general system results in efficient information aggregation is very important for two fields: economics and engineering. On the economics side, we would like to better understand markets and eventually be able to design markets. Once we do that, we can design markets in different arenas where there are no markets now. From the engineering perspective, we're interested in learning from the principles of how markets work how to generally manage distributed information and be able to aggregate it in a meaningful way.

---

The question is whether we can leverage new advances in information science and technology to design new markets.

---

LEDYARD: Economists would suggest that perhaps they know something about markets already, that 200 years of study have produced remarkable insights about them. What's of importance in this Center, however, is the role of technology in the way markets operate. There are barter markets, which have been around for thousands of years, which are not very efficient. The information technology underlying the New York Stock Exchange is still primitive in that humans are crucial at many points in the process. Many aspects of markets work wonderfully. If I'm fixing my house and I need a nail, I know I can go to the hardware store, and the nail is sitting there waiting for me. How did they know I would need a nail that day? It's not centrally planned. It's not managed the way engineers like to manage things. It's dispersed, disorganized, decentralized, but it does compute some pretty incredible things.

There are other pieces that don't work very well: supply chains, for instance, and public good kinds of problems. Markets don't work very well in these cases, partly because there aren't very many participants. They're very specialized and may not have much volume, so you can't rely on immediacy. The question is whether we can leverage new advances in information science and technology to design new markets. Economists have generally attacked these problems assuming computation was free and easy—which it's not. Bringing the reality of information processing into market design is really important. The role of SISL is to bring the exper-tise of engineers and information scientists together with the exper-tise of economists—each has something the other doesn't. Working together, something really special will emerge.

ENGENIOUS: Will you be inventing new computational tools to deal with these problems?

CHANDY: At this point, I don't think we really know. That's why SISL is so interesting. From my point of view, the research of this center will bring "power to the people." Economic power has two parts: resources and information. Information technology today is at a place where one half of the economic power equation—information—is widely available. And this represents a significant dispersal of power from the few to the masses. I'll give you three examples of how this is going to change your life.

When the defense department wants to buy planes, it puts out a request for proposals, companies respond, and they finally choose a plane. DOD can afford to do that because DOD budgets billions of dollars for a plane. If you want to buy a car, you don't have the same flexibility. You don't request proposals for cars that fit your specifications. Nor, if you want to travel, do you put out a request for proposals for tours with certain specifications. You can't do that because the cost of the transaction is high. But apply computational resources to this scenario, and things will change dramatically.

The second example is futures markets. We are familiar with the futures market on things like wheat, oranges, pork bellies, and so on. But what if there were a futures market on services like carpentry, plumbing, and electrical work when you add on to your house?

The third example is the creation of financial derivatives. Today, large financial services companies create financial derivatives tailor-made for companies doing shipbuilding in Poland, for example. Financial services companies create custom-made derivatives and sell them for lots of money. But with the kind of technology we will develop, companies will want to sell you derivative products for yourself based on your personal situation.

So these are a few examples of how the Center's research will help economic power devolve to "the people."

LEDYARD: Here's a sort of common theme in the story: let's say you want to build or buy something, a car or house or vacation. Today, you have to go to somebody who's packaged everything up without your particular needs or desires in mind. You can have people specially build your cars for you, specially build your house, but it's expensive. With computational capability, you can allow people to express what they really want to buy in a marketplace. So, rather than hiring a project manager to build your house, the computer organizes schedules, locks in the futures contracts on carpenters, masons, roofers, and locks in a schedule. This is going to require some interesting theoretical work in terms of how you capture what are essentially "metaphorical" ideas—the idea that I want a house overlooking a lake, with three stories, etc.

The classic example of where this gets mishandled is the California electricity market. That was a designed new market. Somebody said "Let there be markets," and voilà! They did that in Russia and it was a disaster because they forgot they needed banks and property rights and various other things. In California, they forgot to integrate engineering, electricity, and the laws of physics with the market. They also made some bad assumptions about how people behave. There's been research, a lot of it at Caltech over the last 30 years, which could have prevented this problem from occurring. Simon Wilkie had a very nice article in Engineering and Science [Economic Policy in the Information Age, E&S, Vol. LXIV, No. 1, 2001, page 28] on just this problem. Engineers like to control everything. Economists hate to control anything. Integrating these two kinds of approaches is going to be interesting, but it's required for a successful energy market. Give SISL up to ten years, and we'll pull it off.

---

Designers of distributed systems can control the rules of the game, but they cannot control the players.

---

With experimental economics, we have a way of demonstrating to people how these things really work. We can actually bring science to bear on it. The combined energies of those working in this Center will create an intellectual core that anybody working in these fields simply won't be able to ignore.

ABU-MOSTAFA: I'd like to add something to the idea of exotic derivatives: one of the biggest advantages of having the computational technology to price these things is being able to communicate the derivative to so many players, thus creating a commodity. It becomes a real market—a place of exchange between buyers and sellers—because of the number of players and because of their ability to come to an agreement on price and to communicate instantly.

CHANDY: John said that engineers like to control things... but a true distributed system is one in which you don't know the participants, or even how many there are. Designers of distributed systems can control the rules of the game, but they can't control the players. So there are two parts to a distributed system: the visible hand, or the rules by which all the participants play, and then the invisible hand—how many participants, and how participants operate provided they play by the rules. Markets are beautiful examples of this, and we need to understand better how we get global behavior from these policies. This is very much an engineering problem.

LEDYARD: The process Mani is describing is what economists call mechanism design. It's also very much an economics problem, where we recognize the incentives people have to follow the rules or not.

---

Engineers like to control everything. Economists hate to control anything.

---

ENGENIOUS: What other Caltech faculty do you anticipate being involved?

CHANDY: In computer science, there are two relevant areas: applying economic principles to distributed systems, and applying technology to economic principles. For the first part, we have Steven Low's work on the internet and algorithms, and also John Doyle's theories on control and robustness applied to non-traditional applications like markets.

LEDYARD: We have been using markets as examples, because many people have contact with markets. But the same conceptual structures and questions arise in issues of voting and elections, committees, and organizing large organizations. In my Division, we have Tom Palfrey working on political processes. Peter Bossaerts studies the dynamics of financial markets and the process of price discovery. Charles Plott studies information aggregation processes. Matthew Jackson does fundamental research on networks. All of them will be involved, as well as others.

---

If we knew what would happen two years from now, it wouldn't be research.

---

CHANDY: We will also work with people from the Center for the Mathematics of Information. We share an interest in the growth of data, the extent of data. Essentially, data come in three forms. There are structured data, like the price of a car. There are totally unstructured data, like news about an explosion in Azerbaijan near an oil well. And then there are semi-structured data, for instance, auction information like you would find on E-Bay. All three kinds are increasing everyday, so the work of that Center—creating efficient representation choices—will be useful to the work of SISL.

ENGENIOUS: This work, taken as a whole, sounds like it could be an entirely new intellectual discipline.

LEDYARD: It has the potential.

ABU-MOSTAFA: When you design a research enterprise like this, you have to have a gut feeling about it being special. But then these things create a life of their own. If we knew what would happen two years from now, it wouldn't be research. Once the collaborations begin, who knows what can happen? We've been discussing markets because they are tangible, and have real and immediate impact on people, but there is a wide range of applications for this research, including the organization of corporations, the health-care system, etc.

CHANDY: I really believe that SISL will have a direct impact on society, on ordinary people in addition to large institutions. This confluence of economics and information technology will impact everybody. ENG