In April, the National Academy of Sciences will bestow its 2018 John J. Carty Award to a trio of Stanford economists: David Kreps, the Adams Distinguished Professor of Management at Stanford Graduate School of Business; Paul Milgrom, the Shirley and Leonard Ely Professor in the Department of Economics; and Robert Wilson, the Adams Distinguished Professor of Management, Emeritus.
In the citation for the prize, the National Academy cites two different contributions: "For making fundamental advances to game theory by showing how incomplete information alters equilibrium outcomes, [which] enhance our understanding the impact of reputation and the emergence of cooperation,” and "[for] the analysis and application of auctions for allocating scarce resources." We sat down with them to get their story.
What are some of your proudest achievements?
Kreps: In 1990, roughly, then-Dean Chuck Holloway told Jim Baron and me to develop and teach a course on human resource management. It was to be a “he said, she said” affair. I’d say what economics has to say on some question; Jim would say what sociology and social psychology had to say; and the students would be left to sort out the conflicting viewpoints. But we found that economics — the right variety of economics — and sociology and social psychology were complementary. Each filled in pieces of the puzzle that the others omitted.
In game theory, for example, you need to know what people want and how they form their beliefs about what’s going on. Social psychology looks at how your tastes and beliefs are shaped by experiences. So while economics provides a framework for analysis, social psychology supplies the human element.
We wrote a textbook on the subject, and I’ve just completed a trade book on the topic of workforce motivation (The Motivation Toolkit, W. W. Norton). I’m most proud of this work, because it takes ideas from game theory and transaction-cost economics and blends them with the other social sciences to provide actionable insights for practicing managers. That is the right goal for economists in schools of management (and, in fact, more generally): Use theory and empirical analysis to answer practical questions.
Milgrom: I am an enthusiastic fellow, so I’m always most excited about what I’ve been doing recently. In 2017, the U.S. government completed something called the “incentive auction,” in which it purchased enough broadcast rights from TV stations to free up 84MHz of bandwidth for mobile broadband uses. That auction had historic levels of complexity — more than 1 million constraints.
I led the auction design team, working especially closely with [Stanford professor of economics] Ilya Segal. The complex algorithms we used to analyze the constraints were designed and implemented by [University of British Columbia professor with a Stanford PhD in computer science] Kevin Leyton-Brown. I’m proud that our team helped create so much value, and I’m hopeful that the theory we developed to guide our design of the auction will have wider application.
Wilson: I am proud of my work with Dave in adapting game theory to dynamic interactions among parties with different information, and I am also glad that I got to work with Paul on the initial design of spectrum auctions. I’m also fascinated by my current work, which is on the features for sustaining cooperation in long interactions, such as [business] partnerships.
What did your field look like when the three of you got started?
Wilson: Economics became much more mathematical in nature following the Second World War. The first topics that were addressed were “classic” topics of competitive markets, the economics of supply and demand. By 1970, all the low-hanging fruit in these topics had been harvested, and economic theorists moved on to information economics. For example, what happens to incentives when some people have private information? Think here of the “lemons market,” where a car dealer knows which cars are “lemons” but the customers don’t. This broke important new ground.
Milgrom: At the same time, pioneers in game theory such as Robert Aumann, John Harsanyi, John Nash (whose life was portrayed in the book A Beautiful Mind), and Reinhard Selten were developing highly mathematical tools to analyze problems of social interaction.
Those tools offered new ways to describe much older models, such as those dealing with oligopolies. But the work by Nash and others wasn’t about economics. It was highly abstract mathematics about game theory itself. For economists to embrace game theory as a tool, we needed to do more than repackage old theories into a new language. We needed to develop models that would solve real-world economic problems. So we used game theory to analyze issues like bargaining and bidding in auctions. We dealt with problems around the “prisoners’ dilemma,” which is about the factors that determine whether people cooperate, and we wrote models on the impact of reputation and even uncertainty about intentions.
What were the key breakthroughs that kicked off the prolific work that followed? What got the ball rolling?
Kreps: Bob began the process with his early models of auctions in which bidders have private information. Paul, in his thesis and with Bob Weber, continued the study of auctions. Other scholars developed the “revelation principle,” in which you reveal your information to a neutral “referee” who sets the rules of the game.
Paul and our colleague [Stanford GSB economics professor] John Roberts wrote a beautiful paper on the topic of entry deterrence: Early models had assumed that those considering entry into an industry would be scared off by what were, essentially, bluffs by incumbent firms that slashed their prices in advance.
Milgrom and Roberts showed the illogic of this argument. Why would anyone care about the price before entry? What you really care about is what prices will be after entry. In their model, they found that new entrants would not be fooled and would enter the market just as often if the existing big player lowered prices. Yet the incumbent was forced to cut its price anyway. Why? Because if the incumbent failed to cut its price, then potential entrants would mistakenly believe that the incumbent was very weak and therefore would enter in droves. Incomplete information put the incumbent on a treadmill, forcing it to run just to stay in the same place. This was a new theory.
Wilson: Then came the Gang of Four (we three plus John Roberts) and the two papers on the “chain-store paradox.” One was by Dave and me, and the other was by Paul and John. These papers showed how a little bit of uncertainty about the motivations of the players, in a situation in which the players interacted many times, could completely upset what a game theoretic analysis implied if there was no uncertainty about the players’ motivations.
Simply explained, suppose you and I repeatedly play the famous Prisoners’ Dilemma Game, a game in which we’re better off if we cooperate, but our individual interests lead us to the non-cooperative outcome. If you believe there is a small chance that I want to cooperate with you, even if it is costly to me, and if we interact repeatedly, you want to see early on if I’m truly a cooperative fellow. And, I want to encourage you to think that I am. So, for most of our interaction, we cooperate. Our papers, by exposing this logic, showed how game theory can help explain phenomena we see in real life. This got the attention of economists, who went on to apply this logic to a variety of topics, for instance, how central bank independence might promote credible monetary policy.
Milgrom: Private information is very important, but what makes it really powerful is the dynamic interaction between parties. If you marry private information with a tiny bit of uncertainty about the motivations of one player, it can be very powerful. All of this evolved into a model for the impact of a reputation.
What is economic engineering?
Kreps: Economic theory works with relatively simple and stylized models. The idea is to generate interesting insights, but in contexts that are simple enough so that you can say for sure what is going on.
Of course, the real world is never that simple. Someone must adapt what is learned from the stylized models and apply them to real-world contexts. That’s the job of “economic engineers.”
Milgrom: For instance, in early real-life spectrum auctions, we discovered that bidders were sending signals to one another using the last few digits of their bids. A bit of $13,001,645 was basically a bid of $13 million, with the $1,645 a signal to other bidders, which might mean “if you stop bidding for license 16, I’ll stop bidding for license 45.” This possibility was not covered by the original stylized models, and it required modifications to the auction design, to prevent this sort of behavior.
Kreps: Bob and his students have been leaders in economic engineering in real-world applications ranging from auctions of the radio spectrum to models for school choice to kidney exchanges. In particular, the Carty Prize recognizes the work of Bob and Paul in designing government auctions of licenses for mobile communications and other uses. These auctions are incredibly complex, with many different bands for many different geographic areas being auctioned simultaneously. They had to deal with all kinds of practical issues that were not foreseen from the outset.
Looking forward, what do you see as the most important opportunities and challenges in economics and economic engineering?
Kreps: One hears a lot about “behavioral economics” and “behavioral finance.” Indeed, the most recent Nobel Prize in Economics was given to Richard Thaler for his pioneering work in behavioral economics. Behavioral economics provides outstanding opportunities to gain a better understanding of how economic motivations and social concerns interact with each other and how individuals and organizations deal with problems that are too complex to solve completely.
It’s challenging because it takes economists out of their comfort zone, but economists have to be ready to be uncomfortable. Good social science should bring us deeper insights about real-world phenomena, and behavioral economics, done right, can accomplish that.
Milgrom: I agree with Dave about behavioral economics, but there are other cases in which computer science becomes key to effective engineering of markets and mechanisms. Here’s one vivid example. When we were designing that auction for old television frequencies, there were so many different elements to consider that the computational challenges were enormous. We used machine learning and artificial intelligence programs, in which the computer itself learned which algorithms would work best in solving the problems we faced.
Wilson: Like Paul, I foresee changes in market organizations due to rapid advances in digital technologies. Some are incremental, as in a stock market where simple limit orders are replaced by computer programs that implement trading algorithms. Others offer opportunities for major advances, as in auctions that allow bids for packages of several maturities of Treasury bonds, or packages of securities and related options and swaps. Market designs must also address newly enabled strategies, such as “front running” large orders to give certain traders an unfair advantage, or “dark pools,” which are private forums for trading securities. These examples from financial markets are indicative of developments in many other contexts affected by pervasive conversion to electronic commerce.