Takuya Homma's Library tagged hal →  View Popular

A: I’m a big fan of experimental economics. In fact, we’re using it at Google to study various sorts of market designs. The goal in economic experiments, as in all experiments, is to make the results reproducible.

A: Yes, I think so. One problem is that academic programs tend to be oriented toward producing more academics, and they don’t pay as much attention to non-academic jobs as they should. Business schools are somewhat better than economics departments in this regard.

Q: Your job sounds extremely interesting. What jobs would you recommend to a young person with an interest, and maybe a bachelors degree, in economics?

A: If you are looking for a career where your services will be in high demand, you should find something where you provide a scarce, complementary service to something that is getting ubiquitous and cheap. So what’s getting ubiquitous and cheap? Data. And what is complementary to data? Analysis. So my recommendation is to take lots of courses about how to manipulate and analyze data: databases, machine learning, econometrics, statistics, visualization, and so on.

Q: You once said that “marketing is the new finance,” because of the data and tools that are now available. In your opinion, what are the three most useful quantitative techniques for analyzing this kind of data? To what extent does Google use this type of analysis?

A: The three most important techniques, in my opinion, are: 1) experimental design and analysis; 2) regression; and 3) Bayesian methods. We use the first two extensively, and are beginning to use Bayesian techniques more.

Q: Do you feel that more traditional economic models do an adequate job of representing what economists would like them to? Do you think that emerging fields of study, like behavioral economics, will become increasingly important and/or potentially integrated into the core of economic focus? Given the assumptions we make in economics with respect to human rationality, optimal decision making, and large-scale generalizations, how can anyone in the field feel confident that a theoretical conclusion/result/finding is truly significant?

A: I think that the “traditional model” is a good starting point for economic analysis, but I don’t think that it is necessarily the ending point. You have to explore different alternatives to see which model seems to explain the data best. However, I will say that I think that formulating a model mathematically is quite important, as it serves as a check for internal consistency. Furthermore, a mathematical model allows you to draw out the consequences of an assumption that often allows for better understanding.

Back in 2001 when I stepped down as dean at Berkeley, I thought I would take a leave for a year and do something in the real world. But then the real world seemed a little scary so I went to Google instead. There were only 400 people at Google back in 2002 [there are now nearly 14,000] and I had a great time working with them on a number of projects, primarily involving quantitative analysis of one sort or another. For example, I’ve studied the Google ad auction quite a bit and that turned out to be very interesting from an economic point of view.

After a 1 1/2 year leave I returned to Berkeley to teach on a part time basis, but I’ve continued to consult with Google for a few days a week. As the company has grown, the demands on my time have grown, and I finally decided to devote my full attention to Google. Everything you hear about Google is true: it’s a very exciting place to work. So after studying the tribes of Silicon Valley for several years, I’ve finally gone native.

WSJ: What does the job entail?

Varian: During my time at Google we have built up a world-class group of quantitative analysts, and the economics team will complement these existing resources. Google has a great infrastructure for data analysis, and a management team that is very receptive to quantitative methods and willing to invest in this area.

In addition to working on analytics, I’ve also worked on various business strategy and public policy issues, and will continue to do so as the occasion arises. This set of issues will only get more important to Google as time goes on, so I expect that this will also involve a fair amount of my time.

WSJ: In the past, promising new economics PhDs who didn’t want to work in government or academia probably aspired to work on Wall Street. In the future, will they aspire to work at companies like Google?

Varian: I think marketing is the new finance. In the 1960s and 1970s [we] got interesting data, and a lot of analytic fire power focused on that data; Bob Merton and Fischer Black, the whole team of people that developed modern finance. So we saw huge gains in understanding performance in the finance industry. I think marketing is in the same place: now we’re getting a lot of really good data, we have tools, we have methods, we have smart people working on it. So my view is the quants are going to move from Wall Street to Madison Avenue.

WSJ: Is that because the Web can now provide companies with the density of information once available only for things like stock prices and interest rates?

Varian: Absolutely. Adaptive forecasting, how I revise my forecast to take account of updated information, you use that a lot on Wall Street, where you have time series of stock prices. And some of those things carry over into things that Google is doing, that have this real-time flow of data. How do I detect unusual events, and react to them?