An Interview with Glenn Ellison

By TAP Staff Blogger

Posted on February 4, 2010


Jacques Cremer of the Toulouse Network for Information Technology (TNIT) interviewed Glenn Ellison this past October. They discussed Professor Ellison's thoughts about the consequences of the Internet for the efficiency of market places, as well as his recent works on a search cost model of obfuscation and position auctions. Glenn Ellison is the Gregory K. Palm Professor of Economics and Associate Department Head with the Massachusetts Institute of Technology.

TNIT: Your latest paper in Econometrica, co-written with Sara Ellison, deals with electronic market places. Yet, we remember you saying, when you edited Econometrica, that you were surprised at how many papers you received on auctions and how few on Walmart, given their relative importance for the economy. Do we spot a contradiction there?

 Ellison: Wow, that’s a tough question. I have to say I’d assumed this would be a puff piece … But no, I don’t think I’m hypocritical on this. … I do still think Walmart deserves more attention than it gets in our profession. It’s remarkable to have two million employees in a single company. … But e-commerce is really important too. Online sales are already about one-third of Walmart’s sales. Given how big Walmart is, this makes e-commerce an important part of the economy too. And just counting online revenue vastly understates the impact of the Internet. You see statistics saying things like 95% of car purchasers do some research online even though the ultimate purchase is almost always offline.

Within e-commerce, the particular paper you mention is studying a small and quite atypical online marketplace. But I won’t apologize for this. Many of the most important studies in empirical IO have examined atypical environments and I think this is not a coincidence. In my view, industrial organization is an inherently theoretical field. Many questions we need to address are often one-time questions. We can’t be like labor or development economists and compare what happened in the years following the previous 100 times when two satellite radio companies were or were not allowed to merge. We need to have a set of theoretical models we can apply. Empirical work should play an important role in helping us improve our modeling. Which of the many effects theorists have identified are important and which are not? Are there other more important factors that we should be modeling? To gain such insights, it’s often a good idea to study atypical environments. There are two reasons for this. First, many real-world environments are very complex. It’s easier to do a thorough study if you pick an atypically simple environment. The retailers we study, for example, essentially just buy big boxes of computer parts from a wholesaler, put them in individual-sized cardboard boxes, and mail them. This makes it atypically easy to correctly account for costs. Second, if you’re trying to learn about some factor, it’s advantageous to find an atypical environment where that factor is unusually important. In our case, we were trying to learn about search costs, so it was useful to study an environment where there was hardly any product differentiation. This makes demand patterns and markups as affected as possible by search costs.

TNIT: Actually, we have really liked the paper, but one interpretation would be that the same type of trickery that we find in traditional markets can be found in electronic markets - we would have expected them to be more transparent. What is your overall feeling about the consequences of the Internet for the efficiency of market places?


Ellison: I certainly think the Internet is making markets more efficient. I remember what it was like when you had to call each airline separately on the phone or go out to the travel agent’s office and hope there wasn’t a long line.

But the point of our article was to argue that the Internet probably won’t make markets as transparent as people had hoped. The Internet is a great consumer search tool. But it’s a multipurpose tool and firms will also harness it to make search more difficult. We call this obfuscation. In some cases the Internet makes obfuscation a lot easier – it used to be cost effective for firms to employ a used-car salesman if they were selling something really expensive like a car, whereas now you just pay a fixed cost to design a website that runs consumers through a long sales pitch and the server will deliver it at almost no cost. In other cases, the Internet isn’t directly facilitating the obfuscation, but is still relevant because of the role it’s played in reshaping the market. For example, I think the ease of online price search is one factor that’s leading airlines to adopt extra fees for snacks, checked luggage, etc.

TNIT: Many online markets use reputation systems of buyers and sellers. What do you think about these? What are their problems and how could they be improved? Do you think that they could have attenuated the “obfuscation” you describe?

Ellison: I think online reputation systems are useful, but it’s difficult to get them to work well. Many sites with consumer comments don’t get large volumes of comments. It’s hard to know whether the 0.1% of people who left comments had experiences that were at all representative and they’re also prone to abuse by merchants who can leave fake feedback on themselves (or rivals). Systems that generate more feedback more frequently might seem better, but can be really cluttered in a world where disreputable merchants still deliver 95% of the time. In two-sided settings like eBay it’s also hard to get people to give accurate negative feedback because there’s the fear of retaliation.

In the markets we studied consumers could have looked up ratings separately on independent ratings sites, but I think few did. I don’t know that it would have made much difference if the ratings had been easier to access. In theory, yes consumers probably would have been more likely to click on a firm’s link if there was feedback saying “This firm is honest, has an easy-to-use website, and doesn’t spend a lot of time making you sit through automated sales pitches pushing add-ons.” But if a retailer actually behaved this well, then it probably couldn’t have made a profit at the prices you saw on Pricewatch. So in practice, there wouldn’t have been any firms with that feedback and things probably would have worked out similarly.

TNIT: In collaboration with Susan Athey, you also have done work recently on position auctions. What are the big issues there?


Ellison: The sale of “sponsored-link” advertising on search engines is becoming very important. It’s made Google a $150 billion dollar company and continues to grow as other forms of advertising are suffering. At the broadest level the goal of our work is to develop a complete model of sponsored-link advertising – we’d like to understand how the business works, to see what economics has to say about how search engines should want to design their sponsored-link auctions, and to see what insights we can get on how the incentives of search engines are and are not aligned with what is best for consumers and advertisers.

Our work in this area came after we saw previous papers on the topic which had developed an elegant analysis of the “position-auction” sales mechanism. The primary thing that we wanted to add to this literature was to incorporate that the sponsored-link positions are not just generic “objects” that are being auctioned. They’re advertisements and being a sponsored link only has value if consumers believe that sponsored links are sufficiently likely to be of interest to make clicking on them worthwhile.

The most basic conclusion of our paper is that the reason why this business works so well is that the winners of a well-designed position auction are generally the advertisers who have the most to offer consumers, which makes consumers want to click on the links. Our auction analysis is mostly showing that the elegant analysis of the previous papers carries over fairly directly to our more complete model. The fun part then comes in when we talk about auction design. There are a number of new effects in our model that make optimal auction designs very different from what they’d be if you were auctioning generic “objects”. But the basic principles behind auction design in our model turn out to be pretty easy to understand and we can give lots of advice. 

TNIT: Some personal and fun questions
When buying electronics for yourself. What do you rely on: online pricecrawler, your usual electronics dealer or advice from friends and family?

Ellison: I read online reviews to try to figure out what I want. What happens next depends on who’s doing the buying. If it’s my wife, she pretty much always buys from Amazon. I may do this too, but am also open to buying from firms, even ones I’ve never heard of, to get a better price.

TNIT: What is the most expensive thing that you have bought using an online trading platform? Textbook, laptop or car? (Dog?)

Ellison: Not quite sure what “online trading platform” means. If you mean something like eBay, it would be my foosball table. Our youngest daughter was fascinated when the big truck carrying it pulled up outside our house. If you include online purchases from any retailer, it would be a handmade oriental rug my wife bought a couple of years ago. 

I also found our house online – we were living in Princeton at the time – and did all the negotiating to buy my car via e-mail to buy from a dealer I’d never visited, but these were not pure online sales. My wife and I obviously went to look at the house in person before making an offer and I picked up the car and filled out paperwork at the dealership after we’d agreed on a price.

TNIT: Just quickly respond to the following opposites.
Getting rid of unwanted presents: ebay or yard sale?

Ellison: Neither. Our house has a big basement and it's full of stuff that we’ve meant to get rid of one way or the other but have never gotten around to selling.

TNIT: Facebook, LinkedIn or address book?


Ellison: I have a Facebook page, but don’t use it much. Not on Linked In. So I guess address book.

TNIT: Touch Type or Secretary?

Ellison: Touch type.

TNIT: JSTOR or paper copies in library?

Ellison: I love browsing through journals. But I rarely have time for browsing these days. So it’s almost always JSTOR.

TNIT: GPS or map?

Ellison: I think I don’t actually have either one in my car right now. I’m pretty good at knowing my way around or figuring out which direction I must need to go to get somewhere.

TNIT:  coffee or mineral water?

Ellison: Don’t drink coffee, so mineral water. I have been known to have a Coke Zero in the morning though.

TNIT: Twitter or not?

Ellison: Not

TNIT: (La)TeX or Scientific Word?

Ellison: Latex. I hate the way Scientific Workplace rewrites Latex code and try to forbid my coauthors from using it.