In this paper, they discuss some of the history behind GSP as well as do some experimentation and economic theory on the subject.

They show some of the differences between Internet advertising and most auction markets. Bidding takes place continuously, meaning there is no obvious set begin and end time and bidders can even change their bids at any time to put themselves at an advantage. Second, the paper claims that advertising space is a somewhat perishable item. If no advertisements are there, those possible clicks are lost forever. Third, the definition of what is a “unit” of what is being sold, is totally undefined. Advertisers would rather not pay unless a purchase is made through that advertisement. That is the unit they see as a result. However, the search engines would think that every time they display the advertisement is a reasonable unit. This last point has been currently narrowed down because for the most popular search engines, they compromise and have every click equal one “unit” of advertising.

Overture originally had a first price auction for advertising slots, but that was incredibly unstable due to bidding being continuous and everyone knowing each others bids. Google saw this problem and used instead the GSP, and Overture quickly followed. The GSP is especially interesting because it has “emerged in the wild”. It was not really used and therefore well-studied in any way and was basically developed for the market in which it was thrown into use.

The paper discusses a concept they call “envy-free equilibrium”. This means it is an equilibrium as well as no one wants to switch bids with any other bidder. This is an interesting concept to me, remembering what we discussed in class how their can be equilibriums which are out of order and have bidders stuck in a situation where they have way less payoff than they should. Through experimentation and through some theorems on how GSP might work, the paper finds that GSP would result in VCG payoffs for the advertisers. This is also the same as the worst possible envy-free equilibrium for the search engine interestingly enough.

Paper:

http://rwj.berkeley.edu/schwarz/publications/gsp051003.pdf