Estimation of Reserve Price for Sponsored Search Auctions

Jun 17, 2009

An important objective of Yahoo Labs Sponsored Search is to implement a balanced marketplace where both publishers and advertisers have incentives to participate in the marketplace. Typically, advertisers have greater control over their participation in the marketplace - they can determine which keywords to bid on, how much to bid and how much to spend in aggregate. They can also block traffic from certain publishers sources. Market Reserve Price (MRP) is a product that benefits publishers by providing price support in thin markets. It also benefits users by improving the relevance of results by filtering out ads with lower values from the marketplace.

The calculation of reserve prices is based on auction theory. Yahoo economists have played a key role in developing the theory of Generalized Second Price (GSP) auctions, which is the formal name for sponsored search auctions. However, application of the theory is not so straightforward. One key issue is that the value distribution of advertisers is unknown. Another is that the sponsored search auction marketplace is quite dynamic - advertisers enter and leave the marketplace, change their bids, etc. Yahoo scientists have developed methods to aggregate bid data and to estimate advertiser value distributions, and have adapted the reserve price calculation methodology pioneered by Nobel prize-winning economist Roger Myerson to sponsored search auctions. A third issue is that tail keywords do not have sufficient data. For these keywords, Yahoo Labs scientists are exploring a statistical test methodology to collect data to train a machine learned model to estimate reserve prices.

And in support of this innovation, research engineers in Yahoo Labs have developed innovative solutions to enable fast aggregation and calculation on 100's of millions of search history records to set reserve prices for 10's of millions of keywords and 100's of millions of ads.