Author(s)
Susan Athey and Denis Nekipelov
Source
Working Paper, 2012
Summary
This technical article develops a mathematical model of search term auctions and presents some implications.
Policy Relevance
Earlier models of auctions may be inadequate for evaluating behavior of search engine advertisement market participants.
Main Points
- Space for sponsored links, displayed in search engine results, is auctioned for each search term in real time. Bidders receive some information about the “quality” of the query from an advertiser’s perspective, but this information is not perfect.
- The authors develop a mathematical model of these auctions; in this model, bidders are not able to create customized bids for each query. This model feature reflects reality, but has been absent from previous models.
- Results from the model contrast with previous understandings of search engine advertising auctions in several ways:
- Bidders in the new model are incentivized to place bids that are lower than what they would be willing to pay for the advertising slow. This is not always the case in auction models.
- Bidder profits are predicted to be lower in the new model.
- The allocation of ad space is economically inefficient, but only very slightly.
- Search engines may be able to increase revenue from an ad auction by providing less-accurate estimates of the quality of a search term. The extent to which this “coarsening” increases revenues can only be determined by long-term experiments.
- In an analysis of search auction data, winning bidders seem to value ads at 40-90% more than they paid.