The auction that runs the internet
Love it as an information highway or loathe it as an “electronic asylum of babbling loonies”, you can’t help but be amazed by the internet. And yet, our ability to trawl through billions of webpages, connect to a global community, and carry out a multitude of commercial transactions is driven by a simple economic mechanism—the auction for online ads whose revenue, amounting to over $50 billion per year, fund a significant portion of the infrastructure.
In hindsight, if things had turned out a little differently, all this might never have been.
Before 1998 web advertising consisted of banner advertisements for which an advertiser paid a cost per thousand “impressions”, i.e. the number of times users opened pages on which the ad was displayed. This was an untargeted and inefficient system in which the impact of advertisements was difficult to measure.
Things improved when GoTo.com created the first “pay per click” auction in which an advertiser paid for the number of times users clicked on the ad to access its web page.
Advertisers would select keywords they thought their prospective customers would use and submit bids for each phrase, (for instance, “electric kettle”), indicating the maximum amount they were willing to pay per click. When a user entered a selected keyword into the search box, the advertiser’s ad showed up in the slot determined by the size of its bid relative to other advertisers.
The advertiser who bid the highest won the top slot and paid what it bid. The second highest bidder won the second slot and paid its bid amount, and so forth (higher slots commanded higher clicks). This was called a “first price auction”.
Advertisers now had a medium which they could use to serve targeted ads whose impact in terms of traffic could be measured in real time. But there was a glitch—every advertiser wanted to pay the minimum amount needed to secure its ranking.
Consider a situation in which there were two slots on a page and three advertisers Alpha, Beta, and Charlie with values per click of Rs10, Rs4, and Rs2, respectively. If advertiser Beta bid a little above Rs2, it was guaranteed a slot since Charlie would never go above Rs2. Then Alpha would want to bid just a little above Beta’s bid. Given Alpha’s bid, Beta has room to raise its bid to secure the first slot (as its value is Rs4) so a bidding war ensues.
In order to handle responses and counter responses of competitors, advertisers started making socially inefficient investments into bidding robots and passing on the costs to the search engine.
The use of robots added an interesting twist to the scenario.
Suppose that the robot of Alpha was much faster than the robots of Beta and Charlie. Then Beta knew that Alpha’s robot could always place a counter bid against its bid. Therefore it could never acquire the first position given that it did not have the capacity to go beyond a bid of Rs4. Hence, it was content to outbid Charlie, the lowest value advertiser, with a bid of Rs2.01 while Alpha placed a bid of Rs2.02, giving a low value to the search engine.
The fortunes of search engines and of the internet as a whole were given a fillip by an idea that Nobel prize winning economist, William Vickrey, had come up with in a paper in the Journal Of Finance in 1961.
To address the issue of bidders bidding below their true value in first price auctions, he devised a “second price auction”. In this auction, the winner (the highest bidder) has to pay the amount bid by the second highest bidder, not the amount it has bid.This is true even if it bids less than its value and wins. However, by bidding less than true value it runs the risk of losing an object which it would have liked to win in case the winning bid is less than its true value. Indeed, bidding an amount exactly equal to the true value is the most sensible thing to do in this auction.
In line with the principle of the Vickrey auction, Google’s sponsored search auction introduced in 2002 required that instead of paying its own bid, a winning bidder pays the amount bid by the bidder just below it. In the example given above, suppose that Alpha bids Rs10, Beta bids Rs4, and Charlie bids Rs2. Then Alpha would get the first position in the advertised search returns but pay only Rs4 per click as bid by Beta. Beta would get the second position and pay Rs2 per click as bid by Charlie. And Charlie would not get anything.
This innovation stabilized the advertising market and paved the way for a long spell of growth. Google declared that its “unique auction model uses Nobel Prize-winning economic theory to eliminate... that feeling that you’ve paid too much”.
It was an empty boast. The principle of truth-telling established by Vickrey only operates when there is just one good being sold. This principle does not apply in an auction for spaces for multiple ads.
In the example above, suppose that the number of clicks received by advertisers in the first two positions is almost the same. Now Alpha might be better off bidding lower than its true value in order to end up in second place, where it would have to pay the amount bid by the third highest bidder and receive almost the same number of clicks as it would have got in the first place.
More on this and new complexities in the auctions for AdWords in the next piece. Meanwhile, bid wisely!
Rohit Prasad is a professor at MDI, Gurgaon, and author of Blood Red River. Game Sutra is a fortnightly column based on game theory.