THE IMPORTANCE OF COMPETITION
It is widely recognized that revenue realized in an auction depends critically on how much competition there is in that auction. Klemperer (2002) describes the case of the June 2000 auctions for mobile phone licenses in the Netherlands where there were five licenses and six bidders.
One bidder threatened another with legal action if it continued bidding, inducing it to drop out, thus leaving only five bidders for five licenses - not much competition! In fact the auction raised less than 30 percent of what the Dutch government had forecast.The same principle holds true for the position auction: revenue doesn’t really take off until there is competition. In the Google auction there are eight slots for ads on the right-hand side of the page and up to three slots on the top of the page. As mentioned earlier, the ordering of ads is determined by bids and click-through rates, but the ads that are ‘promoted’ (moved to the top of the page) have to satisfy some additional criteria involving ad quality.
To simplify a bit, if an auction has fewer bidders than available slots, or just enough bidders to fill the available slots, we say it is ‘undersold’. If it has more bidders than slots we say it is ‘oversold’. If an auction is undersold, the price paid by the last bidder on the page is the reservation price, which we will take to be five cents.8 If the page is oversold, the price paid by the last bidder on the page is determined by the bid of the first excluded agent, which can easily be at least ten times higher than the reserve price.
Consider a simple example where all bidders have the same value r and the reserve price is r. Let ps be the price paid for slot s and let xs be the number of clicks that slot s receives. If the page is undersold, each bidder has to be indifferent between paying ps and receiving xs clicks versus paying r and receiving xm clicks, where m is the last ad shown on the page.
This implies:
Equation (18.2) says that the expenditure on slot s has to be the expenditure on the last slot plus the incremental value of the clicks in position s.
On the other hand, suppose the page is oversold so that there is at least one excluded bidder with value r. Then each bidder has to be indifferent between what it is paying and
the profit from being excluded - which is zero. This gives us (v - ps)xs = 0, which implies Ps = v.
Note the big leap in revenue in going from a partially sold page to an oversold page. In the first case, everybody is indifferent between being in the slot they are in and being in the worst slot. In the second case, everybody is indifferent between being shown and not shown at all, which means prices are competed up to the equal value. To drive this point home, consider a simple example.
Suppose that there are two slots. The top one gets 100 clicks per day, the second one 80 clicks per day. There are two advertisers, each of whom values a click at 50 cents. In this model, one advertiser occupies slot two and gets 80 clicks per day, for which he or she pays five cents per click = $4.00 in total spend. The second advertiser occupies the top slot getting 20 additional clicks per day. Competition forces him or her to pay $10 more for those clicks than the advertiser in slot two. Thus he or she spends $14 = $4 + $10 in total. Total revenue from the two advertisers is $18.
Now suppose there are three advertisers who value clicks at 50 cents each, but there are still only two slots. The equilibrium bid is now 50 cents per click, there are 180 clicks in total, so the total revenue from the two advertisers is $90. The addition of one more advertiser increases revenue from $18 to $90!
This example illustrates the important point that oversold pages are far more profitable than partially sold pages not just because there are more bidders, but also because the forces of competition are much stronger.
It is this insight that explains the rather remarkable 100 percent ROI described in the previous section. Most auctions only have a few participants, so the reserve price is important in determining revenue. However, the reserve price tends to be set quite low. Hence advertisers in these auctions end up with a tremendous deal. On the other hand, some auctions are highly competitive. That competition translates into higher prices and hence more revenue. In net, search engine advertising turns out to be exceptionally attractive on average.This point also illustrates the importance of the matching algorithm used for displaying ads. The user enters a ‘query’ and the advertiser buys ‘keywords’. The advertiser can specify ‘exact match’, which means that the ad is only shown if the user’s query exactly matches the advertiser’s keywords. But it is more common for advertisers to specify ‘broad match’, which means that the query will match various expansions of the keyword such as synonyms and substrings. The additional ads due to broad match benefit the user and the advertiser, since they make it more likely that the user will click. But they also increase the competition in the auction, raising prices.
18.8