In December 2020, nearly forty states filed a lawsuit claiming that Google misled advertisers and publishers by using inside information to manipulate auctions in its own favor. Now, the United States Department of Justice (“DOJ”) is preparing a second major antitrust suit against Google that would focus on the company’s command of the digital advertising market. These lawsuits reflect concerns over how internet platforms use sponsored search auctions (“SSA”) and highlight the importance of seller and bidder information in online advertising auctions.
Many internet platforms and networks generate a significant part of their revenue through the sale of advertising space. Most online platforms organize their space for advertisements in a list form, with different ads competing for user attention. Users engage with ads in the top of the list more often than with ads in lower slots. Therefore, it is potentially more valuable for an advertiser to place her ad in a slot in the top of the list, because the ad will receive more clicks from platform users. In platforms such as Google, Tripadvisor, and Yahoo, the advertisement slots are allocated with the help of an auction mechanism (an SSA). In such an SSA, the advertisers become bidders and submit bids that reflect their valuation of advertisement. A common form of these SSAs is for advertisers to submit a single bid which is then used to determine which advertisers are allocated to which slots and the prices paid when an advertisement in a particular slot receives a click.
The economic literature studying SSAs typically assumes that the click through rate (“CTR”) of each slot and the probability that a click will convert to a sale (“conversion rate”) is known to all advertisers bidding for a slot. However, this assumption may not be realistic. For example, on Tripadvisor, the CTR and conversion rate for the top advertising slot on a hotel listing page can vary quite dramatically over time. This variation is in part a result of how many viewers come to the page by clicking on Tripadvisor’s paid search advertising compared to how many viewers come to the page as Tripadvisor’s own members. Viewers who come to a hotel listing page through clicking on Tripadvisor’s paid search advertising tend to have quite different click and conversion patterns compared to Tripadvisor’s own members, who arrive at the hotel listing page by performing searches directly on Tripadvisor’s platform.
Anecdotal evidence suggests that different advertisers, such as online travel agencies, may differ in their ability to predict the CTR for the top slot and may believe, systematically, that the average number of clicks for a given slot on a hotel listing page is higher or lower than it really is. This raises the question of whether the platform could increase its revenues by providing advertisers who are bidding for slots with accurate information about the CTR, and, if so, by how much.
I investigated this question by estimating a structural model of bidding behavior for Tripadvisor SSAs. In my model, I allow advertisers to have asymmetric information about the CTR of the top slot in the auction. The auction operates as a General- ized Second Price Auction (“GSPA”). Before the auction takes place, each advertiser receives a signal about the top slot’s CTR, which she updates, based on what happens as the auction plays out. I also allow advertisers to have different prior beliefs about the CTR in order to capture their biases, if any. A prior belief centered close to the true number of clicks of the top slot reflects an advertiser with the ability to predict the CTR well. Respectively, a prior belief centered further away from the true number of clicks of the top slot shows an advertiser that often fails to predict the platform’s CTR.
I use data from Tripadvisor to address whether a platform such as Tripadvisor could increase its revenues by providing more information to advertisers who are bidding for slots. The Tripadvisor data include information on the bids, auction results, user clicks, and conversion data for more than 150,000 auctions. I also observe the margin that each bidder extracts from each hotel when a room is booked through its advertised slot. Estimates from the Tripadvisor data indicate that advertisers are biased in their beliefs about the expected number of clicks that their listing will receive.
I also consider a counterfactual analysis in which Tripadvisor reveals its information about the CTR. I find that revealing the CTR for each auction erases bidder bias. This results in higher revenues for two main reasons. First, advertisers who had a down- ward bias bid higher when they realize that the expected number of clicks in an auction exceed their expectations. Second, advertisers who had an upward bias no longer interpret low bids from their pessimistic counterparts as an indication of a low CTR and thus increase their bids.
If there is asymmetric information and biased beliefs about CTRs in the SSA, platforms may be able to increase their revenues by revealing information about the CTR. An analysis of Tripadvisor SSA data indicates that revealing the CTR would raise platform revenues by an average of seven percent. This analysis also indicates that revealing CTR-related information benefits advertisers with biased beliefs. Thus, these findings suggest that increasing the transparency of CTRs may benefit both advertisers who are bidding for slots and the platforms, such as Tripadvisor, on which they are bidding for these slots.