The transition of the largest online advertising platforms to auction-based automated real-time designs has transformed the advertising industry. The advertisers had an opportunity to design flexible goal-specific advertising campaigns, target focused small groups of consumers, and perform fast and efficient experimentations. At the same time, consumers can be exposed to a smaller number of higher quality advertisements. However, in order to achieve the design goals, such as optimal placement of ads that maximizes the sum of utilities of users or revenue generated by the auction platform, the auction needs to be carefully optimized. To optimize the auction, the auctioneer typically chooses a relatively small number of auction parameters that determine the allocation of items and prices based on submitted bids. Typical auction parameters include reserve prices, which determine the minimum bid and can correspond to any allocation. Another parameter used in sponsored search auctions is the quality score, which is a positive bidder-specific weight used to discount or inflate submitted bids before they are ranked. The setting of auction parameters requires a knowledge of advertisers' preferences and consumers behavior, which can be acquired from data. This makes econometric inference from observed data of high importance for the design and analysis of online advertising auctions.
The structure of online advertising exchanges is becoming significantly complex, and requires multiple parameters to be input by auction designers. These parameters are required to yield consistent advertising allocations, relevant user ad experiences over time and catering ad placements to advertisers' goals. This operational structure of online ad exchanges has been incorporated into the algorithmic implementation of advertising auctions, see Muthukrishnan,29 Aggarwal and Muthukrishnan,30 and Muthukrishnan31 for details of such an implementation. However, the increased heterogeneity and dynamics of the marketplace call for quick "on demand" adjustments of auction parameters in order to pursue auction platform revenue goals and relevant ad experiences for users. Recent evidence from the sponsored search advertising marketplace indicates that there are significant gains (both in terms of overall welfare and ad exchange revenue) from adopting data-driven designs. For example, Ostrovsky and Schwarz34 described the experiment on Yahoo!'s search engine where the advertisers `per-click values were recovered from observed bids. The distribution of recovered values was then used to set search query-specific reserve prices in the advertising auctions. The data shows a significant increase in the search engine's revenue resulting from this switch.
No entries found