Price cycles in online advertising auctions

Paid placement in search engines has become one of the most successful and rapidly growing sectors of the online advertising industry. The innovative use of auctions to sell keyword-related advertisement positions is perhaps the most important factor driving the success of this market. There has been no systematic analysis, however, of the advertisers' strategies to bid for ranks in a dynamic environment, where each bidder's bid can be updated and observed by the competitors in real time. We capture this dynamic setting using a Markov process and identify the Markov perfect equilibrium. We find that in such a dynamic environment, bidders' bidding strategies follow a cyclical pattern (Edgeworth cycle) similar to that conjectured by Edgeworth (1925) in a totally different context. A new data set that contains a detailed bidding history of all advertisers for sample keywords in a leading search engine makes it possible for us to study the real-world behavior of bidders. We propose an empirical framework based on maximum likelihood estimation of latent Markov state switching to confirm the theory. We also discuss the theoretical and practical significance of finding such cycles in an online market place.