Online Causal Inference for Advertising in Real-Time Bidding Auctions

Real-time bidding (RTB) systems, which leverage auctions to programmatically allocate user impressions to multiple competing advertisers, continue to enjoy widespread success in digital advertising. Assessing the effectiveness of such advertising remains a lingering challenge in research and practice. This paper presents a new experimental design to perform causal inference on advertising bought through such mechanisms. Our method leverages the economic structure of first- and second-price auctions, which are ubiquitous in RTB systems, embedded within a multi-armed bandit (MAB) setup for online adaptive experimentation. We implement it via a modified Thompson sampling (TS) algorithm that estimates causal effects of advertising while minimizing the costs of experimentation to the advertiser by simultaneously learning the optimal bidding policy that maximizes her expected payoffs from auction participation. Simulations show that not only the proposed method successfully accomplishes the advertiser's goals, but also does so at a much lower cost than more conventional experimentation policies aimed at performing causal inference.