Statistical plant modeling and simulation in online advertising

We propose a plant modeling and simulation methodology for ad exchanges that supports systematic design of feedback controllers, and that enables analysis and simulation-based evaluation of candidate feedback controllers. The methodology is a bottom-up technique that considers the market clearing mechanism and is based on statistical modeling of performance, performance estimates, competitive bids, and impression supply volume on different inventory segments. The modeling approach is developed in the context of cost control, but can easily be extended beyond that particular objective.