Title: Functional impact into Google AdWords
Authors: Christoph Rust - University of Regensburg (Germany) [presenting]
Abstract: One of the core challenges of the online advertising industry is to explain sales conditional on various attributes of sponsored search advertisements (such as clicks, impressions, ranking, length of ad words). Functional regression with its typically involved dimension reduction techniques seems to provide a very suitable approach. Though, by contrast to classical functional regression problems, we need to focus not only on the functional explanatory variables, but also on specific point-wise information. For instance, time-point specific market events can have important effects that are contrary to the general temporal market evolution. To be able to capture both effects, we adopt a recent method for functional linear regression. Specifically, we analyze a big data sample of an AdWords retailer and propose a functional model for the whole dependence structure, from auction positioning to expected clicks and sales. Furthermore, we interpret findings resulting from the above mentioned method and compare the performance with common models.