Econometric Advances in Diffusion Models Defended on Friday, 2 December 2011
This thesis gives new and important insights in modeling diffusion data in marketing. It addresses modeling multiple series instead of just one series such that one can learn from the differences and similarities across products and countries. Additionally, this thesis addresses the current availability of higher frequency diffusion data. The two issues provide challenges for modeling of diffusion processes.
In this thesis we provide solutions to these challenges, and we also suggest another look at some older issues with a particular focus on parameter estimation. In the first chapters we deal with the estimation of diffusion parameters for a single series. We start with an overview of currently used estimation methods and we suggest that a new method is needed. In fact, our new method does not suffer from bias as it properly incorporates the source of noise and the observational frequency. In the next chapters we focus on modeling high-frequency diffusion data, where we specifically address mixed-frequency diffusion data and seasonality. Finally, we propose a new approach to jointly modeling many diffusion series, where we allow for cross effects between products and countries.
new product diffusion, econometric models, estimation, high-frequency data, mixed-frequency data, seasonality, international diffusion, hierarchical bayes estimation