Stochastic Models for Multivariate Time Series of Counts, with a Marketing Application

• Activity Number: 317
• Type: Invited
• Date/Time: Tuesday, August 2, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Business and Economic Statistics Section
• Author(s): Nalini Ravishanker*+ and Rajkumar Venkatesan and Shan Hu
• Companies: University of Connecticut and University of Virginia and University of Connecticut
• Keywords: Attitudes ; Bayesian modeling ; Discrete-valued time series ; Pharmaceutical marketing data

Abstract

In several applications, there is an increasing need for accurate modeling of multivariate time series of counts for several subjects as functions of relevant covariates (subject-specific and time-varying), incorporating dependence over time and dependence between the components of the response vector. This talk describes a hierarchical dynamic non-linear model framework with a marketing application, using data from a multinational pharmaceutical firm. We model multivariate count data responses on the number of monthly prescriptions made by physicians for the focal drug from this firm, and for drugs from its competitors. We incorporate into the analysis customer attitudes from a survey obtained at irregular times, and discuss imputation of missing attitudes and customer lifetime value computation.