Bayesian state space modeling approach for measuring the effectiveness of marketing activities and baseline sales from POS data

Analysis of Point of Sales (POS) data is an important research area of marketing science and knowledge discovery, which may enable marketing managers to attain the effective marketing activities. To measure the effectiveness of marketing activities and baseline sales, we develop the multivariate time series modeling method in the framework of a general state space model. A multivariate Poisson model and a multivariate correlated auto-regressive model are used for a system model and an observation model. The Bayesian approach via Markov Chain Monte Carlo (MCMC) algorithm is employed for estimating model parameters. To evaluate the goodness of the estimated models, the Bayesian predictive information criterion is utilized. The proposed model is evaluated with its application to actual POS data.