USING THE MULTIREGRESSION DYNAMIC-MODEL TO FORECAST BRAND SALES IN A COMPETITIVE PRODUCT MARKET

This paper uses the multiregression dynamic model, designed to model multivariate time series which exhibit a conditional independence structure, to forecast the brand sales in a competitive product market. Following work of Smith and Queen, an undirected graph is derived representing the competition which is believed to exist between the various brands in this market. It is shown how this undirected graph suggests a suitable influence diagram which represents the conditional independence structure between transformed time series of brand sales. It is discussed how the complex multivariate time series of brand sales can then be forecasted by using a multiregression dynamic model, so that each component is simply modelled by a univariate dynamic linear model.