HLL 2002: Bayesian Statistics in Marketing

A Bayesian approach was taken to address, streamline and standardize the methodologies for solving three specific problems faced by the market research group of HLL. The problems were, 1) estimation of price elasticity of demand by incorporating marketeers' subjective input into the observed data; 2) building forecasting models for sales using price, promotion and macro-economic data; and 3) detection and estimation of a so-called ``price-threshold''. While the first problem involved a straight-forward Bayesian regression, after incorporating the marketeers' beliefs in the prior distribution; the second one involved development of Bayesian co-integration models with Bayesian model averaging suggested as the end-result. The third problem was somewhat non-trivial and a change-point regression model was proposed as a solution, a Bayesian analysis of which, as is well known, yielded the most satisfactory and easily interpretable answers. The ``model'' data sets provided by HLL were analyzed using BUGS, which churned out a lot of interesting and unexpected findings after a thorough Bayesian analysis.