Accounting for heterogeneity and nonstationarity in a cross-sectional model of consumer purchase behavior

When calibrating a brand choice model cross-sectionally, a measure of brand loyalty is often introduced into the utility function to account for differences in utility across households and over time. One of the most widely used measures of brand loyalty, proposed by Guadangi and Little, is an exponential smoothing model of past choice behaviour by the household. The paper argues that the exponential smoothing model of brand loyalty cannot properly distinguish between sources of variation in utility due to heterogeneity (across households) and sources of variation due to nonstationarity (within household over time). It introduces a new measure of brand loyalty, derived from a nonstationary Dirichlet-multinomial choice model, in which heterogeneity and nonstationarity are handled distinctly.