Longitudinal models for dynamic segmentation in financial markets

Purpose - Dynamic market segmentation is a very important topic in many businesses where it is interesting to gain knowledge on the reference market and on its evolution over time. Various papers in the reference literature are devoted to the topic and different statistical models are proposed. The purpose of this paper is to compare two statistical approaches to model categorical longitudinal data to perform dynamic market segmentation. Design/methodology/approach - The latent class Markov model identifies a latent variable whose states represent market segments at an initial point in time, customers can switch to one segment to another between consecutive measurement occasions and a regression structure models the effects of covariates, describing customers' characteristics, on segments belonging and transition probabilities. The latent class growth approach models individual trajectories, describing a behaviour over time. Customers' characteristics may be inserted in the model to affect trajectories that may vary across latent groups, in the author's case, market segments. Findings - The two approaches revealed both suitable for dynamic market segmentation. The advice to marketer analysts is to explore both solutions to dynamically segment the reference market. The best approach will be then judged in terms of fit, substantial results and assumptions on the reference market. Originality/value - The proposed statistical models are new in the field of financial markets.