Relevance of dynamic variables in multicategory choice models

Abstract We investigate the relevance of dynamic variables that reflect purchase history a household as independent in multicategory choice models. To this end, we estimate both homogeneous and finite mixture variants multivariate logit model. consider two types variables. Variables first type, which previous publications on models have ignored, are exponentially smoothed category purchases, simply call loyalties. second type log-transformed times since last any category. Our results clearly show adding improves statistical model performance with loyalties being more important than times. The majority coefficients marketing (features, displays, price reductions), pairwise interactions, cross-category relations differ between either including or excluding also measure effect probabilities same (own effects) other categories (cross effects). This exercise demonstrates without tends to overestimate own effects many product categories. positive omitted variable bias provides another explanation for well-known problem “overpromotion” retailing.