Mathematical model for buying behavior of international travelers : a multinomial logistic regression approach

Abstract

Predicting the future sales is required continuous attention in order to fulfill the consumer requirements. Disaggregating the sales in to micro level of the business would increase the prediction accuracy since the ground level requirements, trends and patterns are captured. Multinomial logistic regression is a technique that is to be used to model the outcomes with categorical response variable with more than two levels. In the study, the significant determinants for the brand wise purchasing decision of chocolates in a travel retail chain and its consequences are investigated. Multinomial logistics regression found that nationality of the consumers, time of purchase, preference for promotions and preference for weight of the products have significant impact on the chocolate brand choice. It was also found that these fours variables have no multicollinearity effect. The pseudo R2 value of the model confirms that only 44.9% of the variability is absorbed by the final model. The model has overall brand classification accuracy of 52.4. The buying preference for any brand of cholate is maximized during the 1st quarters. Mix and match promotion maximizes the preference for purchase of Mars and Mondelez brands while buy 3 get 1 free become the promotion that maximizes the buying preference for Nestle. Preference for weight category is variant for the 3 brands. The relative nationality wise probabilities of selecting a brand of chocolate for fixed levels of promotional preferences and product weight preferences are derived with the multinomial logistic transformation equations. When comparing the nationality wise brand selection probabilities, no significant changes to the probabilities were found according to the nationalities of the customers. It is recommending to carry out similar studies for other sales as well.