Dynamic Pricing for Revenue Management in Retailing Using Support Vector Machine, Poisson Regression and Nonlinear Programming

In recent years, dynamic pricing studies which depend on price-based revenue management have increased significantly due to the devolopments in predictive modeling softwares. Accordingly, studies dealing with the prediction of demand functions and price optimizations have also increased. In this research, a new methodology which could be used in retailing is suggested. In this context, support vector machine which depends on statistical learning and poisson regression which deals with count data is used separately in a comparative manner. In the result of comparisons, using the demand functions of the better forecasting model which has the lowest forecasting errors among them, price based revenue functions are generated. After this, in the case of unlimited capacity, taking the derivative of these previously obtained price based revenue functions or alternatively by using unconstrained nonlinear programming, optimal sales prices which maximized the relevant revenue functions are determined. In the case of limited capacity, price based revenue functions are rearranged according to the relation between price and demand and these rearranged revenue functions are proposed to be the objective function of nonlinear programming model given in this study. Adding capacity constraints to the model, similarly, optimal dynamic price policy which maximized revenue function of the retailer are constructed for the limited capacity conditions.