Solving A Pricing-newsvendor Problem Driven By Big Data

According to the traditional stochastic inventory control theory,customer demand is assumed to be subject to a distribution family with unknown parameters,and the decision maker would use statistics theory to estimate these unknown parameters.Nowadays the uncertainty of customer demand has greatly increased,so these traditional methods are facing huge challenges.At the same time,companies and merchants can also collect more purchasing-relevant data about customers,so as to carry out product pricing and inventory management more efficiently.In the existing research literature on the data-driven newsvendor problem,most studies only consider the total customer demand in each period.In recent years,some researchers have introduced environmental feature data into the newsvendor problem.This paper further adds the customer's individual-level data to this problem.From a data-driven perspective,using machine learning related theories,an optimization model is established to solve the pricing-newsvendor problem.Assuming that for each store visit of a potential customer,the decision maker can obtain the environmental features(such as seasons,weather,market prices,etc.)and individual features(such as age,gender,price sensitivity Degrees,etc.)related to the purchasing decision.If one customer purchased the product,it will be marked as a positive sample,otherwise it is marked as a negative sample.Therefore,we assume that all purchase decisions made by customers can be characterized by a target concept,i.e.,a mapping from the Cartesian product of the environmental and personal feature domain to the Boolean domain.We then develop a model that integrates learning customer arrival and purchasing patterns,forecasting the aggregated demand based on the learned knowledge,and solving the pricing-newsvendor problem.Our solution adapts to the evolving environment and learns how to act effectively within it.Given the accuracy and confidence levels,we also show the sample complexity required to learn an approximately correct concept under the framework of probably approximately correct learning.Based on our model,we generate some sample data by assuming the customer's arrival and purchasing patterns in each period,as well as the different values and distribution of customer characteristics visiting the store.Based on these data,we show the result of the algorithm.The decision maker can learn the customer's arrival and purchasing patterns,estimate the demand distribution,and solve the pricing-newsvendor problem.We also calculated the sample complexity required for learning under the PAC(probably approximately correct)framework based on the set customer characteristics.At the same time,this paper also assumes three distributions of customers' price sensitivity,and gives corresponding pricing strategy recommendations for each distribution.