Data-driven Ordering Strategy Of Retailers Based On CVaR Criterion

This thesis mainly studies the order decision problem of risk-averse retailers.In production and operation,retailers tend to avoid risks and seek to minimize the risk of loss.Under this setting,when the demand distribution is known,there has been a lot of related research.However,when the market demand distribution function is unknown,and the related costs in the ordering are related to the market demand distribution,it is difficult to follow the traditional ideas which fit the demand distribution by observed data,and then determine the optimal ordering strategy.This brings great challenges to practice.In order to solve the above problems,this paper introduces the idea of "data-driven" and the concept of "learning-optimizing",which use data to take the place of traditional assumptions,and effectively integrate the two independent stages of learning and optimization.So an order strategy was designed in this paper to solve the risk-averse retailer when the demand distribution function is unknown.Through this strategy,retailers can quickly respond to data updates and dynamically adjust ordering decisions.At the same time,this study enriches the application scope of ordering decision model and promotes the widespread use of several theoretical models of management science in production practice.This thesis uses the CVa R criterion in finance to measure the risks faced by retailers,and studies the data-driven order decision algorithm for risk-averse retailers.This thesis assumes that the market demand distribution function is unknown,with the help of SAA algorithm and KM algorithm,we gives two policies for solving the optimal order quantity of risk-averse retailers under the unknown demand distribution function.Among them,the SAA algorithm is mainly optimized using the retailer's historical demand data,and the KM algorithm is mainly optimized using the retailer's historical sales data.This is due to the censoring effect of demand data,which cause certain incompleteness.At the same time,this paper also considers the problem of algorithm design in the case of unknown inventory parameters,and introduces the idea of "learning-optimizing" to give an ordering algorithm when the demand distribution function and the unit out-of-stock cost are unknown.Subsequently,this paper demonstrates the convergence of the algorithm through mathematical derivation,and verifies the effectiveness of the algorithm from the numerical experiments.