Demand estimation considering inventory availability under community group buying
With the development of e-commerce,new retail business models have recently emerged in China.For example,community group buying is a novel e-commerce retail model that has developed rapidly in China in recent years.However,as community group buying remains in its infancy,there are currently no methods for demand estimation in community group buying.Demand estimation supports enterprises'sales planning,which is essential for enterprise survival and development.Therefore,it is critical for community group-buying platforms to predict accurately how customers react to the products.The key to accurate demand estimation is to develop a customer choice model that captures customer purchasing behavior and evaluates the spillover effect between substitute products.The traditional customer-choice model based on a multinomial logit(MNL)model can accurately explain the impact of sales availability.However,it cannot describe the impact of inventory availability on demand estimation,which is a specific influencing factor in the context of community group buying.Based on the above discussion,this paper explores(1)how to build a customer choice model that considers inventory availability;and(2)how to design an efficient algorithm to achieve demand estimation.To this end,we first review studies on customer choice behavior modeling and demand estimation;we take the classic MNL model as the benchmark for the analysis of traditional customer choice behavior.Subsequently,we analyze customer choice behavior in the context of community group buying.Customer choice behavior in this context is affected by both sales availability and inventory availability,so it cannot be examined using the traditional customer choice model.Therefore,we build an improved customer choice model that considers inventory availability and has the following two key advantages.First,as real-life sales data are not always consistent with the assumption that no product is out of stock during a sales period,the improved customer-choice model relaxes this assumption and thus has strong practicability.Second,the improved customer-choice model considers inventory availability,and it retains the properties of the MNL model,which is convenient for building an estimation model and designing a solution method.Like the traditional customer choice model,the improved customer-choice model faces the problem of incomplete data,as the product information and the corresponding actual sales are the only available data.Therefore,we cannot estimate the demand parameters directly,which is a common limitation in demand estimation.We solve this problem by building a maximum likelihood function of incomplete data based on the improved customer-choice model.Moreover,we introduce a market share parameter to constrain the range of optimal solutions of the maximum likelihood function.This method solves the problem of multiple optimal solutions,but solving the maximum likelihood function is nevertheless a complex optimization problem,highlighting the need for a more efficient algorithm than is currently available.Accordingly,we design a novel two-stage algorithm and describe the pseudo-code of this algorithm.The algorithm operates via the following two-stage process.In the first stage,a splitting algorithm decomposes the sales data and obtains the optimal segmentation scheme of the sales data.The algorithm divides the sales data of the same sales period into the sales data of the different sales periods to ensure that the product is completely available or not sold in each different sales period.In the second stage,an expectation maximization algorithm is applied to estimate the relevant demand parameters,thereby restoring the real demand of customers and providing a decision-making reference for sales planning.Finally,using the actual sales transactions of a domestic community group-buying platform,a real-world example is examined to verify the effectiveness of the improved customer-choice model and novel two-stage algorithm.The results show that considering the inventory availability in the customer choice model can more accurately describe the substitution effect between similar products,and the proposed two-stage algorithm shows obvious performance advantages.In summary,we make the following three contributions.First,we analyze and develop a novel model for customer choice behavior in a community group-buying context that considers inventory availability.Second,we design a novel two-stage estimation algorithm to solve the demand parameters in a maximum likelihood estimation model,and then verify the effectiveness of the model and algorithm by using them to examine the real sales data of a large domestic community group-buying platform.Third,our findings not only have a particular guiding significance for sales planning in a community group-buying context but can also be applied to enhance other areas of e-commerce operational management.
万方数据
维普
