Product combination purchase choice model based on item-level sales data

Deng, Shiming, Ouyang, Bohan, Xu, Mingyu, Lai, Xin
Journal of Revenue and Pricing Management · Springer Nature · 2026-10-31 · DOI: 10.1057/s41272-026-00638-w

Abstract

Demand estimation is fundamental to revenue management and operations management, supporting inventory replenishment, product bundling, and assortment decisions. However, upstream supply-chain firms may observe only repeated period-level aggregate item sales, which reveal how many units of each product are sold but not how products are purchased together. This paper develops a latent-variable model to recover product-combination purchase probabilities from such aggregated sales data. Treating customer-level purchase configurations as latent variables, we formulate the observed-data likelihood and derive an EM estimation procedure for a given candidate combination set. We establish identifiability of the combination probability vector and consistency of the global maximum likelihood estimator under correct specification. To address the exponential growth of the combination space, we propose an augmentation procedure that iteratively adds promising combinations and removes low-probability ones. Numerical experiments show that the proposed method recovers latent combination-purchase patterns and closely approximates the full-combination benchmark using a substantially smaller candidate set.

Read at publisher →

Similar articles