Forecasting Demand of New Products: A Hybrid Structural and Data Driven Approach
We study a new product demand forecasting problem faced by a cosmetic retailer. The retailer’s business strategy requires frequent launching of new products with new functionalities (e.g., facial mask) or new regimes (e.g., Tea Tree Oil). The main challenge is data availability/sparsity: there is no historical data for new products, and it is not obvious how to utilize data about existing products. Pure data-driven approaches in the literature can extrapolate the data about existing products to generate forecast for new products, however, they fail to capture subtle substitution effects between new and old products within or across product categories (in terms of functions and regimes). On the other hand, pure structural models of product substitution requires historical data to calibrate. We propose a hybrid approach that overcomes the disadvantages of both. The approach’s forecast accuracy is back-tested on real dataset.
Tong Wang is an Associate Professor at Department of Analytics & Operations (formerly known as Department of Decision Sciences), National University of Singapore Business School. His research is mostly on data-driven retailing operations and more recently on football analytics. He teaches Business Analytics, Statistical Learning, and Machine Learning for undergraduate, MSc, and PhD programs at NUS. He holds a PhD degree in Decision Sciences from INSEAD, France and Singapore, an MPhil in Systems Engineering and Engineering Management from the Chinese University of Hong Kong, and a B.Eng in Industrial Engineering from Shanghai Jiao Tong University.