An integrated data envelopment analysis and regression tree method for new product price estimation

• Published in: OR Spectrum Vol. 46; no. 4; pp. 1189 - 1211
• Main Authors: Dellnitz, Andreas, Kleine, Andreas, Tavana, Madjid
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.12.2024; Springer Nature B.V
• Subjects: Business and Management; Calculus of Variations and Optimal Control; Optimization; Data analysis; Data envelopment analysis; Machine learning; Measurement methods; Negotiations; Operations Research/Decision Theory; Original Article; Product development; Regression analysis; Negotiation; Product price estimation; Regression trees; Product innovation; Data envelopment analysis
• ISSN: 0171-6468; 1436-6304
• DOI: 10.1007/s00291-024-00774-y

Data envelopment analysis (DEA) is a well-established method for measuring efficiency among a comparable group of decision-making units (DMUs). DMUs comprise entities with time-related activities—i.e., inputs and outputs. The concept of DMU is not reserved only for business entities; it can also be a project or product. This study focuses on the latter by using DEA efficiency scores to estimate the product price from suppliers’ perspective of newly developed products. These prices are then used as a basis for negotiation. However, DEA-based estimations of such product-related purchasing can only account for deterministic input and output relationships and cannot handle unobservable negotiation behavior. Therefore, we propose a two-stage estimator in which DEA is a deterministic baseline estimator that captures production-related price components. We then train regression trees to estimate the behavioral bargaining surplus. We present a real-world application stemming from the automotive supplier industry to demonstrate the applicability of our approach. Most importantly, we confirm the effectiveness of our approach by substantiating the hypothesis that our method provides better estimates than one-step machine learning methods, especially when there is little knowledge about new products, i.e., when data availability is limited.