Abstract：For the precise delineation of market boundaries and uncover implicit relationship networks among firms, this paper proposes a method for identifying peer firms based on high-dimensional semantic vectors. We employ a neural-network language model to vectorize product descriptions in annual reports, generating semantic embeddings that capture firms’ economic similarities. Building on resource dependence theory, firms exhibiting high economic similarity are then identified as peer firms of the focal firm. We validate the method’s effectiveness using linear mean models. Results indicate that our approach significantly outperforms traditional industry classification methods in explaining focal firms’ performance and decision-making behaviors, while also exhibiting higher frequency in the composition of peer firms. Robustness checks based on stock-price co-movement further confirm the superiority of our method in characterizing inter-firm economic linkages, and we explore how the number of peers influences the measurement of economic similarity.
王伟，陈逢文，王冰. 基于高维语义向量的同群企业识别方法研究[J]. 管理学报, 2025, 22(9): 1626-.
WANG Wei，CHEN Fengwen，WANG Bing. Research on the Identification of Peer Firms Based on High-Dimensional Semantic Vectors. Chinese Journal of Management, 2025, 22(9): 1626-.
