A Deep Learning Approach to Industry Classification

• Activity Number: 441 - Novel Statistical and Machine Learning Approaches for Business and Financial Services
• Type: Topic Contributed
• Date/Time: Thursday, August 6, 2020 : 10:00 AM to 11:50 AM
• Sponsor: Section on Statistical Computing
• Author(s): Xiaohang Zhao* and Xiao Fang and Jing He and Olivia Sheng
• Companies: University of Delaware and University of Delaware and University of Delaware and The University of Utah
• Keywords: deep learning; industry classification; document embedding; fintech

Abstract

Industry classification systems (ICSs), aiming at identifying economically related firms as peer firms, play important roles in both academia and industry. The traditional expert-driven approach of designing ICSs have several limitations such as high development and maintenance cost as well as limited granularity. While the alternative text-based algorithm-driven approach is promising in circumventing the limitations, one need to solve the challenging problem of text representation, that is, how to accurately represent the semantics of textual documents with numerical vectors. In this paper, we propose a novel text representation algorithm based on deep learning. The proposed algorithm is capable of handling arbitrary long documents with heterogeneous and drifting concepts. It provides a better trade-off between memorizing local details and capturing global concepts. Based on the proposed algorithm, we further develop a novel ICS. Using the Item 1 and 1A sections of firms' annual 10K reports, the effectiveness of the proposed ICS is empirically validated by showing that it generates better peer groups than existing ICSs.