Latent Pattern Extraction and Factorization of Firm Bankruptcies and Metabolism in Japan

Abstract

Maintenance and replacement are key elements of transaction management for companies to survive in a highly volatile market. Companies that have too low or too high rates of transactional continuity tend to leave the market, especially in severe times, such as in the aftermath of The Great East Japan Earthquake. In other words, the metabolism of transactions is the key to companies' survival. However, the question is what types of companies that were replaced by other companies and disappeared from the partner companies' list. In this study, we focused on business bankruptcies, which affect the structure of partner firms' transaction networks. Using the data of 1,406,690 cumulative companies over the 10 years from 2007 to 2016 in Japan, provided by Tokyo Shoko Research, Ltd., we investigated the features of companies to identify any patterns in the cases of bankruptcy. We analyzed the business status related to bankruptcy, including discontinuance, suspension, merger, and resumption. To illustrate the latent features of bankruptcies with various corporate features, we used tensor factorial analysis, which reduces data complexity and enables a reasonable interpretation. The analysis results of factorization seem to clarify previously hidden knowledge to support transaction management.