Global Journal of Management and Business Research, A: Administration and Management, Volume 22 Issue 1

A Study on Machine Learning Prediction Model for Company Bankruptcy using Features in Time Series Financial Data Akira Otsuki α , Shohei Narumi σ & Masayoshi Kawamura ρ Abstract- Based on such methods as a discriminant analysis and logistic regression, corporate bankruptcy prediction models have been developed as a means to determine the soundness of a company’s operational status based on its financial statements. However, such analytical methods work with binary variables, and thus, as the only outcome of machine learning, the company in question is considered either likely or unlikely to go bankrupt. However, this is insufficient for business operators who would need to know the possible risk factors of a bankruptcy, allowing them to plan and implement measures to avoid any misfortunes. We have therefore developed a prediction model that not only predicts but also identifies the financial variables that can possibly drive the company to bankruptcy. Keywords: machine learning; corporate bankruptcy prediction; time-series financial statement data analysis. I. I ntroduction t is extremely important for a company and its stakeholders to have a clear understanding of the operational standing of the company. According to Tasaka [1], to identify corporate credibility based on an analysis of financial statements, studies on “credit analysis” began in the second half of the 19th century, and the Great Depression in the 1930s prompted in- depth research into the prediction of bankruptcies in the United States. As stated in Section 2, many bankruptcy prediction models have been proposed in recent years using methods such as a discriminant analysis and logistic regression. However, these analytical methods only return binary outcomes; in most cases, they run machine learning on financial data and predict whether the company in question will or will not go bankrupt. A few existing studies have discussed factors that explain the possible causes of bankruptcy in the given cases. Nevertheless, they either carried out the explanatory consideration manually or employed a different method for the explanatory analysis, falling short of developing a comprehensive (automated) process model. However, from the viewpoint of business operators, knowing those factors that may lead to a bankruptcy is crucial for the preparation of countermeasures. Given this background, we developed a model that facilitates not only a prediction but also the identification of financial variables that may drive the company to bankruptcy. To evaluate the model, from databases such as kabupro.jp (an online database on listed businesses in Japan), we obtained financial data on financially sound companies and those that went bankrupt. For the operating companies, we referred to the business classification table issued by the Japan Exchange Group, and for each of the 10 primary business categories listed, 10 business entities were randomly selected as the samples. As a result, we verified that the model succeeds in organizing bankrupt companies based on their bankruptcy factors. Furthermore, the model demonstrated its ability to cluster a mixture of sound and bankrupt companies based on their financial patterns, and based on the analyses of financial variables in these clusters, predict specific financial variables that may be exacerbated and lead to bankruptcy. II. E xisting and R elevant S tudies Considering Japanese companies, this study deals with a bankruptcy prediction model, and this section presents an overview of existing studies on prediction models targeting businesses within the Japanese context. The new aspect introduced in this study will be described with reference to such studies. Table 2 lists some of the previous Japanese studies on corporate bankruptcy prediction models. Konoet al. [2] plotted the mean values of their data organized by fiscal year and compared their sample data (of bankrupt companies) with the mean values taken from five listed companies to propose a bankruptcy prediction model. Okubo [3] proposed a model that evaluates the business management status based on eight patterns of combinations of positive (+) and negative (−) values for the chosen variables, as shown in Table 1; for example, if the operating cash flow yields a positive value and the investing and financing cash flows yield a negative value, the company in question is in a sound state of business management and will unlikely to go into bankruptcy. I Author α : Nihon University, College of Economics, Japan. e-mail: otsuki.akira@nihon-u.ac.jp Author σ : NTT Data System Technologies. Author ρ : MK Future Software. 9 Global Journal of Management and Business Research Volume XXII Issue I Version I Year 2022 ( ) A © 2022 Global Journals