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

III. P roposed C oncept a) Definition of bankrupt company and analyzed data This study uses the definition of a bankrupt company provided by the Teikoku Databank [10]. We researched bankrupt companies on the Delisting website [11] and obtained the financial statement data of these companies from either kabupro.jp [12] or COSMOS1 [13] (a corporate financial database administered by Teikoku Databank). Regarding the financial statement data of non-bankrupt companies, we referred to the business classification table issued by the Japan Exchange Group and randomly selected 10 companies for each of the 10 primary business categories defined therein. A total of 84 FS data of bankrupt company datasets and 100 FS data of non- bankrupt company datasets were obtained. Each dataset consists of financial statements of the previous 5 consecutive years, counting from the year of bankruptcy for the 84 bankrupt companies, and from fiscal 2020 for the 100 non-bankrupt companies. Note that, whereas the bankrupt companies were selected to ensure that their corporate sizes and types of trade were unbiased, the same could not be ascertained for non- bankrupt companies because they were randomly selected according to the primary business categories; thus, bias control will be recommended for future evaluative experiments with additional datasets. b) Indexes for bankruptcy prediction We drew on the data used in the existing studies shown in Table 2, that is, the data in the “Data used” column, based upon which we identified bankruptcy prediction indexes (explanatory variables) for employment in our proposed model. Table 3 lists these indexes, together with the rationale for the choice. Table 3: List of bankruptcy prediction indexes (explanatory variables) for the prediction model used in this study and selection rationale Explanatory variable Selection rationale Current ratio This expresses a company’s liquid assets against its liabilities due within the current year- period, and was chosen because the ratio is considered to decrease as the company nears its bankruptcy. It may be noted that a quick ratio was not selected because the scope of current assets was too narrow. Operating cash flow This was chosen because it is considered that, in the case of bankruptcy owing to a poor operational performance, the operating cash flow from the main business diminishes. Investing cash flow This was chosen because the investing cash flow is likely to increase when a company struggles to settle its liabilities, which is attributed to sales of assets such as facilities and company vehicles. Operating cash flow/current liabilities This indicates a company’s ability to settle the liabilities due within the current year from the cash derived from its business activities. This is chosen because the ratio is considered to decrease when the company’s performance declines. Inventory turnover (sales revenue/inventory) A poor performance will lead to a decline in sales revenue, resulting in an increase in inventory (in this case, dead inventory), hence the choice. Operating cash flow/sales revenue This was chosen because it is possible that bankruptcy may result from a company being excessively short of cash to fulfill its obligations owing to too many illiquid assets such as collectibles despite realizing a large sales revenue. Return on equity (net profit/equity) Did a bankrupt company practice efficient business management? Was its operating efficiency decreasing over the years prior to the bankruptcy? Knowing the answers to these questions is considered important in formulating preventive measures. Equity/total liabilities A company likely to go bankrupt undoubtedly has its equity minimized (and in some cases, its liabilities increased), resulting in a decrease in this ratio, and hence the bankruptcy decision. As another reason, it indicates the company’s reserve of capital without obligations after offsetting the liabilities. c) Extraction of features from time-series financial data We will now describe the model used for extracting the features of each of the aforementioned explanatory variables, observed during a 5-year period. Dealing with time-series data, it is common practice to use the logarithm rate of change (logarithmic return) [14, 15]. However, this cannot be obtained if a negative value is involved when obtaining a natural logarithm. For this reason, we decided to calculate, instead of the logarithm, the rate of change of the financial indexes over a 5-year period, as shown below: In Formula 1, (Y t ) expresses the change rate between the fiscal year t and the preceding year t−1, and X is each of the eight variable specified as evaluation indexes in the previous section. Y t = (X t −X t−1 ) X t−1 (1) 12 Global Journal of Management and Business Research Volume XXII Issue I Version I Year 2022 ( ) A © 2022 Global Journals A Study on Machine Learning Prediction Model for Company Bankruptcy using Features in Time Series Financial Data