Companies’ intangible asset ratios have been increasing year by year, and their importance is growing. Although valuation methods for intangible assets have been extensively studied in accounting, it is difficult to accurately evaluate intangible asset values, because no clear rules for doing so exist. Consequently, balance sheets do not record latent intangible fixed assets, meaning that companies’ balance sheets do not accurately represent the value of their assets. In this study, I developed a method that uses machine learning to calculate the predicted value of intangible fixed assets, then proposed and verified a method to extract industries with high potential for unrecorded intangible fixed assets based on the deviation rate between the predicted value of intangible fixed assets and the value of intangible fixed assets recorded on the balance sheet. First, I constructed a machine learning model that provides a highly accurate prediction of intangible fixed asset value for a sample of Nikkei 225 stocks by comparing several algorithms using automated machine learning. Next, I used the coefficients of determination, prediction error plots, and learning curves to evaluate the constructed model and confirm that it met an acceptable performance level. The deviation rate between the constructed model’s predictions and the value of intangible fixed assets recorded on the balance sheet was used to identify companies with a high probability of having intangible fixed assets that were not recorded on the balance sheet. Finally, by calculating the mean deviation rate for each industry type, I identified industries with a relatively high probability of latent unrecorded intangible assets.
View full abstract