재무제표 주석은 서로 얼마나 유사한가

The adoption of IFRS has increased the length and importance of the financial statement notes. However, the report which investigated the status of notes of 20 listed companies showed that companies fail to include useful information in their notes and heavily refer to the note templates provided by auditors or Financial Supervisory Service(Lee and Han 2018). This study expands the investigation of notes preparation status to all listed companies using an intuitive measure derived from the cosine similarity method. We examine the cross-sectional and longitudinal similarities of financial statement note sentences for all companies. The cross-sectional similarity measures how similar the sentences in note are across companies. First, we measure the cosine similarity of each sentence combination between the sentences in the focal firm’s note with all the sentences in the other firm’s financial statement note by year. If the maximum cosine similarity of the sentence pair is larger than 0.886 (less than cosine 30˚), we define the focal firm’s sentence has a ‘significantly similar sentence’ in other firms. Second, for each focal firm’s sentence, we count the number of other firms having significantly similar sentence with the focal firm’s sentence in the note. The number shows how many firms share the significantly similar sentence with each specific note sentence of the focal firm. Third, for each focal firm’s sentence, we compute the ratio of firms containing similar sentence through dividing the number of prior stage by the total number of comparing firms. Forth, to measure firm level note similarity, the ratios of each sentence are averaged by year. Our analysis reports that any sentence in financial statement notes in general has ‘significantly similar sentence (the same sentence)’ in 11.8% (3.3%) of other companies notes. The longitudinal similarity measures the similarity of note sentences with sentences in the prior year’s notes. It differs from the cross-sectional similarity that the note sentences are compared with its own note sentences of the previous year. Each note sentence is compared with all the sentences in the previous year’s note. Then, we measure the maximum cosine similarity between the sentences. The larger this value is, the more similar the sentence is to the company’s previous year’s note. Since this value is for each sentence, the longitudinal similarity of the firm is measured as the average maximum cosine similarity of each sentence. The mean value of the longitudinal cosine similarity by firm-year is 0.8448 and the median is 0.8881. The longitudinal similarity shows that most of notes’ sentences are similar to the notes’ sentences in the prior year’s financial statements. The result confirms that firms prepare the note with reference to the previous year. Additionally, we expect that the cross-sectional similarity would be higher for the firms in the same industry because the businesses are similar. The cross-sectional similarity within the same industry is higher than that of all companies as expected, but the difference is very marginal. This study contributes to the prior literature by suggesting how to measure the note similarity more objectively. The methodology can contribute empirical studies which investigate the factors that may affect note preparation. The limitation is that the specific contents of the note are not examined because the cosine similarity mechanically measures the similarity of the sentences.