Applying the Textual Mining Technique to Study the Association between the Information Content of Key Audit Matters of Taiwan Listed Companies and Financial Performance

Abstract

According to Statement on Auditing Standard(SAS) No 58 auditors shall disclose the Key Audit Matters(KAMs) in the audit reports to reduce the audit expectation gap and improve the information disclosure transparency However the content of audit reports is difficult to understand for some investors Today artificial intelligence (AI) is a booming field with numerous practical applications and on-going research topics The sentiment analysis achieved the great performance in predicting for digital currency and real estate market therefore this study applies the textual mining technique to examine the association between the emotional valence of KAMs and the financial performance of companies which provides another intuition index for investors This study analyzes the information content of KAMs of Taiwan listed companies for sentiment analysis with deep learning by using BERT which is a technique for NLP (Natural Language Processing) pre-training developed by Google There are 3 366 KAMs from 2016 to 2017 we use 80% of dataset and 5-fold cross-validation to train the model and use 20% of the dataset as the test dataset And the remaining 1 527 KAMs in 2018 are predicted based on the trained model Overall this study shows that the emotional valence of KAMs has a positive correlation with the marketing performance of the current year and the coming year And the emotional valence of KAMs has a positive correlation with the financial accounting performance of the coming year while no correlative outcome for the current year

Date of Award	2020
Original language	English
Supervisor	Meng-Feng Yen (Supervisor)