The effect of corporate governance on the likelihood of fraud through textual analysis: examining the moderating effect of ownership structure

The increasing incidence of fraud worldwide has affected not only organizations but also the broader economy. Fraud is more challenging to detect as perpetrators increasingly adopted innovative methods to conceal their activities. Recently, studies have shifted from traditional fraud detection methods to advanced methods such as textual analysis that provide deeper insight into likelihood of fraud in corporate disclosures. Therefore, the objectives of this study are to analyze the compliance of MCCG 2017, identify and compare narrative patterns using textual analysis, determine the relationship between corporate governance mechanisms and likelihood of fraud, and examine the moderating role of ownership structure in this relationship. This study examines the likelihood of fraud in the Management Discussion and Analysis (MD&A) sections of annual reports using a two-stage text mining approaches which are K-Means clustering to identify document groupings and LDA to extract dominant topics, while sentiment is measured using the Loughran & McDonald and SentiWordNet dictionaries. The identified fraud likelihood is then regressed against corporate governance variables to assess their influence. Based on the sample of 30 likelihood of fraud vs 30 non-likelihood of fraud firms from the year 2017-2022 of public listed firms in Malaysia, the findings of textual analysis reveals that the likelihood of fraud firms are focused on risk management activities, and loss & regularization reflecting the firms emphasized on the risk related issues. Meanwhile, non-likelihood of fraud firms focused on the non-performing activity, investment activity, and market and product activity reflecting the effort of the firms to improve business performance and profitability. This suggests that likelihood of fraud firms are more active in addressing risks and losses, whereas non-likelihood of fraud firms are more concentrating on business growth. Furthermore, multivariate analysis shows that corporate governance mechanisms such as board independence, board gender, and board meetings and ownership structure reduce the likelihood of fraud based on textual analysis, highlighting the role of corporate governance in reducing likelihood of fraud. Nevertheless, this effect is weakened in family ownership, while institutional ownership does not significantly moderate this relationship. This study contributes to literature by integrating textual analysis into fraud prediction, highlighting how corporate governance interacts with ownership structures. The results of this study provide practical implications for regulators, auditors, corporate managers, and investors by offering tools to identify early signals of fraud, enhance corporate transparency and governance practices, and support better decision-making processes, thereby reducing the likelihood of fraud.