Influence of Institutional Investors' Investment Periods on Earnings Management in Korea: Focusing on the Selection of the Variables Using AI

• Published in: Global business and finance review Vol. 29; no. 9; pp. 86 - 101
• Main Authors: Lee, Soojun, Yeo, Hyeopgoo, Baek, Sujin, Choi, Young Huan
• Format: Journal Article
• Language: English
• Published: Seoul People and Global Business Association 31.10.2024
• Subjects: Earnings management; Institutional investments; Investors; Machine learning; Variables
• ISSN: 1088-6931; 2384-1648
• DOI: 10.17549/gbfr.2024.29.9.86

Purpose: This study investigates the influence of the investment period of institutional investors on earnings management in the Korean capital market, with a particular focus on selecting control variables using explainable AI indicators. Design/methodology/approach: This research utilized machine learning models, specifically the AdaBoost model, to identify control variables within our analytical framework. This analysis is based on 10,392 firm-year observations of non-financial entities listed on the Korea Exchange from 2011 to 2020. Findings: The research found that the investment holding periods of institutional investors had no significant correlation with accrual-based earnings management. However, a substantial negative association was observed with real earnings management, indicating that prolonged investment durations by institutional investors enhance their ability to monitor management effectively. This increased oversight appears to mitigate real earnings manipulations. Additionally, the use of the AdaBoost machine learning model for selecting control variables demonstrated enhanced predictive capabilities. Research limitations/implications: While the use of AI, particularly the AdaBoost model, has proven effective in selecting control variables for analyzing the influence of institutional investors on real earnings, further research is needed. Future studies should aim to identify the most effective methods of control variable selection to enhance the explanatory power of models in diverse economic contexts. Originality/value: This study examines the characteristics of institutional investors beyond shareholding ratios, focusing on their monitoring abilities. It employs advanced machine learning techniques to analyze their impact on earnings management through real activities manipulations. The research challenges traditional methods of measuring investor oversight and proposes a new, dynamic model. Additionally, the findings support policy development for enhanced institutional investor engagement.