An empirical analysis of machine learning-based M&A and violation prediction in Chinese enterprises

Abstract

The merger and acquisition (M&A) market has entered a golden age against the backdrop of global economic recovery, but along with unprecedented development opportunities, the risk of listed companies resorting to improper means to pursue rapid expansion has also significantly increased, providing a breeding ground for fraudulent activities. Therefore, this thesis, consisting of three empirical chapters, adopts machine learning techniques to thoroughly explore and analyse this important and distinctive economic phenomenon and focusses on the prediction of corporate M&A activities and their violations in the Chinese A-share market from 2000 to 2022.

The first empirical chapter examines the effectiveness and advantages of machine learning methods for forecasting in the Chinese M&A market. By systematically analysing the factors that affect the occurrence of takeover activities and constructing a comprehensive indicator system containing 60 factors, we employ supervised learning models, such as the random forest and support vector machine, to predict the release of acquisition announcements. We find that the machine learning models notably outperform the traditional logit model in terms of prediction accuracy and robustness when comparing their forecasting outcomes. Meanwhile, the variable importance ranking reveals the significance of key factors such as acquisition experience and ownership structure in M&A forecasting, and points out that non-state-owned firms and enterprises with high equity dispersion have better predictive efficacy. Moreover, M&A prediction is stable across different economic cycles and has a positive impact on the economic prospects of businesses.

The second empirical chapter further explores the feasibility and validity of machine learning in detecting diversified violations in enterprises. After a comprehensive analysis of the motives and influencing factors of corporate fraud, we build a multi-dimensional fraud prediction framework containing a total of 86 financial and non-financial factors based on the GONE framework. Using two ensemble learning algorithms, random forest and gradient-boosted decision tree, we find that the machine learning models demonstrate superior performance in various types of breach prediction. Besides, this study emphasises the core role of non-financial factors, especially Exposure variables, in violation forecasting and indicates the differences in the relative materiality of non-financial and financial variables in predicting different types of irregularities, highlighting the key signalling role of non-financial factors.

The third empirical chapter delves into the relationship between corporate M&A decisions and their subsequent propensity to engage in fraudulent behaviour. According to the reputation system and agency theory, this study finds that acquirers who issue acquisition announcements have an increased risk of non-compliance in the following year, suggesting that frequent takeovers may be used as a signal to imitate successful operations by companies in order to conceal their potential violations. Further analyses show that factors such as state-owned property, financial status, disclosure quality, and the external governance environment have a significant influence on the correlation between takeovers and fraud tendencies. These findings not only enrich the research perspectives on M&A activities and irregularities of Chinese listed companies, but also provide valuable decision-making references and risk warnings for investors, as well as important guidelines for enterprise compliance management and the formulation and implementation of regulatory policies.

Date of Award	17 Mar 2025
Original language	English
Awarding Institution	University of Nottingham
Supervisor	Xiaogang Bi (Supervisor), Qing-Ping Ma (Supervisor) & Eric Opoku (Supervisor)