VIETNAM NATIONAL UNIVERSITY, HANOI INTERNATIONAL SCHOOL GRADUATION PROJECT PROJECT NAME FINANCIAL DISTRESS PREDICTION: EVIDENCE FROM VIETNAM Student’s name Le Tuan Tung Hanoi - 2024 VIETNAM NATIONAL UNIVERSITY, HANOI INTERNATIONAL SCHOOL GRADUATION PROJECT FINANCIAL DISTRESS PREDICTION: EVIDENCE FROM VIETNAM SUPERVISOR: PhD. Truong Cong Doan STUDENT: Le Tuan Tung STUDENT ID: 21070006 COHORT: QH-2021-Q SUBJECT CODE: INS401101 MAJOR: Management Information System Hanoi - 2024 Table of Contents List of Figures. 1 List of Tables. 1 List of Abbreviations.
Financial Distress Prediction. Research Objectives and Contributions. Machine Learning in Financial Distress. Random Forest Classification.
Support Vector Machine. Gradient Boosting Machine. Artificial Neural Networks. Experimental Set-up .5 Principal Component Analysis (PCA).
Class Imbalance: Over-sampling. Result and Discussion .1 Finding on total.2 Findings by sectors. Demo Streamlit App. Conclusion and Future Research.46 2 List of Figures Figure 1.
Distribution of business types. Distress level by business type. Correlation matrix of financial ratios. Distribution of financial ratios.
Distribution of financial distress levels over the years. Flowchart of financial prediction model. ROC curves for total data. Home page - App Demo.
Prediction Distress Interface. Insert data to web - demo. Result from web demo. 42 List of Tables Table 1.
Summary of Literature Review. List of Independent Variables. Hyperparameters optimization for manufacturer company. Accuracy Before and After tuning for total data.
Distribution of training and testing data. Accuracy Before and After Tuning the Hyperparameters. Evaluation indexes for SVM. ROC curves of SVM for three types of company.
AUROC of models for three types of company. 38 1 List of Abbreviations • ANN: Artificial Neural Network • AUC: Area Under the Curve • AUROC: Area Under the Receiver Operating Characteristic Curve • BNK: Bankrupt • CL: Current Liabilities • DPD: Days Past Due • EBIT: Earnings Before Interest and Taxes • FCF: Free Cash Flow • FDP: Financial Distress Prediction • IE: Interest Expense • LSTM: Long Short-Term Memory • MCC: Matthews Correlation Coefficient • ML: Machine Learning • MVE: Market Value of Equity • NOCF: Net Operating Cash Flow • PCA: Principal Component Analysis • RE: Retained Earnings • ROC: Receiver Operating Characteristic • RNN: Recurrent Neural Network • SVM: Support Vector Machine • TA: Total Assets • TL: Total Liabilities • XGBoost: Extreme Gradient Boosting • Z-Score: Altman's Z-Score • FDP: Financial Distress Prediction 2 Abstract This thesis investigates the application of machine learning techniques to predict financial distress among companies in Vietnam, utilizing a comprehensive dataset encompassing financial ratios from 509 firms, resulting in 7,126 observations spanning from 2006 to 2019. Financial distress prediction (FDP) is crucial for stakeholders, including investors, creditors, and policymakers, as it aids in identifying early warning signs and mitigating potential financial crises. To create prediction models, the study uses a variety of machine learning techniques, such as ANN, XGBoost, gradient boosting, random forests, and support vector machines.
The financial distress levels are labeled using the Altman Z-score, a well- established metric for assessing a company's financial health. These models are trained and validated on historical financial data, focusing on key financial ratios that serve as indicators of a company's financial health. To guarantee solid and trustworthy predictions, the performance of these models is assessed using the following metrics: accuracy, precision, recall, and the area under the receiver operating characteristic (ROC) curve. My findings demonstrate that machine learning models can significantly enhance the accuracy of financial distress predictions compared to traditional statistical methods.
The results indicate that specific financial ratios, such as liquidity, profitability, and leverage ratios, play a pivotal role in predicting financial distress. Additionally, the study highlights the importance of selecting appropriate machine learning algorithms and features to improve predictive performance. This research contributes to the existing literature by providing empirical evidence from Vietnam, a rapidly developing economy with unique financial characteristics. The insights gained from this study can be instrumental for corporate managers, investors, and policymakers in making informed decisions to prevent financial distress and promote economic stability.
Keywords: Financial Distress Prediction, Altman Z-score, Machine Learning 3 Author’s Declaration I certify that I am the author of this thesis. This is an accurate copy of the thesis that my examiners have approved, along with any necessary last-minute changes. I give the Vietnam National University's International School permission to lend this thesis to other organizations or people for academic purposes. Additionally, at the request of other organizations or persons for scholarly research, I grant permission to the International School, Vietnam National University, to replicate this thesis in whole or in part through photocopying or other means.
I understand that the public will have electronic access to my thesis. Le Tuan Tung 4 Acknowledgements First and foremost, I would like to express my deepest gratitude to my advisors, Dr. Truong Cong Doan and Dr. Nguyen Thi Kim Oanh, for their constant assistance, direction, and inspiration during my research.
Their deep expertise and perceptive criticism have been extremely helpful in developing my argument. Their patience and dedication to my academic growth have been a source of inspiration, and I am immensely grateful for their mentorship. Their expertise in financial analysis and machine learning provided the foundation for this study, and their contributions were crucial in the development and completion of this research on financial distress prediction among listed companies in Vietnam. I would also like to thank the faculty members of International School - Vietnam National University, Hanoi for providing a stimulating and supportive academic environment.
Additionally, I extend my appreciation to my friends who have provided support and encouragement throughout this journey. Their companionship and discussions have enriched my learning experience and made this endeavor more enjoyable. The collaborative environment allowed me to explore different perspectives and improve my understanding of machine learning techniques and their application in financial distress prediction. I am also grateful to the financial institutions and data providers that made the data for this research available.
The comprehensive dataset from the FiinPro Platform was essential in conducting a thorough analysis and developing robust predictive models. Without their cooperation, this research would not have been possible. Lastly, I want to express my sincere gratitude to my family for their constant understanding and support. Their support, affection, and faith in my skills have been my inspiration and motivation.
The countless hours they spent listening to my progress and providing emotional support were instrumental in the completion of this thesis. Without the assistance and efforts of each of these people, my thesis would not have been feasible, and for that I am very grateful. Their collective efforts have not only made this research possible but have also enriched my academic journey, providing me with the skills and knowledge to pursue future endeavors in the field of financial distress prediction and beyond. Le Tuan Tung 5 I.
Introduction The prediction of financial distress is a crucial aspect of financial analysis, particularly relevant for stakeholders such as investors, creditors, and policymakers. Accurately predicting financial distress helps in mitigating risks and taking preventive actions to avoid significant economic consequences, including bankruptcy, loss of employment, and broader economic downturns. This thesis focuses on predicting financial distress within the Vietnamese context, leveraging a combination of financial ratios and machine learning techniques to enhance prediction accuracy and reliability. Financial Distress Prediction Financial distress prediction involves identifying firms likely to encounter financial difficulties in the near future.
Such difficulties often manifest as an inability to meet financial obligations, declining profitability, and liquidity problems. Traditional methods, such as Altman Z-score, have been widely used to predict financial distress level. The Altman Z-score uses financial ratios to classify firms into different distress levels (Altman, 1968). In this thesis, the Z-score will be employed to label firms into three categories of distress level: 2 (non-distressed), 1 (moderately distressed), and 0 (severely distressed).
However, the emergence of machine learning has introduced more advanced approaches capable of handling complex patterns and large datasets, potentially improving prediction performance. Based on sixteen financial ratios, this thesis uses a variety of machine learning methods, such as Random Forest, Support Vector Machine (SVM), Artificial Neural Networks (ANN), Gradient Boosting, and XGBoost, to predict financial hardship. These algorithms are chosen for their ability to capture non-linear relationships and interactions among variables, which traditional methods might miss. Problem Background Vietnam's shift from a centrally planned economy to a market-oriented one has driven rapid economic growth and heightened global integration.
This transformation has also exposed Vietnamese firms to various financial risks, necessitating robust financial distress prediction models. The evolving regulatory and economic landscape in Vietnam makes it imperative to develop accurate and reliable prediction models tailored to local conditions. This thesis specifically focuses on manufacturing, service, and trade companies in Vietnam. These sectors are critical to the Vietnamese economy, contributing significantly to GDP, employment, and international trade.
However, they also face 6 unique financial challenges due to fluctuations in market demand, supply chain disruptions, and changing regulatory environments. Predicting financial distress in these sectors can provide valuable insights and preemptive measures to mitigate adverse economic impacts. Earlier research on predicting financial distress in Vietnam has predominantly utilized traditional statistical methods, including logistic regression and discriminant analysis (Nguyen & Tran, 2019). While these methods have provided valuable insights, they may not fully capture the complexity of financial distress in the modern business environment.
This thesis aims to address this gap by incorporating machine learning techniques, presenting a novel approach to improving the predictive accuracy of financial distress models. Additionally, this thesis seeks to create an intuitive web application using Streamlit, a Python library that facilitates the development of interactive web applications. The application will allow users to input financial ratios and receive a prediction of the distress level based on the trained machine learning models. This practical tool will provide valuable assistance to stakeholders in making informed decisions.
Problem Statement The Vietnamese economy, characterized by its rapid transformation and integration into the global market, presents unique challenges and opportunities for financial distress prediction. Manufacturing, service, and trade companies, which are essential to Vietnam's economic foundation, are particularly vulnerable to financial instability due to various internal and external influences, such as market volatility, changes in regulations, and inefficiencies in operations. Despite the significant role these sectors play, existing studies on financial distress prediction in Vietnam have predominantly relied on traditional statistical approaches, which may not adequately reflect the intricate and ever-changing nature of financial distress in today's business landscape. These traditional methods often fail to account for the intricate interactions between multiple financial variables, leading to suboptimal predictive performance.
There is a critical need for more advanced and accurate predictive models that can leverage the vast amount of financial data available today. Machine learning methods, known for their capacity to manage extensive datasets and identify intricate patterns, present a promising solution to this issue. However, their application in the Vietnamese 7 context, especially for manufacturing, service, and trade companies, remains underexplored. This thesis seeks to bridge this gap by creating and assessing machine learning models for predicting financial distress in Vietnamese companies, using a comprehensive set of financial ratios and the Altman Z-score for labeling distress levels.
Additionally, the development of a web application using Streamlit will provide a practical tool for stakeholders to easily input financial data and obtain distress level predictions, enhancing decision-making processes. Research Objectives and Contributions The primary objectives of this thesis are as follows: ● To identify the key financial ratios that significantly impact financial distress prediction in Vietnamese manufacturing, service, and trade companies: ○ This involves selecting and analyzing 17 financial ratios to determine their relevance and predictive power in the context of financial distress. ● To use the chosen financial ratios to create, evaluate, and compare different machine learning models for financial distress prediction. ○ This includes applying algorithms such as Support Vector Machine (SVM), Random Forest, XGBoost, Artificial Neural Networks (ANN) and Gradient Boosting Machine to create models.