MINISTRY OF EDUCATION AND TRAINING STATE BANK OF VIETNAM BANKING UNIVERSITY OF HO CHI MINH CITY GRADUATION THESIS ANALYZING FINANCIAL METRICS FOR PREDICTING DEFAULT PROBABILITY IN SMALL AND MEDIUM ENTERPRIES MAJOR: FINANCE & BANKING CODE: 7340201 HOANG ANH DUC HO CHI MINH CITY, 2024 MINISTRY OF EDUCATION AND TRAINING STATE BANK OF VIETNAM BANKING UNIVERSITY OF HO CHI MINH CITY GRADUATION THESIS ANALYZING FINANCIAL METRICS FOR PREDICTING DEFAULT PROBABILITY IN SMALL AND MEDIUM ENTERPRIES MAJOR: FINANCE & BANKING CODE: 7340201 Full name: HOANG ANH DUC Student code: 030632160346 Class: HQ4-GE01 THESIS ADVISOR Ph. NGUYEN MINH NHAT HO CHI MINH CITY, 2024 REVIEWS OF THESIS ADVISOR. Ho Chi Minh City, January 17, 2024 Thesis advisor ABSTRACT Commercial banks rely on their internal credit rating system to evaluate the credit risk of customers and make informed decisions about extending credit while managing risks. Meanwhile, the Government is working on developing a legal framework for credit rating to improve transparency, help banks control initial credit risk, facilitate capital mobilization through the stock market, and protect investors' rights and interests.
This initiative aims to strengthen the overall financial ecosystem by providing clear guidelines for assessing and mitigating credit risks. Researching and choosing appropriate rating models will greatly enhance the advancement of credit rating activities in Vietnam. Therefore, it is important to address the specific constraints and ongoing discussions surrounding existing models for forecasting default likelihood. There is a lack of consensus regarding the reliability of these models, posing challenges in selecting an appropriate model for predicting business default probability.
Furthermore, identifying which financial ratios influence the rating outcomes remains a critical objective that necessitates exploration within default prediction research. To date, there has been limited published research in Vietnam on selecting models to predict enterprises' default probability based on financial indicators. Thus, the thesis centers on the problem of "Analyzing financial metrics for predicting default probability in Small and Medium Enterprises" in order to systematically offer theoretical support and empirical data to commercial banks regarding the choice of a suitable business bankruptcy prediction model, thereby enhancing the bank's future credit risk management efficiency. In light of its significance and imperative, the primary aim of this research is threefold: (i) to ascertain the criteria for an apt forecasting model; (ii) to delineate the process for selecting a model proficient in forecasting the default probability of Small ii and Medium Enterprises (SMEs) within Vietnamese commercial banks, grounded in financial indicators.
The outcomes derived from this inquiry are intended to furnish supplementary quantitative empirical substantiation, elucidating the optimal predictive model for ascertaining the probability of default among medium and small-scale enterprises within Vietnamese commercial banks. (iii) A paramount contribution of this research lies in the foundational development of the utilization of financial indicators for prognosticating the default probability of SMEs, thereby augmenting the efficacy of credit risk management in Vietnamese commercial banks in the foreseeable future. Small and Medium Enterprises (SMEs) exert a substantial influence on the economies of numerous nations, particularly those in the developmental stage. Globally, SMEs constitute the predominant proportion of businesses and significantly contribute to both employment generation and the advancement of the worldwide economic landscape.
Referred to interchangeably as micro, small, and medium enterprises, these entities are characterized by their modest scale in terms of capital, labor force, or turnover. Classified based on their dimensions, SMEs can be categorized into three distinct groups: micro enterprises, small enterprises, and medium enterprises. As per the criteria established by the World Bank Group, a micro enterprise is defined by an employee count of fewer than 10 individuals; a small enterprise encompasses 10 to less than 200 employees, accompanied by a capital of 20 billion or less; medium enterprises, on the other hand, involve 200 to 300 employees and a capital ranging from 20 to 100 billion. The Probability of Default (PD) holds a pivotal role in various credit risk analysis and risk management endeavors.
In accordance with the Basel II framework, it serves as a crucial parameter employed in the computation of economic capital levels, essential for mitigating risks within credit institutions. PD stands out as an indispensable metric for the categorization of borrowers, mandating all banks, irrespective of their utilization iii of standard or advanced methods, to furnish regulatory authorities with an internal PD estimate corresponding to the borrower's profile. The hierarchical outcome derived from PD rankings is deemed notably accurate, given its calculation grounded in the firm's authentic financial ratios, thereby offering a practical reflection of the business's current state. The judicious consideration of PD can significantly ameliorate credit risk when comprehensively incorporated into risk management strategies.
A comprehensive examination of both domestic and international studies reveals that financial institutions possess a repertoire of diverse credit rating models to prognosticate the likelihood of enterprise default. These predictive models encompass polynomial models, logit models, probit models, and artificial neural network models. Furthermore, these ranking models incorporate various inputs and financial indicators to anticipate business bankruptcy, with commonly employed financial ratios including short-term solvency, rate of return on total assets, and total liabilities to total assets. However, disparities in conclusions arise across researchers when employing diverse datasets spanning distinct periods, leading to variations in the selection of appropriate credit rating models and the identification of financial indicators influencing the probability of default.
Notably, the application of such models in predicting the default likelihood of Small and Medium Enterprises (SMEs) customers in Vietnam, as posited by the author, introduces a novel perspective. The analysis, comparison, and synthesis of the aforementioned studies, along with related issues, underscore several research gaps. Consequently, the author proposes a research model and anticipates employing a specific methodology to address these gaps and contribute to the existing body of knowledge on the subject. To attain the research objectives, the author executed a structured methodology comprising four stages, each delineated by specific procedural steps.
In the initial stage, data collection and processing were undertaken. Subsequently, during the second phase, iv a meticulous selection of input variables for the model was conducted. The third stage involved the execution of regression analyses on chosen credit rating models, namely the logistic regression model, the decision trees model, the gradient boosting model and the artificial neural networks model. The final stage encompassed the utilization of the Confusion matrix and F1-Score to assess the accuracy of each outcome of the models.
Through this comprehensive process, the objective was to discern an appropriate credit rating model demonstrating efficacy in predicting the probability of default among customers. The research utilized a dataset extracted from the annual financial statements of approximately 400 companies spanning the period from 2020 to 2022. Rigorous auditing processes were employed to ensure the integrity and quality of the data source. Among the sampled businesses, 31 were engaged in consumer goods trading, 35 in the petroleum sector, 39 in the automotive industry, 40 in the construction and installation sector, 43 in the pharmaceutical and medical equipment domain, 45 in the textile and garment sector, 47 in fisheries (including fish, shrimp, clam, etc.), 54 in the iron and steel industry, and 66 in agriculture (covering rice, coffee, pepper, etc.
A meticulous selection process led to the inclusion of 14 financial indicators as independent variables in the credit rating models under scrutiny. The analysis of regression outcomes derived from parametric models, coupled with the application of metrics calculated from the confusion matrix (Accuracy, Sensitivity, Specificity, Precision, F1-Score), facilitated a comparative evaluation of each model's proficiency in predicting the default probability of enterprises. This systematic approach aimed to identify an optimal model with the capability to effectively forecast the likelihood of default among businesses. The conclusive findings of this research underscore the significance of four out of thirteen variables in effectively predicting the default probability of customers.
v Specifically, these influential variables encompass Total revenue/Total assets, Income before tax/Net revenue, Income before tax/Total assets, Receivables/Average Revenue. These outcomes offer practical insights for commercial banks, empowering them to assess and strategically select customers, thereby mitigating the risk associated with loan repayment uncertainties. Building upon these research outcomes, the author proffers recommendations for commercial banks to enhance their internal credit rating systems in the future. The thesis introduces a predictive model tailored to forecast the solvency (default probability) of Small and Medium Enterprises (SMEs) customers within commercial banks in Vietnam.
This model serves as a strategic tool for maintaining credit quality and minimizing the incidence of Non-Performing Loans. Customers assigned a favorable credit rating (classified as A or higher), coupled with a positive evaluation of repayment capacity based on the model, exhibit a diminished likelihood of incurring Non-Performing Loans. Consequently, the credit risk associated with this subgroup of customers is deemed negligible. The model, as delineated, serves as a instrumental tool for commercial banks in the realm of credit provision, ensuring the maintenance of credit quality, and fostering a judicious, secure, and sustainable trajectory of expansion and growth.
Its utility extends to aiding banks in the strategic selection and retention of a robust customer portfolio, facilitating the implementation of marketing strategies tailored towards low-risk clientele, and fostering the cultivation of a network populated by reputable customers, thereby safeguarding the repayment of debts. Consequently, the results derived from the model form the foundation for commercial banks to strategically tailor their credit allocation, directing resources away from clients deemed weak (with a high probability of default) and towards those exhibiting strong performance (with a low probability of bankruptcy). Simultaneously, vi the model informs the formulation of a credit policy attuned to the nuances of each customer type, encompassing credit terms, interest rates, fees, and security requirements, thereby ensuring operational safety. Furthermore, the information encapsulated in the assessment of solvency and model outcomes unveils insights into nuanced issues pertaining to the business performance of enterprises and specific industries.
Consequently, the model emerges as a substantive source of information for the forthcoming analysis, assessment, forecasting, and administration of credit policies. vii DECLARATION This thesis represents the original research of the author, presenting accurate and authentic research outcomes. The content contained herein is devoid of any previously published material or contributions from others, with the exception of appropriately cited references acknowledged within the thesis. The author Hoang Anh Duc viii ACKNOWLEDGEMENTS Foremost, I extend my sincere appreciation to the esteemed faculty at Banking University of Ho Chi Minh City for their dedicated and insightful instruction.
Their commitment to imparting knowledge has been instrumental in fortifying my academic foundation, enabling the successful fulfillment of the university curriculum. In particular, I wish to express profound gratitude to Mr. Nguyen Minh Nhat for his meticulous guidance and wholehearted assistance throughout the completion of this graduation thesis. His thoughtful support has proven indispensable, significantly contributing to the successful execution of this academic endeavor.
Acknowledging the constraints of my limited practical experience, I anticipate and welcome further guidance from esteemed educators to refine and enhance the content of the graduation thesis. I am confident that these insights will be invaluable for my ongoing academic and professional development. I extend my sincere thanks for the support and guidance received. ix TABLE OF CONTENT ABSTRACT.
viii TABLE OF CONTENT. ix LIST OF ABBREVIATION. xi LIST OF FIGURES. xi LIST OF TABLE.
xii CHAPTER 1: INTRODUCTION. The Urgency of The Research. The Structure of the Research .8 CHAPTER 2 LITERATURE REVIEW. Small And Medium Enterprises (SMEs).
Probability Of Default (PD).1: 4 popular groups of financial indicators. Overview of probability of default models. Probability of default models. The difference between Logistic Regression model, Decision Trees model, Gradient Boosting model and Artificial Neural Networks model.
Related studies in Vietnam. The other related studies .22 CHAPTER 3: DATA AND METHODOLOGY OF RESEARCH. Data collection and processing .1: Synthesize the number of businesses - business lines. Selection of input variables in the default prediction model .2: Independent variables in the probability default prediction model.
Models for Estimating the Likelihood of Default. Logistic Regression model. Decision Trees and Gradient Boosting model .