Working capital over total assets WCAPTA, Retained Earnings over Total assets REAT, Total liabilities to total assets ratio DTA and Asset turnover ratio TAT are four strong variables whi
RESEARCH INTRODUCTION
Rationale of the topic
Vietnam’s real estate market has endured unprecedented turbulence in recent years In 2023, the sector experienced a record wave of dissolutions, with an average of 107 companies ceasing operations each month, totaling 1,284 closures (Lao Động, 2023) As 2024 begins, market liquidity remains weak, and pressures from maturing corporate bonds are mounting, signaling that the downturn shows no signs of relief (Kinh Tế Đô Thị, 2024).
To address the volatile Vietnam real estate market, we need a robust method that accurately predicts bankruptcy risk while comprehensively evaluating the performance of listed real estate companies Traditional approaches, though widely used, are less effective in such dynamic environments with country-specific characteristics Furthermore, current research has not adequately integrated bankruptcy prediction with performance evaluation, leaving managers, investors, and regulators without reliable tools to identify early warning signals and design timely policy responses A unified framework that combines bankruptcy risk modelling with performance metrics can provide actionable insights for risk management, investment decisions, and regulatory oversight in the listed real estate sector.
Excessive reliance on financial leverage, weak risk management capabilities, and a lack of advanced analytical tools tailored to the domestic market are the core drivers of the current downturn As a result, the sector’s overall performance has weakened, with ROE declining from 15% in 2018 to about 5% in recent years.
2023 Without timely intervention, the wave of bankruptcies may continue to expand, eroding investor confidence and negatively affecting the capital market, the real estate market, and the broader economy
Beyond its practical value, this study makes an academic contribution by filling a gap in applying machine learning to bankruptcy prediction and performance evaluation in emerging markets, with Vietnam as a focal example The findings offer a valuable data foundation and a robust methodological framework to guide future research in ML-driven financial risk assessment for developing economies, and they introduce a novel approach to corporate governance in the real estate sector under volatile economic conditions.
These factors render the topic urgent, scientifically valuable, and of significant practical relevance, as shown by recent evidence Recent studies illustrate this trend by applying machine learning models to predict default risk In particular, researchers Hung and Binh (2021), Thùy and Lê Hải (2023), and Nguyễn Minh Nhật (2024) utilized various machine learning techniques to forecast default risk, underscoring the growing importance of AI-driven financial risk assessment in both theory and practice.
In response to the urgent need for effective risk forecasting, this research project, titled “Machine Learning Applications in Forecasting Bankruptcy Risks of Real Estate Businesses Listed on the Vietnam Stock Exchange,” has been initiated The study investigates how machine learning can be applied to predict bankruptcy risk for listed real estate companies and to help these firms operate more effectively, with the aim of contributing to the stabilization of Vietnam's real estate market.
Research Objectives
This study aims to identify and forecast the bankruptcy risk of real estate enterprises listed on the Vietnam stock market, with findings that highlight actionable managerial implications These implications are designed to help these companies operate more effectively and contribute to the stabilization of the real estate market in Vietnam.
To achieve general goals, the thesis has to implement three objectives including:
(1) Identifying some variables affecting to bankruptcy risk of real estate enterprises
(2) Assessing and comparing all used-model in the study to suggest a model with the highest exact
(3) Proposing some managerial implications to assist these companies in operating effectively and contributing to the stabilization of the real estate market in Vietnam.
Research Questions
- What are variables affecting to bankruptcy risk of real estate enterprises?
- Which machine learning model is the most effective in prediction of bankruptcy risk in the study?
- What are managerial implications helpful to assist these companies in operating effectively and contributing to the stabilization of the real estate market in Vietnam?
Subjects and Scope of Research
The subject of this thesis is the probability of bankruptcy risk among real estate businesses listed on the Vietnam stock market
The study begins with a dataset of 215 real estate companies The analysis excludes firms not listed on the Vietnam stock exchange (for example, OTC, private, and UPCOM listings) The final sample comprises 57 real estate companies listed on the Vietnam stock market, representing 26.51% of the original dataset All selected companies provide complete and audited financial data.
This study uses secondary data spanning 2011–2024 from the Fiinpro-X platform The period was chosen because it captures the COVID-19 pandemic’s negative impacts on the Vietnamese real estate sector and the broader economy, including significant disruption and recession driven by shifts in the international economy and geopolitics During this time, many Vietnamese real estate companies faced serious challenges, with a number of firms going bankrupt.
Research Methods
Using a quantitative approach with Python, this study forecasts the bankruptcy risk of real estate enterprises by evaluating five supervised learning models: decision tree classification, random forest classification, Naive Bayes, gradient boosting, and logistic regression (LR) Naive Bayes offers a simple, efficient classification framework, while the random forest comprises hundreds of decision trees built with bagging and provides measures of variable importance Gradient Boosting sequentially adds trees to minimize residuals, enabling the model to capture complex, non-linear relationships with high predictive accuracy Logistic regression remains a classic statistical method for modeling the relationship between independent variables and binary bankruptcy outcomes Together, these models support robust bankruptcy risk assessment and illuminate key predictors in real estate finance.
Contributions of the study
This study aims to deliver practical insights by identifying and forecasting the bankruptcy risk of real estate enterprises listed on the Vietnam stock market, while evaluating their operational performance By analyzing these risk factors and performance metrics, the research helps managers gain a clearer understanding of the drivers of bankruptcy risk for real estate companies The results provide actionable managerial implications designed to help these enterprises operate more effectively and mitigate distress Ultimately, the study supports the stabilization of Vietnam's real estate market by informing strategic decisions that enhance financial resilience and market confidence.
Structure of the Thesis
Chapter 1 provides a concise overview of the study related to the research topic, outlining the rationale for choosing the topic, the research objectives, and the key research questions; it also defines the scope and subject of the research, describes the chosen methodology, and explains the structure of the study, along with the expected contributions and the overall outline of the thesis.
LITERATURE REVIEW
Bankruptcy
In his 1968 work, Altman defines bankruptcy as a situation in which a firm's liabilities exceed its assets, leaving it unable to meet obligations as they come due and triggering formal insolvency proceedings The analysis centers on using financial ratios and discriminant analysis to predict corporate bankruptcy, outlining the indicators and methods that reveal impending failure.
5 categories of financial ratios particularly short-term liquidity, operational efficiency, leverage, profitability and sales activity Among them, short term debt obligation is the important one
Bankruptcy, as defined by Merton (1974) in On the Pricing of Corporate Debt, is the exercise of creditors' put option when a firm's asset value falls below its debt value, triggering default Rather than relying on book value, the study emphasizes market-driven thresholds—asset volatility, debt maturity, and leverage—as the primary determinants, rather than formal legal rules governing bankruptcy.
In his 1980 work, Financial Ratios and the Probabilistic Prediction of Bankruptcy, Ohlson proposed that when a firm's cumulative financial distress pushes its O-Score above a threshold of 0.5, this indicates imminent bankruptcy irrespective of legal filing status, and he introduced these probability thresholds to examine and quantify bankruptcy risk.
Under Article 4 of Vietnam's Bankruptcy Law (2014), bankruptcy is defined as the status of an enterprise or cooperative that has lost solvency and has been declared bankrupt by a court A company is considered bankrupt when it fails to meet its payment obligations for three consecutive months and a court has issued a bankruptcy decision.
Bankrupt enterprises are typically identifiable by weaknesses across key financial dimensions—liquidity, leverage, profitability, efficiency, and market value—which together signal deteriorating financial health These firms often cannot meet current debt obligations due to weak cash conversion, excessive debt, or an imbalanced capital structure As a result, they confront economic distress and potential legal challenges, with investors’ expectations and debt maturities left unmet Bankruptcy can occur when the market value of assets falls below liabilities, underscoring the gap between asset worth and financial obligations.
Financial ratios
Financial ratios are essential quantitative analysis tools that help organizations make informed decisions by assessing their financial health By evaluating liquidity, leverage, profitability, and efficiency through metrics drawn from the balance sheet, income statement, and cash flow statement, companies can measure performance and benchmark against competitors in the same industry Regular monitoring and analysis enable early risk detection, trend forecasting, and the optimization of operational efficiency to guide strategic actions.
Mentioned in Damodaran (2012), Palepu (2012), Brigham (2016) financial measures are divided into 5 categories:
Liquidity ratios serve as the most immediate warning signals of potential bankruptcy for real estate firms by measuring a company’s ability to meet current and short-term debt obligations without relying on external capital They hinge on current assets such as cash, cash equivalents, and accounts receivable, and reveal whether a firm has enough liquid assets to cover short-term obligations even if it owns valuable long-term assets like property projects When liquidity is weak, banks and creditors may accelerate repayment demands or tighten new lending, funding shortages can delay construction, and customer confidence can deteriorate, potentially leading to cancellations of pre-sale contracts Together, these dynamics create a negative cash-flow cycle that exacerbates liquidity stress and markedly heightens bankruptcy risk for real estate developers.
Leverage ratios measure how much current and short-term liabilities a business uses to fund its operations and whether operating earnings can cover interest expenses and debt service A high debt load combined with low liquidity signals a higher risk of bankruptcy, as highly indebted companies must meet fixed debt payments regardless of performance, leaving little room to absorb losses if revenues decline Even short-term disruptions can trigger severe cash-flow shortages for these firms By contrast, companies with lower debt levels tend to weather periods of reduced sales more comfortably and face less immediate insolvency pressure.
Profitability ratios measure a company's ability to generate income over a specific period, offering an overall view of how effectively it uses current resources and manages expenses to produce earnings When profitability is low, bankruptcy risk increases, because insufficient earnings can prevent timely coverage of interest obligations and other fixed costs Persistent low profits indicate revenues are being eroded by high expenses or unfavorable pricing, undermining operational sustainability As profitability declines, lenders may lose confidence, restrict credit access, and constrain investment in improvements, ultimately heightening the likelihood of financial distress and bankruptcy.
Efficiency ratios, also known as activity ratios, measure how effectively a company uses its assets—such as inventory, receivables, and other resources—to generate sales and cash These metrics help evaluate management's ability to convert assets into cash and optimize cash flow Low efficiency ratios increase bankruptcy risk by signaling that the business is underutilizing assets and not generating adequate sales from its resources Slow inventory turnover ties up capital in unsold goods, reducing cash available for daily operations, while poor asset utilization keeps costs high without corresponding revenue growth Over time, these inefficiencies erode cash flow, limit the ability to meet financial obligations, and raise the likelihood of insolvency.
Market value ratios offer meaningful insights for external stakeholders, including investors and market analysts, and help companies assess their position in the stock market and manage value per share Low market value ratios signal higher bankruptcy risk by reflecting negative investor sentiment about a firm’s future performance and financial stability When stock prices decline, the market often anticipates reduced profitability, asset overvaluation, or undisclosed risks, which can erode investor confidence, constrain access to equity financing, and impair capital-raising ability, thereby increasing the likelihood of financial distress and bankruptcy.
Theories
Modigliani and Miller's 1958 paper, "The Cost of Capital, Finance and the Theory of Investment," is a foundational work in corporate finance that helped earn them the Nobel Prize in Economics in 1985 They argue that in a perfect market with no taxes, no bankruptcy costs, no agency costs, perfect information, and identical borrowing costs for individuals and corporations, a firm's value is independent of its capital structure Yet these idealized conditions are unrealistic, and in the real estate sector firms face substantial taxes, significant bankruptcy risk, and a heavy reliance on debt, which can make capital structure matter for firm value.
Modigliani and Miller's 1963 extension of their original model incorporates corporate income taxes to reflect real-world conditions, showing that debt financing provides a tax shield by deducting interest from taxable income, which can increase firm value Yet higher leverage also signals greater financial and bankruptcy risk, so firms should determine an optimal debt level where the tax benefits of the shield are balanced against the costs of financial distress, such as losing customers, legal fees, or forced asset sales at discounted prices.
Two main types of risk are identified: financial risk, the possibility that a firm may not generate enough cash flow to meet interest obligations, and bankruptcy risk, which occurs when total liabilities exceed asset value and can lead to bankruptcy These risks are especially relevant for firms with unstable earnings before interest and taxes (EBIT) and a high proportion of short-term debt, as such debt increases pressure on cash flow This dynamic is particularly pronounced in real estate, where robust cash flow is essential.
This framework has made a significant contribution to modern bankruptcy prediction models The Altman Z-score, first developed in 1968 and refined in 1993, uses leverage as a key indicator to forecast bankruptcy risk The Ohlson O-score model, introduced in 1980, emphasizes that financial leverage and liquidity are important factors in assessing corporate bankruptcy risk.
The theory of working capital management has its roots in early financial studies Brigham (1962) emphasized the importance of balancing current assets and current liabilities to ensure liquidity and support stable business operations Building on these foundations, Beranek and subsequent researchers expanded the framework, illustrating how effective management of working capital affects liquidity, risk, and profitability and guiding firms in optimizing cash flow, inventory, and short‑term financing decisions.
(1966) published the book named “Working Capital Management” Opler et al (1999) suggests that higher liquidity reduces bankruptcy risk by ensuring short-term solvency It is considered one of the earliest systematic contributions to this area, establishing the basis of the traditional approach to working capital management
This traditional theory focuses on managing four key parts: cash, inventory, accounts receivable, and short-term debt The main goals are to ensure sufficient liquidity, optimize the turnover of working capital and avoid financial problems However, ineffective management may lead to two major risks First, liquidity risk occurs when a company cannot convert short-term assets into cash in time to meet its short-term obligations Second, efficiency risk rises when excessive capital is locked in inventories, receivables and stock reducing profitability and limiting reinvestment capacity Poor working capital management can cause serious financial problems While this traditional theory was useful in the past, it does not fully consider modern business challenges in the market volatility, globalization, and technological advancement in managing cash flow
Since the 2000s, the extended working capital theory has evolved to reflect the complexity of modern business environments and to offer more suitable models for capital-intensive, long-cycle industries such as real estate Researchers such as Gitman (2009) and Harris (2005) broaden the theory by incorporating systemic risks—including interest-rate changes, market volatility, and supply chain problems—and the use of technology to monitor and optimize cash flows The modern framework identifies three types of risk: liquidity risk, indicated by a quick ratio below 1.0 or negative net working capital; efficiency risk, reflected in low inventory and receivables turnover; and structural risk, which occurs when long-term investments are funded with short-term debt, potentially causing cash shortages or higher interest costs, a particularly acute issue in real estate where projects are long-term but often financed with short-term loans.
The DuPont analysis framework, developed by DuPont Corporation in the 1920s, is used to evaluate internal business performance and has become a trusted tool in financial analysis for assessing profitability, operating efficiency, and financial risk It decomposes return on equity (ROE) into three drivers—profit margin, asset turnover, and financial leverage—so each component can be analyzed separately By isolating these elements, DuPont analysis explains how a company generates its profits and supports more informed decision-making for managers and investors.
Profit margin shows a company's ability to control costs and generate profit from its core business activities, while asset turnover gauges how efficiently the business converts its assets into revenue—a crucial metric in capital-intensive sectors like real estate Financial leverage assesses the extent of debt in the firm's capital structure and the related financial risk that can arise, highlighting how leverage impacts overall performance and stability.
DuPont analysis provides a holistic view of ROE and its sustainability by breaking ROE into profitability, asset efficiency, and leverage A high ROE driven mainly by elevated debt can create serious risk if market conditions shift, whereas when ROE arises from strong profit margins and efficient asset use, it tends to be more stable and less risky The DuPont framework shows that higher profit margins (return on sales) and a higher ratio of retained earnings to total assets signal stronger profitability and a lower bankruptcy risk Empirical evidence supports this relationship: firms with ROS below 5% often face a higher likelihood of bankruptcy (Altman, 1968), while retained earnings and related internal capital accumulation, as discussed in Ohlson’s work, reflect a company’s ability to withstand financial distress.
In the real estate sector where debt is commonly used, DuPont analysis can help identify companies that are coping with financial problems or risk of bankruptcy
John Burr Williams's Theory of Investment Value (1938) laid one of the earliest rigorous foundations for modern investment analysis, defining the intrinsic value of a financial asset as the present value of all expected future cash flows it will generate within the discounted cash flow (DCF) framework This approach emphasizes two core valuation factors: the time value of money and the risk or uncertainty surrounding future cash flows Williams advocated a long‑term investment philosophy, urging decisions to be driven by fundamental factors rather than short‑term market fluctuations He noted that market prices may diverge from intrinsic value in the short run but are expected to converge toward true value over time Although the original theory did not account for cash flow growth or cross‑industry differences, it became the cornerstone for later methods, including fundamental analysis and value investing.
From the 1970s through the early 2000s, researchers extended Williams’ theory to real estate, an asset class characterized by long cycles, high risk, and volatile cash flows In this context, Fisher (1993) developed a specialized real estate valuation model that explicitly accounts for location, taxes, and depreciation, providing a more nuanced measure of property value Damodaran (2012) further refined the Discounted Cash Flow (DCF) framework by applying distinct discount rates for different real estate segments—commercial versus residential—and by incorporating risks such as legal and zoning issues These refinements are particularly relevant for Vietnamese emerging markets, where market dynamics benefit from a valuation approach that is more realistic and precise in the face of complex, volatile real estate environments.
Williams’ foundational theory and its modern extensions inform the fair value assessment of real estate assets, supporting sound investment decisions and effective financial management According to Valuation Theory, a high price-to-book (P/B) ratio may signal potential overvaluation, indicating mispricing risk or uncertainty about future growth.
Previous studies
Hung and Binh (2021) evaluated six predictive models—Logistic Regression, Bayesian Network, K-Nearest Neighbors, Artificial Neural Network, Support Vector Machine, and Decision Tree—within a Z-Score Altman framework to forecast default risk for 4,693 firms in the banking, securities, and insurance sectors from 2009 to 2020 The study analyzes 30 financial indicators spanning liquidity, capital budgeting, profitability, efficiency, market, and leverage ratios Results show that Logistic Regression, SVM, Decision Tree, and ANN achieve high prediction accuracy of about 96–98%, while KNN and Bayesian Network reach roughly 84–85% Three indicators—inventory turnover ratio, debt-to-equity ratio, and overall debt ratio—emerge as significant drivers of corporate bankruptcy prediction Given its high accuracy, the model is recommended for deployment in the Vietnamese market to assist businesses and investors.
Vân Trang and Dương (2021) investigate the factors shaping financial risk in 48 listed real estate companies, with debt structure, liquidity, solvency, profitability, operational efficiency, and capital structure as key independent variables and firm age, size, and revenue growth as controls Using a combination of Feasible Generalized Least Squares (FGLS) and Quantile Regression, they assess how these determinants affect financial risk across different quantiles The results indicate that higher quick liquidity and stronger self-financing capacity reduce financial risk across all quantiles Accounts receivable turnover is negatively associated with financial risk at the 0.1, 0.25, 0.5, and 0.9 quantiles Return on total assets is inversely related to financial risk at the 0.1 and 0.25 quantiles, while asset turnover is positively related to financial risk at the 0.1 and 0.25 quantiles Overall, quick liquidity and self-financing capacity emerge as the most strongly linked factors for the financial risk of listed real estate companies, signaling that risk managers should prioritize liquidity improvement and self-financing ability to mitigate financial risk in the real estate sector.
Minh (2022) investigates the application of the Z-Score and H-Score models to predict bankruptcy risk among listed real estate companies on the Vietnamese stock market Based on data from 56 firms listed on the Ho Chi Minh City Stock Exchange (HOSE) and the Hanoi Stock Exchange (HNX) from 2017 to 2020, the study finds the Z-Score model achieves a predictive accuracy of 99.11%, compared with 63.84% for the H-Score model The analysis identifies profitability, working capital, and financial leverage as the key factors shaping bankruptcy risk for these companies.
Dũng et al (2022) examine bankruptcy risk in 55 real estate enterprises listed on Vietnam’s HOSE and HNX from 2015–2020, using data from Vietstock, Cafef, World Bank, and ADB Their results show that a higher debt-to-asset ratio, higher return on equity, higher current ratio, and higher interest rates are positively linked to default risk, while higher return on assets, a larger net working capital to total assets ratio, a higher quick ratio, faster accounts receivable turnover, and GDP growth are associated with lower insolvency risk To minimize bankruptcy risk, real estate firms should optimize financial leverage, pursue transparent financing, and rigorously assess project feasibility, while improving profitability metrics and maintaining an appropriate capital structure with solid liquidity Additional emphasis on effective cash flow management, prudent net working capital control, growth through scale and diversification, and vigilant macroeconomic monitoring can further strengthen resilience against insolvency.
Thùy and Lê Hải (2023) research 200 Vietnamese companies from 2017 to 2019 titled
The study titled "Machine Learning Based Bankruptcy Prediction of Vietnamese Companies" compares several machine learning methods—Logistic Regression, Random Forest, Extreme Gradient Boosting (XGBoost), K-Nearest Neighbor, and Naive Bayes—using financial indicators such as debt, liquidity, profits, company size, and growth to forecast bankruptcy risk The results indicate XGBoost is the best at predicting bankruptcy, achieving the highest accuracy and the fewest mistakes (13.86% and 10.81% errors), with Random Forest placing second and showing slightly higher error rates (21.78% and 11.71%) The research suggests that machine learning methods outperform traditional approaches like logistic regression, and boosting methods such as XGBoost often outperform other models.
Anh (2024) conducted a study titled " Forecasting the bankruptcy risk of Vietnamese listed real estate companies through the Z-Score model" The research analyzed data from
From 2019 to 2023, 41 real estate companies were listed on Vietnam's stock market, including 29 on the Ho Chi Minh City Stock Exchange and 12 on the Hanoi Stock Exchange A Z-Score model using four financial ratios—working capital, retained earnings, earnings before interest and taxes (EBIT), and the market value of equity relative to total liabilities—was employed to assess solvency The analysis found that five of the 41 companies faced bankruptcy risk Based on these findings, the authors propose several risk-reduction strategies for listed Vietnamese real estate firms, such as optimizing working capital, improving the net profit margin on total assets, and using financial leverage prudently.
Nguyễn Minh Nhật (2024) conducted a study titled "Predicting Default Risk for Small and Medium Enterprises in Vietnam Using Machine Learning Models" that developed a predictive model to assess default risk for 1,200 SMEs in Vietnam using machine learning methods such as Logistic Regression, Decision Trees, XGBoost, and Artificial Neural Networks (ANN) Data were collected from the financial reports of businesses that borrowed from commercial banks and listed companies on the Vietnamese financial market from 2010 to 2022 Default risk predictions were based on financial indicators including profitability, financial leverage, liquidity, interest payments, long-term debt payments, and operational efficiency Results showed that Decision Trees, XGBoost, and Artificial Neural Networks outperformed Logistic Regression, with the ANN achieving an F1 score of 0.756 and an accuracy of 0.9345, indicating very strong predictive performance.
This research suggests that Artificial Neural Networks (ANN) is a potential model for identifying customers with high default risk, helping to optimize credit risk management processes
Table 2.1: Summary of domestic empirical research
Predict default risk based on the combination of Z-Score approach of Altman and machine learning methods
4693 companies in banking, securities and insurance sectors from
Logistic Regression, Bayesian Network, K- nearest neighbor, Artificial Neural Network (ANN), Support Vector Machine
Logistic algorithms, Support Vector Machine
(SVM), Decision Tree and Artificial Neural
Network achieve a high accuracy of nearly 96 to
(SVM), and Decision Tree based on Z- Score approach of Altman
Network achieve 84- 85% in the accuracy of prediction Đỗ Thị Vân
Figure out the factors impacting on financial risk of real estate corporations
Assess the quantile regression models
48 real estate businesses listed on
Feasible Generalized Least Squares (FGLS) method and the Quantile
An empirical analysis shows that quick liquidity and robust self-financing capacity reduce financial risk across all quantiles, highlighting the protective role of liquidity management Furthermore, accounts receivable turnover is negatively correlated with financial risk at the 0.1, 0.25, 0.5, and 0.9 quantiles, indicating that higher turnover of receivables consistently lowers risk across different points in the risk distribution.
Across the dataset, return on total assets (ROA) is inversely related to financial risk at the 0.1 and 0.25 quantiles, indicating that higher ROA corresponds to lower risk in these extreme ranges Conversely, asset turnover is positively correlated with financial risk at the 0.1 and 0.25 quantiles, suggesting that higher turnover accompanies greater risk in these quantiles Overall, quick liquidity and self-financing capacity emerge as the most closely linked indicators of financial risk for listed real estate companies.
Predicting the bankruptcy risk of listed real
Ho Chi Minh the Z-Score and H-Score models
Predictive accuracy of the Z-Score model estate companies on the Vietnamese stock market
City Stock Exchange and the Hanoi
2017 to 2020 is 99.11%, while the H- Score model achieves an accuracy of 63.84% The main factors influencing the bankruptcy risk of these companies include profit, working capital, and financial leverage
Factors affecting bankruptcy risk of enterprises in real estate industry
55 real estate enterprises listed on two Vietnam stock market HOSE and HNX
The debt-to- asset ratio, return on equity, current ratio, and interest rate all have a positive impact on the default risk of real estate businesses
Conversely, the return on assets, net working capital to total assets ratio, quick ratio, accounts receivable turnover, and macroeconomic indicators GDP have a negative relationship
Machine Learning Based Bankruptcy Prediction of Vietnamese Companies
Logistic Regression, Random Forest, Extreme Gradient Boosting (XGBoost), K- Nearest
XGBoost model was the best at predicting bankruptcy, with the highest accuracy and the fewest mistakes
The top model achieved error rates of 13.86% and 10.81%, while the Random Forest model—the second-best—showed slightly higher errors at 21.78% and 11.71% These results indicate that machine learning methods generally outperform traditional approaches such as logistic regression for predictive accuracy, and boosting methods like XGBoost often deliver the best performance among the tested algorithms.
Forecasting the bankruptcy risk of Vietnamese listed real estate companies through the Z- from 41 real estate companies listed on the Vietnamese stock market from 2019 to
Z-Score model 5 out of the 41 companies faced bankruptcy risk
29 companies from the Ho Chi Minh City Stock
12 from the Hanoi Stock Exchange
Predicting Default Risk for Small and Medium
Enterprises in Vietnam Using Machine
Logistic Regression, Decision Trees, XGBoost, and Artificial
Decision Trees, XGBoost, and Artificial
Neural Networks (ANN) performed better than Logistic
Regression Specifically, the ANN achieved an F1 score of 0.756 and an accuracy of 0.9345 which means it was very good at making predictions
Source: Compiled by the author
According on the previous study Beaver (1966), Altman (1968), Ohlson (1980), Altman (1993) was expanded the innovative models by the application of statistical analysis methods
Beaver (1966) study, "Financial Ratios as Predictors of Failure," was among the first to contribute modern bankruptcy forecasting models by applying univariate analysis to evaluate the predictive ability of financial ratios The research used a dataset of 158 U.S industrial firms from 1954 to 1964, comprising 79 bankrupt and 79 matched non-bankrupt firms, paired by industry and size for exact comparability From an initial set of 30 financial ratios, six were identified as the most informative differentiators between the two groups, selected using classification accuracy and the Gutenberg coefficient The most predictive variable was cash flow to total debt, which correctly predicted 90% of bankruptcies one year before they occurred and 87% up to five years in advance Other key ratios included debt to total assets, net working capital to total assets, current ratio, return on assets (ROA), and accounts receivable to sales The study showed that companies at risk of bankruptcy typically exhibited lower cash flow and weaker liquidity three to five years prior to failure, underscoring the method’s simplicity, practical applicability, and early warning capability, with particular usefulness in cash-flow‑sensitive industries like real estate However, limitations included reliance on univariate analysis, the absence of a specific warning threshold, a relatively small sample, and the exclusion of macroeconomic factors such as interest rates or exchange-rate volatility.
Introduced by Altman in 1968, the Z-Score is a widely used bankruptcy-prediction tool derived from a Multiple Discriminant Analysis of 66 U.S manufacturing firms (33 bankrupt and 33 healthy) from 1946–1965 It identifies five financial ratios—working capital to total assets (short-term liquidity), retained earnings to total assets (profit retention), EBIT to total assets (operational efficiency), market value of equity to total liabilities (financial leverage), and sales to total assets (asset utilization)—and weights them to separate bankrupt from solvent firms The model achieved high accuracy, correctly predicting 95% of bankruptcies one year before and 72% three years in advance However, the original Z-Score is limited to publicly listed manufacturing companies and is not suitable for industries with different financial characteristics such as real estate or services In 1993, Altman introduced the Z''-Score to extend bankruptcy prediction to non-manufacturing firms and emerging markets, based on data from over 120 U.S firms; yet this version still relies on linear methods and omits potential nonlinear relationships as well as non-financial or macroeconomic factors.
The original Z-Score model, published in 1968 in the Journal of Finance to evaluate the bankruptcy risk of manufacturing firms, combines five key financial ratios: working capital over total assets, retained earnings over total assets, earnings before interest and taxes over total assets, market capitalization over total liabilities, and sales over total assets.
The threshold of failure is measured by Altman (1993) following:
- If Z‖>2.99, corporations operate and run effectively The financial health is still in the safe zone
- If 1.81 ≤ Z‖ ≤ 2.99, companies are in the warning area which is called as ―gray zone‖
- If Z‖ < 1.81, enterprises are facing some significant bankruptcy risk and financial distress
The extended Z-Score model, officially published in 1993 in Corporate Financial Distress and Bankruptcy, provides a framework to evaluate bankruptcy risk for non-manufacturing firms It extends the original Z-Score to non-manufacturing contexts by using four ratios: working capital over total assets, retained earnings over total assets, earnings before interest and taxes over total assets, and market capitalization over total liabilities Together these indicators assess liquidity, profitability, asset efficiency, and leverage, offering a concise signal of financial distress for investors and creditors analyzing non-manufacturing companies.
The threshold of failure is measured by Altman (1968) following:
- If Z‖>2.6, corporations operate and run effectively The financial health is still in the safe zone
- If 1.1< Z‖ ≤ 2.6, companies are in the warning area which is called as ―gray zone‖
- If Z‖ ≤ 1.1, there are some ongoing potential bankruptcy risk and financial distress
Ohlson (1980) contributed to the development of a forward-looking bankruptcy risk model in his book Financial Ratios and the Probabilistic Prediction of Bankruptcy The study analyzed 2,163 U.S companies, including 105 failures, from 1970 to 1976 It employed logistic regression to evaluate nine variables: firm size; total liabilities over total assets; working capital over total assets; short-term liabilities over short-term assets; negative owner equity; net income over total assets; funds from operations over total liabilities; two consecutive years of negative net income; and the change in net income The model correctly predicted bankruptcy 96% of the time one to two years before failure.
RESEARCH METHODOLOGY
Research Process
Source: Created by the author
Figure 3.1: The diagram of research process
The process of this study is presented as follows:
Step 1: Examine the research problems
This thesis concentrates on forecasting the bankruptcy risk of real estate enterprises listed on the Vietnam stock market To maintain consistency across the research process, it lays a solid foundation by clearly defining the study's objective, setting the scope, selecting relevant data sources, and establishing the methodological framework that guides the bankruptcy risk forecasting model By focusing on real estate firms within the Vietnamese market, the work aims to produce actionable insights for investors and regulators, using robust statistical techniques and transparent criteria to assess financial distress risk.
Review relevant literature and previous study
Build model and determine research methods
Collect, aggregate and handle the research data
Apply machine learning algorithms and model evaluation
Conclusion, propose some appropriate managerial implications research, research questions and research subject and scope Besides, it provides some expected beneficial contributions
Step 2: Review related literature and previous study
The study examines a wide range of relevant domestic and international studies Based on that, it aggregates theoretical frameworks and conducts a literature review to identify gaps in existing research
Step 3: Build model and determine research methods
According on the found theories from Chapter 2, the thesis builds appropriate model and research methods
Step 4: Collect, aggregate and handle the research data
The dataset is audited and collected from financial statements obtained from a reliable source, specifically the Fiinpro-X platform, covering 215 real estate companies from
Between 2011 and 2024, the study focuses on firms listed on the Vietnam Stock Exchange, excluding those not listed on the main exchange (OTC, private, and UPCOM) After rigorous data cleaning, filtering, and handling of missing values, the final sample comprises 57 companies, representing 26.51% of the original dataset Descriptive statistics and Pearson correlation analyses are conducted on the variables to mitigate multicollinearity and support robust inference.
Step 5: Apply machine learning algorithms and model evaluation
After data preprocessing, the workflow applies machine learning algorithms—Decision Tree Classification, Naive Bayes, Gradient Boosting, and Logistic Regression—implemented in Python on Google Colab The study evaluates model performance using a comprehensive set of metrics, including the confusion matrix, accuracy, precision, recall, F1 score, and AUC, to compare the effectiveness of each algorithm.
This thesis systematically presents the research results by applying and evaluating a range of predictive models for bankruptcy risk in real estate firms, comparing their performance to identify the most effective approach based on empirical results Through a comprehensive model-by-model assessment, the study demonstrates which method best forecasts insolvency risk and under which conditions, offering clear guidance for practitioners evaluating credit risk in the real estate sector.
Step 7: Conclusion, propose some appropriate managerial implications
The study concludes and suggests suitable managerial implications to assist these companies in operating effectively and contributing to the stabilization of the real estate market in Vietnam.
Research Model and hypotheses
Based on empirical evidence and a robust theoretical framework, this thesis utilizes Altman’s Z-Score, the well-known bankruptcy risk classification method (1993), to assess corporate insolvency risks It builds on and integrates findings from classic studies—Beaver (1966), Ohlson (1980), and Altman (1993)—as well as newer contributions by Vân Trang and Dương (2021) and Dũng et al., to refine the understanding of bankruptcy indicators and their predictive performance.
(2022), Máté et al (2023), Thùy and Lê Hải (2023), Anh (2024), Nguyễn Minh Nhật
(2024) with the related topic, the model is developed following:
Dependent variables identifies the enterprises with two values (0,1) ―0‖ stands for ―non-bankrupt firms‖ and ―1‖ is understood ―bankrupt firms‖ based on Altman (1993) through Z-Score
- If Z‖>2.6, corporations operate and run effectively The financial health is still in the ―safe zone‖
- If 1.1< Z‖ ≤ 2.6, companies are in the warning area which is called as ―gray zone‖
- If Z‖ ≤ 1.1, there are some ongoing potential bankruptcy risk and financial distress
The enterprises belong to ―safe zone‖ which are marked as ―0‖ And the rest are considered that exist bankruptcy risk (marked as 1)
Groups Independent variables Empirical evidence Expected results
Liquidity ratio Working capital over total assets (WCAPTA)
(2022), Trương Thị Thùy Dương & Lê Hải Trung
Phan Trần Trung Dũng & et al (2022)
Leverage ratio Total liabilities to total assets ratio (DTA)
(1980), Phan Trần Trung Dũng & et al (2022)
Short term debt to long-term debt ratio (SDR) Đỗ Thị Vân Trang & Phan Thùy Dương (2021), Phan Trần Trung Dũng & et al
Profitability Ratio Return on sales (ROS) Ohlson (1980), Đỗ Thị Vân
(2021), Phan Trần Trung Dũng & et al (2022)
Retained Earnings over Total assets (REAT)
Altman (1993), Bùi Quang Minh (2022), Trương Thị Thùy Dương & Lê Hải Trung (2023), Nguyễn Văn Quang & Mai Tuấn Anh
Inventory turnover (IT) Đỗ Thị Vân Trang & Phan
Thùy Dương (2021), Phan Trần Trung Dũng & et al
Receivables turnover ratio (RT) Đỗ Thị Vân Trang & Phan Thùy Dương (2021), Phan Trần Trung Dũng & et al
Asset turnover ratio (TAT) Đỗ Thị Vân Trang & Phan Thùy Dương (2021), Phan Trần Trung Dũng & et al
(2022), Domician Mate & et al (2023), Nguyễn Minh Nhật & Ngô Hoàng Khánh Duy (2024)
Market value ratio Price to Book (PB) Phan Trần Trung Dũng & et al (2022)
Source: Compiled by the author
Liquidity ratio measures a company's ability to meet its current and short-term debt obligations without raising external capital It assesses the balance of liquid assets—cash, cash equivalents, and accounts receivable—relative to short-term liabilities to show how quickly the firm can cover obligations A fall in liquidity ratios increases the risk of financial distress and potential bankruptcy, highlighting the relevance of liquidity management for financial stability and creditworthiness.
Working capital over total assets (WCAPTA) =
From the traditional working capital management theory (Brigham, 1962), (Beranek,
Maintaining liquidity depends on balancing current assets with current liabilities to keep working capital healthy and support stable operations Seminal studies by Beaver (1966) and Altman (1968) laid the foundation for understanding liquidity's role in bankruptcy risk, a view echoed by later research such as Opler et al (1999) and Thim et al (2011) These works, along with contributions from Vân Trang and Dương, indicate that higher liquidity reduces bankruptcy risk by improving short-term solvency In particular, Opler et al (1999) show that greater liquidity lowers the probability of insolvency by ensuring firms can meet near-term obligations, underscoring the importance of effective working-capital management for financial stability.
(2021), the author suggests the hypothesis 1: the liquidity ratio exerts the negative impact on the bankruptcy risk of real estate companies
Leverage ratio shows how much of a company’s current and short-term liabilities are used to fund its operations It indicates whether the revenue left after operating expenses is sufficient to cover interest expenses A combination of high debt levels and low liquidity signals a higher risk of bankruptcy.
Total liabilities to total assets ratio (DTA) =
Short term debt to long-term debt ratio (SDR) =
According to the capital structure theory of Modigliani and Miller (1958) in a perfect market, capital structure does not affect firm value However, Modigliani (1963) incorporated corporate income taxes and showed that interest expenses create a tax shield that increases firm value, while also raising financial and bankruptcy risks, especially in the real estate sector with volatile EBIT and high debt ratios Bankruptcy prediction models such as the Altman Z-score (1968, 1993) and O-score (1980) both highlight leverage as a key indicator of bankruptcy risk Based on this theory and prior studies Beaver (1966), Ohlson (1980), Dũng et al (2022), the author suggests the hypothesis 2: the leverage ratio exerts the positive impact on the bankruptcy risk of real estate enterprises
Profitability ratio is considered to estimate the ability of creating income over a specific period of time The figure gives me an overall sight whether the enterprises effectively use current resources and expenses to generate earnings
Retained Earnings over Total assets (REAT) =
The model shows that higher profit margin ratios, such as return on sales (ROS) and retained earnings to total assets (REAT), signal stronger profitability and lower bankruptcy risk for real estate enterprises Guided by the theory of Vân Trang and Dương (2021), Dũng et al (2022), and Thim et al (2011), the analysis supports Hypothesis 3: profitability ratios exert a negative impact on the bankruptcy risk of real estate enterprises.
Operational efficiency ratios, also known as activity ratios, are key financial metrics used to assess how efficiently a company manages its resources They focus on how quickly inventory moves, how effectively assets are deployed to generate sales, and how rapidly accounts receivable are converted into cash By measuring metrics such as inventory turnover, asset turnover, and receivables collection, these ratios illuminate management performance, highlight opportunities to improve working capital, and support decisions aimed at boosting profitability and cash flow.
DuPont analysis defines operating efficiency as a firm’s ability to convert assets into revenue Key indicators—total asset turnover (TAT), inventory turnover (IT), and receivables turnover (RT)—capture how effectively assets are utilized Higher values of TAT, IT, and RT signal more efficient asset use, contributing to stronger profitability, better liquidity, and lower bankruptcy risk Building on this framework, Vân Trang and Dương (2021) and Dũng et al (2022) propose Hypothesis 4: the operational efficiency ratio is negatively related to the bankruptcy risk of real estate enterprises.
Market value ratio is a valuable metric for external stakeholders, including investors and market analysts, offering a clear view of a company's relative worth and market position It aids rigorous financial analysis by highlighting valuation, performance, and potential risk in the capital markets For companies, monitoring the market value ratio helps assess their standing in the stock exchange and guides strategies to optimize value per share and overall shareholder value.
Intrinsic value, as Williams (1938) argued, is the present value of expected future cash flows, grounded in discounted cash flow analysis, the time value of money, and risk Fisher (1993) and Damodaran (2012) later extended the model to real estate by incorporating location, taxes, depreciation, and sector-specific risks, enhancing its applicability in volatile markets From this valuation perspective, a high price-to-book (P/B) ratio can signal overvaluation, mispricing risk, or uncertainty about future growth, which can heighten bankruptcy risk for real estate firms Studies by Agarwal and Taffler (2008) and Dũng et al further illuminate how these valuation dynamics play out in property markets.
(2022), the author suggests the hypothesis 5: the market value ratio exerts the positive impact on the bankruptcy risk of real estate corporations
RESEARCH RESULTS AND DISCUSSION
Data Reading and Processing
Read and process data in the Pandas library to detect null or non-null variables Figure 4.1 shows that both dependent and independent variables have values
Figure 4.1: Data Reading and Processing Results Source: Author’s statistics
Figure 4.2: Research sample Source: Author’s statistics
Table 4.1 presents the research data, comprising 798 observations of financial indicators and market indicators derived from secondary data collected from audited consolidated financial statements of 57 real estate enterprises listed on the Vietnam stock exchange, spanning 2011–2024.
Descriptive statistics
An analysis table presents eight independent variables along with key descriptive statistics, including count, mean, median, standard deviation, minimum, and maximum The dataset comprises 798 observations of real estate companies listed on the Vietnam Stock Exchange, providing a robust overview of distribution and central tendency across firms.
Regarding Working Capital over Total Assets (WCAPAT), its minimum and maximum values are -0.369 and 0.999, respectively, indicating that some businesses operate with negative working capital—a sign of bankruptcy risk—while others show a very high WCAPAT For the Total Liabilities to Total Assets (DTA) ratio, the mean, median, minimum, and maximum are 0.499, 0.523, 0, and 0.923, revealing that the average corporation carries about 50% liabilities relative to assets, with some firms exhibiting notably low leverage Many highly leveraged firms expose themselves to higher bankruptcy risk if cash flow worsens The short-term debt to long-term debt ratio (SDR) has a median value of 2.10, meaning over 50% of firms have short-term debt twice as large as long-term debt The quantile results show that 25% of companies have SDR < 0.83, indicating a sustainable debt structure, 50% have SDR ≤ 2.10, and the rest have SDR > 5.87, which may pose significant financial risks due to heavy reliance on short-term debt.
The table shows ALR liquidity with a median of 1.90, a 25th percentile of 1.52, and a 75th percentile of 2.68, suggesting that most businesses maintain an adequate liquidity position Return on Sales (ROS) has a mean of -0.105 and a standard deviation of 3.656, meaning the average ROS is negative and many firms are unprofitable Retained Earnings over Total Assets (REAT) averages 0.046 (SD 0.076), reflecting a low retained earnings ratio on average, while a minimum of -0.81 reveals some firms experience negative retained earnings and accumulated losses Inventory Turnover (IT) shows that 25% of firms have IT below 0.18, signaling a risk of slow-moving inventory, and 50% below 0.455, indicating generally low inventory management efficiency Receivables Turnover (RT) has a minimum value of 0, highlighting that some firms struggle to collect receivables.
TAT, with a mean of 0.247 and a standard deviation of 0.203, indicates low efficiency in using total assets, and poor asset utilization often signals operational inefficiency and potential risk The Price-to-Book Ratio (PB) has a median of 0.91 and a 25th percentile of 0.48, suggesting many firms are undervalued by the market, while the low valuation could be an early warning sign of financial trouble.
Correlation analysis
Figure 4.3: Correlation matrix Source: Author’s statistics
The correlation analysis in the bankruptcy risk estimation model shows that three variables have the strongest associations with the bankruptcy outcome (Y) WCAPAT (Working Capital over Total Assets) has a strong negative correlation (-0.653), meaning that lower working capital relative to assets increases bankruptcy risk and makes it harder to meet short-term debt obligations DTA (Total Liabilities to Total Assets) exhibits a moderately strong positive correlation (0.325), indicating that higher financial leverage raises the likelihood of bankruptcy REAT (Retained Earnings over Total Assets) is negatively correlated (-0.324), so lower profitability is linked to higher bankruptcy risk Together, these relationships underscore the impact of liquidity, leverage, and profitability on bankruptcy risk in the model.
Among the independent variables, the strongest relationship is between REAT and Total Asset Turnover (TAT), with a correlation coefficient of 0.384 All other variable pairs show very low and statistically insignificant correlations Overall, every correlation coefficient stays below 0.8 in absolute value, indicating there is no serious multicollinearity in the model.
Research results
Table 4.2: Confusion Matrix of the Decision Tree Model
Actual class Non-bankruptcy Bankruptcy
Among 160 real estate firms, the Decision Tree classifier correctly identifies 89 non-bankrupt and 54 bankrupt companies Evaluation across Accuracy, Recall, F1 Score, Precision, and Specificity shows an overall accuracy near 90% and a Specificity near 90%, indicating a strong ability to identify non-bankrupt firms The model achieves a recall of 88.52%, correctly flagging 88.52% of truly bankrupt firms, while the F1 score stands at 86.40%, signaling a stable balance between Precision and Recall; however, Precision is notably lower than the other metrics.
(84.38%) That means among 100 firms predicted as bankrupt, 15.62% are false alarms (False Positives)
Table 4.3: Confusion Matrix of the Random Forest Model
Actual class Non-bankruptcy Bankruptcy
The table shows that Random Forest model correctly forecasts 92 non-bankrupt and
Using 55 bankrupt firms, the Random Forest model achieves an overall accuracy of 91.88% and a specificity of 92.93% It correctly predicts 91.88% of the total sample and 92.93% of non-bankrupt firms, while minimizing false positives among healthy companies The model's recall (90.16%) and precision (88.71%) balance to an F1 score of 89.43% The high recall means it detects 90.16% of companies actually at risk of bankruptcy, and the 88.71% precision indicates relatively few false alarms Compared with a Decision Tree, Random Forest provides higher precision (88.71% vs 84.38%), and only 11.29% false alarms, reducing misclassification of non-bankrupt corporations Overall, Random Forest is superior to the Decision Tree model, especially in precision and specificity, for bankruptcy risk assessment.
Table 4.4: Confusion Matrix of the Naive Bayes Model
Actual class Non-bankruptcy Bankruptcy
Among 160 firms, the Naive Bayes model exactly predicts 88 non-bankrupt and 21 bankrupt businesses, delivering a specificity of 88.89% in identifying healthy companies Its overall accuracy is 68.13%, markedly lower than the Random Forest's 91.88% accuracy The classifier's F1-score is 45.16%, signaling a poor balance between precision and recall, with a recall of 34.43%—it misses 65.57% of actual bankruptcies (about 40 of 61 cases) This represents a severe weakness in bankruptcy risk alerts On precision, Naive Bayes averages 65.63% (34.37% of bankruptcy predictions are incorrect, 11 of 32) Overall, the model is unreliable for risk assessment due to the high miss rate.
Table 4.5: Confusion Matrix of the Gradient Boosting Model
The Gradient Boosting model achieves strong overall performance in bankruptcy prediction, misclassifying only 6 bankruptcies (false negatives) and 8 non-bankruptcies (false positives), resulting in an accuracy of 91.25% It maintains a balanced profile with a high F1-score of 88.71% and a recall of 90.16%, meaning it detects about 90% of actual bankruptcies Precision stands at 87.30%, corresponding to 12.7% false alarms Additionally, the model shows strong specificity of 91.92% in correctly identifying healthy companies.
Table 4.6: Confusion Matrix of the Logistic Regression Model
Using Python, a confusion matrix for a Logistic Regression model was built to evaluate bankruptcy prediction on 160 cases The model correctly classified 95 non-bankrupt and 55 bankrupt firms, achieving an overall accuracy of 93.75% With a precision of 93.22%, only 4 of 59 bankruptcy warnings were incorrect, and specificity reached 95.96% The F1-score was 91.67%, indicating a strong balance between precision and the other metrics and making it suitable for both risk warning and investment decision-making Moreover, logistic regression is easier to explain than gradient boosting and random forest models, highlighting its superiority in precision and specificity.
Models comparison
Logistic Regression emerges as the top performer for bankruptcy prediction, delivering an accuracy of 93.75%, precision of 93.22%, recall of 90.16%, and specificity of 95.96%, effectively minimizing false alarms Naive Bayes registers the weakest results, with accuracy 68.13% and a recall of 34.43%, missing 65.57% of bankrupt cases Random Forest and Gradient Boosting produce similar outcomes, with Random Forest slightly ahead in the F1-Score at 89.43% compared to Gradient Boosting at 88.71% Based on recall, Logistic Regression, Random Forest, and Gradient Boosting all reach 90.16%, indicating roughly 90% detection of actual bankrupt businesses In terms of precision, Logistic Regression achieves 93.22% (about 6.78% false alarms), while the Decision Tree model yields 15.62% incorrect bankruptcy predictions Considering the balance between precision and recall reflected by the F1-Score, Logistic Regression offers the best overall trade-off.
(91.67%) but Nạve Bayes is severely imbalanced (45.16%) With Specificity, Logistic Regression is the most accurate at identifying healthy businesses (95.96%) whereas Decision Tree model is the worst at 89.90%
Figure 4.4: Receiver operating characteristic ROC curve Source: Author’s statistics
The figure shows the AUC performance of four classification models as the True Positive Rate (TPR) and False Positive Rate (FPR) vary Logistic Regression achieves the highest accuracy at 98.33%, followed by Random Forest at 97.97%, Gradient Boosting at 97.88%, and Decision Tree at 97.78%, with Naive Bayes significantly lower at 62.03%.
Through the findings from Chapter 4 It presents that the Logistic Regression (LR) is the most effective on Four parameters achieve the best such as: Accuracy, Recall, Specificity, Precision, AUC
*The results of research model:
Figure 4.5: Confusion Matrix of the Decision Tree Model Source: Author’s statistics
The logistic model achieves an accuracy of nearly 93.75% and records the highest scores across Precision, Recall, Specificity, F1-score, and AUC, underscoring its strong capability to predict bankruptcy risk.
Regarding type I and type II errors, the results show that the model has a smaller type
II error (5.94%) than type I error (6.78%), which will not cause great damage to the business if the forecast is wrong
Therefore, this model is suitable for predicting the bankruptcy risk for real estate businesses.
Discussion of research results
Table 4.8: Summary of research results
Liquidity ratio Working capital over total assets (WCAPTA)
Leverage ratio Total liabilities to total assets ratio (DTA)
Short term debt to long-term
0,0000385 Positive (+) Accept debt ratio (SDR)
Retained Earnings over Total assets (REAT)
Hypothesis 1: the liquidity ratio exerts the negative impact on the bankruptcy risk of real estate companies
Working capital over total assets (WCAPTA) and Total Liquidity Ratios (ALR) show a negative relationship with bankruptcy risk (Y), with regression coefficients of approximately -13.2646 for WCAPTA and -0.000213 for ALR Put differently, weaker WCAPTA and lower ALR amplify the bankruptcy risk for these companies The real estate sector requires substantial working capital to sustain operations while awaiting funds from long-term projects, so when WCAPTA and ALR are low, firms are more likely to struggle with meeting short-term debt obligations, especially during market downturns when unsold properties are hard to dispose of Empirical studies (Altman, 1968; Beaver, 1966; Thim et al., 2011; Van Trang & Duong, 2021) support the view that liquidity ratios adversely affect the bankruptcy risk of real estate companies, consistent with Hypothesis 1 outlined in Chapter 2 of the thesis.
Hypothesis 2: the leverage ratio exerts the positive impact on the bankruptcy risk of real estate enterprises
The effects of the total liabilities to total assets ratio (DTA) and the short-term debt to long-term debt ratio (SDR) on bankruptcy risk are positive, with regression coefficients of 5.5517 and 0.0000385 respectively, indicating a positive relationship between these leverage measures and bankruptcy risk for real estate firms In other words, as DTA and SDR rise, the likelihood of bankruptcy increases, reflecting higher debt burdens When a real estate company takes on excessive debt (high DTA), especially via short-term borrowings (high SDR), it faces heavier repayment pressure This risk is amplified because real estate assets cannot be sold quickly to generate cash for repayment, a danger that grows during market downturns This perspective is supported by several empirical studies, including Beaver (1966), Ohlson (1980), and recent work by Dũng et al (2022), despite differences in samples and measurement of leverage ratios and bankruptcy risk.
Hypothesis 3: the profitability ratio exerts the negative impact on the bankruptcy risk of real estate enterprises
The findings show that Return on Sales (ROS) and Retained Earnings over Total Assets (REAT) negatively affect Y, with regression coefficients of -0.04189 and -8.3402, respectively Specifically, lower ROS and lower REAT are associated with higher bankruptcy risk for these firms, since reduced profitability and weaker capital buffers diminish the financial cushion needed to endure crises This is especially critical in the real estate industry, which hinges on substantial and stable financial resources Moreover, the results support Hypothesis 3 These conclusions align with prior studies, including Van Trang & Dương (2021), Dũng et al (2022), and Thim et al (2011).
Hypothesis 4: the operational efficiency ratio exerts the negative impact on the bankruptcy risk of real estate enterprises
Inventory turnover (IT), receivables turnover ratio (RT), and asset turnover ratio (TAT) have a negative effect on Y, with regression coefficients of -0.001354, -0.004893, and -1.4988, respectively In particular, lower IT, RT, and TAT values correspond to a higher bankruptcy risk for these real estate companies When turnover ratios are low, it signals difficulty in converting assets into cash, leading to cash flow shortages that constrain ongoing operations and the ability to meet debt obligations This result supports Hypothesis 4 discussed in Chapter 2 and aligns with empirical findings from Vân Trang & Dương (2021) and Dũng et al.
Hypothesis 5: the market value ratio exerts the positive impact on the bankruptcy risk of real estate corporations
Regression results show Price to Book (PB) has a negative relationship with bankruptcy risk (Y), with a regression coefficient of -0.78298, meaning higher PB is associated with lower bankruptcy risk This supports hypothesis 5 as outlined in Chapter 2 By contrast, studies such as Agarwal and Taffler (2008) and Dũng et al (2022) report that the market value ratio positively affects bankruptcy risk for real estate firms, highlighting divergent evidence in the literature Real estate book values often reflect historical cost and are recorded at original cost rather than current market prices, which better capture income potential from rents, location premiums, and appreciation over time due to location and planning Additionally, firms with high PB can more easily raise capital through equity offerings during downturns without triggering financial distress.
Regression results identify five variables with the strongest impact on Y, the bankruptcy risk of real estate businesses Specifically, working capital over total assets (WCAPTA) reduces risk with a coefficient of -13.26, and retained earnings over total assets (REAT) reduces risk with a coefficient of -8.34 Increases in total liabilities over total assets (DTA) raise risk, with a coefficient of +5.55 Asset turnover (TAT) and price-to-book (PB) also show moderate risk-reducing effects, with coefficients of -1.50 and -0.78, respectively.