Corporate Financial Distress Prediction in Vietnam Using Calibrated and Explainable Machine Learning

FORECASTING CORPORATE CASH HOLDINGS IN VIETNAM: COMPARING MACHINE LEARNING MODELS AND THE IMPACT OF THE WORLD UNCERTAINTY INDEX

This study forecasts the cash holdings (CASH) of listed companies in Vietnam based on data from 2018-2025 using a machine learning approach. Using panel data with 4,752 observations, the study compares seven algorithms and identifies HistGradientBoosting as the optimal model (R2 approx. 0.7145), far surpassing traditional linear regression (R2 approx. 0.28). This result demonstrates strong non-linearity in liquidity management behavior in the Vietnamese market. Using SHAP and Permutation Importance, the study successfully elucidates the underlying predictive mechanisms. The main drivers are net working capital (NWC) and the current ratio (CR). Notably, short-term debt (STD), interest expense (IE), and the World Uncertainty Index for Vietnam (WUIVN) consistently showed positive effects, strongly reinforcing the Precautionary Motive. The study also found a moderating effect of size, indicating that small businesses are more sensitive and vulnerable to systemic shocks. The application of machine learning models not only improves forecasting accuracy but also provides a transparent view of corporate financial strategies in an uncertain economic environment.

Do Global Uncertainty Indices Predict Corporate Bankruptcy Risk? Evidence from Indonesian Listed Firms

Background: Increasing global shocks threaten corporate financial stability, particularly in emerging markets. This study measures uncertainty using the World Uncertainty Index (WUI), World Pandemic Uncertainty Index (WPUI), Climate Policy Uncertainty (CPU), and Economic Policy Uncertainty (EPU) across current, lag-1, and lag-2 periods. Bankruptcy risk is proxied by the Altman Z-Score and Distance-to-Default (DTD). Objective: This study examines the effect of global uncertainty on the bankruptcy risk of companies listed on the Indonesia Stock Exchange. Methods: Secondary data from 2000–2024 were analyzed using fixed-effects panel regression in STATA 17. Purposive sampling produced 9,377 firm-year observations for the Z-Score model and 9,156 for the DTD model. Results: EPU and CPU significantly reduced Z-Scores across all time specifications, indicating increased financial distress through an aggravation effect. WPUI had a significant negative effect at lag 2, while WUI was insignificant. In the DTD model, only current-period EPU had a significant negative effect. State-owned enterprises and the Commodity and Cyclical sectors were most vulnerable, whereas the Defensive sector was most resilient. Conclusion: Global uncertainty affects bankruptcy risk heterogeneously. The findings support sector-specific risk mitigation, strategic portfolio allocation, and stronger corporate financial resilience.

Predicting firm profitability in post-COVID India: A machine learning approach using fundamental financial ratios

FINANCIAL DISTRESS PREDICTION USING THE DECISION TREE METHOD IN MANUFACTURING COMPANIES IN INDONESIA

This study aims to predict financial distress in Indonesian manufacturing companies using the Decision Tree method. A quantitative predictive design was applied to secondary data from the annual financial statements of manufacturing companies listed on the Indonesia Stock Exchange during 2020–2024, yielding 612 firm-year observations. Financial distress was measured as a binary outcome (financially distressed versus non-financially distressed), with Current Ratio, Debt to Asset Ratio, Return on Assets, Total Asset Turnover, Sales Growth, and Firm Size as predictors. Using Python, a pruned Decision Tree achieved 84.6% accuracy, 64.7% precision, 75.9% recall, and a 69.8% F1-score for the financially distressed class. Return on Assets was the most influential predictor, followed by Debt to Asset Ratio and Current Ratio. The resulting rules show that distress reflects interacting conditions of weak profitability, high leverage, low liquidity, inefficient asset utilization, and declining sales growth. Theoretically, the study extends financial distress prediction research by demonstrating the value of interpretable machine learning in an emerging-market manufacturing context. Practically, its transparent rules provide an actionable early-warning tool for investors, creditors, managers, and regulators to identify financial vulnerability and support timely intervention.

Predicting Financial Distress: A Comparative Analysis of Explainable Machine Learning Models for Cross‐Market Bankruptcy Prediction

Despite substantial advances in machine‐learning approaches to bankruptcy prediction, less is known about whether explainable models remain reliable across markets and during structural breaks. This study examines that issue by comparing two explainable architectures, TabNet–SHAP and NGBoost–SHAP, for 1‐year‐ahead bankruptcy prediction in the United States and China during 2015–2022. Using a balanced panel of listed firms from both markets, we evaluate out‐of‐sample performance before and after COVID‐19 and benchmark both models against conventional linear and ensemble classifiers. Model uncertainty is assessed through cross‐validation and bootstrap confidence intervals. The results show that NGBoost–SHAP provides more stable and better‐calibrated probabilistic predictions across regimes, whereas TabNet–SHAP produces more variable results and is more sensitive to regime shifts and preprocessing choices. We interpret this pattern as an architecture‐specific trade‐off between stability and adaptability rather than as an unconditional ranking of the two models. SHAP‐based interpretation indicates that profitability and capital‐structure variables account for most of the explanatory signal within each market and model. The post‐COVID Chinese subsample shows the most concentrated feature‐attribution structure, although SHAP magnitudes are interpreted only as within‐model rankings rather than as directly comparable cross‐market quantities. The study contributes by proposing an adaptability–stability–interpretability framework for selecting bankruptcy prediction models under different institutional and market conditions. It also provides evidence that explainable ensemble and deep tabular models can recover economically meaningful distress signals across markets while differing in their calibration, stability, and sensitivity to economic regime change.

Shariah Banking Stock Prices: Expected Return, Risk Level and Macroeconomics Conditions In Indonesia

The present study evaluates how Expected Return (ERT), Risk Level (RLV), alongside Macroeconomic Conditions (MCN) impact the Shariah Banking Stock Price (SBS). The scope focuses specifically on Islamic banking institutions actively listed on the Indonesia Stock Exchange (IDX) between October 2021 and August 2024. All shariah-compliant banks featured on the IDX constituted the target population. To select the final subjects, a purposive sampling approach was applied. The core inclusion criteria required these firms to maintain a consistent listing status and provide uninterrupted data throughout the designated timeframe. Consequently, four specific entities met these parameters: BRIS, BTPS, BANK, and PNBS. Given the 35-month timeline, the empirical testing processed a final dataset comprising 140 distinct observations. Analytical procedures relied on SPSS software to conduct descriptive statistics and classical assumption evaluations—specifically assessing normality, multicollinearity, heteroscedasticity, and autocorrelation. Following this, hypothesis verification was carried out via simultaneous (F-statistic) and partial (t-statistic) assessments, complemented by the coefficient of determination (R²). Empirical findings reveal that, collectively, the three independent components (ERT, RLV, and MCN) significantly drive SBS movements, evidenced by an F-value of 64.321 at a 0.000 significance threshold. When evaluated individually, only Macroeconomic Conditions (MCN) demonstrated a distinct, adverse effect on the stock prices (t-value = -8.290, p = 0.000). Conversely, neither Expected Return nor Risk Level showed any statistically meaningful impact on the dependent variable. The generated Adjusted R² stood at 0.741. This metric implies that the formulated predictors account for exactly 74.1% of the fluctuations observed in the shariah stock prices, while unidentified external factors outside this framework are responsible for the remaining 25.9% variance. Moving forward, subsequent investigations could benefit from extending the observational timeframe. Scholars are encouraged to incorporate alternative macroeconomic indicators—such as fluctuating interest rates or inflation metrics—and potentially broaden the sampling frame to encompass conventional banking or entirely different industries to yield more comprehensive insights.