A TIME-ORDERED APPROACH TO PROFITABILITY INDEXING AND PREDICTION: EVIDENCE FROM VIETNAM
Tuyen Le Nam, Tam Phan Huy
University of Economics and Law, Ho Chi Minh City, Vietnam and Vietnam National University, Ho Chi Minh City, Vietnam
Abstract: This study revisits a common issue in corporate finance measurement and forecasting: profitability is fundamentally multi-dimensional, yet research and practical screening often rely on single ratios that may deliver incomplete or even conflicting signals. To address this, the paper constructs an objective composite profitability index by combining correlated profitability proxies, and then tests whether a firm’s profitability class in year t+1 can be predicted using only information observed in year t within a strictly time-ordered design that guards against information leakage. The analysis uses firm-year financial statement data for Vietnamese listed non-financial firms on HOSE and HNX from 1998–2025, provided by the Institute for Development and Research in Banking Technology (Vietnam National University, Ho Chi Minh City). The empirical pipeline applies outlier treatment and feature scaling solely on the training sample, builds the composite index through principal component analysis (PCA), and performs out-of-sample classification with nonlinear machine learning models. Training is restricted to targets up to 2020, and performance is evaluated on a holdout window spanning 2021–2024. Results indicate that the profitability proxies share sufficient common structure to yield a stable composite measure, and that next-year profitability status remains meaningfully predictable beyond the training period. In the baseline specification, an RBF-kernel SVM achieves the highest F1-score (0.7635), while ensemble models deliver the strongest AUC (Random Forest 0.8788; XGBoost 0.8779). Overall, the paper proposes a practical end-to-end framework that links measurement to prediction and argues that composite profitability indicators—implemented with leakage-safe preprocessing—should be considered a minimum standard for credible forecasting, screening, and monitoring.
VOLUME 10 ISSUE 03 2026: 14 – 48