Forecasting ROA and ROE for Retail Companies in Vietnam by Using Machine Learning Techniques

[Purpose] This study aims to forecast the financial performance of Vietnamese retail companies by predicting both Return on Assets (ROA) and Return on Equity (ROE) by using artificial intelligence (AI) models, thereby enhancing predictive decision-making in the retail sector. [Design/methodology/approach] Financial statements from publicly listed retail firms covering the period 2010 to 2024, together with macroeconomic variables (CPI, exchange rate, gold price, VN-Index, oil prices), are combined. Three machine learning algorithms—Random Forest, XGBoost, and Multilayer Perceptron—are trained and tested by using an 80/20 split. Model accuracy is assessed by using RMSE, MAE, MAPE, Pearson’s R, and Theil’s U. [Findings] Random Forest achieved the lowest RMSE of 0.0926 and delivered the highest forecasting accuracy for both ROA and ROE, followed by XGBoost, while MLP performed less effectively. Decision tree–based ensemble models better capture non-linear relationships in financial data than neural networks in this context. [Research limitations/implications] The study covers only listed retail firms and annual data, omitting private firms and intra-year fluctuations. The models identify associations, not causality. [Practical Implications] From a Decision-Sciences perspective, the proposed AI-driven forecasting framework operates as a data-rich decision-support system that transforms multi-source information into actionable guidance on capital allocation, risk mitigation, and performance control. [Originality/value] This is the first study to jointly forecast both ROA and ROE for Vietnamese retailers using three complementary ML families, integrating firm-level and macro-financial variables, and providing reproducible benchmarks for AI-driven financial forecasting in emerging markets.