Comparative analysis of the forecasting ability of AI and Arima models for the Vn-index

Abstract

This study compares the forecasting performance of a traditional econometric model (ARIMA) and artificial intelligence (AI)-based models, namely Multilayer Perceptron (MLP) and Extreme Gradient Boosting (XGBoost), in predicting the VN-Index during the period from 2015 to June 2025, which was characterized by heightened volatility in Vietnam’s stock market. Daily VN-Index closing prices were employed and divided into an 80% training set and a 20% testing set for out-of-sample evaluation. Forecast accuracy was assessed using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). The Diebold–Mariano test was further applied to examine the statistical significance of differences in predictive performance among the models. The results indicate that ARIMA produced the highest forecasting errors, reflecting its limitations in capturing nonlinear dynamics and market volatility. The MLP model significantly improved forecasting accuracy, while XGBoost achieved the lowest error values across all evaluation metrics, demonstrating superior performance in handling noisy and volatile financial time series. AI-based models, particularly XGBoost, outperform the traditional ARIMA model in forecasting the VN-Index during volatile periods. The findings provide useful insights for investors and financial analysts by highlighting the effectiveness of advanced machine learning models in improving short-term market forecasting and investment decision-making in emerging markets.