The Integration of Artificial Intelligence in Economic Modelling: A Comparative Study of Traditional and AI-Based Forecasting

Abstract:

Economic forecasting has traditionally relied on econometric and time-series models such as ARIMA, VAR, and DSGE, which offer interpretability and theoretical grounding but face limitations in capturing nonlinear dynamics, structural breaks, and the increasing volatility of modern economies. Recent advances in artificial intelligence (AI), including machine learning, deep learning, and hybrid techniques, have introduced new possibilities for improving predictive performance and adaptability. This study presents a comparative analysis of traditional econometric models and AI-based forecasting approaches neural networks, ensemble learning, and reinforcement algorithms, evaluating their predictive accuracy, robustness to economic shocks, and applicability across key macroeconomic indicators. Using historical data from advanced and emerging economies, model performance is assessed through RMSE, MAE, and Theil’s U. The findings show that AI-driven models outperform conventional methods in capturing nonlinearities and responding to high-frequency shocks, especially during periods of economic uncertainty. Nevertheless, challenges related to interpretability, data demands, and overfitting persist. The study concludes that hybrid models combining structural interpretability with adaptive learning capabilities offer a promising direction for improving forecasting accuracy and enhancing the practical value of economic policy analysis