AI-Enhanced Inflation Forecasting in Emerging Economies

Abstract

This study examines an AI-enhanced framework for forecasting inflation in Ghana and addresses the limitations of traditional models like ARIMA and VAR in capturing nonlinear and volatile pricing dynamics. The study uses monthly data from January 2010 to September 2025 and sourced data from the Ghana Statistical Service, Bank of Ghana, World Bank and IMF. The study compares traditional models like Vector Autoregression (VAR) and ARIMA against AI models such as Random Forest, XGBoost and AdaBoost and Long Short-Term Memory. Model accuracy was assessed using MAE, RMSE, MAPE and SMAPE. The results show that AI models, particularly XGBoost (MAE = 0.12, RMSE = 0.17) and Random Forest (MAE = 0.57, RMSE = 1.45), achieved the lowest forecasting errors and produced more stable and realistic inflation trajectories than traditional models such as VAR (MAE = 1.29, RMSE = 1.78) and ARIMA (MAE = 1.87, RMSE = 4.08). The findings indicated that incorporating global factors such as oil prices and food indices increased model accuracy. These findings highlight the importance of external shocks in Ghana’s inflation behaviour, as well as the potential for artificial intelligence to improve forecast accuracy. The study suggests that incorporating AI-driven models into the Bank of Ghana’s inflation-targeting framework could boost early warning systems and policy effectiveness.