Global inflation forecasting and Uncertainty Assessment: Comparing ARIMA with advanced machine learning

Abstract

This study analyzes the effectiveness of two techniques for forecasting global inflation rates. The first is the Autoregressive Integrated Moving Average, and the second is the gradient-boosted regression based on machine learning. It is worth noting that the study introduces the Gradient Boosting for univariate time series analysis. The results reveal that the Autoregressive Integrated Moving Average performs better than the cross-validation using the Autoregressive Integrated Moving Average, with the root mean square coefficient of 2.53, the mean average moving average error of 2.21, and the mean average moving average error of 0.48. Conversely, the gradient-boosted regression outperforms in testing on train and test datasets, achieving the root mean square coefficient of 0.078, the mean average moving average error of 0.27, and the mean average moving average error of 0.22, highlighting its potential for predictive tasks. The study concentrates on short-term forecasts of global inflation rates, thereby minimizing exposure to long-term macroeconomic risks (political and economic shocks). Both models expect global inflation rates to remain stable or decline from 2023 to 2025, providing stability in decision-making among stakeholders such as consumers and producers.