Seasonal inflation dynamics: Insights from probabilistic and hybrid forecasting models

Abstract

Inflation significantly influences economic growth and policy decisions, making accurate forecasting essential. This study analyzes seasonal inflation patterns using monthly data covering the period 2003–2024. It also evaluates a wide range of forecasting methods, including autoregressive integrated moving average (ARIMA), artificial neural networks (ANNs), exponential smoothing techniques (Brown and Holt–Winters), BATS (Box-Cox transformation ARMA errorsTrend), TBATS (Trigonometric seasonality Box-Cox transformation ARMA errorsTrend, and Seasonal components), and hybrid models, in addition to probabilistic distributions such as the Weibull, exponentiated Weibull, Kumaraswamy Weibull, exponential, and inverse Weibull. Results show that ANNs achieved the highest predictive accuracy, while the Weibull distribution best captured seasonal dynamics. The 10-month forecast horizon, extending from June 2024 to March 2025, reveals a stable yet gradually rising inflation pattern. The findings demonstrate the superior forecasting accuracy of ANN-based models and underscore the study’s unique integration of machine-learning techniques with probabilistic models to improve inflation-forecasting performance and support informed economic policy.