A Comparative analysis of Arima and Exponential smoothing models in forecasting the Zimbabwe’s terms of trade.

Abstract

This study conducted a comparative analysis of Autoregressive Integrated Moving Average (ARIMA) and Exponential Smoothing (ETS) models for forecasting Zimbabwe's Terms of Trade (TOT) using monthly data from January 2020 to December 2024 (60 observations). The data were split into training (80%, 2020-2023) and testing (20%, 2024) sets. Diagnostic tests revealed that the TOT series was non-stationary (ADF p=0.2068), requiring first differencing (d=1), with moderate volatility (CV=14.41%) and heavier tails than normal (kurtosis=2.69). The ARIMA (0,1,1) model achieved a MAPE of 8.90%, RMSE of 12.60, and MAE of 9.63, while the ETS(A,N,N) model achieved a MAPE of 8.92%, RMSE of 12.62, and MAE of 9.66. The negligible differences across all metrics indicate that both models produce statistically equivalent and highly accurate forecasts (MAPE < 9%). Despite parameter instability (CV=0.15) and the April 2024 currency redenomination structural break, both models maintained high accuracy and outperformed a Random Forest alternative (MAPE=10.04%). Diagnostic tests confirmed no residual autocorrelation (Ljung-Box p=0.527) and constant variance (Breusch-Pagan p=0.860), though slight non-normality was detected (Shapiro-Wilk p=0.047). The 2025 forecast predicts stabilization of TOT around 106-110 index points. The study concludes that both models are equally suitable for forecasting Zimbabwe's TOT, with ETS recommended for its simplicity and ARIMA for its traditional statistical framework. These findings provide evidence-based guidance for policymakers at the Reserve Bank of Zimbabwe and the Ministry of Finance