ARIMA vs. ETS for RMG Export Forecasting of Bangladesh: A Comparative Study on Model Accuracy

Abstract

This study conducts a comparative analysis of ARIMA and ETS models to forecast Bangladesh RMG export trends. Utilizing annual RMG export data from 1983 to 2023, sourced from BGMEA’s website, the research applies a logarithmic transformation to stabilize variance, with differencing exclusively for ARIMA to address non-stationarity. Automated model selection via R’s functions – auto.arima() for ARIMA and ets() for ETS – was implemented to optimize parameter configurations. The optimal ARIMA (2,2,0) and ETS (A, Ad, N) models was trained on data from 1983 to 2015 and validated on the 2016-2023 testing subset. Accuracy metrics, revealed ETS’s superior performance, yielding a lower MAPE (8.47% vs 19.74%) and RMSE (3794.72 vs 8334.12) compared to ARIMA’s. The Diebold-Mariano test confirmed ETS’s statistical superiority at a 15% significance level. The ETS’s adaptability to non-linear trends and damped volatility in RMG underscore its efficacy, while ARIMA’s reliance on linear assumptions limited its applicability. Forecast for 2024-2028 project sustained RMG export growth, emphasizing sector’s economic resilience. These findings advocate for policymakers to inform strategic planning, while highlighting the need for future research integrating external factors through hybrid or machine learning models.

Dhaka Univ. J. Sci. 73(2): 101-105, 2025 (July)