Novel BetaSutte for forecasting: Indonesian export prediction compared to random forest, XGBoost, and ETS

Purpose This study aims to evaluate the efficacy of the novel BetaSutte forecasting methodology for enhancing export management decision-making in Indonesia, benchmarking its performance against established machine learning and statistical approaches, including random forest, eXtreme gradient boosting (XGBoost), and ETS (Error, Trend, Seasonality) exponential smoothing models. Design/methodology/approach Monthly Indonesian export data spanning January 2022 through September 2024 was analyzed using a methodologically rigorous framework. The data set was partitioned into training (22 months) and testing segments (11 months). The BetaSutte model, which explicitly decomposes time series into trend and residual components, was implemented alongside random forest, XGBoost, and ETS models. Performance was evaluated through multiple accuracy metrics, including mean absolute percentage error (MAPE), root mean square error and mean absolute error, with particular attention to forecast stability across different time horizons. Findings The BetaSutte model demonstrated superior forecasting accuracy (MAPE: 7.41%) compared to random forest (MAPE: 9.32%), ETS (MAPE: 11.32%) and XGBoost (MAPE: 16.73%). While machine learning models exhibited strong in-sample performance, the BetaSutte approach yielded more reliable out-of-sample predictions. Analysis of volatile periods revealed that the BetaSutte methodology was particularly effective at capturing sudden directional changes in export values, offering enhanced predictive capabilities during periods of market instability when forecasting accuracy is most crucial for trade planning. Practical implications For trade policy practitioners and export-oriented businesses in Indonesia, this research provides actionable guidance on selecting appropriate forecasting methodologies based on performance characteristics and decision contexts. The superior performance of the BetaSutte model during volatile periods offers particular value for strategic planning in uncertain economic environments. Originality/value To the best of the authors’ knowledge, this study represents the first comprehensive comparison of the BetaSutte methodology against both machines learning and statistical approaches for export forecasting in Indonesia. The findings contribute to forecasting literature by demonstrating the value of explicit trend-residual decomposition in time series prediction while offering practical insights for enhancing decision-making in trade management through improved forecasting precision.