Forecasting the Global Aviation Industry Index with Machine Learning Methods (Economic Analysis)

Summary

• XGBoost reduces RMSE by approximately 21.8% compared to Random Forest and by 64.5% compared to Facebook Prophet when forecasting the U.S. Global Jets ETF (JETS).p.1
• The dataset consists of 1,016 trading-day observations of daily JETS closing prices covering the post-COVID-19 recovery phase.p.5
• XGBoost achieves the strongest predictive performance, with RMSE = 0.3715 and MAPE = 1.31%.p.9
• Random Forest shows stable predictive behavior, with a relatively low MAPE of 1.66%.p.8
• Facebook Prophet was run with a forecasting horizon of 120 future periods using its standard configuration.p.9
• Facebook Prophet shows weaknesses in capturing nonlinear market dynamics during structurally sensitive periods, while Random Forest performs consistently and competitively.p.1

Key insights

1. XGBoost lowers RMSE by approximately 21.8% versus Random Forest and by 64.5% versus Facebook Prophet.p.1
2. XGBoost achieves the lowest error values across all three evaluation metrics (MSE, RMSE and MAPE).p.8
3. Random Forest delivers competitive accuracy with slightly higher errors than XGBoost, and a MAPE of 1.66%.p.8
4. XGBoost reaches RMSE = 0.3715 and MAPE = 1.31%, which the authors attribute to sequential error correction in boosting.p.9
5. Ensemble tree-based learning methods outperform additive time-series modeling in forecasting JETS during structurally volatile periods.p.8
6. The study uses only daily closing prices in a univariate framework, with 1,016 trading-day observations and no missing values.p.5
7. Prophet's accuracy remains inferior to tree-based ensemble methods even though it captures structural temporal patterns.p.10