Improving Energy Price Forecasts: Autometrics With Indicator Saturation
Journal of Forecasting · 2026-11-11
DOI: 10.1002/for.70239 ↗At a glance
Journal article in Journal of Forecasting, published online on 2026-11-11. Written by Manish Aggarwal and 2 co-authors. 42 references.
Abstract (original language)
In this study, we evaluate Autometrics, an automated model selection method with controls for outliers and location shifts, against artificial neural networks (ANN) and random forests for six monthly energy price series (WTI, Brent, Dubai, US and Russian gas, Australian coal) from IMF data. Autometrics generally delivers lower RMSE/MAPE and higher directional accuracy than the ANN and random forest specifications considered here, particularly in the more volatile oil and coal markets. Autometrics also has the benefit of being interpretable in a familiar regression context as opposed to more “blackbox” machine learning. The CARDt method often delivers predictive accuracy which is not statistically different from Autometrics, but there is some evidence that simple model averaging of the two with intercept‐corrected Autometrics yields further gains relative to either alone.
Cite
APA
Aggarwal, M., Aggarwal, S., & Stillwagon, J. R. (2026). Improving Energy Price Forecasts: Autometrics With Indicator Saturation. Journal of Forecasting. https://doi.org/10.1002/for.70239
Vancouver
Aggarwal M, Aggarwal S, Stillwagon JR. Improving Energy Price Forecasts: Autometrics With Indicator Saturation. Journal of Forecasting. 2026. doi:10.1002/for.70239
BibTeX
@article{Aggarwal2026,
author = {Aggarwal, Manish and Aggarwal, Shipra and Stillwagon, Josh R.},
title = {Improving Energy Price Forecasts: Autometrics With Indicator Saturation},
journal = {Journal of Forecasting},
year = {2026},
doi = {10.1002/for.70239}
}
EUVIMED