“One Out of Many” consolidating a long-term trend forecast for investing in energy commodities

Abstract

Accurately predicting the unobservable long-term trend of energy commodity prices is vital for economic and financial stability, yet single forecasts are often unreliable. This study proposes and evaluates a combined forecasting approach to consolidate a single, robust trend estimate for Crude Oil (CL) prices. First, we generate five individual trend estimates using diverse methods: a Cubic Polynomial, Symmetric Moving Average, the Hodrick-Prescott Filter, the Maximum Overlap Discrete Wavelet Transform (MODWT), and the Kalman filter estimation of the Schwartz and Smith (2000) model. Second, we combine these estimates using seven distinct algorithms: simple average, median, weighted average, Multi-Linear Regression (MLR), Principal Components Analysis (PCA), a Feed-Forward Neural Network (FFNN), and a novel modified Multi-Ensemble Time-Scale (METS 2.0) algorithm designed to filter high-frequency noise. We find that combined forecasts consistently outperform individual estimates, with METS 2.0 algorithm demonstrating the best performance by minimizing errors and remaining stable across volatile market conditions. This is validated by long-term investment strategies, where trend-guided approaches significantly outperform those based on price forecasts. This key result confirms the value of isolating the underlying trend for long-term decisions.