Forecasting oil commodity spot price in a data-rich environment.

Area of Science:

• Econometrics
• Time Series Analysis
• Machine Learning

Background:

• Time series forecasting is challenged by statistically varying properties.
• Accurate crude oil price prediction is crucial for market participants.

Purpose of the Study:

• To propose a Change Point-Adaptive Recurrent Neural Network (CP-ADARNN) framework for enhanced crude oil price forecasting.
• To address the challenge of time-varying statistical properties in financial time series data.

Main Methods:

• Structural breaks in predictors were identified using change point detection techniques.
• An adaptive recurrent neural network (ADARNN) model was trained for prediction.
• The framework utilized 310 high-dimensional economic series as exogenous factors from 1993 to 2021.

Main Results:

• CP-ADARNN demonstrated superior performance in predicting WTI crude oil prices, outperforming benchmarks by 12.5% in root mean square error.
• A high correlation of 0.706 was achieved between predicted and actual WTI crude oil returns.
• The model's effectiveness was robust for Brent oil prices and during the COVID-19 pandemic.

Conclusions:

• The CP-ADARNN framework offers a significant advancement in crude oil price forecasting.
• The findings provide valuable insights for investors and researchers in the global oil market.
• The method effectively handles structural breaks and high-dimensional data for improved predictive accuracy.