Please rotate your device to landscape mode to view the charts.

Do global COVOL and geopolitical risks affect clean energy prices? Evidence from explainable artificial intelligence models

Journal: Energy Economics

Date: 2025

Author: Sami Ben Jabeur, Yassine Bakkar, Oguzhan Cepni

Abstract:
We investigate the impact of global common volatility and geopolitical risks on clean energy prices. Our study utilizes daily data from January 1, 2001, to March 18, 2024. Using a new framework based on explainable artificial intelligence (XAI) methods, our findings demonstrate that the COVOL index outperforms the geopolitical risk index in accurately predicting clean energy prices. Furthermore, the Extreme Trees algorithm shows superior performance compared to traditional regression techniques. Our findings indicate that XAI improves transparency, thereby making a substantial contribution to agile decision-making in predicting clean energy prices. Practitioners, including investors and portfolio managers, can enhance investment decisions and manage systemic risks by incorporating COVOL into their risk assessment and asset allocation models.

Link: Google Scholar

Background and Context

Research Focus

The study investigates how global common volatility (COVOL) and geopolitical risks affect clean energy prices using daily data from 2001-2024.

Market Context

Global investment in energy transition reached $1.8 trillion in 2023, with clean energy markets attracting diverse investments amid increasing volatility.

Methodology

Uses explainable artificial intelligence (XAI) methods including LASSO regression, Elastic Net, K-Nearest Neighbors, LightGBM, XGBoost and Extra Trees to analyze predictive power of risk factors.

COVOL Outperforms Traditional Models in Predicting Clean Energy Prices

  • Extra Trees (ET) model achieves highest prediction accuracy with R-squared value of 0.9336
  • Advanced machine learning models significantly outperform traditional statistical approaches
  • LASSO and Elastic Net show poorest performance with R-squared values below 0.1

Model Performance Metrics Show Superior Results with COVOL

  • Model incorporating COVOL shows lowest error rates across all metrics
  • Mean Absolute Percentage Error (MAPE) of just 7.9% indicates high prediction accuracy
  • Low RMSLE value of 0.1515 suggests consistent performance across different price ranges

Performance Improvement with COVOL vs Base Model

  • Adding COVOL reduces prediction errors by up to 26%
  • R-squared value improves by 2.57%
  • Consistent improvement across all performance metrics

Comparative Performance During Ukraine Crisis Period

  • Extra Trees maintains best performance during crisis period with lowest MAE of 7.1797
  • Traditional models show significantly higher prediction errors
  • Advanced machine learning models remain robust during market turbulence

Contribution and Implications

  • First study to comprehensively examine how clean energy assets respond to COVOL and geopolitical concerns
  • Demonstrates superiority of XAI framework in capturing complex market relationships
  • Provides practical guidance for policymakers in creating risk management frameworks for clean energy markets

Data Sources

  • Model comparison chart based on Table 4 showing performance metrics across different models
  • Error metrics visualization derived from Table 2 performance statistics
  • Improvement percentages from Table 7 showing model enhancement with COVOL
  • Crisis period performance from Tables 17 and 18
  • Feature importance visualization based on SHAP analysis results discussed in Section 4.2