Advancing Renewable Energy in the EU: Financial Analysis, Policy Evaluation, and Machine Learning Forecasts

Abstract

This thesis dives into the evaluation of the renewable energy (RE) sector through the lens of financial, macroeconomic, and policy impacts in the European Union (EU) from 2000 to 2024, which have allowed this region to encounter substantial sector growth and establish itself as a global leader in the RE field. The research combines traditional econometric techniques and machine learning (ML) algorithms to provide a novel and multi-dimensional analytical approach to investors, policymakers, and stakeholders engaged in investing and developing the sector.

The second chapter investigates the key drivers of the EU RE stock returns and explore the evolution of their relationship over the last decade. To model the interaction, the analysis incorporates a five-variable VAR model, along with Granger causality, impulse response functions, historical variance decomposition, and structural breaks tests. The results suggest that there has been a change in the RE market interaction with the adopted financial and macroeconomic variables over the sample period, especially since the Paris Accord of 2016 and the COVID-19 pandemic.

The third chapter focuses on the Environmental Policy Stringency (EPS) and isolated policy reforms in 2014 and 2016 to evaluate their effect on the RE supply across 21 EU countries. The analysis adopts panel data and employs econometric methodology that encompass both static and dynamic approaches. The results reveal that whilst the Energy Security policies of 2014 and the EPS index demonstrated a strong and positive impact on the RE supply, the Paris Accord (2016) policies created an unexpected negative impact, highlighting the importance of domestic targeted policies.

The fourth chapter adopts and compares three tree-based ML algorithms to forecast the direction of stock prices in the RE sector, along with identifying the most significant predictors. The analysis incorporates technical indicators for RE stocks, as well as new macroeconomic and financial variables. The Random Forest algorithm was identified as the most accurate for RE stock predictions over all key time horizons. The technical indicators were found to be the most important features whilst the technology index was the most influential among non-technical features.

Date of Award	10 Dec 2025
Original language	English
Awarding Institution	University of Bath
Sponsors	Economic and Social Research Council
Supervisor	Bruce Morley (Supervisor) & Alistair Hunt (Supervisor)