The Utility Of Machine Learning In The Analysis Of The Clean Energy Transition: The Case Of Germany

One of the main components of the clean energy transition process in the EU are its liberalized electricity markets. Since most of the electricity is traded in day-ahead closed auctions, reliable and accurate electricity price prediction has become a question of paramount importance. This has led to the extensive use of machine learning algorithms, which have become increasingly powerful in the last decade, in predicting the movement of key economic variables in the energy sector. However, their use is currently for the most part limited to producing black-box predictions, with no attempt to produce explanations or economic insight. The purpose of this paper is to attempt to see whether a bridge can be built between the disconnected realms of economic analysis and machine learning. We use decision tree-based techniques to analyse the variability of hourly prices in the German electricity market from 2015-2020. We then compare the results with coefficient magnitudes from a linear regression framework. Our results indicate that the two approaches end up in substantial agreement on variable importance. We conclude that this is an area worth exploring further, since it can lead to expanding the energy sector analysis toolkit, which could lead to more informed energy policy.