Multiresolutional Statistical Machine Learning for Testing Interdependence of Power Markets: A Variational Mode Decomposition-Based Approach

In the increasingly interconnected and digitized world, the field of electricity price forecasting has benefited from growing research especially due to the market liberalization and the connectedness between electrical systems. This study defines a novel multiscaled forecasting model based upon the Variational Mode Decomposition (VMD) to quantify multiscaled cross-correlation between two important European markets during COVID-19 pandemic. The VMD is known to be a strong information processing tool which is localized in both frequency and time, and is especially used for capturing nonstationary and nonlinear behaviors of time series. The set of new VMD techniques is applied on hourly electricity spot prices from the Nord Pool and MIBEL energy exchanges for the period ranging from January 2019 to March 2020. The sampled time series include a period of specific recession in the financial system, coinciding with the Brexit and COVID-19 event, which was accompanied by a significant collapse in the world's economic sphere. The empirical results reveal a significant dependence between electricity markets across long- and medium-run investment time horizons, with evidence for dynamic lead–lag relationships at some frequency sub-bands. However, over the short-term (daily and intra-daily intervals), we notice a kind of independence between markets, especially in times of crisis, which offers investors different investment diversification opportunities. On the other hand, the accuracy of generated forecasts prove the interest of a conjoint modeling and the reliability of this new tool, in particular when the approach is adequately coupled with feedforward neural networks.