Fast change point detection for natural gas market analysis

Change point analysis is an important part of the time-series analysis, as the existence of a change point denotes an abrupt and significant change in the data generating process. While the methodologies for change point detection are widely studied and there exist many different algorithms to detect them, little attention has been given to evaluating the performance of these algorithms on real world time-series. This thesis aims at applying change-point methodologies to a real time-series. The dataset employed to perform the change-point analysis has been provided by ARERA (Autorità di regolamentazione per energia reti e ambiente) and is composed by multiple time-series regarding natural gas prices that cover the period between the 4th of January 2010 and the 25th of March 2021. The time-series include the Italian virtual hub prices (PSV, Punto di Scambio Virtuale), and the Dutch virtual hub prices (TTF, Title Transfer Facility). Natural gas is still very important to meet global energy needs, therefore being able to identify anomalies and irregularities would be an important achievement. This is why the analysis focuses on this particular dataset. The model employed to conduct the change-point analysis is the non-parametric multivariate and multiple change-point model. This analysis has been provided with the R-studio software using the “ecp” R package. To verify if the change points detected by the algorithms do correspond to structural changes in the natural gas market over the years, the results from this statistical analysis have been checked against a set of a priori expected change-points. The interpretation of the obtained change-point has been made through the methodology of in-dept interviews of people with the highest skills in the natural gas market like traders, energy regulators and academics. Results demonstrate that the non-parametric change-point detection model can effectively identify structural changes in the Italian and Dutch natural gas markets from the price information alone. The changes detected include natural, technical, and economic changes. Therefore, the model has proved to be useful in monitoring market performance.