Modeling electricity price formation during Poland’s energy transition using market data and sentiment analysis

Electricity-price dynamics in energy-transitioning power systems reflect the interaction of fuel costs, carbon prices, financial conditions, regulatory information, and market expectations. This study examines whether text-derived news indicators provide incremental predictive information for Polish electricity prices. A curated English-language European policy-news corpus identified through a GDELT-assisted collection procedure is processed using two separate natural-language-processing branches. TF–IDF is used to characterize the thematic prominence of selected energy and climate-policy terms, while VADER is applied directly to article text to derive an interpretable news-tone indicator, referred to as the Climate Change News Index. These indicators are aligned with electricity, natural-gas, coal, oil, carbon-allowance, and equity-market variables. Ordinary Least Squares and Elastic Net models are evaluated using chronological out-of-sample forecasts. Market fundamentals remain the main predictive inputs, while the contribution of text-derived variables is small and specification-dependent. The results therefore indicate limited incremental predictive information rather than systematic forecasting superiority. The study is observational, and the reported relationships are interpreted as associative and predictive rather than causal. The proposed workflow provides a transparent way to incorporate policy-news information into electricity-price analysis. However, it does not replace physical power-system variables but it provides additional predictive power for better forecasting which can be useful for the market players.