Analysis of crude oil and natural gas market dynamics through a text mining approach

The aim of this study is to systematically model the news-based information flow driving price formation in crude oil and natural gas markets by employing text mining, natural language processing and machine learning techniques and to econometrically examine the effects of these news-based indicators on price dynamics. A large corpus of news articles obtained from the OilPrice platform is classified according to theoretically defined dimensions, including supply–demand direction (increase/decrease), realization versus expectation, geopolitical risk content and energy-market-specific themes. Based on this classification, daily standardized text-based indices are constructed for each category. These multidimensional textual indicators are then linked to Brent crude oil and natural gas prices within an AutoRegressive Distributed Lag (ARDL) framework, allowing for an empirical assessment of both short-run and long-run effects. The findings indicate that news-based supply and demand signals exhibit statistically significant and strong explanatory power, particularly in the crude oil market, with demand-oriented news generating substantially larger price effects than supply-oriented news. Long-run estimates reveal that the impact of the Demand Increase Index on Brent prices is approximately 11.5 times stronger than that of the Supply Increase Index, highlighting a pronounced asymmetry in the market's response to demand-driven information. Moreover, global macro-financial indicators such as the VIX (financial market volatility index) and the DXY (U.S. dollar index) exert statistically significant and negative effects on oil prices, confirming that periods of heightened financial uncertainty tend to suppress oil price dynamics. In contrast, the results for the natural gas market suggest that news effects are relatively weaker, reflecting the market's asymmetric information structure and region-specific pricing mechanisms. ARDL-based price forecasting results further indicate that, over a two-year forecast horizon, Brent crude oil prices are expected to converge to a lower equilibrium band relative to the reference period, a long-run downward price scenario that is consistent with international institutions' projections of excess supply and subdued demand growth. By contrast, natural gas price forecasts display a relatively flat average trajectory accompanied by widening confidence intervals, suggesting heightened structural uncertainty and continued exposure to potential shocks. Overall, this study offers a novel methodological framework that integrates textual information flows into energy price formation processes and provides comprehensive theoretical and empirical evidence on the role of news-based sentiment in energy markets. The results underscore the critical importance of news dynamics for price discovery, with direct implications for market participants, policymakers and risk managers.