Monday, March 2nd, 6.00 pm, Room B217, LAPE, Université de Limoges.

Presented by Aref Mahdavi Ardekani, , DCU Business School, Dublin City University, Dublin, Ireland. Raul Gomez Martinez, Universidad Rey Juan Carlos. Beatriz Garcia Costa, Universidad Rey Juan Carlos. Damien Dupré, DCU Business School, Dublin City University, Dublin, Ireland.

The relation between the dissemination of financial news and financial markets has mostly be investigated in terms of returns prediction. However, the way how news convey information also influence the market in terms of volatility. This research leverage a Large Language Models (OpenAI’s ChatGPT4) to extract six dimensions of narrative sentiment from The Economist between 2023 and 2025. By the application of non-linear econometric techniques, the study is analyzing the information flow to the S&P 500 and the Euro Stoxx 50 indexes. The results are indicating that the narrative structure is not serving as a linear predictor of price direction. Instead, the application of Shannon Transfer Entropy is revealing a significant reduction of uncertainty from narrative Polarity to market volatility. Furthermore, the estimation of a Markov Switching Dynamic Regression is demonstrating that this relationship is highly dependent on the regime. The impact of the sentiment on the market risk is found to be significantly stronger during the high-volatility “crisis” states than in the calm periods. These findings are suggesting that in addition to information, tone and sentiment in authoritative journalism is acting as a state-dependent early warning system for the structural instability of the market.