News sentiment and volatility dynamics: a HAR-X analysis of Brazilian stock markets
Carregando...
Arquivos
Data
2025-06-24
Autores
Orientador(res)
Pinto, Afonso de Campos
Botelho, Marcos Antonio
Métricas
Título da Revista
ISSN da Revista
Título de Volume
Resumo
The objective of this work is to apply to stocks traded on the Brazilian stock exchange the Heterogeneous Autoregressive model integrated with the use of news sentiment scores as an exogenous variable in order to estimate the effect of news on stock volatility. The volatility of the assets is obtained from the opening, minimum, maximum, and closing prices over one-hour intervals, using the Garman-Klass estimator. The sentiment scores are derived from a zero-shot classifier and proposed hypotheses, using as input approximately 2500 financial news from May 2023 to March 2025, with the intention of estimating the impact of news on stocks. Machine learning metrics such as R 2 and Mean Squared Error were compared between the models with and without exogenous variables for each evaluated asset. The inclusion of sentiment scores led to an overall improvement in performance. Furthermore, the analysis revealed a consistent lag structure across all evaluated stocks, characterized by similar linear coefficients associated with the lagged news sentiment variables. This pattern held irrespective of the industry in which each company operates, suggesting a generalized temporal response to news sentiment in volatility modeling.