This paper investigates whether sector-specific financial sentiment lexicons improve the analysis of Brazilian stock-market news in Portuguese. We use 19,460 news items about 38 firms in the financial, electric, and manufacturing sectors, published between 2020 and 2025. Starting from OpLexicon, we build sector-specific lexicons through unsupervised expansion based on PMI and conditional word-polarity association. The lexicons are evaluated by lexical heterogeneity and through LSTM price forecasting. Results show sectoral variation and moderate classification performance, but limited predictive gains. Sentiment improves forecasting only in manufacturing, while other sectors do not benefit from its inclusion. The findings indicate that sector-specific lexicons are useful resources, although their predictive value are context dependent.