This thesis presents the design and implementation of a reproducible software pipeline for financial sentiment analysis, combining Natural Language Processing and concepts related to Generative AI. The system collects real public financial, macroeconomic, and geopolitical news, applies FinBERT, a financial language model, to classify headlines as positive, negative, or neutral, aggregates sentiment at daily frequency, and combines the resulting features with stock return data. From a financial perspective, the objective is to study whether sentiment extracted from real news can provide useful information for predicting the short-term direction of stock returns. The research focuses on ten large US stocks, mainly from the technology sector, and compares market-based models with models enriched by company-level and macroeconomic sentiment features. The empirical results do not clearly confirm the research hypothesis in the selected sample. The sentiment variables do not provide robust evidence of significant incremental predictive power beyond past returns and trading volume. However, this result is also influenced by the limited amount of available public news, the short time window, and the use of daily aggregated headlines. The main contribution of the thesis is therefore the construction of a clear, reproducible, and scalable framework for financial sentiment analysis. With larger datasets, more complete news sources, and longer observation periods, the same pipeline could be extended to support more detailed and potentially stronger empirical analyses.
This thesis presents the design and implementation of a reproducible software pipeline for financial sentiment analysis, combining Natural Language Processing and concepts related to Generative AI. The system collects real public financial, macroeconomic, and geopolitical news, applies FinBERT, a financial language model, to classify headlines as positive, negative, or neutral, aggregates sentiment at daily frequency, and combines the resulting features with stock return data. From a financial perspective, the objective is to study whether sentiment extracted from real news can provide useful information for predicting the short-term direction of stock returns. The research focuses on ten large US stocks, mainly from the technology sector, and compares market-based models with models enriched by company-level and macroeconomic sentiment features. The empirical results do not clearly confirm the research hypothesis in the selected sample. The sentiment variables do not provide robust evidence of significant incremental predictive power beyond past returns and trading volume. However, this result is also influenced by the limited amount of available public news, the short time window, and the use of daily aggregated headlines. The main contribution of the thesis is therefore the construction of a clear, reproducible, and scalable framework for financial sentiment analysis. With larger datasets, more complete news sources, and longer observation periods, the same pipeline could be extended to support more detailed and potentially stronger empirical analyses.
Generative AI for Finance: The Signaling Role of Sentiment Analysis in Financial Markets
PADOVAN, DAVID FRANCESCO
2025/2026
Abstract
This thesis presents the design and implementation of a reproducible software pipeline for financial sentiment analysis, combining Natural Language Processing and concepts related to Generative AI. The system collects real public financial, macroeconomic, and geopolitical news, applies FinBERT, a financial language model, to classify headlines as positive, negative, or neutral, aggregates sentiment at daily frequency, and combines the resulting features with stock return data. From a financial perspective, the objective is to study whether sentiment extracted from real news can provide useful information for predicting the short-term direction of stock returns. The research focuses on ten large US stocks, mainly from the technology sector, and compares market-based models with models enriched by company-level and macroeconomic sentiment features. The empirical results do not clearly confirm the research hypothesis in the selected sample. The sentiment variables do not provide robust evidence of significant incremental predictive power beyond past returns and trading volume. However, this result is also influenced by the limited amount of available public news, the short time window, and the use of daily aggregated headlines. The main contribution of the thesis is therefore the construction of a clear, reproducible, and scalable framework for financial sentiment analysis. With larger datasets, more complete news sources, and longer observation periods, the same pipeline could be extended to support more detailed and potentially stronger empirical analyses.| File | Dimensione | Formato | |
|---|---|---|---|
|
Padovan_DavidFrancesco.pdf
accesso aperto
Dimensione
4.71 MB
Formato
Adobe PDF
|
4.71 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/114278