Cargando...
Fecha
2025-07
Derechos de acceso
info:eu-repo/semantics/openAccess
Título de la revista
ISSN de la revista
Título del volumen
Editorial
Citas
Resumen
Este Trabajo de Fin de Máster aborda el desafío de cuantificar la información cualitativa contenida en el Libro Beige de la Reserva Federal de Estados Unidos (FED) mediante técnicas de Procesamiento del Lenguaje Natural (PLN) y transfer learning. El Libro Beige, un documento relevante sobre las condiciones económicas, tradicionalmente ha sido objeto de interpretaciones subjetivas. El objetivo central es desarrollar una metodología sistemática para construir un índice de sentimiento económico y utilizarlo para modelar la relación con el Producto Interior Bruto (PIB) de EE.UU. y predecir el comportamiento del índice bursátil S&P 500. La metodología principal empleada es el transfer learning, crucial dada la escasez de datos etiquetados específicamente para el Libro Beige. Se aplican modelos preentrenados basados en la arquitectura Transformer, ajustados con datos etiquetados de comunicaciones de bancos centrales. Se exploran distintos modelos con el fin de generar un índice cuantitativo de sentimiento del Libro Beige para epígrafes relevantes desde octubre de 1996. Los resultados del estudio indican una relación de valor informativo entre el índice de sentimiento del Libro Beige construido y la tasa de crecimiento del PIB real de EE.UU., validando su capacidad para reflejar la actividad económica. El análisis de la relación dinámica con el índice bursátil S&P 500 (1996-2022) muestra un movimiento unísono general. Se observa que los momentos de mínimos en los índices de sentimiento parecen indicar oportunidades de entrada en bolsa, especialmente en un horizonte temporal amplio. Se evalúa la capacidad predictiva del sentimiento del Libro Beige sobre el S&P 500 utilizando una red neuronal LSTM. En la predicción para el año 2022, la inclusión del índice de sentimiento mejora el desempeño predictivo respecto a un modelo base. Si bien los experimentos de optimización de hiperparámetros no siempre son concluyentes, estos hallazgos iniciales sugieren que el sentimiento extraído del Libro Beige puede aportar valor predictivo. Este trabajo demuestra la aplicación efectiva del transfer learning para cuantificar la información cualitativa contenida en el Libro Beige. El índice de sentimiento desarrollado ofrece una valoración más objetiva de este documento y muestra potencial como indicador económico y financiero adelantado y como herramienta complementaria para decisiones de inversión.
This Master’s Final Project addresses the challenge of quantifying the qualitative information contained in the Federal Reserve’s (FED) Beige Book through Natural Language Processing (NLP) techniques and transfer learning. The Beige Book, a relevant document on economic conditions, has traditionally been subject to subjective interpretations. The central objective is to develop a systematic methodology to build an economic sentiment index and use it to model the relationship with the U.S. Gross Domestic Product (GDP) and predict the behavior of the S&P 500 stock index. The main methodology employed is transfer learning, crucial given the scarcity of data specifically labeled for the Beige Book. Pre-trained models based on the Transformer architecture are applied, fine-tuned with labeled data from central bank communications. Different models are explored in order to generate a quantitative Beige Book sentiment index for relevant epigraphs since October 1996. Once this quantitative Beige Book sentiment index is generated, correlations with GDP are analyzed and predictive capacity on the S&P 500 is evaluated using an LSTM neural network. The results demonstrate the effective application of transfer learning to quantify qualitative information in the Beige Book, offering a more objective assessment of this document. The study results indicate an informative relationship between the constructed Beige Book sentiment index and the U.S. real GDP growth rate, validating its ability to reflect economic activity. The analysis of the dynamic relationship with the S&P 500 stock index (1996-2022) shows a general unison movement. It is observed that moments of lows in the sentiment indices seem to indicate stock entry opportunities, especially over a broad time horizon. The predictive capacity of the Beige Book sentiment on the S&P 500 is evaluated using an LSTM neural network. In the prediction for the year 2022, the inclusion of the sentiment index, improves predictive performance compared to a base model. While hyperparameter optimization experiments are not always conclusive, these initial findings suggest that sentiment extracted from the Beige Book can provide predictive value. The developed sentiment index shows potential as a leading economic and financial indicator and as a complementary tool for investment decisions.
This Master’s Final Project addresses the challenge of quantifying the qualitative information contained in the Federal Reserve’s (FED) Beige Book through Natural Language Processing (NLP) techniques and transfer learning. The Beige Book, a relevant document on economic conditions, has traditionally been subject to subjective interpretations. The central objective is to develop a systematic methodology to build an economic sentiment index and use it to model the relationship with the U.S. Gross Domestic Product (GDP) and predict the behavior of the S&P 500 stock index. The main methodology employed is transfer learning, crucial given the scarcity of data specifically labeled for the Beige Book. Pre-trained models based on the Transformer architecture are applied, fine-tuned with labeled data from central bank communications. Different models are explored in order to generate a quantitative Beige Book sentiment index for relevant epigraphs since October 1996. Once this quantitative Beige Book sentiment index is generated, correlations with GDP are analyzed and predictive capacity on the S&P 500 is evaluated using an LSTM neural network. The results demonstrate the effective application of transfer learning to quantify qualitative information in the Beige Book, offering a more objective assessment of this document. The study results indicate an informative relationship between the constructed Beige Book sentiment index and the U.S. real GDP growth rate, validating its ability to reflect economic activity. The analysis of the dynamic relationship with the S&P 500 stock index (1996-2022) shows a general unison movement. It is observed that moments of lows in the sentiment indices seem to indicate stock entry opportunities, especially over a broad time horizon. The predictive capacity of the Beige Book sentiment on the S&P 500 is evaluated using an LSTM neural network. In the prediction for the year 2022, the inclusion of the sentiment index, improves predictive performance compared to a base model. While hyperparameter optimization experiments are not always conclusive, these initial findings suggest that sentiment extracted from the Beige Book can provide predictive value. The developed sentiment index shows potential as a leading economic and financial indicator and as a complementary tool for investment decisions.
Descripción
Categorías UNESCO
Palabras clave
Libro Beige, Transfer Learning, Análisis de sentimiento, PIB, S&P 500, Arquitectura Transformer, Beige Book, Transfer Learning, Sentiment Analysis, GDP, S&P 500, Transformer architecture
Citación
Alcalá Gutiérrez, Luis Alberto. Trabajo fin de Máster: "Transfer Learning para el Análisis de Sentimiento: Aplicación de Libro Beige". Universidad Nacional de Educación a Distnacia (UNED), 2025
Centro
E.T.S. de Ingeniería Informática
Departamento
Lenguajes y Sistemas Informáticos