Enhancing financial time series prediction and asset allocation with machine learning and artificial intelligence

• Autores: Daniel Alejandro González Cortés
• Directores de la Tesis: Enrique Onieva Caracuel (dir. tes.), Iker Pastor López (dir. tes.)
• Lectura: En la Universidad de Deusto ( España ) en 2025
• Idioma: inglés
• Tribunal Calificador de la Tesis: Eneko Osaba (presid.), Laura Arjona Aguilera (secret.), Igone Porto Gómez (voc.)
• Programa de doctorado: Programa de Doctorado en Ingeniería para la Sociedad de la Información y Desarrollo Sostenible por la Universidad de Deusto
• Materias:
  • Matemáticas
    • Ciencia de los ordenadores
      • Inteligencia artificial
      • Código y sistemas de codificación
      • Sistemas de información, diseño y componentes
  • Ciencias económicas
    • Econometría
      • Modelos econométricos
    • Actividad económica
      • Dinero y operaciones bancarias
• Enlaces
  • Tesis en acceso abierto en: DeustoTeka
• Resumen

  • This thesis explored the integration of machine learning algorithms, deep learning, and reinforcement learning in the analysis of financial time series for prediction and portfolio construction. The main objectives were to enhance the accuracy of predictions, optimize asset selection to maximize risk-adjusted returns, improve the interpretability of market forecasts, and optimize portfolio management strategies.

    This research began with a comprehensive review of traditional asset pricing models, regression algorithms, and neural network architectures. It then explored state-of-the-art machine learning applications in finance, highlighting trends and methodologies for financial time series forecasting and clustering.

    A neural network-based approach for predicting financial market time series was proposed, demonstrating significant improvements in forecasting accuracy. Additionally, an autoencoder-enhanced clustering method was introduced to facilitate dimensionality reduction in financial datasets, leading to a more efficient data analysis.

    The application of explainability in predicting digital asset prices was examined, emphasizing the importance of model transparency and interpretability. Various machine learning models were evaluated, and their performance was analyzed in terms of accuracy and explainability.

    The thesis also presented a novel approach to portfolio construction using explainable reinforcement learning. This method leveraged reinforcement learning algorithms to optimize portfolio allocation while ensuring the model's decisions were interpretable and understandable to stakeholders.

    The findings contributed to the field of financial machine learning by providing adequate and explainable methodologies for time series forecasting, clustering, and portfolio management.