Simple trees, smarter portfolios: Black-Litterman meets machine learning

The thesis analyses the structural limitations of classic asset allocation models, from Markowitz to CAPM, highlighting how input errors and portfolio instability reduce their applicability. In this context, the Black-Litterman model is introduced, which, through a Bayesian approach, allows for the integration of a prior equilibrium with subjective views, mitigating the critical issues of traditional models. The methodological contribution of the thesis is the use of machine learning techniques, in particular Random Forests, to generate quantitative views consistent with the Bayesian framework. This allows for the exploration of a hybrid approach that maintains theoretical soundness and expands the possibilities for application, with practical implications for asset management.