A non-supervised asset selection for portfolio optimization

• Ano de defesa: 2023
• Autor(a) principal: Gularte, Ana Paula [UNIFESP]
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de São Paulo
• Palavras-chave em Português: Optimization; Project Portfolio; Cluster Analysis; Financial Market; Risk Assessment; Investments; Computing; Economics
• Link de acesso: https://repositorio.unifesp.br/handle/11600/69685

As a fundamental component of the global financial system, investment portfolios drive individual and institutional wealth and play a pivotal role in economic growth and resource allocation. Optimization of these portfolios, while maximizing returns and minimizing risks, is influenced by various economic factors, market dynamics, and investor sentiments. Achieving a precise balance in asset allocation among these complexities is a formidable challenge. This research introduces a two-tier approach that combines dimensionality reduction with clustering techniques to enhance portfolio optimization. The methodology encompasses: (1) the Uniform Manifold Approximation and Projection (UMAP) technique, juxtaposed with Kernel Principal Component Analysis (KPCA) and t-distributed Stochastic Neighbor Embedding (t-SNE), to extract the essential characteristics of asset data. Subsequent clustering via K-means, Partition Around Medoids (PAM), Agglomerative Hierarchical Clustering (AHC), and the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) to enable effective asset pre-selection; (2) Hierarchical Risk Parity (HRP) model implementation for optimized asset allocation, compared against the Mean-Variance (MV) and Inverse-Variance Portfolio (IVP) models. The process begins with a meticulous curation of asset data, undergoes UMAP-driven dimensionality reduction, is refined through clustering for asset selection, and culminates in portfolio optimization via the HRP model. Empirical evaluation of the Ibovespa and Standard and Poor’s 500 (S&P 500) indexes from January 1, 2016, to December 31, 2021, indicates that the integrated approach outshines traditional models, achieving a Sharpe ratio of 1.11. In particular, portfolios also demonstrated resilience to market adversities, such as the pandemic, surpassing benchmarks in terms of cumulative returns. This research underscores the potential of combining data-driven techniques with traditional financial insights, ushering in a new era of robust portfolio management.