Enhancing asset allocation and portfolio rebalancing through dynamic theme detection

Abstract

Traditional industry classification systems, such as the Global Industry Classification Standard (GICS) and the North American Industry Classification System (NAICS), have long served as the backbone of financial analysis, portfolio construction, and risk management. However, these systems rely on static categorizations, typically assigning each company to a single sector based on its main revenue source. In today’s interconnected and rapidly evolving economy, this approach fails to capture the complexity of corporate activities. Modern companies increasingly operate across multiple industries and emerging themes, which cannot be adequately represented by rigid taxonomies. This thesis introduces and evaluates Dynamic Industry Classification (DIC), a novel framework that leverages artificial intelligence and natural language processing to provide a multidimensional and continuously updated view of corporate activities. Unlike static models, DIC identifies a company’s exposures across multiple industries and themes, offering a more accurate representation of its business reality. The research compares DIC with GICS by examining their effectiveness in portfolio optimization, risk management, and thematic investing. By constructing and testing portfolios based on both systems, the study assesses whether DIC delivers better diversification, improved risk control, and earlier recognition of emerging trends. The findings contribute to both academic literature and practical applications, providing insights for investors, regulators, and policymakers in a highly dynamic financial landscape.