A Diverse Group Trading Strategy Portfolio Optimization Algorithm Based on Network Modularity

Strategically allocating capital and safeguarding investors from potential adverse conditions are important challenges in finance. Building upon Markowitz's Modern Portfolio Theory (MPT), which advocates diversifying investments across uncorrelated assets, we extend this principle to encompass diversified trading strategies. Previous works in portfolio optimization have explored various assets and methodologies, including machine learning and evolutionary algorithms. The Group Trading Strategy Portfolio (GTSP) framework, which utilizes the grouping genetic algorithm (GGA) to optimize portfolios of trading strategies, has also been proposed. However, challenges persist in the GTSP framework, such as the absence of a mechanism in fitness evaluation to ensure dissimilarity among trading strategies within the same group, leading to the inclusion of highly correlated trading strategies in the portfolios generated. Additionally, in the GTSP framework, strategies rely on a single stock series, which poses a risk, as the underlying risk profile remains uniform across strategies, limiting diversification opportunities. To overcome these challenges, we integrate the GTSP approach with network theory, specifically modularity, to create a model we call GTSP-Modu. By introducing diverse underlying assets, ensuring similarity within groups, and re-designed the fitness function, experiments on the real datasets show that the proposed approach is better than the previous approach in terms of risk-adjusted returns.