Multiobjective portfolio management based on dynamic link prediction

Abstract Current research increasingly emphasizes the use of intelligent algorithms to enhance decision-making in portfolio management. However, treating each stock as an independent entity neglects the interdependencies in returns and risks arising from complex relationships among stocks. To address this limitation, we propose a link prediction portfolio management model that redefines portfolio management by incorporating network-based insights. The model extracts a diverse set of link features and introduces a novel sampling technique to dynamically predict future co-movement relationships between stock pairs. It also integrates multiple investment theories to develop effective stock preselection strategies. Furthermore, we extend the traditional mean value-at-risk framework by introducing a new objective function that simultaneously captures dependency-based returns and risks across stocks. This enhancement offers improved risk control and greater flexibility in maximizing profits. Empirical evaluation on real-world Chinese stock market data demonstrates the model’s superior performance, achieving cumulative returns that are 91.66 to $$-$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mo>-</mml:mo> </mml:math> 217.19% higher than the benchmark model, along with annual Sharpe ratios that are 1.99 to $$-$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mo>-</mml:mo> </mml:math> 4.7 times greater. Additionally, by incorporating stock price synchronicity theory, we assess our investment strategies across various dimensions-including investor preferences and market conditions-offering a comprehensive analysis of both theoretical and practical aspects of portfolio management.