Graph centrality and deep learning for enhanced portfolio management

Mohammad Mahdi Haji Abbasi, Zahed Rahmati, Mostafa Abbaszadeh,
Graph centrality and deep learning for enhanced portfolio management,
Egyptian Informatics Journal,
Volume 32,
2025,
100858,
ISSN 1110-8665,
https://doi.org/10.1016/j.eij.2025.100858.
(https://www.sciencedirect.com/science/article/pii/S1110866525002518)
Abstract: Financial market investors always try to manage their portfolios with regard to stock prices for maximizing returns with minimum risk in a constrained amount of time. Instead of selecting stocks at random, we present a novel method in this paper that creates a graph of the stock market. We determine the most significant stocks using this graph, centrality metrics, and graph representation techniques. Next, a hybrid deep learning model incorporating the Conv1D, GRU, and RNN layers predicts future prices and evaluates their risks. The S&P 500 index is used as a benchmark for these outcomes. These chosen stocks have been evaluated for their risk and return over three time periods — short, medium, and long-term — and contrasted with the S&P 500 benchmark. According to these findings, the suggested algorithms perform better than the S&P 500 over these time periods. Second, compared to selecting stocks at random, this method is more generalizable and can be used in a variety of markets and contexts.
Keywords: Portfolio optimization; Stock price prediction; Graph representation learning; Graph centrality; Deep learning; Recurrent neural networks; GRU; Ensemble learning