Browsing by Author "Dlamini, Nondumiso"

Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    Open Access
    The construction of optimal socially responsible investment portfolios in South Africa using traditional and artificial intelligence techniques
    (2025) Dlamini, Nondumiso; Charteris, Ailie
    For decades, scholars and practitioners have sought optimal portfolio construction methods. Traditional approaches, like mean-variance, face challenges with complex non-linear and non-convex models. Recently, meta-heuristic artificial intelligence (AI) algorithms have enhanced portfolio construction by addressing such constraints. Socially responsible investment (SRI) has gained popularity for its focus on sustainability, but using Environmental, Social and Governance (ESG) criteria in constructing SRI portfolios can introduce estimation risks, increasing the uncertainty of the input parameters and reducing diversification compared to non-SRI portfolios. This study evaluates six portfolio construction methods for SRI portfolios in South Africa, including traditional (mean variance, naïve and risk parity) and AI (particle swarm optimization, simulated annealing and genetic algorithm) methods. Portfolios are compared based on risk-adjusted returns, diversification and stability. On average, AI algorithms produced optimal SRI portfolios with higher risk-adjusted returns. During a period of positive market returns, the genetic algorithm approach performed best, while the mean-variance approach dominated during a period marked by downturns in the market. Across all periods, the genetic algorithm consistently outperformed other methods for SRI portfolios. In contrast, for non-SRI portfolios, the mean-variance method led, followed by genetic algorithm and simulated annealing. Overall, meta-heuristic approaches yielded superior performance for both constrained (SRI) and non-constrained (non-SRI) portfolios, although with higher concentration and less stable weights. Based on the Sharpe ratio, SRI portfolios initially outperformed non-SRI portfolios but lagged in the second period. Non-SRI portfolios ultimately outperformed, suggesting that while AI approaches enhance portfolio construction, SRI strategies may not always match conventional investments.