Abstract
This study investigates the dynamics of financial contagion and its implications for portfolio design across the BRICS (Brazil, Russia, India, China, South Africa) equity markets and key global commodities under three systemic crises: the Oil Market Collapse (OMC), the COVID-19 (C19) pandemic, and the Russia–Ukraine conflict (RUC). The main objectives of the study are to move beyond correlation-based contagion measures by employing transfer entropy (TE) to capture nonlinear and directional information flows, mapping these into directed network topologies, and integrating the resulting structural insights into a centrality-augmented (CA) portfolio optimisation framework. Transfer entropy is estimated using a non-parametric kernel density approach, from which directed weighted networks are constructed and analysed using sender-oriented centrality measures. Portfolio strategies are evaluated through constrained optimisation and rolling out-of-sample testing with transaction costs. The analysis is conducted using daily adjusted closing price data for BRICS equity indices and global commodity benchmarks obtained from Yahoo Finance over the period 2013–2022.
Results show that TE networks successfully detect directional contagion channels that vary across crises, with Oil and Russia dominating as transmitters during the OMC and RUC periods, and decentralised spillovers prevailing during COVID-19. Incorporating network-derived centralities into portfolio optimisation improves risk-adjusted performance, reduces maximum drawdowns and average rebalancing costs across all crises. The optimal CA model depends on the crisis type: CA minimum conditional value-at-risk (CA-MinCVaR) performs best under commodity-driven collapses, CA minimum variance (CA-MV) under globally synchronised shocks, and CA equal risk contribution (CA-ERC) or CA minimum variance with turnover constraint (CA-MinVar+TO) under prolonged geopolitical stress. Similarly, the optimal centrality measure depends on the crisis topology: Betweenness is most effective when contagion is directional, CheiRank and Out-Strength perform best under decentralised shocks, and HITS Hub stabilises portfolios when contagion is prolonged and symmetric. Moreover, pre-crisis calibration consistently outperforms in-crisis retraining, confirming the predictive value of pre-shock information structures.