Beyond Covariance: Network-Based Portfolio Construction for Global Bonds

This article proposes an implementable framework for global bond portfolio construction that integrates yield curve modeling, network-based currency allocation, and network-based bond selection within a unified portfolio construction process. Portfolio decisions are separated into currency allocation and security selection, with yield-curve-implied expectations of bond and currency returns serving as common inputs across both stages. Currency allocation is conducted through a network of local currency markets designed to capture cross-market interactions and macro-financial risk exposures, while bond selection is performed using a complementary network of individual securities defined by duration, spread, roll-down, momentum, and liquidity characteristics. Local currency yield curves provide a consistent term-structure foundation for both expected return estimation and risk modeling. The results show that network-based risk representations provide a richer characterization of currency and bond interdependencies than traditional covariance-based approaches do and lead to more structurally diversified portfolios. From an implementation perspective, the framework provides a systematic process for integrating currency allocation, bond selection, and factor exposure control within a single portfolio construction methodology. By integrating network structure into both allocation and optimization decisions, the framework provides a transparent and flexible alternative to traditional global fixed-income portfolio construction methods. The proposed approach illustrates how network representations can complement traditional factor and risk models in global fixed-income portfolio management.