A Motif-Based Framework for Decomposing Risk Spillovers

Battle Count: 71

The paper directly addresses portfolio construction through the Minimum Structural Similarity Portfolio (MSP), which outperforms minimum correlation and minimum connectedness benchmarks on risk-adjusted returns. The orbit-position diversity metric identifies tail-specific net transmitters, useful for risk-sensitive allocation. However, the primary contribution is structural/analytical rather than a complete trading system. The portfolio strategies are long-only, equal-weight-constrained, and the performance gains, while statistically meaningful, are moderate. The framework is most relevant for risk management overlay and diversification rather than alpha generation.

The framework involves multiple complex stages: (1) QVAR estimation at multiple quantiles with rolling windows, (2) GFEVD computation and extended joint connectedness normalization, (3) multiscale backbone extraction via disparity filter with dual-perspective testing, (4) enumeration of 13 directed triadic motifs and 30 orbit positions with significance testing against degree-preserving null models, (5) colored motif analysis under multiple sector partitions, (6) orbit-position diversity computation and correlation analysis, and (7) portfolio optimization using structural similarity matrices. Each stage requires specialized computational tools and careful parameter selection.

The paper provides detailed mathematical formulations for QVAR, GFEVD, multiscale backbone extraction, and motif analysis. Data is sourced from Investing.com (publicly available). However, no code repository is mentioned, and the specific implementation details for motif enumeration, orbit counting, and portfolio optimization are not fully specified. The rolling window parameters (w=200, w=250) and backbone thresholds (alpha=0.05, 0.10) are clearly stated.