Multi-Paradigm Analysis of Thai Capital Market Linkages: Bivariate/Vine Copulas, Granger Causality, Network Centrality, and Graph Neural Network/Graph Embedding Approaches

Analytically thorough understanding of causal, probabilistic, and informational linkages amongst modern, highly-interconnected capital markets is fundamental to the promotion of capital-market innovation, efficiency, and resilience; whereupon innovative, efficient, and resilient capital markets are fundamental to the sustainable economic development of any nation and the robust financial stability of her economy. Whereas analytics derived essentially from some form of linear correlation matrix remains to this day the archetype model representation of return co-movements within a single stock exchange, analysis of financial random multi-variates, especially across domestic/international/regional/global/ domains, warrants a generalised multi-paradigm investigation. In lieu of employing a particular quantitative technique to test a specific theoretically-motivated hypothesis, our aim here is to apply a multitude of analytical tools from semi-/non-parametric statistics, from time-series econometrics, from graph-theoretic network analysis, and from Graph Neural Network (GNN)/Geometric Deep Learning (GDL) paradigms to investigate quantifiable linkages amongst 14 selected variables representing Thai/Asia/Emerging Market/US foreign-exchange, fixed-income, and equity market movements, as well as foreign portfolio-investment flows into Thailand’s domestic equity and bond markets. In particular, Bivariate/Vine Copulas, Granger Causality, Network Centrality, and GNN/Graph Embedding together enable us to tease out subtle quantitative relations amongst said linked capital-market variables, which could in turn be exploited, for instance, in the form of better specified Monte Carlo Value-at-Risk/Expected Shortfall estimation for hypothetical portfolios comprising multi-national assets. Our experiments thus far delivered incremental, rather than game-changing, improvements w.r.t. market insights and accuracy gains. Nonetheless, our multi-paradigm framework, along with the publicly shared codebase, can be applied to any other “linkage” phenomena of future research interest and real-world application.