Structure and dynamics of economic networks across scales

Recent economic disruptions have underscored the urgent need to understand the structure and dynamics of supply- and value-chains. Yet, empirical characterization remains hin- dered by data scarcity and aggregation biases. This disser- tation addresses these challenges through a series of studies that apply concepts from network theory, statistical physics and machine learning. It begins with an analysis of the global automotive supply-chain, aimed at assessing the statistical significance of the observed structural patterns, by employ- ing a null model belonging to the Exponential Random Graphs (ERGs) family. The results reveal a complementarity-driven organization and distinct geographical patterns across the net- work. The focus, then, shifts to the Ecuadorian production network, on which different ERG-like reconstruction models are evaluated in terms of their ability to recover firm-level systemic risk profiles. The results highlight the minimal amount of information required to conduct reliable stress-tests on na- tional production networks. The dissertation next introduces a granular, product-level value-chain constructed from firm- level data and examines its hierarchical organization. This structure is used to characterize countries’ imports and ex- ports, shedding light on industrial capabilities, specialization patterns and growth potential. The final contribution bridges Economic Complexity with interpretable machine learning by proposing the Feature Importance Product Space (FIPS), a predictive framework that surpasses traditional network approaches in forecasting export diversification, while main- taining transparency and interpretability in its predictions.