Is the World Getting Smaller? Network‐Driven Heterogeneity in Gravity‐Model Elasticities

ABSTRACT Gravity models constitute the workhorse of empirical international trade research, yet the temporal stability of their core parameters—distance and GDP elasticities—remains underexplored, particularly in relation to the evolving structure of trade networks. This study investigates the bidirectional Granger‐causal relationships between gravity model coefficients and trade network topology using two complementary databases: the OECD Inter‐Country Input–Output tables (1995–2020, 45 sectors) and the CEPII‐BACI bilateral trade database (1995–2022, 97 product groups). We employ three methodologically distinct causality testing approaches—traditional panel Granger tests, Bayesian model comparison via Bayes factors, and Random Forest‐based variable importance analysis—to establish robust directional relationships while accounting for potential nonlinearities. A key methodological innovation is the incorporation of fractal dimensions and local lacunarity distribution moments (mean, variance, skewness, kurtosis) as novel network indicators capturing multiscale heterogeneity, alongside conventional centrality and modularity measures. Using network‐based dimensionality reduction (Generalized Network‐based Dimensionality Analysis), we identify two to four distinct temporal clusters of sectors and products exhibiting heterogeneous gravity coefficient dynamics, with forecasts extended to 2030 using Bayesian ARIMA models. Our results reveal significant Granger‐causal relationships between network structural properties and gravity coefficients, but the three methods diverge sharply in how much structure they detect: the Random Forest approach identifies three‐ to four‐fold more significant sector‐and product‐level relationships than the two linear methods. At the aggregate level, centralization measures and heterogeneity indicators (lacunarity, fractal dimension) Granger‐precede GDP elasticities broadly and consistently across sectors and products, whereas their relationship with distance elasticity is concentrated in specific sectors and product groups rather than operating uniformly across the aggregate network. Distance elasticity shows declining magnitude over the study period at the aggregate level, but substantial heterogeneity persists across product groups, with some clusters exhibiting stable or even increasing distance sensitivity. The cross‐database triangulation confirms that these patterns are robust to alternative trade flow conceptualizations (gross vs. value‐added trade). These findings contribute to international trade theory in three ways. First, they provide systematic evidence of temporal precedence—established via Granger‐style tests, not structural causal identification—showing that the extensive margin of trade—captured by network topology—consistently leads the intensive margin reflected in gravity coefficients, consistent with heterogeneous firm models predicting that firms' entry/exit decisions reshape network structure before aggregate trade volumes adjust. Second, and most robustly, the documented sector‐level heterogeneity offers a structural account of the ‘distance puzzle’: ICT and services sectors exhibit genuine distance compression through network decentralization, while heavy manufacturing maintains geographic constraints through persistent regional modularity. Third, the Granger‐causal precedence of network fragility indicators over gravity coefficients establishes an empirical bridge between the production network literature on shock propagation and the gravity framework, with direct relevance for understanding supply chain resilience amid geopolitical fragmentation.