Related diversification, not generic breadth: industrial composition and urban resilience to the COVID-19 shock

Industrial diversity is widely held to buffer urban economies against economic shocks, yet evidence remains mixed and the operative mechanisms poorly understood. Using establishment-count data for 131 wards across nine Japanese designated cities and Tokyo (2016--2021), we apply Bayesian spatial error models to examine how three diversity dimensions---within-ward breadth ($\alpha$-diversity), between-ward compositional heterogeneity ($\beta$-diversity), and related/unrelated variety---independently predict ward-level resilience during the COVID-19 shock. Related diversity is robustly protective while unrelated diversity is harmful---a sign reversal relative to long-run growth studies that we attribute to the sector-discriminating character of the pandemic: cross-sector portfolio breadth provides no hedging mechanism when proximity-dependent industries collapse simultaneously. Shannon entropy's positive association with resilience masks this opposition between related and unrelated variety. $\alpha$-diversity and sectoral concentration (HHI) exhibit mutual suppression, implying that industrial breadth confers resilience specifically when organised around an anchor sector, while $\beta$-diversity proves to reflect spatial clustering of the information and communications technology sector rather than general compositional heterogeneity. These results imply that the policy-relevant target is strategic \emph{related} diversification---broadening within-sector-family portfolios---rather than maximising generic industrial breadth, and that the operative diversity mechanism depends critically on the type of shock a city faces.