When resilience erodes itself: double machine learning evidence on extreme precipitation and supply chain resilience in the textile value chain

We examine a capability paradox: extreme precipitation heightens the need for supply chain resilience (SCR) while depleting the resources required to sustain it. Using Chinese A-share firms in the textile value chain (2010–2024), we construct a five-dimensional SCR index and measure extreme precipitation using daily meteorological observations and city-specific percentile thresholds. Partially linear double machine learning, implemented with random forests, cross-fitting, and orthogonalization, indicates that extreme precipitation significantly weakens SCR. Mechanism tests reveal higher operating cost and selling expense ratios, more severe financing constraints, and increased debt financing costs. Cash holdings, sponge city construction, and political connections attenuate the adverse effect, while extreme precipitation also inhibits value chain upgrading. Results remain robust to alternative samples, model specifications, learning algorithms, resilience measures, and permutation-based placebo tests. The findings inform differentiated climate risk management by textile firms and government policies for climate adaptation.