The role of institutional elements in enterprise innovation resilience: Explainable machine learning evidence from China

Abstract The institutional environment is an important external condition shaping enterprise innovation resilience ( EIR ), particularly enterprises’ capacity to sustain innovation and adapt to external shocks. Yet existing studies have largely focused on single institutional factors or linear average effects, leaving the relative importance, nonlinear patterns, and interactions of multidimensional institutional elements ( IE ) underexplored. Drawing on institutional theory and a sample of Chinese listed agriculture- and forestry-related enterprises, this study combines multiple machine-learning algorithms, including CatBoost, with SHAP to examine how regulative, normative, and cognitive IE predict one-period-ahead enterprise innovation resilience ( EIR t+1 ). The results show that current-period enterprise innovation resilience ( EIR t ) is the strongest predictor of EIR t+1 , indicating pronounced dynamic persistence. Among the three institutional dimensions, environmental governance ( RG ), institutional investor shareholding ( INST ), and innovation culture ( EIC ) are the most important predictive features, while the Cognitive Institutional Index ( CII ) has the highest predictive importance among the three composite institutional indices. IE also exhibit nonlinear, stage-dependent, and cross-dimensional interaction patterns, with substantial heterogeneity across regions, ownership types, enterprise sizes, and agriculture- versus forestry-related enterprises. These findings extend institutional theory to EIR research and highlight the predictive relevance of multidimensional, nonlinear, and context-dependent institutional configurations.