How does trade development affect environmental performance? New assessment from partially linear additive panel analysis.

Assessing the linkage between trade development and environmental quality has recently attracted increasing attention because of the rapid advancement of trade globalization along with the severe degradation of the environment. However, neglecting the possible nonlinear relationship between the variables, the existing studies mainly use linear models to explore the trade-environment linkage. To fill this gap, this paper introduces a novel partially linear additive panel model that highlights nonlinear and heterogeneous features to investigate the direct and indirect influences of trade development on environmental performance over the period 2005–2017 in China. The results show that trade has an inverse "U-shaped" direct effect on the environment. In contrast, trade development can indirectly improve environmental quality through economic growth. Nonlinear marginal analysis demonstrates that such an impact not only presents individual differences but also embodies the trait of temporal dynamics. Empirical results further reveal the importance of our model, which can automatically identify the variable nexus and thus reduce the setting error of the parametric models. Model evaluation also confirms the superiority of the used model over existing parametric and nonparametric models. The partially linear additive panel model provides a new evaluation method for environmental impact regarding trade development, and can be applied to other fields where an assessment of the effect of socio-economic factors on the environment is needed. • Investigate the effects of trade development on environmental quality. • Execute the latest method of partial linear additive panel model. • Reveal an inverse "U-shaped" direct effect of trade development on environmental quality. • Show a serious regional mismatch between trade-related pollutants and economic gains. • Evaluate the performance of different methods by model evaluation and selection. [ABSTRACT FROM AUTHOR]