Regional Trade Agreements (RTA) play an important role in shaping agricultural trade and market access. Compared with manufactured goods, agricultural trade is more sensitive to tariffs, SPS measures, and other trade barriers. As a result, existing studies report mixed findings on the effects of RTA on agricultural trade. This study argues that such differences may result not only from real economic heterogeneity, but also from model specification issues. Previous studies often use highly complex structural gravity models with many variables and interaction terms, which may improve in-sample fit but also create overfitting problems and weaken predictive performance. To address this issue, this study introduces the Least Absolute Shrinkage and Selection Operator (Lasso) method into a structural gravity framework. By selecting only the most relevant extended variables while preserving core theoretical constraints, the model reduces unnecessary complexity and improves robustness. Empirical results show that the regularized model improves out-of-sample predictive performance and reduces sensitivity to alternative specifications in evaluating the effects of RTA on agricultural trade.
