Carbon emissions trading and high-quality agricultural development in China: evidence from staggered adoption

China’s Carbon Emissions Trading (CET) pilots may affect agriculture through input costs, investment, technology adoption, and environmental governance. Using provincial data for 2007–2020, this study estimates cohort-time effects of CET exposure on High-Quality Agricultural Development (HQAD), an index covering innovation, coordination, greening, openness, and sharing. The analysis covers 30 province-level regions and three adoption cohorts defined by trading commencement in 2013, 2014, and 2016. The five-year average treatment effect on the treated is 0.02091 HQAD index units (standard error 0.00873; 95% confidence interval [0.00380−0.03802]), indicating a positive average response over the launch year and the following four years. In the candidate-channel regressions, CET exposure is positively associated with environmental regulation intensity in all three samples, with 95% confidence intervals above zero. The CET coefficients for agricultural energy total factor productivity (AETFP) are positive in both samples, with intervals that include zero. All five intermediate-outcome coefficients in the next-year HQAD equations have intervals that include zero. These coefficients describe conditional associations. Separate simulation benchmarks compare first-difference Double Machine Learning (DML) specifications, with coefficients ranging from 0.00628 to 0.01066 across nuisance learners. CET incentives are linked to five HQAD dimensions through competing productive-adjustment and adjustment-cost hypotheses.The five-year HQAD effect is 0.02091, with a 95% confidence interval of [0.00380−0.03802].The design distinguishes the DML projection coefficient from the cohort-time average treatment effect.Trading-launch dates define four provincial adopters in 2013, two in 2014, and one in 2016.The Type-3 architecture specifies four scalar consequent bounds per rule.All six HQAD constructions yield positive ATT estimates; only the baseline 95% confidence interval excludes zero. CET incentives are linked to five HQAD dimensions through competing productive-adjustment and adjustment-cost hypotheses. The five-year HQAD effect is 0.02091, with a 95% confidence interval of [0.00380−0.03802]. The design distinguishes the DML projection coefficient from the cohort-time average treatment effect. Trading-launch dates define four provincial adopters in 2013, two in 2014, and one in 2016. The Type-3 architecture specifies four scalar consequent bounds per rule. All six HQAD constructions yield positive ATT estimates; only the baseline 95% confidence interval excludes zero.