GMO approvals under pressure: How climate shocks shape policy across nations

PLOS One | https://doi.org/10.1371/journal.pone.0332298 October 22, 2025 9 / 20 technologies. By estimating impulse response functions directly, the method accounts for the dynamic nature of policy adoption while avoiding the complexity of specifying and estimating a full vector autoregressive (VAR) model. Following prior work [31], the empirical model regresses the level of the outcome (number of GMO appovals) at horizon h (i.e., t + h) on its lagged values and current covariates, i.e., climate shocks, enabling us to trace out the effect of a shock over time. The impulse response is estimated as the following model: θij,t+h = βh 1 ωi,t + βh 3 (θij,t–1 ) + βh 4 σi + δh j + εh i,t (12) where ωi,t denotes the number of shocks in country i at year t, which is the main variable of interest in this paper. The dependent variable is θij,t+h which denotes the number of GMO approvals in levels at horizons t + h for product j . Horizon h is the estimation time period, ranging from horizon 0 to horizon 7, which therefore captures the contemporaneous and cumulative effects up to 7 years after the shock. The coefficient βh 1 captures the impact of one additional climate shock on GMO approvals h years later. The lagged dependent variable θij,t–1 controls for prior approval trends. To address potential endogeneity, the analysis adopts several strategies. First, climate shocks—such as droughts, floods, or extreme cold—are treated as plausibly exogenous, as defined by EM-DAT’s physical criteria, and are unlikely

to be driven by GMO policy. Second, the local projection framework helps mitigate simultaneity concerns by estimating delayed effects of shocks over time, which aligns with the regulatory nature of GMO approval processes. Third, the model includes a set of country-year control variables σi,t to address omitted variable bias. These controls are selected for their presumed exogeneity and relevance to GMO regulatory capacity. Specifically, the model includes: (1) an indicator for membership in the International Union for the Protection of New Varieties of Plants (UPOV), which reflects alignment with global biosafety and intellectual property frameworks such as the Codex Alimentarius and the Cartagena Protocol on Biosafety; (2) the agricultural share of GDP; (3) arable land per capita; and (4) an index of infra- structure quality. These variables capture differences in regulatory institutions, agricultural dependence, and a country’s capacity to adopt and implement GMO technologies. Prior research indicates that countries with weaker infrastructure or a higher reliance on agriculture may respond differently to climate shocks in the context of GMO adoption [32–35]. Other potentially relevant variables, such as fertilizer use, are excluded due to endogeneity concerns. Fertilizer application may reflect unobserved factors—such as crop mix, farming practices, or technological capacity—that also affect GMO approval decisions. Including such variables could introduce multicollinearity with existing variables or bias the estimated relation-

ship between climate shocks and regulatory responses. Forth, the country fixed effect σi absorb unobserved, time-invariant heterogeneity such as climate, geography, dietary preferences, or baseline regulatory stances. Commodity fixed effects δh j account for persistent differences across prod- ucts—for example, cotton may require fewer GMO traits than maize. Standard errors are clustered at the country-product level to allow for serial correlation and within-cluster heteroskedasticity. 5. Results To assess the long-term impact of climate shocks on GMO approvals, Table 3 presents cumulative effects from the first year up to seven years after a shock. These results incorporate lag structures and control variables, capturing how an additional climate shock influences the number of GMO approvals over time. In the first panel, which includes country and commodity fixed effects but omits control variables, we find that climate shocks significantly reduce GMO approvals four years after the event. Effects in later years are statistically insignificant. In the second panel, which adds key control variables, the effect remains: a single shock is associated with a 0.031-unit decline in approvals in year four. This timing aligns with average GMO approval timelines, typically ranging from 1.75 to 5.9 years [36]. Among the controls, a higher agricultural share of GDP is negatively associated with approvals in years six and seven, suggesting that countries with smaller agricultural sectors may be more inclined to adopt GMOs to address productivity concerns. UPOV membership