Evaluating Estimator Performance in Unbalanced Climate Panels: A Simulation-Based Comparison of FE, IV, Difference-GMM, and System-GMM

Abstract

The aim of this study is to evaluate estimator performance in unbalanced dy namic climate
panels; a simulation-based comparison of FE, IV, Difference-GMM, and System-GMM.
The study established that the data generating process (DGP) effectively reproduced key
features of climate–economy datasets, enabling systematic evaluation of estimator performance.
Using simu lated panels of 135 countries over 50 periods with 1,000 replications, the
design incorporated GDP persistence, CO2 emissions, precipitation and temperature effects,
country-specific heterogeneity, and varying levels of endogeneity. The results confirm that
the DGP successfully captured the primary features of the climate–economy datasets by
including persistence, heterogeneity, endogeneity, and imbalance. The study concluded that
estimator performance varied sys tematically with the strength of endogeneity. Fixed Effects
(FE) and Instru mental Variables (IV) became increasingly biased and unreliable under
higher endogeneity, while Difference-GMM showed moderate improvements but insta bility
in CO2 estimation. System-GMM consistently achieved the lowest bias and RMSE across
weak, moderate, and strong endogeneity levels, making it the most robust and preferred estimator
for analyzing dynamic climate–economy relation ships. The study recommends the use
of simulation frameworks in future climate economy research that incorporate persistence,
heterogeneity, endogeneity, and unbalancedness.