Modeling Agro-Environmental Dynamics with VAR under Climate and Volcanic Variability: A Simulation-Based OLS–GLS Comparison from East Flores, Indonesia

Abstract

Vector Autoregressive (VAR) modeling provides a flexible framework for examining dynamic interdependencies among multiple time-series variables within agro-environmental systems. Reliable estimation of VAR parameters is critical for accurate inference and forecasting, particularly when vegetation indicators and climatic variables are affected by correlated and heteroskedastic disturbances. This study conducts a simulation-based evaluation of Ordinary Least Squares (OLS) and Generalized Least Squares (GLS) estimators for VAR models under alternative error variance structures. A trivariate VAR(1) specification is employed to characterize agro-environmental dynamics involving vegetation condition, represented by the Normalized Difference Vegetation Index (NDVI), total precipitation, and near-surface air temperature. Monte Carlo simulation experiments are performed to assess estimator performance under independent and correlated disturbance scenarios using bias, variance, and mean squared error as evaluation metrics. To demonstrate empirical relevance, the methodological framework is illustrated using monthly agro-climatic observations from East Flores Regency, East Nusa Tenggara, Indonesia, spanning January 2015 to November 2024. The dataset is compiled from MODIS NDVI products and NASA POWER climatic records. The study setting also reflects recent environmental disturbances linked to volcanic activity in East Flores, which may amplify variability and cross-dependence among agro-environmental variables. Simulation results indicate that OLS estimators remain unbiased and exhibit satisfactory efficiency when error terms are independent. In contrast, under correlated or heteroskedastic disturbances—conditions frequently encountered in agro-environmental time-series data—GLS provides more efficient parameter estimates with lower estimation error. The empirical illustration corroborates the simulation outcomes by demonstrating the role of climatic interactions in shaping vegetation dynamics. Overall, the findings underscore the importance of incorporating appropriate error structures in multivariate agro-environmental time-series analysis and offer methodological insights for agricultural and life science applications employing VAR models.