Forecasting of Carbon Emissions in India Using Bayesian ARIMA and BSTS Approaches: Evidence from Environmental Sustainability

This study uses a completely Bayesian method to analyze and forecast per-capita CO2 emissions in India by comparing the performance of the Bayesian Autoregressive Integrated Moving Average (ARIMA) and Bayesian Structural Time Series (BSTS) models. This study intends to show that the Bayesian formulation of the ARIMA model can provide better predictive performance in specific situations, even though prior research has frequently emphasized the advantages of the BSTS model, particularly in capturing intricate structures in environmental and economic time series. This investigation, which focuses on long-term historical per-capita CO2 emissions data from 1858 to 2023, takes a different modelling approach and comparison framework than previous studies. Choosing the best ARIMA model order is an important initial step based on the Leave-One-Out Information Criterion (LOOIC). The rstan package was used to perform parameter estimates for the ARIMA and BSTS models using the Hamiltonian Monte Carlo technique. Bayesian criteria, including the Widely Applicable Information Criterion (WAIC) and Leave-One-Out Information Criterion (LOOIC), were used to assess the model performance. The findings show that, in terms of forecast accuracy for India’s per-capita CO2 emissions, the Bayesian ARIMA model routinely outperforms the BSTS model, even with its more straightforward structure.