Statistical Modeling of Non-stationary Carbondioxide Emissions versus Income Data

Abstract

This paper explores the relationship between per capita carbon dioxide emissions and per capita income for Sri Lanka, Japan and United States. Unit root tests carried out showed that these variables are nonstationary. Since non-stationary data outperform regression models developed with them, regression models were developed in this study with stationarized variables. For the best fitted models for Sri Lanka, Japan and United States, it was found that the per capita emissions were driven mainly by its own autoregressive term, GDP per capita and its autoregressive term. For the best fitted models, statistical characteristics such as Mean Square Error, Akaike Information Criterion and Bayesian Information Criterion took the least values among the models studied. Also, the residuals of the best fitted models were found to posses the characteristics of white noise and they were normally distributed.