Assessing Non-Linearity and Stationarity in the Time Series of Albania’s Annual Emissions of CO2 from Land-Use Change

Abstract

The annual emissions of CO₂ from land-use change in Albania are the main focus of this research. The aim is to analyze the presence of non-linearity and stationarity. A mixed-methods strategy is used, which combines descriptive, inferential, and exploratory data analysis in time series data. A data sample was obtained from the Our World in Data website, spanning from 1850 through 2022. After the Isolation Forest technique was employed to identify outliers in the time series, the Long-Short-Term Memory model was used to impute them. Exploratory data analysis was applied to the original and imputed time series to ensure that the basic characteristics of the initial data distribution were preserved. Non-linearity and stationarity were checked in the imputed time series before and after applying the first differences. Non-linearity was assessed using the BDS test and the Teräsvirta Neural Network test. In the presence of non-linearity, stationarity was analyzed using the KPSS test, the Zivot-Andrews Unit Root test, and the Breitung test. The first differencing application transformed the non-stationary series into a stationary one, but it was insufficient to eliminate non-linearity. This highlights the complex nature of CO₂ emissions data and the need for sophisticated modeling techniques.