Building an ARIMA Model for Predicting Time Series in Python

Abstract

This study showcases the practical application of the Box-Jenkins model, specifically ARIMA, to predict forthcoming values of a short-term economic indicator in the Republic of Kazakhstan. Data extracted from the Bureau of National Statistics website, covering the period from January 2009 to July 2021, served as the foundation for this analysis. Leveraging the Python programming language, the authors constructed the ARIMA model and conducted thorough time series analysis to uncover temporal patterns within the data. Validation of the model's performance was carried out using data from August 2021 to July 2022. The article presents a comprehensive methodology for model development, encompassing data preprocessing, parameter estimation, and model evaluation stages. Emphasis is placed on the necessity of regular data updates to uphold the accuracy of forecasts, underscoring the practical significance of this study within the domain of time series modeling and forecasting methodologies. As a result, using the constructed model, future values of the series were obtained and a comparison of the predicted values with real data was carried out. To check the error, the mean absolute error in percent (MAPE) was calculated, which was 7.2%. Checking the residual errors showed that the residuals have a normal distribution. This research contributes valuable insights into the application of advanced statistical techniques for economic forecasting, particularly in dynamically evolving contexts like Kazakhstan's economy.