TIME SERIES FORECASTING USING ARIMA MODELS

Abstract

This article explores the application of ARIMA (AutoRegressive Integrated Moving Average) models for forecasting the population of Ukraine. Time series forecasting plays a crucial role in demographic analysis, long-term socio-economic planning, and the development of effective public policies. The study applies the classical Box-Jenkins methodology to build ARIMA models of different orders and assess their forecasting accuracy. The modeling was performed using official data on the population of Ukraine for the period from 1990 to 2019, provided by the State Statistics Service of Ukraine. A population forecast for the year 2020 was generated based on various ARIMA configurations, and the results were compared to actual statistical data.

The study evaluates twelve ARIMA models with different combinations of autoregressive (p), differencing (d), and moving average (q) parameters. The selection of the optimal model was based on two main criteria: the root mean square error (RMSE) of forecast accuracy and the absolute difference between forecasted and actual values for 2020. The research highlights the advantages and limitations of ARIMA models in short-term population forecasting and proposes the use of RMSE minimization as a reliable criterion for model selection. The findings confirm the effectiveness of ARIMA models in demographic forecasting tasks, which are essential for data-driven decision-making in public administration, urban development, and social policy.

Keywords: time series, population forecasting, ARIMA model, Box-Jenkins method, model selection criteria.