Time Series–Based Population Forecasting: A Comparative Analysis of Linear Regression and Exponential Smoothing

This study compares linear regression and exponential smoothing based on time-series methods for the estimation of long term populations at national and regional levels. The UN World Population Prospects provide annual population figures for the time period 1950-2023. The countries included in this analysis are Asia, Africa, Europe, the U.S.A., and Turkey. The results of linear regression (LR) show that it has a very good explanatory value in regions that have shown relatively stable population development (Asia, the U.S.A., and Turkey, R² > 0.99). On the other hand, the explanatory value of LR is less good in regions that have shown nonlinear population development, especially in Africa (R² = 0.946) and Europe (R² = 0.877), as evidenced by large forecast errors (up to 6.99 × 10⁹ MSE). In comparison to LR, the exponential smoothing (ES) model was able to produce better forecasts for all of the analyzed regions and reached R² values greater than 0.9999; furthermore, it produced smaller forecast errors (for example, MSE = 4.10 × 10⁵ for Africa and MSE = 2.50 × 10⁴ for Turkey). Based on these findings, it can be expected that the population will continue to grow in Africa (approximately 1.98 billion people) and in Asia (approximately 5.67 billion people) until 2043; however, there will be moderate population increases in Turkey (approximately 98 million people) and in the U.S.A. (approximately 360 million people); and there will be no significant changes in the European population (approximately 700 million people).