Application of dynamic models in forecasting the total population of the United States

Dynamic models have been widely cited in predicting criminal population, residential electricity consumption, food prices and other objects. However, for total population predictions, dynamic models are rarely used. In this study, we will analyse the relationship between 13 variables such as CPI, grain prices, and medical expenditures and the total population of the United States, then combine it with the ARIMA model to generate a time series dynamic regression model. The conclusion is that, according to the parameters of the final model, two predictors (CPI and the number of crimes) and one interaction term (the product of the poverty rate and unemployment rate) are significantly related to changes in the population. Ultimately, the model performed well on the test set and was remarkably accurate for population prediction five years later. This report screens various factors influencing the total population and provides a broader background for applying dynamic models. In addition, this study also provides directions for subsequent research on more efficient dynamic models.