| Shanice Douglas |
This study explores the application of the ARIMA (Auto Regressive Integrated Moving
Average) model for forecasting U.S. unemployment rates using monthly data from 1948 to 2015.
The ARIMA (1,2,0) model was selected based on the data's non-seasonal and non-stationary
properties and was used to generate a 48-month forecast, which was then compared to actual
unemployment data from 2016 to 2019. While ARIMA is well-documented as a robust tool for
short-term forecasting in economic data, the model in this study exhibited significant limitations
in capturing the cyclical nature of U.S. unemployment trends. The forecast produced by the
model remained relatively flat, failing to reflect the actual downward trend observed during the
period, with increasing confidence intervals indicating growing uncertainty over time. Diagnostic
tools, such as the Akaike Information Criterion (AIC) and residual analysis, confirmed the
model's adequate fit to historical data but highlighted its inability to account for cyclical
economic influences necessary for accurate long-term predictions. Given the U.S. unemployment
rate's cyclical behavior, this study recommends considering alternative time series models, such
as Exponential Smoothing (ETS), Holt-Winters, or SARIMA models, to enhance the accuracy of
future forecasts. These approaches could provide more reliable and actionable insights for
policymakers and economists by accounting for cyclical and seasonal trends in the data. |