Evaluating the Performance of Time Series Models in Forecasting the Unemployment Rate in Libya

Abstract

This paper aims to analyze and predict the unemployment rate in Libya using a set of time series models. The analysis depended on annual unemployment rate data for the period 1991-2024 and applied trend regression models, exponential smoothing models, a random walk model, and ARIMA models. Series stationarity test were carried out using the ADF test, and the results showed that the series becomes stationary after the 2nd difference. The estimated models were evaluated using residuals diagnostic tests, including Ljung-Box test, Jarque-Bera test, and ARCH-LM test. The results showed that trend regression models failed to meet the residual independence condition while the other models meet all the required conditions. Forecasting performance was evaluated using RMSE, MAE, and MAPE measures. The results showed that Brown exponential smoothing model out-of-sample predictive accuracy was superior. Accordingly, the chosen model was used to predict unemployment rate over the next five years, with confidence intervals at 95% level. The study's results confirm the efficiency of exponential smoothing models, especially the Brown model, in forecasting regular-trend time series in practical applications.