This study provides a comprehensive empirical analysis of unemployment dynamics in Somalia using advanced time series forecasting techniques. The research evaluates and compares the performance of both single and hybrid time series models in capturing the behavior of unemployment in a fragile and conflict-affected economy. Annual unemployment data covering the period 1991–2024 were obtained from reliable secondary sources, including World Bank Open Data based on ILO estimates. The study applies a range of univariate time series models, including ARIMA, ETS, TBATS, Theta, ARFIMA, and NNAR, alongside multiple hybrid model combinations. The dataset was divided into training (1991–2018) and testing (2019–2024) subsets to ensure robust out-of-sample validation. Stationarity was assessed using ADF, PP, and KPSS tests, and forecasting performance was evaluated using MAPE, sMAPE, and Theil’s U statistics. The findings indicate that hybrid models outperform most single-model approaches in forecasting unemployment in Somalia. In particular, the ETS–NNAR hybrid model demonstrated the highest predictive accuracy, while ARFIMA performed best among individual models. Forecast results for the period 2025–2030 suggest a relatively stable unemployment trajectory, with moderate uncertainty reflecting the influence of external shocks such as economic instability, climate variability, and political disruptions. This study contributes to the empirical literature by introducing a robust hybrid forecasting framework tailored to high-volatility and data-constrained environments. The results provide valuable evidence for policymakers to support forward-looking labor market planning, macroeconomic stabilization strategies, and employment policy design in Somalia.
Darod²* et al. (Thu,) studied this question.