Title: Robust Time-Series Forecasting of Unemployment Rates Using Adaptive Quintile Regression and Ensembles
Authors: Maha Hasan Sultan
Abdul amir Taima Bandar
Ahmed Turki Abd Ali
Abdulhussain Hasan Altae
Volume: 10
Issue: 5
Pages: 8-14
Publication Date: 2026/05/28
Abstract:
Precise probabilistic forecasts of the state of the labour market are the basis for macro fiscal strategy and risk management. This study creates a hybrid model that cascades adaptive Quintile Regression Forests (QRF), Gradient Boosting Machines (GBM) and Random Forests (RF) by cascading them in a linear blender to predict the aggregate unemployment rate for nineteen advanced economies. With the World Bank World Development Indicators panel (2011-2021), k-nearest neighbor imputes missing values and three lags of each predictor per year are included to capture dynamics. The hybrid has 1.39 pp mean absolute error and an 80 % prediction interval coverage of 79 % on the pandemic-affected hold-out year 2021 and maintains reliability while having sharper bands than conformal single-model baselines in the literature. Recursive predictions for 2022-2025 depict diverging post-COVID trajectories, Spain remaining more than 13 % and Germany less than 5 %. Results demonstrate the advantage of integrating non-parametric quintile estimation with boosting and bagging learners for consistent socio-economic prediction.