Forecasting Qatar’s real GDP Growth: Horizon-dependent Performance of classical and Machine Learning Models

Abstract

Accurate forecasting of real GDP growth is essential for macroeconomic planning, particularly in small, hydrocarbon-dependent economies where data availability is limited and economic dynamics are structurally volatile. This study evaluates the relative forecasting performance of classical and machine learning models for Qatar’s real GDP growth. The analysis uses quarterly macroeconomic data spanning 2010Q1-2025Q3 and includes 18 variables capturing output, prices, external sector conditions, fiscal indicators, and financial dynamics. A traditional SARIMA benchmark, specified via AutoARIMA, is compared with three nonlinear models: Extreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM), and Gated Recurrent Units (GRU). Forecasts are generated for both quarter-on-quarter and year-on-year growth at horizons of 1, 4, and 8 quarters ahead using an expanding-window, multi-horizon evaluation framework. Feature engineering incorporates lag structures, rolling statistics, and macroeconomic predictor blocks designed to capture temporal persistence and structural dynamics. The results reveal clear horizon-dependent differences in model performance. SARIMA achieves the highest accuracy at short horizons (one-quarter ahead), reflecting the advantages of parsimonious models in small-sample environments. In contrast, machine learning and deep learning models, particularly XGBoost and LSTM, become increasingly competitive at longer horizons, where nonlinear relationships and richer predictor information can be exploited. Overall, the findings suggest that model performance is horizon-dependent rather than universally dominated by any single model class. The study contributes to the macroeconomic forecasting literature by providing a multi-horizon comparison of classical and machine learning models in a data-constrained emerging economy, and by highlighting the importance of evaluation design, predictor selection, and robustness to structural change. These insights have direct policy relevance for Qatar, where timely and reliable forecasts of non-oil GDP are critical for economic diversification and planning.

Date of Award	2026
Original language	American English
Awarding Institution	HBKU College of Science and Engineering