Enhancing GDP Growth Forecasting with LSTM, GRU, and Hybrid Model: Evidence from South Korea

Abstract

This study examines the application of Long Short-Term Memory (LSTM) networks, Gated Recurrent Units (GRU) along with traditional econometric models in forecasting South Korea’s GDP growth. A hybrid framework is also developed, integrating these models through a meta-learner to capitalize on their complementary strengths. LSTM, with its ability to model nonlinear relationships and capture long-term dependencies, demonstrates accuracy improvements, especially during periods of economic volatility, such as the COVID-19 pandemic. The hybrid model further enhances forecasting performance by dynamically combining the strengths of LSTM and GRU with traditional approaches. This study provides a robust methodological contribution by uniting machine learning and econometric techniques, demonstrating their combined potential for enhancing forecasting accuracy and effectively addressing the complexities of diverse economic conditions.