Advanced Machine Learning Approaches for Enhanced GDP Nowcasting in Syria Through Comprehensive Analysis of Regularization Techniques

Abstract

This study addresses the challenge of nowcasting Gross Domestic Product (GDP) in data-scarce environments, with a focus on Syria, a country facing significant economic and political instability. Utilizing a dataset from 2010 to 2022, three machine learning algorithms Elastic Net, Ridge, and Lasso were applied to model GDP dynamics based on macroeconomic indicators, commodity prices, and high-frequency internet search data from Google Trends. Among these, the Lasso regression model, noted for its variable selection and sparsity promotion, proved most effective in capturing Syria's complex economic realities, achieving the lowest Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE). This accuracy highlights the Lasso model's capability to identify robust economic relationships despite limited data, thereby reducing overfitting and improving forecast generalizability. The study underscores the significant impact of non-traditional indicators, such as Google Trends Agriculture (GTA) and Google Trends Consumption (GTC), on GDP growth, offering valuable insights for policymakers and analysts in data-scarce environments. The findings support the use of machine learning techniques, particularly Lasso regression, as powerful tools for economic forecasting, enhancing informed decision-making in challenging settings.