Luca Attolico

KEYWORDS: Economic growth; China; Greater Bay Area; Nowcasting; Machine Learning

Abstract

Economic policy requires a real-time view of the economy. This research builds a nowcasting pipeline for quarterly GDP using regularized regressions, dimension-reduction methods, ensemble learners, and neural networks. Expanding-and-rolling-window out-of-sample validation, Bayesian hyperparameter optimization, and a compute-aware design ensure robustness and efficiency. Uncertainty is quantified via regime-aware prediction intervals and model-specific explainability with feature-importance rankings and confidence intervals. Model selection uses the Model Confidence Set; aggregation employs simple, weighted, and exponentially weighted averages. Formal predictive-ability tests confirm outperformance relative to traditional econometric benchmarks.

• EMAIL: luca.attolico@gmail.com
• LINKEDIN PROFILE
• ORCID: 0000-0002-8649-5440