GDP Forecasting: GDFM vs Econometric Benchmarks

This repository features a professional MATLAB framework designed for macroeconomic forecasting. The project specifically focuses on predicting GDP growth by leveraging high-dimensional datasets through factor models.

Objective

The goal of this project is to perform h-step ahead forecasting of GDP (Gross Domestic Product). It evaluates whether complex dynamic factor models can outperform simpler univariate and multivariate econometric benchmarks.

Models and Comparison

The framework performs a rigorous comparison between:

• GDFM (Generalized Dynamic Factor Model): The primary model, which extracts dynamic common components from a large panel of variables to capture the underlying signal of the economy.
• Static PCA: A benchmark that uses standard Principal Component Analysis to build a factor-based forecast without accounting for lead-lag relationships.
• VAR (Vector Autoregression): A multivariate benchmark where the additional variables are selected via a data-driven correlation ranking with the target's common component.
• AR(1) (Autoregressive): The standard univariate benchmark for time-series persistence.
• Random Walk (RW): The naive benchmark (no-change forecast).

Structure

• data/: Contains the dataset processed_data copia.xlsx.
• script/: Contains the main forecasting script main_gdp.m.
• tools/: Directory for external libraries and functions.
  • gdfm copia/: Core GDFM estimation and forecasting functions.
  • ABC_crit_fast/: Alessi-Barigozzi-Capasso (2010) factor selection criterion.

Citations and Acknowledgments

This project implements methodologies from the following seminal works:

Generalized Dynamic Factor Model (GDFM)

• Forni, M., Hallin, M., Lippi, M., & Reichlin, L. (2005). "The Generalized Dynamic Factor Model: One-Sided Estimation and Forecasting."
• Forni, M., Hallin, M., Lippi, M., & Reichlin, L. (2000). "The Generalized Dynamic Factor Model: Identification and Estimation."

ABC Factor Selection Criterion

• Alessi, L., Barigozzi, M., & Capasso, M. (2010). "Improved Performance of the Hellin and Liška Test for Determining the Number of Factors." Journal of Applied Econometrics.

Data Source

• The dataset used in this framework is sourced from Prof. Matteo Barigozzi's official website.

Methodology Summary

• Evaluation: Out-of-Sample (OOS) rolling-origin evaluation.
• Rolling Window: 80 quarters (fixed).
• Forecast Horizon: 8-step ahead (h=8).
• Evaluation Metric: Root Mean Square Error (RMSE) and Relative RMSE (benchmarked against AR1).