Machine-Learning-Project

A Comparison Between Several Approaches to Forecasting Using Many Predictors

Replication Files for the Paper

“A Comparative Study of Machine Learning Models for Forecasting Key Macroeconomic Indicators: How COVID-19 Betrayed Expectations”

Authors:

• Chiara Perugini, University of Bologna
• Davide Delfino, University of Bologna
• Mohammad Kouzehgar Kaleji, University of Bologna
• Sara Tozzi, University of Bologna

---

Overview

This project trains and compares the forecasting performances of alternative forecasting methods for high-dimensional macroeconomic data. The main R scripts involved are:

• Bayesian_Shrinkage.R
• FarmSelect.R

The outputs include:

1. Factor model forecasts (Principal Components Regression, PCR).
2. Bayesian regression with i.i.d. normal prior (Ridge Regression).
3. Bayesian regression with i.i.d. Laplace prior (LASSO Regression).
4. Factor-Adjusted model (FarmSelect).

The data used in this project were retrieved from:
Barigozzi, M., & Lissona, C. (2024): EA-MD-QD: Large Euro Area and Euro Member Countries Datasets for Macroeconomic Research (Version 12.2023).
Zenodo DOI: 10.5281/ZENODO.10514668.

---

Scripts

Main Scripts

• Bayesian_Shrinkage.R: Trains and compares forecasting methods for PCR, Ridge, and LASSO.
• FarmSelect.R: Implements the Factor-Adjusted Regression Model (FarmSelect).
• Forecasting_methods.R: Produces out-of-sample forecasts for all models.
• Covid_predictions.R: Produces counterfactual forecasts for the best-performing models during the COVID-19 pandemic (2020).

Supporting Scripts in R/functions:

• Bayesian_shrinkage_functions.R: Defines parameters and computes forecasts for PCR, LASSO, and Ridge regression.
• FarmSelect_functions.R: Defines parameters and computes forecasts for FarmSelect.

---

Instructions

Loading Data

Data is stored in the data/EA-MD-QD/ folder.
Use the script Bayesian_Shrinkage.R to load and preprocess the data, ensuring the correct working directory is set.

Troubleshooting source():

If the source() function does not load the R functions correctly, manually run each function in the R/functions folder before executing the main scripts.

---

Additional Scripts

• Descriptive_analysis.R: Provides a descriptive analysis of the dataset.

---

How to Run

1. Clone this repository to your local machine.
2. Set the working directory to the project folder:

   path <- "C:/Users/Davide/Desktop/Alma Mater/SECOND YEAR/Machine Learning/Machine-Learning-Project"
   setwd(path)