This repository contains the replication files for the paper Backtesting Global Growth-at-Risk by Christian Brownlees and Andre B.M. Souza which is available on SSRN at the address https://ssrn.com/abstract=3461214

Christian Brownlees and Andre B.M. Souza

MATLAB The code has been tested with the MATLAB releases R2017a and R2019a

To replicate the out-of-sample results run the script gar_replication.m. The script will create Tables 4 to 6 of the paper. The tables will be stored as individual CSV files in the directory tables.

Important Disclaimer: The data used in this study was downloaded from the following sources in June 2019.

• GDP from the OECD Database
• NFCI from the IMF.
• ST_INTEREST from the OECD database
• LT_INTEREST from the OECD database
• BAA_CREDIT from the St. Louis Fed.
• AAA_CREDIT from the St. Louis Fed.
• CREDIT_STATISTICS from the BIS database
• PROPERTY PRICES from the BIS database
• WUI from the Policy Uncertainty website
• GPR from the Policy Uncertainty website
• EPU for several countries, all of which can be found in the Policy Uncertainty website:
  • Australia
  • Canada
  • Europe
  • Mexico
  • South Korea
  • United Kingdom
  • United States
  • Japan
  • Greece
  • Ireland
  • Netherlands
  • Sweden

• rq.m: Function to compute quantile regression. Source: Vulnerable Growth Replication Files (Adrian et al, 2019)
• QuantilesInterpolation.m: Function to interpolate quantiles and get a Skewed T.

• covnw.m: Function to estimate HAC covariance matrices.
• olsnw.m: Function to perform inference on ols parameters with HAC.
• normloglik.m: Log Likelihood for the standard normal.

The following files are modified versions of the above. We use them to perform DQ tests for h>1 step ahead forecasts.

• covnw_dq.m: Function to estimate HAC covariance matrices for binary hit variables.
• olsnw_dq.m: Function to perform robust inference on binary hit variables.

• L1QR.m: L1 penalized quantile regression. Depends on SDPT3-4.0. See below.

• SDPT3-4.0