Nowcasting GDP using machine learning methods

This repository provides code to reproduce results in the paper "Nowcasting GDP using machine learning methods" by Dennis Kant, Andreas Pick (EUR) and Jasper de Winter (DNB) that is forthcoming in AStA Advances in Statistical Analysis. The scripts are most easily ran in Rstudio following these steps: Download or clone the repository and double-click AstA.proj to open the project in RStudio. Next, open Main and run the models and/or calculate DM-tests and report RMSFEs.

• The outcome of all Figures and Tables are collected in the directory Results. The folders dm and rmsfe contain the outcomes in Table 2, 3 and 4 of our paper and show the outcome of the Diebold-Mariano test and RMSFE, respectively. The folder fcst contains the raw backcasts, nowcasts and forecasts for all models underlying these results. The folder graphs contains pdf-files for Figure 3, 4 and 5;
• The setups in the folder Set-ups_publiccan be used to reproduce all modeloutcomes and figures in our paper;
• Run R-file MAIN.R to reproduce all outcomes in Table 2, 3 and 4;
• Adjust line 16 of the R-file MAIN.R to produce outcomes for one of the models in our paper (adjust index in line 16), i.e. AR(1) model, AR, Random Walk RW, Factor-augmented Mixed Data Sampling regression MIDAS-F, Least Absolute Shrinkage and Selection OperatorLASSO, Elastic Net EN, Random Subspace regression RS, Random Projections RP, Random ForestRF.
• the backcasts, nowcasts and forecasts of the models are written to the directory Results_public\fcst. Figures are written to Results_public\graphs. Auxiliary files written to Results_public\other;
• Run the Matlab file TABLE_5_7.m to reproduces Table 5, 6 and 7 of our paper;
• For confidentiality reasons some series had to be removed from the data-file (notably all data-series in Table A.2 that only have codes and no links. The forecasts, RMSFEs and contributions will therefore deviate from the outcomes in the paper);
• The text-file SessionInfo.txtcontains information on the R-(packages) and Matlab versions used to estimate the models.

DISCLAIMER: The code is provided "as is". The authors make no assertions as to its performance or effects if run, provides no warranties of any kind, and disclaims any implied warranties, express or implied, including but not limited to the warranties of merchantability, fitness for a particular purpose or non infringement.