The Forecasting MacroModelBase

About the project

A common framework for comparing macroeconomic forecasts, understanding uncertainty, and evaluating models in real time.

Developed at the Institute for Monetary and Financial Stability, Goethe University Frankfurt.

Team

This project was developed by the Institute for Monetary and Financial Stability.

Portrait of Luca Schmitz

Luca Schmitz

Development and Research
Website
Portrait of Volker Wieland

Prof. Volker Wieland, PhD

Supervision
Website

Methodology

Our forecasting methodology combines dynamic stochastic general equilibrium (DSGE), time-series, machine-learning, neural network, and foundation models, as well as expert forecasts. The platform offers individual model forecasts and forecast combinations, with performance evaluated across historical vintages and forecast horizons.

Model description

The project includes a diverse set of macroeconomic models, including DSGE models used in central banks and academic institutions, as well as simpler autoregressive models for benchmarking.

DSGE model collection Overview of DSGE models used in the platform All model descriptions Documentation covering the broader model set

Data description

Forecasts are based on historical macroeconomic indicators including GDP, inflation, and interest rates. Data vintages are carefully managed to replicate the real-time information sets available at each forecast date.

DSGE data documentation Detailed overview of DSGE-related data inputs High-dimensional dataset Description of the macro dataset

Reading historical forecast comparisons

Vintage quarters identify the forecast origin, when the forecast was made. The target quarter is the origin plus the horizon; h0 is a nowcast.

  • Foundation models: All foundation models except MacroCast are not vintage-consistent. They are pretrained on large datasets from the 2020s, which can contain information unavailable at historical forecast origins.
  • Growth-at-Risk: Results before 2011 are not vintage-consistent because the NFCI input was unavailable at those historical forecast origins.
  • Fair model: Forecasts are unavailable for some historical origins, including during COVID. Coverage varies across vintages.
  • Greenbook: GBbeforeSPF and GBafterSPF forecasts are unavailable for the most recent five years.

References