What is FRED-QD+?
FRED-QD+ is an extended, real-time version of the FRED-QD dataset by McCracken and Ng. The extension increases vintage history from 2018 to 2001, expands the variable set, and incorporates mixed-frequency nowcasting tools to include curren-quarter data and impute missing values in the dataset.
Approximately 250 variables with a quarterly history back to 1959
Real-time vintages extended back to 2001
15 new variables added
Mixed-frequency nowcasting layer
Improved missing-value imputation via dynamic factor models
FRED-QD+.pdfWhy I built this dataset:
For my Forecasting Projects, I wanted a real-time machine learning ready dataset for macroeconomists. The natural choice here is the FRED-QD dataset by Michael McCracken and Serena Ng. It comprises of around 250 variables and contains quarterly data until 1959. However, its realtime construction so far only dates back to May 2018 which may be too short for proper out-of-sample analysis. Furthermore, due to the quarterly structure, the data has signifcant lags, even for variables only available in monthly frequencies. These current-quarter observations can be used to nowcast the dataset alongside improved missing value imputation, all done with Dynamic Factor Models.
If you are interested, check out the original paper as well as my technical note.
Click here to download the data and replication code.
