midasmlpy

Python implementation of the midasml approach - providing estimation and prediction methods for high-dimensional mixed-frequency time-series data

Authors

• Jonas Striaukas - jstriaukas
• Kris Stern - krisstern

About

The midasmlpy package implements estimation and prediction methods for high-dimensional mixed-frequency (MIDAS) time-series and panel data in regression models. The regularized MIDAS models are estimated using orthogonal (e.g. Legendre) polynomials and the sparse-group LASSO estimator. For more information on the midasmlpy approach there are references in the footnotes¹²³.

The package is equipped with the fast implementation of the sparse-group LASSO estimator by means of proximal block coordinate descent. High-dimensional mixed frequency time-series data can also be easily manipulated with functions provided in the package.

The .f90 source files

These can be found at the midasmlpy/src directory. Compiled .so files can be found at midasmlpy/compiled directory. Please note that we have taken the .f90 source code for sparse-group LASSO from the repo of the R package sparsegl on hosted GitHub at https://github.com/dajmcdon/sparsegl. We have taken their source code in accordance with their GPL-2.0 license as is without any modification as of November 2nd, 2023 UTC. What can be found at midasmlpy/src/sparsegl are copies of their .f90 code.

Compiling the .f90 source files with f2py on a Mac

The gfortran command to compile .mod files from .f90

Run:

gfortran -c fortran_file.f90

The f2py command to compile .so files from .f90

Run:

python -m numpy.f2py -c fortran_file.f90 -m fortran_file

Note: Be sure the python executable above matches the one in your environment.

Software in other languages

• A Julia implementation of the midasml method is available here.
• A MATLAB implementation of the midasml method is available here.
• An R implementation of the midasml method is available here.

Installation

To install the midasmlpy package, download the files and at the directory of files run:

pip install .

Development

To install the midasmlpy package for development, do the following instead in “editable” mode:

pip install -e .

Testing

To run the tests, run:

pytest --pyargs src

Remarks

In case you are running the code on a different platform, you can compile the Fortran code sglfitF.f90 by using f2py which is part of numpy.

Footnotes

1. Babii, A., Ghysels, E., & Striaukas, J. Machine learning time series regressions with an application to nowcasting, (2022) Journal of Business & Economic Statistics, Volume 40, Issue 3, 1094-1106. https://doi.org/10.1080/07350015.2021.1899933. ↩

2. Babii, A., Ghysels, E., & Striaukas, J. High-dimensional Granger causality tests with an application to VIX and news, (2022) Journal of Financial Econometrics, nbac023. https://doi.org/10.1093/jjfinec/nbac023. ↩

3. Babii, A., R. Ball, Ghysels, E., & Striaukas, J. Machine learning panel data regressions with heavy-tailed dependent data: Theory and application, (2022) Journal of Econometrics. https://doi.org/10.1016/j.jeconom.2022.07.001. ↩