covbayesvar: A Python Package for Large Bayesian VAR Models with COVID Volatility

covbayesvar is a Python package designed to estimate large Bayesian Vector Autoregressive (BVAR) models, incorporating COVID-induced volatility. It is a tool to forecast monthly and quarterly macroeconomic and financial data, leveraging advanced econometric methods to address the unique challenges posed by extreme observations and structural shifts during the pandemic. The package contains python functions which are translated one-to-one with the publically-available MATLAB counterparts, where they were originally written in.

This package facilitates various tasks, including:

• Estimating BVAR Models with COVID volatility: Implements hierarchical priors to shrink parameters in high-dimensional systems and adjust for the heightened volatility and structural changes caused by the COVID-19 pandemic.
• Forecasting: Generates both unconditional (baseline), generalized impulse response functions, and joint distribution of variables at any forecast horizon.
• Scenario Analysis: Explores the effects of hypothetical economic shocks on macroeconomic variables by constructing forecasts conditional on shocks and scenario analyses.
• Entropic Tilting: Anchoring forecasts to long-term policy targets, such as inflation or unemployment rates.
• Identification of Shocks:: Generates impulse responses functions after identifying shocks using Cholesky decomposition where variables are ordered in a pre-defined way
• Versatility: The package can be applied to diverse sets of monthly and quarterly datasets to answer policy-related questions, making it a valuable resource for researchers, policymakers, and financial analysts.
• Extensive Documentation: Includes detailed programming examples, documentation, and Google Colab notebooks to guide users through various use cases.

The methodology draws from seminal works in the field of macroeconometrics, including Giannone, Lenza, and Primiceri (2015), Banbura et. al. (2015), Crump et al. (2021), and Lenza and Primiceri (2022).

For a complete overview, refer to the following

• Research paper and Google Colab jupyter notebook files
• Official documentation
• Github repository
• covbayesvar package

Installation

You can install the package directly from PyPI using pip:

pip install covbayesvar

Applications

Examples of questions this package and Google Colab python files can answer:

• Assess the impact of supply, and demand, simulating the response of changes in any fiscal and monetary policy, etc
• Forecast inflation, unemployment, and other macroeconomic and financial indicators.
• Evaluate economic resilience under stress scenarios and extreme uncertainty.
• Examine the presence of structural breaks during and after COVID pandemic

---

Ways to replicate the results of the research paper

To replicate the results presented in the accompanying research paper, you can choose between two primary methods: running Jupyter notebooks in Google Colab or executing Python scripts locally. Each method has its own advantages and limitations, which are outlined below to help you decide based on your preferences and computational resources.

1. Run the Jupyter Notebooks in Google Colab (Not Recommended for higher-frequency data)

This method involves using pre-configured Jupyter notebooks that are optimized for Google Colab, a cloud-based environment that allows you to run Python code without setting up anything on your local machine.

Pros:

• Don't need to set up an environment or install.
• Suitable for preliminary analyses or demonstrations.
• Does not consume your local machine’s CPU or RAM.
• Dependencies are often pre-installed or easy to install.
• Ideal for testing small code snippets or analyzing small-sized datasets with fewer draws defined in the Ndraws parameter.

Cons:

• The performance slows considerably, especially when estimating the model at the monthly level.
• Colab imposes usage limits, especially for free and Pro accounts.
• May generate out-of-memory errors with large Bayesian monthly VAR models if the number of Monte Carlo simulation draws Ndraws are extensive.

How to Use:

• Open the provided Jupyter notebooks in Google Colab.
• Execute cells sequentially

---

2. Run Python Scripts Locally (Recommended method)

This method involves cloning the repository to your local machine, setting up a virtual environment, and executing Python scripts found in the scripts folder.

Pros:

• Leverages your local hardware, speeding computations, especially with monthly datasets.
• Sessions don't timeout unlike seen in the free version of Colab.
• Easier to debug, modify code and unit-test functions.

Cons:

• Requires cloning the repository, creating virtual environments, and installing dependencies.
• Performance depends on your local machine’s CPU, and RAM.

How to Use:

• Follow the steps outlined below on how to proceed with running the python scripts locally.

Cloning the Repository

To get started, clone the repository to your local machine:

git clone https://github.com/joshi27s/Large-BVAR-with-Covid-Volatility.git Large_BVAR
pwd # prints the present working directory to know where the Large_BVAR folder was cloned
cd Large_BVAR

Installing from Source

If you'd like to install the latest version of the package from the source code, follow these steps:

1. Ensure you have Python 3.8 or higher installed.
2. Create a virtual environment and activate it:

python3 -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

3. Install the package and its dependencies:

pip install .

Requirements

The package requires the following Python packages:

• numpy
• pandas
• scipy
• matplotlib
• seaborn
• openpyxl
• plotly

You can also install all dependencies using the requirements.txt file:

pip install -r requirements.txt

How to Execute the Model?

Using monthly data

In addition to the Google Colab notebooks, the scripts folder in the GitHub repository contains Python files to run the model on monthly data in the following order. To have a deeper understanding of the code, refer to the accompanying research paper, and the official documentation of the functions. The package relies on functions stored in the large_bvar module that you can find in the covbayesvar folder. In particular, section 4, and 5 elaborate the method and results obtained from the following scripts

1. Descriptives.py:

   • Generates time series graphs of level and transformed data.
   • Transforms are applied only to variables not measured in percentage terms.

2. main.py:

   • Runs the BVAR model, including:
     • Unconditional (baseline) forecasts.
     • Conditional forecasts with shocks applied to two variables.
     • Scenario analysis and joint predictive densities.

3. Entropic_Tilting.py:

   • Anchors the long-run baseline forecasts to long-run targets mentioned in the Summary of Economic Projections (SEP) by the FOMC.

4. MCMC_Sims.py (optional):

   • Tests the performance of the BVAR model using a simple Monte Carlo simulation.

# example of how to run the scripts via terminal/command prompt
python Descriptives.py
python main.py
python entropicTilting.py

Using quarterly data

Navigate to the scripts/Quarterly folder and run the following scripts that contain examples and results described in section 6 of the research paper:

1. Descriptives.py

2. main_quarterly.py

3. entropicTilting.py

4. increaseGDP_longRun.py

5. increase1Y_unrestricted.py

6. increase1Y_Cholesky.py

7. financialTurmoil.py

8. structuralBreak.py

9. increaseAAA_restricted.py

10. increaseOilPrice_unrestricted.py

Replicating figures 2,3 and 4 from Lenza and Primiceri (2022): How to Estimate a VAR after March 2020

Navigate to the scripts/LP 2021 folder and run the following scripts that contain examples and results described in section 7:

1. Baseline_May2021.py

2. Baseline_June2020.py

3. CV_Feb2020_June2020.py

4. CV_Feb2020_May2020.py

5. CV_May2021.py

6. generate_figures.py

Citation

If you use covbayesvar in your work, please cite the following research papers that inspired its methodology:

1. Banbura, M., Giannone, D., & Lenza, M. (2015). Conditional forecasts and scenario analysis with vector autoregressions for large cross-sections. International Journal of Forecasting, 31(3), 739–756. https://doi.org/10.1016/j.ijforecast.2014.08.013

2. Giannone, D., Lenza, M., & Primiceri, G. E. (2015). Prior selection for vector autoregressions. The Review of Economics and Statistics, 97(2), 436–451. https://doi.org/10.1162/REST_a_00483

3. Crump, R. K., Eusepi, S., Giannone, D., Qian, E., & Sbordone, A. M. (2021). A large Bayesian VAR of the United States economy. NY Fed Staff Report. https://www.newyorkfed.org/research/staff_reports/sr976

4. Lenza, M., & Primiceri, G. (2022). How to estimate a VAR after March 2020. Journal of Applied Econometrics, 37(4), 688–699.https://doi.org/10.1002/jae.2895

5. Clements, M. P., & Galvao, A. B. (2024). Macroeconomic forecasting using BVARs. In Handbook of Research Methods and Applications on Macroeconomic Forecasting (Chapter 2, pp. 15–42). Cheltenham, UK: Edward Elgar Publishing. https://pureportal.strath.ac.uk/en/publications/macroeconomic-forecasting-using-bvars

You can additionally cite the software repository itself:

@misc{joshi_sudiksha_2025_covbayesvar,
  author = {Sudiksha Joshi},
  title = {covbayesvar: A Python Package for Large Bayesian VAR Models with COVID Volatility},
  url = {https://github.com/joshi27s/Large-BVAR-with-Covid-Volatility},
  year = {2025}
}

Contributions

You are welcome to contribute in any capacity to aid in expanding the project from creating an issue, reporting a bug, suggesting an improvement, to forking the repository and creating a pull request to the development branch.

Independence Statement

This package, accompanying research paper, codes and supplementary materials are not affiliated with, endorsed from or sponsored by Amazon.com, its subsidiaries or its employees. I independently wrote the functions of the package for research purposes before joining Amazon.com. Prior to my employment at Amazon, I had mapped the python functions one-to-one with the publicly-available MATLAB functions. I am solely responsible for any omissions or errors.