Estimating Heterogeneous Agent Models With Neural Networks

Example code for the simple 3 equation New Keynesian model with analytical solution from "Estimating Heterogeneous Agent Models With Neural Networks" by H. Kase, L. Melosi, and M. Rottner

Github Repo: https://github.com/tseep/estimating-hank-nn

Modified by Qingyuan Fang (JHU) on Nov 20, 2024

In preparation for JHU Economics Machine Learning Reading Group meeting on Nov 22, 2024

1. ''pyproject.toml'': add dependencies, delete license Reference : https://packaging.python.org/en/latest/guides/writing-pyproject-toml/
2. "solution_QY.ipynb": add more comments to the first part of the authors' notebook ("Extended Model Solution with Neural Network Approach"), and delete the second part ("Particle Filters") Ideal for individuals with a basic understanding of neural networks (like stochastic gradient descend) who are interested in training one using PyTorch.

Structure of the folder

• ./examples
  • ./save
    • ./authors_backup: include the two networks trained by the authors
  • analytical.ipynb: notebook by the authors
  • solution_QY.ipynb: modified notebook by Qingyuan, focusing on using the extended NN to solve the model
• ./src/estimating_hank_nn
  • the source code (with some helper functions) by the authors

Create a virtual environment and install the packages

After cloning the repo

git clone https://github.com/QingyuanFang/KMR22_JHUML_20241122.git

Mac

cd [your path]/KMR22_JHUML_20241122

python -m venv .venv
source .venv/bin/activate
pip install --upgrade pip
pip install .
deactivate

Windows

cd [your path]\KMR22_JHUML_20241122

python -m venv .venv
.venv\Scripts\activate
pip install --upgrade pip
pip install .
deactivate

What does the code above do?

1. First, it creates a virtual environment, named ".venv", inside the "KMR22_JHUML_20241122" folder. Note: on Mac, press "command+shift+." to make it visible in Finder.
2. Then, it activates that environment and installs all python packages the project requires. See the "dependencies" section in pyproject.toml.

numpy
scipy
matplotlib
torch
tqdm

3. It also installs the package "estimating_hank_nn". It is inside the ./src folder and has some helper functions written by the authors.
4. Finally, It deactivates the environment.

To proceed

Open solution_QY.ipynb in your favorite code editor, choose ./venv as the kernel, and you are ready to go!

Reference

Kase, Hanno, Leonardo Melosi, and Matthias Rottner. Estimating nonlinear heterogeneous agents models with neural networks. Centre for Economic Policy Research, 2022.