This repository contains code for solving a New Keynesian model with a Noisy Taylor Rule, as described in "Rough Notes".

1. NK_solution – Script for a single-seed solution with a -1 SD TFP shock.
   Produces:

   • IRF plot
   • IRF CSV file: IRF_shock_z_wide.csv

2. NK_seed_experiment – Runs experiments with 10 random seeds.
   Produces:

   • IRF_seed_* files and plots seed_torch_exp_{date} (for MIT shock IRFs)
   • IRF_noisy_seed_* files and plots seed_torch_exp_{date} (for noisy shock IRFs)

3. NK_solution_seed – Wrapper for NK_solution, used within NK_seed_experiment.

4. NK_NN_architecture – Neural network architecture setup module.

5. NK_residuals – Main module for computing loss functions.

6. NK_simulations – Contains resampling functions used for solving the model (both log-linear and with arbitrary policy functions).

7. NK_saving_plotting – Functions for saving IRF results to CSV and plotting IRFs.

8. params – Parameter setup module.

9. NK_utils – Additional helper functions.

The NK model is solved in four main steps:

1. Simulate the initial grid
   Generate a grid over ((k, v, z)) using the log-linearized model.

2. Deterministic model solution
   Solve the deterministic version of the New Keynesian model using the initial grid.

3. New grid simulation
   Simulate a new grid based on the policy rules from the deterministic solution.

4. Full stochastic solution
   Solve the full (stochastic) version of the model on the new grid, with adaptive refinements.