sw_stresstest's Introduction

System-Wide Stress Test Model

This is the reference implementation of the system-wide stress test model, as described in this paper. This repo uses public data from EBA, while the result in the paper is generated using confidential data. And so this repo is not sufficient to exactly reproduce the paper result. However, the result is qualitatively the same.

This repo uses the Resilience library.

For a self-contained introduction to the model in fewer than 1k lines of code, see https://github.com/ox-inet-resilience/firesale_stresstest. The firesale_stresstest repository implements all of its building block from scratch instead of using the Resilience library. It reproduces the result of Cont-Schaanning 2017.

Usage

Requires Python 3. It is recommended to run everything within a virtualenv.

git clone the repo and cd to it.
Install the dependencies with pip install .
Run the simulation with python sw_stresstest/simulation.py

You will see the results in the plots/ folder.

Remarks:

You can enable/disable parallelization (the parallelization is implemented using Python's multiprocessing library) by searching for fs.parallel in simulation.py and setting it to True/False.
Parallelization seems to not work on Windows. You probably need to disable it to run the simulation.
You can tweak the number of repetitions of the simulations by changing NSIM in the simulation.py file.
To speed up the simulation, you may decrease the value of NPOINTS. This represents the number of points in the x-axis in the simulation plots. Note: for now, changing NPOINTS will cause problem with running simulation FF11 because it is hardcoded to assume that the length is 11. This will be fixed in the future version.