AI4T 2020/21 project about Shock Propagation

Who are we?

About this project

This project is indeed an analysis about the phenomenon of shock propagation from a symbol to another under different and controlled circumstances.

This repo allow you to simulate a trading day with an exchange tracking an ETF and many underlying symbols with different shares (weigths) in it. This environment is of course populated with different agents behaving in different ways according to their strategy: for example, ValueAgents try to follow the fundamental given them by an Oracle, MomentumAgents try to follow the trends, while NoiseAgents simply do controlled random orders in order to add a realistic scent to the simulation.

At the end of each simulation you'll be prompted with some plots of the mid price, spread and transaction volume of each symbol involved alongside the strength of the causalities of each symbol to the other to see if they have a positive, negative or dark correlation.

Beware that PC algorithm is just a sketch and is not to be considered a final release, since it only accounts for positive causalities.

Check the results

To check our result you'll have to install the requirements and run a simulation, at the end of which you'll be prompted with plots and causality stats for each symbol, also saved into data\<simulation_name>.

Install the dependencies

You must have installed:

• Python 3.6 or 3.7 installed (3.8+ gave us some problem concerning matplotlib's package')
• fim image viewer if you want to have the plots prompted at the end of each simulation, else you'll have to check them manually inside data\<simulation_name>\plots

Assuming you've successfully installed a supported version of Python, from inside the root folder (the one containing README.md) type:

sudo apt install fim
pip install -r requirements.txt

This will install all the dependencies needed for the program to work correctly.

Please note that, currently, this project has been made and tested to run correctly in Linux environments (Linux Mint 20 and Ubuntu 20.04) since ABIDES do seem not to work on Windows.

Run a simulation

You can either run a simulation or multiple ones in sequence by using scripts/run_simulation.py with different parameters.

Run a single simulation

From inside the root folder type:

cd scripts
python run_simulation.py
cd ..

This will start a new simulation (with a random seed) of a trading day with a secondary market duration of two hours, a huge shock in the middle, an ETF and three underlying symbols with different shares.

You can view the list of parameters by giving the -h parameter to the script, e.g. python run_simulation.py -h. Some interesting parameters:

• --seed: specify a seed in order to be able to reproduce a particular simulation (default is a random number)
• --hours: number of hours of market, from 1 to 8 (default is 2)
• --scale: multiplier of the number of agents, useful to reduce impact on the RAM (default is 0.3 the full number of agents)
• --num_impacts: number of shocks on the non-ETF symbol with biggest share in it, evenly distributed in time (default is 1)