Researching Financial Market Dynamics through Algorithmic Trading Agents

Document information

Author	phelstab
Status	wip
Classification of Information	open source
Location	GitHub

Change History

Version	Date	Author	Change
1.01	27.07.2024	phelstab	cleanup in process
1.0	24.11.2022	phelstab	init

Table of contents

• 1. Introduction
• 2. ABIDES
  • 2.1. Installation
  • 2.2. Architecture
  • 2.3. Known Bugs
• 3. Appendix
  • 3.1. References
  • 3.2. Short Paper and Poster
    • 3.2.1. Arbeitskreis Wirtschaftsinformatik 2024 (Landshut)

1. Introduction

Our research explores the use of Agent-based Interactive Discrete Event Simulations
(ABIDES) for the analysis of complex financial market scenarios and the identification of
potential anomalies. The goal is to experimentally simulate financial market issues instead
of observing them, for example, in controversial pilot programs on real reference markets.
Our work encompasses various market fees and a synthetic competition between two
exchanges, employing an inter-market spread arbitrage machine to enable price stability
between the exchanges. In other experiments various market scenarios were simulated,
and key metrics such as the impact on execution quality, generated turnaround, and market
share of exchanges were analyzed under different marketplace fee scenarios. This
approach allows the observation of anomalies related to spreads, execution speed, traded
volumes, liquidity, probability of order execution, and the impact on agents' decision-
making.

2. ABIDES

2.1. Installation

# STRONGLY RECOMMENDED:
# USE UNIX BASED SYSTEM (Linux, Mac)
# due problems with windows dt & pd binaries (see 2.4)
#
# Step 0: Install Python version 3.7.14 or higher
sudo apt install software-properties-common -y
sudo add-apt-repository ppa:deadsnakes/ppa #unofficial but trusted
apt-get install python3.7
sudo apt install python3-pip
alias python=python3.7
# check with python --version (should be 3.7.14 or higher)

# Step 1: Clone repository
git clone https://github.com/phelstab/bachelor.git

# Step 2: Switch to abides folder
cd bachelor
cd abides-jpmc-public

# Step 3: Select Python version and create Virtual Environment
# Python 3.7.14 >= recommended (check with python --version)
# Update pip and create venv with example name lambda
# (Used lambda as naming for the venv)
python -m venv lambda

# Step 4: Activate venv
# Activate venv on Linux
source lambda/bin/activate
# Activate venv on Windows
.\lambda\Scripts\activate
# Activate venv on mac
source lambda/bin/activate
# or through fishshell
. lambda/bin/activate.fish

# Step 5: Update pip and install libs
python -m pip install --upgrade pip
pip install -r requirements_main.txt

# Step 6: Additional installs (recommended)
pip install torch
pip install tensorboard
pip install ipykernel
# Set up kernel for jupyter notebook to use venv in jupyter notebook
python -m ipykernel install --user --name=lambda

# Step 7: Run and install in abides root folder
pip install -e .

# Step 8: Example test run
# 8.1 switch to 000_experiments folder
cd 000_experiments
# 8.2 run rmsc05_var.py (variable market fees single exchange experiment)
python rmsc05_var.py
# 8.3 open post analysis dashboard in your browser

Overview of different experiments in the 000_experiments folder:

python rmsc05.py #(base model no fees)

python rmsc05_fix.py #(fix market fees, single exchange)
python rmsc05_var.py #(varaible market fees, single exchange)
python rmsc05_mt.py #(maker taker market fees, single exchange)

2.2. Architecture

Figure 1. ABIDES with extended experiment structure overview

2.3. Known Bugs

Pandas timedelta 64bit is returning wrong values on windows.

• Python Version 3.7.14 on both systems

• Pandas Version 1.2.4 on both systems

• Numpy Version 1.20.3 on both systems

• Windows 11 vs WSL

3. Appendix

3.1. References

Title	DOI	Source
Explaining Agent-Based Financial Market Simulation	10.48550/ARXIV.1909.11650	https://arxiv.org/abs/1909.11650
Towards Realistic Market Simulations: a Generative Adversarial Networks Approach	10.48550/ARXIV.2110.13287	https://arxiv.org/abs/2110.13287
ABIDES: Towards High-Fidelity Market Simulation for AI Research	10.48550/ARXIV.1904.12066	https://arxiv.org/abs/1904.12066
Explaining Agent-Based Financial Market Simulation	10.1145/3490354.3494433	https://doi.org/10.1145\%2F3490354.3494433
How to Evaluate Trading Strategies: Single Agent Market Replay or Multiple Agent Interactive Simulation?	10.48550/ARXIV.1906.12010	https://arxiv.org/abs/1906.12010

3.2. Short Paper and Poster

3.2.1. Arbeitskreis Wirtschaftsinformatik 2024 (Landshut)

Poster

Paper