Market microstructure with ABIDES
Question: How does an execution agent change prices and order-book conditions in a simulated market?
Byrd et al. (2020) introduce ABIDES, an agent-based discrete-event market simulator. The original repository and its RMSC03 experiment guide provide a concrete starting case. RMSC03 is an upstream experiment using the framework; this walkthrough does not identify it as a reproduction of a particular paper figure.
Prepare the historical environment#
The inspected snapshot is c4bf157678928934417aba6073eb0651aeaf6d15:
git clone https://github.com/abides-sim/abides.git
cd abides
git checkout c4bf157678928934417aba6073eb0651aeaf6d15
Inspect the snapshot's requirements before creating a separate environment. Its old NumPy, pandas, and SciPy versions should not be mixed into the lab's modern CPU environment. A working historical Python environment has not been established here. The distinct JPMorgan repository abides-jpmc-public was archived when checked; it is not interchangeable with this original codebase.
Follow the published code example#
Once the upstream dependencies are working, the RMSC03 guide's baseline command is:
python -u abides.py -c rmsc03 -t ABM -d 20200603 -s 1234 -l rmsc03_two_hour
ABM is the simulated instrument. The date labels the simulation; this command does not load that day's real market. Inspect the configuration's agents, latency assumptions, and session times. The guide documents exchange and order-book logs including log/rmsc03_two_hour/EXCHANGE_AGENT.bz2 and ORDERBOOK_ABM_FULL.bz2.
The guide also adds an execution agent using -e and a participation rate such as -p 0.1. Use a new log directory for each case. For a comparison, pair the baseline and execution cases on the same seed, then repeat the pair over several independent seeds. Measure a specified price-impact or execution-cost statistic and retain the raw logs and analysis code. A single stochastic path is not an estimated population effect.
For a smaller mathematical finance exercise that already runs in the lab, use Black–Scholes and QuantLib. That example prices a model payoff; ABIDES models market interactions.
Status: DOI metadata, original source, historical dependencies, and upstream experiment commands inspected on 2026-11-15. Article methods have not received a full-text audit; this historical environment and experiment were not run.