Adaptive Curves for Optimally Efficient Market Making

Empirical Results

The EmpiricalResults folder contains our simulation codebase. This folder has the following important files:

main.py

This file initializes the Glosten-Milgrom model and takes the informedness (alpha) and volatility (sigma) as inputs. It outputs a csv file which stores average modetary losses compared over different policies (Algorithms 1 and 2 in the paper, and the loss oracle algorithm) and plots of how the ask and bid price vary with time, along with plots of monetary loss vs time, mid-price deviation vs time and the spread vs time. By default, the performance of the bayesian policy and the oracle policy is also included in these. To run the file, use the command python main.py --alpha <alpha> --sigma <sigma>.

You may want to change the following variables in this file for running any experiments.

• --max_episode_len : Number of discrete time steps per sample path. Default value is 20000.
• --max_episodes : Number of distinct sample paths to be averaged over. Default value is 120.

Other variables can be changed directly in the file itself. Their function has been indicated in the comments.

run.sh

This file runs multiple simulations of main.py.

plot.py

This file plots the monetary loss comparison averaged over multiple simulations and different values of eta and sigma. We recommend this file be run after run.sh completes.

env.py

This file contains all the code to initialize and run the Glosten-Milgrom model environment. Also implements the Kalman Filter-based market maker.

agent.py

This file contains all the code for initializing and training Adaptive Kalman Filter, Robust Kalman Filter agents on the environment.

requirements.txt

The python packages used have been listed in therequirements.txt. They can be installed using the following command.

pip install -r requirements.txt

Alternatively, they can be installed using conda in a new environment called adaptswap.

conda create --name adaptswap python=3.8
conda activate adaptswap
conda install gym numpy matplotlib torch scipy pandas tqdm

Implementation

The OnChainImplementation folder contains the code for the Axiom circuit (for the Kalman filter-based algorithm) and the Uniswap v4 hook that takes off-chain data.