Combinatorial auctions on CoW Swap

Code for testing different implementations of combinatorial auctions.

Installing dependencies

Installing dependencies is easiest via uv. Just run the script or the jupyter notebook and dependencies will be installed automatically into the environment .venv.

Data fetching

If you want to test historical auctions, you need to provide database credentials in a .env file, see .env.example.

At the moment, the data fetching can take 1 minute the first time it is run. Data is then stored in a 80MB file and not fetched again (for the same range of auctions).

If you have git-lfs installed, a set of auctions should be automatically downloaded with the repo. Otherwise, you can download a set of auctions from the release page. In this way you can run the experiment without access to the database.

Running the script

Running the script and notebook is easiest via uv.

The script for analysing historical auctions is run via

uv run counter_factual_analysis.py

Some parameters of the simulation can be modified using command line arguments.

usage: counter_factual_analysis.py [-h] [--auction_start AUCTION_START] [--auction_end AUCTION_END]
                                   [--efficiency_loss EFFICIENCY_LOSS] [--approach APPROACH]
                                   [--reward_upper_cap REWARD_UPPER_CAP] [--reward_lower_cap REWARD_LOWER_CAP]

Run counterfactual analysis on auction solutions.

options:
  -h, --help            show this help message and exit
  --auction_start AUCTION_START
                        Start block for fetching auctions (default: 10322553 - 50000)
  --auction_end AUCTION_END
                        End block for fetching auctions (default: 10322553)
  --efficiency_loss EFFICIENCY_LOSS
                        Efficiency loss parameter (default: 0.01)
  --approach APPROACH   Approach type for solution splitting (default: complete)
  --reward_upper_cap REWARD_UPPER_CAP
                        Upper cap for rewards in wei (default: 12 * 10^15)
  --reward_lower_cap REWARD_LOWER_CAP
                        Lower cap for rewards in wei (default: 10^16)