Frozen research artifacts for "Dynamic Multi-Pair Trading Strategy in Cryptocurrency Markets with Deep Reinforcement Learning"

Supplementary data for the paper "Dynamic Multi-Pair Trading Strategy in Cryptocurrency Markets with Deep Reinforcement Learning" (Lebiedź & Ślepaczuk, 2026). This deposit contains frozen experiment outputs and intermediate research artifacts used in the study. Source code, configuration, and reproduction instructions are available at:https://github.com/damianlebiedz/pair-trading-with-rl Contents data.zip data/pair_selection/ - monthly ranked cointegrated pair lists used in walk-forward backtests. data/rl_training/ - per-pair backtest trajectories exported for RL training. data/rl_models/ - trained Stable-Baselines3 checkpoints (Recurrent PPO) and VecNormalize statistics. results.zip Parquet outputs for every experiment run cited in the paper (per-pair returns, trades, performance statistics). .hydra/config.yaml and .hydra/overrides.yaml per run - resolved experiment parameters for reproducibility. What is NOT included: data/historical/ - OHLCV from Binance Data Vision. Market data is not redistributed here due to licensing and redistribution concerns. To re-run backtests, download historical prices locally using helpers/data_fetching_pipeline.py from the GitHub repository. execution.log and .hydra/hydra.yaml - omitted from results.zip because they contain local filesystem paths. For detailed instructions on how to automatically download these artifacts, fetch the market data, and reproduce the experiments, please refer to the README file in the project's GitHub repository.