WebCryptoAgent: Agentic Crypto Trading with Web Informatics

This repository accompanies a research workflow focused on backtesting model-driven trading decisions.

Purpose and Scope

The codebase is intended for offline backtesting and analysis. It is not a production trading system and does not include proprietary feeds or credentials.

Data Policy

For all data scrapers, please provide and clearly indicate your own data sources/endpoints. Any datasets used in experiments should be cited or documented by the researcher.

Contact and Access

For a more detailed Node.js implementation and integration of news sensitivity, please contact ali86144@gmail.com for further discussion (privacy reasons).

Backtest Models (Reproducibility)

1) Install dependencies

pip install -r python/requirements.txt

2) Prepare data (OHLCV)

Place CSV files under python/data/ohlcv/ using this naming convention:

SYMBOL_INTERVAL.csv

Example:

BTCUSDT_15m.csv

Accepted columns include:

• iso or ts (timestamp)
• open, high, low, close, volume
• trades (optional)

3) Provide an LLM API key

Create a .env.backtest file in the repo root with at least one API key:

OPENAI_API_KEY=your_key_here
# or GEMINI_API_KEY / DEEPSEEK_API_KEY / QWEN_API_KEY

4) Run the backtest

From the repo root:

PYTHONPATH=python python -m app.backtest_models --symbol BTCUSDT --interval 15m --lookback-days 365

Multiple symbols:

PYTHONPATH=python python -m app.backtest_models --symbols BTCUSDT,ETHUSDT --interval 15m --lookback-days 365

Optional: fetch market data

If you want to auto-fetch OHLCV, set your own ticker endpoint in python/app/backtest_models.py (Ticker_API) and run with --fetch.

Outputs

Results are saved to results/backtests/<timestamp>/, including:

• summary_all.json
• model_comparison.csv
• model_comparison.md
• equity_compare.png (or per-symbol variants)

Sample Results (Figures)

The following figures are included as representative outputs from sample_result/backtest/.