← Back

Strategy Harvest

Built

Signature capability

End-to-end autonomy: in one unattended run it cloned three strategy repos, converted 12 strategies off Freqtrade's dependency graph into a self-contained pandas/TA-Lib engine, backtested and ranked them, and generated production strategy files plus a 16-strategy x 6-pair deployment script.

What it is

Strategy Harvest scrapes proven open-source crypto strategies (Freqtrade official, NostalgiaForInfinity, nateemma's PCA/DWT/Kalman set) and strips them out of Freqtrade's heavy runtime into a minimal `simple_backtest()` engine built on pandas and TA-Lib. Each converted strategy exposes `calculate_indicators()` and `generate_signals()` returning {1 long, 0 flat, -1 short}, plus a `STRATEGY_CONFIG` block of risk params. It backtested 12 strategies on 30 days of ETHUSD 1h data, computed Sharpe / return / drawdown / win-rate, ranked them, wrote the top 3 as standalone production files, and generated a `deploy_strategy_army.py` that fans 16 strategies across 6 pairs with per-tier capital multipliers. Notably the output docs self-flag the statistical thinness rather than hide it.

Highlights

  • Repo harvest: cloned 3 open-source repos totalling ~40 strategies (Freqtrade 24, NostalgiaForInfinity 5, nateemma 10+) and normalised them to one signal interface
  • Dependency-strip: converts Freqtrade strategies into a self-contained pandas/TA-Lib backtester (no Freqtrade runtime), ~1,474 lines of tooling across converter, batch tester and deploy script
  • Deployment layer: deploy_strategy_army.py defines 16 strategies in 3 tiers with capital multipliers (1.25x/1.0x/0.8x) fanned across 6 pairs, backed by a SQLite results store
  • Intellectual honesty: the harvest report explicitly calls out the red flags itself, only 30 days of data, single pair, 1-13 trades per strategy, long-only, and warns live Sharpe will be well below backtest
  • Grounded expectations: projects a realistic $500 to $550-650 over 90 days rather than the headline backtest numbers

Tech · Python 3.11, pandas, NumPy, TA-Lib, ccxt (data pull), SQLite, Freqtrade strategy sources