This fork intruduces simple implemention of meta-labeling module build on top of Backtesting.py library.
The module integrates with Backtesting.py, allowing users to easily add meta-labeling to their existing strategies. It supports various machine learning models and provides utilities for feature engineering and signal filtering.
Key features:
-
Easy integration with Backtesting.py strategies
-
Support for multiple classifier models
-
Ensemble model builder
# clone this repo
cd backtesting.py-metalabel
uv sync
uv run python showcase.py#ASSET = GOOG
class SmaCross(Strategy):
def init(self):
price = self.data.Close
self.ma1 = self.I(SMA, price, 5)
self.ma2 = self.I(SMA, price, 9)
def next(self):
if crossover(self.ma1, self.ma2):
self.buy()
elif crossover(self.ma2, self.ma1):
self.sell()
split_idx = int(len(ASSET) * 0.8)
train_data = ASSET.iloc[:split_idx].copy()
test_data = ASSET.iloc[split_idx:].copy()
rf_parms = {
'n_estimators': 10000,
'max_depth': 15,
'random_state': 223145
}
xgb_params = {
"n_estimators": 700,
"learning_rate": 0.05,
}
ensemble = (
EnsembleBuilder()
.add_model("rf", RandomForrestModel(**rf_parms), weight=0.7)
.add_model("xg", XGBClassifier(**xgb_params), weight=0.3)
.soft_voting()
.build()
)
model = ensemble
curr_strat = SmaCross
ml = MetaLabeler(strategy=curr_strat, model=model, window_size=100,
data=train_data, commission=.002, exclusive_orders=True,
finalize_trades=True, cash=1_000_000)
gate = ml.create_filter(curr_strat, window_size=100, threshold=0.7, runtime_data=test_data)
enhanced_strategy = make_enhanced_strategy(curr_strat, gate)
bt = Backtest(test_data, enhanced_strategy, cash=1_000_000,
commission=.002, exclusive_orders=True, finalize_trades=True)Full script can be found in showcase.py.
Here is a markdown table comparing the results of the Base Strategy and Enhanced Strategy:
| Metric | Base Strategy | Enhanced Strategy |
|---|---|---|
| Start | 2011-06-15 00:00:00 | 2011-06-15 00:00:00 |
| End | 2013-03-01 00:00:00 | 2013-03-01 00:00:00 |
| Duration | 625 days 00:00:00 | 625 days 00:00:00 |
| Exposure Time [%] | 97.2093 | 97.2093 |
| Equity Final [$] | 1,167,385.86 | 1,631,710.29 |
| Equity Peak [$] | 1,380,412.05 | 1,653,523.12 |
| Commissions [$] | 217,971.24 | 141,266.02 |
| Return [%] | 16.74 | 63.17 |
| Buy & Hold Return [%] | 66.98 | 66.98 |
| Return (Ann.) [%] | 9.49 | 33.24 |
| Volatility (Ann.) [%] | 28.80 | 35.60 |
| CAGR [%] | 6.44 | 21.83 |
| Sharpe Ratio | 0.33 | 0.93 |
| Sortino Ratio | 0.59 | 1.96 |
| Calmar Ratio | 0.34 | 1.81 |
| Alpha [%] | 15.82 | 21.85 |
| Beta | 0.01 | 0.62 |
| Max. Drawdown [%] | -27.54 | -18.41 |
| Avg. Drawdown [%] | -8.51 | -4.84 |
| Max. Drawdown Duration | 560 days 00:00:00 | 246 days 00:00:00 |
| Avg. Drawdown Duration | 119 days 00:00:00 | 41 days 00:00:00 |
| # Trades | 46 | 27 |
| Win Rate [%] | 34.78 | 55.56 |
| Best Trade [%] | 23.23 | 17.47 |
| Worst Trade [%] | -6.81 | -6.35 |
| Avg. Trade [%] | 0.34 | 1.83 |
| Max. Trade Duration | 83 days 00:00:00 | 108 days 00:00:00 |
| Avg. Trade Duration | 14 days 00:00:00 | 23 days 00:00:00 |
| Profit Factor | 1.32 | 3.16 |
| Expectancy [%] | 0.48 | 1.97 |
| SQN | 0.40 | 1.90 |
| Kelly Criterion | 0.06 | 0.37 |
| _strategy | SmaCross | Enhanced_SmaCross |