EuroStoxx Trading Strategy Analysis - Machine Learning

📌 Implemented Strategy

Objective: Predict buy/sell signals (1/0) for EuroStoxx based on historical returns. Binary classification.

Methodology:

1. Data Preparation:

   • X: EuroStoxx prices (2003-2023)
   • y: Binarized returns (1 if >0, 0 otherwise)
   • NaN handling and stationarity checks

2. Modeling:

   • Algorithm: Random Forest Classifier
   • Optimization: GridSearchCV for hyperparameters

   param_grid = { 
       'n_estimators': [50,100,200],
       'max_depth': [4,5,6],
       'criterion': ['gini', 'entropy']
   }

3. Validation:

Custom seasonal cross-validation (252 days)

Progressive data purging (5*j days)

4. Evaluation:

Test on last 22 "unseen" days

---

📊 Results

Overall Performance

Metric	Score
Average CV Accuracy	53.5%
Final 22-day Accuracy	63.6%

CV Fold Details:

Score classification fold 0 : 0.8
Score classification fold 10 : 0.56
...
Score classification fold 19 : 0.56

Final Predictions (22 days):

array([0, 0, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1])

🔍 Own Critical Analysis

✅ Strengths

1. Data Processing:

• Clean NaN handling
• Effective returns visualization

2. Methodology:

• Financial-data adapted cross-validation
• Progressive purging to avoid look-ahead bias

3. Optimization:

• Comprehensive GridSearch
• Best params: {'criterion': 'entropy', 'max_depth': 4, 'n_estimators': 100}

⚠️ Limitations Univariate Features:

• Only price used, no technical indicators
• Missing macroeconomic features

Stationarity Issues:

• Volatility clustering not modeled

Potential Overfitting:

• 53.5% accuracy close to random baseline
• Strong bullish bias in predictions (15/22 = 68% 1s)

💡 Improvement I wanted to add

Feature Engineering:

# Ideas of additions
df['MA_50'] = df['eurostoxx'].rolling(50).mean()
df['RSI'] = talib.RSI(df['eurostoxx'], timeperiod=14)

Alternative Models:

• LSTM for temporal dependencies
• XGBoost with early stopping

Additional Metrics:

from sklearn.metrics import classification_report
print(classification_report(y_true, y_pred))

Backtesting:

• Implement walk-forward analysis
• Calculate strategy Sharpe Ratio

🎯 Conclusion

The strategy shows promising initial results (63.6% on test set) but requires:

• More sophisticated feature engineering

• Robust validation framework

• Real P&L performance analysis

Next Steps : Integrate dynamic stops and conduct 6-month forward testing.

---

Full code :

https://htmlpreview.github.io/?https://raw.githubusercontent.com/aurelien7877/EurostoxxStrategy-ML/main/index.html