Model Performance
Backtested evaluation of the prediction models. Historical results — not a guarantee of future performance.
Demo data — last evaluated Oct 14, 01:57 PM.
Demo dataBacktested
Classification & regression metrics
Per-model accuracy, precision, recall, F1, ROC-AUC, log loss, Brier, MAE & RMSE
| Model | Accuracy | Precision | Recall | F1 | ROC-AUC | Log Loss | Brier | MAE | RMSE |
|---|---|---|---|---|---|---|---|---|---|
| Logistic Regression | 52.9% | 52.4% | 49.6% | 0.515 | 0.593 | 0.593 | 0.234 | 1.9 | 2.8 |
| Random Forest | 56.9% | 58.1% | 54.3% | 0.56 | 0.63 | 0.574 | 0.217 | 1.99 | 2.73 |
| XGBoost | 57.9% | 56.2% | 54.4% | 0.572 | 0.622 | 0.554 | 0.201 | 1.85 | 2.43 |
| Ensemble | 59.9% | 61.1% | 57.4% | 0.588 | 0.639 | 0.548 | 0.206 | 1.73 | 2.54 |
Walk-forward validation
Each fold trained on the past, tested on unseen future
2024 Q1
Acc 56.2% · ROC-AUC 0.597Train: rolling 24m → 2024 Q1 · Test: 2024 Q1
2024 Q2
Acc 58.0% · ROC-AUC 0.614Train: rolling 24m → 2024 Q2 · Test: 2024 Q2
2024 Q3
Acc 58.2% · ROC-AUC 0.575Train: rolling 24m → 2024 Q3 · Test: 2024 Q3
2024 Q4
Acc 57.4% · ROC-AUC 0.613Train: rolling 24m → 2024 Q4 · Test: 2024 Q4
2025 Q1
Acc 58.0% · ROC-AUC 0.623Train: rolling 24m → 2025 Q1 · Test: 2025 Q1
Baseline comparison
Multi-source model vs. single-source & naive baselines
| Configuration | Accuracy | ROC-AUC | Note |
|---|---|---|---|
| Naive Baseline | 50.0% | 0.5 | Majority-class / random walk reference |
| Technical-only | 53.7% | 0.57 | — |
| Fundamental-only | 52.3% | 0.541 | — |
| News-only | 53.2% | 0.539 | — |
| Multi-source | 57.2% | 0.607 | Combined feature set |