Advanced Concepts

So far, we've discussed how a model can output a signal for each trading day: buy, sell, or hold. This approach is simple and practical—but it doesn't tell the whole story.

Many models, particularly those trained for classification, can also return probabilities for each class. Instead of simply saying "buy," a model might predict:

• buy: 72% probability
• hold: 20%
• sell: 8%

These probabilities offer deeper insight into the model's confidence, which can be used to build more advanced and nuanced strategies.

---

Why Use Probabilities?

Simple signals treat all predictions as equal. But not all predictions are made with the same level of certainty. A model that outputs buy with 51% confidence is not the same as one with 90% confidence.

Using probabilities allows you to:

• Avoid low-confidence trades by setting a threshold (e.g. only act if confidence > 65%)
• Size positions dynamically based on confidence levels
• Combine multiple models using ensemble logic, where each model's vote is weighted by its confidence
• Detect indecision: If probabilities are balanced, the model is likely uncertain—a good time to avoid trading

---

Example Strategy Using Probabilities

Here's a conceptual example of how probabilities can guide trading decisions:

• If buy probability > 70% → enter long
• If sell probability > 70% → enter short
• If no class exceeds 70% → hold
• If currently in a trade and the opposing signal exceeds 70% → exit and reverse

This strategy avoids acting on weak or ambiguous predictions, reducing unnecessary trades.

---

Where This Fits In the Pipeline

To recap:

1. Market data is collected
2. Targets (buy/sell/hold) are created
3. Indicators are added
4. Sequencing and splitting are done
5. Model is trained
6. Evaluation is performed (ML metrics and backtesting)
7. Predictions are generated
8. Strategies are applied

What we describe here happens after Step 7, when the model produces raw predictions. This page expands your toolbox for building intelligent strategies on top of those predictions.

---

Summary

Using prediction probabilities can:

• Improve trade quality
• Allow adaptive risk control
• Help reduce noise from uncertain signals

We encourage experimenting with these ideas to develop your own trading logic. They can significantly enhance both the robustness and performance of your final strategy.