Every AI-generated trading signal comes with a confidence score. But what does this number really tell you, and how should you incorporate it into your trading decisions?

What Is a Confidence Score?

A confidence score is a numerical value — typically between 0 and 100 percent — indicating how certain the AI model is about a particular prediction. It is not a guarantee of trade success, but a measure of the model statistical certainty.

Think of it like flipping a coin: the probability of heads is 50%. An AI signal with 85% confidence means the model was correct in 85 out of 100 cases in similar historical situations.

How Are Confidence Scores Calculated?

The calculation depends on the model used:

Ensemble Methods: Multiple models vote. The higher the agreement, the higher the score. If 9 out of 10 sub-models are bullish, that yields a score of 90%.
Probabilistic Models: Bayesian networks or Gaussian Processes directly output probability distributions. The confidence score corresponds to the probability of the predicted direction.
Historical Backtesting: The signal is tested against historical data. The score reflects the hit rate in comparable market situations.

Practical Application: The Score Ladder

Experienced traders use confidence scores as filters for position sizing:

90-100%: Maximum position size (e.g., 5% of portfolio). Such signals are rare but historically highly profitable.
75-89%: Standard position size (2-3%). The sweet spot for most trades.
60-74%: Reduced position (1%). Additional confirmation through own analysis recommended.
Below 60%: Ignore signal or use only as observation.

Common Interpretation Mistakes

The biggest mistake is equating high confidence with high returns. A signal with 95% confidence for an upward move says nothing about how large that move will be. It could be 0.1% or 10%.

Another mistake: blind trust. Even high-confidence signals can be wrong — especially in extraordinary market situations. Use the score as one of several decision factors, not the only one.

Tip: Keep a trading journal and track the actual hit rate of different score ranges. This way you learn how reliable the scores are in your specific trading setup.