What is model confidence in a stock signal?
Model confidence measures how far a signal's underlying probability sits from a coin-flip (50/50), rescaled to a 0–100% reading. A probability of 80% (or 20%) yields 60% confidence; a probability of 50% yields 0% confidence. It is a statement about how strong the model's view is, independent of whether the direction of that view is correct — a confident model can still be confidently wrong, and confidence is not the same as historical accuracy.
Confidence vs. track record
Confidence is computed at signal time, from the current probability estimate; it says nothing about how that ticker's signals have performed historically. Quantustik separates the two deliberately: pair a given signal's stated confidence with that ticker's own backtest win rate/CI90 accuracy (a genuinely separate number) before sizing a position — a high-confidence signal on a ticker with a poor historical track record is a different bet than the same confidence on a well-calibrated name.
Live example: AAPL's current 3-month swing signal carries a stated model confidence of 1%. See the full AAPL forecast for the signal and trade plan this confidence attaches to.
Why low-confidence signals get smaller position sizes
Recommended position size scales with confidence: a low-conviction signal (confidence near 0%) gets little to no allocation even if the raw direction call is technically BUY, because the model itself is signaling it isn't sure. This is a deliberate risk-first design choice — a weak signal traded at full size is a common way to turn a marginal edge into a losing one.
Frequently asked questions
Does high model confidence mean the signal will be correct?
No. Confidence measures how strong the model's current view is, not whether that view has historically been right — pair it with the ticker's own backtest track record before sizing a position.
How is confidence calculated?
As the distance of the underlying probability from 50% (a coin-flip), scaled to 0-100% — an 80% or 20% probability both yield 60% confidence.
Why does low confidence reduce position size?
Recommended position size scales with confidence by design, so a weak, low-conviction signal never gets sized as if it were a strong one — a deliberate risk-first guardrail.
Educational research only — not investment advice.