The point estimate is half the forecast
"Ethereum will be at $3,850 in 30 days" — a sentence like that is the most quoted and least useful output a forecast can produce. The information is not in the point; it's in the band around the point and the probability attached to it. This guide explains how to read those two numbers like a professional, and how to use them for position sizing rather than vibes.
What "85% confidence band" actually means
When a forecast says "expected range $3,300–$4,100 at 85% confidence," it means: across many similar forecasts, the realized price at horizon landed inside the stated band about 85% of the time. Three consequences follow directly, and all three are commonly misunderstood:
- It is not a price target. The middle of the band is the expected path; the band is where the price is allowed to wander while the forecast remains "correct." A forecast can be right 85% of the time and still be wrong on the magnitude — a coin that lands just inside the top of the band validated the forecast while feeling like a rocket.
- The band width is a feature, not a hedge. A wide band in a high-volatility regime is honesty. A narrow band in the same regime is fiction. If you compare two services, the one with wider bands in choppy conditions is usually the one with more calibration discipline, not less skill.
- 15% of the time, it breaks. That's the deal. A system whose bands never break is overfit or lying; a system whose bands break at exactly the stated rate is calibrated — and that's the only kind worth using.
Calibration: the confidence score you can audit
The confidence percentage on a forecast (say, 72% up) is only meaningful if the system's historical statements have matched reality. That property is called calibration, and it is measurable: bucket all past forecasts by stated confidence, then check the realized hit rate per bucket. A calibrated "70%" bucket wins about 70% of the time; an overconfident one wins 55%.
Calibration is why confidence must vary. Real edge concentrates: sometimes 30 models and the fundamental layer all agree, and 80%+ confidence is earned; sometimes they split evenly, and the honest output is barely-better-than-a-coin at 55%. A service whose confidence badge never moves is selling theater. On FARAZ forecasts the confidence number is explicitly a fusion output — model agreement raises it, channel conflict suppresses it — and it's graded after the fact on the My Predictions page after you sign in.
Using bands for position sizing
Here is where forecasting becomes risk management. A band plus a confidence score is enough to run a volatility-aware sizing rule:
- Distance to invalidation — the band's far edge in the wrong direction is a natural stop reference. If the forecast says 85% inside $3,300–$4,100 and you're long from $3,700, the band tells you the market can travel to $3,300 while the forecast remains valid. Placing stops tighter than the band means getting shaken out by noise the model explicitly priced in.
- Volatility-scaled sizing — position size should shrink as the band widens: the same 5% account risk behind a forecast with a 12% band and one with a 40% band are entirely different trades. A simple heuristic: risk a fixed fraction of account equity, divided by the distance to invalidation.
- Confidence-gated exposure — nothing forces you to trade every forecast. Taking only 65%-plus-confidence setups with agreeable bands trades frequency for quality, at the cost of patience.
None of this guarantees profit — bands are probabilities, and the 15% still happens — but it converts a forecast from a "prediction" into a structured decision input, which is the only durable way to use one. For the full framework, pair this with the risk-management guide.
Three habits of people who read forecasts well
- They read the band before the arrow. Direction is the headline; the range is the news.
- They track the forecaster, not the forecast. One outcome proves nothing; fifty logged outcomes say everything. Insist on an auditable record before adjusting your trust.
- They distrust constant confidence. Markets alternate between regimes where prediction is genuinely possible and regimes where it is barely better than chance. Confidence that never breathes is ignoring that alternation.
For how the FARAZ engine computes and grades its bands and confidence, see how AI crypto predictions work — and run a live example with the reasoning breakdown in the dashboard.