A forecast band says where price should close after a given number of candles. An 80% band makes a simple promise: four closes in five should land inside it. Before putting a number like that on a product page, we wanted to know whether our band kept the promise on data it had never seen.
The setup
The band was learned on Kraken BTC/USD candles from 2018 to 2023 and then frozen. It describes the historical spread of moves in units of average true range (ATR), adjusted for whether the market is quiet or loud and for the day of the week. Nothing was refitted for this test.
We scored it on 1h, 2h and 4h candles in two windows. The first runs from January 2024 to June 2025. The second, from July 2025 to June 2026, was sealed: it played no part in building the band or in any decision about it.
As built, the band was too narrow
The 80% band caught 76 to 81% of closes, and the 50% band caught 43 to 46%. The shortfall was three to five points almost everywhere. Split by market regime, the sealed year ranged from 68% to 92%, well outside the 78 to 82% we had set as the target for every regime. The adjustment for loud markets overcorrected and landed at 73 to 75%.
| Band | Target | As built | Recalibrated, sealed year |
|---|---|---|---|
| 50% band | 50% | 43–46% | within 1.5 points |
| 80% band | 80% | 76–81% | within 1.5 points |
One correction, tested once
We fitted a single width scale for each timeframe and horizon on the 2024 to 2025 window, then applied it unchanged to the sealed year. Coverage landed within 1.5 points of target at 50, 80, 85 and 90%. The 85% band caught between 84.5% and 86.5% of closes.
The sealed year has now been used. Any further tuning can only be checked going forward, on new quarters of data and in the live record.
Coverage is a dial
Any coverage figure can be reached by making the band wider, so a percentage says little without the width next to it. After recalibration, the 85% band ten candles ahead is roughly ±1.9% on the 1h chart and ±3.9% on the 4h chart. Those widths belong in the same sentence as the 85%.
The claim that did not survive
We also wanted to say that price reaches the inner half of the band about three times in four. Read strictly, it does not. The candle at the horizon trades into the 25 to 75% zone 66% of the time five candles ahead, 58% at ten and 55% at twenty. Staying inside for the whole path happens 50%, 41% and 32% of the time.
A looser statement does hold. Price sits inside the band on about 77% of the candles along the way, which would rise to around 85% with the recalibrated band. That is a different claim, and it should be worded as one.
What a band claim needs
A claim about a forecast band needs three parts: the coverage, the width at which that coverage was reached, and the period it was checked on. Without the width, the number can be made to say anything. Without a period the band never saw, it describes the fit and not the forecast.