簡単な回答

Options Band keeps a public record of its own expected-move calculations. Every trading day the site snapshots what the options market implies for each stock — before the outcome is known — and scores each snapshot after its expiration passes. NVDA's history page is one example.

The record is append-only: entries are never edited, deleted, or backfilled. The archive begins in August 2026, so early pages show short histories that simply lengthen with time.

Each page leads with two statistics: the percentage of expiration closes that landed inside the implied range, and the mean absolute error between the implied move and what actually happened.

詳しく見る

A worked entry makes the scoring concrete. Say a snapshot records NVDA at $600 with an expected move of ±$30, an implied range of $570 to $630. At expiration the stock closes at $612. Two facts get written down: the close landed inside the range, and the realized move of $12 differed from the implied $30 by $18. Averaging that second number across every scored snapshot produces the mean absolute error.

The percentage-inside figure needs a benchmark to be readable. A one-straddle range is not built to contain the close every time — under the model's own assumptions it should succeed roughly 60% of the time. A page showing 60% means the market's implied ranges have been about right for that stock; readings near 55% or 70% sit within the ordinary wobble of limited samples, and neither is a failure. The number exists to reveal persistent, large drift across many snapshots — say, a stock whose closes land inside only 40% of the time over two years. The derivation of the 60% benchmark is in the probability methodology.

The database rules do the quiet work. Snapshots are recorded before outcomes are known and are never revised afterward, so the statistics cannot be flattered after the fact; how the archive is stored and scored is documented in the historical database methodology. Because records begin in August 2026, a stock's early history may cover only a handful of expirations — small samples deserve matching caution.

詳細説明

These pages report calibration, which is why the site avoids the phrase "win rate." A win rate implies a bet that someone placed and either won or lost. An expected move is not a bet; it is a stated range with a stated coverage frequency, and the honest question is whether closes land inside it about as often as the model claims. Landing inside the range is not a profit, and landing outside is not a loss — hence percentage inside range, not wins.

Formally, each snapshot stores the spot price, the implied move, and the range endpoints at recording time. Scoring compares the expiration close against those endpoints and logs the absolute difference between implied and realized move sizes; mean absolute error is the average of those differences, in dollars, over all scored snapshots.

Limitations worth knowing: scoring uses the expiration close only, so a stock that left the range mid-week and returned still counts as inside; mean absolute error pools calm and volatile regimes into one figure; and the inputs come from delayed quotes. The append-only rule prevents survivorship editing, but it also means early samples are small — a dozen scored weeks says little in either direction.