Why this page exists
A forecast nobody checks is just an opinion. Every forecast we make is saved before the trading day starts, scored automatically when the day closes, and kept here for good, including the ones that were wrong. Nothing can be edited or deleted afterwards.
Live record
The first live forecast is logged at the start of the next trading day. Until forecasts have been scored, the only evidence is the back-test below. This page updates itself every day.
The back-test (simulated on the past)
To check the method before trusting it, we ran it over 3,595 past trading days. For each day it used only what was known that morning, so it could not peek at the future. Nothing was tuned on the days being scored.
| What we checked | Next day | Next two days | What good looks like |
|---|---|---|---|
| Direction skill vs the long-run average | 0.16% | 0.27% | Above 0% means better than a plain average |
| 50% range held | 49% | 49% | Close to 50% |
| 80% range held | 78% | 78% | Close to 80% |
Back-test calibration: chance of touching a level
What the model said vs how often it happened
How the numbers are scored
- Brier score measures probability forecasts: it rewards being right and being honest about uncertainty. Skill compares it with always quoting the long-run average.
- Range held is how often the real close fell inside the stated range. A 50% range that holds 50% of the time is honest.
- Calibration groups forecasts by the chance we stated and checks how often those things happened.
Frequently asked questions
Are the forecasts edited after the fact?
No. Each forecast is saved with a time stamp before the day starts, and the log cannot be overwritten by our software.
Why do the direction odds look so close to the average?
Because in our back-test daily direction was almost unpredictable. The tested strength is the likely range and the chance of touching a level.
Sources and method
- Back-test: 3,595 days, nearest-neighbour method with 250 similar days, daily bars from 2010, scored with the Brier score against the long-run average.