Are AI stock predictions accurate? An honest answer
Short answer: our forecast bands are honestly calibrated as of the most recent recalibration, but calibration is not the same question as "does it predict direction correctly" — and we publish both numbers, including a past mistake we corrected in public.
The correction: what we got wrong, and fixed
An earlier headline claimed 97% CI90 accuracy on our TOP-20 backtest. That figure turned out to be a single curated snapshot that was not reproducible from the committed backtest artifact — we retracted it in #907. Re-running the model honestly on the same committed artifact showed real overconfidence: raw coverage of roughly 74.7% (3-month), 93.3% (6-month), and 80.8% (1-year) against a 90% target — two of three horizons were meaningfully below where a well-calibrated model should sit. Rather than quietly loosening the bands to hit the target number, we shipped a horizon-dependent band recalibration (#1088) and published the before/after difference.
Where the numbers stand now
- 3 months: 91.0% CI90 coverage (well-calibrated)
- 6 months: 93.3% CI90 coverage (well-calibrated)
- 1 year: 90.8% CI90 coverage (well-calibrated)
All figures fall inside our own pre-registered well-calibrated range of 85–95%. Read the full per-ticker breakdown, misses included, on the live calibration page.
Coverage is not a directional hit-rate
These are band-calibration figures: the percentage of realized prices that landed inside the published 90% range, not a measure of whether the model's directional call (up or down) was correct. Directional accuracy for any model in efficient large-cap markets is typically only modestly better than a coin flip — we never present a forecast mean as a standalone buy/sell signal, and per-ticker dispersion around the aggregate figures above is still wide: structural-break names like companies mid-acquisition or undergoing a business-model shift remain poorly covered, especially at the 1-year horizon.
Why we publish this instead of only good news
A forecasting product that only shows favorable numbers is not trustworthy, and trust is the entire basis on which anyone would pay for a forecast. The live calibration page and the live track record update automatically as the backtest window rolls forward and as trades close — both show the losses alongside the wins, not a cherry-picked highlight reel.
Frequently asked questions
Are Quantustik's stock predictions accurate?
Our 90% confidence bands are honestly calibrated as of the #1088 recalibration — coverage sits within our own well-calibrated 85-95% range at every horizon. That is a different, narrower claim than "the direction call is usually right" — see the next answer.
What is the difference between calibration and directional accuracy?
Calibration measures whether a stated 90% band actually contains 90% of realized prices. Directional accuracy measures whether the forecast correctly called up vs. down. A model can be well-calibrated while its directional hit-rate is only modestly better than a coin flip — the two are graded and published separately.
Did Quantustik ever publish an inaccurate accuracy claim?
Yes — an earlier "97% CI90 accuracy" headline was a non-reproducible curated snapshot, retracted in #907. Honest re-testing showed real overconfidence at two of three horizons, which a public band recalibration (#1088) corrected.
Where can I see the current live numbers?
The calibration page and the track record page both self-update from the committed backtest artifact and the closed-trade aggregate, respectively — including the misses.
Educational research only — not investment advice.