Do the 90% bands hold?
A well-calibrated 90% confidence (CI90) band should contain the realized price about 90% of the time. Whether ours does depends heavily on when you start, so we measure it separately at each quarterly start date and publish every one. Universe: the TOP-20 S&P 500 tickers by market cap.
| Start date | 3 months | 6 months | 1 year |
|---|---|---|---|
| 31 Mar 2024 | 84% | 92% | 90% |
| 30 Jun 2024 | 87% | 82% | 96% |
| 30 Sep 2024 | 78% | 88% | 90% |
| 31 Dec 2024 | 87% | 75% | 86% |
| 31 Mar 2025 | 80% | 84% | 94% |
| 30 Jun 2025 | 83% | 88% | 90% |
| 30 Sep 2025 | 94% | 71% | 90% |
| 31 Dec 2025 | 95% | 86% | 96% |
| 31 Mar 2026 | 91% | 91% | 90% |
We publish no single averaged coverage figure. The quarterly windows overlap at 6 months and 1 year, so averaging the rows would state an accuracy this measurement cannot support, and a confidence interval printed on top of one would be read as a range the price might land in, which it is not. The band held between 78% and 95% at 3 months, between 71% and 92% at 6 months and between 86% and 96% at 1 year. We hold ourselves to a pre-registered [85%, 95%] tolerance around the 90% target — our own tolerance, not an external standard — and several start dates sit outside it in both directions. The worst individual windows: UNH started 30 Jun 2025 (10%), NVDA started 30 Jun 2024 (6%) and NVDA started 31 Dec 2024 (2%). These are coverage figures, not a directional hit-rate.
Educational research only — not investment advice.