Can we tell at 09:30 whether today will be a trend day (RTH body ratio ≥ 0.6)?
Criterion not met — BSS -1.9 % [-3.6 ; -0.3], AUC 50.3 %, accuracy 61.6 % against a base rate of 36.2 %, n = 904 days pooled forward.
Criterion: BSS > 2 and AUC ≥ 57, pooled forward across 2023–2026
| Decision time | 09:30 New York — bars up to t−1 complete, the opening print and every tick of session t before 09:30 |
| Target | body ratio br = |rc − ro| / (rh − rl) ≥ 0.6 |
| Data window | 2020-05-29 to 2026-08-14 · 1607 daily bars · last day 2026-08-14 |
| Walk-forward | Dev ≤ 2022 (definitions only), test 2023 / 2024 / 2025 / 2026, expanding with a yearly refit |
| Model | L2 logistic regression on 15 preregistered set C features, standardised on the training window; C from inner validation (the last training year). Features: gap_atr, abs_gap_atr, on_range_atr, ln_on_relr, on_vol_rel, on_delta_norm, open_pos_on, on_ext_recency, pre_range_share, on_high_vs_rh1_atr, on_low_vs_rl1_atr, open_pos_va, prev_ln_relr_rth, prev_ln_relr_mean5, prev_atr5_atr20 |
| Holdout | The PX holdout from 2026-05-01 concerns the PX theses on the tick store. This study is a walk-forward over 2023 to 2026 on daily data — explicitly approved by the user for regime rounds 1 and 2. Every test year is computed from data before it only; no fit ever sees its own test slice. |
Every row is a forecast over the same days. Base rate, majority and persistence are the baselines; the model has to beat them, otherwise it does not count.
| Row | Year | n | Base Rate % | Acc % | AUC % | Brier | BSS % |
|---|---|---|---|---|---|---|---|
| Base Rate | 2023 | 246 | 39.8 | 60.2 | 50.0 | 0.2397 | 0.0 |
| Majority | 2023 | 246 | 39.8 | 60.2 | — | — | — |
| Persistence | 2023 | 246 | 39.8 | 60.2 | 54.7 | 0.2375 | 0.9 |
| Model (LogReg) | 2023 | 246 | 39.8 | 54.1 | 48.5 | 0.2516 | -5.0 |
| Base Rate | 2024 | 250 | 33.2 | 66.8 | 50.0 | 0.2262 | 0.0 |
| Majority | 2024 | 250 | 33.2 | 66.8 | — | — | — |
| Persistence | 2024 | 250 | 33.2 | 66.8 | 50.5 | 0.2283 | -0.9 |
| Model (LogReg) | 2024 | 250 | 33.2 | 65.2 | 54.6 | 0.2265 | -0.1 |
| Base Rate | 2025 | 251 | 37.8 | 62.2 | 50.0 | 0.2353 | 0.0 |
| Majority | 2025 | 251 | 37.8 | 62.2 | — | — | — |
| Persistence | 2025 | 251 | 37.8 | 62.2 | 55.3 | 0.2328 | 1.0 |
| Model (LogReg) | 2025 | 251 | 37.8 | 61.8 | 46.3 | 0.2390 | -1.6 |
| Base Rate | 2026 | 157 | 32.5 | 67.5 | 50.0 | 0.2226 | 0.0 |
| Majority | 2026 | 157 | 32.5 | 67.5 | — | — | — |
| Persistence | 2026 | 157 | 32.5 | 67.5 | 46.8 | 0.2264 | -1.7 |
| Model (LogReg) | 2026 | 157 | 32.5 | 67.5 | 53.6 | 0.2232 | -0.3 |
| Base Rate | pooled | 904 | 36.2 | 63.8 | 49.9 | 0.2317 | 0.0 |
| Majority | pooled | 904 | 36.2 | 63.8 | — | — | — |
| Persistence | pooled | 904 | 36.2 | 63.8 | 52.6 | 0.2317 | 0.0 |
| Model (LogReg) | pooled | 904 | 36.2 | 61.6 | 50.3 | 0.2362 | -1.9 |
Block bootstrap (20-day blocks, 1000 runs, pooled forward): BSS 95 % interval [-3.6 ; -0.3].
Confusion matrix pooled: true 0 correct 532 of 577, true 1 correct 25 of 327 (recall 0 = 92.2 %, recall 1 = 7.6 %).
Regularisation chosen per refit — 2023: C = 0.1, n_train = 580 · 2024: C = 0.01, n_train = 826 · 2025: C = 0.01, n_train = 1076 · 2026: C = 0.01, n_train = 1327.
| Decile | n | mean p % | observed % | difference |
|---|---|---|---|---|
| 1 | 91 | 30.7 | 41.8 | 11.1 |
| 2 | 91 | 35.2 | 29.7 | -5.5 |
| 3 | 91 | 36.8 | 30.8 | -6.1 |
| 4 | 91 | 37.8 | 39.6 | 1.7 |
| 5 | 90 | 38.8 | 40.0 | 1.2 |
| 6 | 90 | 40.0 | 33.3 | -6.6 |
| 7 | 90 | 41.3 | 37.8 | -3.5 |
| 8 | 90 | 43.0 | 31.1 | -11.9 |
| 9 | 90 | 45.7 | 36.7 | -9.0 |
| 10 | 90 | 53.4 | 41.1 | -12.3 |
Sharpness: p ≥ 0.7 on 0.0 % of the days (n = 0), hitting — % there · p ≤ 0.3 on 3.2 % (n = 29), hitting 58.6 % there.
| Base Rate | the unconditional frequency in the training window, and at the same time the constant comparison forecast. |
| Baseline | the number a model has to beat. Three of them here: base rate, majority (always the more frequent class) and persistence. |
| Persistence | the forecast “today like yesterday” — the value of the same target on the previous day, turned into a rate. |
| BSS | Brier skill score: what percentage of the base rate constant's Brier score the model saves. 0 means equally good, negative means worse. |
| AUC | the probability that a random positive day is scored above a random negative one. 50 is a coin flip. |
| Brier | the mean squared error of the probability. Smaller is better. |
| Sharpness | on what percentage of the days the model says something clear (p ≥ 0.7 or ≤ 0.3) and how often it hits there. A calibrated model without sharpness is useless. |
| Value Area | the price band [rval, rvah] in which 70 % of the RTH volume traded. Always the previous day's here. |
| Trend Day | an RTH with a body ratio ≥ 0.6 — the body of the daily candle fills at least 60 % of its range. |
| Body Ratio | br = |rc − ro| / (rh − rl), computed on the RTH session. |
| relr | RTH range divided by the median of the 20 previous RTH ranges. relr ≥ 1 means: today is bigger than the typical one of the last four weeks. |
| Walk-Forward | every test year is computed from a model that has only seen data before it; the training window grows with every year. No fit sees its own test. |
The numbers in this report are recomputed on every publish — from data/daily-bars-*.json and data/open-features-*.json, which the extractor builds from the tick store.
Run: 1607 daily bars, 0.6 s.