When Does the High or Low of the Day Form? 16 Years of ES Data
The high or low of day often forms earlier than a finished chart makes obvious. In MADDOG’s archived ES scorecard, at least one side of the session’s final range was already present in 58.3% of sessions after 35 minutes and 81.2% after 90 minutes. That is a timing base rate—not a way to know which side has formed.
Quick answer
Across 3,946 archived ES regular sessions from 2010–2026, the first five-minute bar already contained either the final session high or the final session low 26.3% of the time. The share reached 58.3% by 35 minutes and 81.2% by 90 minutes.
“Either” is the crucial word. The measurement does not identify in real time whether the printed extreme is the high or the low, and it does not say the other side cannot extend.
When one final session extreme was in place
“In place” means the final session high wick or low wick had first appeared by the checkpoint. Only one side is required. If the same price is touched again later, its first appearance still counts as early.
| Checkpoint | Brooks framework | Archived (n = 3,946) | Reconstruction (n = 3,942) |
|---|---|---|---|
| First 5-minute bar | ~20% | 26.3% | 26.3% (1,037 / 3,942) |
| First 35 minutes (7 bars) | ~50% | 58.3% | 58.3% (2,300 / 3,942) |
| First 90 minutes (18 bars) | ~90% | 81.2% | 81.2% (3,202 / 3,942) |
The reconstruction reproduces all three rounded percentages on 3,942 complete 81-bar sessions. Its four-session difference from the archived table is why the two sample counts remain separate. In clock time, 35 minutes ends at 10:05 ET and 90 minutes ends at 11:00 ET.
On bull and bear days, which extreme tended to form early?
A second calculation conditions on how the session ultimately closes. In the reconstruction, a “bull day” means the final close is above the session open and a “bear day” means it is below. The first third is the first 27 of 81 bars, about 135 minutes.
| Final day condition | Question | Archived rate | Reconstruction |
|---|---|---|---|
| Session closes above its open | Low forms in first third | 83.3% | 83.3% (1,805 / 2,166) |
| Session closes below its open | High forms in first third | 80.4% | 80.9% (1,413 / 1,747) |
The archived scorecard stored 3,946 as the source-dataset size for both conditional claims, not the number of eligible bull or bear days. Its conditional denominators were not retained. The reconstruction supplies its own eligible counts; 29 additional sessions closed exactly at their open and enter neither row.
The hindsight boundary
The bull/bear table is descriptive, not a live forecast. A session is placed in the bull row only after its final close is known, then the earlier low is inspected. Reading the table backward—“the low formed early, therefore this will be a bull day”—would be a different conditional probability and is not measured here.
The checkpoint table has a related limit. At 90 minutes it says one of the two final extremes was already present in 81.2% of sessions. It does not reveal which extreme, whether price will revisit it, or how far the unfinished side will travel. Those questions require additional features and a separately tested model.
Why MADDOG reads after 90 minutes
Ninety minutes is a useful information checkpoint because one side of the final range was already present in about four of five historical sessions. It is one input to the product timing, not a claim that 90 minutes is an optimal entry time. MADDOG waits for 18 five-minute bars so its day-type model can combine that opening structure with the session’s other measured features, then reports a probability distribution rather than declaring that a high or low is final.
The same distinction applies to the opening range breakout strategy research: a structural base rate is evidence for a product design choice, not proof of a profitable trade.
Methodology and provenance
The archived timing scorecard covers ES five-minute regular sessions from 2010–2026. The strict reconstruction covers 2010-06-07 through 2026-07-17 on a 09:30–16:15 ET grid. That is the historical research convention, not the current live endpoint’s 09:30–16:00 window. The original executable scorer was not retained. The reconstruction over 3,942 complete 81-bar days reproduced the three checkpoint percentages exactly at one-decimal precision and reproduced the bull-day conditional rate exactly; the bear-day rate was 80.9% versus the archived 80.4%.
Because the original conditional denominators are missing, this page never presents 3,946 as the number of bull days or bear days. The archived figures and reconstructed counts stay visibly separate instead of being merged into a cleaner-looking but false record.
Four 2020 sessions contain only 79–80 non-empty bars. Including days with at least 78 bars produces 3,946 sessions and still reproduces 26.3%, 58.3% and 81.2%, which may explain the archived sample size. Without the original scorer, that remains an explanation rather than a confirmed historical rule.
Read the full methodology and provenance record, or inspect the machine-readable reconstruction record. You can also compare where extremes print across the full tape in the ETH versus RTH study.