Sixteen years of E-mini Nasdaq-100 futures — 5.43 million one-minute bars, 3,992 complete trading sessions — mined to answer the oldest question on the tape. Forty statistical tests declared in advance, confidence intervals on every number, false-discovery control throughout, and three rounds of hostile review whose corrections are printed beside the results. These are the published findings.
This brief didn’t come from a research team — it came from Kings Palace running unattended: declare the hypotheses, mine the tape, catch its own errors, revise. Four passes, forty tests, ~3 hours, zero manual correction. What follows is the output.
Traders read the opening one-minute bar for shape, for size, and for direction. Sixteen years of data grade each reading separately — and the grades could not be more different.
| Finding | Estimate | Uncertainty | Plain-language verdict |
|---|---|---|---|
| Candle shape (close location) | mean 0.507 of range | CI 0.496–0.518 | statistically dead-center; wick-and-close lore finds no signal |
| Candle extremes as levels | median break time 1 min | — | the open’s own high/low are consumed within minutes, not respected |
| First-minute width → day width | elasticity 0.30 net of regime | partial r ≈ 0.29 | survives the volatility-persistence control — shrunken, real, scale-free spread ×1.41 between extremes |
| Direction persistence (green vs red) | +4.21 pp | q = 0.0094 | positive in all 11 regime cuts; two orders of magnitude below costs |
| Up-gap continuation (≥50bp) | 56.9% | CI 52.7–61.1 | faintly positive vs the post-open mark; supports neither fade nor follow |
| Opening volume → range (implementable form) | no gradient | p = 0.19 | a celebrated inverse relation shown to be a future-leaking artifact |
| Location → barrier races / completion | all null | q ≥ 0.69 | where the candle closed inside itself predicts nothing about completions |
An earlier study in this series reported that quiet openings precede wide days — a monotone inverse gradient across sixteen years, robust to every cut. Its own review suspected the finding. This study ran the suspect test, and the result is the cleanest methodological lesson we know:
The gradient survives only in a statistic that leaks the future — opening volume as a share of the day’s total volume, which cannot be known at 9:31. Measured in the form a trader could actually implement — opening volume against a trailing window — the gradient is flat (p = 0.19). Meanwhile the candle’s width, which needs no normalization at all, does carry forward range information — though most of that, too, turns out to be yesterday’s volatility speaking through today’s open (R² = 0.59 from the trailing prior alone; the candle adds +0.035).
“Quiet opens precede wide days” was true in the data — and unactionable in the market. The difference between those two sentences is the entire content of this research program.
Research earns trust by what it does when it is wrong. This study’s full build log is published in its appendix; the arc:
Every headline statistic carries a cluster-bootstrap confidence interval; every p-value is published alongside its false-discovery q-value; negative results are reported as findings; and each pass lists exactly what changed and why — including the numbers that got smaller.
The opening minute tells you about amplitude — and mostly tells you what yesterday already knew. It murmurs four percentage points about direction, positive in every era we can construct and worth approximately nothing after costs. About location — the geometry chart readers trade on — it says nothing that survives honest multiplicity control. The practitioner instinct that the open matters is correct; the instinct about which dimension matters is wrong three times out of three.
Standing disclosure. All results are descriptive statistics on historical futures data. No configuration described herein has been validated as a trading strategy; nothing here is investment advice; past structure does not guarantee future structure. Known limitations, recorded as open work in the underlying papers: stronger volatility benchmarks, era-split controls, gap-direction deconfounding, cross-instrument replication, event-day calendar controls, and computed cost arithmetic.