NVIDIA's IPO chart vs. 27 years of price

The chart has a place but no time — so it has no angles

NVDA first traded on Friday 22 January 1999 on the Nasdaq National Market. The date is certain. The time is not recorded anywhere, and the assumption most people would reach for is wrong: IPOs do not trade at the opening bell. Nasdaq runs a price-discovery cross first, and the first print lands anywhere from half an hour to several hours later — Google in 2004 and Meta in 2012 both printed more than two hours after 09:30.

That is not a rounding error. Computed with our own engine at the exchange, across the plausible 09:30–13:00 window:

The Ascendant sweeps 78.2° across four signs and the Midheaven 49.8° across three. Their sign is undetermined, not merely their degree. So this study uses no Ascendant, no Midheaven and no houses — only transiting planets against natal planets.

What was fixed before we looked

The hypothesis list, the orb, the aspect set, the horizon and the threshold were written into _data/prereg.py and committed to git before the price-fetching script existed (commit e68e557d, 2026-08-24 10:59:49 UTC). The commit timestamp is the evidence. A pre-registration you cannot date is a claim about our own discipline, which is worth nothing to a stranger.

• 7 transiting bodies (Sun, Mars, Jupiter, Saturn, Uranus, Neptune, Pluto) × 2 aspect classes = 14 hypotheses. That number is the multiple-comparison denominator, so the bar is p < 0.00357, not p < 0.05.
• Orb 1°, and the aspect set stated. Both are researcher degrees of freedom: twenty transits at 1° and twenty at 8° are forty hypotheses, not twenty. Hard = conjunction, square, opposition. Soft = sextile, trine.
• A "big move" is a 10.0% absolute log return over 7 calendar days — identical to the Bitcoin study on purpose. A replication that quietly moves its threshold is answering a different question. NVDA is far less volatile than Bitcoin, so this buys a lower base rate and weaker power, and we report that rather than fix it.
• The natal Moon is excluded entirely. Across the unknown-time window it drifts 2.025°, which is larger than the 1° orb — so a natal-Moon aspect cannot be said to happen at all. Every other body drifts under 0.234°, which is why the rest of the chart is safe to use.
• Market-specific rules, fixed in advance: only split- and dividend-adjusted closes; a transit perfecting on a weekend or holiday anchors to the next trading session; the horizon is calendar days, not sessions. Each of those is a choice someone could otherwise make after seeing which version helped.

The base rate comes first

Over all 6,937 sessions, NVDA moved more than 10.0% in a 7-day window 1,083 times — 15.6% of days, unprovoked. That is the number every row below has to beat. Roughly one week in six was a big week for this stock with no astrology involved at all, which is exactly the fact that makes "Saturn squared its Sun and it crashed" so easy to write and so empty.

All 14 results

Sorted by p-value, nulls included at the same size as everything else. n is how many times the transit occurred; k how many were followed by a big move.

Nothing survived correction. Bonferroni never had to do any work: not one row reached even an uncorrected p < 0.05 except one, and that one is taken apart below. With 14 tests you expect 0.7 rows under p < 0.05 by chance alone. We got 1.

The one row that would have made a good thread

Transiting Neptune, hard aspects came in at 38.9% against a 15.6% base — more than double — with an uncorrected p of 0.0146. It is the only row under 0.05, it does not survive correction, and it is published here rather than in isolation because one row in fourteen landing there is what chance looks like.

But there is a better reason to disbelieve it, and it is worth more than the p-value. Those 18 "events" are not 18 independent trials. They are 6 natal contacts visited 18 times, because an outer planet crosses the same point up to three times — direct, retrograde, direct again — over a matter of months. The sky did 6 things; we counted them 18 times.

Read the dates rather than the p-value. The big moves land in 1999, in 2003–04 and in 2008–09 — the dot-com run when NVDA was a penny stock, and the financial crisis. Neptune was not picking those years out; those years were volatile for everyone, and a slow planet that sits on one natal point for months is guaranteed to overlap whichever epoch it happens to fall in.

A robustness check we did not pre-register

The retrograde problem above applies to every slow-planet row, so we measured it across all 14. Passes of the same natal contact that belong to a single approach collapse into one trial.

The window has to be per body, and getting this wrong in the obvious way deletes real data. A single global window merged the Sun's waxing and waning square to the same point — about 182 days apart — when the Sun has no retrograde at all and every one of its crossings is a separate event. So each body gets a window above its retrograde loop, where one approach's crossings merge, and below its synodic return, where the next genuine recurrence survives: nothing for the Sun, 250 days for Mars, Jupiter and Saturn, 400 for Uranus, Neptune and Pluto — those three do not return to a natal degree for centuries, so a wide window cannot merge two real ones. The Sun's rows come through untouched at 1.0×, which is the check that the rule is not simply eating data.

This is a post-hoc diagnostic and is not one of the 14 pre-registered tests. It is shown beside them, never instead of them — choosing after the fact which analysis to headline is the precise move pre-registration exists to prevent.

And it is worth being exact about what it can do, because the tempting thing to claim is that a correction like this only ever tightens a result. That is not true, and our own numbers say so. Collapsing changes the hits and the trials together, so a p-value can move either way: 2 of these 14 rows come out more nominally significant, not less (Mars · hard, Neptune · soft). What makes the diagnostic honest is not its direction. It is that the pre-registered number stays the published one, and this never gets promoted into a second attempt at significance.

Neptune's hard aspects go from 0.0146 to 0.2380 once the same contact stops being counted three times. Under this correction not a single row reaches even an uncorrected p < 0.05.

The natal chart we used

22 January 1999, computed at 16:00 UTC — the midpoint of the plausible first-trade window. Because the time is unknown, the uncertainty is published rather than hidden: the right-hand column is how far each body could move across the whole window.

What this does and does not show

• It does not show astrology is false. It shows that these 14 specific claims, at this orb, on this horizon, on this stock, do not beat the base rate. A different orb or a different outcome could be tested — and would need its own pre-registration, not a re-run of this one until something lands.
• It is now a replication, which is the point. Bitcoin came back null on a chart with a perfect timestamp; NVIDIA comes back null on a chart with a perfect location. Two different subjects, two different markets, the same 14 hypotheses, the same answer.
• Power is honestly limited on the slow planets. Pluto's hard aspects have 26 events in 27 years, and after collapsing retrograde passes, 7. A rare transit cannot be ruled out by a sample this size, and we would rather say so than pretend the null is stronger than it is.
• This is history, not a forecast. Nothing here is financial advice, and nothing here predicts NVDA's next week.