Research
Who Said It
Trump and Musk are supposed to move markets by speaking. Measured against the right control, the speaker turns out to be worth almost nothing.
These are not the posts themselves.
No archive of historical tweets was available, so every event here is a newswire report of something Trump or Musk said — filed minutes to hours after the statement. That is a real limitation, and it is stated up front rather than buried. It matters less than it sounds, because the test that carries this page enters at the following market open regardless.
Eighteen tests across both speakers at one, two and three days out. None survived
correction for multiple comparisons; the best reached q = 0.994.
The interesting part is not that null. It is the one result that looked like it might
be real.
Musk moves Tesla — apparently
A companion study found that news filed before the opening bell produces a gap that keeps drifting through the session. Applying that test to statements, split by speaker and by stock:
| group | events | drift edge | p | q |
|---|---|---|---|---|
| Musk → TSLA | 435 | +30.9b | 0.019 | 0.094 |
| Trump → NVDA | 80 | +38.7b | 0.115 | 0.192 |
| Trump, all symbols | 659 | +16.5b | 0.058 | 0.144 |
| Musk, all symbols | 897 | +10.5b | 0.161 | 0.202 |
Musk on Tesla: +30.9 basis points at p = 0.019,
across 435 pre-market events. Half again the size of the general news effect. This is the
number that would headline a different version of this page.
Compared against what?
+30.9 bps is only meaningful against a baseline, and the baseline decides the answer. Measured against the whole market it looks like a Musk effect. But Tesla is not the whole market. The honest comparison is other pre-market news about the same stock — same company, same mechanism, same hour of the morning, different speaker.
| Tesla, pre-market news | events | drift |
|---|---|---|
| driven by Musk | 454 | +61.8b |
| everything else | 389 | +55.0b |
| difference | — | +6.8b, CI [−26.3, +39.1], p = 0.691 |
| All symbols, pre-market news | events | drift |
|---|---|---|
| driven by Trump | 797 | +23.2b |
| everything else | 6,169 | +21.3b |
| difference | — | +1.9b, CI [−12.9, +18.4], p = 0.806 |
Tesla drifts roughly 2.6× as far after pre-market news as the average large cap — 55 to 62 basis points against 21. That is Tesla being Tesla: high beta, heavy retail ownership, a crowded overnight tape.
Musk adds +6.8 bps on top, with a confidence interval running from −26 to +39. Trump adds +1.9 bps. Neither is distinguishable from zero. Swap in any other newswire story about Tesla that morning and the price behaves the same way.
They do not even move prices further
| group | |gap| | vs. control | |intraday| | vs. control |
|---|---|---|---|---|
| Musk on TSLA | 132b | 1.09× (p=0.286) | 200b | 1.10× (p=0.113) |
| Trump, all | 114b | 1.11× (p=0.093) | 140b | 1.00× (p=0.986) |
Direction aside, a statement from either man might at least be expected to shake the price harder than routine coverage. It does not, by any margin the data can separate from noise. Roughly a tenth larger, with p-values that would embarrass a horoscope.
When they talk
72.9% of Musk-driven Tesla headlines land before the opening bell, against 45.7% for news generally. He posts at hours when no one can trade, which is the single clearest way his behaviour differs from an ordinary corporate newswire.
It does not create an opportunity. It means the market has hours to digest each statement before anyone can act, and the opening auction prices it alongside everything else that happened overnight. The reason the speaker does not matter may simply be that by 9:30 the speaker is old news.
What would undercut this
Reports, not posts
Every event is a wire story about a statement, not the statement. A direct archive with original timestamps would test the seconds after a post — a window this data cannot see at all.
Power
The Musk–Tesla comparison detects a difference of about 37 basis points at 80% power. A genuine speaker premium smaller than that would not show up here. The finding is “no large effect”, not “no effect”.
Thirty large caps
Statements aimed at small caps, private companies, crypto or currencies are outside this universe entirely — and those are exactly where a single post plausibly matters most.
A crude classifier
Direction comes from pattern matching on headline text. Sarcasm, hedging and political framing all read poorly to it.
Headlines mentioning each figure are classified for direction, then measured from the opening auction to the close on the following session, minus SPY over the same window. The control is other pre-market news on the same symbols over the same five years. Significance is a Welch t-test with a bootstrap interval, corrected with Benjamini–Hochberg.
This is a measurement of historical data. It is not investment advice, a forecast, or a recommendation about any security.