Do Trump's tweets move the market?

Five things "everyone knows" about his posts and the stock market — each one tested against the check that decides it. No finance background needed.

generated 2026-07-15T19:13:47+00:00 · every figure computed at build time from data/real/bars.csv + corpus_v3.csv + the registered study — nothing hardcoded

Three ideas, and you can read the whole thing

1 · “beat the market”A stock going up isn't impressive if everything went up that day. So we always ask: did it beat the S&P 500 — the average of the 500 biggest US companies? Rising 3% on a day the market rose 3% counts as zero.

2 · “could it be luck?”Prices jump around on their own. So for every result we ask: how often would we see this by pure chance? We answer it by measuring the same thing on random days with no tweet. If the tweet days look like the random days, there's nothing there.

3 · “ask lots of questions…”…and a few will look amazing by luck alone — like flipping dozens of coins and bragging about the one that hit five heads. So we wrote every question down before looking at any answer, and every result is adjusted for how many questions we asked.

That's it. Everything below is those three ideas applied to real prices.

THE HYPOTHESIS: "Trump tweeted about Intel and the stock ran for months." Every broker has heard it. It is the single most-cited example of tweet-driven alpha.

The tweet

2026-04-29T22:20 — Intel Stock continues to rise. I’m very proud of that Company in that I am responsible for making the United States of America over 30 Billion Dollars in the last 90 days on that s

Note the tense. "continues to rise" — he is describing a move that has already happened, and claiming credit for it.

What INTC did BEFORE the tweet

window	INTC	SPY	INTC excess
prior 5d	+41.88%	+0.44%	+41.44%
prior 10d	+38.32%	+1.41%	+36.91%
prior 21d	+114.71%	+9.42%	+105.29%
prior 42d	+108.24%	+3.95%	+104.29%
prior 63d	+94.72%	+2.81%	+91.91%

What INTC did AFTER the tweet

abnormal return	1d	3d	5d	10d	21d	42d
INTC post-tweet	-3.02%	-2.11%	+8.94%	+11.30%	-2.42%	+13.82%

VERDICT — THIS TWEET IS REVERSE CAUSALITY. Intel had already doubled (+105% in 21 sessions) before the post. In the 21 sessions after it: −2.4%. The stock moved, then he tweeted — and memory encodes the correlation as a sequence.

Every Intel-the-company mention in the corpus (8 unique of 8,317 posts)

The famous tweet is the highlighted row. But it is not the whole story, and we are not going to show you only the row that suits us — so here is every mention, with what Intel did in the 21 trading days before and after each one. Both columns are measured against the S&P 500.

date	INTC in the 21 days BEFORE	INTC in the 21 days AFTER	the post
2025-08-07	-14.32%	+18.10%	The CEO of INTEL is highly CONFLICTED and must resign, immediately. There is no other solution to this problem
2025-08-11	-13.15%	+12.90%	I met with Mr. Lip-Bu Tan, of Intel, along with Secretary of Commerce, Howard Lutnick, and Secretary of the Tr
2025-08-22	+18.52%	+13.55%	It is my Great Honor to report that the United States of America now fully owns and controls 10% of INTEL, a G
2025-08-25	+18.52%	+13.55%	I PAID ZERO FOR INTEL, IT IS WORTH APPROXIMATELY 11 BILLION DOLLARS. All goes to the USA. Why are “stupid” peo
2025-12-08	+3.80%	-9.41%	I have informed President Xi, of China, that the United States will allow NVIDIA to ship its H200 products to
2026-01-08	+0.26%	+15.71%	I just finished a great meeting with the very successful Intel CEO, Lip-Bu Tan. Intel just launched the first
2026-04-29	+105.29%	-2.42%	Intel Stock continues to rise. I’m very proud of that Company in that I am responsible for making the United S
2026-06-18	+8.31%	n/a	The Technology the World relies on was invented in America. We all remember "Intel Inside." Stupid Presidents

Read this honestly. For the famous post the order is unmistakable: a double first, a tweet second, nothing after. But 3 of these 8 posts land the other way round — Intel was falling when he demanded the CEO resign, and rose afterwards. Those posts are also the ones where he was announcing something real: a president calling for a CEO's head, or the US government taking a 10% stake. That is news, and news moves prices.

So the honest claim is narrower than "his tweets never matter": this famous tweet described a move that had already happened. Eight mentions cannot settle the general question either way — eight is an anecdote, and picking the flattering ones out of eight is how the myth got built in the first place. The registered test in tab 4 is what settles it, and it finds nothing.

THE HYPOTHESIS: "After a big Iran/Hormuz tweet oil spikes on fear, then recovers — you can trade the arc." The shape is vivid and everyone remembers it.

Oil (USO): how it moved vs the S&P 500

which tweets	1d	3d	5d	10d	21d	42d
top-5 'strongest' tweets	-5.34%	-3.99%	+0.01%	+4.03%	+8.03%	+12.71%	the anecdotes
all 65 tweet-days	-0.21%	+0.41%	+0.53%	+2.47%	+5.62%	+5.34%	the population
random days (n=120)	+0.13%	+0.12%	+0.51%	+0.48%	+2.30%	+5.34%	no tweet, any day
BIG move, NO tweet (n=3)	+2.40%	+3.14%	+10.06%	+14.09%	+22.34%	+30.10%	the control that matters

top-5 tweets @42d

+12.71%

big move, NO tweet @42d

+30.10%

VERDICT — MEAN REVERSION. The dip-then-recovery arc is real — and it is stronger on big oil moves with no tweet at all. It is what volatility does, not what Trump does. All 65 tweet-days are indistinguishable from random days (+5.34% vs +5.34% @42d).
Control n=3 is small; it agrees with the population comparison, which is flat.

THE HYPOTHESIS: "Just look at the biggest moves — the tweets that moved oil are obviously geopolitical." So we sorted the oil events by biggest day-1 move and read the tweets. Here they are, unedited.

2026-03-09 — USO -13.23% on day 1

Short term oil prices, which will drop rapidly when the destruction of the Iran nuclear threat is over, is a very small price to pay for U.S.A., and W

2025-06-23 — USO -8.76% on day 1

I am very happy to report that I have arranged, along with Secretary of State Marco Rubio, a wonderful Treaty between the Democratic Republic of the C

2025-04-08 — USO -4.52% on day 1

“Rates are plummeting, oil prices are plummeting, deregulation is happening. President Trump is not going to bend..."

2026-03-03 — USO -4.41% on day 1

The Radical Left Democrats, a Party that has completely lost its way, are complaining bitterly about the very necessary and important attack, by the U

2026-03-13 — USO +4.23% on day 1

The Iran National Soccer Team is welcome to The World Cup, but I really don’t believe it is appropriate that they be there, for their own life and saf

VERDICT — SELECTION BIAS. A soccer-team tweet and a "Radical Left Democrats" tweet sit in the top five "oil movers." Neither is about oil. With posts on 65 of 312 sessions (21% of the calendar), every large move has a same-day tweet available to blame. Sorting on the outcome manufactures a greatest-hits reel out of coincidence — this is the anecdote generator, visible in one table.

The registered test

We sorted every tweet into a type — war/geopolitics, company, or ordinary — using only the words in the tweet, never the result. For each type we asked three questions about its matching assets over 1, 3 and 5 days: did the price move in the direction expected? how big was the move? was there unusual trading?

Two rules keep this honest. We wrote down all 63 questions before looking at a single answer — otherwise you can ask a hundred things and report the three that worked. And every answer is compared against the same asset on random days, so "oil moved" only counts if it moved more than oil normally does.

questions asked
(fixed in advance)

questions with a
real answer

best result: chance
it was luck

76%

how many tweets of each type

war/geopolitics

1158

ordinary

company

The six that came closest — and still failed

tweet type / asset / days	how many tweets	after these tweets	on random days	chance it's luck	...after testing 63 things
war/geopolitics USO · 1d · how much trading?	193	+1.9728	+1.6505	0.038	0.755
war/geopolitics USO · 3d · how much trading?	193	+1.9726	+1.6931	0.064	0.755
war/geopolitics GLD · 5d · did it move as expected?	193	-0.0064	-0.0030	0.065	0.755
war/geopolitics ITA · 5d · did it move as expected?	193	-0.0045	-0.0022	0.074	0.755
ordinary SPY · 3d · did it move as expected?	92	+0.0032	+0.0002	0.079	0.755
war/geopolitics GLD · 3d · did it move as expected?	193	-0.0049	-0.0025	0.080	0.755

The last column is the one that matters. Ask 63 questions and a few will look impressive by chance alone — like flipping 63 coins and celebrating the one that came up heads five times. That column adjusts for how many questions we asked. Nothing survives it.

THE ANSWER: 0 out of 63 questions found anything real. Not the direction of the move, not its size, not the trading volume.

But "nothing happens" is the wrong way to read this. The right reading is: by the time he posts, the market already knows. The news that moves a price is the actual event — the strike, the tariff, the earnings — and prices react to that, usually before he writes about it. His post is a description of something already priced in.

What this terminal is not

This is not a trading tool and it gives no buy/sell advice. The live service will read a tweet and tell you what it's about and which assets it touches — but it deliberately quotes no accuracy number and no confidence score, because nothing we measured earned one. Research output. Not investment advice.

WHAT THIS TAB DOES: a machine reads one of his tweets — just the words, no prices — and says which assets it will move and in which direction. Then we check it against what the market actually did. Expected vs reality, side by side.

Pick a tweet, then drag the two sliders to choose how far back and how far forward to look. Every number is a real measured price from that exact day — the slider looks it up, it never estimates or draws a smooth curve.

THIS TAB IS A LOOKING-GLASS, NOT A TEST. 20 instruments across 22 tweets is hundreds of comparisons, and at that number something always looks like a pattern. Nothing here is evidence — it is where you form a hunch. Tabs 4 and 6 are where hunches get checked properly.

1 · anchor tweet asset (same panel for every tweet)

2 · sliding windows

lookback — T − 21 sessions

lookforward — T + 21 sessions

Everything is measured relative to the S&P 500: if a stock rose 3% on a day the whole market rose 3%, that counts as zero — it just went along for the ride. Zero on the chart is the moment of the first market open after the tweet — the earliest point anyone could actually have acted on it.

Event trace

T − 21 sessions T₀ — first session after the post T + 21 sessions

Hover the line for the exact date and value on any day. Up means it beat the S&P 500; down means it lagged. The gold line is the tweet — left of it is before, right of it is after.

move BEFORE the tweet
(vs the S&P 500)

—

move AFTER the tweet
(vs the S&P 500)

—

what the two numbers say

—

This line is worked out live from the two numbers above it — if the price had already moved a lot before the tweet, and barely moved after, the tweet didn't cause it. It's not a saved conclusion; change the sliders and it changes.

“When do we predict correctly?”

We split the tweets by date: the model was studied on the earlier ones ("train") and then checked against later ones it had never seen ("test") — the honest way to find out if a pattern is real or imagined.

Each row is a group of tweets. "0.304" means it got 30% of them right; 0.5 is a coin flip. If a group were genuinely easier to predict, its early score should hold up on the later tweets. None of them do.

group of tweets	earlier tweets (studied)	later tweets (unseen)	change
when he posted weekend	0.304 n=46	0.765 n=17	+0.460
what it's about everything else	0.420 n=50	0.714 n=7	+0.294
how loud it is level 9 of 10	0.387 n=31	0.571 n=7	+0.184
when he posted before the bell	0.372 n=43	0.545 n=22	+0.173
how loud it is level 6 of 10	0.435 n=23	0.600 n=10	+0.165
what it's about economy/policy	0.469 n=32	0.600 n=5	+0.131
what it's about war/geopolitics	0.364 n=118	0.491 n=55	+0.127
when he posted market open	0.429 n=63	0.316 n=19	-0.113
how loud it is level 8 of 10	0.394 n=142	0.490 n=49	+0.095
when he posted after the bell	0.460 n=50	0.444 n=9	-0.016

ANSWER: nowhere that holds. Every pattern flips when you check it on tweets the model never saw — weekend posts go from the worst group (30% right) to the best (76% right). Learn “avoid weekends” from the earlier tweets and the later ones punish you; learn “weekends are 76%” from the later ones and the next batch punishes you. We also tried the obvious next move — a second model whose only job was to spot when the first one could be trusted. It looked promising while it was being built and then scored worse than a coin flip on tweets it had never seen. These groups aren't real patterns — they're random wobble, and random wobble never repeats.

THE BELIEF BEING TESTED: when he posts about Iran, or a military strike, oil and the "fear gauge" should jump within the hour — and looking at whole days is too slow to catch it. Zoom in to the hour and the real market-movers should stand out.

Design

We compare the hour after each tweet with the hour before the same tweet. That matters: he only sometimes posts while the market is open, and those moments might be special for other reasons — using the same tweet's own "before" picture cancels that out.

We also compare against his ordinary posts (the ones about nothing in particular). "War tweets move oil" only means something if they move oil more than his everyday posts do.

Every price is measured against the S&P 500, so a day when the whole market moved doesn't get mistaken for a tweet effect. Prices come from free public data (anyone can re-run this), and we wrote down every question before looking at any answer — so we can't quietly pick the ones that worked.

tweets posted while
the market was open

2371

questions asked
(fixed in advance)

questions with a
real answer

best result: chance
it was luck

56%

The eight that came closest — and still failed

tweet type / asset	what we measured	how many tweets	after these tweets	after ORDINARY tweets	difference	smallest gap we could detect	chance it's luck	...after testing 30 things
war/geopolitics USO	how big was the move?	228	+0.0070	+0.0053	+0.0017	0.0025	0.025	0.558
war/geopolitics XLE	how big was the move?	228	+0.0036	+0.0030	+0.0006	0.0011	0.051	0.558
war/geopolitics VIXY	how much trading?	228	+1.3023	+1.5686	-0.2663	0.0037	0.077	0.558
economy/policy TLT	did it beat the market?	123	-0.0005	+0.0002	-0.0007	0.0013	0.103	0.558
economy/policy GLD	how much trading?	105	+1.1494	+1.3491	-0.1998	0.0015	0.126	0.558
economy/policy TLT	how big was the move?	123	+0.0030	+0.0026	+0.0005	0.0013	0.130	0.558
war/geopolitics ITA	how much trading?	226	+1.4038	+1.2081	+0.1957	0.0008	0.143	0.558
economy/policy UUP	did it beat the market?	123	-0.0003	+0.0003	-0.0006	0.0013	0.162	0.558

Why the "smallest gap we could detect" column matters. With 228 war-tweets, a real effect of 0.25% on oil would have shown up clearly. What we actually measured was 0.17% — smaller than our detector, and well inside normal random wobble.

So this isn't "our instruments were too blunt to tell." It's: if an effect this size existed, we would have seen it. We didn't.

THE NEAR-MISS — how we almost published the opposite. Our first run said exactly what everyone expects: 9 solid findings, odds of luck about 1 in 550 — "war tweets spike oil trading by 79%".

It was a counting mistake. He posts in bursts — five tweets in ten minutes. Our data has one price reading per hour, so all five tweets pointed at the same price move. We counted one event five times and called it five pieces of evidence. Two posts a minute apart were both credited with the same gold move.

Count each hour once and the 9 findings become zero. We only caught it because the "biggest movers" list had a tweet about a renovated Palm Room at the top — the statistics looked perfect.

The biggest movers — now read the tweets

size of move vs a normal hour	asset	tweet type	tweet
-11.16σ	GLD	NOISE	The renovated Palm Room!
+10.02σ	USO	NOISE	Governor Gavin Newscum should immediately go to Northern California an
+8.26σ	XLE	GEOPOLITIC	This morning, on my Orders, U.S. Military Forces conducted a SECOND Ki
-8.25σ	GLD	GEOPOLITIC	Iran has officially responded to our Obliteration of their Nuclear Fac
+8.23σ	VIXY	NOISE	Everybody watch Sean Hannity tonight at 9 P.M. Bibi Netanyahu will be
-8.05σ	XLE	NOISE	I am thrilled to report that the U.S. Space Command headquarters will

Look at which tweets top this list: a renovated Palm Room. A plug for a TV show. These are not market news — they're just big market moves that happened to have a tweet in the same hour.

And that will always happen: he posts during roughly 1 in 4 market hours. So every big move has a tweet sitting next to it, and you can always find one that looks like the cause. That's the whole illusion, in one table.

THE ANSWER: 0 out of 30 questions found anything real. War and geopolitics tweets do not move oil, gold, defence stocks or the fear gauge more than his ordinary posts do — not the price, not the trading volume, not within the hour.

Sorting tweets into categories doesn't find the "big movers" either. And this was our most sensitive test: it could have caught an effect five times smaller than anything our day-scale tests could see. It still found nothing.

Research output. Not investment advice. Reproduce: make dashboard. Full method: experiments/event_study/REPORT.md.