Do social media posts move the stock market? Published studies of social media posts and the stock market, most of them on posts from one U.S. president's account, find effects that depend on the topic, the tone, the asset and how finely prices are measured. Several found no significant average effect [10, 11, 35], and many of the effects that did appear faded within hours or days [8, 9, 12, 14].
This report sets the event studies side by side and gives null results the same space as effects.
What studies find on social media posts and the stock market
Six studies tested whether a U.S. president's posts that named listed companies moved those companies' stocks [8, 9, 10, 11, 12, 13].
| Study | Posts | Prices | Result |
|---|---|---|---|
| Born, Myers and Clark | 15 [11] | Event-day open and close | Moves with the post's tone, not significant within five trading days [8] |
| Ge, Kurov and Wolfe | 48 [11] | Not stated in the abstract | Moves in price, volume and volatility, some reversal in the next few trading days [9] |
| Juma'h and Alnsour | 58 [11] | Daily | No significant effect on average [10] |
| Brans and Scholtens | 100 [11] | Daily | Average abnormal return of 0.1% on the post day and -0.07% the next, neither significant [11] |
| Ajjoub, Walker and Zhao | 513 [12] | Not stated in the abstract | Declines after negative posts about non-media firms, largely reversed the next day [12] |
| Machus, Mestel and Theissen | Not stated in the abstract | Minute by minute | More trading, no lasting price effect [13] |
The samples range from 15 to 513 posts [11, 12], and the rules for choosing them differ. Brans and Scholtens kept 100 of over 5,600 posts after excluding retweets and companies with mergers, splits, profit warnings or top-management changes [11]. Ajjoub, Walker and Zhao kept 513 of 6,983 posts and found that the response depended on the firm: for media firms positive posts had a pronounced positive price impact, and for non-media firms negative posts were followed by declines [12].
Trading responded more consistently than prices [8, 9, 13]. Born and coauthors found trading volume and search activity rising with the posts, which they read as transitory trading by small or noise traders [8]. Ge and coauthors found higher volume, volatility and institutional investor attention [9].
Tone changes the answer
Splitting posts by tone gives conflicting results. Brans and Scholtens scored posts with SentiStrength, a word-list tool, and found strongly negative posts followed by lower company value, with supportive posts showing no significant effect [11]. The gap between negative and positive posts was significant at the 10% level [11]. Ge and coauthors, who coded tone by hand, found no significant difference between negative and positive posts [11]. Kinyua and coauthors, with 30-minute windows on the Dow Jones Industrial Average and the S&P 500, found positive market reactions to strongly positive and strongly negative posts alike [15].
The ways of scoring tone, from word lists to language models, are compared in Dictionaries, FinBERT or language models for financial sentiment.
Effects measured in minutes
Gjerstad, Meyn, Molnár and Næss matched 8,686 posts to minute-by-minute futures on the S&P 500, the VIX, WTI crude, the Hang Seng and gold to September 2019 [14]. Pooled over all posts, the S&P 500 effect was negative, not very significant and gone within an hour, while trading volume stayed higher for at least an hour [14]. After posts containing "products", returns fell about 30 basis points over 60 minutes, and after "tariff" about ten basis points, with both effects gone within two hours [14]. Most topics did not affect markets. Trade-war posts were followed by lower U.S. prices, more trading, more uncertainty, a negative response in Chinese stocks and a positive one in gold [14].
Burggraf, Fendel and Huynh found that posts on the U.S.-China trade war negatively predicted S&P 500 returns and positively predicted the VIX [16]. Nishimura, Dong and Sun found that posting raised volatility and jumps in the later part of their sample [17], and Nishimura and Sun found higher volatility in German, French and UK stocks, rising with attention to the posts [18].
Currencies, oil and interest rates
Mexico-related policy posts raised the peso's volatility [19]. Escalating negative sentiment in Russia-related posts was followed by a ruble depreciation of 4% to 10% over about three days, while sanctions not accompanied by negative posts did not cause depreciation [20]. Posts with macroeconomic policy or trade content were followed by a stronger dollar and lower intraday volatility [21]. Some posts raised the likelihood of a wider spread between WTI and Brent crude [22].
Bianchi, Kind and Kung timed posts criticizing the Federal Reserve to the second against tick-by-tick fed funds futures and found that expected policy rates fell, while placebo tests with randomly selected other posts gave coefficients close to zero [23]. Their design is one of the cases in the event-study methods note. Tillmann found that the president's statements on the Fed lowered long-term interest rates on average, with an impact that declined over time [24].
A chief executive's posts as disclosure
In April 2013 the SEC said companies can announce key information on social media in compliance with Regulation FD, provided investors have been alerted to the channels [2]. The occasion was a post by a streaming company's chief executive on his personal Facebook page, set out in company disclosure timing and the 23-hour trading day [3].
In 2018 the SEC alleged that an electric-car maker's chief executive posted false and misleading statements about taking the company private at $420 per share with funding secured, and that the posts caused the stock to jump by over 6% on 7 August 2018 [4]. He and the company settled without admitting or denying the allegations [5]. The terms required him to step down as chairman for three years and each to pay a $20 million penalty, and the company, which had told the market in 2013 that it intended to use his account to announce material information, was charged with lacking disclosure controls over his posts [5]. A 2023 Second Circuit summary order upheld the district court's refusal to modify or end the consent decree, which requires him to follow the company's policy of pre-approval of covered posts by company lawyers, and the Supreme Court denied a petition for a writ of certiorari on 29 April 2024 [6, 7].
Crypto assets and one executive's posts
Ante studied 47 crypto-related posting events by the same chief executive [25]. Price effects were significant on average for Dogecoin-related posts and not for Bitcoin, where positive and negative effects cancel out, and individual posts moved Bitcoin's price up by 16.9% or down by almost 11.8% [25]. With daily data, Huynh found that the tone of 10,850 posts had a Granger causal relation with Bitcoin returns, with reversal effects and no link to volatility [27].
Crowd posts and mood indexes
Bartov, Faurel and Mohanram found that the aggregate opinion in individual posts predicted firms' quarterly earnings and announcement returns from 2009 to 2012 [29]. A 2011 paper reported 87.6% accuracy in predicting the daily up and down changes of the Dow Jones Industrial Average from a Twitter mood series [31]. Lachanski and Pav could not reproduce its p-value pattern and found no evidence that mood helped predict the market out of sample [32].
In January 2021, the SEC staff found, GameStop and other stocks rose sharply "as bullish sentiments of individual investors filled social media", and positive sentiment sustained the weeks-long rise [34]. A Federal Reserve discussion paper found that overnight sentiment in posts helped predict next-day stock returns [33].
Why the studies disagree
The studies differ in four ways that change results:
- Selection. Posts picked by hand or by content, against every post except retweets [11, 14, 21].
- Data frequency. Daily closes against minute and tick data. The pooled and keyword effects in the minute-level data faded within one to two hours, while the trade-war topic's effect on the S&P 500 stayed significant through the full two-hour window [14]. One daily study calls the lack of high-frequency data a limitation [11], and another does not rule out intraday reactions [35].
- Timing and cause. Prices and trading began to move before posts in two minute-level studies [13, 14]. One reads this as posts commenting on events that had already drawn attention [13], the other as consecutive posts on the same topic [14].
- Missing posts and outliers. One archive missed posts deleted within a day before 27 January 2017, and one large drop was removed from that sample [11].
For 2025 posts, a May 2026 bachelor's thesis at Lund University used daily data and found no significant association between tariff posts and daily VIX changes [35]. Tariff-related posts were assigned to 56 of 85 trading days, leaving 29 comparison days [35].
What a study of posts needs
Each of these practices comes from a study above: timestamps to the minute or the second [14, 23], every post kept as issued, deletions included [11], a baseline of ordinary windows or placebo posts [14, 23], a stated rule for choosing posts [11, 12, 21], null results reported in full [10, 11, 35], and a test outside the sample that produced the finding [32].
Rillor's event and news response work runs event studies on a time-aligned corpus of public communications, each statement kept as issued with its timestamp and paired with the price moves in a declared window around it.
Read the full report
The full report adds each study's sample, data and window, and the SEC record through the 2024 Supreme Court docket.