Blog
Do finance YouTubers beat the market? Four channels, checked
By Callsight · Published Oct 9, 2026 · Updated Oct 9, 2026
Figures as of Nov 3, 2026. They change as new windows settle.
As of Nov 3, 2026, these figures cover only the four finance YouTube channels Callsight tracks. At one year, the 95% range of every channel’s average return against its benchmark includes zero. On calls with a base rate, hit rates run from 46.9% of 160 (base rate 49.2%) to 69.1% of 68 (61.2%), mostly reflecting how often the called moves happened anyway.
Do finance YouTubers beat the market?
Callsight tracks four channels: Graham Stephan, Andrei Jikh, Joseph Carlson and Coin Bureau. What follows is their record, and it says nothing about the many other finance channels on YouTube, which Callsight has not checked. So far the record has a verdict on 766 of their calls, each counted once however many of its windows have closed.
For each channel the answer is a single average. Every price call that has a benchmark is measured from the first close after the video went public: the return on the price, dividends included, less the benchmark’s return over the same days, reversed for a call that a price would fall. The benchmark is the S&P 500 for shares, funds, Bitcoin and Ethereum, and Bitcoin for other coins.
| Channel | Calls decided at one year | Price calls with a benchmark | Average against the benchmark | 95% range |
|---|---|---|---|---|
| Graham Stephan | 104 | 13 | +9.9 pts | −67.8 to +87.5 pts |
| Andrei Jikh | 51 | 15 | −6.4 pts | −31.4 to +18.6 pts |
| Joseph Carlson | 70 | 47 | −5.1 pts | −12.6 to +2.4 pts |
| Coin Bureau | 177 | 114 | −5.3 pts | −20.4 to +9.8 pts |
Every channel’s 95% range runs from below zero to above it, so on these calls none of the four can be told apart from its benchmark. As more calls are checked, an average could settle on either side. A few large moves can pull an average of returns a long way, which is why some ranges are wide. The same holds at every other window from 30 days to 5 years, wherever an average can be given.
This answer covers price calls with a benchmark only: 189 of the 402 calls decided at one year. Calls on the economy have no benchmark. Neither do price targets, which are judged on whether a level was reached, or calls on the S&P 500 itself, which is the benchmark. The article on how often the channels’ price targets were hit counts the targets channel by channel.
Four of the calls behind these averages follow, one from each channel, picked by the rule at the end of this article. Each card shows the return beside the benchmark’s over the same days, and the verdict at each window up to a year.
“just as quickly as it could go up it can and will go back down just as people bought in gamestop at 450 a share and subsequently lost 65 of their investment people will lose money on dogecoin if they don't know what they're doing”Entry price (first close after the video)
$0.365443
Apr 16, 2021
Price at close
$0.1435
Apr 16, 2022
Return
−60.7%
Bitcoin −34.2%
DOGE closed at $0.365443 on Apr 16, 2021. On Apr 16, 2022 it closed at $0.1435, a return of −60.7%. BTC returned −34.2% over the same window. The call was that it would fall.
30 days
Incorrect
90 days
Correct
180 days
Correct
1 year
Correct
No deadline was stated, so the headline is the one-year window, where it is Correct. At 30 days it was Incorrect.
“for Bitcoin I think the next 12 months are going to be super interesting because Bitcoin tends to lag behind Gold's performance so first gold goes up and then Bitcoin goes up”Entry price (first close after the video)
$61,760
Sep 18, 2024
Price at close
$117,074
Sep 18, 2025
Return
+89.6%
S&P 500 +19.4%
BTC closed at $61,760 on Sep 18, 2024. On Sep 18, 2025 it closed at $117,074, a return of +89.6%. SPY returned +19.4% over the same window. The call was that it would rise.
30 days
Correct
90 days
Correct
180 days
Correct
1 year
Correct
The words speak of the next 12 months, the same length as the one-year window, where it is Correct.
“I think that Google will continue to go higher.”Entry price (first close after the video)
$230.66
Sep 3, 2025
Price at close
$342.48
Sep 3, 2026
Total return, dividends included
+48.9%
S&P 500 +21.4%
GOOGL closed at $230.66 on Sep 3, 2025. On Sep 3, 2026 it closed at $342.48, a total return of +48.9% (dividends included). SPY's total return over the same window was +21.4%. The call was that it would rise.
30 days
Correct
90 days
Correct
180 days
Correct
1 year
Correct
No deadline was stated, so the headline is the one-year window, where it is Correct.
“so while circle CRCL stock could rise further from current levels
alongside the broader crypto market we wouldn't expect the kind of crazy returns we've seen up to
now”Entry price (first close after the video)
$222.65
Jun 24, 2025
Price at close
$70.98
Jun 24, 2026
Total return, dividends included
−68.1%
S&P 500 +22.2%
CRCL closed at $222.65 on Jun 24, 2025. On Jun 24, 2026 it closed at $70.98, a total return of −68.1% (dividends included). SPY's total return over the same window was +22.2%. The call was that it would rise.
30 days
Incorrect
90 days
Incorrect
180 days
Incorrect
1 year
Incorrect
No deadline was stated, so the headline is the one-year window, where it is Incorrect. The words pair a possible rise with a caution against expecting returns like those before, and the record reads the call as one for a rise.
How accurate are finance YouTubers?
At one year, the calls of three of the four channels Callsight tracks were Correct more often than the base rate of the same calls, and those of one less often; the differences run from −2.3 to +15.4 points. The table gives the hit rate at 30 days, 1 year and 3 years, three of the six windows every call is checked at. Beside it are the base rate, which is how often the move a call named happened anyway over a span that long, and how far the hit rate stands above that base rate, both taken on the same calls.
| Channel | Window | Correct | Incorrect | Pending | Hit rate | Base rate | Above the base rate |
|---|---|---|---|---|---|---|---|
| Graham Stephan | 30 days | 49 | 48 | 1 | 50.5% (97) | 33.0%On the 60 of 97 calls that have a base rate: 43.3% correct, base rate 33.0% | +10.3 pts |
| 1 year | 63 | 41 | 20 | 60.6% (104) | 47.2%On the 84 of 104 calls that have a base rate: 57.1% correct, base rate 47.2% | +10.0 pts | |
| 3 years | 64 | 25 | 42 | 71.9% (89) | 51.9%On the 73 of 89 calls that have a base rate: 69.9% correct, base rate 51.9% | +17.9 pts | |
| Andrei Jikh | 30 days | 22 | 27 | 1 | 44.9% (49) | 32.1%On the 43 of 49 calls that have a base rate: 37.2% correct, base rate 32.1% | +5.1 pts |
| 1 year | 30 | 21 | 7 | 58.8% (51) | 41.7%On the 49 of 51 calls that have a base rate: 57.1% correct, base rate 41.7% | +15.4 pts | |
| 3 years | 20 | 14 | 28 | 58.8% (34) | 48.7%On the 33 of 34 calls that have a base rate: 57.6% correct, base rate 48.7% | +8.9 pts | |
| Joseph Carlson | 30 days | 29 | 42 | 0 | 40.8% (71) | 50.0%On the 66 of 71 calls that have a base rate: 37.9% correct, base rate 50.0% | −12.1 pts |
| 1 year | 49 | 21 | 7 | 70.0% (70) | 61.2%On the 68 of 70 calls that have a base rate: 69.1% correct, base rate 61.2% | +7.9 pts | |
| 3 years | 31 | 10 | 42 | 75.6% (41) | 67.5%On the 40 of 41 calls that have a base rate: 75.0% correct, base rate 67.5% | +7.5 pts | |
| Coin Bureau | 30 days | 76 | 103 | 1 | 42.5% (179) | 45.7%On the 161 of 179 calls that have a base rate: 43.5% correct, base rate 45.7% | −2.2 pts |
| 1 year | 82 | 95 | 43 | 46.3% (177) | 49.2%On the 160 of 177 calls that have a base rate: 46.9% correct, base rate 49.2% | −2.3 pts | |
| 3 years | 50 | 40 | 267 | 55.6% (90) | 47.2%On the 56 of 90 calls that have a base rate: 60.7% correct, base rate 47.2% | +13.5 pts |
The one-year window is the headline for every call on the site, fixed in advance so that nobody can pick, after the fact, the window that reads best. A call can be Incorrect at 30 days and Correct at a year, and each channel’s page has every window.
Is the most accurate finance YouTuber the one with the highest hit rate?
A hit rate alone cannot say, because some calls come true most of the time whoever makes them. The base rate measures how often. For each call, it is the share of past spans of the same length, in the instrument’s own history up to the day of the call, in which the move the call named happened. A call that a price will rise starts with a high base rate when that price has risen over most such spans.
The largest sample of stock calls among the four channels is Joseph Carlson’s. At one year, 36 of those calls on stocks and funds were Correct and 18 Incorrect, a hit rate of 66.7% on 54. The base rate on the same 54 calls is 68.7%, so a call like these came true more often than not whoever made it, and these calls came true less often than their base rate.
A table sorted by hit rate would therefore mislead, and the record page orders the channels by a score. The score sets the hit rate against the base rate on the calls that have one, adds the return against the benchmark and calibration, and shrinks each part when there is little evidence, so that 50 means no different from the base rate. A channel with fewer than 20 independent calls is listed below the ranked ones, and none is hidden. The tables in this article keep the record page’s order, and the methodology page defines the score.
How much of what they say can be checked?
On each of the four channels, more than half of the statements made in the channel’s own words were too vague to check, from 53.4% of 506 statements to 61.3% of 302. Such a statement is counted and never scored, so a vague remark is never marked Incorrect and never adds to a hit rate.
| Channel | Own statements | Too vague to check | Counted calls | Read out from others | Own calls held out |
|---|---|---|---|---|---|
| Graham Stephan | 506 | 53.4% | 226 | 169 | 155 |
| Andrei Jikh | 302 | 61.3% | 103 | 55 | 74 |
| Joseph Carlson | 366 | 54.1% | 129 | 19 | 97 |
| Coin Bureau | 1,539 | 60.3% | 586 | 205 | 144 |
Two more kinds of call are listed and left out of every rate and average in this article. A forecast that a channel reads out from someone else is shown with its source and counted for no one. A channel’s own call is held out for reasons such as a position the channel already holds or a direction in doubt, and its card says why. This is how a forecast read out from a bank appears:
“Goldman Sachs just raised their year-end S&P 500 target to 8,000, up from 7,600, which implies solid upside from where we are today.”Graham Stephan is reading out a forecast by Goldman Sachs. The card names the source, and the forecast is in no channel’s figures.
Are crypto calls more accurate than stock calls?
Hit rates from different subjects cannot be set side by side, because the base rates differ: at one year they run from 50.0% to 59.7% on crypto, from 50.5% to 68.7% on stocks and funds, and from 29.8% to 38.8% on the economy. On crypto, all three channels’ calls were Correct less often than their base rate. On stocks and funds, the calls of two channels were Correct more often than their base rate and those of two less often. On the economy, all four channels’ calls were Correct more often than their base rate.
| Subject | Channel | Correct | Incorrect | Hit rate | Base rate | Above the base rate |
|---|---|---|---|---|---|---|
| Crypto | Graham Stephan | 3 | 7 | 30.0% (10)fewer than 20 calls | 50.0%On the 8 of 10 calls that have a base rate: 37.5% correct, base rate 50.0% | −12.5 pts |
| Andrei Jikh | 6 | 6 | 50.0% (12)fewer than 20 calls | 59.7%on 12 of 12 calls | −9.7 pts | |
| Coin Bureau | 58 | 72 | 44.6% (130) | 53.0%On the 118 of 130 calls that have a base rate: 44.1% correct, base rate 53.0% | −8.9 pts | |
| Equities and funds | Graham Stephan | 14 | 7 | 66.7% (21) | 68.0%on 21 of 21 calls | −1.3 pts |
| Andrei Jikh | 6 | 4 | 60.0% (10)fewer than 20 calls | 52.4%on 10 of 10 calls | +7.6 pts | |
| Joseph Carlson | 36 | 18 | 66.7% (54) | 68.7%on 54 of 54 calls | −2.1 pts | |
| Coin Bureau | 7 | 4 | 63.6% (11)fewer than 20 calls | 50.5%On the 9 of 11 calls that have a base rate: 66.7% correct, base rate 50.5% | +16.2 pts | |
| The economy | Graham Stephan | 46 | 27 | 63.0% (73) | 38.8%On the 55 of 73 calls that have a base rate: 56.4% correct, base rate 38.8% | +17.5 pts |
| Andrei Jikh | 18 | 11 | 62.1% (29) | 29.8%On the 27 of 29 calls that have a base rate: 59.3% correct, base rate 29.8% | +29.5 pts | |
| Joseph Carlson | 13 | 3 | 81.3% (16)fewer than 20 calls | 32.0%On the 14 of 16 calls that have a base rate: 78.6% correct, base rate 32.0% | +46.5 pts | |
| Coin Bureau | 17 | 19 | 47.2% (36) | 35.2%On the 33 of 36 calls that have a base rate: 51.5% correct, base rate 35.2% | +16.4 pts |
Five rates in the table rest on fewer than 20 calls and are marked; on so few, one call either way moves a rate by more than 5.0 points.
The base rates for the economy are low because they count how often a change of that size happened in the series’ own history, and for the Fed’s rate a move of at least 25 basis points in one direction within a year is the less common outcome. They do not reflect what markets expected on the day of the call, which the method does not measure, so a call on a Fed move that markets already expected starts from a base rate that understates how likely the move was.
What happened to their recession calls?
No channel has a Correct one-year verdict on a call that a recession would come. A recession call is judged on the months the National Bureau of Economic Research dates as recession. When none falls inside the window, the verdict waits 15 months after the window closes, because the dating comes late. At one year, each channel’s own calls on a recession stand as follows:
- Graham Stephan: 4 counted calls said a recession would come, and at one year they stand at 0 Correct, 3 Incorrect and 1 still open; 6 more are held out because their direction is in doubt or under review.
- Andrei Jikh: 8 counted calls said a recession would come, and at one year they stand at 0 Correct, 6 Incorrect and 2 still open; 15 more are held out because their direction is in doubt or under review.
- Joseph Carlson: 3 counted calls said a recession would come, and at one year they stand at 0 Correct and 3 Incorrect; 6 more are held out because their direction is in doubt or under review.
- Coin Bureau: 9 counted calls said a recession would come, and at one year they stand at 0 Correct, 8 Incorrect and 1 that could not be checked; 1 said none would come and stands at 0 Correct, 0 Incorrect and 1 still open; 2 more counted calls are judged on an economic series and stand at 0 Correct, 1 Incorrect and 1 still open; 13 more are held out because their direction is in doubt or under review.
Every recession call from every channel is on the recession page, each held one with its reason on the card, and the economy section has the other economic subjects. The article on who predicted the recession reads the recession and crash calls one by one.
What do academic studies of finfluencers find?
Both studies below measure what acting on finfluencers’ recommendations would have returned, as a portfolio or a trading strategy. Callsight checks single statements against fixed windows. The two kinds of figure answer different questions, so this article does not set them side by side.
VideoConviction (Galarnyk and co-authors, 2025) is a dataset of finfluencers’ videos with more than 6,000 expert annotations. Its abstract reports that stock recommendations made with high conviction did better than those made with low conviction and still did worse than an S&P 500 index fund. Betting against the recommendations beat the S&P 500 by 6.8% in annual returns, with greater risk: a Sharpe ratio of 0.41 against 0.65.
Finfluencers (Kakhbod, Kazempour, Livdan and Schürhoff, Swiss Finance Institute working paper 23-30, 2023) uses tweet-level data from a social media platform. Its abstract reports 28% of the finfluencers skilled, their advice generating abnormal returns of 2.6% a month, 16% unskilled, and 56% with negative skill, generating −2.3% a month. Those with negative skill had more followers and more influence on retail trading than the skilled ones.
Should you trust finance YouTubers?
Whom to trust is for the reader to decide, and nothing on this page is investment advice. The data checks statements one at a time: each channel’s page lists every call with the words that were said, a link to the second and the verdict at each window, so a reader can weigh a channel’s record call by call.
The comparison pages set two channels’ calls side by side, and the signals page records when several channels with a measured record agree on one instrument and none calls it the other way. Four articles look at single channels: Graham Stephan’s calls on housing, Andrei Jikh’s Bitcoin predictions, Joseph Carlson’s stock calls and Coin Bureau’s price predictions. Another sets every channel’s Bitcoin calls side by side.
How was this measured?
Each call is quoted word for word from the video and linked to the second it was said. A price call is measured from the first closing price after the video went public, and an economic call on the published series it names, as first published. Every call is checked at six windows, from 30 days to 5 years, and the one-year window is the headline. A call that a price will rise or fall is Within band when the move is smaller than 0.5% either way, and a price target is Correct when the price reaches the level inside the window. A forecast read out from someone else is shown with its source and never counted. Once a window has closed, its verdict is frozen: it changes only under a new version of a rule, which keeps the old verdict on record, or after an upheld correction. The rules are the same for every channel, and the methodology page gives them in full. If you think a call was misread, the corrections page says how to report it.
How these examples were chosen
The calls in the first section were chosen on Oct 7, 2026 by one rule: for each channel, the most recent of its own price calls that has a benchmark and a one-year verdict of Correct or Incorrect. A call was passed over if its reading is being checked again, if its quoted words do not name the coin, stock or fund, if they describe the present or do not say which way the price will go, or if they set a span other than a year, such as the short term, the long term or a date. The rule gave three Correct and one Incorrect. The forecast read out from a bank is there to show how such a call appears, and it counts for no one. Every call is listed on its channel’s page.
Figures are computed by code from the same record as the rest of the site. This article was drafted with the help of AI tools and reviewed by a person before publication. It is not investment advice.