Should you buy the stock a channel says is the next one? The full report

The question this report answers is the viewer’s own: I missed Nvidia. Should I buy the stock a channel says is the next one? No one can audit every pick, so we tested the class. At seven public hype moments we bought every company we could identify in the hot stock’s industry, listed that day, the ones that later died included and the hot stock itself left out, and held each basket three years against the S&P 500 (the SPY fund, dividends reinvested), before trading costs.

Where the “If you missed X, this is bigger” formula was made (we tested the class, not these picks):

• Fin Tek, If You Missed Nvidia, This is Even Bigger
• Ticker Symbol: YOU, If You Missed NVIDIA Stock, This is Way Bigger.
• BWB - Business With Brian, If You Missed Palantir at $25... It's Happening Again
• ZipTrader, If You Missed Palantir, THIS IS BIGGER.
• Ticker Symbol: YOU, NVIDIA Made Me Rich. I'm Buying This Stock Next.

We tested no pick these videos make as a pick, and nothing here says whether any stock in them will succeed or fail. Of 19 earlier pitch videos in this formula from six channels, 1 carried a ticker in its title or description, and that one was a false match, so a pick-by-pick check would have rested on about 15 transcripts, too few to carry a verdict.

What we grant, and what we tested

Granted, not measured: some stocks in these industries became giant winners. In our baskets: AAOI (+10,270.9% from the day after Nvidia’s breakout), Credo (+1,174.9% from ChatGPT’s launch), SoundHound (+841.4%, packaged software from the same day) and Enphase (+7,360.4% from 2017, a solar company filed among the chip makers). The hot stocks at ChatGPT’s launch went on to return +946.9% (Nvidia) and +2,146.0% (Palantir) over three years, after falling 48.2% and 63.7% in the year before. Nothing in our method could have singled any of them out in advance; picking them is hindsight. A skilled stock-picker inside a theme may beat it; we did not test skill.

Tested: the class, mechanically: every identified listed member of the hot stock’s industry on the moment’s date, identity checked against the company’s own SEC filings as of that date, dead ones kept at their last exchange price, the hot stock itself excluded because the pitch is about the next one.

The locked plan

We wrote the plan down and committed it (commit ecec8d39, 2026-10-06 08:02:06 UTC) before any return of these baskets or of the S&P 500 was computed; the returns engine first ran at 08:02:11 UTC. An independent blind review read the draft plan and every lab script before the lock, computed no return, and returned 14 findings, each fixed or answered before the lock. Two dated addenda follow the lock: a serialization fix after the first run (the rerun printed 14 of 14 moment lines identical) and a correction of descriptive prose (the cannabis basket’s makeup). The plan’s rules are unchanged since the lock. These are repository timestamps, not an independent external attestation.

The moments, as registered

A basket is the hot stock’s whole industry as the SEC codes it, not the narrower theme a pitch names: the cannabis basket held 3 cannabis growers, 5 supplement makers and 2 drug developers; the biotech basket is mostly biotech that never touched a vaccine. The second rule kept 5 of 13 candidate days; the others failed its filters before any of their returns was computed (listed in the design download).

Our written priors, before any basket return

We knew roughly how these eras went, so we wrote our guesses down first. Every confident guess matched, so those results are not discoveries. We were unsure only about the chip basket at ChatGPT’s launch and about both lines at Facebook’s IPO. The second rule carries no per-moment guess.

Deliberately NOT tested

1. Any stock a channel pitched, as a pick. Whether any channel’s picks worked.
2. Over-the-counter or foreign-only listings (SK hynix had no US listing at any of our moments).
3. Themes whose industry code did not name the business as of their own moments (solar, crypto) or whose companies share no code (3D printing).
4. Moments after 2023-06-22: three years of prices do not exist yet.
5. The narrower themes a pitch names (“physical AI”), and stock-picking skill.

The registered result: seven named moments

All returns below are three-year total returns with dividends reinvested, from the close on the moment date, before trading costs (the costed variant is in the sensitivity table under Every moment, and every purchase). A gap is the basket’s return minus the S&P 500’s, in percentage points.

The one win is a data error we caught. The registered rule counts one basket as beating the index: internet and data-processing companies at Facebook’s IPO. Its best member, Autobytel, shows a one-day +413.1% jump on 2012-07-12, the day the company filed an 8-K with items 3.03 and 5.03, and our data holds no split record for it. We infer an unadjusted reverse split; the inference is ours. The registered rule keeps data-check-flagged members, so it reads 1 of 7. With the flagged member removed (the registered data-check variant) that basket trails the S&P 500 by 39.0 points, and the count is 0 of 7. Treating the jump instead as a share change, a check added after the results and not pre-registered, Autobytel returns +239.0% and the basket trails by 27.7 points; it would beat the index only if Autobytel’s true move that day had been +147.1% or more. That basket’s own figures stay withheld on this page and in the downloads.

The typical stock, the one in the middle from worst to best, beat the index in 0 of 7. Across the seven baskets, 114 of 589 stocks (19.4%) beat the S&P 500 over three years, and the median stock trailed it by 78.3 points. Over one year (report-only), the whole basket beat the index in 1 of 7, the same moment and the same data error, and the typical stock in 0 of 7.

After the run. In the 3 moments bought after the hot stock had already more than doubled in the year before (Tesla’s S&P 500 entry, the vaccine news, Zoom’s quarter), all three baskets and all three typical stocks trailed, and so did the three hot stocks: Tesla +17.5% against the index’s +34.4%; BioNTech -4.0% against the index’s +28.0%; Zoom -84.3% against the index’s +33.9%. That is n = 3. Across every after-the-run moment in the lab (those three, the two side checks, and the second rule’s 2017 chip makers and 2020 car makers), the whole basket beat the index in 3 of 7 and the typical stock in 0 of 7 (a pooled count chosen after the results; report-only).

Listings that ended within three years: biotech 23 of 141 (13 collapsed, exiting at a fifth or less of the entry price), packaged software 60 of 251 (6 collapsed); our data does not record why the rest ended. The three cannabis growers alone trailed the index by 139.5 points (n = 3); Moderna and Novavax alone by 79.6 points (n = 2).

Why we call it fragile: a second rule picks the moments

We chose the seven moments ourselves. So a second, simple rule picked more: the first day in our data (from 2004) a stock became the most-traded US stock (by 20-day dollar volume), with the rest of its industry bought the same way. Of 13 such days, 5 survived the plan’s filters (no catch-all codes, no industry repeated within three years, at least 10 members).

The whole basket beat the index in 4 of 5; the typical stock in 1 of 5 (the 2016 catalog retailers). All four basket wins fall below the index once their best few members are removed: the 2017 chip makers without Enphase (+7,360.4%) returned +34.2% against +38.9%; the 2020 car makers without NIO +29.2% against +33.5%; the 2016 catalog retailers without their best two +49.7% against +50.9%; the 2008 refiners without their best three +50.8% against +52.6%. Those removal counts were measured after the results and are not pre-registered.

This rule finds attention, not only hype: the 2008-2009 bank volume was crisis volume, and its early dates carry the most companies we could not identify. With those unidentified companies counted at -100%, its baskets beat the index in 1 of 5 (only the 2017 chip makers). The disagreement is why we call the answer fragile; it is not a reversal.

The closest test to the pitch: the day after Nvidia’s breakout

Fixed in advance as a side check, so it changes neither the count nor the answer: the day after Nvidia’s May 2023 earnings report (2023-05-25) we bought every company we could identify under Nvidia’s industry code (chip makers, solar-cell makers included, as at ChatGPT’s launch) except Nvidia, 97 of them. Over three years the basket returned +290.5% against +86.9% for the index, and +107.8% without its five best stocks. Yet the typical chip stock returned +55.3%, 31.6 points behind, and 54 of 97 trailed. Among the 77 that traded at least $1 million that day, the typical stock trailed by only 2.3 points (for the whole basket, by 9.3 points with unidentified companies counted at its 90th-percentile member, and by 9.0 points bought 63 trading days later).

Its two best stocks, AAOI (+10,270.9%, about 104 times the money) and AXTI (+4,116.5%, about 42 times), are winners nothing in our method could have singled out in advance. Both price paths were checked: full-year session counts, no split records, no data-check flag.

Every check we ran on the answer

The label stayed AGAINST under every fragility family (unidentified companies at -100% and at the 90th percentile, flagged members removed, an extra -30% on collapsed exits, buying 63 trading days earlier and later) and with any one of the seven moments dropped (7 of 7 leave-one-out labels AGAINST).

Timing can move the whole basket. Bought 63 trading days after ChatGPT’s launch (2023-03-03), the chip basket beat the index by 19.7 points while its typical stock still trailed (by 67.9). Bought 63 trading days before Tesla’s S&P 500 entry (2020-09-22), the car makers’ typical stock beat the index by 1.4 points. In our seven moments, timing moved the typical stock ahead once, narrowly.

Weighting can move it too. Weighted by trading activity instead of equally, the chip and software baskets at ChatGPT’s launch beat the index by 64.1 and 11.9 points, so the whole-basket count would read 2 of 7; these are the two moments before their hot stocks ran. Weighting cannot move the typical stock. The hot stock put back: the launch-day chip basket with Nvidia still trails by 17.5 points, the software basket with Palantir by 29.5.

Limits, in both directions

• Industry codes are blunt: the chip code holds solar-cell makers beside AI chips, and the software codes hold every kind of software and internet firm.
• Companies we could not identify (their listing ended and their ticker was later reused, or they left no usable filing) are left out, which flatters the baskets: 463 rows on ChatGPT’s launch day, 1,177 on Facebook’s first day. Those counts include warrants, units and funds, so they overstate the missing companies. We reran every basket with the expected number of them wiped out, and with each beating nine in ten of its industry; the label held both ways.
• Exits at the last exchange price flatter the baskets too; cash at 0% after an exit and spin-offs booked as drops cut the other way.
• US exchange listings only. Over-the-counter and foreign-only stocks are outside.
• The moments overlap in time (three fall within 16 weeks of each other in 2020; two share ChatGPT’s launch day). Seven overlapping moments are a count, not a law, and no significance test is claimed.
• The typical stock is not the stock a channel picks. A pitch is a selected stock; the typical member stands in for a pick made with no skill, nothing more.
• Because a basket holds the whole industry, a stock pitched today can sit in an older basket as one member. Its row describes that past window only and is not a test of any pitch.
• Inside the 2022 and 2023 chip baskets, a one-day Wolfspeed move on 2025-09-29 is flagged by our data check, cause not established. It stays in the registered result; with it removed, both chip baskets keep the same sign.
• The S&P 500 over these windows includes strong bull runs. Beating it is the viewer’s real alternative, not an easy bar.

Every moment, and every purchase

The ledger below has one row per moment: the seven named moments, the two side checks and the five rule-picked moments. The purchase download has every stock we bought, 938 rows in all (589 of them at the seven named moments), with its identity, how it was held and its returns. The data-error member’s returns, and the one basket’s figures that rest on it, are withheld; every other value is as computed.

Ledger columns: row; id; arm (named moment, side check or rule-picked moment); moment_date; moment; industry_code and industry (what the basket really is); hot_stock_left_out; members; basket_1y_pct and basket_3y_pct (the whole basket, equal weight); typical_stock_3y_pct (the median member); sp500_3y_pct; basket_gap_pts and typical_gap_pts (minus the S&P 500, in points); members_beat_sp500; listings_ended_collapsed and listings_ended_other; basket_beat_registered and typical_beat_registered (the two counts); basket_gap_flagged_removed_pts (members flagged by the data check removed); basket_without_best_1_pct and basket_without_best_5_pct; hot_stock_3y_pct and hot_stock_prior_year_pct (its 252 sessions before the moment); after_the_run; note. Percent values are returns times 100. Blank means withheld or not applicable, never zero; the note says which.

Purchase columns: row, id, moment_date, ticker (on the moment date), sec_cik (the SEC registrant number), identity (own filings, linked to a renamed ticker, or successor issuer), held_how, handoffs (successor ticker, kind, first date), listing_ended, collapsed, return_1y_pct, return_3y_pct, sp500_3y_pct, beat_sp500_3y, traded_1m_that_day (at least $1 million traded on the moment day), data_check_flag_dates, note. Sensitivity columns: row, id, variant, what_changes, members, entry_date, basket_gap_pts, typical_gap_pts, basket_beat, typical_beat, changes_basket_answer, changes_typical_answer, note.

What would change our mind

1. A re-run with better company identifiers (the research-grade kind that follow a company through every rename) that moves either count enough to change the label.
2. Today’s “next Nvidia” moments, once three years of prices exist, registered in advance under this same rule.
3. A list of hype moments chosen by someone else (for example dated theme-fund launches, or a search-interest peak series) run under this plan: if it reads SUPPORT, the claim deserves a second look, and we would say so.
4. A corrected split table for our data (the 2012 Autobytel jump and the 2025 Wolfspeed jump), rerun with both results reported.

Data, provenance and reproduction

Prices: Massive.com (formerly Polygon.io), Stocks Advanced, a survivorship-free daily store of every US stock and fund that traded each day, pulled 2026-06-22 and split-adjusted as of that pull; dividends from the same vendor, scaled for later splits and reinvested. Company identity and industry codes: the SEC’s bulk submissions file (sha256 066aff2f3ec2…) and 34,438 filing headers read from sec.gov. EODHD supplied no data to this experiment. Affiliate links support tools, not a result; we never take money from anyone whose claims we test.

For this report we re-aggregated the stored member returns before writing a byte: 224 of 224 stored moment figures reproduced across 14 of 14 moments, after a planted change was rejected first. The registered counts, the rule-picked counts, the pooled lines, the side check’s liquid line, the Autobytel correction and the four rule-picked thresholds reproduced too. This is re-aggregation, not a fresh price run; the design download lists the source file hashes.