There is a version of this question that flatters everyone who asks it, and a version that answers it. The flattering version picks a handful of famous listings, notices that a couple of them went up a lot, and concludes that you should have bought those. The problem is that the handful is chosen after the fact, and every name in it is still trading, which quietly deletes the ones that died.
So we did it mechanically instead.
How the 50 were chosen
The rule, fixed before any price was pulled:
The 50 largest US listings whose first trading day fell on or after 1 January 2020, ranked by offer size at listing, where offer size is the shares on the cover of the listing prospectus multiplied by the offer price. Ordinary equity, US primary listing. No blank-check shells, no closed-end funds.
The universe came from the Nasdaq IPO calendar for all 81 months from January 2020 to September 2026: every deal marked priced, 2,752 of them, deduplicated and sorted by deal size. We did not choose the names. The sort did.
For the top 85 candidates we then opened each company's own 424B4 prospectus on SEC EDGAR and read the offer price off the cover. Seventy-three of the seventy-four machine-readable prices matched the feed to the cent.
Four things were excluded, each for a stated reason rather than a preference. SK hynix, which at $26.5bn would have ranked third, because its prospectus names the Korean exchange as its principal trading market. Pershing Square, because its own cover shows an offer price of $0 and it is paired with a closed-end investment company. Hertz, because its 2021 deal was selling stockholders offloading a stock already quoted over the counter, not a first listing. And thirteen SPACs, which matter more than they sound: the feed lists a blank-check deal under the name of whatever company later merged into it, so Lucid, Paysafe, Ginkgo, Joby and SoFi all look like conventional IPOs until you check.
How it was measured
Every name is measured from its own first trading day close, not from the offer price. The offer price is what institutional allocations paid. The first-day close is roughly what a reader could actually get.
The benchmark is $1 into SPY on each of those same 50 dates, held to the same 50 end dates. A single index window would have made the answer depend on when you started counting, so the comparison is date for date. Prices are split-adjusted. Dividends are excluded on both sides, which if anything flatters the IPOs.
What the basket actually did
| Measure | Value |
|---|---|
| Equal-weight basket of all 50 | +34.4% |
| Same-dated SPY | +71.0% |
| Gap | 36.6 points behind |
| Median name | -27.5% |
| Beat its own-window SPY | 8 of 50 |
| Above its first-day close at all | 16 of 50 |
| Down 50% or more | 18 of 50 |
The mean is +34.4% and the median is -27.5%. That 62-point gap between the two is the entire subject of this piece. The average is being carried; the typical name lost about a quarter of its value.
The weight on one name
Put $50 into this basket, $1 per name, and you finish with $67.22. The profit is $17.22.
Palantir alone turned its $1 into $19.53. Its profit, by itself, was $18.53.
Which means the other 49 names, added together, lost $1.31. One listing out of fifty produced more than the entire gain, and the remaining 98% of the portfolio was a net drag on it.
Stated as shares of the basket's total profit:
| Names | Share of the basket's entire profit | |
|---|---|---|
| Top 1 | PLTR | 107.6% |
| Top 2 | PLTR, APP | 129.1% |
| Top 3 | PLTR, APP, ARM | 150.5% |
| Top 5 | HOOD, VIK | 175.7% |
| Top 10 | CRWV, EBC, CRBG, PPD, RPRX | 194.9% |
Those shares exceed 100% because everything else nets out negative. That is not a rounding artifact or a presentational trick. It is the finding.
Take the winners out
| Basket | Return | Same-dated SPY |
|---|---|---|
| All 50 | +34.4% | +71.0% |
| Without the top 1 | -2.7% | +69.9% |
| Without the top 2 | -10.4% | +69.6% |
| Without the top 3 | -18.5% | +69.6% |
| Without the top 5 | -29.0% | +69.9% |
| Without the top 10 | -40.9% | +67.7% |
Remove Palantir and a +34% basket becomes a -3% basket while the index does +70%. Remove the top five and it is -29%.
Note what this table does not show: a point where the basket starts beating the index. It never stops beating it, because it never starts. Even with every winner included, the group loses by 37 points.
So would concentrating in the few have worked?
Yes. Enormously. After the fact.
| Portfolio | Return | Same-dated SPY |
|---|---|---|
| Best 1 only | +1852.6% | +127.7% |
| Best 3 only | +863.8% | +93.5% |
| Best 5 only | +604.9% | +81.2% |
And the identical bet placed on the wrong names:
| Portfolio | Return | Same-dated SPY |
|---|---|---|
| Worst 1 only | -97.6% | +133.5% |
| Worst 3 only | -95.8% | +110.0% |
| Worst 5 only | -92.7% | +93.5% |
The part that decides it
"Buy the few that carry it" is only a strategy if the few can be named in advance. So we measured what picking from this group actually paid, with no foresight.
| If you had picked | Share that beat the same-dated SPY |
|---|---|
| 1 name at random | 16% (8 of 50) |
| 2 names at random | 18.8% |
| 3 names at random | 19.3% |
| 5 names at random | 17.5% |
| 10 names at random | 21.4% |
| 20 names at random | about 32% |
The one and two and three name rows are exact, every combination enumerated. The rest are 200,000 random draws each.
Sixteen percent of single picks beat the index. Eighty-four percent did not, and the median pick lost 27.5% while its own-window SPY returned +76.8%.
Spreading the bet inside the group does not rescue it either. Even a random twenty-name portfolio, which is most of the cohort, beat the index only a third of the time. The problem here is not variance that diversification can smooth away. It is a negative median.
The same concentration that produced Palantir's +1853% produced Lufax's -97.6%, and there was nothing observable on either listing day that separated them.
What this does not say
It does not say IPOs are bad, and it does not say the concentration is an illusion. Both halves are real: the weight on one or two names is as extreme as anyone claims, and the strategy that follows from noticing it had roughly a one in six hit rate.
We ran it two other ways to see whether any of it depended on a judgement call. Restricting to conventionally underwritten deals only, which drops the six direct listings including Coinbase, Roblox and Palantir itself, the basket returns -0.4% against SPY's +70.8%: the conclusion gets stronger, not weaker. Requiring at least a year of trading, which drops the newest listings, leaves the basket at +48.3% against +87.5%, with Palantir still accounting for 96% of the total.
Two limits worth stating. The enumeration comes from a single calendar source, so a large deal that source missed would be missed here too, though it would have to exceed about $1.48bn to change the membership. And both price vendors silently drop the first trading session for eight of the fifty names, so those are measured from session two; a sensitivity test moving them 15% in either direction leaves the top-1 share between 103% and 112%.
One correction we had to make to our own data along the way, since it would catch anyone repeating this: asking Yahoo for a maximum range at daily interval silently returns weekly bars. That put Airbnb's first-day close at $139.25, the Friday, instead of the correct $144.71.
Educational commentary, not investment advice. Marks as at 29 September 2026.