Buying US stocks at an all-time high, stock by stock: small stocks since 1998, with the failed companies counted

How this study happened

An earlier study of four US indexes found that only the small-cap one, the Russell 2000, did worse than usual in the year after an all-time high, though the gap could have been luck. On X, @mrs_btc_ pointed out that adding the companies that left the exchange "will gut small cap returns". An index stops counting a failing company once it drops out, so its final loss never shows up. So I ran it stock by stock: 8,496 US stocks that started trading in 1998 or later, including the ones that later went bust.

Before computing a single return I wrote down how I'd count and how I'd describe each possible result, so I couldn't go looking for the version that looks best afterwards. For the average, the comment held: the estimated final losses of the companies that went bust pulled small stocks' average down, more after an all-time high than after an ordinary day. The median small stock still did as well as usual.

Wojciech, 8 October 2026

The findings at a glance

+6.8% / +10.9%

Small US stocks, on average in the year after a new all-time high and after any day, the ones that went bust included. Luck alone gives a gap this big about 2 times in 100

+3.7% / +3.4%

The same for the median small stock: almost no difference

−4.1 points

The gap, in percentage points. It depends on how much a company that went bust lost after its last recorded price. I used the usual estimates from research on delisted stocks (30% on NYSE, 55% on Nasdaq; a takeover adds no loss). With no extra loss the gap is 2.2 points, within what luck gives; with a total loss, 6.7 points

8,496

US stocks that started trading in 1998 or later, including those that later went bust or were taken over

Claims and verdicts

Four claims and what the data says

Every claim from every study, in one list →

Claim tested	Finding	Verdict
Small US stocks have a weaker year after an all-time high, once the companies that later failed are counted	Up 6.8% on average a year after a record, against 10.9% after any day; with the record dates shifted at random, a gap this large came up in 2 of 100 tries. It depends on the loss assumed for a failed company: with none the gap is 2.2 points, within chance, with a total loss 6.7	On average, yes
The median small US stock has a weaker year after an all-time high	Median +3.7% a year after a record, +3.4% after any day; the stock was up a year later 55% of the time against 53%. Over 3 years it did lag: +7.1% against +12.8%	No: about the same
Large US stocks have a weaker year after an all-time high	11.4% on average a year after a record, against 12.0% after any day, within chance. Across all stocks from 2003 the gap is 2.1 points, on the edge of chance. Over 3 years they did lag on average: 34.5% against 38.6%	No clear gap
US companies not yet profitable lag most after an all-time high	1.9% on average a year after a record, against 7.8% after any day, the widest gap of the three profit groups; a gap this large came up by chance in 3 of 100 tries. The median one did better after a record: −3.8% against −9.8%	On average, yes (side result)

1 · Small stocks

Small US stocks did worse than usual on average in the year after an all-time high, with the failed companies counted.

Every day I ranked each listed stock by its market value that day: the 1,000 largest are large, the next 2,000 small (roughly what the Russell 2000 holds), the rest micro. Among small stocks there were 98,218 record closes with a full year after them, and 5.9 million ordinary stock-days to compare them with. A year after a record they were up 6.8% on average, with dividends; after any day, 10.9%.

Before seeing the result, I'd written down how I'd word a negative gap with a chance below 5 in 100: small caps still lagged after an all-time high with the failed companies counted. With the whole calendar of records shifted at random, a gap this large came up in 2 of 100 tries. It's still an average over many overlapping years, so it describes the past and says nothing about the next record.

These are companies that started trading in 1998 or later, so younger ones by construction. Across all stocks, with records from 2003, the gap is almost the same: −4.2 points, chance 3.1%.

Average 12-month total return after a record against any day: large stocks +11.4% against +12.0%, small +6.8% against +10.9%, micro −2.0% against +6.0%. Median: large +9.0% against +9.2%, small +3.7% against +3.4%, micro +3.3% against −6.7%. — WB Trading Labs · Fig. 112-month total return, stocks that started trading in 1998 or later, start dates from 31 December 1998 to 13 August 2025. Size by market value, ranked each day. Companies that later failed are included, with a final loss by reason. Data: Sharadar, my calculations.

Every size and profit group, 12 months after an all-time-high close

Group	Days at a record	Average after a record	Average after any day	Chance	Median after a record	Median after any day	Chance (median)
Large (the 1,000 largest)	76,724	+11.4%	+12.0%	17%	+9.0%	+9.2%	13%
Small (the next 2,000)	98,218	+6.8%	+10.9%	2.4%	+3.7%	+3.4%	91%
Micro (the rest)	53,754	−2.0%	+6.0%	4.1%	+3.3%	−6.7%	73%
Profitable	180,703	+7.6%	+10.6%	5.3%	+6.0%	+6.2%	8.2%
Not yet profitable	39,164	+1.9%	+7.8%	3.2%	−3.8%	−9.8%	93%
No recent filing	13,465	+4.9%	+6.5%	17%	+5.5%	−1.7%	70%

WB Trading Labs · Table 1"Days at a record": with a full year after them. "Chance": how often a randomly shifted calendar of records gives a gap at least as large, the same way round (see the method). "No recent filing": no report in the last 15 months.

2 · The median small stock

The median small US stock did as well after an all-time high as after any day.

The median, the middle result, tells a different story from the average. A year after a record the median small stock was up 3.7%, after any day 3.4%, and small stocks ended the year higher 55% of the time, against 53%. That difference is well within chance: it came up in 91 of 100 shifted calendars.

So the gap comes from the tails, not the middle. After any day the average (10.9%) is far above the median (3.4%), because a minority of very large winners pulls it up; after a record the two are closer (6.8% and 3.7%). On top of that, the final losses of failed companies weigh more on the year after a record and make up about half the gap (section 3).

Counting only the first record after at least three months (63 sessions) without one, as in the earlier index study, leaves 6,379 such highs. Their median year was +4.6%, above the +3.4% after any day; random draws of the same size from ordinary days came that high 3 times in 100. Their average, 9.3%, was still below 10.9%.

Over three years the picture changes: there even the median small stock lagged, +7.1% after a record against +12.8% after any day. And in the wider check, every stock from 2003 with highs counted from 1998, the median small stock was 1.2 points behind after a year, within chance.

Small US stocks after an all-time-high close and after any day, by horizon

Horizon	Average after a record	Average after any day	Chance	Median after a record	Median after any day	Chance (median)
1 month	+0.4%	+0.8%	3.2%	+0.4%	+0.4%	3.8%
3 months	+1.6%	+2.4%	6.3%	+1.3%	+1.1%	90%
6 months	+3.7%	+5.1%	8.4%	+2.3%	+1.9%	87%
12 months	+6.8%	+10.9%	2.4%	+3.7%	+3.4%	91%
3 years	+20.3%	+32.7%	below 0.1%	+7.1%	+12.8%	below 0.1%

WB Trading Labs · Table 2Total returns, stocks that started trading in 1998 or later. The horizons overlap, so these are not five separate results. The data ends on 14 August 2026.

3 · The failed companies

For small US stocks, the gap depends on what a failed company is assumed to lose.

The price data stops at a company's last trade, so the final loss of a company that goes bust is missing. I chose the rule before seeing any results. A company delisted after a bankruptcy, a regulator's decision or by its own choice loses another 30% if it traded on the NYSE or NYSE American and 55% if on Nasdaq, the typical delisting losses in academic studies (Shumway 1997; Shumway and Warther 1999). A company taken over or merged loses nothing more. After delisting, the money earns the S&P 500's total return to the end of the 12 months. The same rule applies after any day, so failures lower both sides.

With no final loss at all, the small-stock gap shrinks to 2.2 points, and a gap that size came up in 8 of 100 shifted calendars: within chance. With a total loss it widens to 6.7. Under the rule, the final losses lowered the average after a record by 2.3 points and after any day by 0.4: the failed companies weigh more on the year after a record.

The other variations, all listed on 8 October, give −4.1 to −4.4 points, each with a chance of 1.4% to 3.1%. Counting each stock once instead of each day makes the gap larger, 7.7 to 16.3 points depending on which stocks the comparison uses. A stock with a single record right before a collapse then counts as much as one with hundreds, so I don't read that as a sharper result.

Average 12-month total return for small stocks after a record against any day: with a final loss of 0%, +9.1% against +11.2%; under the rule, +6.8% against +10.9%; with a final loss of −100%, +3.6% against +10.4%. — WB Trading Labs · Fig. 2Average 12-month total return, small stocks (ranked 1,001 to 3,000 by market value), that started trading in 1998 or later. Companies taken over or merged lose nothing more in every version.

Small US stocks, 12 months: the answer under each variation I'd listed beforehand

Variant	Average after a record	Average after any day	Gap, points	Chance
The main rule	+6.8%	+10.9%	−4.1	2.4%
Final loss 0%	+9.1%	+11.2%	−2.2	7.6%
Final loss −100%	+3.6%	+10.4%	−6.7	below 0.1%
Delisted with no stated reason: 0%	+6.8%	+10.9%	−4.1	2.4%
Cash after delisting, not the S&P 500	+5.9%	+10.4%	−4.4	1.4%
Over-the-counter prices after delisting kept	+6.8%	+10.9%	−4.1	2.4%
No $1 price floor	+6.8%	+11.0%	−4.2	2.5%
A high since 1998, all stocks, from 2003	+8.5%	+12.7%	−4.2	3.1%
Each stock counted once, against all small stocks	−8.8%	−1.1%	−7.7	not shown
Each stock counted once, against the same stocks on any day	−8.8%	+7.6%	−16.3	not shown

WB Trading Labs · Table 3"Each stock counted once": each stock's own average first, then the average across stocks; short-lived stocks then weigh as much as long-lived ones, so the levels fall. Of these two rows, the upper one compares the stocks that had a record with all small stocks, the lower one with the same stocks on any day; a stock that never reaches a record is often one that only fell, so the stocks that do reach one have better days on average. No chance figure: the shifted-calendar test fits only the lower row, and with each stock counted once it would mostly measure stocks with only a few records. "A high since 1998": all stocks, records from 2003, highs counted only from 1998.

4 · Large and micro stocks

Large US stocks showed no clear gap in the year after an all-time high; micro stocks did, but only on average.

A year after a record, large stocks were up 11.4% on average against 12.0% after any day, and the median one 9.0% against 9.2%; a gap at least that large came up by chance in 17 of 100 tries. That matches the earlier index study, where the S&P 500, the Nasdaq-100 and the Russell 1000 showed no gap. Over three years they did lag on average, 34.5% against 38.6%, and in the wider check from 2003 the one-year gap is 2.1 points, on the edge of chance (5.0%).

Micro stocks, everything below the largest 3,000, show the widest average gap: −2.0% a year after a record against +6.0% after any day. But the median micro stock did better after a record, +3.3% against −6.7%, so here too the average and the middle point opposite ways. I report micro stocks as a description, not a tested claim.

5 · Profit

US companies not yet profitable lagged most in the year after an all-time high, on average.

I split stocks by their net income over the last 12 months, from the latest report filed before the day. Companies not yet profitable were up 1.9% on average a year after a record, against 7.8% after any day: a 5.9-point gap that came up by chance in 3 of 100 tries. Profitable companies lagged by 3.0 points, on the edge of chance (5.3%). Among small stocks only, the gap for companies not yet profitable is 6.4 points.

Here too the median points the other way: for companies not yet profitable it was −3.8% in the year after a record and −9.8% after any day. Measured by operating income instead of net income, the gap for companies not yet profitable widens to 14.1 points and the one for profitable companies disappears (0.0).

Method and caveats

What this study measures, and what it doesn't

Data: daily prices and financial reports from Sharadar (Nasdaq Data Link) for 15,701 US stocks (common shares listed on the NYSE, NYSE American and Nasdaq), from 31 December 1997 to 14 August 2026, including the ones later delisted. Returns include dividends; the S&P 500 total return from Yahoo Finance. Prices and corporate actions from the extract of 17 August 2026, the list of stocks and the reports from the September 2026 refresh. I publish aggregates only.
Record: a close (adjusted for splits, not dividends) at or above every earlier close of the same stock. A stock's first 252 sessions (about a year) never count, as a record or as any day, and a day counts only if the price is at least $1.
Why from 1998: my data starts at the end of 1997, so for a stock listed earlier I can't see its highs before 1998. So the test uses the 8,496 stocks first priced on or after 5 January 1998: younger companies by construction. A wider check takes every stock, counts records from 2003 and calls them "a high since 1998"; its small-stock gap is −4.2 points, chance 3.1%.
Failed companies: the rule in section 3. Under it, among the delisted stocks 7,097 were taken over or merged, 3,554 went bust or were removed by a regulator, 157 left by choice, and 531 have no stated reason within 45 days of their last price; those count as failures, the cautious side.
Size and profit: only what was known on the day: market value from that day's price and the share count in the latest report, profit from the latest report filed before it. Never today's values.
"Chance": records come in clusters, and many stocks hit records in the same weeks. So I shift the whole calendar of records by the same random number of sessions for every stock, longer than the horizon, 4,000 times, and count how often that gives a gap at least as large, the same way round. The same for medians. For first records only, 20,000 random draws of the same size from ordinary days. It's a rough gauge, not a formal test.
Many comparisons: 3 sizes, 3 profit groups and 5 horizons make dozens of comparisons, so some would look unusual by luck alone. Before seeing any results, I chose just one to carry the answer: small stocks, 12 months, stocks that started trading in 1998 or later, the delisting rule. Everything else on this page is a description.
Weighting: each stock-day counts once, so stocks that sit at records for long stretches weigh more. Counting each stock once widens the small-stock gap to between 7.7 and 16.3 points, depending on the comparison (Table 3).
Rules written first: on 8 October, before computing any return, I wrote down the method, the delisting rule and the wording for each possible answer, and committed them to my private research repository (commit b41de2fc0). Where the data forced a change, it's listed below.
Change 1, market value: the share counts in the reports are restated for every later split, so the formula as written would have overstated, for example, Apple's 1998 market value about 112-fold. I used the shares outstanding on the day instead: the reported count divided by the splits after the day. On 60,000 sampled reports that matches Sharadar's own market value 99.4% of the time, the written formula 75.8%.
Change 2, prices after delisting: 425 delisted stocks kept trading over the counter, some for years. I end each one on the day it left the exchange, as the study covers exchange listings only; keeping those prices gives −4.1 points (Table 3).
Change 3, counts: the coverage table in the written method mixed two versions of the data: 531 delisted stocks have no stated reason, not 145. They count as failures, as the rule says; with no final loss for them the gap stays at −4.1 points.
Smaller choices the written method left open (the order of delisting reasons, how sessions are counted, 37 stocks with a new ticker between data versions, 39 stocks whose prices stop early without a delisting, which tests run on medians) are recorded with the data.
Not tested: markets outside the US, or why a stock was at a record. The averages describe 1998 to 2026 and say nothing about what follows the next record.

The short version goes up on X, in the thread where I promised it.

Follow @WojciechBernPL

The same question on four US indexes All studies

In the year after hitting their own all-time high, small US stocks lagged on average - the median stock didn't