AI-generated research, checked but not peer-reviewed. It may contain errors.

E802　 KILL　 No signalTested · October 2026

Buying a token after its treasury company's SEC filing does not pay

Buying a token 15 minutes after its treasury company's SEC 8-K earned +6.6 bp over 24 hours, the same as a week earlier (+6.1 bp), and -25.5 bp once the alt-market move is removed.

digital-asset-treasury sec-filings 8-k perpetual-futures alt-tokens event-study

Readable summary Full research record JSON recordFeedback on E802

Listed companies now raise money from ordinary investors to buy a single crypto token and keep it on their balance sheet. The belief is that their SEC filings show buying that has to happen and that the token then rises. We read 592 candidate filings from 21 such companies, kept 118 for 15 alternative-token companies, and bought the token's perpetual future 15 minutes after each filing for 24 hours. After a 10 bp cost the average was +6.6 bp, with an interval from -85 to +89 bp. The same trade one week earlier made +6.1 bp.

Filing events11815 companies, 11 tokens, 2025-04 to 2026-10

Net per trade+6.6 bp95% company-clustered interval -85 to +89 bp; required +30 bp

Same hour, 7 days earlier+6.1 bpthe filing adds +0.6 bp, paired per event

Market-adjusted-25.5 bpafter subtracting the average move of the other tokens of a 15-token list

01 · The idea being tested

When a company that holds a crypto token for its treasury files with the SEC about a new raise or new holdings, that token rises as the money is deployed.

Why people believe it

These companies sell shares or notes to traditional investors and are expected to turn the proceeds into the token. The filing is public and arrives as plain text on the filing system. Most crypto derivatives traders do not read it, so the buying should reach the token price after the filing, not before.

Why it would persist The argument, as stated before testing

Parsing hundreds of filings takes work, the tokens involved are thinner than the largest coins, and people argue the purchases are spread over days, so the price response would be slow enough to remain available.

02 · How it was tested

We fixed a list of 16 alternative-token treasury companies and a set of text rules before reading any filing or price. Every 8-K in the search window (from January 2025; the first qualifying event was in April 2025) that mentioned the company's token and read like a closed financing or a holdings update became an event. We bought the token's perpetual future 15 minutes after the filing was accepted, sold 24 hours later, and charged 10 bp for the round trip. Each trade was compared with the same token at the same clock time one week earlier, and with the average move of the other tokens of a 15-token list.

Pass required

At least 50 events from at least 12 companies; mean net at least +30 bp; company-clustered 95% interval above 0; median above 0; control below half the mean; if at least 30 raise events have a parsed dose, the top dose third above the bottom third. The market-adjusted mean also had to be above 0.

Killed if

Mean below +10 bp, or the interval includes 0.

Costs

10 bp round trip; funding ignored (stated).

Controls

Same token, same clock time, 7 days earlier (decides the verdict). Alt-adjusted: subtract the equal-weight 24-hour return of the other tokens of a 15-token list (5-minute prices), excluding the traded token. Secondary: Ether treasuries.

03 · What happened

Why it was rejected

The mean (+6.6 bp) is below the +10 bp kill line and the company-clustered interval (-85 to +89 bp) includes 0; the control earned +6.1 bp.

Across 118 filing events (one per company-day, 15 companies) the long earned +6.6 bp net over 24 hours; the median trade lost 6.4 bp, 48.3% of trades made money, and the worst lost 1,147 bp (the last event in the sample).
The same trade one week earlier earned +6.1 bp. Paired per event, the filing adds +0.6 bp (bootstrap by company, 4,000 draws, interval -126 to +98 bp).
Market-adjusted, subtracting the average 24-hour move of the other tokens of a 15-token list (14 when the traded token is on it), the mean is -25.5 bp (interval -83 to +22 bp): the raw average sat inside normal alt-market movement.
The 122 Ether-treasury filings (secondary) averaged +0.9 bp.
The planned test of larger raises against smaller ones could not be run: proceeds were parsed from the text for too few events. A hand audit of 10 random classified filings found about 7 sensible; misfires were risk-factor text and an in-kind exchange of a locked token.

The filing adds nothing to a normal day

Mean 24-hour net return of the long with 95% company-clustered intervals. Event and control sit at +6.6 and +6.1 bp, their paired difference at +0.6 bp, and the market-adjusted event at -25.5 bp. Every interval includes 0 and the +30 bp pass line is out of reach.

Chart data (bp)

	Mean	95% low	95% high
After the filing (n 118)	6.6	-85	89
Same hour, 7 days earlier	6.1	-45	59
Filing minus control (paired)	0.6	-126	98
After the filing, market-adjusted	-25.5	-83	22

Wide, and the same with or without a filing

After the filing7 days earlier

Trades by 24-hour net return in bp, filing against the same hour a week earlier (118 each). Both are wide and centred near zero, with no visible shift to the right after a filing.

Chart data (number of trades)

Bin	After the filing (count)	7 days earlier (count)
< -800	2	1
-800..-400	11	16
-400..-200	19	16
-200..-100	11	13
-100..0	18	18
0..100	12	14
100..200	17	10
200..400	12	16
400..800	11	8
> 800	5	6

04 · What it means

Why it failed Interpretation · not established

The filing date may simply be too late: a raise may be announced by press release first, and the token may be bought before the filing is accepted, so the filing may record news the market already has (not measured here). 58% of events fall in the second half of 2025, one alt-market regime, where the market-adjusted figure is the more informative one and it is negative.

Lesson

A filing may be a late record of news already public (press-release timing was not measured here); an information event is best timed at the first public disclosure. A raw average over a single market regime mostly measures that regime. When the 24-hour spread across events (several hundred bp) dwarfs the effect sought, even 118 events can only rule out large effects.

Scope of this result

Covers 8-Ks of 15 alternative-token treasury companies (11 tokens) from April 2025 to October 2026, a 24-hour long on the perpetual future with entry 15 minutes after acceptance and a 10 bp cost, funding ignored. The text classifier was automatic and imperfect. Ether treasuries are reported as a secondary group. The result does not cover other filing forms, other tokens, or other entry delays.

A rejected experiment applies to the tested conditions and does not prove an idea can never work. bp = basis point, 0.01%.

1. Test design

Pre-registered

Yes: rule and criteria fixed before the test data was examined

Data

8-K filings (items 1.01, 3.02, 7.01, 8.01) with exhibits, and acceptance timestamps, for a frozen list of treasury companies — the public SEC EDGAR filing system; per filing; 2025-01..2026-10; public | Perpetual futures 1-minute and 5-minute prices — largest USD-margined crypto perpetual venue, public market data; 1 minute; 5 minutes for the alt index; 2025-01..2026-10; public

Universe

Primary: 16 companies on the frozen list, each holding one alternative token (11 distinct tokens); 15 produced events. Secondary, reported only: 5 companies holding Ether. Bitcoin treasuries and foreign filers of form 6-K were excluded. 592 candidate 8-Ks from 21 companies in total.

Signal

Candidate: an 8-K or 8-K/A accepted from 2025-01 to 2026-10 with item 1.01, 3.02, 7.01 or 8.01 whose main document or exhibit 99 contains the company's token symbol or name. Class "raise": the word closed, closing, consummated or completion within 300 characters of private placement, offering, PIPE, financing or subscription, and the word proceeds or cash anywhere in the text. Class "holdings": a number followed within 15 characters by the token symbol or name, with hold, holdings, acquired, purchased or treasury in the 200 characters before it. Both: raise. Automatic rules, no manual reclassification after prices were seen. One event per company per day (earliest acceptance). Event time is the acceptance timestamp. Classifier result: 106 raise, 138 holdings, 348 neither.

Execution

Long the token's perpetual at the first 1-minute open at or after acceptance + 15 minutes, exit at the first open at or after entry + 24 hours; simple return.

Costs

10 bp round trip; funding ignored (stated).

Controls

Sample split

Rule fixed before any filing text or price was read; results reported by quarter. No train/test split: one sample, one pass.

Pass criterion

Kill criterion

Mean below +10 bp, or the interval includes 0.

Power

Pre-registered minimum detectable effect about 145 bp at 60 events (24-hour market-adjusted standard deviation about 4.5%): only a large effect could be seen.

2. Result

Observations

118

Instruments

Net mean

+6.6 bp

95% interval

-85 bp to +89 bp

Median

-6.4 bp

Worst

-1,147 bp

Finding 1

Across 118 filing events (one per company-day, 15 companies) the long earned +6.6 bp net over 24 hours; the median trade lost 6.4 bp, 48.3% of trades made money, and the worst lost 1,147 bp (the last event in the sample).

Finding 2

The same trade one week earlier earned +6.1 bp. Paired per event, the filing adds +0.6 bp (bootstrap by company, 4,000 draws, interval -126 to +98 bp).

Finding 3

Market-adjusted, subtracting the average 24-hour move of the other tokens of a 15-token list (14 when the traded token is on it), the mean is -25.5 bp (interval -83 to +22 bp): the raw average sat inside normal alt-market movement.

Finding 4

The 122 Ether-treasury filings (secondary) averaged +0.9 bp.

Finding 5

The planned test of larger raises against smaller ones could not be run: proceeds were parsed from the text for too few events. A hand audit of 10 random classified filings found about 7 sensible; misfires were risk-factor text and an in-kind exchange of a locked token.

3. Interpretation

Why it failed

Lesson

Scope

4. Reproducibility

Steps

1) Fix a list of treasury companies and their EDGAR identifiers; download each company's filing index and every 8-K with items 1.01, 3.02, 7.01 or 8.01 accepted from 2025-01 to 2026-10, plus exhibit 99 text. 2) Apply the raise and holdings text rules above; keep one event per company per day, timed at acceptance. 3) Using 1-minute perpetual prices, go long at the first open 15 minutes after acceptance, exit 24 hours later, subtract 10 bp; repeat at the same clock 7 days earlier and subtract the equal-weight return of the other tokens of a 15-token list (5-minute prices). 4) Bootstrap by company; report by class and by quarter.

Grade

COMPLETE