Three games, mashed together
The note sounds like one dashboard. It is three desks that do not share a file.
| Game | Who plays it | What they see | Horizon | Public? |
|---|---|---|---|---|
| IR surveillance | Issuer IR, Nasdaq IR Insight, Broadridge, Markit | Daily DTC participant positions, NOBO lists, transfer-agent book | This week | No |
| Event / ownership | Event funds, activists, our parser | Form 4, 13D/G, 13F, N-PORT | Days to quarters | Yes |
| Tape / squeeze | Stat arb, stock-loan desks, Twitter | Dark-pool prints, borrow fee, short interest | Minutes to weeks | Partial / paid |
Form 4 — the only “NOW” that is real
Officers, directors, and 10% holders generally file by the end of the second business day after the trade. Once it is on EDGAR, we can see it in seconds. That is the only item on the original list with a clean public clock.
Retail feeds (Quiver, Unusual Whales, OpenInsider) dump “CEO sold $2m.” Serious users throw most of it away. They keep open-market P and S, drop M/F/A/G (exercise, tax withhold, grant, gift), then classify the person, not the headline.
The academic filter
Cohen, Malloy, Pomorski, Journal of Finance 2012: if an insider trades in the same calendar month for three years, call them routine. Everything else is opportunistic. Routine trades: ~0 abnormal return. Opportunistic portfolio in-sample: +82 bp/month value-weighted, +180 bp equal-weighted. They also predict subsequent firm news. That is the canonical Form 4 “secret.” It is a 1975–2008-ish portfolio with costs, capacity, and post-publication decay unaccounted for — a research flag, not a printing press.
| Keep | Toss |
|---|---|
| Open-market buy (P), especially CFO / operator, not just CEO | Option exercise + same-day sale (M then S/F) |
| Several different insiders buying after a drawdown | One scheduled 10b5-1 sale |
| Large % of that person’s remaining stake | Tiny sale vs a huge leftover position |
| Breaks their own calendar | Same month, same size, every year |
| Clustered purchases | Raw “cluster selling” headlines |
Raw sales are usually not a short. Insiders sell for houses, taxes, and diversification. Only sales that are large and a large fraction of holdings look bearish. Form 144 is “I might sell up to X.” Form 4 is “I sold Y.” Do not add them.
What the edge looks like
Stock is down 25% into a quiet period. Three officers who have not bought in years file open-market P’s in two days, no 10b5-1 box. Treat that as a research flag, size small, hold through the next print. Not “insider sold, short it.”
13F — idea generation that got crowded
Quarter-end long sleeve, filed up to 45 days later. No shorts, no swaps, many non-US names missing. Confidential Treatment can hide the actual idea for another 3–12 months, then it shows up on a 13F-HR/A.
WhaleWisdom / Fintel / our Popular Portfolios already do celebrity books via FMP. That product exists. The leftover public edge is not “Buffett still owns AAPL.”
Best ideas, not the book
Copy only a manager’s largest active overweights. Screen for low-turnover, concentrated, long-only or activist names. Alert on new positions and full exits, not levels.
Confidential amendments
Agarwal, Jiang, Tang, Yang (JoF 2013): names hidden then later amended in outperformed the same manager’s public book by roughly 5–6% annualized in sample. Watch the cover-page omit box, then the amendment.
Why the easy clone died: the lag, the missing short book, copycats creating a filing-day pop, and 2020s crowding. A 2012–2025 lag-respecting backtest of popular “copy 13F” recipes found no significant factor alpha. Use 13F to answer who is the shareholder base, not what to buy tomorrow.
13D / 13G — “should I be concerned?”
This is the form the Slack note is actually about. A 13D is a 5%+ holder who will not call themselves passive. Deadline is now five business days after the trigger. A lot of the move happens in the run-up, before the file.
Brav, Jiang, Partnoy, Thomas (JoF 2008): about 7% abnormal return in the [−20, +20] window around the first activist 13D. Median ~5.6% when the 13D is the first public demand. Greenwood/Schor: a lot of the long-run return sits in names that later get acquired. Independent-and-stuck often goes nowhere. Item 4 “we may talk to the board” is not “sale of the company.”
The high-signal version is a 13G → 13D conversion: a passive holder just said they might fight.
How to use it
- New 13D or 13G→13D hits.
- Read Item 4 + exhibits. Tag: board, sale-process, capital return, or boilerplate.
- If it is a real campaign on a small/mid name that has not already run +15% on the rumor, that is an event — sized for a binary outcome, not a 7% average.
N-PORT — confirmation, not a tape
Mutual-fund / ETF book. Public month is the third month of the quarter, +60 days. The original note said 30. That is wrong; do not productize a wrong lag. The delay exists to kill front-running. Copycat papers say you can roughly match a fund after costs if you copy recent winners with representative disclosures. You do not get a nowcast.
Use: did a persistent active mutual fund show up in this name last quarter? Complements 13F (funds vs managers).
Short interest and cost to borrow
Two hypotheses people mash together:
Crowded short → later underperformance
High SI, especially with scarce borrow or rising short demand, has historically predicted lower returns. Days-to-cover is a bearish crowding proxy, not a squeeze timer.
Buy the squeeze
Rare, nonlinear, needs a catalyst + fragile borrow + fuel. High CTB more often means the stock is expensive to short because the news is bad. GME covering was not even the main driver.
FINRA SI is twice a month, published T+7 business days. CTB is a stock-loan vendor number. FINRA SLATE daily loan stats are 2027. Not in EDGAR.
Dark pools and block “alerts”
A print has two sides. You cannot see “institution bought.” Midpoint dark crosses are common; Lee/Ready-style inference is wrong a lot. Desks mark the price level and wait for the lit tape to accept or reject it.
Public FINRA ATS data is weekly, delayed 2–4 weeks, no side. The Slack alert is a Terminal / proprietary thing.
DTC #0050, solicitors, NOBO
This is the job the Slack author used to have. It is not a public alpha book.
Weekly IR stock watch: pull the issuer-only DTC Security Position Report. Participant 0050 is Morgan Stanley & Co. LLC clearing, not “hedge funds.” Rank weekly Δ shares. Overlay NOBO (retail who did not object), the transfer-agent DRS book, Bloomberg holders, last 13F, any 13D, Form 4s. Then call the top holders. The call is the data.
A drop at 0050 can be prime-broker client sell, MS prop, retail, an ETF AP, collateral, or NSCC netting. IR says “fast money” the way fishermen say “they’re biting.” Sometimes they are right because they also called the fund.
What the leftover edge actually is
confirmed, timely Form 4 open-market P (opportunistic) new 13D / 13G→13D with a real Item 4 inferred, delayed 13F new / exit / confidential amendment N-PORT persistent active-fund add context, not a trade SI / days-to-cover (stale, usually bearish) holder mix (index vs active vs activist) private, not ours DTC SPR, NOBO, solicitor vote reports cosplay dark-pool “buy” badges, 0050 = fast money
Four rules that survive a backtest
- Filter harder than the feed. Most Form 4s and most 13F rows are worthless. The literature’s entire point is the residual after you throw 50–80% away.
- Respect the clock. T+2 ≠ 5 days ≠ 45 days ≠ 60 days ≠ biweekly+7. One “smart money now” tile is how you lie.
- Asymmetry. Buys and activist 13Ds have a long literature. Raw sells, celebrity 13F clones, and squeeze-chasing do not.
- Confirm across games, don’t average them. Useful combo: opportunistic insider buys + new 13D. Slack combo (volume + 0050 + dark pool + borrow) is an IR desk with a Terminal, not a parser.
What we already have, and what to build
sec-api is an earnings machine: ticker → CIK → 8-K 2.02 / 6-K → EX-99.1 → street EPS. It finds every form on this list in submissions. It parses almost none of them in production.
- Submissions, archive index, 10 rps governor
- Popular Portfolios = FMP 13F by CIK
- App insider endpoints = FMP Form 4
- Warehouse regex stubs for Form 4 + 13F XML
- Atom filter already matches 4 and SC 13D/G
- New form family = new ingest job
- Deterministic XML, not LLM — except 13D Item 4 tags
- Replace warehouse regex with schema-driven XML
- Gold-set against edgartools fixtures
| Ship | Why it’s the real edge | Never say |
|---|---|---|
| Opportunistic Form 4 buys (CMP + code P + % of holdings) | Only timely public ownership signal with academic teeth | “Insiders are dumping” from raw S |
| 13D / 13G→13D + Item 4 tags | Event premium; answers “should I worry” | Every 13D is a 7% gift |
| Issuer-centric 13F: new / exit / confidential amend | What’s left after clone strategies died | “Buffett just bought” on a 45-day file |
| SI + days-to-cover, labeled stale | Context next to the above | Squeeze alert |
Sequence: keep the earnings spine alone. Warehouse Form 4 live ingest next. Then DERA 13F backfill + issuer-centric exits. Then 13D Item 4. FINRA SI only if we still want their recipes. Never claim DTC, dark-pool side, or NOBO from this repo.
There is no hidden EDGAR form that tells you who is selling this morning. The people who sound like they have that are the issuer, a prime broker, or someone selling a story.