A.8.4 Insider & Ownership Data

Basic Prerequisites: 02 · SEC EDGAR & XBRL (the filings these datasets are parsed from).

1. Intuition & Practical Objective

Corporate filings fall into two kinds: what the company says about itself (10-K, 10-Q) and what insiders and large holders say about their own positions (Form 4, 13-D/G, 13-F, DEF 14A). The second family is a different kind of data - it is timestamped by law, tied to specific people, and it answers questions the financial statements cannot: who owns this, and what have the people closest to it been doing with their own money?

Three datasets, three questions:

Form 4 - insider transactions. Filed within 2 business days of a transaction by officers, directors, and >10% holders. The most timely, least aggregated dataset in this folder. Aggregators like OpenInsider make it screenable (cluster buys, open-market purchases vs. routine grants).
Schedule 13-D / 13-G - beneficial ownership. Filed when a holder crosses 5%; Schedule 13D within 5 business days (activist-flavoured); Schedule 13G (passive/qualified-institutional) within 5 business days for passive filers (or 45 days after quarter-end for Qualified Institutional Investors), with all 13G amendments due quarterly (SEC 2024 amendments).
Form 13-F - institutional holdings. Every institutional manager with >$100M AUM reports its long positions within 45 days of quarter end. Aggregators like WhaleWisdom turn the quarterly deluge into ownership percentages and fund-level views.

The single most important fact about this family is a timing fact: each has a legally mandated filing lag, and the lag is data. A “13-F ownership” number is never today’s - it is at best 45 days old, and often 4.5 months old by the time the next one lands.

2. Mathematical Ground Truth & Derivations

The lag schedule (the $t_{filed} - t_{event}$ that governs usability):

Form	Who files	Trigger	Deadline	Typical staleness
Form 4	officers, directors, >10% holders	any transaction	2 business days	days
13-D	any holder crossing 5%	crossing 5%	5 business days	days–weeks
13-G	passive/qualified-institutional >5% holder	eligible for short-form	5 business days (passive) / 45 days after quarter end (QII); quarterly amendments	up to ~1 quarter
13-F	institutional manager >$100M AUM	quarter end	45 days after quarter end	45–135 days
DEF 14A (proxy)	the company	annual meeting	varies	months

Net insider flow and the cluster signal. For a window $[t_{0}, t_{1}]$ , with transactions $i$ carrying code $c_{i} \in {P, S, A, M, \dots}$ , shares $n_{i}$ , price $p_{i}$ :

NetInsiderFlow = i : c_{i} = P \sum n_{i} p_{i} - i : c_{i} = S \sum n_{i} p_{i} .

The cluster buy signal is a count condition, not a dollar condition - the theory being that several insiders independently choosing to buy open-market ( $P$ ) in a short window carries more information than one large purchase:

# {distinct insiders with a P inside a 30-day window} \geq 3 ⟹ cluster buy .

Ownership concentration. With $k$ reporting institutions holding $n_{j}$ shares and $N$ shares outstanding:

InstitutionalOwnership = \frac{\sum _{j = 1}^{k} n _{j}}{N}, Concentration (HHI) = j \sum (\frac{n _{j}}{N})^{2} .

The critical trap - codes carry opposite meanings. $A$ (grant/award) and $M$ (option exercise) are compensation events; only $P$ (open-market purchase) is a discretionary open-market buy. Counting $A$ and $M$ as “insider buying” is the single most common error in this dataset - and aggregators exist largely to let you exclude them.

3. Computational Implementation - Form 4 flow, cluster detection, and 13-F ownership

Offline and fully runnable (stdlib only). It parses a small embedded Form 4 ledger, computes net open-market flow, detects a 30-day cluster buy with real date arithmetic, and reports institutional ownership.

Read the output carefully: the firm shows a modestly positive net insider flow (+ $100.5 k) an d a g e n u in e * * c l u s t er b u y * * (3 d i s t in c t in s i d ers in s i d e M a rc h) - b u t a l so a l a r g e CEO * * s a l e * * in A p r i l, an d a 3, 000 - s ha re g r an tt ha t anai v e p a rser w o u l d a dd t o " in s i d er b u y in g ." T h es i g na l i so n l y l e g ib l eo n ceyo u (a) k ee p$ P $an d$ S $se p a r a t e, (b) e x c l u d e$ A/M$, and (c) look at distinct insider counts within a window rather than a raw dollar total.

Where the data comes from, and what to use.

Dataset	Upstream filing	Free route	Convenience route
Insider transactions	Form 4	EDGAR full-text / daily index	OpenInsider (free)
5% beneficial ownership	13-D / 13-G	EDGAR	aggregators
Institutional holdings	13-F	EDGAR	WhaleWisdom (freemium)
Compensation & holdings	DEF 14A	EDGAR	vendored “insider” feeds

4. Failure Modes & First-Principles Breakdowns

Counting grants and awards as buying (first principle: only $P$ is discretionary). Codes $A$ (grant) and $M$ (option exercise) are compensation, mechanically scheduled, and carry no bullish information. A “net buying” screen that includes them is measuring the compensation calendar, not insider conviction.
Ignoring the 45-day 13-F lag (first principle: the lag is the data). “Institutional ownership was 55%” is a statement about a quarter that ended ≥45 days ago, and the live number may be materially different. Backtesting on 13-F data without lagging by the filing date leaks ~1.5–4.5 months of information, exactly as in 01.
Double-counting in 13-F aggregation (first principle: holdings are a set, not a sum). The same shares are reported by multiple managers; naive summation overstates ownership, and the same manager appears under different filer names/CIKs across quarters. Always key on CIK, and reconcile against shares outstanding.
10b5-1 plans and routine sells (first principle: a pre-scheduled sale is not a vote). Many senior insiders sell on automatic 10b5-1 schedules regardless of outlook. Treat programmatic sales as noise; weight discretionary $P$ buys (and clusters) much more heavily.
Survivorship in the insider universe. Tickers delisted since a Form 4 was filed often drop out of aggregator screens - reintroducing exactly the bias 05 formalizes.
Amended Form 4s (4/A). Reports are corrected; a parser that ignores amendments can double-count or miss the true transaction. Key on accession number and take the latest version.

5. References

SEC, EDGAR filing-type documentation
OpenInsider (openinsider.com)
WhaleWisdom (whalewisdom.com)
Loughran & McDonald (JF, 2011)
Jensen & Meckling, “Theory of the Firm…” (JFE, 1976) and Jensen, “Agency Costs of Free Cash Flow…” (AER, 1986)

6. Connected Graph Bridges

Back: 03 · Commercial Providers · Index Hub
Forward: 05 · Failure Modes · 06 · Advanced Extensions
Theory: Capital Structure & Corporate Finance (agency, ownership, governance) · Accounting Quality & Red Flags (insider selling as a red-flag input)

Kwant Atlas

Explorer