Corporate Actions Data: Why Your Position Deltas Lie

Why snapshot diffs are fragile

Most institutional position analytics is subtraction. A 13F is a quarter-end snapshot of a manager's long US equity positions, due 45 days after quarter end, and "what did this fund buy" is computed by diffing two consecutive snapshots. The subtraction silently assumes the security is the same object in both quarters: same CUSIP, same share basis, same issuer. Corporate actions break exactly that assumption, which makes them the single richest source of false signals in holdings data.

At scale this is constant, not occasional. Arkolith's Q1 2026 dataset covers 1,824 institutional filers reporting 1.87 million long positions worth $53.7 trillion. In a universe that size, some fraction of issuers is always mid-split, mid-merger, or mid-spin-off, so a delta pipeline with no corporate-action layer is permanently emitting fiction somewhere. The SEC's own 13F FAQ defines what filers must report; it normalizes nothing across quarters. That is the dataset builder's job, and it is where naive pipelines fail (we catalogued the broader failure family in how accurate is 13F data).

There is one structural consolation. Because hundreds or thousands of filers report the same issuer, corporate actions leave a statistical fingerprint that real trades do not: trades are idiosyncratic, corporate actions are universal. One fund multiplying its share count is a decision. Every fund multiplying its share count by the same ratio in the same quarter is paperwork. Nearly every signature below is a variation on that rule.

Stock splits: the phantom buying spree

A forward split multiplies every holder's share count and divides the price, leaving position value roughly unchanged. A naive quarter-over-quarter share delta reads that as a massive coordinated buy. Reverse splits produce the mirror image: an apparent mass liquidation in which nobody sold a share. The widely covered Nvidia split in 2024 is the canonical recent case: every 13F holder of NVDA showed a share count that leapt by the same multiple while dollar exposure moved only with the market. (The SEC's Investor.gov has plain-language primers on the mechanics.)

The detection signature has three parts. First, the share ratio is suspiciously clean: a small integer for forward splits, a fraction like one tenth for reverse splits. Second, the position's value change is decoupled from the share change; it tracks the stock's market move, not the apparent "buy." Third, and decisively, the same ratio appears across most of the issuer's filer base in the same quarter. Any single test can false-positive. The conjunction of all three almost never does.

Two complications deserve respect. Reverse splits commonly trigger assignment of a new CUSIP, which converts a unit problem into an identifier problem (more below). And insider data inherits the same distortion: a Form 4 is filed within two business days of a trade with the share counts of that moment, so an insider's pre-split transaction history is not directly comparable to post-split holdings without adjustment. An agent comparing raw share counts across a split boundary, in either dataset, will hallucinate conviction changes that never happened.

Spin-offs: a new position nobody bought

In a spin-off, a parent distributes shares of a subsidiary pro-rata to its own shareholders. No one places an order, yet next quarter every institutional holder of the parent reports a brand-new security. Diff logic reads this as hundreds of funds simultaneously initiating the same position, which looks like the strongest consensus-buy signal imaginable. It is not a signal. It is arithmetic.

The corruption runs both directions. The parent's reported value drops by roughly the carved-out subsidiary's worth, which a delta pipeline scores as broad trimming of the parent. One spin-off therefore fabricates two stories at once: phantom accumulation of the child and phantom distribution of the parent.

The signature: an issuer with no filing history appears in the same quarter across a large fraction of the parent's existing holder base, with position sizes roughly proportional to each filer's parent stake. Genuine accumulation never looks like that. New conviction positions arrive at idiosyncratic sizes, at different times, sometimes accompanied by 13D or 13G filings when stakes cross ownership thresholds. Pro-rata proportionality across an entire holder base is the tell of a distribution, not a trade.

Confirmation is cheap because the event itself is filed. The parent's 8-K and the subsidiary's registration documents sit on EDGAR, findable in minutes through the SEC's full-text search. A position-change pipeline that cannot point at the corporate filing behind an anomaly is guessing.

Mergers and re-listings: exits that were not sales

A completed cash acquisition deletes the target's CUSIP from every 13F at once. Naive deltas record a universal exit, and ranking pages dutifully report that institutions "dumped" a company that simply ceased to exist. The proceeds land as cash, which is not 13F-reportable, so the capital appears to evaporate. Stock-for-stock deals are subtler: the target vanishes while every former holder's position in the acquirer inflates without a single buy order, polluting accumulation metrics on the acquirer's side too.

Re-listings are the quietest corruptor of the four. Redomiciliations, SPAC combinations, bankruptcy emergences, and plain name changes can each assign a new CUSIP to a continuing business. Identifier-keyed pipelines then manufacture an exit-and-entry pair: the "old" company abandoned and a "new" one accumulated, by the same filers, in the same quarter, at nearly the same value. We covered why identifiers churn in what is a CUSIP; the practical consequence is that a CUSIP keys a security's paperwork, not its economic identity.

Signatures, in rough order of reliability: a universal simultaneous exit (every filer, same quarter, no partial trimming in prior quarters) is a deletion event, not a sell-off. A universal exit paired with correlated same-filer increases in one specific other issuer indicates a stock deal. Matched exit-and-entry pairs of similar value within the same filer set indicate an identifier change. Any quarterly mass move should be checked against deal filings before being narrated as conviction; our Q1 2026 hedge fund moves analysis applies exactly this filter.

Detection signatures at a glance

The patterns compress into a small lookup table. Keep it next to any quarter-over-quarter screen, because every cell describes a case where the naive reading is the exact opposite of reality.

The unifying heuristic is worth restating: real trades disagree with each other, corporate actions agree perfectly. When the holder base moves in lockstep, with ratios too clean and timing too synchronized, the cause is almost always an event in the issuer's filing history rather than a thousand independent decisions. The anomalies are also much easier to debug once you have seen the raw XML a filer actually submits, rather than only the aggregated table downstream of it.

One key, with the paperwork attached

Arkolith treats corporate actions as a data-integrity problem rather than the consumer's problem. Position rows carry provenance back to the SEC EDGAR accession number they came from, so when an agent flags an anomaly it can cite the actual filing instead of asserting a trade that never happened (the grounding approach behind stopping AI from hallucinating market data). The same discipline extends to the 51,000+ insider transactions we track, where split boundaries corrupt comparisons just as readily.

# Resolve an issuer across ticker and identifier changes
curl -H "Authorization: Bearer YOUR_KEY" "https://arkolith.com/api/v1/search?q=nvidia"

# Browse covered institutional filers and their CIKs
curl -H "Authorization: Bearer YOUR_KEY" "https://arkolith.com/api/v1/funds"

# Pull a filer's holdings, then diff with the signatures above in mind
curl -H "Authorization: Bearer YOUR_KEY" "https://arkolith.com/api/v1/funds/<cik>/holdings"

The practical workflow: before narrating any large position change, run the three checks (ratio cleanliness, cross-filer universality, paired exit-and-entry), then confirm against the issuer's own EDGAR filings. Full endpoint and MCP tool reference lives in the docs.

Frequently asked questions about corporate actions data

What is corporate actions data?

It is the record of events that change a security's structure or identity: splits, spin-offs, mergers, name and identifier changes, and re-listings. For position analytics, the actions that matter most are the ones that alter share counts or CUSIPs between snapshots, because those silently break quarter-over-quarter comparisons.

Do stock splits change a company's CUSIP?

Forward splits usually do not; the share count changes but the identifier survives. Reverse splits commonly trigger a new CUSIP assignment, and mergers, redomiciliations, and name changes almost always do. Any pipeline keyed purely on CUSIP needs a mapping layer for these transitions.

Why does a merger look like a mass exit in 13F data?

Because the target's security stops existing, it disappears from every holder's next quarterly filing at once. A delta pipeline reads simultaneous universal disappearance as everyone selling. The tell is the universality: real sell-offs unwind over multiple quarters at fund-specific paces, with partial trims along the way.

How can an AI agent avoid misreading corporate actions?

Apply the universality rule before narrating any large position change: check whether the same anomaly appears across most filers holding the issuer in the same quarter. Then confirm against the issuer's own EDGAR filings rather than inferring intent from the delta alone. Data with per-datapoint provenance lets the agent cite the source filing instead of asserting an unverifiable trade.

This article explains public filings and data concepts. It is not investment advice.