ZipLime/fund-holdings-nport

US Fund and ETF Holdings — Form N-PORT

Every position of every US mutual fund and ETF, monthly, including the bonds, loans, asset-backed paper and derivatives that 13F does not report at all.

145 630 731 positions · 17 513 funds · 341 049 monthly reports · 2019-09-30 to 2026-05-31

The pipeline lives in recipe/ at the same revision as the data. See PIPELINE.md for the method.

Why this and not 13F

13F is what everyone uses because it is what everyone knows about. It is also a narrow window: long US equity positions only, once a quarter, filed 45 days late, and only by managers above $100M.

	13F	N-PORT
What is reported	long US equities	every position the fund holds
Bonds, loans, ABS	no	yes
Derivatives	no	yes
Short positions	no	yes, as payoff profile
Frequency	quarterly	monthly
Fund identity	manager, all funds merged	each fund separately
Net assets, flows, returns	no	yes

Equity is a third of what funds hold. The rest is invisible in 13F:

Asset class	Positions
Equity	48 494 795
Loans	38 436 253
Debt	36 851 600
Mortgage- and asset-backed	14 114 701
Derivatives	2 664 045
Short-term investment vehicles	617 875

The two dates

report_date      the month end the portfolio is as of
filed_date       when it reached EDGAR, and could be read

The gap is 57 days at the median, 60 at the ninetieth percentile, and it is published as disclosure_lag_days. Never key a backtest on report_date: a portfolio dated 31 December was not knowable until late February.

pit is keyed on knowledge_date and partitioned by it, so a point-in-time window prunes partitions rather than scanning years wider than it asked for.

What makes it joinable

entity_id in the pit config is the issuer CIK of the security held — the same key used by company-fundamentals, insider-trading and the rest of the family. So "which funds owned this company, when did the market learn it, and what were its fundamentals at the time" is one join, not a research project.

The link comes from the CUSIP bridge in security-master, and only high and medium confidence matches are attached. 96.5% of the 28 036 284 US equity positions resolve, and 28 851 456 positions in all carry an issuer. The rest cannot, by construction:

• foreign equities — a Japanese or British line has no CUSIP and no SEC filer;
• exchange-traded funds — a fund is not an operating company;
• municipal bonds and private placements — the issuer never files.

Measured over all 48.5M equity positions the rate reads 58.6%, which looks like a broken bridge and is not: the missing mass is Tokyo, London and Shanghai.

import polars as pl

pit = pl.read_parquet("data/pit/knowledge_quarter=*/**/*.parquet")
owners = pit.filter(
    (pl.col("entity_id") == "0000320193")          # Apple
    & (pl.col("knowledge_date") <= "2020-06-30")   # as known then
).sort("value_usd", descending=True)

Configs

Config	What one row is
holdings	one position: a security a fund held on a month end, with balance, value, weight, asset class, payoff profile and country
filings	one monthly report by one fund: registrant, series, report and filing dates, total and net assets
returns	one share class, one month, one total return
flows	one fund, one month: sales, reinvestment, redemptions and the net
pit	one point-in-time position keyed on the issuer held, plus a Delta table for ziplime
quarters	one source archive consumed, with its SHA-256

flows is worth a second look. Money into and out of every US fund, every month, as reported by the fund itself. It is not a sentiment index inferred from prices — it is a measurement, and nobody publishes it free.

The fund, not the trust

A registrant like Vanguard Index Funds files for dozens of series. series_id is the fund; keying a holdings table on the registrant would merge the S&P 500 fund with the small-cap one and produce a portfolio nobody holds.

Known gaps

• Nothing before 2019 Q4. DERA's N-PORT sets begin there, with the first quarter after the form became mandatory.
• Derivative legs are not expanded. A position's derivative_category and notional are here; the counterparty and reference-instrument tables in the source archive are not parsed.
• Foreign positions carry no issuer CIK. See above.
• Filing acceptance is a date, not a time. The N-PORT data sets carry no acceptance timestamp, so knowledge_date is the end of the filing day — which errs late and therefore never leaks.
• Positions sum to 1.0001 of net assets at the median. Cash, receivables and derivative netting sit outside the holdings table, so a fund's positions are not expected to add to exactly its net assets — but the median says the value column is read right.
• Amendments are kept, not merged. An NPORT-P/A is a separate filing beside the original; deduplicate on (series_id, report_date) taking the latest filed_date if you want one portfolio per month.

Provenance and updates

Source: SEC Form N-PORT Data Sets, DERA. US Government work, public domain. The quarters config lists every archive consumed with its SHA-256.

Rebuilt weekly, Monday 08:40 UTC, after the security master that supplies the issuer link. DERA publishes an archive once a quarter, so weekly is already more often than the source moves.