ETF overlap: how much of any two funds is the same stock

Any two of the 687 funds. Answered in the page from the filings — no account, no request to anything.

For each security both funds hold, take the smaller of its two weights, and add those up. If VOO has Apple at 6.6% and QQQ has it at 6.7%, that pair contributes 6.6 points of overlap: that much of your money is in Apple either way. It is not the count of shared names — two funds can share 500 companies and overlap by very little if the weights differ, and share 30 and overlap by most of the money.

• Positions are matched on CUSIP, then ISIN, then name — so a company held through two identifiers counts once.
• Securities-lending cash collateral is excluded and the weights renormalised (ex-collateral-v2), because it is not an investment the fund chose.
• Combined effective holdings assumes equal money in each fund: it says how many equal positions the pair behaves like, and it is often lower than either fund alone.
• Each fund files on its own date. A pair compares two filings that may be weeks apart; both dates are in the per-fund file.

Different tickers, different issuers, all but identical portfolios. Of the 687 funds measured, 73 have something holding 90% of the same money, and 379 — more than half — have something at 50% or more. Holding two of these together is not diversification.

The other end: funds whose highest overlap with any of the other 686 is small. Only 45 funds have nothing above 10%. If a portfolio needs genuinely separate exposure, this is the short list — and it is short.

16 funds are left out of that list on purpose. They hold other funds rather than securities — AOA, AOM, AOR, BNDW, BSCP, BUFD and 10 more — so their position-level overlap with anything is near zero while their actual exposure may be identical to the fund they wrap. The first run of this dataset ranked one of them the most distinct fund on the site, which is how the check got written. They stay in the matrix, flagged, and out of this table.

• Overlap is not correlation. Two funds holding nothing in common can still move together; two funds overlapping by 90% can differ if the remaining 10% is leveraged. This measures shared holdings, not shared behaviour.
• High overlap is not a verdict. Two near-identical funds may differ on fee, tax treatment, lending policy or spread. This says they hold the same things, not that one is better.
• Funds below $1bn are not here. The floor is a judgement about where the filings thin out; the pair count grows as the square of the fund count.
• Leveraged and inverse funds are excluded as not comparable — notional exposure would make the arithmetic meaningless.
• Every pair is re-derived before each release and checked against the 266 pairs measured independently for the comparison pages; a disagreement stops the deploy.

• overlap_matrix.csv.gz — all 235,641 pairs: ticker A, ticker B, overlap %, shared count, combined effective holdings.
• overlap_matrix.json — the index: every fund's closest match, median overlap and the flags.
• /data/overlap/TICKER.json — one fund's forty nearest, with dates.
• Agents: the MCP server answers compare_funds directly.
• Free to reuse with attribution. If a figure looks wrong, say so and it will be corrected in place.