What Moves Like AI

What this is for

Three questions, one number behind all of them.

• Harvesting a loss without leaving the market. Sell a position at a loss to book it, and the wash-sale rule stops you claiming that loss if you buy back something substantially identical within 30 days. The usual move is to buy something that behaves similarly without being the same security. What counts as substantially identical is a question for a tax professional, not for a correlation table — but finding the candidates is what this page does.
• Picking between near-duplicates. A dozen funds track the same index. If two of them correlate at 0.999, the choice is about expense ratio, spread and tax treatment, not exposure.
• Finding out you hold one position twice. Two tickers that feel like separate holdings and correlate at 0.97 are one bet wearing two names. That is the most common way a portfolio ends up less diversified than its owner thinks.

How to read the two correlation columns

The one-year figure is what the relationship looks like now. The three-year figure is what it has looked like over a longer stretch. Comparing them is the point: when they agree, the relationship is structural and likely to hold. When the one-year number is far below the three-year one, something has changed — a shift in business mix, an index reconstitution, a sector that has stopped moving together — and the recent figure is the one to trust less, not more.

Correlation measures whether two things move together, not whether they move by the same amount. That is why the volatility column is there. Two tickers can correlate at 0.99 while one swings half again as hard, and swapping one for the other would quietly change how much risk you are carrying.

What was deliberately left out

Ranking purely by correlation produces answers that are mathematically perfect and practically useless. The closest matches to AAPL by raw correlation are leveraged AAPL funds at 0.999 — correct, and not replacements for anything. Four filters run before a ticker is allowed to appear:

• Comparable risk. A candidate has to sit within a band of the subject's own volatility. Note that the band is relative: looking up an ordinary stock or fund, this removes the leveraged products, because they move two or three times as far. Looking up a leveraged product itself, the comparable-risk answer is other leveraged products — and that is what you get.
• Three years of history. A replacement you intend to hold should have traded through more than one market regime. This removes the single-stock covered-call and option-income funds as a class, since almost all are recent launches.
• No structured products. What survives both of the above is the rare instrument that tracks another almost exactly but with visibly damped volatility — the signature of a capped-upside product. Genuine share-class pairs do not look like that, so they come through untouched.
• A liquidity floor. Everything listed trades at least $1.0M a day. A substitute you cannot get filled in is not a substitute.

Where all of that leaves a ticker with no close match, the page says so. A stock with no substitute is a real and useful finding, and inventing a loose one to fill the table would be worse than an empty answer.

Limits, stated up front

• Correlation is backward-looking. It describes the window it was measured over and carries no promise about the next one. Relationships that held for years break, usually at the moment people are relying on them.
• It says nothing about direction or returns. Two tickers can correlate at 0.99 and one can be down 40% over the same stretch. This page ranks co-movement, not performance, and nothing here is a recommendation.
• Daily returns only. Measured on a different frequency — weekly, monthly — the numbers move, sometimes a lot. Daily is the standard choice and it is the one used throughout.
• Substantially identical is a legal test, not a statistical one. A high correlation does not make two securities substantially identical for tax purposes, and a low one does not make them safe. Talk to a tax professional.

Where the numbers come from

Correlations are computed in-house from a daily price history and rebuilt every night. What gets published is the derived statistic — a correlation coefficient, a volatility ratio, an observation count — and nothing from which a price or return series could be reconstructed. The rest of the Markets section is on the Markets overview, including futures positioning extremes and SEC Form 4 insider transactions.