Pair trading, and how it differs from buying

The short version

Buying asks one question: will this go up?

A pair asks a different one: will these two come back together? Direction does not appear in it. Both assets can fall, both can rise, and the pair settles on the distance between them — nothing else.

That is the entire idea. Everything below is what it costs to hold that idea in practice.

What the position actually holds

A classic pair is long one asset and short the other, sized so the two legs cancel. The proportion is not one-to-one: it is the hedge ratio β, how much the second asset historically moved for each move of the first.

What that cancellation removes is the market. If everything falls 30% together, both legs of the pair move with it and the result barely notices.

What is left is the residual — the part of one price the other does not explain. Measured against its own recent range, that residual becomes a z-score, and a z-score is the thing a pair trader actually watches. The dollar price of either leg has stopped being the subject.

For someone who simply holds coins, the same idea has a version without shorting: rotation. Sell the one that has become expensive relative to the other, buy the cheap one, hold the new position until the distance closes. The result is relative — more of the other asset, not more dollars — which is exactly what the next section shows.

A worked example, from this site's data

Take BNB and NEAR, and the first months of 2022.

On 7 January 2022 the spread between them crossed −2σ: NEAR had become expensive relative to BNB, further than it had been at any ordinary point of the preceding six months. BNB traded at $448, NEAR at $15.06. One BNB bought about 30 NEAR.

Then the whole market fell.

By 22 May 2022 the spread was back inside its norm, and the episode closed after 135 days. BNB was $319 — down 28.7%. NEAR was $6.24 — down 58.5%. Anyone holding either coin lost money, and the one who picked "the better project" lost more.

The distance, though, did exactly what it was supposed to do. One BNB now bought 51 NEAR instead of 30: the pair moved 72% in favour of the side that was cheap.

A holder who rotated NEAR into BNB on 7 January was still down 28.7% in dollars at the end of it. They also finished with 72% more value than if they had kept the NEAR. Both statements are true, and the second one is the only thing a pair ever promises.

Where pairs go wrong

The spread usually widens before it closes. The same pair, three years later: an episode opened on 3 December 2024 and did not close until 19 November 2025 — 351 days. Anyone who acted on the first crossing of ±2σ spent a year being early. On this site the median episode across every pair in the showcase runs around two months, and the distribution has a long tail to the right.

The relationship can simply end. In early 2022 one BNB bought thirty NEAR. The ratio is a multiple of that today. The spread kept returning to "normal" along the way, and normal kept moving. A rolling window has no opinion about whether the level it is averaging still means anything — see correlation and cointegration for the difference between two assets that move together and two assets that belong together.

Two legs cost twice. Fees on both sides, funding on the short, and the risk that the short leg is unavailable or expensive to borrow exactly when the opportunity looks best. Rotation avoids the borrow, but pays the spread twice and realises a taxable event in most jurisdictions.

Being market-neutral is not being safe. It removes one risk and keeps another. A pair that diverges and stays diverged loses money steadily while the market goes nowhere, which is a far less comfortable way to be wrong than a position that simply fell with everything else.

What a pair is actually good for

It answers comparative questions, and only comparative questions. Which of these two is unusually priced against the other right now. How far from normal it is. How long that gap has historically taken to close.

It says nothing about whether either asset is worth owning, and it is not a strategy on its own — it is a measurement that a strategy can be built on. For a holder the useful form of the question is the plain one: of the things I could hold instead, is anything unusually cheap against what I hold now, and how long did that usually take to pay?

Where to look next

The screener ranks every pair on the site by that distance, so the comparative question can be asked across hundreds of pairs at once instead of one. The regression spread explains how the distance is measured in the first place, and the BTC/ETH ratio walks through the same reading on the pair most people already follow.