Pairs Trading: What an Honest Test Would Need
Pairs trading bets that two linked instruments will diverge and converge again. On our data it can only be tested formally, and such a test would mislead for three separate reasons.
A relationship that may not exist
The approach rests on a stable statistical relationship, usually cointegration. Testing it on two years of daily data is meaningless: on a short sample a relationship can be found between almost anything, and its stability is the entire question. A strategy built on a link discovered inside a two-year window tests the sample's capacity to produce coincidences, not the market.
Searching pairs as a trap
The second problem is deeper. Twelve instruments form sixty-six pairs, and testing all of them and reporting the best produces an excellent result in the complete absence of any edge. This is classic multiple testing: the more combinations examined, the higher the chance of finding an attractive coincidence. An honest test requires a correction for the number of variants examined, and with that correction little usually survives.
Swap changes the arithmetic
The third reason is specific to forex. A pairs trade is a long and a short position held simultaneously, each with its own carrying cost. The strategy assumes holding for days or weeks, so swap stops being a detail and becomes part of the result. A test that ignores swap on each leg separately measures the wrong quantity.
What an honest test would need
A long history — years rather than two years — to test stability outside the window in which the relationship was found. A correction for multiple testing when selecting pairs. And real swap rates for each leg across the whole period. All of that is obtainable, but we do not have it, so we publish no result rather than publishing a wrong one.
We publish these explanations alongside the runs themselves. Rules that can be tested are tested and published with their numbers, including the negative ones — see the MACD backtest or the weekly range breakout. The absence of figures here is a result, not a gap.
This material is educational and is not individual investment advice. Backtested results do not guarantee similar results in the future. Trading forex carries the risk of losing capital.
Frequently asked questions
Why can't pairs trading be tested on two years of data?
Because on a short sample a statistical relationship can be found between almost any instruments. The question is not whether the link exists inside the window but whether it survives outside it.
What is wrong with searching across pairs?
Twelve instruments produce sixty-six combinations. Reporting the best of them yields an excellent result even with no edge at all — that is multiple testing, and it requires a correction for the number of variants examined.
Why does swap matter for pairs trading?
Because the trade holds a long and a short position simultaneously, each with its own carrying cost, over days or weeks. Without accounting for swap on both legs, the calculation measures the wrong thing.
New backtest studies — to your inbox
When a new reproducible study comes out or the data is updated, we send a short email. No spam, no "signals", no selling your address.
By leaving your email you agree to receive occasional emails from the portal. Unsubscribe in one click from any of them.