RSI oversold/overbought (21/25/80)
What has fallen too far tends to come back and what has risen too far tends to correct: entering as price leaves each extreme captures the reversal from the start. It works above all in a sideways market, where price spends most of its time.
Measured in crypto — Binance spot · 540 pairs, delisted ones included · 0.2% per round trip
- ✓invariance
- ✓costs
- ✓placebo
- ✓benchmark
- ✓out of sample
- ✓multiple testing
What was measured
How many of those trades actually count
What it paid, before any deductions
Why it passed
- the profit spread across the trades instead of coming from a handful
- the edge survived the broker's fee
- it paid more than entering on randomly drawn dates
- it paid more than buying and holding for the same stretch
- the result came back in the separated pieces of the data
- it survived the discount for having tested many versions
What this result does NOT say
- One market, one universe
- Measured on 54 spot cryptocurrency pairs, delisted ones included. It says nothing about futures, equities or indices, nor about how the same technique behaves in another market.
- One window of time, not every window
- The measured period runs from 2017-08-17 to 2026-07-28. A market moves through regimes, and a technique can work in one and fail in another — the card measures the regimes that fit inside this window, not the ones still to come.
- One cost structure
- The cost charged is 0.10% per leg, in and out. Anyone paying more than that gets a worse result, and anyone paying less gets a better one — the verdict holds for this fee.
- One exit rule
- The trade was closed by: the technique itself (held until the opposite signal). The same entry measured with a different exit is a different strategy, and can earn a different verdict — it happens in this archive.
- The number of trades is not the sample size
- There are 87 trades, but only 71 independent market episodes: a single move fires the technique across dozens of assets at once, and counting those as separate observations inflates any result. It is the smaller number that governs the arithmetic. With 71 episodes, what the data supports is a range from +29.99% to +125.36% per trade — the published average is the centre of it, not the exact measurement.
- Daily bars
- Measured at the daily close. Nothing here measures what happens inside the day, and an intraday technique is not auditable with this data.
Numbers and reproducibility
The six controls
| control | status | t | threshold | episodes |
|---|---|---|---|---|
| invariance | passed | — | — | — |
| costs | passed | 3.24 | 2.00 | 71 |
| placebo | passed | 2.90 | 2.00 | 71 |
| benchmark | passed | — | — | — |
| out of sample | passed | — | — | — |
| multiple testing | passed | — | — | — |
- invariance87 signals across 140496 bars
- costsgross +77.674% · cost 0.200% · net +77.474% (t=3.24)
- placeboactual +77.474% · placebo -17.962% · excess +95.437% ± 32.871% (t=2.90, 71 real groups, 4,350 sham dates)
- benchmarktechnique +77.47% · buy and hold (same horizon) +53.87% · excess +23.61%
- out of sampleasset half A: +108.000% (t=3.36, 39 groups) · asset half B: +48.984% (t=1.52, 37 groups) · earlier period: +96.183% (t=2.04, 35 groups) · recent period: +58.331% (t=4.85, 36 groups)
- multiple testing108 variation(s) tested · t=3.24 · p≈0.0012 · false positives expected by chance ≈ 0.13
Reproducibility
- period
- 2017-08-17 to 2026-07-28
- assets that traded
- 54
- variations tested before this one
- 108
- gross per trade
- +77.67%
- net per trade
- +77.47%
- exit rule
- the technique itself (held until the opposite signal)
- median duration bars
- 451
- mean duration bars
- 627.5
- max duration bars
- 2341
- fee per leg
- 0.001
- RSI period
- 21
- oversold
- 25
- overbought
- 80
Binance spot klines (delisted pairs included) · collected from 2026-07-27 23:31 to 2026-07-27 23:31 · 106 assets · 140,496 bars · 2017-08-17 to 2026-07-28
Seed not recorded: audits before 2026-07-29 used 20260728 by convention in the code, and the value is not in the record.
Hypothesis, filed before the result
The 'rsi' family contains no combination with a real edge: the ones that look good are the tails of the grid itself. I expect the sweep's best `t` not to survive the multiple-testing control.
the filing date is not in this audit's record
The original, as it was filed
A família 'rsi' não contém uma combinação com vantagem real: as que parecem boas são as caudas da própria grade. Espero que o melhor `t` da varredura não sobreviva ao controle de múltiplos testes.
Pre-registration exists to keep prediction apart from rationalisation: written after the number, every hypothesis is right.
Earlier audits of the same technique
Each variation an author teaches enters as its own test, so that whatever might work in the strategy gets covered. The verdict changed along the way — a fix in the engine changes the number, and that is a fact that has to stay visible. No earlier record was deleted.