THE LAB · TEST 001

RSI 80/20 — the internet's favourite entry, tested

Buy when RSI leaves oversold. Sell when it leaves overbought. It is taught in a thousand videos, almost never with evidence. We ran it through a deterministic backtest across nine pairs — up to 30 years of daily data and 700 days of hourly data — with the rules declared before the results were seen.

Published by the desk behind a live, verifiable trading record · free · nothing sold · no sign-up


VERDICT · NO USABLE EDGE AFTER COSTS

The win rate is real. The edge is not. With the classic RSI-50 exit, 63.4% of daily trades and 62.8% of hourly trades won — and it still barely paid. Across nine pairs and roughly two decades of daily data the strategy produced +523.2 pips in total — about +4.7 pips per trade, less than the day's spread on some pairs — while at its worst sitting 9,632.1 pips below its own high-water mark. On the hourly chart: +865.5 pips across 325 trades, +2.7 pips per trade — smaller than one extra pip of slippage. Change the unstated exit and the sign flips. A 63.4% win rate that cannot reliably beat its own spread is not a strategy. It is a coin with better marketing.

The claim

RSI (the Relative Strength Index) measures how one-sided recent price movement has been, on a scale of 0–100. The popular teaching: below 20 a market is "oversold" and due a bounce, above 80 it is "overbought" and due a fall. The classic entry waits for RSI to leave the extreme — cross back above 20, or back below 80 — and trades the snap-back. What the videos rarely tell you: when to get out, what it costs, or what happened when someone actually ran it.

The rules we tested

Declared in code before any result was seen. The exit is almost never specified by the people teaching this, so we tested three reasonable readings and publish all three — no picking the winner afterwards.

RSI(14), Wilder smoothing, on bar closes

LONG   when RSI crosses UP through 20      # leaves oversold
SHORT  when RSI crosses DOWN through 80    # leaves overbought
ENTRY  at the NEXT bar's open              # no look-ahead
       one open trade per pair; signals while in a trade are skipped

EXITS  (three schemes, all published):
  RSI-50 exit   close at next open after RSI crosses 50,
                time-capped at 30 bars
  5-bar hold    close at the open 5 bars after entry
  10-bar hold   close at the open 10 bars after entry

COSTS  round-trip spread subtracted from every trade:
       EURUSD/USDJPY 1.0 · GBPUSD/USDCHF/USDCAD/AUDUSD/NZDUSD 1.5
       EURJPY 2.0 · GBPJPY 2.5 pips

Data: Yahoo Finance price history — the same source our live record's outcome scoring uses. Daily bars back to 1996, hourly bars for the last ~700 days (the source's hourly limit). 167 daily signals and 451 hourly signals in total.

Finding 1 — on the daily chart, it barely ever fires

RSI(14) on daily bars spends almost all its life between the famous levels. EURUSD closed with RSI past 20 or 80 on only a handful of days — and produced 13 entry signals in 23 years: one trade roughly every 21 months. Anyone teaching this on a daily chart is teaching a strategy that almost never trades.

PAIRYEARS OF DATA TIME PAST 20/80SIGNALS SIGNALS / YEAR
GBPUSD22.71.1%200.9
EURUSD22.70.5%130.6
USDJPY29.80.9%361.2
USDCHF22.90.8%190.8
USDCAD22.90.9%190.8
AUDUSD20.20.7%160.8
NZDUSD22.70.3%70.3
GBPJPY22.70.8%261.1
EURJPY23.50.3%110.5

"Time past 20/80" = share of days the RSI closed beyond either famous level.

Finding 2 — the win rate is seductive, and it means nothing

Both timeframes show a win rate near 63% with the RSI-50 exit. That is not evidence of an edge — it is the shape of the exit. Snap-back trades bank many small quick wins, and the losers are carried for up to 30 bars. High win rate, small average win, occasional deep loss: the profile every martingale-adjacent strategy wears. Our expectancy calculator will show you a 63% win rate losing money all day long once the loss size outgrows the win size.

Finding 3 — the results

Daily bars, nine pairs, pooled (30 years max span)

EXITTRADESWIN RATE GROSS PIPSNET PIPS AVG/TRADEPROFIT FACTOR MAX DDWORST RUN
RSI-50 exit11263.4%+699.7+523.2+4.71.09,632.17
5-bar hold14053.6%+3,172.6+2,955.6+21.11.32,203.88
10-bar hold12551.2%-459.0-653.5-5.21.07,096.48

All figures in pips, net of the disclosed spread unless marked gross. MAX DD = deepest peak-to-trough fall of the pooled cumulative curve. WORST RUN = longest streak of consecutive losers.

-8,000-6,000-4,000-2,00002,000Feb 1997Aug 2026RSI-50 exit +5235-bar hold +2,95610-bar hold -654

Pooled cumulative net pips, daily bars, all nine pairs, by exit scheme. Solid ink = RSI-50 exit · vermilion = 5-bar hold · dashed = 10-bar hold.

Hourly bars, nine pairs, pooled (~700 days)

EXITTRADESWIN RATE GROSS PIPSNET PIPS AVG/TRADEPROFIT FACTOR MAX DDWORST RUN
RSI-50 exit32562.8%+1,369.5+865.5+2.71.11,083.66
5-bar hold39248.2%-267.6-872.1-2.20.81,413.99
10-bar hold35748.7%-28.1-581.6-1.60.91,507.08
-1,00001,0002,000Sep 2024Aug 2026RSI-50 exit +8665-bar hold -87210-bar hold -582

Same strategy, hourly bars, last ~700 days. Note the 5-bar hold — the best exit on daily data — is the worst here (-872.1 pips). That instability is the finding.

Per-pair breakdown, RSI-50 exit (daily)
PAIRTRADESWIN RATE NET PIPSPROFIT FACTOR
GBPUSD1681.2%+832.51.4
EURUSD966.7%+1,100.33.2
USDJPY2045.0%-1,272.20.7
USDCHF1376.9%+2,139.514.1
USDCAD1154.5%-396.70.7
AUDUSD955.6%-779.60.5
NZDUSD785.7%+574.34.7
GBPJPY1968.4%+1,899.81.6
EURJPY837.5%-3,574.70.1
Per-pair breakdown, RSI-50 exit (hourly)
PAIRTRADESWIN RATE NET PIPSPROFIT FACTOR
GBPUSD4158.5%+41.51.1
EURUSD3363.6%-169.60.7
USDJPY4757.4%-100.30.9
USDCHF3669.4%+101.31.3
USDCAD2860.7%+165.51.5
AUDUSD2766.7%+86.91.2
NZDUSD3655.6%-103.40.8
GBPJPY3661.1%+305.71.3
EURJPY4173.2%+537.91.8

Finding 4 — the profit depends on a choice nobody teaches

On daily bars the 5-bar hold made +2,955.6 pips and the 10-bar hold lost -653.5. On hourly bars the 5-bar hold lost -872.1. Same entry, same data — the sign of two decades of results flips on an exit rule the strategy's teachers never state. When an entry only makes money under one particular unstated exit, the entry is not where the money is. Mostly, there is no money.

What this test cannot tell you

What would change our mind

A pre-declared variant — exact parameters, exact exit, costs included — that survives on data it was not fitted to, across pairs, with enough trades to mean something. That is the bar we hold our own system to, in public, with real money. It is the bar any strategy you are taught should meet before it touches yours.

Next: how backtests fool people — The Backtest Trap (free course) · run your own numbers with the expectancy calculator · or see the live record this site actually stands on. ← All Lab tests