Why a 44% win rate at 1.2R still loses — and why changing exits won't fix it
In short
- "I win 44% of trades and my target is 1.2R — why am I losing?" Because the average win is not 1.2R. Partial profits and breakeven stops cut it to about +0.6R.
- We replayed the strategy's real code over 1,621 trades on 30 coins and tried five exit rules. Win rates ranged from 41% to 53% — yet every rule ended between −0.066R and −0.054R a trade.
- Holding every trade until it hits the stop or target, with no time limit, was slightly worse still.
- Before fees the strategy made +0.032R a trade. Fees cost 0.094R. With a 1.3% stop on a 15-minute chart, costs eat any small edge.
On 6 November we started a paper-trading bot on the 15-minute chart. It traded a "market read" that looks for a sweep of a key level — the previous day's high, the Asian-session low, the day's open — followed by a close back inside and order flow turning in the trade's favour. After one day it had taken 18 trades and won 8 of them: 44%. With a target of around 1.2R in mind, that sounds close to breakeven. It was down 3.8R.
The obvious question: what is wrong? And the obvious fix: change the exits. Take the full target instead of half. Stop moving the stop to breakeven. Let winners run. We tested all of those properly, and none of them was the answer.
The arithmetic of a win rate
A strategy breaks even when win rate × average win = loss rate × average loss. With a true 1.2R on every winner and −1R on every loser, the breakeven win rate is 1 ÷ (1 + 1.2) = 45.5% — before fees. So 44% at a clean 1.2R would lose a little. But the bot's winners were not clean 1.2R winners.
Its exits worked like most "safe" trade management: close half at the first target (TP1), move the stop to the entry price, let the rest run to the second target (TP2). Of its 8 winning trades, 5 took the first half and then got stopped at breakeven on the rest. Each of those earned about +0.2R. Its average winner was +0.61R; its average loser −0.87R. At those sizes the breakeven win rate is 0.87 ÷ (0.61 + 0.87) = 59%.
That much is just arithmetic, and it is the most common reason a "high win rate" strategy loses: the win rate is measured on trades that include tiny wins. Always look at the average win and average loss alongside the win rate. But 18 trades on one day prove nothing about the strategy itself, so we went further.
How we tested it
The strategy reads live market data — order flow, open interest, the day's key levels — so it cannot be tested with a simple formula. Instead we ran the actual server code against history: a simulated clock steps through every 15-minute close from 9 Oct 2026 to 6 Nov 2026, and a simulated exchange answers each data request with only what was known at that moment. Candles still forming are built from the minutes that had passed.
As a check, the replay reproduced the bot's real Ethereum trade from 6 November exactly: the same stop, the same first target and the same second target, with an entry within 0.05% of the live one.
- Coins: the 30 highest-volume Binance USDⓈ-M perpetuals that had been listed for at least 90 days — the same universe the bot scans.
- Entries: every signal the strategy rated a valid setup, filtered exactly as the bot filters them: once per level and side per day, one position per coin, no coins moving less than 0.25% or more than 1.8% per 15 minutes, stops no wider than 5%.
- Trades: 1,621, with a median stop 1.31% from entry.
- Exits simulated on 1-minute candles, stop first if a candle touches both. Fees: 0.05% taker plus 0.01% slippage on market fills and stops, 0.02% maker on target fills.
Five exit rules, same trades
Every exit rule below was applied to exactly the same 1,621 entries, so any difference comes from the exit alone. Each trade is closed after 8 hours if it has not finished, as the bot does.
Average result per trade by exit rule
Same 1,621 entries, after fees · 9 Oct 2026 – 6 Nov 2026
Dots are the average R per trade; lines are 95% ranges from resampling whole trading days. The five rules overlap almost completely.
Show the numbers as a table
| Exit rule | Won | Avg win | Avg loss | R per trade | 95% range |
|---|---|---|---|---|---|
| Current: half at TP1, breakeven, rest at TP2 | 52% | +0.77 | −0.96 | −0.062 | −0.13 to +0.01 |
| No partials: all to TP2 | 41% | +1.28 | −0.99 | −0.066 | −0.16 to +0.04 |
| Half at TP1, no breakeven move | 45% | +0.95 | −0.88 | −0.063 | −0.14 to +0.01 |
| No partials: fixed 1.2R target | 48% | +0.97 | −0.98 | −0.054 | −0.11 to +0.01 |
| No partials: all at TP1 | 53% | +0.75 | −0.99 | −0.060 | −0.12 to 0.00 |
Look at the win rates and the averages together. Taking everything at the first target wins 53% of trades with a small average win. Letting the whole position ride to the second target wins only 41%, but its average win is twice as big. Removing the breakeven move turns some small wins into full losses and some breakeven exits into big wins. Each change trades win rate for win size — and they cancel. Every rule lands within 0.015R of the others.
This is the general rule worth remembering: exits reshape a strategy's results; they rarely create an edge. If an entry has no edge, the price after it moves randomly, and no exit — tight, wide, partial or trailing — can reliably extract money from a random walk. You can choose between many small wins or a few large ones, but the average stays where the entry put it.
What about holding until stop or target?
Removing the 8-hour time limit, so every trade stays open until it hits its stop or its target, made the results slightly worse, not better: letting the whole position ride to the second target dropped to about −0.09R a trade, against −0.05R with the time limit on the same trades. Half the trades finished within 3–4 hours anyway; the time exit mostly closes trades that were going nowhere, often at a small profit, before they drift back to the stop.
The real problem: costs against a small edge
Before fees, the strategy's trades made +0.032R each on average. That is a real but tiny edge: the same trades taken in a random direction lost −0.102R after fees, so the strategy picks direction slightly better than a coin flip. Fees and slippage cost 0.094R per trade — three times the edge.
Why so expensive? A round trip of market orders costs about 0.12% of the position. With a stop 1.3% away, that is 0.12 ÷ 1.3 ≈ 0.09 of the risk — 0.09R, on every trade, win or lose. The tighter the stop, the bigger the bite: on a 0.5% stop, the same fees cost 0.24R a trade. Short-timeframe strategies need a much bigger edge than slow ones just to pay their costs.
Result per trade by the strategy's own quality rating
Current exits · after fees
The setups the strategy rated highest did worst — the only group whose range is clearly below zero.
Show the numbers as a table
| Rating | Trades | R per trade | 95% range |
|---|---|---|---|
| low | 865 | −0.026 | −0.12 to +0.08 |
| medium | 451 | −0.042 | −0.15 to +0.07 |
| high | 305 | −0.196 | −0.32 to −0.05 |
One more warning sign: the setups the strategy itself rated "high quality" — trend aligned, at a good level, normal volatility, at least 2R of room — did the worst, at −0.20R a trade. When a rating system's top grade underperforms its bottom grade, the rating is not measuring what it claims to.
What to take from this
- Read win rate together with average win and average loss. A 44% or 52% win rate means nothing on its own. Work out the breakeven win rate from your real average win and loss.
- Breakeven stops and partials don't add money. They change how your results feel — more winners, smaller wins. Use them if they help you stick to a plan, not because they improve the maths.
- Fix the entry before the exit. If changing exits doesn't move your average, the entry has no edge to manage.
- Put your fees in R. Round-trip cost ÷ stop distance = the R you pay on every trade. If that number is close to your expected edge, the strategy cannot work at that stop size.
- Judge a strategy on hundreds of trades, not one day. The bot's first day (−0.21R a trade) and the 4-week replay (−0.06R) tell the same story, but only the replay is evidence.
Limits of this test
Four weeks is short, and most of it was a falling market. The coin list is the top 30 by volume today, so it leans towards coins that are still popular. The replay enters at market at the 15-minute close; the live bot often uses a resting limit order, which would save about 0.03R a trade in fees — not enough to change the conclusion. Open-interest and long/short data only go back about 30 days at Binance, which is why the window is not longer.
Method: 1,621 trades on 30 Binance USDⓈ-M perpetuals, decisions at every 15-minute close from 9 Oct 2026 to 6 Nov 2026, exits on 1-minute candles. Random-direction control: 4 coin-flip twins per trade with the same stop and targets. Confidence ranges from 2,000 bootstrap resamples of trading days. Related: our wider test of 4H chart patterns.