A 90% win rate is easier to get than it sounds, and on its own it says very little about whether a strategy makes money. We sorted every rule test on our backtested strategy pages by win rate. Of 84 tests, none won 90% of its trades, 13 won more than 60%, and 8 of those 13 lost money. The one strategy in our data that won every trade it closed, a DCA bot, lost 19.64% of the account.
Every test here ran on Hyperliquid perpetual futures from 2024-08-15 to 2026-09-29, after trading fees, slippage and funding. The backtested pages were run on 2026-11-17, and this post uses those runs.
How to achieve a 90% win rate in trading, and what it costs
Win rate counts how often a trade closes in profit. It says nothing about how much each win makes or how much each loss costs. Two habits push it up:
- Take profit early. A small target is hit often, so most trades close as wins.
- Hold on to losers. A trade that never closes at a loss never counts as a losing trade. A wide stop, or no stop at all, does this.
Both habits raise the win rate and shrink the payoff. A strategy that wins small and loses big can win most of its trades and still lose money. The win rate you need depends on how your average win compares with your average loss, and the win rate calculator works out the break-even point for any risk and reward.
The DCA bot that never closed a losing trade
The DCA bot backtest buys BTC, ETH, SOL, XRP and DOGE when the 14-period RSI on 4h candles is below 40. If the price then falls 3% from its last buy, it buys again, up to 4 more times. It sells when the price is 3% above its average entry. It has no stop loss.
It closed 52 trades and every one of them was a win: a win rate of 100%. Over the same window it lost 19.64% of the account, and its worst drawdown was 49.72%.
The loss sat in the trades it had not closed. A deal only closes at its target, so in a long fall every extra buy fills and the position waits at a loss. At the end of the test the open positions were down by more than all the closed trades had made. Over the whole test, funding on the positions it held cost 8.33% of the account, and trading fees took another 0.59%.
This is a common shape for a very high win rate: many small wins, and the losses kept open until they are too big to ignore.
84 rule tests, sorted by win rate
Each backtested strategy page runs one well-known rule on the 10 most traded Hyperliquid perps, on 4h and 1d candles, long only and long and short. That makes 4 tests for each of 21 rules, 84 tests in all. The chart puts each test's win rate against its total return.
Three things stand out.
No rule test came close to 90%. The highest win rate was 69.23%.
High win rates lost money more often than not. 13 tests won more than 60% of their trades, and 8 of them lost money. Their median return was a loss of 35.98%. All 13 came from 4 rules that buy dips and short rallies: Bollinger Bands, RSI, stochastic and mean reversion. Most of their trades close at a small profit, and the few that keep going the wrong way are the big losses. The clearest case is Bollinger Bands on 1d candles, long and short: it won 65.06% of 279 trades and lost 51.45%.
Low win rates often made the most. 56 tests won less than 40% of their trades, and 40 of those made money, with a median return of 31.65%. The 200 EMA rule on 4h candles, long and short, won 13.83% of 1,499 trades and returned 109.13%. Trend rules like this one take many small losses and make their money on a few long moves.
| Rule | Timeframe | Direction | Win rate | Return | Trades | Max drawdown |
|---|---|---|---|---|---|---|
| Bollinger Bands | 1d | Long only | 69.23% | +15.47% | 130 | −30.73% |
| RSI | 4h | Long only | 67.46% | +23.65% | 335 | −26.33% |
| Mean reversion | 1d | Long only | 67.37% | +19.20% | 95 | −30.39% |
| Stochastic | 1d | Long only | 66.67% | +22.42% | 198 | −37.32% |
| Bollinger Bands | 1d | Long and short | 65.06% | −51.45% | 279 | −54.82% |
| RSI | 4h | Long and short | 63.91% | −40.77% | 691 | −43.85% |
| Bollinger Bands | 4h | Long and short | 63.88% | −63.15% | 1,704 | −65.77% |
| Mean reversion | 1d | Long and short | 63.64% | −47.69% | 208 | −52.74% |
| Stochastic | 1d | Long and short | 63.61% | −35.98% | 369 | −48.81% |
| Bollinger Bands | 4h | Long only | 62.46% | −4.58% | 851 | −25.79% |
| Mean reversion | 4h | Long and short | 61.46% | −54.38% | 1,262 | −57.76% |
| Mean reversion | 4h | Long only | 61.09% | +10.63% | 623 | −22.70% |
| Stochastic | 4h | Long and short | 60.49% | −69.60% | 2,244 | −72.72% |
| Rule | Timeframe | Direction | Win rate | Return | Trades | Max drawdown |
|---|---|---|---|---|---|---|
| 200 EMA | 4h | Long and short | 13.83% | +109.13% | 1,499 | −37.45% |
| 200 EMA | 4h | Long only | 14.96% | +107.13% | 752 | −39.20% |
| Ichimoku Cloud | 1d | Long and short | 23.24% | +105.77% | 350 | −44.84% |
| Momentum (trend following) | 1d | Long only | 31.73% | +99.79% | 385 | −41.19% |
| Ichimoku Cloud | 1d | Long only | 20.28% | +94.49% | 153 | −33.25% |
Single coins tell the same story
Each backtested page also runs its rule on every coin alone. We counted the single-coin tests with at least 30 trades, 295 of them. 43 won more than 60% of their trades, and 41 of those 43 lost money.
A low win rate is no promise either. Of the 194 single-coin tests that won less than 40% of their trades, 65 made money. Win rate just doesn't sort winners from losers.
Where a high win rate did pay
Our study of 30 crypto bot strategies shows the other side. There, 9 strategies won more than 60% of their trades, and none of them lost money. The Regime-Gated Crypto Index won 80.47% of its trades and returned 196.85%. The best return in the study, 408.84%, came from the MACD crossover strategy, which won 60.2%.
Two things set this set apart from the rule tests:
- It is not a random sample. 7 of those 9 strategies are in our strategy library, where we publish strategies that held up in testing. They come with their own sizing and exits, not a bare textbook rule.
- The study counts trades differently. Every order that makes a position smaller counts as a trade, so trimming a winning position back to its target size counts as a win. Strategies that rebalance often, like the index ones, collect many of those.
Low win rates did well here too: of the 6 strategies that won less than 40% of their trades, 5 made money.
| Strategy | Win rate | Return | Profit factor | Trades |
|---|---|---|---|---|
| Regime-Gated Crypto Index (Risk-Managed, Top-15) | 80.47% | +196.85% | 2.30 | 394 |
| Regime Rotation (BTC Trend Switch) | 78.16% | +102.36% | 3.00 | 91 |
| Supertrend Crypto B | 73.81% | +30.69% | 1.44 | 379 |
| AI Trading Bot Strategy (Trend Wrapper) | 70.56% | +213.38% | 1.50 | 1,141 |
| Supertrend Strategy (Crypto Perps) | 67.22% | +105.03% | 1.42 | 679 |
| RSI Oversold-Recovery Strategy (Crypto) | 65.62% | +33.66% | 8.73 | 32 |
| ADX Trend-Strength Strategy (Crypto) | 63.62% | +81.64% | 1.21 | 1,403 |
| Vol Targeting | 63.53% | +85.56% | 2.55 | 339 |
| MACD Crossover Strategy (Crypto) | 60.20% | +408.84% | 1.35 | 2,746 |
Win rate vs risk reward: what to check instead
A high win rate trading strategy can work, as long as its average loss stays small next to its average win. The habits that raise a win rate tend to make the losses bigger, so check these before you trust one:
- Profit factor: gross profit divided by gross loss. Above one means the wins outweigh the losses. The profit factor calculator works it out from your own trade list.
- Return after costs: fees, slippage and, on perps, funding. The DCA bot paid more in funding than it paid in fees.
- Worst drawdown: the biggest drop from a peak. A strategy that wins 100% of its closed trades can still fall 49.72%, as the DCA bot did. The max drawdown calculator shows what a given drop takes to recover.
- Open trades: a strategy with no stop can hold its losses in positions it hasn't closed. Look at the equity curve, not only the closed trades.
- Trade count: a win rate from a handful of trades says little, which is why we only counted single-coin tests with at least 30 trades.
So, does a trading strategy with a 90% win rate work? In our tests the only one that got there did it by never closing a losing trade, and it lost money. Judge a strategy by what it makes after costs and how far it falls on the way, and treat the win rate as a detail.
How we tested
- Data: Hyperliquid perpetual futures from 2024-08-15 to 2026-09-29.
- Costs: 4.5 basis points of trading fee and 4.5 of slippage on every trade, plus Hyperliquid's hourly funding on every open position.
- Rule tests: the 21 rules on our backtested pages, each on the 10 most traded Hyperliquid perps, on 4h and 1d candles, long only and long and short. Each coin trades its own signal with a tenth of the account.
- Win rate: on the backtested pages, the share of round trips (opening a position to closing it) that closed at a profit. In the bot study, every order that makes a position smaller counts as a trade.
- No new runs: every number here comes from the published pages, as run on 2026-11-17. Those pages are re-run when our backtest engine changes, so a page you open later can show different numbers from this post.
A backtest is evidence about one window, not a promise. Every strategy here can lose money, and none of this is investment advice.
Test your own strategy
Keel runs the backtests behind these pages. Describe a strategy in plain words, from the Keel app or an assistant such as Claude or ChatGPT, and Keel backtests it on Hyperliquid data with fees, slippage and funding. You see the win rate next to the return, the drawdown and every trade, and you can paper trade it on live prices before any money is at risk.
See more than the win rate
Backtest your strategy on Hyperliquid data with fees, slippage and funding, and read its win rate next to its return and worst drawdown.
Backtest a strategy free