Win rate tells you how often a strategy closes in the green. Average profit per trade tells you whether each trade was worth taking. A strategy with a 70% win rate can still lose money, and a strategy with a 35% win rate can compound steadily. The question that actually matters is the product of the two: expectancy. Test any candidate strategy on app.gt-protocol.io and you will see both numbers side by side, because in isolation neither is enough.
Most retail traders fixate on win rate because it feels intuitive. Seven wins out of ten reads as competence. Three wins out of ten reads as failure. That reading is wrong often enough to matter. This article shows what each metric captures, where it lies to you, and how to combine the two so a backtest result actually predicts live performance.
What does win rate tell you about a strategy?
Win rate is the percentage of closed trades that finished profitable. A strategy that closes 100 deals with 62 in the green and 38 in the red has a 62% win rate. The metric measures one thing only: how often the entry-and-exit logic is correct. It says nothing about how much money you make when right, how much you lose when wrong, or how much capital each trade tied up. Win rate is psychologically dominant because frequent small wins are easy to live with — drawdowns hurt less when they arrive on a 1-in-3 cadence instead of a 1-in-10 cadence. That comfort is real and worth something, but comfort is not edge. A strategy can have a 90% win rate and bleed capital if the 10% of losing trades are each five times larger than a typical win.
The standard formula:
Win rate = winning trades / total closed trades × 100%
Common pitfalls when reading this number:
- Sample size. A 75% win rate over 20 trades means very little. The same number over 500 trades means a lot. The law of large numbers only kicks in once you have enough trades for noise to cancel.
- Hidden survivorship. If unclosed losing positions are not counted, win rate is overstated. A bot that holds a 30% drawdown indefinitely is not winning — it just has not closed.
- Fee blindness. A "winning" trade that nets +0.05% after a 0.1% round-trip fee is actually a loss. Win rate calculated on gross PnL hides this.
What does average profit per trade tell you?
Average profit per trade is the net dollar (or percentage) result of a strategy divided by the number of closed trades. It answers a different question: when this strategy fires, is the outcome worth the capital lockup, the slippage, and the risk? A scalping bot might average $1.20 per trade over thousands of trades and compound nicely. A breakout bot might average $40 per trade with far fewer fires. Both can be valid. What kills a strategy is when the average drifts toward zero or negative — at that point the wins, however frequent, no longer cover the losses plus the cost of doing business. Average profit per trade is also what determines whether a strategy survives a regime change. When volatility compresses or fees rise, the strategies that die first are the ones whose average was always borderline.
The formula:
Average profit per trade = total net PnL / number of closed trades
Two things to watch for:
- Mean vs median. One huge outlier trade can inflate the mean while the typical trade still loses money. Always check the median next to the mean. If they diverge by more than 2×, the strategy is being carried by a small number of lucky fills.
- Percentage vs absolute. Average profit in dollars depends on position size. Average profit in percentage of capital deployed is what compares across strategies and account sizes.
Why neither metric works alone
The mistake that wipes out the most accounts is treating either number as a verdict. A 70% win rate with an average win of $10 and an average loss of $35 is a losing strategy: over 100 trades the math gives 70 × 10 − 30 × 35 = 700 − 1,050 = −$350. A 35% win rate with an average win of $80 and an average loss of $25 is a winning strategy: 35 × 80 − 65 × 25 = 2,800 − 1,625 = +$1,175. The trader who picks by win rate alone takes the first one home. The trader who picks by average trade size alone takes a strategy that wins big once a month and bleeds the rest of the time. The only number that fuses both honestly is expectancy, and that is what serious backtesting reports first.
Expectancy formalized:
Expectancy = (win rate × average win) − (loss rate × average loss)
This is the standard metric in systematic trading literature — see Van Tharp's expectancy framework, which has been the reference for two decades. Expectancy expressed in R-multiples (units of initial risk) lets you compare a scalping bot to a swing bot to a portfolio strategy on the same axis.
Quick reference table
| Strategy profile | Win rate | Avg win | Avg loss | Expectancy / trade |
|---|---|---|---|---|
| High-frequency scalper | 72% | $8 | $15 | +$1.56 |
| Trend follower | 38% | $120 | $40 | +$20.80 |
| Mean reversion grid | 85% | $5 | $60 | −$4.75 |
| Breakout swing | 45% | $95 | $55 | +$12.50 |
The mean reversion grid in row three is the classic trap: a comfortable 85% win rate hiding negative expectancy. Most accounts die here.
How GT's backtest engine treats the two metrics
When the GT AI builds a strategy from a natural-language prompt, it does not show the result to the user until a backtest has run. The acceptance gate is intentionally two-sided: the engine rejects any candidate whose total PnL is negative or whose win rate falls below 50%, and it iterates up to three times before giving up. The PnL check enforces average-profit-per-trade discipline. The win rate floor enforces consistency. Either condition by itself would let bad strategies through — a strategy with one huge winning trade and many small losses could clear the PnL bar with a 20% win rate, and a high-win-rate grid bot could clear the win rate bar while bleeding capital overall. Requiring both turns out to be a simple, effective filter.
The same logic should guide how you read any backtest report, whether produced by GT or by another tool. If only one of the two numbers passes, treat the result as uncomfirmed. Look for the second number, look at the sample size, and look at the distribution shape.
Reading a backtest report without fooling yourself
A trustworthy backtest report contains at least seven fields, and a strategy needs to pass all of them — not just the two headline metrics. Number of closed trades (under 100 is anecdote, not data). Win rate and median trade outcome side by side. Average win and average loss, separately, so you can compute the reward-to-risk ratio. Maximum drawdown in percent of capital, because a strategy that returns 40% per year but draws down 60% mid-year is not deployable. Profit factor (gross profits ÷ gross losses), which catches strategies with one freak winner carrying everything. Time in market, because a strategy that is only active 4% of the time and matches a buy-and-hold return is not adding edge.
If the report only shows you win rate and total return, it is selling you something. The fields it omits are the ones that would have killed the marketing pitch.
Frequently Asked Questions
What is a good win rate for a crypto trading bot?
There is no single answer because it depends on the reward-to-risk ratio. A trend-following bot can be profitable at 35-40%. A mean-reversion bot usually needs 65%+ to overcome its asymmetric loss profile. The right question is not "is the win rate high" but "is the win rate high enough given how much I lose when I am wrong."
Can a strategy with a low win rate still be profitable?
Yes. Trend-following systems historically run at 30-45% win rate and remain profitable because winning trades are several times larger than losing ones. The defining trait is positive expectancy, not high accuracy. A strategy that is right 30% of the time but wins 4x what it loses has the same expectancy as one that is right 70% of the time and wins 0.4x what it loses.
What is expectancy and how do I calculate it?
Expectancy is the average amount you expect to win or lose per trade once you account for both win rate and the size of wins versus losses. The formula is (win rate × average win) − (loss rate × average loss). Positive expectancy means the strategy makes money over enough trades. Zero or negative expectancy means it does not, regardless of how comfortable the win rate looks.
How many trades should a backtest include before I trust the win rate?
At minimum 100 closed trades, and 500+ is much better. Below 100 trades, statistical noise dominates and the reported win rate can be off by 10-15 percentage points from the strategy's true value. The same applies to live trading: do not draw conclusions from your first 20 trades on a new bot.
Does fee inclusion change which metric matters more?
Fees compress average profit per trade without touching win rate directly. High-frequency strategies are hit hardest because they trade often and each trade pays the spread plus the maker/taker fee. After fees, many high-win-rate scalping strategies that look profitable on paper turn negative. Always compute both metrics on net PnL, not gross.
Is average profit per trade the same as profit factor?
No. Average profit per trade is total net PnL divided by trade count. Profit factor is gross profit divided by gross loss. The two answer different questions: average profit per trade tells you the typical outcome, profit factor tells you the ratio of money made to money lost. A strategy with profit factor below 1.0 is losing money regardless of its win rate.
How does GT App handle these metrics when proposing a strategy?
The GT AI runs an internal backtest before showing any strategy to the user. It rejects strategies where total PnL is negative or where win rate falls below 50%, and retries up to three times with different parameters. Both filters are applied together, so a candidate that passes only one is discarded.
The takeaway
Win rate alone is a comfort number. Average profit per trade alone is a wishful number. Together — and combined into expectancy — they are the start of an honest answer. The next time you evaluate a bot, a signal service, or a backtest report, write both numbers down. Compute expectancy in your head. If the strategy clears that bar with a sample size above 100 trades and a drawdown you can stomach, it deserves a closer look. If it only clears one of the two bars, you have found a marketing artifact, not an edge. Open app.gt-protocol.io, run a strategy through the backtest, and read both numbers before you commit capital.
