What Is a Good Win Rate for a Trading Bot? The Number Alone Is Not Enough

What win rate actually measures

Win rate is the percentage of completed trades that finish with a profit. If a bot closes 100 trades and 60 are profitable, its win rate is 60%. The calculation is simple, which is why it often becomes the first number shown in a strategy report.

The number says nothing about magnitude. It does not reveal whether the average winning trade earned one dollar or one hundred dollars, and it does not show whether one losing trade could remove the gains from the previous twenty winners.

This makes win rate a useful description of trade frequency, but an incomplete measure of strategy quality. It must be read beside the amounts won and lost.

The break even win rate formula

Expected value connects win rate to money. Let p represent the probability of a winning trade, W the average gain, L the absolute value of the average loss, and C the average total cost per completed trade. The expected result is pW minus (1 minus p)L minus C.

Set that expected result to zero and the break even win rate becomes (L plus C) divided by (W plus L). If average gains and losses already include every cost, the formula simplifies to L divided by (W plus L). A strategy needs a higher rate than this threshold to have positive expectancy under the same assumptions.

The examples below use R as one unit of initial risk. They show why a lower win rate can be sufficient when the average gain is larger, and why realistic costs change the answer.

Why an 80% win rate can still lose money

Imagine a bot that wins eight trades out of ten. Each winner earns 1 unit and each loser loses 5 units. The winners produce 8 units, while the two losses remove 10. Before costs, the strategy has lost 2 units despite an 80% win rate.

This pattern can appear in systems that take small profits quickly but allow losing positions to grow. Averaging into a falling position can also record many small wins while concentrating risk in the rare trade that does not recover.

A high win rate is not automatically suspicious. It simply requires the same scrutiny as any other result. Inspect the largest loss, average loss, exit rule, leverage and behavior during extreme market moves before treating it as evidence of safety.

How a 40% win rate can be profitable

Now consider 100 trades with 40 winners that earn 2 units each and 60 losers that lose 1 unit each. Gross gains equal 80 units and gross losses equal 60, leaving 20 units before costs. The system loses more often than it wins, yet its payoff pattern creates positive expectancy.

Trend following strategies can have this shape because many attempts end with small losses while a smaller number of sustained moves produce larger gains. The experience may be uncomfortable even when the arithmetic is sound. Several losses can occur in sequence.

A suitable strategy therefore needs both positive expectancy and a loss pattern the operator can tolerate. Abandoning a valid process during a normal losing streak can turn a statistical edge into a real loss.

Trading costs move the target

Fees, spread, slippage and funding are paid in money, not percentages on a marketing page. Their effect is especially large when a strategy trades frequently or captures small price movements.

In the equal gain and equal loss example, the break even rate is 50% before costs. If each completed trade costs 0.1R on average, the threshold rises to 55%. A reported 53% rate would look positive under a cost free model and negative under the more realistic assumption.

Costs also vary. A market order placed during a volatility shock may fill farther from the expected price than it did during normal conditions. A credible test uses realistic assumptions and checks whether the result survives worse ones.

Sample size and market variety matter

A win rate from twenty trades is fragile. A few different outcomes can change it sharply. More trades reduce that sensitivity, but trade count alone does not solve every problem. One thousand nearly identical trades from one calm market can provide less useful evidence than a sample that spans advances, declines, ranges and volatility shocks.

The development process matters too. If settings were selected after viewing the entire history, the result may reflect chance rather than a durable relationship. Unseen data, frozen rules and forward observation help separate discovery from evaluation.

No single trade count guarantees reliability. Confidence should rise when the sample is larger, the regimes are more varied, costs are realistic and similar results appear outside the development period.

Read these metrics beside win rate

Average gain and average loss reveal the payoff ratio. Expectancy combines that ratio with win and loss probabilities. Profit factor compares all gross profit with all gross loss. Maximum drawdown describes the largest decline from a previous equity peak.

Trade count and test length show how much evidence supports the numbers. The largest loss can reveal tail risk hidden by averages. Exposure, leverage and recovery time show how much capital and patience the strategy required.

Together these measures answer a better question: did the system produce a worthwhile return for the risk and uncertainty it carried? Win rate alone cannot answer it.

• Average gain divided by average loss
• Expected value per trade after costs
• Profit factor
• Maximum drawdown and recovery time
• Largest loss and longest losing sequence
• Trade count, test period and market regimes

How to read the TensorParadigm simulation

The TensorParadigm historical simulation beginning on 1 January 2026 reported 907 profitable trades from 1,373 completed trades, equal to a 66.06% win rate. The same report showed a profit factor of 3.252, a 79.70% return and a 16.77% maximum drawdown.

The win rate becomes more informative when those additional figures are visible. Profit factor provides context about gross gains relative to gross losses, while drawdown describes part of the risk experienced along the path.

The simulation charged 0.12% commission on each entry and exit plus 0.03% slippage, the same rates as the live trades. The figures still describe one historical simulation, not a promise of future performance.

What our live record shows, and what it hides

Our agents’ live trades show why a win rate needs context. At a snapshot on 26 October 2026, 88 trades had been posted in our public Telegram channel since 6 October. All 78 closed trades were profitable after commission: a 100% win rate on closed trades.

That figure alone would mislead. The other 10 trades were still open, and all of them were below their entry prices, by 0.7% to 11.1% before leverage. The median closed gain was +1.19% before leverage, so the largest open loss was about nine times a typical win.

When winners close quickly and losing positions stay open, the closed-trade win rate can look perfect while real losses are still being carried. The honest measure includes open positions at their current price. Our live case study lists every open trade so readers can check it against the market.

The honest benchmark

There is no universal good win rate for a trading bot. The right threshold comes from the size of the average gain, the size of the average loss and the total cost of trading.

A good reported rate is one that remains above its break even threshold across enough trades and varied conditions, while drawdown and loss sequences stay within defined limits. The number should support an evidence based decision, not replace one.