Skip to content
How to evaluate a bot

The 90% win rate trap

Plenty of trading platforms lead with a win rate: 85%, 90%, sometimes higher. It sounds like the most important number and it is close to the least. A high win rate is easy to manufacture while losing money, and the platforms quoting it rarely publish the number that actually matters.

The maths

Expectancy is the number that decides everything

Expectancy = (win rate × average win) − (loss rate × average loss). A strategy is profitable when that figure is positive, and win rate is only one of its four inputs.

90% winners can still lose money

Win nine trades at $10 and lose one at $200, and you are down $110 with a 90% win rate. Nothing about the headline number reveals that, which is precisely why it gets advertised.

40% winners can print money

Lose six trades at $50 and win four at $200, and you are up $500 with a 40% win rate. Several of the strongest systematic strategies look like this: frequently wrong, and very profitable, because the winners are far bigger than the losers.

The mechanics

Four ways a high win rate gets manufactured

Asymmetric exits

Take profit at a fraction of a percent, and either use no stop loss or a very wide one. Almost every trade eventually closes green until the one that does not, and that one takes back everything.

Averaging down and grid systems

Martingale, DCA-down and grid bots add to losing positions until the average closes green. Every closed trade counts as a win. The blowup is rare and account-sized when it comes.

Counting only closed trades

A position sitting 40% underwater is not a loss until it is closed, so a bot holding losers indefinitely can advertise a near-perfect record while the account bleeds.

Choosing the window

Measure a long-only strategy during a bull run and a high win rate falls out for free. The test is how it behaved through a downturn.

What to ask instead

Four questions worth more than any win rate

What is the maximum drawdown?

The worst peak-to-trough fall tells you what holding the strategy actually feels like, and whether you would have stayed with it.

Can I verify the record independently?

Ask how you would check the published results without trusting the company. If there is no answer, the results are marketing. AutoCoin publishes a signed, chained, on-chain-anchored record you can recompute in your own browser.

Is this paper or live, and is it labelled?

Simulated results miss slippage, liquidity and timing. Any honest platform labels paper and live at equal weight rather than blending them.

Are costs included?

Spreads, funding and fees turn many impressive backtests negative. A result that only survives with optimistic fills is not a result.

FAQ

Frequently asked questions

Is a 90% win rate possible in trading?

Yes, and it is usually engineered rather than earned. Taking tiny profits while holding losers indefinitely, or averaging down until positions close green, produces very high win rates while expectancy stays negative. The win rate is real; the implication that it means profitability is not.

What is a good win rate for a trading bot?

There is no good number in isolation. Trend and momentum strategies often win 35% to 45% of trades and profit because winners are multiples of losers. Mean-reversion strategies often win 70% or more and can still lose money if a single loss is large enough. Judge expectancy and drawdown, not the ratio.

How can I tell if a trading bot's results are real?

Check whether you can verify them without the company's cooperation. Look for results published as a signed, tamper-evident record, ideally anchored to a public blockchain, with paper and live labelled at equal weight and costs modelled in. If verification depends on trusting a screenshot, treat the numbers as advertising.

See a record built to be checked

AutoCoin publishes daily results you can independently verify, instead of a flattering ratio.

FinCEN-registered
MSB #31000322777334
US company
AutoCoin LLC, Florida
Non-custodial
Your keys, your funds. Trade-only access.
Verifiable
Daily results hashed, signed, and independently checkable