Win rate is not skill
A trading system that never uses a stop and adds to a position when price moves against it will show a win rate of 98–100%. It does so with random entries. The win rate is a property of the exit rule, not of whoever chooses the trades. The cost arrives as a rare loss that takes the whole account, and in the test below most of the profit came from gold rising, not from the system.
A schema first
Take a price that moves like a fair coin: no drift, no costs. Close the trade at +2, or at −100. The chance of reaching +2 first is
and the expected value is exactly zero:
A 98% win rate and a zero expectation are the same fact. The rare loss is fifty times the usual win, and it balances the wins one for one. This is a schema, with one position and a fixed loss limit. The real mechanic below has no loss limit at all.
The mechanic
The systems sold to retail traders in 2026 that I tested work like this: open a position, and if price moves against it by a fixed step, open another; close the whole basket when the average price shows a small profit. There is no stop.
On one-minute gold prices from 2019 to 2025, with fixed-lot additions, the basket win rate was 98.4–100% in all 16 settings tested.
Same machine, random signals
One product published its own settings: enter after an adverse move, no stop, add in fixed steps. I kept those settings and replaced the product's signals with random ones.
- The target was reached in 88.4% of trades with random direction, and in 99.9% with long-only random signals, although some trades waited one to two years.
- Random signals wiped the account 7 times in 7 years.
- In every one of 80 windows of 180 days, the worst floating loss was larger than the figure the product advertised. The median worst floating loss was $2,543.
These tests stop at the end of 2025, so they say nothing about either product's actual 2026 trading. They say that the headline win rate needs no signal to produce.
Where the loss is
The loss does not arrive trade by trade. It arrives all at once.
Under a common compounding rule of 0.01 lot per $1,000 of balance, the price drop that wipes the account does not depend on the balance:
dollars of gold price, about $100 for a $5 step.
I simulated opening an account every Monday under this rule, 312 accounts in total. Depending on the step, 52–97% were still alive after 4 weeks and 0–39% after 52 weeks. A four-week track record from such a system is mostly a record of not yet having met the drop.
One account shows the shape. Opened with $1,000 on 1 January 2019, it grew to $7,300 by 11 March 2020 and was wiped on 12 March 2020. 873 of its 874 baskets closed in profit (99.89%). That is one price path, an illustration rather than a typical result; the typical result is the 312-account figure.
Across all 24 compounding settings, the net result against the total money deposited was negative. The best setting had 10 wipes in 7 years.
"Top up the account when it is under strain" makes this worse. Under the compounding rule with a $1,000 base, no top-ups meant $37,000 deposited and a net loss of $35,279 over 36 wipes. With up to three top-ups: $97,000 deposited, a net loss of $93,711, over 24 wipes. Fewer wipes, a loss three to ten times larger across the settings tested.
Where the profit came from
To be fair to the mechanic: a wide fixed-lot grid (step of $10 or more, no compounding) was net positive from 2019 to 2025 with 0–2 wipes. Gold roughly tripled over that period.
Removing gold's trend shows what that profit was. With the 2019–2025 trend subtracted linearly, the long-only random-signal version reached its target 87.5% of the time instead of 99.9%, was wiped 5 times instead of 0, and its median gain in windows without a wipe fell from $1,231 to $112. The random-direction version lost money with or without the trend.
The profit was being long a rising market. The no-stop structure turned it into a win rate.
Choosing the best setting
One product's site reported results for the best of 250 settings, chosen and measured on the same 180 days. I repeated that procedure across 14 blocks of 180 days. With random-direction signals, the chosen setting's "worst floating loss" was exceeded in the following block 8 times out of 13, and the account was wiped 5 times out of 13. The median result moved from +$80 in the block where the setting was chosen to −$665 in the next.
The long-only version kept gaining in the next block, because of the trend, but its stated worst floating loss was also exceeded 8 times out of 13. The number that fails is the risk figure.
Winners without skill
Selection produces winners by itself.
- Among 10,000 traders with zero skill and a 15% chance of passing each of 20 attempts, about 673 pass at least 6 times.
- If 100 people each call 10 coin flips, the chance that at least one gets 9 or 10 right is 66%.
- The best of 100 zero-edge strategy variants has an expected z-score of 2.51.
These assume independent attempts. They do not say that every winner is lucky; they say that a winner on camera is what zero skill predicts.
What a win rate does not tell you
A win rate says how often trades close in profit. It does not say how large the losses are, how long losing positions are held, or how many settings were tried before the one on screen. For a no-stop system, those are the numbers that matter.
Method: one-minute XAUUSD prices 2019–2025, deterministic simulations with fixed seeds. The products' actual step, target and 2026 trading were not available. Code and outputs are in a private research repository and available on request.