Win rate vs. risk-reward: what 57,819 random entries on gold teach
"What's your win rate?" is often the first question traders ask each other. On its own it is the wrong question. A test on ten years of gold shows why.
The test
We used exactly the rule from the game's TP/SL mode:
- Enter at the close of a 1-hour candle.
- Place the stop 1 × ATR(14) away and the target 1, 1.5, 2 or 3 × ATR away.
- Whichever level is touched first decides the trade. After 50 candles the trade closes at the market.
- If one candle touches both levels, the stop counts.
We did this at every 1-hour candle from October 2016 to August 2026, once as a buy and once as a sell. No chart reading was involved, so this is what pure chance looks like. Spreads and fees are left out.
| Stop : target | Win rate needed to break even | Buy: target hit first | Buy: average R | Sell: target hit first | Sell: average R |
|---|---|---|---|---|---|
| 1 : 1 | 50.0% | 49.9% | −0.001 | 48.1% | −0.040 |
| 1 : 1.5 | 40.0% | 40.5% | 0.012 | 38.4% | −0.040 |
| 1 : 2 | 33.3% | 34.3% | 0.029 | 31.9% | −0.044 |
| 1 : 3 | 25.0% | 25.9% | 0.053 | 23.5% | −0.049 |
| An entry at the close of every 1-hour candle, Oct 2016 – Aug 2026 (57,819 entries each way). Stop = 1 × ATR(14), target = 1, 1.5, 2 or 3 × ATR, maximum 50 candles. If the stop and target are both touched in one candle, the stop counts. No spread or fees. | |||||
What the table says
Win rates fall as targets grow. Random buys won 49.9% of the time with a 1:1 target, 40.5% at 1:1.5, 34.3% at 1:2 and 25.9% at 1:3.
Each one sits right on break-even. If the target is k times the stop, you need a win rate of 1 ÷ (1 + k) to break even: 50%, 40%, 33.3% and 25%. The random results are within about two percentage points of those lines. A bigger target does not create an edge. It only trades win rate for payoff.
Buys did slightly better than sells, averaging between 0.00R and +0.05R per trade against −0.04R to −0.05R for sells. That gap is not skill. Gold rose from about $1,300 to over $4,300 during the period, and random buys rode that drift.
Costs tip the balance. A spread costs a fraction of R on every trade. Add it, and the random results turn negative in both directions.
Why win rate alone misleads
What counts is expectancy, the average result per trade in R:
Expectancy = win rate × k − (1 − win rate) × 1
- 70% wins with a target of 0.3R: 0.70 × 0.3 − 0.30 = −0.09R per trade. It feels great and still loses.
- 40% wins with a target of 2R: 0.40 × 2 − 0.60 = +0.20R per trade. It feels like losing most of the time, yet it wins.
So whenever you see a win rate, ask what the payoff was. And when you see a big payoff, ask how often it happened.
How to use this in the game
- In close-after-N mode, 50% is the baseline. Holding 55% over a few hundred calls is meaningful.
- In TP/SL mode, compare your win rate with the break-even column, not with 50%. At 1:2, a 38% win rate is good and 45% would be excellent. Total R in the record panel is the number to watch.
- Mind the sample size. Over 100 calls, chance alone moves a win rate by about ±10 percentage points. Judge yourself after 300 or more.
- Expect small edges. Real edges are usually a few points above break-even, not 20. A few points, held steadily over many calls, is the goal.