I trade four gold setups regularly. Until this summer, I couldn't have told you the expectancy of any of them individually. I knew the overall figure for gold, and I had a feel for which setups were good. The feel turned out to be partly wrong.
What I did: tagged every XAUUSD ticket since January 2025 by setup, then calculated for each one the win rate, average winner and average loser in R, and expectancy per trade after costs.
The London open range break: 118 trades, 46% win rate, average winner 1.9R, average loser 1.0R. Expectancy around +0.33R. My best setup and the one I'd have guessed.
The round-number fade: 74 trades, 61% win rate, but average winner only 0.8R against average loser 1.1R. Expectancy close to +0.06R. After a bad week, this goes negative. I thought this was my second-best setup. It's not.
The post-US-data pullback: 39 trades, 44% win rate, average winner 2.3R. Expectancy about +0.45R, but small sample and wide variance.
The afternoon trend continuation: 61 trades, slightly negative expectancy. I'd stopped noticing that it wasn't working because the winners were memorable.
What changed: the afternoon setup is suspended. The round-number fade is now half size until it earns its place back. The post-data pullback gets more attention, carefully, because 39 trades is not much.
Win rate was misleading me. Expectancy wasn't.
Advertisement
Desk deep dive: linking my gold setups to expectancy
-
LondonScalper
- Posts: 965
- Joined: Sat Sep 05, 2026 7:54 am
Recommended broker for automated trading & scalping
Re: Desk deep dive: linking my gold setups to expectancy
Useful breakdown, and worth one more column: how sure each number is. The uncertainty on an average is roughly the spread of the individual results divided by the square root of the trade count. For the round-number fade, the results spread by a bit under 1R, so 74 trades gives about plus or minus 0.11R around +0.06R. That's honestly indistinguishable from zero, which fits your point that one bad week tips it negative. The open range break has a wider spread because of the 1.9R winners, around 1.45R, so 118 trades gives about 0.13R of uncertainty around +0.33R, and that one is clearly positive. The 39-trade setup will be near a quarter of an R either way, so it's probably too early to rank it until it's past a hundred.
It’s Fairman 
-
Shadow Trader
- Posts: 959
- Joined: Wed Sep 09, 2026 5:19 pm
Re: Desk deep dive: linking my gold setups to expectancy
LondonScalper, the round-number fade looks to me like a target problem more than an entry problem. A 61 percent win rate with winners at 0.8R suggests the exits sit too close to the level, so before halving it I would check how far price ran in your favour on those 74 trades after you closed.
On my own gold log a similar fade had a median favourable run of about 1.3R, and I was banking 0.7R. Moving the target to 1.1R dropped the win rate to around 52 percent, but expectancy went from roughly zero to about +0.12R.
Costs deserve their own column on gold too. A 25 cent spread plus commission is a bigger share of a tight fade stop than of a range break stop, so the fade pays more in R per trade before it even starts.
On the afternoon continuation, I would split it by whether the US data had already printed. Mine was negative overall but fine after 15:30 London once the data was out, which pointed at timing rather than the setup.
What is the average stop in dollars on each of the four? If the fade is the tightest, that alone explains part of the gap.
On my own gold log a similar fade had a median favourable run of about 1.3R, and I was banking 0.7R. Moving the target to 1.1R dropped the win rate to around 52 percent, but expectancy went from roughly zero to about +0.12R.
Costs deserve their own column on gold too. A 25 cent spread plus commission is a bigger share of a tight fade stop than of a range break stop, so the fade pays more in R per trade before it even starts.
On the afternoon continuation, I would split it by whether the US data had already printed. Mine was negative overall but fine after 15:30 London once the data was out, which pointed at timing rather than the setup.
What is the average stop in dollars on each of the four? If the fade is the tightest, that alone explains part of the gap.