Evaluation pass odds from a simulation: 45% winners at 1.5R on a $100k challenge at four risk levels

[Fairman]

Evaluation pass odds from a simulation: 45% winners at 1.5R on a $100k challenge at four risk levels

I simulated a $100,000 challenge with rules that are typical but not any one firm's: 10% profit target, 10% static max loss, 5% daily loss, 30 trading days, 3 trades a day, independent trades, no costs. 20,000 runs per row. The system wins 45% of the time at 1.5R, an expectancy of +0.125R before costs.

Risk per trade, then pass, breach, time out:
0.25%: 1% pass, 0% breach, 99% time out
0.5%: 33%, 1%, 66%
0.75%: 57%, 5.5%, 37%
1.0%: 70%, 13%, 17%

The 0.25% row is the odd one. Ninety trades at 0.125R and 0.25% each is an expected gain of about 2.8%, nowhere near 10%, so it's safe and it fails. More risk gets you to the target in time, and also closer to the floor.

A coin flip system, 50% at 1R with zero expectancy and no costs, at 1% risk: about 30% pass, 29% breach, 41% time out. Add a cost of 0.1R per trade and that becomes 7% pass and 60% breach.

So a pass doesn't prove an edge, because luck at 1% risk passes about three in ten attempts, and a tiny cost turns that into a firm's profit. Simplified assumptions, my own code, illustrative only.

[FXS HongKong Scalper]

The 0.25% row is a nice illustration that being too careful can fail you just as surely as being reckless, only more slowly. What I would love to see added is a column with realistic costs, maybe 0.1R per trade for a scalper on a raw account. With a gross edge of about an eighth of R, that leaves almost nothing, and I suspect the pass rates at every risk level would drop sharply while time outs dominate. It would show how much of the challenge outcome depends on costs rather than on the risk setting.

[Fairman]

A slightly better reward to risk changes the picture a lot. Same simulation with 45% winners at 1.7R, an expectancy of +0.215R. At 0.5% risk: 57.5% pass, 0.3% breach, 42% time out. At 1% risk: 85% pass, 6.5% breach, 8% time out. Going from 1.5R to 1.7R at 1% risk lifts the pass rate from 70% to 85% and halves the breach rate from 13% to 6.5%. A small change in reward to risk moves the result a lot, which shows how much an evaluation rewards a real edge and how little it forgives a zero one.

[Fairman]

What if there's no time limit? At 0.25% risk, each trade contributes 0.125R times 0.25%, which is 0.031% on average. The 10% target needs about 320 trades, around 107 trading days at 3 trades a day, ignoring variance. So with an open ended evaluation, small risk can work, just slowly, and each extra month of waiting costs a monthly fee on some platforms. With a 30 day limit, as in the simulation, the same risk level has no chance. It's why the time limit matters as much as the drawdown limit when you choose a risk level, and why firms that drop the time limit change what sizing makes sense.

[Fairman]

In this simulation the 5% daily limit almost never matters at 1% risk, because three losers in a day lose 3%, which is under 5%. The static 10% is the one that does the damage. Raise the risk to 2% and three losers in a day lose 6%, so the daily limit becomes live. The point is that two limits set different ceilings on risk per trade, and the lower one wins. Work out how many consecutive losses each limit allows at your size, and compare: at 1% it's 5 losses in a day for the daily rule and 10 in total for the static one, so the two interact in a way that depends on trades per day.

[Fairman]

Re-run it with your own figures instead of trusting mine. Win rate is the most sensitive input. Same 1.5R system at 1% risk with different win rates: 42% passes 50% of the time and breaches 26%. 45% passes 70% and breaches 13%. 48% passes 85% and breaches 5.5%. Three points of win rate swings the outcome from a coin flip to a likely pass. So a win rate estimated from 40 trades, which has an uncertainty of several points either way, is too rough an input to plan around. Use your last 100 or more trades, and run the sizes against the low end of what the data allows.