forex-scalping.com

How deep a drawdown should you expect from a system that really works? Simulated: 45% winners at 1.5R

System: 45% winners, 1.5R target, 1R stop. Expectancy is +0.125R per trade. I ran 20,000 simulated runs of 200 trades each, fixed risk, no costs, independent trades.

Final result after 200 trades: median +25R, 10th percentile +2.5R, 90th percentile +47.5R. About 6.5% of runs end negative.

Deepest peak to trough drawdown in a run: median 12R, 90th percentile 20R. In 74% of runs the drawdown reaches 10R or more at some point. In 54% it reaches 12R, and in 31% it reaches 15R.

Longest losing streak: median 8 trades, 90th percentile 11.

At 1% risk per trade, a 12R drawdown is 12% of the account. That's a system with a genuine edge by construction, with a typical drawdown about half the size of its typical total gain.

Many traders stop a system at a 10% drawdown, taking it as proof that it's broken. In this simulation, a working system has a roughly three in four chance of passing through 10% at some point in 200 trades. A drawdown of that size isn't, by itself, information about whether the edge is real.

Simplified model, my own code, illustrative.

Numbers like these should be shown to everyone before they judge a strategy. A 12R median drawdown on a working system means that at 1% risk a trader should expect a 12% dip at some point, and a 20% dip is not unusual. Most people would abandon the system long before then. It also explains why so many prop challenges fail with genuinely positive systems: a 10% max loss sits right inside the normal drawdown range at 1% risk.

Translate the losing streaks into account terms. At 1% risk, 8 losses in a row, the median longest streak, take the account down by 7.7% (0.99 to the eighth power is 0.9227). 11 losses, the 90th percentile, take it down 10.5%. These are streaks that a profitable system produces on its own. A trader who sees 8 in a row and decides the system has stopped working would have abandoned it at about the point a working system is most likely to be sitting at its normal worst. It helps to know the expected streak before it arrives, rather than after.

The spread of final results is the other half of it. One run in ten ends 200 trades with less than +2.5R, which at 1% is under 2.5% for the whole period, and about 6.5% end negative. Meanwhile the top tenth reach +47.5R. Same system, same edge, same rules. The gap is only luck, and it shows how much of a short track record is noise. Someone with 200 trades and +40R and someone with +3R might be running exactly the same strategy. Comparing them as though one found something better and the other didn't would be an error.

Use the 90th percentile drawdown to choose the risk. If 20R is a bad but plausible drawdown and you can tolerate 10% at most, then risk per trade should be 10% divided by 20, which is 0.5%. At 1% the same bad case is 20%, and at 2% it's 40%. The numbers come from this one system, so substitute your own win rate and reward to risk. The principle stands: pick the drawdown you can live with in the bad case, divide by the bad case in R, and that's the risk per trade, not a number chosen because it sounds usual.

Costs deepen all of this. Add 0.1R of spread and commission to each trade and the expectancy falls from 0.125R to 0.025R. Over 200 trades the expected final result drops by 20R, from +25R to about +5R, and I'd expect the drawdowns to get deeper since there's less drift to climb out with, though I haven't re-run it. So a drawdown of 12R against an expected gain of 5R is a poor ratio. A thin edge means long periods under water. It's another reason to measure costs exactly: a tenth of an R is 80% of the pre cost edge of 0.125R in this example.