The worst monthly drawdown in my portfolio history is 277 USD on the reference balance. The number is correct. I still do not plan with it.
Here is why.
History gave us one order of months: this losing month, then that winning month, in one exact sequence. The market could easily have delivered the same months in a different order. And if the order changes, the worst losing period changes too.
So I changed it. I shuffled the same monthly results into a new random order, and I repeated that 3,000 times. The result is a cloud of drawdowns that could have happened.
In that cloud my real historical number sits at the 14th percentile. Only about one history in seven was kinder than the one I actually lived through. The 95th percentile of the cloud is more than twice as deep as what actually happened.
That deeper number - 609 USD, not 277 - is my planning number. I use it for position sizing, for the "can I sleep through this" test, and for the risk line in my product description. I plan from the tail of the cloud, not from the single path that history happened to produce.
This has an uncomfortable meaning for every backtest you read, including mine. The drawdown in the backtest report is not the risk. It is one sample from a distribution, and usually a friendly one. Ask any equity curve what its brothers and sisters looked like. The answer is always worse than the photo.
The EAs I developed account for this statistical analysis, their backtest reports do not just look at one backtest report. They risk according to 3000 variations of drawdown so that I can sleep well.
