I've posted something similar to this in the past, so pardon if this feels at all redundant to anyone. . . I feel like my mistake previously was being overly specific and technical, so perhaps this approach will work a bit better in terms of garnering response/discussion. . .
I find most of the traditional statistical measures of trading performance (most specifically automated trading, involving backtests/optimizations, but this is applicable to discretionary trading as well) to be woefully lacking. . I'm curious as to why there's been so little innovation (at least publicly) in this context, for such a long period of time.
An example might be 'drawdown'. . most of us use a 'Max Drawdown' statistic, to help us better ascertain the risks involved in a certain trading strategy or approach. This irks me deeply, it tells us so little of real value. . . for example, if I have two automated trading strategies I'm considering running on a live account, the first with a max drawdown of 11.3%, the second with a max drawdown of 10.5%. . . the second obviously seems superior thus far, even if just a bit. However, lets say, hypothetically, that this second trading strategy had three other separate drawdowns of 9.7%, 9.5%, and 8.5%, in its historical record. . . whereas the first has an equity graph that is extremely stable, and has no other historical drawdown over 5%, with its 11.3% drawdown being near the beginning of the historical record in the backtest, and thus far less relevant than it otherwise could be. . .
This is just an example, but I feel like trusting a single 'max' over an entire historical record is ridiculous, akin to using a blunt club when there are scalpels at the ready.
What if one were to measure the drawdown distance (amount) from EVERY peak, throughout the historical record, and sum all of these together. . . and divide this by a sum of every INCREASE of the former equity high? Wouldn't this be a much, much more accurate measure of actual likely drawdowns, and overall consistency/robustness?
Just thinking aloud here, but I'm anxious to hear thoughts, if any. . . and very much anxious to hear other suggestions of meaningful statistical data points that may be more 'telling' than the existing handful most of us seem to rely on.