Picking the strategy with the best return-to-drawdown ratio
Return persists weakly out of sample and the drawdown ratio does not, so rank on return and correlation and let the portfolio shape the drawdown.
- Verdict
- Settled by the data
- Decided by
- No out-of-sample content
- The number
- Rank persistence of $/DD: −0.019
The hypothesis, as pre-registered
Return per unit of drawdown is the statistic that matters to whoever has to hold the position, so a shortlist should be ranked on it.
What the test found
Across 64 strategy streams over 944 sessions cut into six contiguous blocks, the rank correlation of return-per-drawdown between one block and another averaged −0.019 over all fifteen disjoint pairs, positive in six of them. That is zero. The same measurement on dollars per year gives +0.136 — weak, but real and repeatable.
The single train-versus-holdout cut is more vivid than the average. The three best full-sample drawdown ratios ranked 28th, 38th and 48th of 64 on the training period. The training leaders inverted: one fell from 14.49 to 2.54.
It was confirmed a second time on an unrelated set of candidates: a median training ratio of 5.55 became 1.20 out of sample, rank correlation −0.071, and five of seven candidates reading above 5.0 in training collapsed. A screen asked to deliver five strategies above a drawdown-ratio threshold would have delivered exactly that, and the number would have meant nothing.
Bootstrapping the drawdown before ranking does not rescue it, which was the obvious repair and was tested at up to 4,000 draws. The problem is not the estimator. The realised drawdown path is itself a single draw, and no resampling of one path recovers the distribution the next one comes from.
What would make this work
Rank on the return, which is weakly forecastable, and on correlation against what is already running, which is far more stable — then let the portfolio produce the shallow drawdown. The ratio is a property of a book, not of a single strategy. Report the drawdown always; just never sort on it.
Every insight ends with this line: the exact condition that would turn the finding into a tradeable edge. That is what makes each entry a next experiment rather than a full stop.
Screened September 4, 2026
Run the arithmetic on your own numbers with the deflated Sharpe ratio calculator.
Testing something similar?
The screener puts your version of this idea through the same gate stack, on your market, and answers in seconds. Your first verdict is free.
Related insights
An edge that only exists in the last three years
Ninety-eight percent of a six-stream book's lifetime profit sat after 2022, which is why every result is now split by era before it ships.
SettledA strategy that only works on the 30-minute chart
Hold the entry instant and the stop distance fixed and an 18x span of bar sizes pays the same, which makes the strategy far more robust than it first appeared.