Performance figures cited below are from the locked five-year backtest. They are not a forecast of any forward performance, and live execution will produce a different trio of numbers. The argument is about how to evaluate any strategy, not a guarantee about ours.
The conventional way to evaluate a trading strategy is to pick the metric that flatters it most and lead with that. A high CAGR ignores the volatility cost of getting there. A high Sharpe ratio ignores whether the strategy actually deployed enough capital to be worth running. A small drawdown ignores whether the strategy did anything at all. None of the three numbers is sufficient by itself. The well-adjusted strategy is the one that improves all three together, and a change that improves one at the cost of another is rejected.
The trinity is the practice of judging a strategy on three numbers at once, compound return, the Sharpe ratio, and maximum drawdown , and accepting a change only when it improves all three, because each number is trivially gameable on its own.
Why any single metric is gameable
Each of the three big numbers has at least one trivial attack that produces a flattering value at the cost of the actual strategy.
- Return alone. Any strategy’s headline return can be inflated by widening the stops, increasing the position size, or running with leverage. The cost is paid in drawdown and Sharpe; the headline looks better. A return number quoted without the drawdown that produced it is a number you cannot evaluate.
- Sharpe alone. Deleveraging doesn’t flatter Sharpe, blend a book with cash and both its excess return and its volatility scale down together, leaving the ratio unchanged. Areported Sharpe is inflated a different way: by manufacturing a small, smooth premium, harvest carry, sell tails, or vol-target into quiet names so realised volatility collapses faster than the edge. Sharpe is blind to skew and penalises upside volatility, so it flatters anything short-vol-shaped until the tail it ignored arrives. The risk hasn’t gone; it moved somewhere Sharpe can’t see.
- Drawdown alone. The smallest drawdown available to any strategy is zero, achieved by never deploying capital. Sit in cash; report a flat curve; drawdown is zero. The number is unimpeachable. The strategy is not a strategy.
Each metric, optimised in isolation, produces a worse product than the joint optimisation. That is not a novel observation. What is striking is how often published strategies lead with one of these numbers and leave the other two off the page entirely. The reader cannot evaluate what they are not shown.
The trinity
Three numbers, reported together, contain almost all of the operationally-relevant information about a strategy. Not because they are individually sacred, they each have known failure modes, but because their joint distribution constrains each one’s manipulability.
Return tells you the strategy did something. A meaningful CAGR is the floor for any further discussion. If return is small, the rest of the metrics are answering the wrong question.
Sharpe tells you the strategy produced that return at a defensible level of volatility. Not small volatility, defensible volatility. A long-bias equity strategy will have meaningful volatility because the underlying instruments are volatile; the Sharpe asks whether the return is reasonable given that floor.
Drawdown tells you the worst the path was. The CAGR is the endpoint; the drawdown is the memory. A strategy with a beautiful endpoint but a thirty-percent drawdown midway will be redeemed out of before it gets to display the endpoint. Drawdown is what determines whether the operator is allowed to keep running the system.
The three together: did the strategy do something (return), did it do it sensibly (Sharpe), did the path get there honestly (drawdown). Any one of them misleading; all three of them together, almost impossible to game.
The locked baseline
Shishin’s published trinity over the five-year backtest:
- CAGR · +137.2%
- Sharpe · 2.44
- Max drawdown · −15.9%
None of those numbers stands alone. The CAGR matters only because the drawdown that produced it stayed contained. The Sharpe matters only because the return is large enough to be worth the volatility. The drawdown matters only because the strategy was actually deployed enough to have meaningful returns to draw down from. The trinity, taken together, describes the strategy. Any one of the three, taken alone, describes only a sliver of it.
What this changes about the design process
Every parameter change, every weight adjustment, every proposed new feature is evaluated against the trinity jointly. A sweep that produces a meaningful CAGR boost but worsens drawdown by enough basis points to matter is rejected even when the headline number looks impressive. A change that improves Sharpe by an academically-interesting fraction but does not improve the CAGR or the drawdown is examined skeptically, it is likely an in-sample artefact. The only changes that ship to production are those that improve, or at worst leave unchanged, all three numbers at once.
This is harder than it sounds. There are many changes that improve the headline CAGR. There are also many changes that improve the drawdown. The intersection of “changes that improve all three together” is a small set, and most candidate parameter movements do not land in it. The design loop is mostly an exercise in finding the rare changes that do, while rejecting the larger population of changes that flatter one number at the cost of the others.
The trade-off discipline is uncomfortable. It is the reason the published backtest moves slowly. It is also the reason the strategy that finally shipped does not have a hidden weakness on the dimension we were not looking at.
What we don’t do
We do not aggregate the three into a single composite score (MAR ratio, Calmar ratio, Sharpe×CAGR, whatever is fashionable). The temptation is real, a single score is easier to optimise against, easier to report in a marketing deck, easier to plot on a chart. The cost of the single score is that it hides the trade-off the trinity is supposed to surface. A composite metric tells you the strategy looks good. It does not tell you which of the three components is carrying the result and which one is being quietly traded against to produce it.
We report the three numbers. The reader does the trade-off mathematics. That is the institutional convention because it is the convention that survives contact with allocators who know what they are evaluating.
What this means for any backtest you read
Three questions to ask of any published track record, anywhere on the open web. They take ten seconds. They will tell you most of what you need to know about whether the author actually evaluates their own work or just markets it.
- Is the CAGR published alongside the max drawdown? If not, the CAGR is uninterpretable. A high CAGR can be levered up, but only so far: past the growth-optimal point, more leveragelowers it and eventually wipes the account out. The question is whether the path got there inside an envelope the operator could actually live in.
- Is the Sharpe ratio quoted alongside the return? If only the Sharpe is reported and not the return, the strategy might be a low-deployment curiosity. If only the return is reported and not the Sharpe, the strategy might be a high-leverage gamble that happened to work in the sample period.
- Is the worst drawdown reported across the full backtest window, not just the most recent year? Drawdowns shown only in calm windows are not drawdowns. They are the absence of drawdowns. The thing you want to know is what the strategy did in the worst window it lived through, not the calmest.
Any one of those three answered with anything other than yes is a flag. Not a deal-breaker, everyone publishes selectively from time to time, but a signal that the author has chosen which metric to market on rather than letting all three speak for themselves.
The well-adjusted strategy is the one that does not have to choose. The trinity is honest because it cannot be optimised for one face at the expense of the others without the other faces showing it. That is the only reason to publish all three.
Sources & further reading
- Sharpe, W. F. (1994). “The Sharpe Ratio.” Journal of Portfolio Management, 21(1), 49 to 58.
- Magdon-Ismail, M. & Atiya, A. F. (2004). “Maximum Drawdown.” Risk, 17(10), 99 to 102.