What is a normal maximum drawdown for a trading strategy?

There is no universal figure, because drawdown scales with position size and with the strategy's style. What is more useful than a benchmark is the ratio between return and drawdown, and the honest comparison of your own backtested figure against a resampled distribution of the same trades — which usually shows the historical number sitting well inside a much wider range.

Is drawdown duration more important than depth?

For most traders it is the one that ends the experiment. A deep, fast drawdown is frightening; a shallow one lasting a year is corrosive, because it removes the evidence that the strategy works for long enough that abandoning it starts to look rational. Both should be reported, and duration is the one usually left out.

Should I stop trading a strategy that hits a new maximum drawdown?

A new maximum is expected eventually — over a long enough run, the previous worst case is always exceeded. That makes it a poor stopping rule on its own. A more defensible approach is to decide in advance what would constitute evidence the edge has changed, in terms of trade count and expectancy, and to write it down before the drawdown makes the decision feel urgent.

Maximum Drawdown: How Much Is Too Much?

What is maximum drawdown?

Maximum drawdown is the largest fall from an equity peak to a subsequent trough, expressed as a percentage of the peak. It answers the question every trader eventually asks in the worst possible circumstances: how bad did it get?

It is the most psychologically honest metric available, because it is the number you actually experience. Annual return is an abstraction; a 34% drawdown is eleven weeks of watching an account shrink while wondering whether the strategy has stopped working.

There is a second figure that matters as much and is quoted far less: drawdown DURATION. A 20% fall that recovers in three weeks and a 20% fall that takes fourteen months are the same headline number and completely different experiences, and the long one is what makes people abandon a working system.

Both are properties of a specific historical path, which is the limitation worth internalising before treating either as a bound.

Why is the backtested drawdown always too small?

The historical maximum drawdown is the worst thing that happened in one sample, and there is no reason the future is bounded by it — three separate effects push the real figure higher.

It is a maximum, not an average. Given a longer sample, the maximum can only stay the same or grow. A strategy tested over five years has had five years of chances to produce its worst run; over twenty it gets four times as many.

Execution is worse live. Slippage, widened spreads during the exact volatile periods when a strategy is already struggling, and fills that do not happen all compound during a drawdown rather than averaging out across it.

The strategy may be partly fitted. To whatever extent parameters were chosen on this history, the historical drawdown was minimised by the same process that maximised the return. The out-of-sample drawdown is the one to look at, and it is usually deeper.

The practical rule most systematic traders use is to plan for something meaningfully worse than the backtest showed, and to size accordingly rather than hope.

How do you estimate a drawdown you have not seen yet?

Resample the trade sequence. The order in which wins and losses arrived is one draw from a distribution, and reshuffling it thousands of times produces a range of drawdowns the same strategy could plausibly have produced.

This is the most useful thing a Monte Carlo does, and it is genuinely sobering the first time. A strategy whose historical worst run was 18% will routinely show simulated runs past 30% simply from a less lucky ordering of the same trades.

The technique has an important caveat: reshuffling assumes trades are independent. If your strategy clusters losses — most trend-following systems do, because they lose during ranges — then shuffling breaks that structure and understates the risk of a long bad patch. Block resampling, which keeps runs of consecutive trades together, is the honest refinement.

Neither version tells you what will happen. Both tell you that the single historical number is one draw from a wide distribution, which is the thing worth knowing before sizing a position around it.

How much drawdown should you accept?

Size the strategy so its plausible worst case is a loss you will still be trading through, because a drawdown you abandon at the bottom converts a temporary fall into a permanent one.

That reframing matters more than any specific percentage. The question is not what drawdown is statistically acceptable; it is what drawdown you will hold through without intervening. Those are different numbers, and the second one is usually much smaller than people estimate in advance.

The arithmetic of recovery is worth knowing while making that decision. A 20% drawdown needs a 25% gain to recover, 33% needs 50%, and 50% needs a double. The cost of a deep drawdown is not linear, and the strategies that produce them need proportionally more edge to justify themselves.

Position size is the lever. Halving the size roughly halves the drawdown and roughly halves the return, which is a trade many traders should take and few do — the account that survives a bad year at half size is still there for the good one.