What drawdown control means in practice

Drawdown is a distribution, and a backtest shows one draw from it

A backtest reports a single maximum drawdown, and that number is usually treated as the worst case. It is one observation of a random quantity, and treating it as a bound is the most common error in risk planning.

The reason is that maximum drawdown depends on the order of outcomes as much as on the outcomes themselves. Take a strategy's actual trades, keep every one of them, and reorder them. The total return is unchanged, because addition does not care about order. The maximum drawdown changes considerably, because a run of losses arriving consecutively produces a deeper trough than the same losses spread apart.

Resampling the trade sequence thousands of times produces a distribution of maximum drawdowns. The number that appeared in the backtest sits somewhere in that distribution, and it frequently sits in the favourable part of it. A strategy whose backtest showed a maximum drawdown of ten units may have a median across resamplings of fifteen, and a tail beyond twenty.

The tail of that distribution is the planning figure. A strategy sized so that only the observed maximum drawdown is survivable will breach its limit in the ordinary case where the same trades arrive in a less kind order.

Sizing is the only reliable lever

Once a strategy's rules are fixed, its return and its drawdown are locked together. Risk per trade scales both.

A strategy risking one percent of the account per trade and producing a thirty percent maximum drawdown becomes, at half a percent per trade, a strategy producing roughly a fifteen percent drawdown and roughly half the return. The ratio between them is a property of the strategy. The level is a choice.

This is why the sizing decision has to be made against a drawdown limit rather than against a return target. Choosing the return first and discovering the drawdown afterwards inverts the constraint, and the constraint is the part that ends accounts.

Where a hard external limit exists, such as a maximum loss threshold in a funded account, the arithmetic runs backwards from it. If the threshold is ten percent and the strategy's tail drawdown is twenty-four units of risk, then risk per trade has to be small enough that twenty-four units stays inside ten percent. That gives roughly four tenths of a percent per trade, and the achievable return follows. Sizing above that level is a bet that the tail will not arrive during the period that matters.

Attempts to control drawdown after the fact mostly fail

The appealing alternative to smaller size is a rule that reduces exposure once losses begin. Cut size after a losing day, stop after three consecutive losses, pause until a new equity peak.

Most of these rules test poorly, and the reason is structural. A strategy with positive expectancy has the same expectancy after a loss as before one, unless losses genuinely cluster for an identifiable reason. Reducing size after losses therefore reduces exposure precisely when the expectancy is unchanged, which lowers return without lowering risk proportionally. The equity curve looks calmer in the middle and recovers more slowly, and the ratio of return to drawdown often gets worse.

The rules that survive testing tend to share one feature: they respond to something with a demonstrated relationship to forward performance, rather than to the recent sign of the equity curve. A rule keyed to a measured change in the strategy's own behaviour, held to a threshold set before the test and evaluated on data that took no part in choosing it, can add value. A rule keyed to the last three trades usually does not.

The test that matters is whether the rule improves return divided by maximum drawdown, on data outside the window used to select the rule's parameters. Many drawdown rules improve maximum drawdown alone, which is trivially achievable by trading less, and the ratio reveals whether anything was gained.

What to measure

Four quantities describe drawdown behaviour usefully, and none of them is total return.

Maximum drawdown, in units of risk. Measuring in units of risk rather than in currency separates the strategy from the sizing decision, which allows the two to be reasoned about independently.

Return divided by maximum drawdown. This makes strategies of different aggressiveness comparable. A strategy returning 600 units with a drawdown of 15 is a different proposition from one returning 900 with a drawdown of 60, and the raw returns rank them the wrong way round.

The drawdown distribution from resampling, with its median and its tail. This is the number to plan against.

Time to recovery. Drawdowns cost time as well as capital. A shallow decline that takes fourteen months to recover consumes more of an allocation period than a deeper one that recovers in six weeks, and it is harder to hold through.

Risk of ruin follows from size and edge together

The probability of losing a given fraction of an account depends on the expectancy per trade, the variability of individual trades, and the risk taken per trade. Increasing risk per trade raises return linearly and raises the probability of ruin faster than linearly, because ruin is an absorbing state. An account that reaches its limit stops trading and does not participate in the recovery that the strategy would otherwise have delivered.

This asymmetry is the entire argument for conservative sizing. The strategy does not know or care that the account has a floor. The floor is a feature of the account, and respecting it is a sizing decision made in advance.

The practical form of drawdown control

Stated as a sequence, drawdown control is four decisions taken before any capital is at risk.

1. Establish the drawdown distribution by resampling the trade sequence, and take the tail as the planning figure rather than the observed maximum.
2. Identify the hard limit that must not be breached, whether imposed externally or chosen.
3. Set risk per trade so that the tail figure stays inside the hard limit, and accept the return that follows.
4. Fix that size and change it only for reasons established in advance and tested outside the window that produced them.

Every one of those steps happens before a loss occurs. By the time a drawdown is under way, the decisions that determine whether it is survivable have already been made.