Drawdown

What drawdown is

Take the highest value the account has reached so far, the peak. Drawdown at any moment is how far below that peak the account sits now.

Three related ideas are worth separating:

• Depth. How big the fall was, in percent. A 35 percent maximum drawdown means that at one point the account was 35 percent below its prior high.
• Duration. How long it took to fall, and how long to get back above the old peak. Two strategies with the same depth can differ by years in duration.
• Frequency. How often drawdowns of a given depth occur, which a single maximum figure cannot show.

Drawdown can be measured on closed trades only, or on equity including open floating losses. The second is what the account really did. The first can hide a position that went deep underwater and then recovered.

The asymmetry of recovery

A loss is measured against a larger number than the gain that repairs it. Lose 50 percent of 100 and you hold 50. To get back to 100 you must make 50 on 50, which is 100 percent. The formula is exact:

Calculated by us on 27 Oct 2026, exact. Up to about 20 percent the table is fairly kind. Past 50 percent the gain needed rises steeply, and by 90 percent it needs the account to grow tenfold. Nobody plans for a 90 percent loss, but this is why a limit on drawdown is usually set well before it. The deeper the hole, the less likely a realistic strategy is to climb out.

Why it matters more than the average return

An average return says where the strategy ends up over a very long time if it survives. Drawdown says whether you survive the trip. Three things get in the way:

• Margin and stop out. A deep drawdown on a leveraged account can reach the margin call before it ever recovers.
• Behaviour. People switch off bots in drawdown. A strategy that needs a 40 percent hole to be endured is only as good as the willingness to endure it.
• The arithmetic above. Recovery needs a bigger gain than the loss.

A worked example: the smooth curve and the one big drop

This is a made-up bot of the martingale or grid kind. It starts at $1,000 and makes 0.4 percent on each step for 60 steps in a row. That is the picture a short backtest would show: a smooth line up. Then one event, a losing ladder, takes 40 percent off the account in one step. After that it carries on at 0.4 percent per step.

Calculated by us on 27 Oct 2026, exact for these made-up inputs. The peak is $1,000 x 1.004^60 = $1,271. A 40 percent loss leaves $762. To recover, the account must gain 66.7 percent, and at 0.4 percent a step that takes 128 steps, more than twice as long as the 60 steps it took to build the peak. The drawdown lasts 129 steps from the peak to the new high. The 60 small wins added $271. The one big loss took away $508, about 1.9 times as much.

How it hits bots

• Martingale bots win small and often and lose rarely and hugely. The equity curve is the one above. Their drawdown figure depends heavily on whether the test happened to include the rare event. See martingale.
• Grid bots hold a growing stack in a trend. Closed profit looks steady while the floating loss on the stack grows. A drawdown that counts only closed trades does not see it.
• DCA bots buy as price falls, so drawdown on the whole position can reach very deep levels before the average cost helps, and there is no recovery if the asset does not come back.
• EAs are often reported with a maximum drawdown that came from one test period. Different periods can produce a very different number.

Prop-firm style daily loss rules

Some evaluation or funded-account programmes use drawdown limits as rules: for example, a limit on loss in one day and a limit on total loss from the starting balance. The details differ by programme, so read yours. In general terms, two things catch bots out. The limit is often measured on equity including floating losses, not only on closed trades. And a breach usually ends the account on the spot, even if the position would have recovered later. Take a made-up $10,000 account with a 5 percent daily limit, which is $500. A ladder that is $600 under water at some point in the day has breached it, whatever it closes at that night. A bot that wins small and holds large floating losses is the kind of bot that this kind of rule penalises.

How to read a backtest drawdown

• Depth. Is the maximum drawdown measured on equity or on closed trades? Is it a percent of the peak or of the starting balance?
• Length. How long did the longest drawdown last, peak to new high? Could you run the bot for that long?
• Margin. Did the test include margin, leverage and the stop out? A test with unlimited margin can survive a hole that a real account would not.
• Costs. Spread, commission and slippage deepen every drawdown.
• How long the test ran. A short record may have missed the event. Run a Monte Carlo simulation on the trade list to see deeper drawdowns the shuffled order could have produced.
• Compare with a forward test. Backtest drawdowns are usually shallower than live ones.

What it does not tell you

The maximum drawdown in a record is the worst that happened, not the worst that can happen. A longer record, or a different market, can show a deeper one. It also says nothing about whether the strategy has an edge: a flat line has no drawdown and no profit. Use it alongside expectancy and the margin limits.