What is a trading bot drawdown?
Drawdown is the distance from an equity peak to the lowest point that follows, measured in percent. If your bot grows the account to $12,000, then a losing streak drags it down to $9,600, that is a 20% drawdown. It stays "open" until equity prints a new high above $12,000.
Every strategy has drawdowns. A grid bot bleeds when price trends hard against the grid. A momentum bot chops out in a range. A copy-trading portfolio inherits the drawdowns of whoever you follow. The question is never whether you will have one, it is how deep, how often, and whether the bot survives long enough to recover.
How do you read max drawdown and recovery time?
Max drawdown (often written MDD) is the single largest peak-to-trough drop over a test or live period. It answers one question: what is the worst this thing has done? A backtest showing 8% annual return with a 45% max drawdown is not an 8% strategy, it is a strategy that once asked you to sit through nearly half your account gone.
Recovery time, or "time under water," is how long equity stayed below the old peak. Two bots can share a 25% max drawdown, but one claws back in three weeks and the other takes eleven months. The slow one is far more likely to get abandoned mid-recovery, which locks the loss in for real.
- Max drawdown: depth of the worst drop.
- Time under water: duration of that drop until a new high.
- Average drawdown: the typical pain, not just the worst day.
- Calmar ratio: annual return divided by max drawdown, or reward per unit of worst-case pain.
If a vendor or a backtest shows return but hides these, treat the number as marketing. A curve fit on old data usually has a suspiciously shallow drawdown that vanishes in live trading, the classic sign of backtest overfitting.
Why does drawdown recovery get so brutal?
Because losses and gains are not symmetric. A 20% loss does not need a 20% gain to get back, it needs 25%, because you are now growing a smaller base. The deeper you fall, the more the math turns against you.
| Drawdown | Gain needed to recover |
|---|---|
| 10% | 11.1% |
| 20% | 25% |
| 30% | 42.9% |
| 40% | 66.7% |
| 50% | 100% |
| 60% | 150% |
| 70% | 233% |
This is why "it will bounce back" is dangerous past a point. At 50% down you need to double the account just to break even. At 70% you need to more than triple it. Very few bots, or humans, do that before running out of capital or nerve. Keeping drawdowns shallow is not caution for its own sake; it keeps the recovery math achievable.
How deep is too deep? Setting a kill-switch threshold
Decide the number before you deploy, not during the panic. A kill switch is a hard equity level where the bot stops opening new positions, closes what is open, or both. The point is to remove the in-the-moment decision, because in the moment you will always find a reason to hold on.
Rough starting bands, adjust for your own risk tolerance:
- Spot and low-leverage bots (grid, DCA): a 15 to 25% kill line. These bleed slowly, so a moderate threshold gives room without letting a trend run you to zero.
- Leveraged perp bots: 10 to 20%, tighter. With leverage a 20% account drawdown can mean the position is far closer to liquidation than you think, see leverage and trading bots.
- Copy-trading allocation: set a per-trader stop, not just a portfolio one, so a single blowup does not drag the whole book down. More on that in why most copy traders lose money.
Two practical rules. First, base the kill line on your own risk budget, not the strategy's advertised drawdown, since advertised numbers are best case. Second, tie position size to that budget so a normal drawdown never breaches it; the math for that lives in position sizing for bots.
Max drawdown vs realized drawdown: which to watch?
There are two drawdowns and people confuse them constantly. Unrealized (mark-to-market) drawdown is the paper loss on open positions, and it swings with every candle. Realized drawdown is what you actually locked in by closing. A grid or DCA bot can sit on a scary unrealized drawdown that never becomes real if price mean-reverts, while a momentum bot's stops turn drawdown into realized loss fast.
Watch both, but set your kill switch on total equity including unrealized losses. Ignoring open losses is how accounts get liquidated, the position does not care that you were planning to hold. On perp venues, a deep unrealized drawdown plus an adverse funding rate can quietly compound the bleed while you wait.
Common drawdown mistakes that blow up bots
- Averaging down with no floor. Adding to a losing position feels like recovery and is the fastest route to ruin when the trend is real.
- Turning off the stop during a drawdown. If you disable the kill switch the first time it fires, you never had one.
- Judging a bot by return alone. A 200% year with an 80% max drawdown is a coin flip you got lucky on, not an edge.
- Restarting after a loss with more size. Revenge-sizing turns a recoverable dip into a terminal one.
- Trusting a backtest's shallow drawdown. Live slippage, fees (often 2 to 10 bps a fill), and funding make real drawdowns deeper than the curve promised.
None of this guarantees profit. Most retail bots lose money after costs, and a disciplined drawdown rule mainly keeps you in the game long enough to find out whether yours has an edge. For the bigger picture on that, see do trading bots actually work.
Frequently asked questions
What is a good max drawdown for a trading bot?
There is no universal number, but many traders treat anything under 20% as manageable, 20 to 35% as demanding real conviction, and above 40% as a red flag. What matters more is the ratio of return to drawdown: a bot returning 30% a year with a 15% max drawdown is far healthier than one returning 60% with a 55% drawdown.
How long does drawdown recovery take?
It varies from days to over a year, depending on the strategy and how deep the hole is. Because recovery is asymmetric, deeper drawdowns take disproportionately longer. A 10% dip may recover in weeks, while a 50% drawdown needs a 100% gain and can take years, if it happens at all.
What drawdown percentage should trigger a kill switch?
Set it before deploying, based on your risk budget rather than the strategy's best-case numbers. Common bands are 15 to 25% for low-leverage spot bots and 10 to 20% for leveraged perp bots. The exact figure matters less than actually enforcing it; a kill switch you override the first time it fires is not a kill switch.
Is max drawdown the same as risk?
No, but it is one of the clearest windows into it. Max drawdown shows the worst realized pain, while risk also includes leverage, liquidation distance, correlation between positions, and how fees and funding erode returns. Use drawdown alongside position sizing and leverage limits, not as your only risk measure.
Can a bot recover from a 50% drawdown?
Mathematically it must double the account to break even, which is why 50% is often treated as a point of no return for automated systems. Some do recover, but many run out of capital, get liquidated, or are switched off first. A kill switch is meant to stop the bot long before it reaches this depth.