Drawdowns and Variance
Trading Mastery
9 min read
maxDrawdownRavgDrawdownR
Understand and survive the dark side of trading by learning to navigate drawdowns and the natural variance in any trading system.
Loading
9 min read
Understand and survive the dark side of trading by learning to navigate drawdowns and the natural variance in any trading system.
Every trader loves a winning streak. But what separates winners from long-term losers is how they handle the inevitable drawdown.
Drawdowns are not if — they're when. They test your discipline, mindset, and trust in your system.
In this lesson, we cover:
A drawdown is the drop from your peak equity to a valley before the next new high.
Four distinct numbers — do not collapse them:
If you collapse these four into "I had a bad week," you lose the diagnostic power. Each one points at a different failure mode. (See MaxDD, Calmar, MAR, Ulcer Index for the full reference.)
MaxDD% = (Peak − Trough) / Peak × 100
Worked example: account peaks at $10,000, drops to $8,000, then recovers. MaxDD = (10,000 − 8,000) / 10,000 × 100 = 20%.
A drawdown of d requires a gain of 1 / (1 − d) − 1 to break even. The relationship is not linear, and most traders underestimate it badly:
Recovery math is non-linear: a 50% drawdown demands a 100% gain; a 75% drawdown demands 300%.
| Drawdown | Gain to recover | Practical note |
|---|---|---|
| 10% | 11.1% | Annoying, ignorable |
| 20% | 25.0% | Within most edges |
| 33% | 49.3% | Pause and audit |
| 50% | 100.0% | Career-defining; rebuild at half size |
| 75% | 300.0% | Typically terminal for retail |
This is why position sizing matters more than entry signals. (See Risk Per Trade & Position Sizing for the math.)
Variance is the natural randomness in your trading outcomes — even with a valid edge.
For any strategy with non-zero per-trade standard deviation σ, expected max drawdown scales roughly as σ × √N over N trades. This is mathematics, not bad luck. The corollary: two traders running the same edge with identical risk can experience radically different equity curves purely from path dependence — the order in which winners and losers arrive.
Variance is what causes losing streaks — not bad strategy. (See Aaron Brown, Red-Blooded Risk, and Ralph Vince's work on path-dependent leverage for the underlying mathematics.)
A profitable edge does not win every time. It wins over time.
| Concept | What it measures | Units | When it matters |
|---|---|---|---|
| Drawdown | Peak-to-trough equity decline | % | Position sizing, ruin risk |
| Losing streak | Consecutive losers, count | trades | Psychology, system trust |
| Variance | Spread of trade-by-trade returns | σ² | Edge stability |
| Ulcer Index | RMS depth × duration of drawdown | unitless | Comparing strategies |
During drawdowns, traders often:
This is how temporary loss turns into long-term failure. Kahneman & Tversky's prospect theory: losses are felt roughly 2.25× as intensely as equivalent gains. A 10% drawdown does not feel "small" — it feels like a 22% gain would feel good. That asymmetry, not the math, is what breaks discipline.
Kahneman and Tversky (1979): losses are felt roughly 2.25x as intensely as equivalent gains. A 10% drawdown does not feel like a 10% gain — it hurts more than a 22% gain feels good.
If you've tracked 100+ trades (and if you haven't, measure your edge first), you should know:
The expected longest losing streak ≈ log(N) / −log(1 − p_loss), where N is the number of trades and p_loss is your loss rate. Concretely:
Match your tolerance to your distribution, not to a folk number. When drawdown hits, the question is not "what's wrong?" — it's "is this within the expected range?"
Bootstrap your historical trade list 10,000 times. Pause the strategy only if live drawdown exceeds the 99th percentile of simulated max drawdowns, or if live trade count exceeds 2× the longest historical losing streak. Below those thresholds, you are inside variance — do nothing.
| Drawdown vs backtest distribution | Action | Rationale |
|---|---|---|
| < P50 | None | Within median variance |
| P50–P95 | Review entries, no rule changes | Variance, watch closely |
| > P95 | Halve size | Outside historical norms |
| > P99 | Pause, full forensic review | Edge may be broken or regime changed |
If drawdown psychology threatens to make you abandon the system entirely, a mechanical filter can buy you time:
Caveat: equity-curve filters reduce variance but typically cost 10–20% of expectancy in backtests. They are a discipline tool, not a profit tool.
If you break the rules during a drawdown, the stats no longer apply.
The inverse failure mode is just as real: refusing to stop a strategy whose live behaviour is outside its backtest distribution. The P95/P99 rule above is what tells you which side of the line you are on.
Drawdowns are the cost of doing business — until they aren't. Two failure modes:
The 1% rule plus a hard account-level halt (e.g. −20% equity → stop, review, restart at half size) is what separates the two. A 25% drawdown at 4× leverage is account zero — leverage compounds drawdowns, and the recovery table above stops being a guide and starts being an obituary.
Drawdowns don't just test math. They test belief.
Traders who survive:
(See Mark Douglas, Trading in the Zone, for the canonical treatment of belief and discipline; Kahneman & Tversky 1979 for the prospect-theory foundation of mid-drawdown errors.)
A drawdown is the percentage drop from peak equity to a subsequent valley before a new high. It is calculated as (Peak − Trough) / Peak × 100. The maximum drawdown over a trading record is one of the most important inputs to position sizing and ruin-risk analysis.
Recovery is asymmetric. A drawdown of d requires a gain of 1 / (1 − d) − 1 to break even. So a 20% drawdown needs +25%, a 33% drawdown needs +49.3%, a 50% drawdown needs +100%, and a 75% drawdown needs +300%. This is why avoiding deep drawdowns through position sizing matters more than maximising trade-level returns.
The expected longest losing streak is approximately log(N) / −log(1 − p_loss), where N is the number of trades and p_loss is your historical loss rate. A 50% win-rate system over 100 trades expects 6–7 consecutive losses; a 35% win-rate trend system over 500 trades expects ~12. Calibrate to your own distribution, not to a generic range.
Bootstrap your historical trade list 10,000 times and compare live drawdown to the simulated distribution. Below the 95th percentile, you are inside variance — do nothing. Between P95 and P99, halve size and watch closely. Above P99, pause and run a forensic review — your edge may have decayed or the regime may have changed.
No. Don't adjust rules during a cold streak — once you break the rules, your historical stats no longer apply. Use backtesting and simulation to test changes before going live. The exception is a hard account-level halt (e.g. −20%) that pauses trading entirely for review, which is a sizing rule, not a strategy change.
Related lessons
Drawdowns are not failures of the edge — they are evidence the edge is being measured.
A 30% drawdown costs 43% to recover. A 50% drawdown costs 100%. The job is to make sure the first number never becomes the second — by sizing, not by hoping.