# Surviving a Drawdown
Every trading system hits drawdown. The question isn't if, but when and how deep. Here's how we think about, measure, and survive drawdowns.
What Drawdown Actually Is
Drawdown is the decline from your peak equity to your current equity. If your account was at $100,000 and is now at $93,000, you're in a 7% drawdown.
Drawdown is not a loss. It's a decline from a peak. This distinction matters because:
- A new equity high resets the drawdown clock
- You can have positive absolute returns and still be in drawdown
- The drawdown from peak is what brokers and prop firms care about
Types of Drawdown
Peak-to-Trough Drawdown
The standard measure. From highest equity point to lowest point. Used by prop firms and most platforms.
Example: Account peaks at $107,000. Drops to $97,000. Peak-to-trough = 9.3%.
Current Drawdown
Current decline from the peak. Changes every day.
Example: Same account rebounds from $97,000 to $101,000. Account is still down 5.6% from $107,000 peak. This is the current drawdown.
Intraday Drawdown
The lowest point reached during a trading day, relative to the previous day's close.
Example: Account closed at $100,000. During the day, price drops and the account hits $97,000. That's a 3% intraday drawdown, even if the account recovers to close at $99,500.
Maximum Drawdown (MDD)
The worst peak-to-trough drawdown ever experienced by the system.
Example: In 4 years of operation, the worst the account ever got was 14.2% down from a peak. MDD = 14.2%. This is the number that matters for risk planning.
Why Drawdowns Happen
Statistical Drawdown
Even with a 55% win rate and 2:1 R:R, losing streaks happen.
Expected losing streaks:
- 5 consecutive losses: Expected roughly every 100 trades
- 8 consecutive losses: Expected roughly every 1,000 trades
- 10 consecutive losses: Expected roughly every 5,000 trades
If your average trade risk is 1% and you get 10 consecutive losses, that's a 10% drawdown. Statistically expected over 5,000 trades (about 2 years for a day trading system).
Regime Drawdown
The market changes character. Your model was trained on one regime, and now it's in a different one.
Examples:
- A mean-reversion strategy works in 2023 (range-bound) and fails in 2024 (trend)
- A trend-following strategy works in 2020 (COVID crash/recovery) and fails in 2021 (grinding higher)
- A volatility strategy that works in any regime stops working when the Fed changes policy
Regime drawdowns are dangerous because they're not random — they represent a structural change. The model's edge may be gone until the regime shifts back.
Execution Drawdown
The model is fine. The data is fine. But execution issues cause losses.
Examples:
- The broker changes their fill policy during volatile periods
- The API stops responding, causing missed entries or exits
- Latency issues cause systematic slippage that wasn't in simulation
- NinjaTrader crashes during a key trade
These drawdowns are fixable. They're infrastructure problems, not model problems.
Catastrophic Drawdown
Market events that no model was prepared for.
Examples:
- COVID crash (March 2020): ES circuit breakers, 10% daily moves, halted trading
- Swiss Franc crisis (January 2015): 30% move in minutes
- Flash crash (May 2010): ES dropped 10% in 5 minutes, recovered 10% in 5 minutes
No model that was trained on normal data will handle these events well. The best you can do is have circuit breakers that stop trading during extreme conditions.
Our Drawdown Response Protocol
Tier 1: Normal Fluctuation (0-5% drawdown)
Action: Nothing.
Drawdowns this size are normal statistical noise. The model is working as designed. Intervening at this level means you shouldn't be running a system.
Checklist:
- [ ] Nothing. Go about your day.
Tier 2: Alert (5-10% drawdown)
Action: Reduce position sizes by 25%. Monitor daily.
Checklist:
- [ ] Verify data quality (no feed issues, no data gaps)
- [ ] Verify execution (fills matching expectations)
- [ ] Verify the model is still receiving signals
- [ ] Check that market regimes haven't changed (compare recent volatility/trend to training period)
- [ ] Increase confidence threshold by 10%
At this level, we're looking for execution issues and regime changes. If everything is clean, the drawdown is likely statistical and will revert.
Tier 3: Warning (10-15% drawdown)
Action: Reduce position sizes by 50%. Begin formal review.
Checklist:
- [ ] Full regime analysis: has the market regime changed since the model was trained?
- [ ] Feature analysis: which features are contributing to losses?
- [ ] Walk-forward check: did the model perform in regimes like this during walk-forward testing?
- [ ] Drift analysis: has feature importance distribution changed?
- [ ] The model is reviewed, not changed.
This is the most dangerous psychological zone. The drawdown is large enough to hurt but not large enough to force a stop. The temptation to change something is strongest here. Don't change the model. Gather data. Plan. Execute the plan when the drawdown is over.
Tier 4: Critical (15-20% drawdown)
Action: Stop trading. Full system review.
Checklist:
- [ ] Stop all trading immediately
- [ ] Go flat on all positions
- [ ] Full model audit: retrain from scratch, test against current market conditions
- [ ] Hypothesis: is this a regime shift, overfitting, or data quality issue?
- [ ] If regime shift: wait for regime to change, or retrain on current regime
- [ ] If overfitting: simplify the model, reduce features
- [ ] If data quality: fix data pipeline, retrain
- [ ] Resume only after model passes fresh walk-forward on recent data
Tier 5: System Review (20%+ drawdown)
Action: The system as designed has failed. A Board vote is required to continue.
Checklist:
- [ ] Board convenes (CEO, COO, CTO weighted vote)
- [ ] Full post-mortem: what went wrong and why
- [ ] Decision: retrain, redesign, or shut down
- [ ] If redesign: start from hypothesis, not from the existing model
- [ ] If shut down: document what was learned, archive the model
20% drawdown means something fundamental is broken. Either the model was never valid, or market structure has changed so dramatically that the old model cannot adapt.
The Emotional Reality
Drawdowns feel worse than they are.
A 10% drawdown with 1% risk per trade means you survived 10 consecutive losses. That's statistically expected. But it feels like:
- "The model is broken"
- "I'm an idiot for trusting this"
- "I should have stopped at 5%"
- "This is going to keep going"
**It won't keep going.* Statistically, it won't. The losing streak will end. The model's edge will reassert. The drawdown will recover.
The question is whether you can hold. Most traders can't. They stop at the bottom of the drawdown, right before the recovery. That's the pattern. Recognize it. Prepare for it. Survive it.
The Recovery
After a drawdown recovers, most people feel relief. They shouldn't feel relief — they should feel analysis.
Drawdown recovery is the most important data point for improving your system.
- What caused the drawdown?
- Were there warning signs you missed?
- Did the model behave as expected during the drawdown?
- What would you do differently next time?
Document the answers while the drawdown is fresh. This is the most valuable data your system will produce. Don't waste it by celebrating the recovery.