AI Risk Management for Traders: The Complete Guide
Position sizing, drawdown control, Kelly criterion, and margin-of-safety adjustments — the full risk framework, with copy-paste AI prompts.
Risk management is where edge gets preserved or destroyed. A mediocre strategy with disciplined sizing compounds. A great strategy with sloppy sizing blows up.
Most retail traders treat risk management as an afterthought — “I'll never risk more than 2%” — without understanding why, without adjusting for their setup's actual edge, and without building drawdown circuit-breakers.
This guide is the full framework. Three layers, with copy-paste AI prompts that produce concrete numbers, not platitudes.
Why fixed-fractional sizing is incomplete
“Risk 1-2% per trade” is the most repeated rule in retail trading. It's also one of the laziest. Here's why:
- It ignores edge variance. A setup with 65% win rate and 1:2 R/R deserves a much larger position than a setup with 50% win rate and 1:1.5 R/R. Fixed 1% sizes both identically.
- It ignores correlation. Sizing 5 tech longs at 1% each isn't 5% risk — it's ~3-4% effective risk during normal markets and 5%+ during a tech sell-off.
- It ignores drawdown geometry. Risking 2% per trade with 70% win rate has a different tail-risk profile than 2% per trade with 45% win rate. Same %, very different ruin probability.
The fix isn't to abandon fixed-fractional. It's to layer on top of it.
Kelly criterion explained without the math
Kelly criterion is the math-derived “optimal” position size given a known edge. It maximizes long-term geometric growth.
The intuition: if your win rate is 55% and your average winner is 1.8× your average loser, Kelly says risk roughly 11% per trade for geometric growth. Don't do this.
Full Kelly maximizes growth but produces 50%+ drawdowns en route. No human can survive that psychologically. The practical version is half-Kelly (~5.5% in the example) — captures 75% of the geometric growth with half the drawdown.
And in practice, even half-Kelly is often too aggressive because your stated win rate is probably overestimated by 5-10 percentage points (recency bias). Hence the “margin-of-safety” adjustment in the next section.
The 3 layers of risk
Real risk management is three layers, not one. Each layer catches failures the layer above missed.
Layer 1 — Per-trade sizing
Use half-Kelly with a 5pt downward win-rate adjustment. The math looks like overkill until you realize it's the difference between doubling your account and blowing it up.
For most retail traders trading liquid US equities or major crypto, this lands around 0.5-1.5% risk per trade.
Layer 2 — Per-day cap
Max daily loss: 3% of account. Stop trading for the day when you hit it. No exceptions.
This single rule prevents 90% of catastrophic blowups. The blowups don't come from the first losing trade. They come from the third or fourth trade taken on tilt to “make it back.”
Layer 3 — Per-month circuit breaker
Max monthly drawdown: 8% of account. Hit it, stop trading for the rest of the month. Spend the time reviewing what went wrong instead of digging deeper.
The Risk Calculator runs this for you
Plug in your account size, setup win rate, and R/R ratio. Get half-Kelly position size, drawdown projection, and margin-of-safety adjustments — with the math shown. Pro tier.
Prompt #1 — Personalized sizing calculation
Run this every time you size a position. The math is too easy to mess up under emotional pressure — let the AI handle it.
You are a portfolio risk manager. Calculate position sizing for this trade. Account size: $[AMOUNT] Trade: long [TICKER] at $[ENTRY], stop $[STOP], target $[TARGET] Setup win rate (last 100 of this setup): [%] Avg winner / avg loser ratio: [RATIO] Max risk per trade rule: [%] Show me: 1. Half-Kelly optimal position size (% of account) 2. Position size in shares and dollars 3. Dollar amount at risk if stop hits 4. Maximum drawdown if I take 5 consecutive losses at this size 5. Position size if my stated win rate is overestimated by 5pts (margin-of-safety adjustment) 6. Breakeven win rate given this R/R ratio 7. The single edge metric I should track to know if my setup is degrading Show your math. Use the Kelly formula explicitly: f* = (bp - q) / b
Prompt #2 — Drawdown stress test
Run this monthly. Forces you to confront how bad it could get before it actually does.
You are a tail-risk analyst. My account: $[AMOUNT] My win rate: [%] My avg R per trade: [RATIO] My current per-trade risk: [%] Trades per month I take: [NUMBER] Calculate: 1. Probability of a 10% drawdown in any given month 2. Probability of a 20% drawdown in any given month 3. Worst-case drawdown over 1 year at 95% confidence 4. The position size that would reduce 1-year drawdown probability by half 5. Three concrete behavioral rules I should adopt to survive a 15% drawdown without breaking discipline Use Monte Carlo logic. Show your reasoning. Don't just give numbers — explain what each means in dollars.
Correlation: the silent risk multiplier
Five “independent” trades that are all long tech is one trade with 5× the size. Most retail traders ignore this because correlation isn't intuitive when each ticker has a different chart.
Quick rule: never run more than 3 positions in highly correlated assets at full size. If you want more exposure to a sector, size each at 50% of your normal position.
Common correlation traps:
- 5 different tech longs (NVDA, AMD, TSM, AVGO, ASML) — same trade.
- BTC + ETH + 3 large-cap alts — same trade.
- SPY puts + QQQ puts + IWM puts — same trade (worse: levered).
- Long oil + short bonds + long commodities — all reflation.
The risk audit you should run weekly
Five-minute Sunday review. Saves more capital than any indicator.
- Total open exposure as % of account.
- Largest single-name exposure.
- Effective sector exposure (group correlated positions).
- Daily and monthly P&L vs caps.
- Edge metric reading: still pointing the right direction?
If any of these look off, the response isn't to add another rule. It's to size down until they fit. Risk management is mostly subtraction, not addition.