Part IV, Edge and Risk Chapter 17

Risk Management for Futures

The trader who manages risk well outlives the trader who picks setups well. After 200 sessions the difference is unmistakable.

17.0Why this chapter exists

A trader can be wrong about every setup in this book and still survive if they size correctly. A trader can be right about every setup and still go broke if they do not.

That is not hyperbole; it is arithmetic, and the rest of the chapter is the proof.

Position sizing, stop placement, and the daily-loss cap are the only edges that compound across regime changes. Pattern recognition decays. Strategy edges erode. Markets adapt and competitors arrive. What does not change is the math of compounding losses, which punishes leverage with a force most retail-trained traders genuinely have not internalised until the second or third time they have to climb out of a deep drawdown.

So this chapter is dense, and it has to be. We start with the geometry of recovery, because that is the constraint everything else respects. From there: the Kelly criterion and why institutional desks use a quarter or half of what the formula suggests. The sizing formula made explicit, with worked examples across the four primary contracts. Stop placement rules, including the one rule that gets violated more than any other (round numbers as stops). Correlated-position management, because long ES and long NQ are not two trades. A five-tier framework for matching size to context. A drawdown protocol that prescribes specific actions at specific equity-down levels. And the failure modes that consistently destroy accounts.

The material is non-negotiable. Open any working desk's risk handbook and you will find approximately the same rules, because the math underneath them is universal.

Figure D17.1.Drawdown versus recovery required. The math underlying the per-session loss cap: stay out of the gravity well past 50 percent.

17.1The geometry of compounding losses

Before the rules, the math.

The recovery requirement

A loss of L% requires a gain of L / (1 − L) to break even:

Loss	Recovery required
5%	5.3%
10%	11.1%
20%	25.0%
25%	33.3%
33%	49.3%
50%	100%
75%	300%
90%	900%

The asymmetry is the entire point. A 50% drawdown requires doubling to recover. A 75% drawdown requires quadrupling. Above 50% drawdowns, the math becomes catastrophic; the probability of recovering becomes vanishingly small under realistic edge assumptions.

This is why prop desks write daily-loss caps into the contract. A daily limit of 2 to 3% per session keeps the trader in the recoverable zone even after a string of bad sessions; without the cap, a single bad day can put the trader into the catastrophic zone.

Implications for sizing

The geometry of recovery dictates conservative sizing:

Per-trade risk: 0.5 to 1.5% of account. Outliers (high-conviction setups in known-edge regimes) up to 2%. Never above.
Per-session loss cap: 2 to 3% of account. Hard. When hit, stop trading for the day.
Per-week loss cap (institutional): 5 to 8% of account. Stop trading for the week.
Per-month loss cap: 10 to 15% of account. Reduce size or stop trading; investigate root cause.

These are defensive numbers, not suggestions. Real desks enforce them with hard rules, not heuristics.

17.2The Kelly criterion and fractional Kelly

The Kelly criterion (Kelly, 1956) gives the optimal fraction of capital to bet given known edge:

f* = (p × b − q) / b

Where: - p is win probability. - q = 1 − p is loss probability. - b is the odds (R-multiple of win to loss).

For a strategy with 55% win rate and 1.5R:1R odds:

f* = (0.55 × 1.5 − 0.45) / 1.5 = (0.825 − 0.45) / 1.5 = 0.25

Full Kelly suggests betting 25% of capital per trade. This is enormous by retail standards.

Why full Kelly is too aggressive

Full Kelly is mathematically optimal under three assumptions:

The win probability p and odds b are known with certainty.
Returns are independent across trades.
The trader can survive arbitrary drawdown.

In reality: 1. p and b are estimates with substantial error. Overestimating either inflates f*. 2. Returns have positive autocorrelation in some regimes (winning streaks, losing streaks). 3. Drawdowns are emotionally and operationally limited; a 50% drawdown that is "in the math" is rarely tolerable in practice.

Fractional Kelly

The institutional standard: use 0.25 to 0.5 of the Kelly suggestion:

Half Kelly (0.5): captures most of the long-run growth advantage with materially smaller drawdowns.
Quarter Kelly (0.25): even smaller drawdowns; lower long-run growth but higher psychological tolerance.

For the 55%/1.5R example: full Kelly = 25%; half Kelly = 12.5%; quarter Kelly = 6.25%.

Even quarter Kelly is well above the 1% retail standard. The reason: Kelly assumes the edge is real and known; if you are uncertain about your edge, you should size much smaller.

When the edge estimate is uncertain

If p is estimated from limited data, the upper-bound of f* is not the right number to use. A Bayesian-aware adjustment:

Use the lower confidence bound on p rather than the point estimate.
This reduces f* substantially.
For a strategy with 55% observed win rate over 100 trades, the 95% lower bound is around 47% (which gives a negative f* for 1.5R odds, suggesting no edge).

Most trader's strategies have insufficient data to overcome the Bayesian conservatism. Hence: 1% per trade is roughly the right scale for retail-validated strategies; only highly-validated institutional strategies justify larger fractions.

Figure D17.4.Position sizing at 500 dollars per-trade risk across four contracts. Same dollar risk produces different contract counts because tick values differ.

17.3The sizing formula

The operational formula:

contracts = floor(account_equity × per_trade_risk_pct / (stop_distance_ticks × tick_value))

Worked example (repeated from Frameworks doc): - Account: $50,000. - Per-trade risk: 1.0% = $500. - Trade: long ES, entry 6,200, stop 6,196 (16-tick stop). - ES tick value: $12.50. - Risk per contract: 16 × $12.50 = $200. - Contracts: floor($500 / $200) = 2 contracts.

Variations across the four primary contracts at $500 risk:

Contract	Stop ticks	Tick value	Risk per contract	Max contracts
ES	16	$12.50	$200	2
NQ	40	$5.00	$200	2
GC	40	$10.00	$400	1
CL	25	$10.00	$250	2

The contract count differs across contracts at the same dollar risk. This is correct. Sizing in dollars equalizes risk; sizing in contract counts would mis-equalize.

What if the formula gives 0 contracts?

If the stop is so wide that even one contract exceeds the risk budget, the trade is not sized. Either: - Skip the trade (the structural stop is too wide for the account). - Trade in micros (one-tenth notional, allowing fractional sizing). - Reduce the per-trade risk percentage (some traders go to 0.5% in known-edge regimes to fit the trade in).

Never reduce the stop distance to fit the position. The structural stop is what protects the account; sizing should adjust, not the stop.

17.4Stop placement rules

The structural rules for where stops belong.

Rule 1: Behind structure

For a long entry, the stop belongs below the structural support level. For a short, above the structural resistance. The stop is "behind" the level that justifies the trade.

If the entry rationale is "long at the equal-low support cluster," the stop is below the equal-low cluster, plus a buffer for ATR-conditioned noise (typically 1× to 1.5× ATR).

Rule 2: Never at round numbers

Round numbers attract clusters of stops; they are exactly where sweeps occur. Placing your stop at a round number is placing it in the harvest zone.

If the structural stop falls near a round number, move the stop beyond the round number (further from entry) by a few ticks. The cost of slightly wider stop is much smaller than the cost of being swept.

Rule 3: ATR-sized minimum

A stop should be at least 1× ATR(14) wide. Tighter stops on liquid intraday futures are statistically dominated by normal volatility; the win rate on entries with stops below 1× ATR is typically below break-even.

If structural stop is < 1× ATR, either: - Widen the stop to 1× ATR. - Skip the trade (the level is too tight to justify the entry given recent volatility).

Rule 4: No stop is not a strategy

Some traders enter "without a stop" because they "trust the structure" or "expect to manage it manually." This is a recipe for catastrophic loss when the regime shifts unexpectedly. Always have a stop entered with the order, even if it is wider than ideal. The discipline beats the optimization.

Rule 5: Stops do not move down (in long trades)

Once a stop is set, it can move up (to lock in profit, scale out, etc.) but not down. Moving a stop further from entry to "give the trade more room" is the most reliably account-destroying behavior in retail trading.

17.5The per-session loss cap

It is 11:18 ET. You are down 2.4% on the session. Two trades, both range fades, both stopped out cleanly because the regime was not what you thought it was at 10:30. Your per-session cap is 2.5%. You are one bad ten-tick wiggle from hitting it.

The next setup forms in front of you. It looks clean. Equal highs at PDH, sweep-and-rejection, CVD divergence on the wick, footprint shows absorption on the rejection bar. Textbook. A real trade, by every framework in this book.

You take it. You are stopped out. You are now down 3.6%.

This is the moment that decides whether you are a trader who survives ten years or a trader who blows up in eighteen months. Not because that one trade was wrong, the setup was reasonable, and on a different day in a different regime it would have worked. The trade was wrong because you took it after hitting the cap, and the cap is the discipline that protects you from the version of yourself that thinks one more clean trade can fix a bad day.

The math behind the 2 to 3 percent figure is the same compounding geometry from §17.1, applied at the session horizon. A 2 percent loss requires 2.04 percent to recover; a 3 percent loss, 3.09 percent. Both are routinely achievable on the next session. A 5 percent session loss requires 5.26 percent recovery, which already starts demanding above-average performance. A 10 percent session loss requires 11.1 percent, which on a calibrated edge is roughly two weeks of typical gain. The cap is designed to keep you on the easily-recoverable side of that curve, no matter what the day delivers.

The harder part is the psychology. After a losing session, the trader is emotionally inclined to trade more, not less. The cap directly contradicts the impulse, which is exactly why it has to be hard rather than soft. Prop desks enforce it with software (auto-flatten, platform lockout) because they know willpower is not the right tool for this job. The retail trader who has to enforce it on themselves needs a different kind of structure: a sticky note on the monitor, a hard mental rule established before the session opens, a buddy you call before taking the override trade. Whatever the mechanism, the rule has to outrank the in-the-moment judgement of a trader who is already down.

The cap is a floor, not a target

Hitting the cap is not the goal; the goal is to not hit the cap most days. A trader who hits the cap regularly is over-trading or has degraded edge; investigate root cause rather than treating the cap as the daily allotment.

17.6Correlated position management

A trader long ES and long NQ does not have two independent trades; they have one trade in two correlated contracts. Risk should be assessed at the aggregate exposure.

The correlation-aware sizing rule

If two positions are correlated above 0.7, treat them as one position for risk purposes. The combined risk should not exceed the per-trade cap.

Worked example

Account: $50K, per-trade risk 1.0% = $500.
Long ES at 1× sizing: 2 contracts, $200 risk × 2 = ... wait, $500 budget allows 2.5, so size to 2. Risk = $400.
Long NQ also tempting: would size to ~2 contracts with $400 risk.
Correlated: ES-NQ at 0.85+ intraday. The combined position is functionally one bet at $800 risk (1.6% of account), exceeding the 1% per-trade cap.
Correct: either reduce ES or NQ size, or skip the second.

Why this matters

In trending or risk-on days, ES and NQ move together. A 50% adverse move in the index complex hits both positions equally. The "hedged" feeling of having "two different trades" is illusory; the diversification benefit at correlation 0.85+ is minimal.

When correlation is lower

If the trader is long ES and long GC, the correlation is typically low (~0.0 to 0.3 over a month, varies by macro regime). These can be sized somewhat independently, but the trader should still cap total portfolio exposure at 3 to 5% per session.

17.7The five-tier risk framework

From Frameworks doc, expanded:

Tier 1: Conservative (0.5% / 1.0%)

When to use: - New strategy (less than 100 live trades validated). - New contract. - Recovering from drawdown. - Periods of personal stress, sleep deprivation, or distraction. - First 10 sessions on a new platform.

Tier 2: Standard (1.0% / 2.0%)

When to use: - Validated strategy in known regime. - Normal personal state. - 100+ live trades documented.

This is the institutional default for most setups.

Tier 3: Aggressive (1.5% / 3.0%)

When to use: - Highest-conviction setup in your validated playbook. - Regime is the strategy's known sweet spot. - All confluence factors aligned. - You are well-rested and focused.

Aggressive sizing should be rare, perhaps 1 to 2 trades per week at most.

Tier 4: Crisis (auto, 0.25% / 0.5%)

When to use (automatic, when conditions are met): - VIX > 80th percentile of trailing 30 days. - ATR > 2× trailing 30-day median. - Cross-asset correlation convergence. - Crisis regime classification from composite.

Cut size automatically; investigate the regime; consider standing down entirely.

Tier 5: Probationary (0.25% / 1.0%)

When to use: - Recovering from a > 5% drawdown. - After a discipline violation. - After a sequence of L4 losses (Frameworks doc Framework 9).

Stay in Probationary until 5 winning sessions in a row at the smaller size; then return to Standard.

17.8Drawdown management

When in drawdown, the rules change.

The drawdown response sequence

Drawdown < 3%: normal trading; review sessions for systemic errors.
Drawdown 3 to 5%: reduce per-trade risk to Conservative tier. Continue trading. Sustained underperformance flag.
Drawdown 5 to 10%: mandatory pause for 24 hours. Document root cause: regime shift, edge decay, discipline drift, personal factors. Address before resuming.
Drawdown 10 to 15%: mandatory pause for 1 week. Re-validate strategy on recent data. Consider whether the edge has decayed.
Drawdown 15 to 20%: strategic re-evaluation. Reduce capital allocation. Consult with peer or mentor. Investigate whether the strategy still has edge.
Drawdown > 20%: stop trading. Investigate fully. Restart only with significantly reduced size and high confidence.

Why this matters

The trader's capacity to make good decisions degrades during drawdown. Cognitive fatigue, emotional reactivity, and motivated reasoning all increase as drawdowns deepen. The forced pauses prevent compounding mistakes.

The institutional analog

Prop desks have explicit drawdown protocols. A trader at 5% drawdown in a month sees their capital allocation reduced. At 10%, they may be removed from active trading. The discipline is built in; retail traders need to build their own.

17.9The math of position sizing under uncertainty

A common mistake: using the historical win rate as the input to sizing without accounting for estimation uncertainty.

The Bayesian alternative

If the trader's strategy has won 55 of the last 100 trades: - Point estimate of p is 55%. - 95% confidence interval is approximately 45% to 65%. - Lower bound is 45%.

Sizing should use the lower bound, not the point estimate. The principle: do not bet that your future performance equals your historical mean; bet that your future is no worse than your historical lower bound.

The compounding effect

If you size based on point estimate and the true p is at the lower bound, your live performance disappoints. If you size based on the lower bound and the true p is at the point estimate, your performance exceeds expectations. The asymmetry favors conservative sizing.

When the sample is small

For a strategy with 30 trades observed, the confidence interval is wide (e.g., 36 to 73% for 50% observed). Sizing should be at the lower bound, which may suggest no edge. The strategy needs more data before aggressive sizing.

This is why prop desks require 200+ live trades before scaling up size. Less than that is statistically underpowered for sizing decisions.

17.10Failure modes specific to risk management

Sizing in points or ticks, not dollars. Chapter 1 trap. Always size in dollars.
Adjusting size after a loss in the same direction. Martingale-adjacent. Size remains constant per the formula; the budget shrinks with the account, but the percentage stays.
Moving stops further from entry. "Giving the trade more room." Almost always wrong. Stops move closer (lock in profit), not further.
Skipping the per-session cap. The cap is hard. "One more trade" after the cap is the most account-destructive behavior in trading.
Correlation blindness. Trading multiple correlated contracts as if they were independent. Aggregate exposure must be assessed.
Sizing on conviction without validation. "I'm sure this works" is not a sizing input. Validation is. Size based on documented edge, not feeling.
Crisis-regime sizing as opportunity. Crisis volatility looks like opportunity (bigger moves) but is actually higher risk (lower edge per unit of risk). Cut size automatically.
Drawdown denial. Refusing to acknowledge drawdown and adjust accordingly. The protocol exists for a reason.
No stop on "managed" trades. Always have a hard stop with the order. Manual management is psychologically unreliable in stressful moments.
Probabilistic over-confidence. "55% win rate" assumes the past 100 trades are representative. They may not be. Bayesian conservatism beats point-estimate confidence.

17.11The integrated stack treatment

Risk management is Layer 9 of the institutional stack, alongside systematic infrastructure and statistical validation. It is the discipline that makes everything else possible.

The institutional stack from §10 of the Research Summary lists "Mechanical risk constraints" as the ninth layer. This chapter is what fills it in.

A daily flow:

Pre-market: confirm per-trade and per-session risk caps for today.
Open: size every trade per the formula.
During session: stop trading at session cap.
Post-session: review for risk-rule violations; document any.
Weekly: review aggregate exposure, drawdown, and tier-level adjustments.

The risk management layer wraps every other layer. The trade is no good if the risk is wrong.

17.12Diagram concepts referenced in this chapter

D17.1: Drawdown-vs-recovery curve. The asymmetric curve from §17.1, with annotated key drawdown levels.
D17.2: Kelly fraction nomogram. A 2D plot of f* over p and b, with shaded regions for full / half / quarter Kelly.
D17.3: Per-tier sizing formula visualization. The five tiers with characteristic per-trade and per-session caps, shown as a stacked-bar chart of risk allowance.
D17.4: Sizing across contracts. The four-contract table from §17.3 visualized: same dollar risk = different contract counts.
D17.5: Stops at structure vs at round numbers. Two panels: a stop at structure vs a stop at a round number; the round-number stop gets swept, the structural stop does not.
D17.6: Correlated-position aggregate risk. A schematic showing how two correlated positions sum to one effective position, with the implied risk reduction.

17.14Exercises

Exercise 17.1: Sizing fluency. For your account size and 1% per-trade risk, compute the contract count for each of: - ES with 8-tick, 16-tick, 25-tick stops. - NQ with 20-tick, 40-tick, 60-tick stops. - GC with 20-tick, 40-tick stops. - CL with 15-tick, 25-tick, 40-tick stops.

Build the table. The goal is to be able to size any setup in 5 seconds.

Exercise 17.2: Drawdown response sequence audit. Review your trade journal for any drawdown periods exceeding 5%. Did you follow the response sequence? If not, what did you do, and how did it work out?

Exercise 17.3: Kelly calibration. For your most-traded setup, compute the historical p and b. Compute full Kelly. Compute half and quarter Kelly. Compare to your actual sizing. Where is the gap?

Exercise 17.4: Per-session cap compliance. Track over one month: how many sessions did you hit the per-session cap? How many sessions did you violate it (continued trading after hitting)? What was the outcome of the violations?

Exercise 17.5: Bayesian sizing. For a setup with 30 observed trades and 60% win rate, compute the 95% lower confidence bound on p. Use that lower bound to size. Compare to point-estimate sizing.

Next chapter: execution, slippage, spread, queue position, and the cost of being wrong on a fast move.