Position Sizing & Kelly Criterion for Stocks

Advanced Track. Assumes solid risk basics from Risk Management 101, strategy context from Track 3 (e.g. swing, day trading, breakouts), and process literacy from the free stock curriculum.

How Much Dominates What

Of all variables a trader controls, position size is simultaneously the most impactful and the most neglected. Entry timing and indicator choice attract obsessive attention; “how many shares?” is often answered with a round lot or a gut feeling. That is a critical error. A mathematically superior sizing method can take a mediocre edge and produce durable long-run growth; reckless size can take a profitable edge and drive it to ruin through a perfectly normal drawdown sequence.

The Kelly Criterion — and its practical derivative, half-Kelly — provides a rigorous foundation for optimal sizing when you have a statistically estimated edge. This course derives the idea for equity traders, contrasts full Kelly vs half-Kelly vs fixed fractional (1% rule), stresses gap and estimation error, and ties every formula to free tools including the Kelly criterion calculator and risk / position size calculator.

1. The Mathematics of Ruin

Consider a strategy with a genuine edge: 55% win rate, 1:1 reward-to-risk. Sized at 25% of equity per trade, the probability of a devastating drawdown before meaningful compounding is extreme because variance compounds with size. The same edge at 1–2% risk per trade has far lower ruin probability over hundreds of trades. The strategy is identical; outcomes are categorically different.

John L. Kelly Jr. (Bell Labs, 1956), building on Shannon’s information theory, showed that the correct objective for a compounding account is expected geometric growth rate — not average arithmetic return. A 50% drawdown requires a 100% gain to recover; a 25% drawdown needs only ~33%. Sizing that constrains drawdown magnitude accelerates recovery. That insight is why professional risk desks obsess over fraction of capital at risk — not just “whether the setup looks good.” Revisit expectancy framing from Risk Management 101 and track results with the win rate calculator.

2. The Kelly Criterion: Formula

For a sequence of independent bets with known edge, the fraction of capital that maximises long-run geometric growth is:

f* = (b × p − q) ÷ b

• f* = optimal capital fraction to risk (Kelly fraction)
• b = net reward ÷ risk ratio (e.g. average win $ / average loss $)
• p = probability of winning
• q = 1 − p (probability of losing)

Worked example. p = 0.55, b = 1.5, q = 0.45.
f* = (1.5 × 0.55 − 0.45) ÷ 1.5 = (0.825 − 0.45) ÷ 1.5 = 0.375 ÷ 1.5 = 0.25 (25%).

Full Kelly says risk 25% of equity per trade under those exact inputs. That is the growth-maximising fraction in the idealized model — not the recommended retail operating fraction (see half-Kelly). Compute with the Kelly criterion calculator rather than mental math under stress.

3. Full Kelly vs Half-Kelly vs Fixed Fractional

Full Kelly maximises median geometric growth but produces enormous outcome volatility. Drawdowns of 30–50% are common even with genuine edge because the formula assumes infinite trials and zero estimation error — conditions retail traders never enjoy. Most professional managers treat full Kelly as theoretically interesting and practically unacceptable for a single strategy sleeve.

Half-Kelly deploys 50% of f*. It captures a large share of full-Kelly growth (on the order of ~75% of the geometric rate in classic analyses) while cutting drawdown volatility dramatically. Operating bands of 0.25–0.5× Kelly are common in systematic funds. This is the professional default when Kelly inputs are trusted at all.

Fixed fractional (e.g. risk 1% of equity per trade to a structural stop) — the backbone of Risk Management 101 — requires no edge estimate. It is the correct starting point until you have a large, clean sample (often 100–300+ trades) from a journaled strategy. ATR-based dynamic stops (later Track 4 course) still convert to a dollar risk first; Kelly then asks what fraction of equity that dollar risk may represent.

4. Estimation Error: Why Overbetting Kills

Your “true” p and b are unknowns estimated from finite history. Overestimate either and Kelly returns an oversized f* that pushes you into the overbetting regime, where geometric growth falls even though arithmetic expectancy may still look positive. Asymmetry is brutal: underbetting slows wealth; overbetting risks ruin.

Practical defenses:

• Use half-Kelly or quarter-Kelly when samples are modest
• Cap f* at a hard ceiling (e.g. never above 2–5% risk per trade for equities with gaps)
• Recalculate inputs on a rolling window; strategy edge drifts by regime
• Do not apply Kelly to 20-trade “hot streaks”

Journaling and metrics (Track 4 later) feed honest p and b. Until then, fixed fractional dominates. Use the break-even calculator to stress what win rate you need at a given R:R.

5. Equity-Specific Complications: Gaps and Correlation

Classic Kelly assumes independent trials with known payoffs. Equities violate that:

• Gaps: planned $ risk can be exceeded at the open — size smaller for overnight holds (swing trading, gap trading)
• Correlation: three “independent” longs that are all SPY beta are one risk unit (portfolio basics, relative strength)
• Fat tails: earnings, halts, and cascade days break neat win/loss distributions
• Costs: commissions and spreads shrink effective b — use net P&L from the P&L calculator

Session and broker context: How Stock Markets Work, How to Use a Stock Broker. Post-PDT margin reality does not change ruin math — ~$2k margin minimum is not a sizing strategy.

6. Implementation Workflow

1. From journal: estimate p and b on a rolling sample (prefer ≥100 trades).
2. Compute f* with the Kelly calculator; take half-Kelly (0.5 f*).
3. Cap at your policy max (e.g. 1–2% risk per trade for gap-prone equities).
4. For each setup: structural stop distance → dollar risk = min(half-Kelly×equity, policy cap×equity) → shares = dollar risk ÷ stop distance via the risk calculator.
5. Validate exits with the SL/TP calculator.
6. Recalibrate quarterly or after regime change (trend vs chop — see trend following vs mean reversion).

Numeric bridge. Equity $50,000. Half-Kelly risk fraction capped at 1.5% → max risk $750. Long entry $100, stop $97 → $3 risk/share → max size = 250 shares. If half-Kelly raw said 4%, you still risk only 1.5% under the equity policy ceiling.

7. Strategy Type Matters

A scalping book with 60% wins at 0.8:1 R:R yields a different f* than a swing book with 40% wins at 3:1. Both can be Kelly-positive; both demand different size. A 50% win rate at exactly 1:1 implies f* = 0 — no edge, no trade. If calculated half-Kelly exceeds ~10–20% of equity, re-check sample quality before believing the estimate.

Strategy modules that feed samples: breakouts, day trading, momentum. Multi-timeframe consistency: MTF course.

8. Common Sizing Mistakes

• Emotional scaling — size up after wins, revenge size after losses (beginner mistakes)
• Kelly on noise — 30 trades is not a probability estimate
• Ignoring correlation — three Kelly-sized correlated bets = triple risk
• Confusing notional with risk — $25k position is not $25k risk if stop is tight
• Full Kelly because the formula is “optimal” — optimal only under assumptions you do not have

Optional reading: DennTech blog. Free tool stack: all calculators. Measure extension before oversized “conviction” entries with the percentage change calculator.

9. Checklist Before Raising Size

1. Written strategy with structural stops (not mental).
2. Journal sample large enough for rough p and b.
3. f* computed; half-Kelly taken; policy cap applied.
4. Gap and event risk reflected in cap (overnight vs day).
5. Open correlated risk summed, not treated as independent Kelly bets.
6. Recalibration date scheduled.

Foundations of ownership and process: What Is Stock Trading? TA probability framing: Intro to TA. Structure for stops: support & resistance, market structure.

10. Narrative: From 1% to Data-Driven Fraction

A trader runs a pullback-in-trend playbook for a year at fixed 1% risk. After 200 trades, win rate is 48% and average win/loss is 1.8. Kelly f* ≈ (1.8×0.48 − 0.52)/1.8 ≈ 0.19. Half-Kelly ≈ 9.5% — still far too aggressive as a per-trade risk for equities with gaps, so they apply a 2% policy ceiling and treat half-Kelly only as an upper theoretical reference, not a target. They do not jump from 1% to 9%. They may test 1.25–1.5% only after confirming the sample survives a regime change quarter. That is professional sizing: math informs; policy and market structure constrain.

11. Translating Kelly Fraction into Share Count (Full Pipeline)

Traders often stop at “half-Kelly is 4%” and then still buy a round lot. The pipeline must be explicit. First, define the unit of risk as dollars to the structural invalidation, not as position notional. Second, convert the allowed equity fraction into a dollar risk budget: if equity is $80,000 and policy risk is the minimum of half-Kelly and a 1.5% hard cap, the dollar risk budget is 1.5% × $80,000 = $1,200 even if raw half-Kelly argued for more. Third, measure stop distance in price: entry $54.20, stop $52.40 yields $1.80 per share. Fourth, shares = floor($1,200 / $1.80) = 666 shares. Fifth, check liquidity: if average daily volume cannot absorb that size without moving the market, cut further. Sixth, check portfolio correlation caps so this Kelly-derived leg does not stack three high-beta names into one factor bet.

Notice that “optimal” never overrode microstructure or book-level constraints. Kelly informs the ceiling; the stop defines the conversion; the book defines whether you may spend the ceiling at all. Day-trade versions of the same pipeline use tighter caps (for example 0.4% per trade) because trade frequency multiplies decision risk even when overnight gap risk is lower. Swing versions may use slightly higher per-trade caps only if you have already reduced concurrent overnight exposure. In all cases, recompute after large P&L days: a +8% equity spike that is not yet withdrawn still increases absolute dollar risk at a fixed percentage — which is intended — but you must reconfirm that your psychology and liquidity can handle the larger share counts.

12. Sample Size, Confidence Intervals, and Honesty About Edge

A win rate of 58% over 40 trades is a noisy estimate. Sampling variability alone can swing that estimate several points. Professionals therefore distrust Kelly inputs until the sample is large enough that small changes in p or b do not flip f* from “sizable” to “near zero.” A practical discipline is to maintain three estimates: optimistic (top of a reasonable range for p and b), base, and conservative (haircut p by a few points and b by costs). Size using the conservative estimate, or take the minimum of conservative half-Kelly and your fixed fractional policy. If conservative f* collapses near zero, you do not have a robust edge for aggressive sizing — you have a maybe.

Separate edge decay from bad luck. If a strategy’s rolling 100-trade expectancy turns negative after a regime shift (trend system in a year of ranges), Kelly correctly tells you to reduce size toward zero. That is not “giving up”; it is refusing to fund a dead edge. Conversely, a temporary drawdown inside historical bounds is not permission to double size for recovery. Recovery sizing is the opposite of Kelly logic: it increases fraction after losses, exactly when equity is smaller and estimation error may be rising. Document regime labels on each trade (trend / range / event) so you can compute p and b inside regimes rather than mixing incompatible markets into one fake edge.

13. When to Stay on Fixed Fractional Forever

Not every trader should graduate to Kelly-driven fractions. If your edge is discretionary and poorly stationary, if your journal is incomplete, if you change setups monthly, or if you cannot tolerate even moderate multi-week drawdowns, fixed fractional sizing at 0.5–1% (gap-aware) remains the professional choice. Kelly is a tool for systems with measurable statistics and operators who will obey the haircut. It is not a badge of sophistication. Many excellent proprietary traders never compute f* explicitly; they enforce hard risk budgets that approximate conservative Kelly outcomes without the false precision of a three-decimal fraction.

The advanced skill is knowing which camp you are in this quarter. If you lack 100+ clean trades, stay fixed fractional. If you have the sample but live with heavy overnight event risk, keep a low ceiling regardless of f*. If you run multiple correlated strategies, allocate risk budgets at the book level first, then size legs inside those budgets. Math without governance is just a more elegant way to overbet.

Key Takeaways

Tools for This Course

• Kelly Criterion Calculator — compute f* from win rate and payoff ratio; then take half.
• Risk & Position Size Calculator — convert capped dollar risk into share count from stop distance.
• Win Rate + P&L — feed honest inputs from your journal.
• Stock Courses Hub — Track 4 continues with ATR sizing, psychology, backtesting, and journals.