Skip to content
QuantCalc Tools
Directory

Kelly Margin Leverage Sizing

Continuous Merton-Kelly Leverage ((mu - r_f)/sigma^2) & Half-Kelly Math

Under geometric Brownian motion, continuous log-optimal leverage equals excess arithmetic drift divided by variance; prove the exact c*(2-c) fractional Kelly parabola and Sharpe^2 growth ceiling.

By Elena Rostova, FRM, CQFReview not recorded · Updated
Continuous Merton-Kelly Leverage ((mu - r_f)/sigma^2) & Half-Kelly Math — QuantCalc Tools 2026 quantitative & supervisory data card
QuantCalc Tools (Kelly Criterion, Initial/Variation Margin Simulators & Liquidation Math): Merton-Thorp Ito diffusion proof, c*(2-c) Half-Kelly parabola, and borrow kinks.Source: QuantCalc Tools Institutional Research DeskLicense: CC-BY-4.0

Real-world asset returns are continuous distributions rather than discrete coin tosses. In 1969–1971, Edward Thorp and Nobel laureate Robert Merton extended Kelly's logarithmic utility criterion ($U(W) = \ln W$) to continuous-time geometric Brownian motion, producing the single most important equation in institutional leverage management: *$f^ = \frac{\mu - r_f}{\sigma^2} = \frac{\text{SR}}{\sigma}$. Even better, continuous calculus reveals a quadratic miracle at Half-Kelly ($f = 0.5 f^$): cutting leverage in half sacrifices only $25\%$* of theoretical maximum excess compounding while eliminating $75\%$ of portfolio variance! Run live fractional simulations in our Kelly Criterion Calculator, inspect the Continuous Fractional Half-Kelly Engine, and anchor in our Kelly, Margin & Liquidation Hub.

Exact Ito's Lemma Derivation: Why Maximum Log Growth Equals $r_f + \frac{1}{2}\text{SR}^2$#

Let risky asset $St$ follow $dSt / St = \mu\,dt + \sigma\,dWt$ (critical warning: $\mu$ here is the ARITHMETIC instantaneous drift, NEVER historical geometric CAGR $g = \mu - \frac{1}{2}\sigma^2$!) and risk-free cash earn $rf$. Allocating constant fraction $f$ of wealth $Wt$ to the risky asset (borrowing $(f-1)Wt$ at $rf$ when $f > 1$) yields $dWt / Wt = [rf + f(\mu - rf)]\,dt + f\sigma\,dWt$. Applying Ito's Lemma to $\ln Wt$ gives exact asymptotic log-growth rate $g(f)$ as a downward-opening parabola in $f$:

Formula or model from the articleNot independently verified
Continuous Merton-Kelly Optimal Leverage & Fractional c-Kelly Parabola:
  • Asymptotic Log-Growth Rate : g(f) = rf + f * (mu - rf) - 0.5 * f^2 * sigma^2
  • First-Order Condition g'=0 : (mu - rf) - f* * sigma^2 = 0 => f* = (mu - rf) / sigma^2 = SR / sigma
  • Max Theoretical Growth Rate: g(f) = r_f + 0.5 [ (mu - rf)^2 / sigma^2 ] = rf + 0.5 * SR^2
  • Exact Fractional Kelly Scaling Law (Let f = c * f*, for any multiplier c > 0):
  • g(c * f) - r_f = c (2 - c) * [ 0.5 * SR^2 ]
  • Portfolio Vol = c * f* * sigma = c * SR
  • Portfolio Variance = c^2 * SR^2 (At Half-Kelly c = 0.5: Growth keeps 0.5*(1.5) = 75.0%; Variance drops to 0.25 = 25.0%!)
  • Worked InstitutionalCTA Sizing Audit (rf = 4.50%, Sharpe SR = 0.80, Unlevered Vol sigma = 12.0% -> mu - rf = 9.60%):
  • 1. Full Kelly (c = 1.0) : f* = 0.096 / 0.0144 = 6.67x Leverage! | Vol = 80.0% | g(f) = 4.50% + 0.5(0.64) = 36.50% log (CAGR = 44.0%!)
  • (Wait: 80% annual vol means routine -55% drawdowns & massive CME margin calls!)
  • 2. Half Kelly (c = 0.5) : 0.5f = 3.33x Leverage | Vol = 40.0% | g = 4.50% + 0.75*(32.0%) = 28.50% log (CAGR = 33.0%!)
  • 3. Quarter Kelly (c=0.25): 0.25f= 1.67x Leverage | Vol = 20.0% | g = 4.50% + 0.4375*(32%) = 18.50% log (CAGR = 20.3%, institutional sweet spot!)

Look at the Maximum Growth Identity $g(f^) = r_f + \frac{1}{2}\text{SR}^2$: maximum achievable geometric compounding depends ONLY on the strategy's squared Sharpe Ratio $\text{SR}^2$, completely independent of whether the underlying asset has $5\%$ volatility or $50\%$ volatility! A low-volatility bond arbitrage pod with $\text{SR} = 1.40$ has nearly $8\times$ higher* Kelly growth capacity ($\frac{1}{2}(1.96) = +98\%$ log excess) than an equity index with $\text{SR} = 0.50$ ($\frac{1}{2}(0.25) = +12.5\%$ log excess), provided institutional financing spreads remain tight!

Asymmetric risk-reward mechanics of the c*(2-c) fractional Kelly parabola showing why over-estimating drift mu is penalized far harder than under-estimating mu
Asymmetric risk-reward mechanics of the c*(2-c) fractional Kelly parabola showing why over-estimating drift mu is penalized far harder than under-estimating muVerified Exhibit

Exact Adjustment When Margin Borrowing Rate $rb$ Exceeds Cash Rate $rf$#

Textbook Merton-Kelly assumes you can borrow at the exact same risk-free rate $rf$ that cash earns. On a real brokerage desk (SOFR Carry Calculator), cash earns $rf = 4.50\%$ while margin loans cost $r_b = 5.75\%$ (+125 bps prime spread). That creates a three-zone kinked Kelly solution:

  1. If $(\mu - rf) / \sigma^2 \le 1$: You hold positive cash (no borrowing), so *$f^ = (\mu - rf) / \sigma^2$**.
  2. If $(\mu - rb) / \sigma^2 \ge 1$: Marginal leverage clears the higher borrow cost, so *$f^ = (\mu - rb) / \sigma^2$**!
  3. If $(\mu - rb) / \sigma^2 < 1 < (\mu - rf) / \sigma^2$: *Optimal leverage clamps at exact `f = 1.00x` ($100\%$ unlevered equity, zero margin loan)**!

Complete Fractional Kelly Scaling Schedule ($c \in [0.20, 2.25]$)#

Inspect the exact algebraic trade-off across fractional multiplier $c = f / f^*$ under continuous Brownian motion:

Quantitative Distribution · Excess Log Growth Kept $c(2-c)$
Excess Log Growth Kept $c(2-c)$Portfolio Vol ($c \cdot \text{SR}$)
010.25x Quarter-Kelly$43.75\%$020.50x Half-Kelly$75.00\%$030.75x Three-Quarter Kelly$93.75\%$041.00x Full-Kelly$100.00\%$ (Pea…051.50x Over-Kelly$75.00\%$ (Infe…062.00x Double-Kelly Cliff$0.00\%$ (Wiped…
Fractional Multiplier ($c = f / f^*$)Excess Log Growth Kept $c(2-c)$Portfolio Vol ($c \cdot \text{SR}$)Portfolio Variance ($c^2$)Prob of $-50\%$ DrawdownInstitutional Desk Usage
0.25x Quarter-Kelly$43.75\%$$0.25\,\text{SR}$$6.25\%$$1.56\%$ ((0.5)^6)Multi-strategy / pension standard
0.50x Half-Kelly$75.00\%$$0.50\,\text{SR}$$25.00\%$$12.50\%$ ((0.5)^3)Classic Thorp / Ziemba quant ceiling
0.75x Three-Quarter Kelly$93.75\%$$0.75\,\text{SR}$$56.25\%$$31.50\%$Aggressive prop (high parameter confidence)
1.00x Full-Kelly$100.00\%$ (Peak)$1.00\,\text{SR}$$100.00\%$$50.00\%$ ((0.5)^1)Theoretical upper bound — too volatile live!
1.50x Over-Kelly$75.00\%$ (Inferior!)$1.50\,\text{SR}$$225.00\%$$75.98\%$Strictly dominated by 0.50x (3x vol for same g!)
2.00x Double-Kelly Cliff$0.00\%$ (Wiped!)$2.00\,\text{SR}$$400.00\%$$87.50\%$Zero excess growth; $>2.0\times$ decays to $0$!
Dominance comparison between 0.5x Half-Kelly and 1.5x Over-Kelly proving identical 75 percent growth with 3x lower volatility at Half-Kelly
Dominance comparison between 0.5x Half-Kelly and 1.5x Over-Kelly proving identical 75 percent growth with 3x lower volatility at Half-KellyVerified Exhibit

Subtle Coding Bug Alert: Plugging Historical CAGR $g$ Instead of Arithmetic $\mu$!#

Watch out for a famous formula collision! If you read a long-run equity study quoting historical geometric CAGR $g = 10.0\%$, cash $rf = 4.5\%$, and volatility $\sigma = 18.0\%$ ($\sigma^2 = 0.0324$), DO NOT plug $g = 10.0\%$ into $(g - rf)/\sigma^2$! Remember $\mu = g + \frac{1}{2}\sigma^2 = 10.0\% + 1.62\% = 11.62\%$: plugging geometric $g$ instead of arithmetic $\mu$ silently subtracts $-0.50$ from optimal leverage ((g - r_f)/sigma^2 = f* - 0.5)!

Utility Theory Connection: Fractional $c$-Kelly Equals CRRA Power Utility ($\gamma = 1/c$)#

MacLean, Thorp & Ziemba (2011) proved that trading constant fractional Kelly $f = c \cdot f^$ is 100% mathematically equivalent* to maximizing Constant Relative Risk Aversion (CRRA) power utility $U(W) = \frac{W^{1-\gamma}}{1-\gamma}$ with risk-aversion coefficient $\gamma = \frac{1}{c}$: Full Kelly is $\gamma = 1$ ($\ln W$), Half-Kelly is $\gamma = 2$, and Quarter-Kelly is $\gamma = 4$! Extend single-asset Merton-Kelly to correlated books in multi-asset inverse-covariance Kelly shrinkage and drawdown-constrained fractional Kelly, and test Sharpe inputs in our Sharpe & Sortino Calculator.

Frequently asked questions

Why does continuous Merton-Kelly allow f* > 100% (say 3.0x leverage) when discrete binary Kelly never exceeds 100%?

In discrete binary Kelly where a loss forfeits 100% of your stake, betting f >= 100% carries positive probability of instant total ruin in one step. Under continuous diffusion Brownian motion with continuous rebalancing (zero jumps!), prices move continuously without gapping, allowing f* > 1. In real markets with overnight jump risk, always cap leverage against jump ruin!

How do Poisson crash jumps (Merton Jump-Diffusion) change continuous Kelly leverage?

Even a tiny 5% annual probability of a -25% instantaneous crash jump slashes optimal Kelly leverage by 30% to 50% below the pure Brownian (mu - r_f)/sigma^2 formula, because concave log utility ln(1 - 0.25*f) imposes an infinite barrier at f = 1 / 0.25 = 4.0x regardless of how high normal-times Sharpe looks!

How does estimation uncertainty in sample mean return (Baker & McHale 2013) shrink optimal Kelly?

Because true drift mu is estimated from T years of noisy data with standard error sigma / sqrt(T), integrating over the Bayesian predictive Student-t distribution multiplies plug-in Kelly f* by a shrinkage factor approximately equal to 1 / (1 + 1 / [SR^2 * T] ...)—typically recommending 0.3x to 0.6x plug-in Kelly on 5-year track records.

Why is 1.5x Over-Kelly strictly irrational for every risk-averse investor?

Look at the parabola c*(2 - c): c = 1.50 and c = 0.50 BOTH yield 0.75 * max_growth (exact same compounding rate!), yet c = 1.50 suffers THREE TIMES the volatility and a 76% probability of a -50% drawdown versus 12.5% at Half-Kelly. Every point c > 1.0 lies on the inefficient side of the growth-variance frontier!

Publication record

Published
13 November 2026
Updated
14 November 2026
Author
Elena Rostova, FRM, CQF
Review
Review not recorded
Automated fact check
Passed in the stored pipeline record
Source URLs
4 recorded

AI assistance recorded: anthropic/claude-opus-4.6. This field does not establish human review or source verification.

Academic & Institutional Attribution

Cite This Research & Embed Live Data Card

Open Attribution License (CC BY 4.0)

Standard academic & institutional research paper citation format.

Elena Rostova, FRM, CQF. (2026, November 14). Continuous Merton-Kelly Leverage ((mu - r_f)/sigma^2) & Half-Kelly Math. QuantCalc Tools. https://quantcalctools.com/guides/continuous-merton-kelly-leverage-mu-over-sigma-squared-half-kelly
Explore Kelly Margin Leverage Sizing Hub (20) →

QuantCalc Tools Data, Benchmarks & Contributor Records

Continuous Kelly Leverage Formula ((mu-rf)/sigma^2) Guide | QuantCalc Tools