Skip to content
QuantCalc Tools
Directory

Kelly Margin Leverage Sizing

Ralph Vince Optimal f & Terminal Wealth Relative (TWR): Empirical Sizing Math

Real trading logs contain partial exits, scratch trades, and fat-tailed outliers rather than two neat binary outcomes; derive Ralph Vince Optimal f and largest-loss contract scaling.

By Elena Rostova, FRM, CQFReview not recorded · Updated
Ralph Vince Optimal f & Terminal Wealth Relative (TWR): Empirical Sizing Math — QuantCalc Tools 2026 quantitative & supervisory data card
QuantCalc Tools (Kelly Criterion, Initial/Variation Margin Simulators & Liquidation Math): Ralph Vince TWR/GHPR maximization, stress-padded |W_worst|, and bootstrap Secure f.Source: QuantCalc Tools Institutional Research DeskLicense: CC-BY-4.0

What happens when you export a 400-trade execution log from MT5 or TradingView and try to plug it into the textbook two-state discrete binomial Kelly formula $p - q/R$? You immediately discover that real trade logs are multi-modal empirical distributions—containing +0.1R scratch breakevens, -1.0R normal stops, -2.8R overnight gap losses, and +11.5R home-run runners! Reducing 400 rich empirical P&L outcomes to just two averages (Avg Win and Avg Loss) throws away crucial tail kurtosis! In Portfolio Management Formulas (1990) and The Mathematics of Money Management (1992), Ralph Vince formulated Empirical Optimal f—maximizing exact Geometric Mean Holding Period Return (GHPR) directly over all $N$ historical dollar P&L outcomes normalized by the Largest Historical Loss (|W_worst|)! Benchmark parametric vs empirical fractions in our Kelly Criterion Calculator and Kelly, Margin & Liquidation Hub.

Exact Ralph Vince HPR, TWR, GHPR, and Dollar-Per-Contract Equations#

Let $\{T1, T2, \dots, TN\}$ be the realized net dollar P&L per 1 contract (or per 1 lot, after commissions and slippage) across $N$ historical trades, with strictly negative worst single-trade loss $W{\text{worst}} = \mini Ti < 0$ (so $|W{\text{worst}}| > 0$). For any candidate risk fraction $f \in (0, 1)$, Vince defines trade $i$'s **Holding Period Return (`HPRi(f))**, cumulative **Terminal Wealth Relative (TWR(f))**, and contract count Contracts` as:

Formula or model from the articleNot independently verified
Ralph Vince (1990, 1992) Exact Empirical Optimal f & Contract Sizing Equations:
  • 1. Single-Trade Holding Period Return (Normalized by Worst Loss |W_worst| > 0):
  • HPRi(f) = 1 + f * ( Ti / |W_worst| ) for f in (0, 1)
  • (Note: If f = 1.00 and Wworst recurs, HPRworst = 1 + 1.0*(-1) = 0.00 -> Instant 100% Wipeout!)
  • 2. Terminal Wealth Relative & Geometric Mean HPR across N Trades:
  • TWR(f) = Product{i=1..N} HPRi(f) = Product{i=1..N} [ 1 + f * (Ti / |W_worst|) ]
  • GHPR(f) = [ TWR(f) ]^(1 / N) <=> Maximize L(f) = (1 / N) * Sum{i=1..N} ln( 1 + f * (Ti / |W_worst|) )
  • 3. First-Order Brent/Newton Root for Optimal f* (Unique Concave Peak when Mean(T_i) > 0):
  • L'(f) = Sum{i=1..N} [ Ti / ( |W_worst| + f * T_i ) ] = 0
  • 4. Converting Optimal f* (or Diluted c * f*) into Live Account Dollar Divisor & Contracts:
  • DollarUnitperContract = |Wworst| / fused => Contracts = floor( AccountEquity / DollarUnitper_Contract )
  • Worked 5-Trade Skewed Empirical Audit (Per-Contract PnLs: [+$900, -$300, +$1,800, -$600, -$300]):
  • Mean PnL = +$300/trade (> $0 positive edge!), Worst Loss |W_worst| = $600 (Trade #4):
  • Evaluate TWR(f) = (1 + 1.5f) * (1 - 0.5f)^2 * (1 + 3.0f) * (1 - 1.0f):
  • At f = 0.15 : TWR = (1.225)(0.8556)(1.450)*(0.850) = 1.2919 (GHPR = 1.0526 => +5.26%/trade)
  • At f = 0.236: TWR = 1.3268 (EXACT PEAK Optimal f* = 23.6% => GHPR = 1.0582 => +5.82%/trade)
  • At f = 0.50 : TWR = (1.750)(0.5625)(2.500)*(0.500) = 1.2305 (Over-betting drops TWR & risks -50% on 1 repeat of -$600!)
  • Practical Diluted Quarter-Optimal-f (f_used = 0.25 * 0.236 = 0.059):
  • Dollar_Unit = $600 / 0.059 = $10,169 per contract -> Trade 9 contracts on a $100,000 account!

Single-Trade Tail Shock Under Undiluted Optimal f Versus Quarter-Vince Sizing#

Look closely at Equation 4 (Dollar_Unit = |W_worst| / f_used): if Optimal f* = 0.236, trading undiluted f* = 0.236 means the very next time a loss equal to your historical worst loss -$600 occurs, your account instantly drops -23.6% on ONE single trade—and if the future worst loss is 1.5x worse than your historical sample (-$900), you lose -35.4% in one afternoon!

Four-stage empirical Optimal f calculation from per-contract net trade log to stress-padded Worst Loss, concave log-HPR solver, and Secure f drawdown cap
Four-stage empirical Optimal f calculation from per-contract net trade log to stress-padded Worst Loss, concave log-HPR solver, and Secure f drawdown capVerified Exhibit

Why Sample |W_worst| Is Biased Downward & How Extreme Value Theory Fixes It#

Notice a subtle statistical flaw in raw Ralph Vince Optimal f: the denominator |W_worst| = max_i(-T_i) is a sample maximum over finite $N$ trades! By Extreme Value Theory, the worst loss over your next $1{,}000$ live trades exceeds the worst loss over your first $100$ backtested trades with $90.9\%$ probability ($1000 / (100 + 1000)$)! Institutional systematic desks apply two mandatory upgrades before turning Optimal f live:

  1. Tail-Padded |W_stress|: Replace raw sample |W_worst| with max(1.5 * |W_worst|, Expected_Shortfall_99.5%) modeled via Cornish-Fisher / Generalized Pareto tails.
  2. Leo Zamansky & David Stendahl's Secure f: Maximize GHPR(f) across 10,000 block-bootstrapped resamples of $\{Ti\}$ subject to **95th-percentile Maximum Drawdown $\le D{\text{max}}$** (which seamlessly bridges Vince's empirical framework to drawdown-constrained fractional Kelly!).

Comparison of Position-Sizing Algorithms Across Skewed Empirical Trade Logs#

Compare how four sizing methods handle fat-tailed trade distributions (N = 250 trades, +0.38R expectancy, 1 outlier -4.2R gap loss):

Quantitative Distribution · Typical Undiluted Drawdown
Typical Undiluted Drawdown
012-State Binomial Kelly (p -…$55\%\text{--}8…02Continuous Merton (mu / sig…$45\%\text{--}6…
Position-Sizing MethodUses Full $N$-Trade Distribution?Accounts for -4.2R Outlier Gap?Typical Undiluted DrawdownInstitutional Production Adaptation
2-State Binomial Kelly (p - q/R)No (only p & AvgW/AvgL)No! (Hides -4.2R inside AvgL!)$55\%\text{--}85\%$ (Can hit ruin on gap!)Dangerous on fat-tailed gap logs
Continuous Merton (mu / sigma^2)Uses 1st & 2nd momentsPartial (via inflated $\sigma^2$)$45\%\text{--}65\%$Shrink to 0.25x–0.50x (Half-Kelly)
Ralph Vince Raw Optimal fYes (Every single $T_i$!)Yes (Anchored to `W_worst`!)$50\%\text{--}75\%$ (Unacceptable raw!)Dilute to 0.20x–0.30x + pad `W_worst`
Bootstrap Secure f (Zamansky)Yes (Full path resamples)Yes (`1.5xW_worst` + streaks)Capped at Mandated $15\%\text{--}20\%$!Best-in-class empirical single-system sizing
Terminal Wealth Relative and 95th-percentile Bootstrap Maximum Drawdown across fractions of Ralph Vince Optimal f from 0.2x to 1.2x
Terminal Wealth Relative and 95th-percentile Bootstrap Maximum Drawdown across fractions of Ralph Vince Optimal f from 0.2x to 1.2xVerified Exhibit

Always Verify Dollar_Unit >= Exchange Initial Margin!#

Suppose Secure f outputs Dollar_Unit = $12,000 per E-mini S&P (ES) contract, while CME SPAN 2 Speculator Initial Margin is $15,400 per contract! Exchange Initial Margin is a hard physical floor: your executable divisor is max(Dollar_Unit_Secure_f, 1.5 * Exchange_IM_per_Lot). Cross-verify risk of ruin in Markov absorption vs fixed-fractional ruin math and CAGR volatility drag.

Frequently asked questions

When does Ralph Vince Optimal f give the EXACT same answer as the textbook Discrete Kelly formula?

Whenever your empirical trade log consists solely of two discrete outcomes (+b dollars on fraction p of trades, and -a dollars on fraction q = 1 - p of trades)! In that binary case, |W_worst| = a, and setting Vince’s derivative L’(f) = p*(b/a)/(1 + f*b/a) - q/(1 - f) = 0 solves identically to f* = p - q / (b/a)!

Should I feed percentage returns or 1-contract dollar PnLs into Optimal f for futures and FX?

For futures and FX lots, feed price-normalized 1-contract constant-point/percentage PnLs (or constant $100k notional PnLs)! Why? Because a -$500 point move when Gold was $1,200/oz (-4.2%) is twice as severe in percentage volatility as a -$500 move when Gold is $2,600/oz (-1.9%)—always normalize T_i to current contract price!

What is Ralph Vince’s "Leverage Space Portfolio Model" (2009)?

Vince extended univariate Optimal f to M simultaneous trading systems by maximizing joint N-period geometric HPR over the M-dimensional hypercube f = (f_1, ..., f_M) in [0, 1]^M subject to a global path drawdown probability constraint—the exact non-parametric empirical cousin of Multi-Asset Drawdown-Constrained Kelly!

How does Ryan Jones’s "Fixed Ratio" position sizing differ from Ralph Vince’s Fixed Fractional Optimal f?

Fixed Fractional (Optimal f) scales contract count linearly with equity (Contracts = Equity / Dollar_Unit), whereas Ryan Jones’s Fixed Ratio scales contract count with the SQUARE ROOT of accumulated profits via parameter Delta—growing faster on small accounts and automatically de-leveraging as equity becomes large!

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). Ralph Vince Optimal f & Terminal Wealth Relative (TWR): Empirical Sizing Math. QuantCalc Tools. https://quantcalctools.com/guides/ralph-vince-optimal-f-empirical-trade-distribution-largest-loss
Explore Kelly Margin Leverage Sizing Hub (20) →

QuantCalc Tools Data, Benchmarks & Contributor Records

Ralph Vince Optimal f & Terminal Wealth Relative Formula | QuantCalc Tools