Kelly Criterion Formula

Quick Reference

Formula matematica per calcolare leverage ottimale che massimizza crescita geometrica del capitale.

Formula Base Kelly

f* = (r - b) / s²

Variabili

  • f*: Optimal leverage factor
  • r: Expected annual return (senza leverage)
  • b: Borrowing rate (risk-free rate)
  • s: Annual standard deviation of returns

Optimal Volatility Target

Versione piu utile per trading:

Optimal Vol Target = (r - b) / s = SR

Dove SR = Sharpe Ratio

In parole: Optimal volatility target = Expected Sharpe Ratio

Half-Kelly (Raccomandato)

Half-Kelly Target = SR / 2

Con aggiustamenti realistici:

Recommended Target = (Backtest SR × Discount Factor) / 2

Discount Factors

  • Out-of-sample bootstrap: 0.75
  • Rolling window: 0.60
  • In-sample only: 0.25

Esempio Numerico Completo

Step 1: Backtest Results

Sistema trend-following: - Backtest SR = 1.0 - Out-of-sample bootstrap - Zero skew

Step 2: Apply Discount

Realistic SR = 1.0 × 0.75 = 0.75

Step 3: Half-Kelly

Target Vol = 0.75 / 2 = 0.375 = 37.5%

Raccomandazione finale: 35-40% target volatility

Formula Derivation (Semplificata)

Massimizzare geometric mean growth rate:

Obiettivo:

Maximize: E[log(1 + return)]

Con leverage f:

Return = f × (r - b) - f × volatility shocks

Soluzione:

f_optimal = (r - b) / s²

Optimal risk:

Risk_optimal = f × s = (r - b) / s = SR

Negative Skew Adjustment

Per strategie con negative skew (~-2):

Target Vol = (SR / 2) / 2 = SR / 4

Esempio: - Realistic SR = 0.50 - Negative skew strategy - Target = 0.50 / 4 = 12.5%

Expected Return Calculation

Date leverage/volatility target, expected return:

Expected Return = SR × Vol Target

Esempi:

SR Vol Target Expected Return
0.25 12.5% 3.1%
0.50 25% 12.5%
0.75 37.5% 28%
1.0 50% 50%

Questi sono geometric means attesi.

Overconfidence Example

Trader overconfident: - Crede backtest SR = 2.0 (nessun discount!) - Full Kelly: Target = 200% - Expected return = 400% annuo

Realta: - True SR = 0.5 (delusione!) - Con 200% vol e SR 0.5: - Molto probabile rovina

Lesson: SEMPRE discount backtest e usa Half-Kelly.

Kelly per Diversi Trader Types

Staunch Systems Trader

Target = (Realistic SR × 0.75) / 2  [zero skew]
Target = (Realistic SR × 0.75) / 4  [negative skew]

Max realistic SR assumed: 1.0

Semi-Automatic Trader

Target = (Expected SR) / 2  [positive/zero skew]
Target = (Expected SR) / 4  [negative skew]

Max safe expected SR: 0.5

Asset Allocator

Target = Expected SR / 2

Max realistic SR: 0.4 (large diversified portfolio)

Probability of Ruin

Con optimal Half-Kelly e SR 0.5, 10 anni:

Vol Target Prob Lose 50% Prob Lose 90%
12.5% (1/4 Kelly) <1% <1%
25% (Half Kelly) 10% 1.1%
50% (Full Kelly) 40% 22%
100% 93% 88%

Nota: Anche Half-Kelly ha 10% prob di perdere meta capitale!

Continuous Rebalancing

Kelly richiede continuous position adjustment:

New Position = Current Capital × (SR / Price × σ%)

Questo è rolling up profits and losses.

Non rebalancing → rischio di rovina aumenta drammaticamente.

Kelly con Transaction Costs

Costi riducono optimal Kelly:

Effective SR = Raw SR - (Annual Costs / Vol Target)

Usa effective SR per calcolo target.

Esempio: - Raw SR = 0.60 - Costs = 0.10 SR units - Effective SR = 0.50 - Half-Kelly target = 25% (non 30%)

Fractional Kelly

General formula per frazione f del Kelly:

Target Vol = (Fraction) × SR
  • f = 0.25: Quarter-Kelly
  • f = 0.50: Half-Kelly
  • f = 1.00: Full Kelly

Raccomandazioni: - Retail: 0.25-0.50 - Professional: 0.25-0.50 (!!) - Overconfident: 1.0+ (bankruptcy soon)

Verifica Formula

Sanity check dopo calcolo:

Implied Leverage = Target Vol / Natural Vol

Red flags: - Implied leverage > 50: Strumento troppo low-vol - Target vol > 50%: Troppo aggressivo per most traders - Expected return > 100%: Sei overconfident

Alternative: Fixed Fraction

Se non hai SR estimate, usa fixed conservative fraction:

Ultra-conservative: 10-12% vol target Conservative: 15-20% vol target Moderate: 20-25% vol target

Errors Common

  • Using full Kelly: Recipe for ruin
  • No discount to backtest: Overconfidence
  • Ignoring skew: Negative skew needs even lower target
  • Not rebalancing: Kelly assumes continuous adjustment
  • Wrong SR units: Confondere daily con annual SR
  • Forgetting costs: Use net SR after costs

Concetti Correlati

  • [[Sharpe Ratio]] - input chiave per Kelly
  • [[Volatility Targeting]] - implementazione Kelly in practice
  • [[Geometric Mean]] - Kelly massimizza questo
  • [[Position Sizing]] - output del Kelly criterion
  • [[Skew]] - richiede adjustment al Kelly