Thorp Ch. 1: The Kelly Criterion — Mathematics of Edge & Optimal Sizing
阅读中文版Derivation of the Kelly Criterion, geometric compounding maximization, fractional Kelly sizing, and why over-betting leads to guaranteed ruin.
🔊 Listen to Article (Chinese Audio)
Thorp Quantitative Ch. 1: The Kelly Criterion — Mathematics of Edge & Sizing
"If you bet too much, even with a statistical edge on every single hand, the demon of geometric compounding will eventually reduce you to zero." — Edward O. Thorp
Why the Quant Revolution Began with Thorp
In the 1960s, long before Wall Street embraced complex algorithms, mathematician Edward O. Thorp used an IBM 704 mainframe to launch modern quantitative finance. Thorp's breakthrough was not about predicting macroeconomic trends; it was about using probability theory and capital allocation mechanics to extract deterministic geometric compounding from positive-expectancy bets.
While fundamental value investing analyzes business moats, quantitative investing requires answering an exact mathematical question: Given a quantified edge, exactly what percentage of capital should be allocated to maximize growth without risking ruin?
Betting too little yields suboptimal compounding; betting too much guarantees eventual insolvency due to geometric volatility drag. The Kelly Criterion is the foundational mathematical solution.
Mathematical Derivation of the Kelly Formula
Let $p$ be the probability of winning, $q = 1 - p$ the probability of losing, and $b$ the odds ratio (net profit per dollar wagered).
For initial wealth $V_0$ and fixed betting fraction $f$, wealth after $N$ trials with $W = Np$ wins and $L = Nq$ losses is:
$$V_N = V_0 (1 + bf)^W (1 - f)^L$$
Maximizing the asymptotic growth rate $G(f) = \lim_{N \to \infty} \frac{1}{N} \ln(V_N / V_0)$:
$$G(f) = p \ln(1 + bf) + q \ln(1 - f)$$
Taking the first derivative with respect to $f$ and setting it to zero:
$$\frac{dG}{df} = \frac{pb}{1 + bf} - \frac{q}{1 - f} = 0 \implies f^* = \frac{pb - q}{b}$$
| Parameter | Meaning | Financial Market Equivalent |
|---|---|---|
| $p$ | Win rate | Out-of-sample verified strategy accuracy |
| $q$ | Loss rate | $1 - p$ |
| $b$ | Win/loss payoff | Average winning trade / Average losing trade |
| $f^*$ | Optimal fraction | Maximum portfolio allocation fraction |
The Over-Betting Trap
- Under-betting ($f < f^*$): Suboptimal compounding speed, zero probability of ruin.
- Full Kelly ($f = f^*$): Maximum theoretical growth rate, but severe drawdown volatility (>50%).
- Over-betting ($f^ < f < 2f^$): Higher risk with strictly lower returns.
- Ruin Zone ($f \ge 2f^*$): Negative geometric growth leading to 100% mathematical ruin over time.
Practical Quant Rules
- Always use Fractional Kelly (0.25x to 0.50x): Mitigates model parameter uncertainty and fat-tailed shocks.
- Never exceed Total Portfolio Kelly sum > 1.0: Enforce cross-asset correlation haircuts.