The Kelly Criterion: Why Parallel Bets Increase Risk

The Kelly criterion is one of the most elegant results in probability theory and finance. It tells us exactly how much of our capital to risk on a favorable bet to maximize long-term wealth. However, there’s a critical subtlety: the formula assumes sequential betting. When you run multiple bets in parallel, the risk profile changes dramatically.

In this post, I’ll derive the Kelly criterion and demonstrate mathematically why parallel positions require a more conservative approach.

The Kelly Criterion Formula

Suppose you have a favorable bet with:

• Probability of winning: $p$
• Probability of losing: $q = 1 - p$
• Win pays $b$ to 1 (e.g., $b = 1$ means you double your money)
• Loss loses your entire stake

The Kelly criterion tells us to bet a fraction $f^*$ of our capital, where:

$f^* = \frac{bp - q}{b} = \frac{bp - (1-p)}{b}$

For the special case where $b = 1$ (even-money bet), this simplifies to:

$f^* = 2p - 1$

Derivation

The Kelly criterion maximizes the expected logarithm of wealth. After one bet with fraction $f$ of capital:

$\mathbb{E}[\log(W)] = p \log(1 + bf) + q \log(1 - f)$

To find the maximum, we differentiate with respect to $f$ and set to zero:

$\frac{d}{df}\mathbb{E}[\log(W)] = \frac{pb}{1 + bf} - \frac{q}{1 - f} = 0$

Solving for $f$:

$\frac{pb}{1 + bf} = \frac{q}{1 - f}$

$pb(1 - f) = q(1 + bf)$

$pb - pbf = q + qbf$

$pb - q = pbf + qbf$

$bp - q = f(pb + qb)$

$f^* = \frac{bp - q}{b(p + q)} = \frac{bp - q}{b}$

This is the Kelly fraction.

Why Logarithmic Growth?

The Kelly criterion maximizes $\mathbb{E}[\log(W)]$, not $\mathbb{E}[W]$. Why?

The logarithm makes the criterion focus on geometric growth rather than arithmetic growth. Over $n$ independent bets, your wealth becomes:

$W_n = W_0 \prod_{i=1}^{n} (1 + R_i)$

where $R_i$ is the return on bet $i$. Taking logarithms:

$\log(W_n) = \log(W_0) + \sum_{i=1}^{n} \log(1 + R_i)$

By the law of large numbers, $\frac{1}{n}\sum_{i=1}^{n} \log(1 + R_i) \to \mathbb{E}[\log(1 + R)]$, so:

$W_n \approx W_0 \exp(n \cdot \mathbb{E}[\log(1 + R)])$

Maximizing $\mathbb{E}[\log(1 + R)]$ therefore maximizes the long-run growth rate.

The Parallel Betting Problem

The Kelly criterion assumes you make bets sequentially: bet, observe outcome, reinvest, repeat. But what if you have $N$ simultaneous positions?

Consider the extreme case: if you take the full Kelly fraction on $N$ independent bets simultaneously, your total exposure is $N \cdot f^*$. If $N$ is large enough, you can easily exceed 100% of your capital — guaranteed ruin!

Even with smaller $N$, the probability of simultaneous losses creates fat tails in your return distribution. The sequential Kelly formula doesn’t account for this correlation in timing.

The Adjustment

A common heuristic is to divide the Kelly fraction by the number of parallel bets:

$f_{\text{parallel}}^* = \frac{f^*}{N}$

or more conservatively, by the square root:

$f_{\text{parallel}}^* = \frac{f^*}{\sqrt{N}}$

But the exact optimal fraction depends on the correlation structure.

Theoretical Analysis of Parallel Betting Risk

1. Sequential Kelly: The Ideal Case

With sequential Kelly betting and favorable odds (say $p = 0.55$, giving a Kelly fraction of 10%), you can never lose your entire capital in one round. The wealth grows exponentially according to the theoretical growth rate $g = \mathbb{E}[\log(1 + R)]$, and the risk of ruin approaches zero with proper Kelly sizing.

2. Parallel Bets Increase Bankruptcy Risk

When you naively apply the Kelly fraction to multiple parallel bets, bankruptcy risk increases dramatically. Consider what happens with $N$ parallel bets each using fraction $f^*/N$:

• With 2 parallel bets: The correlation in timing introduces slight additional risk
• With 5 parallel bets: The risk compounds significantly
• With 10 parallel bets: A synchronized run of losses across positions creates serious ruin risk

The problem is that multiple simultaneous losses can compound. Even though each individual bet has the “right” Kelly sizing, the combined exposure creates dangerous scenarios that the sequential formula doesn’t account for.

3. Variance Scales Unfavorably

The variance of returns from $N$ independent parallel bets grows as:

$\text{Var}\left(\sum_{i=1}^{N} R_i\right) = N \cdot \text{Var}(R)$

So the standard deviation of combined returns scales as $\sqrt{N}$. Meanwhile, the expected return grows linearly with $N$:

$\mathbb{E}\left[\sum_{i=1}^{N} R_i\right] = N \cdot \mathbb{E}[R]$

The coefficient of variation therefore increases as $\sqrt{N}$, meaning relative volatility grows with the number of positions. This creates:

• Sequential: Low variance, smooth exponential growth
• 5 Parallel: Higher variance, choppy paths with wider swings
• 10 Parallel: Extreme variance, fat-tailed outcomes

4. Lower Growth Rates Despite Same Edge

Even with the same theoretical edge per bet, parallel strategies achieve lower long-run growth rates. This is because of volatility drag: the increased variance in returns reduces the geometric mean, even when the arithmetic mean stays constant.

For a random variable with mean $\mu$ and variance $\sigma^2$, by Jensen’s inequality:

$\mathbb{E}[\log(1 + R)] < \log(1 + \mathbb{E}[R])$

The gap widens as variance increases. So parallel bets, despite having the same expected return per position, suffer from increased path dependence and volatility drag.

5. Fractional Kelly as a Solution

Using a fractional Kelly approach (e.g., 50% of Kelly) with parallel positions dramatically reduces risk while maintaining reasonable returns. This is the classic risk-return tradeoff:

• Full Kelly with parallel bets: Higher variance, increased ruin risk, lower realized growth
• Fractional Kelly (25-50%): Lower variance, near-zero ruin risk, smoother growth trajectory

Mathematical Intuition: Why This Happens

The key is that Kelly maximizes $\mathbb{E}[\log(1 + R)]$ for a single bet. For $N$ parallel bets, your wealth after one round is:

$W_{\text{new}} = W_{\text{old}} + \sum_{i=1}^{N} f \cdot W_{\text{old}} \cdot R_i$

$= W_{\text{old}}\left(1 + f\sum_{i=1}^{N} R_i\right)$

The sum $\sum R_i$ has variance proportional to $N$ (for independent bets), so the wealth volatility increases with $\sqrt{N}$. But the expected return only scales linearly with $N$.

The Kelly criterion balances expected return against log-variance. When you increase variance faster than expected return, you get suboptimal growth — and risk of ruin.

Practical Implications

For Traders and Investors

1. Don’t naively apply Kelly to a portfolio: If you have 20 uncorrelated positions, using Kelly on each would mean betting 200% of your capital (with $f^* = 10\%$ each). Absurd!

2. Adjust for correlation: If your positions are correlated, the effective number of “independent” bets is lower. You need to account for correlation structure.

3. Use fractional Kelly: Many professional traders use 25-50% of Kelly as a practical heuristic. This provides a safety margin against model error and parallel position risk.

4. Monitor total exposure: Always track $\sum f_i$, the total fraction of capital at risk across all positions.

For Gamblers

If you’re counting cards at multiple blackjack tables, or betting on multiple sports games simultaneously, the same logic applies. Your effective Kelly fraction must be divided across your positions.

Extensions and Further Reading

The Kelly criterion has been extended in many directions:

• Continuous-time Kelly (for Brownian motion models)
• Kelly with leverage constraints
• Multi-asset Kelly portfolios (Markowitz meets Kelly)
• Robust Kelly (accounting for parameter uncertainty)

A particularly elegant result is the continuous-time Kelly formula for a stock following geometric Brownian motion:

$f^* = \frac{\mu - r}{\sigma^2}$

where $\mu$ is expected return, $r$ is the risk-free rate, and $\sigma^2$ is variance. Notice how volatility appears in the denominator — higher volatility means bet less.

Conclusion

The Kelly criterion is a powerful tool, but it comes with an important caveat: it’s derived for sequential, independent bets. When you have parallel positions:

1. Bankruptcy risk increases dramatically
2. Volatility increases faster than expected return ($\sqrt{N}$ vs. $N$)
3. Growth rates suffer due to volatility drag and path dependence
4. Fractional Kelly becomes essential for survival

The mathematics is clear: naively applying full Kelly to parallel positions can increase ruin risk from near-zero to dangerous levels, even though each individual bet has positive expected value. A portfolio of 10 parallel positions using full Kelly on each could easily exceed 100% capital allocation — an obvious path to ruin.

The lesson: context matters. A formula that’s optimal in one setting (sequential) can be dangerous in another (parallel). Always understand the assumptions behind your models.

---

This analysis explores the mathematical foundations of the Kelly criterion and its limitations. For practical applications in trading and portfolio management, always account for correlation structure and consider using fractional Kelly (25-50%) as a safety margin.