--- title: 'Kelly Criterion: Theory & Applications' url: https://www.emergentmind.com/topics/kelly-rule type: topic --- # Kelly Criterion: Theory & Applications to=arxiv_search tool code: {"query":"Kelly criterion rule stop-loss distributionally robust Wasserstein Kelly prediction markets quantile Kelly arXiv", "max_results": 10} to=arxiv_search tool code: {"query":"Kelly criterion arXiv 1311.2550 1603.06183 1812.10371 2302.13979 1201.6655 2604.17577", "max_results": 10} The Kelly Rule, or Kelly criterion, is a prescription for allocating fractions of wealth across repeated gambles or portfolios so as to maximize expected logarithmic wealth and, under multiplicative wealth dynamics, the long-run growth rate. In the canonical binary gamble with win probability \(p\), loss probability \(1-p\), and net odds \(b\), the optimal fraction is \(f^*=\frac{bp-(1-p)}{b}\); in general portfolio form it is the optimization of \(\mathbb{E}[\log(1+f^\top R)]\), and in continuous-time geometric Brownian motion with risky drift \(\mu\), volatility \(\sigma\), and risk-free rate \(r\) it yields the constant risky weight \((\mu-r)/\sigma^2\) [1710.01786; 2508.18868; 1311.2550]. Modern work treats the rule as a growth-optimal principle that extends to continuous-time control, drawdown constraints, ambiguity sets, finite-horizon quantiles, prediction markets, and several non-financial settings. ## 1. Formal definition and canonical formulations In its most general single-period form, the Kelly problem is \[ \max \mathbb{E}[\log W_{t+1}], \] with the feasible portfolio or betting rule chosen so that wealth remains nonnegative [2508.18868]. In repeated IID gambles, this is equivalent to maximizing the expected logarithmic growth rate per play, which is the standard long-run interpretation of the criterion [1710.01786]. For a vector of random returns \(X\in\mathbb{R}^n\), allocation vector \(f\in\mathbb{R}^n\), and wealth recursion \(V(k+1)=(1+f^\top X(k))V(k)\), the admissible set is determined by the survival condition \(1+f^\top x\ge 0\) for every \(x\) in the support; the optimization becomes \[ f^*(P)\in \arg\max_{f\in\mathcal{F}} \mathbb{E}_P[\log(1+f^\top X)]. \] This support-aware formulation is central whenever rare outcomes can force bankruptcy [1710.01786]. Several equivalent parameterizations are standard. In the long-only gambling model of Busseti, Ryu, and Boyd, one chooses \(b\ge 0\) with \(\mathbf{1}^\top b=1\) to maximize \(\mathbb{E}[\log(r^\top b)]\), where \(r\) is a nonnegative return vector and one component is risk-free cash [1603.06183]. In discrete-time stock-market approximations with excess returns \(\tilde r\), a second-order expansion gives the moment system \[ M f=b,\qquad M=\mathbb{E}[\tilde r\tilde r^\top],\qquad b=\mathbb{E}[\tilde r], \] so that \(f^*=M^{-1}b\) in the unconstrained approximation; for Gaussian excess returns this becomes \((\Sigma+\mu\mu^\top)^{-1}\mu\) [1806.05293]. For a single multinomial event with subjective probabilities \(p_i\), state prices \(q_i\), explicit stakes \(x_i\), and cash fraction \(c\), the full Kelly solution can be written in state-price form as \[ x_i=(p_i-cq_i)_+, \] with the activity threshold \(p_i/q_i>c\). In this representation, cash acts as an implicit position in every outcome, and the optimal terminal wealth satisfies \(w_i^*=\max\{c,p_i/q_i\}\) [2603.13581]. ## 2. Continuous-time growth optimality and dynamic control In continuous time, the standard Kelly problem is embedded in a market with one risky asset \(S\) and one risk-free asset \(B\), where \[ dS_t=S_t(\mu\,dt+\sigma\,dW_t),\qquad dB_t=rB_t\,dt. \] With discounted wealth \(\pi_t\) and risky fraction \(\alpha(\pi_t,t)\), the discounted wealth dynamics are \[ \frac{d\pi_t}{\pi_t}=\alpha(\pi_t,t)\big[(\mu-r)\,dt+\sigma\,dW_t\big]. \] For log utility \(g(\pi_T)=\log\pi_T\), the Hamilton–Jacobi–Bellman equation yields the classical constant Kelly fraction \[ \alpha_K=\frac{\mu-r}{\sigma^2}, \] together with the value function \[ J(\pi,t)=\log\pi+\frac{(\mu-r)^2}{2\sigma^2}(T-t). \] The paper "The Kelly growth optimal strategy with a stop-loss rule" shows that the optimal control itself satisfies a nonlinear PDE independent of the terminal utility, \[ \partial_t\alpha(\pi,t)=-\frac{\sigma^2}{2}\alpha(\pi,t)^2\,\partial_\pi\!\left(\pi^2\partial_\pi\alpha(\pi,t)\right), \] or, in terms of \(\gamma(\pi,t)=\pi\alpha(\pi,t)\), \[ \partial_t\gamma(\pi,t)=-\frac{\sigma^2}{2}\gamma(\pi,t)^2\,\partial_\pi^2\gamma(\pi,t). \] This makes the unconstrained Kelly rule a trivial constant solution and permits direct numerical solution once boundary conditions for \(\alpha\) are known [1311.2550]. The same continuous-time logic appears in leverage problems. With broker funding cost \(r\), client borrowing rate \(r_L\), risky weight \(b\), and risky-asset GBM \(dS_t/S_t=\mu\,dt+\sigma\,dW_t\), the instantaneous expected log-growth is \[ g(b;r_L)=r_L+b(\mu-r_L)-\frac{1}{2}b^2\sigma^2. \] Garivaltis shows that if the broker and the Kelly gambler Nash-bargain jointly over leverage and the margin rate, then the investor chooses the risky position as if borrowing at the broker’s call rate, \[ b^*=\frac{\mu-r}{\sigma^2}, \] while the negotiated lending rate depends on the threat point; under zero margin loans as the disagreement point, \[ r_L^*=\frac{3}{4}r+\frac{1}{4}\left(\nu-\frac{1}{2}\sigma^2\right), \qquad \nu=\mu-\frac{1}{2}\sigma^2. \] This preserves growth-optimal sizing while reallocating surplus between borrower and broker [1904.06628]. ## 3. Constraints, drawdown control, and finite-horizon modifications A large contemporary literature modifies the Kelly Rule because pure log-growth optimality can tolerate pathwise risk that is unacceptable operationally. Busseti, Ryu, and Boyd formulate a risk-constrained Kelly problem by adding a drawdown surrogate to the classical program. If \(W_{\min}=\inf_{t\ge 1}w_t\) and the target is \(\mathbb{P}(W_{\min}<\alpha)\le \beta\), they prove that, for \(\lambda=\log\beta/\log\alpha\), \[ \mathbb{E}\big[(r^\top b)^{-\lambda}\big]\le 1 \quad\Longrightarrow\quad \mathbb{P}(W_{\min}<\alpha)<\beta. \] Replacing the original path-dependent constraint by this bound gives the convex Risk-Constrained Kelly program \[ \max \mathbb{E}[\log(r^\top b)] \quad\text{s.t.}\quad \mathbf{1}^\top b=1,\; b\ge 0,\; \mathbb{E}[(r^\top b)^{-\lambda}]\le 1. \] The same paper shows that a quadratic approximation links this formulation to a Markowitz-type mean-variance program [1603.06183]. Periodic stop-loss rules alter the dynamic allocation even in the continuous-time GBM benchmark. If discounted wealth hits a boundary \(\pi_c\), the risky weight is set to zero until the next reset date. In scaled variables \(u=\alpha/\alpha_K\), \(z=\pi_c/\pi\), and \(\theta=(T-t)/\tau\) with \(\tau=2\sigma^2/(\mu-r)^2\), the control PDE becomes \[ \partial_\theta u(z,\theta)=u(z,\theta)^2 z^2 \partial_z^2 u(z,\theta), \] with boundary conditions \(u(0,\theta)=1\), \(u(1,\theta)=0\), and \(u(z,0)=1\). The long-horizon asymptotic is \(u(z,\theta)\to 1-z\), giving the CPPI-like limit \(\alpha(\pi)\to \alpha_K(1-\pi_c/\pi)\) [1311.2550]. Finite-horizon objectives can depart sharply from ordinary log-growth maximization. "Exact Finite-Horizon Quantile Kelly for Repeated Multi-Outcome Events" considers an \(m\)-outcome event repeated \(n\) times in Arrow–Debreu coordinates \(W\), where terminal wealth is the monomial \(X_n(W)=W^N\) for multinomial count vector \(N\). For fixed upper quantile level \(\alpha\), the terminal-wealth quantile is a positively homogeneous piecewise-monomial function on the closed wealth simplex; on each chamber of the multinomial count arrangement, the problem reduces exactly to a one-period Kelly problem for the shadow law \(k/n\). The unique unconstrained one-period Kelly point is \[ W_i^{\mathrm K}=\frac{p_i}{q_i}, \] but exact finite-horizon quantile maximizers can lie on support faces and need not equal \(W^{\mathrm K}\); asymptotically, however, \[ \sup_W G_{n,\alpha}(W)\to \sum_{i=1}^m p_i\log\frac{p_i}{q_i}, \qquad W_n^{(\alpha)}\to W^{\mathrm K}. \] This suggests that finite-horizon quantile control is a genuine alternative objective rather than a perturbation of ordinary Kelly [2604.17577]. A simpler departure from the constant-payoff model occurs when the payoff \(B\) on a winning round is itself random. Then the expected log-growth is \[ g(f)=(1-p)\log(1-f)+p\,\mathbb{E}[\log(1+fB)], \] and the Kelly fraction is characterized by the integral equation \[ p\,\mathbb{E}\!\left[\frac{B}{1+f^*B}\right]=\frac{1-p}{1-f^*}. \] The paper proves that this \(f^*\) is strictly smaller than the classical Kelly fraction computed using the same average payoff \(\mathbb{E}[B]\), unless \(B\) is almost surely constant [1411.3615]. ## 4. Estimation risk, ambiguity, and the problem of overconfidence A recurrent criticism of the Kelly Rule is not its objective but its dependence on a fully specified return law. One line of work argues that under some parametric models the criterion is not too aggressive but too conservative. The "Restricted Betting Theorem" states that any Kelly optimizer \(f^*(P)\) under true support \(\mathcal X\) must satisfy \[ h_{\mathcal X}(-f^*(P))\le 1. \] In the scalar case this confines the optimizer to \[ f^*(P)\in\left[-\frac{1}{X_{\max}},-\frac{1}{X_{\min}}\right]. \] If the support is unbounded above and below, as with a normal distribution, then the theorem forces \(f^*(P)=0\). The paper’s argument is that the true-support solution can therefore be much smaller than the empirical Kelly solution \(\hat f_m\), because finite samples truncate tails and expand the empirical admissible set [1710.01786]. A second response is to replace the nominal distribution by an ambiguity set and maximize worst-case log-growth. Sun and Boyd define the distributionally robust Kelly problem \[ \max_{b\in\mathcal B}\min_{\pi\in\mathcal P}\sum_{i=1}^m \pi_i\log(b^\top r_i), \qquad \mathcal B=\{b\in\mathbb R_+^n:\mathbf 1^\top b=1\}, \] and show that it remains a convex optimization problem. They provide tractable formulations for polyhedral sets, ellipsoidal sets, \(\phi\)-divergence balls, Wasserstein balls, and moment-based uncertainty sets, and extend Breiman-type asymptotic optimality to varying distributions constrained to lie in a fixed uncertainty set [1812.10371]. The Wasserstein variant of this idea is developed further in "Wasserstein-Kelly Portfolios." Rather than robustifying directly over simple returns, that paper places the Wasserstein ball on log-returns \(r\), so that the robust objective becomes \[ \max_{w\in\mathcal W}\min_{Q\in\mathcal P}\mathbb E_Q[\log(\exp(r)^\top w)], \qquad \mathcal W=\{w\ge 0:\mathbf 1^\top w=1\}. \] The modeling shift avoids the \(-\infty\) pathologies that arise if the adversary can place mass on outcomes with nonpositive gross return. The resulting problem admits a finite-dimensional convex reformulation, and the reported S&P 500 experiments show greater stability and better out-of-sample performance than empirical Kelly across multiple metrics [2302.13979]. Estimation risk can also be addressed by enlarging the investment menu. In a discrete-time binomial stock–bond market with a European put, the paper "Tackling estimation risk in Kelly investing using options" shows that fairly priced options are redundant under correct specification, but can be used to construct a class of Kelly portfolios that is asymptotically robust to misspecification of the up/down move. Two Kelly-with-option portfolios with opposite hedges can be convexly combined so that the long-run growth rate converges to that of the better constituent under the realized model [2508.18868]. ## 5. Aggregation, market applications, and institutional implementations When traders themselves follow Kelly, the rule induces an aggregation mechanism. In a prediction market security priced at \(\pi\) and paying \(1\) on the event, an agent with belief \(p_i\) and wealth \(w_i\) chooses demand \[ x_i^*=w_i\,\frac{p_i-\pi}{\pi(1-\pi)}. \] Market clearing implies \[ \pi=\sum_i w_i p_i, \] so the equilibrium price is the wealth-weighted average of beliefs. After the outcome is realized, wealth updates exactly as if the market were applying Bayes’ Law to the index of traders, and the market’s cumulative log loss satisfies the worst-case regret bound \[ L_M\le \min_i L_i + \ln\!\big(1/w_i(0)\big). \] With fractional Kelly, the same paper shows that agent \(i\) behaves like a full-Kelly trader with shrunk belief \(\tilde p_i=\lambda p_i+(1-\lambda)\pi\), and the market price converges to a time-discounted frequency rather than the raw empirical frequency [1201.6655]. Institutional implementations often combine Kelly sizing with additional state variables. In systematic short-dated SPXW put-writing, Kelly sizing is estimated by Monte Carlo under GBM using empirical win probabilities and average gain/loss magnitudes, then translated into contract counts through a margin model. The study compares three schemes—pure Kelly, VIX-rank scaling, and a multiplicative hybrid—and reports that ultra-short-dated, far out-of-the-money options deliver superior risk-adjusted returns, while the hybrid rule better balances return generation and drawdown control, particularly in low-volatility conditions such as 2024 [2508.16598]. The rule is also routinely approximated in feedback-control language. In the single-asset recursion \[ V(k+1)=V(k)(1+KX(k)), \] the exact Kelly objective is \(g(K)=\mathbb E[\log(1+KX)]\). A second-order Taylor approximation gives \[ K_{\mathrm{approx}}^*=\frac{\mu}{\mu^2+\sigma^2}, \qquad \mu=\mathbb E[X],\quad \sigma^2=\mathrm{Var}(X), \] together with closed forms for expected cumulative gain, its variance, and a survivability interval \( -1/X_{\max}