Paper: arXiv 2503.07498
Authors: Vladimir Markov
Abstract
We examine the problem of optimal portfolio allocation within the framework of utility theory. We apply exponential utility to derive the optimal diversification strategy and logarithmic utility to determine the optimal leverage. We enhance existing methodologies by incorporating compound probability distributions to model the effects of both statistical and non-stationary uncertainties. Additionally, we extend the maximum expected utility objective by including the variance of utility in the objective function, which we term generalized mean-variance. In the case of logarithmic utility, it provides a natural explanation for the half-Kelly criterion, a concept widely used by practitioners.
Complexity vs Empirical Score
- Math Complexity: 8.5/10
- Empirical Rigor: 2.0/10
- Quadrant: Lab Rats — high math complexity, limited empirical testing
Why this score: The paper is heavily theoretical, using advanced mathematical frameworks like Bayesian posterior predictive distributions, compound distributions, and deriving closed-form solutions for various utility functions. However, it lacks any implementation details, backtesting results, or empirical data validation, focusing solely on theoretical derivations.
Research Flowchart
flowchart TD
A["Research Goal: Optimal Portfolio Allocation via Utility Theory"] --> B["Methodology: Apply Utility Functions & Compound Distributions"]
B --> C{"Input: Asset Returns & Uncertainties"}
C --> D["Process: Exponential Utility for Diversification"]
C --> E["Process: Logarithmic Utility for Leverage"]
D --> F["Outcome: Maximizing Expected Utility"]
E --> F
F --> G["Finding: Natural explanation of Half-Kelly Criterion"]
F --> H["Outcome: Generalized Mean-Variance Extension"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes a unified utility‑based framework that uses exponential (CARA) utility for optimal diversification and logarithmic utility within a generalized mean‑variance (GMV) objective to obtain optimal leverage, incorporating a compound distribution of outcomes and Bayesian marginalization.
Source sentence
In this paper, we propose a few enhancements to the existing methodologies. By accounting for the variance of utility, we extend the maximum expected utility objective into a generalized mean-variance framework.
Assumes 6
- Asset returns follow a Gaussian (or ALD) distribution with uncertain mean and covariance, modeled via hyper‑parameters and a compound distribution. model
Source sentence
This approach introduces hyperparameters that allow us to model both statistical noise and non-stationary effects in the expected returns and covariance matrices. In the stationary setting, we model univariate returns as r ∼N(µ, σ2) with µ ∼N(µpd, σ2 pd), where µpd and σ2 pd are the posterior mean and variance, respectively.
- Future wealth evolves according to geometric Brownian motion (GBM) when using logarithmic utility. model
Source sentence
Over multiple periods, traders face the risk of ruin inherent in multiplicative processes like geometric Brownian motion (GBM). Even with a positive mean, GBM causes most trajectories to trend toward zero wealth.
- Portfolio allocation is performed sequentially by optimizing only the next time slot and re‑optimizing after each observation. execution
Source sentence
Consequently, we propose to optimize only the next time slot while accounting for the risks ahead. After executing each allocation, we observe the outcome, update the data, and re-optimize.
- The variance of utility (Var[U(X)]) captures the decision‑maker’s aversion to dispersion around expected utility. statistical
Source sentence
In this paper, we investigate an additional component of risk, parameterized by the variance of the utility Var[U(X)], which encodes the decision maker’s aversion to dispersion around the expected utility. This introduces a hyper-utility function that measures the trade-off between maximizing expected utility and minimizing the dispersion around its mean which we call a generalized mean-variance (GMV) problem: w =
Show 2 more
- Hyper‑parameters governing the outcome distribution are conditioned on past data and subjective market views. data
Source sentence
σ parameters of the outcome distribution X ∼P(µ, σ) are parameterized by distributions Gµ and Gσ with corre-sponding hyperparameters θµ and θσ: Xt ∼P(µ, σ), µ ∼Gµ(θµ|Dt−1), σ2 ∼Gσ(θσ|Dt−1) (1) where the hyperparameters θµ and θσ are conditioned on past data and subjective views on the market, Dt−1.
- When the covariance matrix is uncertain, the posterior predictive distribution can be approximated analytically (e.g., Student’s t for Gaussian with Normal‑Wishart priors). model
Source sentence
It is worth noting that the PPD is also known for the univariate and multivariate cases of Gaussian outcomes with Normal-Wishart and Normal-Inverse-Wishart priors. In both cases, the PPD is given by the Student’s t-distribution, which, in the case of exponential utility, can only be studied numerically.
Data
No dataset named in the text.
Costs
Not applicable
Finds 5
- Exponential utility generalizes the mean‑variance approach, providing a diversification rule that remains analytically tractable for ALD‑distributed returns and Gaussian returns with uncertain covariance. theoretical
Source sentence
We show that exponential utility generalizes the MV approach, offering a diversifi-cation rule while maintaining analytical tractability for cases such as ALD-distributed returns and Gaussian returns with an uncertain covariance matrix.
- Optimal leverage can be derived from logarithmic utility within the GMV framework, balancing expected utility and its variance. theoretical
Source sentence
The optimal leverage can be obtained from the logarithmic utility in the GMV approach, which balances the expected utility and its variance.
- The GMV approach naturally explains the half‑Kelly criterion for leverage sizing. theoretical
Source sentence
This naturally explains the half-Kelly criterion.
- Introducing a compound distribution of outcomes and marginalizing over parameters coincides with the Bayesian posterior predictive distribution, enabling Bayesian treatment of uncertainty. theoretical
Source sentence
We enhance existing methods by introducing a compound distribution of the outcome and the GMV. The use of the compound distribution and marginalization over parameters for-mally coincides with the posterior predictive distribution of the outcome, allowing us to use Bayesian machinery when modeling the outcome of an investment.
- The GMV formulation allows control over utility dispersion, which is especially relevant when the probability of ruin is non‑negligible. theoretical
Source sentence
The GMV can be seen as a generalization of the maximum expected utility principle and enables control over the dis-persion of utility—which is particularly relevant in situations where the probability of ruin is non-negligible.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 26 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.