Publication Type

Working Paper

Version

publishedVersion

Publication Date

7-2025

Abstract

Multi-period portfolio optimization is a central problem in finance, yet it is computationally intractable for traditional dynamic programming methods due to the curse of dimensionality. This paper develops a tractable and theoretically grounded 'one-shot' stochastic optimization framework that recasts the sequential decision problem into a single, high-dimensional optimization task. Our approach models the predictive distribution of factor returns using Gaussian Processes (GPs), allowing it to capture complex, non-linear market dynamics. We make three primary contributions. First, for the special case of a linear GP kernel, we derive an analytical solution for the optimal portfolio path, providing a clear economic interpretation that decomposes the policy into myopic and intertemporal hedging components. Second, for the general non-linear case, we provide rigorous theoretical guarantees, establishing the consistency and convergence rate of our Monte Carlo-based optimizer. Third, we demonstrate through extensive backtesting that the one-shot GP strategy significantly outperforms standard benchmarks, such as myopic mean-variance optimization and equalweight portfolios, delivering superior risk-adjusted returns and lower drawdowns. Our work provides an efficient and robust framework for dynamic asset allocation that bridges the gap between theoretical optimality and practical implementation.

Keywords

Multi-period factor investing, one-shot stochastic optimization, long-term planning, Gaussian Processes

Discipline

Finance | Finance and Financial Management

Areas of Excellence

Digital transformation

First Page

1

Last Page

29

Identifier

10.2139/ssrn.5373661

Publisher

SSRN

Download

Share

COinS