Quantum-Assisted Optimal Rebalancing with Uncorrelated Asset Selection for Algorithmic Trading Walk-Forward QUBO Scheduling via QAOA

Abraham Itzhak Weinberg · 2026

Read Quantum-Assisted Optimal Rebalancing with Uncorrelated Asset Selection for Algorithmic Trading Walk-Forward QUBO Scheduling via QAOA by Abraham Itzhak Weinberg (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-16904.txt.

Introduction

Weinberg presents a hybrid classical and quantum framework for portfolio rebalancing. Asset selection uses shrinkage covariance estimates and correlation clustering to pick 10 decorrelated stocks from the S&P 500. Weights come from a genetic algorithm, checked against simple benchmarks. The rebalancing schedule itself is cast as a QUBO problem and solved with QAOA inside a walk forward design. In backtests, the combined strategy reached a Sharpe ratio of 0.588 and a 10.1 percent return. It used 8 rebalances instead of 24, cutting transaction costs by 44.5 percent.

Catalog record

CREATORAbraham Itzhak Weinberg
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE30748
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.16904

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Quantum-Assisted Optimal Rebalancing with Uncorrelated Asset Selection for Algorithmic Trading Walk-Forward QUBO Scheduling via QAOA Abraham Itzhak Weinberg Abstract We propose a quantum-assisted algorithmic trading framework that addresses two core portfolio management problems simultaneously: principled asset selection and optimal rebalancing scheduling. For asset selection, we apply Ledoit-Wolf shrinkage covariance estimation with hierarchical correlation clustering to select n = 10 n{=}10 maximally uncorrelated stocks from the S&P 500 universe without survivorship bias. For weight optimisation, we deploy an entropy-regularised Genetic Algorithm (GA) accelerated on GPU, alongside closed-form Minimum Variance (MinVar) and equal-weight baselines and a three-way ensemble. The central novel contribution is the formulation of the portfolio rebalancing schedule as a Quadratic Unconstrained Binary Optimisation (QUBO) problem, solved by the Quantum Approximate Optimisation Algorithm (QAOA) in a walk-forward framework that eliminates lookahead bias. Backtested on S&P 500 data (train: 2010–2024, test: 2025, n = 249 n{=}249 trading days), the GA + QAOA strategy achieves a Sharpe ratio of 0.588 and total return of 10.1% , outperforming the best classical baseline (GA Rebalance/10d, Sharpe 0.575) while executing only 8 rebalances (6.1 bp transaction cost) versus 24 rebalances (11.0 bp), representing a 44.5% cost reduction for superior risk-adjusted performance. Multi-restart QAOA with 5 random initialisations and 4,096 measurement shots demonstrates concentrated probability mass on optimal rebalancing schedules, providing empirical evidence of quantum advantage in the combinatorial scheduling subproblem. Keywords: Quantum computing, QAOA, QUBO, Portfolio optimisation, Rebalancing, Ledoit-Wolf, Genetic algorithm, Algorithmic trading, S&P 500. 1 Introduction Portfolio management involves two interrelated decisions: which assets to hold and when to rebalance toward target weights. Classical approaches treat these as separate, sequential problems — first optimise weights (Markowitz [ 1 , 2 ] , Sharpe [ 3 ] ), then apply a fixed rebalancing schedule (calendar-based or threshold-based [ 4 ] ). This decoupling ignores the interaction between weight drift and rebalancing cost, and fixed schedules waste transaction budget on unnecessary trades while missing beneficial rebalancing opportunities. Quantum computing offers a promising avenue for combinatorial optimisation problems that are classically intractable at scale. The Quantum Approximate Optimisation Algorithm (QAOA) [ 5 ] provides a near-term, variational approach to solving Quadratic Unconstrained Binary Optimisation (QUBO) problems on current noisy intermediate-scale quantum (NISQ) hardware [ 6 ] . Portfolio-related QUBO formulations have been studied for asset selection [ 7 , 8 , 9 ] , but the question of applying QAOA to the rebalancing schedule problem has not been previously addressed in the literature. The contributions of this paper are: 1. Principled uncorrelated asset selection via Ledoit-Wolf shrinkage [ 10 ] combined with hierarchical correlation clustering — avoiding both the Markowitz estimation error problem and greedy-search survivorship bias. 2. Entropy-regularised GA weight optimisation on GPU, preventing degenerate sparse solutions while maximising annualised Sharpe ratio. 3. Minimum-Variance and ensemble weights as principled classical baselines computed via closed-form solution. 4. Novel QUBO formulation of the rebalancing schedule incorporating marginal Sharpe gain, transaction cost penalty, and over-frequency penalty via an exponentially decaying interaction term. 5. Walk-forward QAOA scheduling with multi-restart optimisation — eliminating lookahead bias while providing robust angle estimation. 6. Comprehensive net-of-cost backtest comparing 16 strategies across Sharpe, Sortino, maximum drawdown, Calmar ratio, rebalance count, and transaction cost. The remainder of the paper is organised as follows. Section 2 reviews related work. Section 3 presents the full methodology. Section 4 describes experimental setup. Section 5 presents results and analysis. Section 6 discusses the quantum advantage argument. Section 8 concludes. 2 Related Work This section reviews prior work in classical portfolio optimisation, machine learning–based allocation, and quantum approaches to portfolio construction. While asset weighting and selection have been widely studied, the optimisation of rebalancing timing remains comparatively underexplored, particularly within quantum frameworks. Our formulation addresses this gap by casting walk-forward rebalancing scheduling as a QUBO problem amenable to variational quantum algorithms. 2.1 Classical Portfolio Optimisation The mean-variance framework of Markowitz [ 1 , 2 ] remains the canonical approach to portfolio weight selection, extended by the Capital Asset Pricing Model (CAPM) [ 3 ] and subsequent factor models [ 11 ] . In practice, the covariance matrix must be estimated from finite samples; Ledoit and Wolf [ 10 ] showed that analytical shrinkage of the sample covariance toward a structured estimator dramatically reduces estimation error and improves out-of-sample performance. Portfolio rebalancing has been studied primarily as a cost-benefit trade-off. Perold and Sharpe [ 4 ] introduced the constant-mix strategy, while [ 12 , 13 ] showed that optimal rebalancing frequency depends on asset volatility and transaction costs. Threshold-based rebalancing [ 14 ] and time-series momentum signals [ 15 ] have been proposed as improvements over calendar-based schedules. 2.2 Evolutionary and Machine Learning Approaches Genetic algorithms have been applied to portfolio optimisation since [ 16 ] , demonstrating competitive performance on non-convex objectives including Sharpe ratio maximisation. Reinforcement learning approaches [ 17 ] and deep neural networks [ 18 ] have been proposed for dynamic asset allocation, though out-of-sample generalisation remains challenging. 2.3 Quantum Portfolio Optimisation Quantum computing approaches to portfolio optimisation have focused primarily on the asset selection problem, formulated as a QUBO. Rebentrost and Lloyd [ 9 ] proposed a quantum algorithm for portfolio optimisation with quadratic speedup. Barkoutsos et al. [ 8 ] applied QAOA and Variational Quantum Eigensolver (VQE) to portfolio selection on small instances. Mugel et al. [ 7 ] demonstrated dynamic portfolio optimisation on D-Wave quantum annealers. More recently, [ 19 ] provided a comprehensive survey of quantum finance algorithms. To our knowledge, no prior work has formulated the rebalancing timing decision as a QUBO problem or applied QAOA to this subproblem. 3 Methodology 3.1 Problem Statement Let \mathcal{U} be a universe of M assets with daily adjusted close prices. We seek a portfolio of n\ll M assets with weights \mathbf{w}\in\Delta^{n} (the n -simplex) and a binary rebalancing schedule \mathbf{x}\in\{0,1\}^{W} over W candidate dates, jointly maximising risk-adjusted return net of transaction costs. 3.2 Data and Universe Construction We construct a survivorship-bias-free S&P 500 universe by starting from the current index constituents and reversing all addition/deletion events that occurred after the training cutoff date (December 31, 2024), recovering the set of stocks that were in the index on that date. Daily adjusted close prices are downloaded for the period January 1, 2010 to December 31, 2025. After removing tickers with missing data throughout the full period, M=422 assets remain. The training period covers January 2010 – December 2024 ( T_{\text{train}}=3{,}774 trading days) and the out-of-sample test period covers January – December 2025 ( T_{\text{test}}=249 trading days). 3.3 Asset Selection via Hierarchical Clustering Ledoit-Wolf Shrinkage Covariance. Let \mathbf{r}_{t}=\log(P_{t}/P_{t-1})\in\mathbb{R}^{M} be the vector of daily log returns. The sample covariance \hat{\Sigma} is estimated with the analytical Ledoit-Wolf shrinkage estimator [ 10 ] : \hat{\Sigma}_{\text{LW}}=(1-\alpha)\,\hat{\Sigma}_{\text{sample}}+\alpha\,\mu_{\text{target}}\mathbf{I}, (1) where \alpha is the data-driven shrinkage intensity and \mu_{\text{target}} is the average sample eigenvalue. This reduces estimation error, particularly in the large- M , finite- T regime prevalent in finance. Angular Distance and Hierarchical Clustering. From the shrinkage correlation matrix \hat{R}_{\text{LW}}=D^{-1/2}\hat{\Sigma}_{\text{LW}}D^{-1/2} (where D=\operatorname{diag}(\hat{\Sigma}_{\text{LW}}) ), we define the angular distance between assets i and j : d_{ij}=\sqrt{\tfrac{1}{2}(1-\hat{\rho}_{ij})},\quad\hat{\rho}_{ij}\in[-1,1]. (2) This metric satisfies the triangle inequality and maps zero correlation to distance 1/\sqrt{2} [ 20 ] . Ward’s linkage hierarchical clustering [ 21 ] is applied to the

Read the complete text