Quantum Portfolio Optimization: An Extensive Benchmark

arXiv:2509.17876 · quant-ph, math.OC · Submitted 2025-09-22 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.

Kai: Today's paper: "Quantum Portfolio Optimization: An Extensive Benchmark".

Mira: A computational study was conducted to evaluate quantum approaches against classical methods for volatility-minimizing portfolio optimization, aiming to quantify the potential for quantum advantage in this specific real-world problem.

Kai: First, who's behind it and why it matters.

Paper summary: Kai: So to recap, this paper, "Quantum Portfolio Optimization: An Extensive Benchmark," sets out to bridge a gap in the literature by providing a computational study that compares quantum approaches against established classical methods for portfolio optimization on a meaningful instance set.

Mira: The core thesis they present is that they are testing whether quantum computing can offer any tangible advantage over classical solvers when tackling real-world portfolio problems, specifically focusing on volatility minimization.

Lev: It's interesting that they chose to focus their comparison specifically on the volatility-minimizing variant because the authors suggested it was the most challenging one for classical optimization methods to solve efficiently.

Kai: They benchmarked two primary heuristic quantum algorithms—quantum annealing and QAOA—against several classical techniques like mixed-integer programming, simulated annealing, and tabu search.

Mira: The results they highlight indicate that all instances can be solved to proven optimality by mixed-integer programming within a few seconds, which sets a very high bar for what we need quantum methods to overcome.

Lev: If the classical exact methods are that fast, it puts the onus entirely on the quantum hardware to provide an exponential speedup just to match that runtime, which seems unlikely given current qubit counts and coherence times.

Kai: Furthermore, they noted some limitations with the quantum approaches themselves; for instance, QAOA often struggled with finding feasible solutions because of decaying feasibility percentages as instances got larger.

Mira: They also pointed out specific technical hurdles for quantum annealing, suggesting that the high density of the QUBO problem combined with all-to-all connectivity embedding led to suboptimal performance.

Lev: That points directly to the physical implementation challenges; having long chains of qubits for each variable due to that connectivity is exactly what we worry about when mapping these abstract problems onto physical hardware.

Kai: So, in short, the paper's main claim is that under the tested conditions, classical heuristics and solvers maintain a clear lead in terms of solution quality and feasibility over the quantum methods they tested.

Mira: This matters because it provides a concrete data set showing where current quantum algorithms fall short when trying to match established classical performance on this type of optimization task.

Lev: For anyone thinking about running this on actual hardware, the results suggest that we're looking at a significant gap in performance that needs to be addressed before we can even hope to see practical utility.

Conclusion: Kai: Reflecting on the entire study presented in "Quantum Portfolio Optimization: An Extensive Benchmark," we see that Eric Stopfer and Friedrich Wagner have provided a very clear picture of the current landscape for quantum optimization in finance.

Mira: They’ve demonstrated that while quantum approaches are theoretically interesting, when tested against rigorous classical methods on a real-world data set, they haven't yet shown the capability to provide a meaningful speedup or superior quality in terms of solution feasibility.

Lev: I think the main implication for researchers is that we need to be very pragmatic about where we invest our efforts; focusing on error correction and qubit stability for these specific problem types might yield more immediate results than trying to force a quantum solution onto this specific benchmark set right now.

Kai: The title itself speaks volumes about what this work achieves: it establishes a necessary benchmark for anyone trying to claim quantum potential in portfolio optimization research.

Mira: It’s important because it clearly delineates the difference between theoretical possibility and practical applicability for the volatility-minimizing variant versus, say, the return-maximizing variants they also studied.

Lev: So, what this means in simple terms is that for now, classical methods are still the reliable workhorse when you need a high-quality result on a problem structured like this.

Kai: Exactly; we aren't seeing clear evidence that quantum algorithms can beat the best classical heuristics in terms of finding a good solution within reasonable time constraints for this specific optimization task.

Mira: The authors' work serves as an important piece of evidence showing that the complexity they mapped onto QUBO problems is currently too demanding for the connectivity and noise levels present in those devices.

Lev: It gives us a concrete reason why we need to keep pushing on improving error correction and increasing qubit counts, rather than expecting a quantum computer to solve this today.

Fraunhofer Institute for Integrated Circuits, Nürnberg

quant-ph, math.OC

Submitted: 2025-09-22

Updated: 2026-10-07

Comments: submission process ongoing

Code: https://github.com/stopfereric/portfolio_opt_

Project page: https://qiskit.github.io/qiskit-aer/stubs/qiskit_

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 79/100

The gist: A computational study was conducted to evaluate quantum approaches against classical methods for volatility-minimizing portfolio optimization, aiming to quantify the potential for quantum advantage

Key concepts

MaxRet
This optimization variant aims to maximize the expected return of a portfolio while keeping its volatility below a certain limit. It is formulated as a quadratic program, which is a standard mathematical problem used in finance to balance risk and reward.
MinVola
This specific portfolio optimization variant focuses on minimizing volatility while ensuring the portfolio achieves at least a minimum required return level. The researchers deemed this variant the most promising for quantum speedup because its structure was found to be particularly difficult for classical solvers.
Quantum Annealing
This quantum algorithm uses physical annealing processes on devices like D-Wave processors to find solutions to optimization problems, often by mapping them onto a QUBO (Quadratic Unconstrained Binary Optimization) format. Performance was limited by the high density of the problem and the overhead of embedding it for all-to-all connectivity.
Approximation Ratio (Θ)
This metric measures how close a solution found by an algorithm is to the absolute best possible solution (fopt). A ratio greater than 1 means fopt is less than or equal to fm, indicating the quality of the found solution. The goal in this study was to see if quantum methods could achieve a better approximation ratio than classical ones.

Terminology

Summary

A computational study was conducted to evaluate quantum approaches against classical methods for volatility-minimizing portfolio optimization, aiming to quantify the potential for quantum advantage in this specific real-world problem.

Problem Variants Considered

The research focuses on three variants of portfolio optimization:

  1. MaxRet: Maximizing expected portfolio return subject to an upper bound on volatility, formulated as a quadratic program (MaxRet).

  2. MinVola: Minimizing volatility while ensuring a minimum return level, formalized by the quadratic program MinVola. This variant was identified as the most promising variant for a potential quantum speedup.

  3. MultiObj: Combining both objectives by maximizing a weighted sum of return and volatility (MultiObj).

Classical Solvers and Hardness Assessment

The study compared quantum methods against several classical approaches, including:

** mixed-integer programming (MIP)**

(simulated annealing)

(steepest descent local search)

(tabu search)

A key finding was that all instances can be solved to proven optimality by mixed-integer programming in the order of seconds. Furthermore, the problem-tailored classical heuristic consistently outperforms quantum approaches in terms of solution quality for fixed runtime. The study identified the MinVola variant as having the largest potential for quantum advantage because it was found to be the most promising variant for a potential quantum speedup.

Quantum Approaches Compared

The paper benchmarked two primary heuristic quantum algorithms:

  1. Quantum Annealing: This method was tested on a D-Wave Advantage 2 processor, utilizing an embedding of the QUBO variables. Results showed that quantum annealing with adjusted chain strength is roughly on par with random sampling in terms of both feasibility percentage and approximation ratio. The authors conjectured that the high density of the QUBO problem is a reason for the suboptimal performance of quantum annealing due to long chains of qubits for each variable caused by all-to-all connectivity embedding.

  2. Quantum Approximate Optimization Algorithm (QAOA): This hybrid algorithm was tested using various parameter determination methods, including grid search, linear ramp QAOA (LR-QAOA), and COBYLA optimization. The results indicated that QAOA often falls short on finding feasible solutions, which results in decaying feasibility percentages in Figure 3b. The authors attributed the performance degradation with larger instances to the rapidly increasing number of swap gates that are required to transpile the QAOA circuit with all-to-all-connectivity to the quantum computer with limited connectivity.

Classical Heuristics Performance

The classical heuristics demonstrated strong performance:

(steepest descent, simulated annealing, and tabu search)

In general, all open-source implementations (steepest descent, simulated annealing and tabu search) are able to find feasible solutions for nearly all problem instances. The problem-specific heuristic performs best with respect to the average approximation ratio. For instance, for 500 assets, the problem-specific heuristic was only able to return roughly 100 out of 500 possible solutions in 60 s.

Benchmark Metrics and Conclusion

The benchmark procedure enforced a time limit of 60 seconds on the execution time of the quantum algorithm on the device to ensure fairness. The primary solution quality metric used was the approximation ratio, defined as Θ:= fm / fopt ≥ 1, where fopt is calculated by an exact mathematical optimization solver. The final conclusion is that classical heuristics like simulated annealing, steepest descent, tabu search and a problem-specific heuristic clearly outperform QAOA and quantum annealing regarding solution feasibility and quality. While quantum methods showed some superiority in finding more samples or slightly better quality under specific conditions (e.g., quantum annealing with adjusted chain strengths), they did not clearly differ from random sampling within the time limitation. The authors conclude that there is only very limited room for a potential quantum advantage for the considered variant of portfolio optimization.

Key Findings Summary

  1. Classical solvers like Gurobi are significantly faster than SCIP, showing a speed-up factor of over 1,000 for large problem instances with n = 1,000 assets.

  2. The MinVola variant is the most promising candidate for quantum speedup due to its classical hardness.

  3. Quantum annealing performance was limited by high density of the QUBO problem and embedding overhead.

  4. Problem-specific heuristics achieved the best results regarding average approximation ratio.

  5. The study confirmed that, for this specific variant, classical methods remain superior in terms of solution quality and feasibility under practical time constraints.

The gist

All instances can be solved to proven optimality by mixed-integer programming in the order of seconds.

Improvements for AI systems

Based on the provided scientific paper, here are specific improvements that could be made to existing or future AI systems, categorized by the area of improvement:

) Improve Portfolio Optimization Accuracy and Robustness

The paper highlights that for a volatility-minimizing portfolio variant (MinVola), classical methods like Mixed-Integer Programming (MIP) can solve instances with up to 1,000 assets to proven optimality in seconds. Conversely, quantum methods (QAOA and Quantum Annealing) struggle with feasibility and solution quality compared to problem-specific heuristics.

The improved AI system should be able to:

  1. Perform high-fidelity portfolio optimization for real-world financial data (up to 1,000 assets) by leveraging the efficiency of classical solvers (like Gurobi/SCIP).

  2. For problems where exact classical solutions are intractable or too slow, the system should employ a Hybrid Solver Strategy:

Competing with quantum methods by using a sophisticated problem-tailored heuristic (as shown in Section 6) that consistently outperforms quantum approaches in terms of solution quality for fixed runtime.

) Enhance Quantum Algorithm Implementation and Parameter Selection

The paper demonstrates that the performance of Variational Quantum Eigensolver (VQE) and QAOA is highly sensitive to hyperparameter tuning, specifically the selection of penalty factors and circuit parameters.

  1. Implement a robust, automated parameter optimization loop for hybrid quantum algorithms (QAOA/VQE). This system would automatically utilize classical optimizers like COBYLA or grid searches (as described in Section 5) to find the optimal set of parameters that maximize the approximation ratio and feasibility percentage on current hardware.

  2. Develop adaptive penalty factor selection mechanisms, informed by statistical analysis of random portfolio returns and volatilities (as detailed in Appendix D), to ensure that constraint violations are penalized effectively without excessively bloating the objective function magnitude compared to typical portfolio values.

) Integrate Quantum-Classical Synergy for Intractable Problems

The benchmark suggests that while quantum methods show promise, they are currently outperformed by classical heuristics for this specific variant. The paper notes that the performance gap in QAOA/Annealing is largely attributed to high qubit overhead during embedding and transpilation of dense problems.

  1. Dynamically determine the optimal problem decomposition or sparsification technique (as mentioned in Related Work, Section 2) based on the real-time structure of the input asset data, aiming to reduce the required QUBO size before attempting a quantum computation.

  2. Utilize quantum methods not for finding the final solution directly, but as a sampling engine for high-dimensional spaces where classical heuristics fail to find any feasible solutions (as suggested by Figure 4b).

) Develop Adaptive Heuristic Search Capabilities

The problem-specific heuristic shown in Section 17 is effective because it iteratively adds weight based on maximizing variance reduction while satisfying constraints. This suggests a powerful, iterative search capability.

  1. Employ reinforcement learning or meta-heuristic search algorithms that learn the structure of the portfolio optimization problem (the MinVola variant) and adapt the weight addition strategy (Section 17) dynamically for new, unseen asset sets, allowing it to generate feasible solutions rapidly without needing a full mathematical model formulation upfront.

Abstract

Recently, several researchers proposed portfolio optimization as a potential use case for quantum optimization. However, the literature is lacking an extensive benchmark quantifying the potential of quantum computers for portfolio optimization. In this work, we contribute to closing this gap. We provide a computational study, comparing quantum approaches against state-of-the-art classical methods on a meaningful, real-world instance set. In particular, we compare quantum annealing and the quantum approximate optimization algorithm against classical mixed-integer programming, simulated annealing, steepest descent local search, tabu search and a problem-tailored heuristic. We consider a variance-minimizing variant of portfolio optimization which we show to be more difficult to solve for classical optimizers than return-maximizing or multi-objective formulations. Our benchmark data set comprises 270 instances with up to 1,000 assets from actual stock data. Due to hardware limitation, quantum methods could only be tested for instances with at most 30 assets. The results show that all instances can be solved to proven optimality by mixed-integer programming in the order of seconds. Moreover, the problem-tailored heuristic consistently outperforms quantum approaches in terms of solution quality for fixed runtime. Thus, we conclude that there is only very limited room for a potential quantum advantage for the considered variant of portfolio optimization.

Sources

Related papers