Pith. sign in

REVIEW 5 major objections 5 minor 1 cited by

Benchmarking of Quantum and Classical Computing in Large-Scale Dynamic Portfolio Optimization Under Market Frictions

T0 review · 5 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper builds a QUBO benchmark for dynamic portfolio optimization with frictions and shows that a GPU-based heuristic matches a leading commercial solver on time-to-solution.

desk verdict Benchmark ambition is real, but the model equations don't encode the stated problem and the benchmark itself isn't released; the numerical comparisons aren't interpretable. read the letter →

arxiv 2502.05226 v1 pith:YZBYQMTS submitted 2025-02-06 math.OC cs.CC

Ying Chen , Thorsten Koch , Hanqiu Peng , Hongrui Zhang This is my paper · ORCID
classification math.OCcs.CC MSC 90C2090C2791G10
keywords dynamicportfoliooptimizationQUBOquantumcomputingbenchmarkingtransactioncostsshortsellingbinaryquadraticprogrammingS&P500
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper builds a dynamic portfolio optimization problem with realistic frictions—transaction costs, short selling, cash interest, and caps on asset counts—and rewrites it as a quadratic unconstrained binary optimization (QUBO) problem, a form that runs on both ordinary computers and quantum hardware. It then solves two publicly released benchmark instances built from S&P 500 data: 200 stocks over 10 trading days and 499 stocks over 15 days. The central result is that a GPU-based QUBO heuristic consistently matches or beats a leading commercial solver on time-to-solution, while final objectives are close on all but the hardest risk-aversion settings. If this holds, the field gets a transparent, fair testbed for claims about quantum or classical speedups in financial optimization.

What carries the argument

The load-bearing object is the QUBO encoding of the dynamic portfolio policy. Binary variables x_{it} mark whether asset i is held at time t, slack binary variables encode the shortfall to the asset-count cap B and cash cap C, and the transaction-cost term uses the identity |x_{i,t-1} - x_{it}| = x_{i,t-1}+x_{it}-2x_{i,t-1}x_{it} to keep costs quadratic in binaries. The short-selling direction is carried by an indicator τ_i, and the whole objective is a sum of risk, profit, transaction, interest, short-fee, and liquidation terms plus quadratic penalties for the two hard constraints. The evaluation machinery is the optimality gap, computed against the relaxed continuous lower bound of an equivalent binary quadratic program, which lets a heuristic QUBO solution be judged even when exact optimality is unproven.

What would settle it

Run a tiny brute-force check on a 2-asset, 2-period instance: enumerate every feasible holding path under the stated rules, compute the objective from equation (9) for each, and compare the solver's reported optimum to the true minimum; any mismatch shows the QUBO equations do not encode the described problem.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is a benchmark, not a new algorithm: a multi-period mean-variance portfolio problem with integer constraints, transaction costs, short-sale fees, and path-dependent cash is encoded as a QUBO, and two instances (n=200, T=10; n=499, T=15) with up to 45,060 binary variables are solved to gaps that can be measured against a relaxed lower bound. The head-to-head shows the ABS2 heuristic in the qoqo solver matching or outperforming the commercial BQP solver Gurobi across most risk-aversion levels, with the largest advantages in time-to-solution; quantum hardware (annealing and gate-based) runs only the smaller instance and returns materially worse objectives. The conclusion is that today's quantum-ready QUBO solvers are already competitive with classical optimization on this testbed, and the released instances give future quantum methods a precise yardstick.

Load-bearing premise

The benchmark is only what it claims to be if equations (6)–(9) faithfully encode the stated portfolio problem—specifically, that the binary variables and the capital constraint capture the up-to-k selections per asset and the price-weighted cash limit as the text describes.

Editorial extensions

If this is right

  • A solver that improves on these benchmark numbers, whether quantum or classical, will have a transparent, reproducible claim to advancement.
  • The QUBO form means the same instances can be run on annealing hardware, gate-based quantum computers, and classical heuristics without reformulation.
  • For extreme risk aversion (q=0.01), both solvers converge to the all-cash strategy, confirming the benchmark is solvable and the penalty encoding is consistent.
  • At the hardest intermediate risk levels, the GPU heuristic reaches a good solution in tens of seconds where the commercial solver takes thousands, so time-limited comparisons should report convergence trajectories.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The 'matched or outperformed' claim mostly rests on time-to-solution and early convergence; on final objective value the commercial solver still wins for the most balanced risk levels (e.g., q=0.001), so a reader should compare both metrics before declaring a champion.
  • The paper's benchmark instances fix one calibration of transaction cost and short-loan rates; regenerating the same instances with different frictions (e.g., taxes, price impact, or no short selling) would test whether the observed solver ranking persists.
  • A natural extension is to verify the QUBO encoding against brute-force enumeration on tiny synthetic instances (n=3, T=3), which would catch any mismatch between the written equations and the intended 2kn-dimensional encoding.
  • The plateau after about 600 seconds suggests that benchmark protocols should report full time-to-objective curves, not just a single final number, to make future claims of speedup interpretable.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper proposes a benchmarking framework for large-scale multi-period dynamic portfolio optimization with transaction costs, short selling, and capital constraints, formulated as QUBO and BQP. It reports experiments on up to 499 S&P 500 stocks over 15 time steps, comparing the BQP solver Gurobi with the QUBO heuristic qoqo/ABS2 and discussing D-Wave, IBM Qiskit, and InfinityQ. The central claims are that the problem is a challenging yet solvable benchmark and that the qoqo (ABS2) algorithm consistently matched or outperformed Gurobi.

Significance. If the model were correctly specified, the benchmark would be a useful contribution: large-scale QUBO instances with real market data, explicit optimality gaps, and time-to-solution metrics are rare, and the comparison with a state-of-the-art classical solver would be informative. The paper also provides a concise literature table and a clear parameter setting. However, the internal inconsistencies in the model formulation mean that the reported numerical results do not support the paper's stated central claim, so the significance cannot be assessed on the current evidence.

major comments (5)
  1. [§2.2, Eqs. (6)–(9)] The prose states that each asset can be selected up to k times and defines a 2kn-dimensional binary vector with k blocks per asset and separate long/short entries, but the displayed model uses only scalar binary variables x_it and sums over i. No block index, k-copy structure, or long/short split appears in the objective or constraints. The experiments labeled n200_T10_k3 and n499_T15_k3 therefore do not instantiate the problem described in the text, and Tables 3–5 cannot be benchmarks for that problem.
  2. [§2.2, Eq. (7)] The capital constraint sums τ_i x_it plus binary slack without multiplying by the price p_it. If x_it is a binary selection or block count, the left-hand side is a count rather than a capital amount; if x_it is instead a dollar position, then the profit term (p_{t+1} - p_t)x_it in Eq. (2) and Eq. (9) is dimensionally inconsistent. In addition, because the profit term does not contain τ_i and is positive for any x_it, a short position is rewarded exactly like a long position, so the modeled objective does not correspond to a long/short portfolio.
  3. [§2.1–§2.2, Eqs. (8)–(9)] The penalty parameter P is introduced as 'sufficiently large' but is never given a numerical value, either in Table 2 or in the experimental section. Since the QUBO solver minimizes the penalized objective, the reported 'QUBO Obj.' values depend critically on P. The manuscript also does not report whether the returned QUBO solutions satisfy constraints (6) and (7), so the reported gaps may correspond to infeasible allocations.
  4. [§5, Tables 3–4] The conclusion that the qoqo (ABS2) algorithm 'consistently matched or outperformed Gurobi' is not supported by the reported numbers. In Table 3 at q=0.0005, Gurobi finds -296,540 versus ABS2's -272,033, and in Table 4 at q=0.001, Gurobi finds -332,172 versus ABS2's -132,613. Gurobi also attains a better final objective at most other values of q; ABS2 shows a shorter time-to-solution in some hard cases, but 'consistently' is an overstatement that the data do not justify.
  5. [Abstract and §5] The abstract and conclusion state that benchmarks are released, but the manuscript contains no repository link, data URL, or code availability statement. For a benchmarking paper, the absence of the released benchmark artifact is a load-bearing omission that prevents readers from using or verifying the claimed resource.
minor comments (5)
  1. [§1] The sentence 'Infact, solvinglarge-scaledynamicportfoliooptimizationproblemsisalsochallenging' has missing spaces and should be corrected.
  2. [§2.3] The phrase 'the optimal solutios' should read 'the optimal solutions'.
  3. [§4.2.1, Table 3] The column labeled 'QUBO Obj.' for q=0.01 reports -1,000 with a gap* of 10,782%; the paper explains the cash-only solution, but it would be helpful to state explicitly why this value is -1,000 and how it relates to C=10.
  4. [§4.2.2, Table 5] The text mentions D-Wave and InfinityQ results for n200_T10_k3_B60_C10, but Table 5 only contains aggregate Sharpe ratios and transaction costs without reporting the underlying objective values or computational times, making those comparisons difficult to assess.
  5. [§2.2, Eq. (6)] The notation for slack variables is inconsistent: the text introduces s_b,t, but later equations use s_ct and y_ct; the indices should be unified.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the QUBO benchmark is built directly from the stated model, and the only self-citation (qoqo/ABS2) is attribution rather than load-bearing.

full rationale

The paper does not derive any prediction from fitted parameters or from its own previous results. The QUBO objective in Eq. (9) and the equivalent BQP in Eq. (10) are constructed directly from the stated multi-period objective, transaction costs, short-selling costs, cash interest, and quadratic penalty terms in Section 2.2; nothing is calibrated to force a target outcome. The numerical experiments compare an external solver, Gurobi, with qoqo/ABS2, and the conclusion that qoqo (ABS2) 'consistently matched or outperformed Gurobi' is supported by the reported Tables 3-5, not by a prior result of the authors. The one self-citation, Rehfeldt et al. (2023) for the qoqo solver, is co-authored by T. Koch, but it is used only to attribute the software employed, not to justify the benchmark outcome. No uniqueness theorem, ansatz smuggled in via citation, or renamed known result is invoked. The apparent mismatch between the described k-block/long-short encoding and the displayed single-index x_it in Eqs. (6)-(9) is a modeling correctness concern, not a circularity of derivation, and therefore does not affect this circularity score.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The model relies on the standard QUBO/Ising equivalence, plus two unstated modeling assumptions about how prices, units, and binary blocks interact. The only free parameter in the QUBO conversion is the penalty weight P, which is never specified. No new physical entities are introduced.

free parameters (1)
  • penalty weight P = not specified
    The QUBO reformulation in Eq. (9) enforces the capital and asset-count constraints through a quadratic penalty with weight P, but no value, scaling rule, or verification of constraint satisfaction is given anywhere in the paper. The comparison between QUBO and BQP objectives depends on this unstated parameter.
assumptions (3)
  • standard math QUBO can be written as an Ising Hamiltonian with x_i = (s_i + 1)/2
    Section 3.2 cites Lucas (2014); this is a standard, machine-checkable equivalence.
  • domain assumption The capital constraint (7) is correctly stated without price weighting
    Equation (7) sums x_it and cash slack variables to equal C, omitting p_it. The text elsewhere uses p_it in risk and profit terms, so this assumption that asset units and cash units are commensurable without prices is unstated and load-bearing.
  • ad hoc to paper Each asset's k blocks are representable by the scalar binary x_it in equations (6) through (9)
    Section 2.2 introduces a 2kn-dimensional binary vector, but the displayed equations use a single x_it per asset per time, making the encoding of k=3 blocks and of long/short variables undefined.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Benchmarking of Quantum and Classical Computing in Large-Scale Dynamic Portfolio Optimization Under Market Frictions." pith.science (2026). https://pith.science/paper/YZBYQMTS

@misc{pith2026250205226,
  author       = {Pith},
  title        = {Pith review of: Benchmarking of Quantum and Classical Computing in Large-Scale Dynamic Portfolio Optimization Under Market Frictions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YZBYQMTS}},
  note         = {Machine review of arXiv:2502.05226}
}
read the original abstract

Quantum computing is poised to transform the financial industry, yet its advantages over traditional methods have not been evidenced. As this technology rapidly evolves, benchmarking is essential to fairly evaluate and compare different computational strategies. This study presents a challenging yet solvable problem of large-scale dynamic portfolio optimization under realistic market conditions with frictions. We frame this issue as a Quadratic Unconstrained Binary Optimization (QUBO) problem, compatible with digital computing and ready for quantum computing, to establish a reliable benchmark. By applying the latest solvers to real data, we release benchmarks that help verify true advancements in dynamic trading strategies, either quantum or digital computing, ensuring that reported improvements in portfolio optimization are based on robust, transparent, and comparable metrics.

Figures

Figures reproduced from arXiv: 2502.05226 by the authors.

Figure 1
Figure 1. Pareto optimal frontier for n200_T10_k3_B60_C10 [PITH_FULL_IMAGE:figures/full_fig_p016_1.png] view at source ↗
Figure 2
Figure 2. Stock positions for n200_T10_k3_B60_C10 with [PITH_FULL_IMAGE:figures/full_fig_p016_2.png] view at source ↗
Figure 3
Figure 3. Stock positions for n200_T10_k3_B60_C10 with [PITH_FULL_IMAGE:figures/full_fig_p017_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Convergence of objective values for different risk aversion parameters. [PITH_FULL_IMAGE:figures/full_fig_p017_4.png]
Figure 5
Figure 5. Figure 5: Stock positions for n200_T10_k3_B60_C10 with [PITH_FULL_IMAGE:figures/full_fig_p018_5.png]
Figure 6
Figure 6. Figure 6: Profits and Variances at different q values. for simpler problems, while QUBO resolves more challenging scenarios at q = 0.0005 and q = 0.001 more effectively. An in-depth analysis of the objective value changes over time for q = 0.0005 and q = 0.001 is depicted in Fig…
Figure 8
Figure 8. Figure 8: Objective Values for n499_T15_k3_B60_C10 at different times. [PITH_FULL_IMAGE:figures/full_fig_p020_8.png]
Figure 9
Figure 9. Figure 9: Stock positions for n499_T15_k3_B60_C10 with [PITH_FULL_IMAGE:figures/full_fig_p020_9.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Benchmarking Quantum Solvers in Noisy Digital Simulations for Financial Portfolio Optimization

    quant-ph 2025-08 reject novelty 4.0 of 10

    On small synthetic portfolio problems, noiseless QAOA fits the known ground-state energy well, but noisy QAOA fails while QITE, pretrained on noiseless simulators, still identifies the optimal portfolio on IBM hardware.

Reference graph

Works this paper leans on

11 extracted references · 9 canonical work pages · cited by 1 Pith paper

  1. [1]

    Optimal control of execution costs

    Bertsimas, D. and A. W. Lo (1998). “Optimal control of execution costs”.Journal of financial markets 1.1, pp. 1–50. Bertsimas, D. and J. N. Tsitsiklis (1997).Introduction to linear optimization. Vol

    work page 1998

  2. [5]

    odzimierz Ogryczak, M

    Mansini, R., W. odzimierz Ogryczak, M. G. Speranza, and E. T. A. of European Oper- ational Research Societies (2015).Linear and mixed integer programming for portfolio optimization. Vol

    work page 2015

  3. [6]

    Quantum algorithms for mixed binary optimization applied to transaction settlement

    Athena scientific Belmont, MA. Braine, L., D. J. Egger, J. Glick, and S. Woerner (2021). “Quantum algorithms for mixed binary optimization applied to transaction settlement”.IEEE Transactions on Quantum Engineering 2, pp. 1–8. Brandt, M. W. and P. Santa-Clara (2006). “Dynamic portfolio selection by augmenting the asset space”.The journal of Finance61.5, p...

    work page 2021

  4. [15]

    Capital asset pricing model (CAPM) with drawdown measure

    Zabarankin, M., K. Pavlikov, and S. Uryasev (2014). “Capital asset pricing model (CAPM) with drawdown measure”.European Journal of Operational Research234.2, pp. 508–517. 25

    work page 2014

  5. [21]

    Optimum consumption and portfolio rules in a continuous-time model

    Springer. McGeoch, C. C. (2022). Adiabatic quantum computation and quantum annealing: Theory and practice. Springer Nature. Merton, R. C. (1975). “Optimum consumption and portfolio rules in a continuous-time model”. Stochastic optimization models in finance. Elsevier, pp. 621–661. Montanez-Barrera, J. and K. Michielsen (2024). “Towards a universal QAOA pr...

    arXiv 2022

  6. [131]

    Portfolio optimization with irreversible long-term investments in renewable energy under policy risk: A mixed- integer multistage stochastic model and a moving-horizon approach

    Gatzert, N., A. Martin, M. Schmidt, B. Seith, and N. Vogl (2021). “Portfolio optimization with irreversible long-term investments in renewable energy under policy risk: A mixed- integer multistage stochastic model and a moving-horizon approach”.European Journal of Operational Research290.2, pp. 734–748. Gilliam, A., S. Woerner, and C. Gonciulea (2021). “G...

    work page 2021

  7. [291]

    Optimal portfolio policies with borrowing and shortsale constraints

    Tepla, L. (2000). “Optimal portfolio policies with borrowing and shortsale constraints”.Jour- nal of Economic Dynamics and Control24.11-12, pp. 1623–1639. Wörner, S. and D. J. Egger (2019). “Quantum risk analysis”.npj Quantum Information5.1, p

    work page 2000

  8. [404]

    A multi-period fuzzy portfolio optimization model with minimum transaction lots

    Liu, Y.-J. and W.-G. Zhang (2015). “A multi-period fuzzy portfolio optimization model with minimum transaction lots”.European Journal of Operational Research242.3, pp. 933–

    work page 2015

Show all 11 references
  1. [428]

    Quantum computational supremacy

    Harrow, A. W. and A. Montanaro (2017). “Quantum computational supremacy”.Nature 549.7671, pp. 203–209. Herman, D., C. Googin, X. Liu, A. Galda, I. Safro, Y. Sun, M. Pistoia, and Y. Alexeev (2022). “A survey of quantum computing for finance”.arXiv preprint arXiv:2201.02773. 23 ...

    2017 arXiv

  2. [941]

    Ising formulations of many NP problems

    Lucas, A. (2014). “Ising formulations of many NP problems”.Frontiers in physics2, p

    2014

  3. [4213]

    Quantumcomputationalfinance:Monte Carlo pricing of financial derivatives

    Pliska, S. R. (1997).Introduction to mathematical finance: discrete time models. Wiley. Rebentrost,P.,B.Gupt,andT.R.Bromley(2018).“Quantumcomputationalfinance:Monte Carlo pricing of financial derivatives”.Physical Review A98.2, p. 022321. Rehfeldt, D., T. Koch, and Y. Shinano ...

    1997

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.