Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

Uncertainty-Aware Strategies: A Model-Agnostic Framework for Robust Financial Optimization through Subsampling

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper claims that replacing the plug-in objective with an uncertainty-aware objective over a distribution of candidate models improves out-of-sample financial decisions, and that a bootstrap-like subsampling approximation delivers…

desk verdict The Gaussian analytic core is solid, but the CVaR-SGD algorithm's proof and implementation target opposite tails, so the paper's main computational claim does not hold as written. read the letter →

arxiv 2506.07299 v1 pith:UGQ4TKLQ submitted 2025-06-08 q-fin.CP cs.LGq-fin.MFq-fin.RM

Hans Buehler , Blanka Horvath , Yannick Limmer , Thorsten Schmidt This is my paper
classification q-fin.CPcs.LGq-fin.MFq-fin.RM MSC 91G1091G8062F4090C15
keywords modeluncertaintyKnightianriskmeasuresCVaRentropicsubsamplingdeephedgingrobustfinance
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 proposes a model-agnostic way to make financial decisions—portfolio allocation, hedging, pricing—robust to the fact that the true probability distribution is never known. Instead of optimizing the objective under one estimated model, it nests the objective inside an outer "uncertainty measure" (entropic risk or CVaR) taken over a distribution of candidate models. The authors show analytically in a Gaussian example that this uncertainty-aware strategy outperforms both the naive plug-in strategy and the mixture-of-measures approach, which averages models without penalizing model risk. Because a principled distribution over models is often unavailable, they introduce a subsampling scheme akin to bootstrapping, plus a memory-efficient CVaR-SGD algorithm, and demonstrate that the approach matches the performance of a Kalman-filter Bayesian method on a real pairs-trading example and robustifies deep hedging on a simulated cliquet option.

What carries the argument

The central object is the uncertainty-aware objective in equation (2), a composition of an inner objective $J$ with an outer uncertainty measure $U$, where $U$ maps a random variable on the model space $\Theta$ to $\mathbb{R}$. For $U$ the paper uses the entropic risk $\mathrm{Entr}_{\lambda'}$ and the conditional value-at-risk $\mathrm{CVaR}_\alpha$; for the sampling distribution $P$ over models it uses subsampling: drawing $n$ events from the estimated measure $\hat P$, giving an empirical measure $P_\theta = \frac1n\sum_{i=1}^n \delta_{\omega_i}$, and treating $\theta$ as random with distribution $\hat P^n$. The proof machinery for convergence is the Rockafellar–Uryasev representation $\mathrm{CVaR}_\alpha(Z)=\min_\zeta (\zeta + \frac1\alpha E[(Z-\zeta)_+])$, which lets the paper write the robust objective as a convex function of the action and apply the standard SGD convergence theorem.

What would settle it

Simulate the exact setting of Theorem 5.1 with $m$ fixed (say $m=50$, $\alpha=0.1$) on a scalar convex quadratic objective where the CVaR gradient is computable in closed form, and compare Algorithm 1's mean update direction to the true CVaR gradient; a systematic finite-$m$ bias that does not vanish as $t$ grows would contradict the unbiasedness premise and with it the $O(1/\sqrt{t})$ rate.

Watch

Extended reading notes

Core claim

The central claim is that the plug-in objective $J(X(a),\hat P)$ should be replaced by the uncertainty-aware objective $U(J(X(a),P_\theta),P)$, where $U$ is a risk measure over the model-indexing variable $\theta$. With $U$ chosen as entropic risk or CVaR, the resulting action shrinks toward zero when the estimated signal is weak, in contrast to the plug-in action, which scales linearly with the estimate and can produce negative out-of-sample utility. The paper derives closed-form strategies and out-of-sample performance in a Gaussian i.i.d. setting (Lemmas 3.1 and 3.2), shows that the mixture measure is essentially a variance adjustment and fails to robustify enough, and proves (Theorem 5.1) that a CVaR-SGD algorithm with step size $B/(\rho\sqrt{t})$ converges at rate $O(1/\sqrt{t})$ to the CVaR-robust objective. Numerically, subsampling-based uncertainty-aware strategies match Bayesian (Kalman-filter) performance and improve deep hedging out-of-sample.

Load-bearing premise

The convergence proof for CVaR-SGD assumes that keeping the worst $k$ of $m$ sampled objectives and averaging their gradients is an unbiased estimate of the true CVaR gradient, which is exactly true only as the number of samples grows; for finite $m$ the $k$-smallest selection is a rank-based statistic with a smoothed inclusion probability, so the practical algorithm's update is only approximately the CVaR gradient.

Editorial extensions

If this is right

  • Plug-in and mixture strategies can produce negative out-of-sample utility in the Gaussian example, while uncertainty-aware strategies with sufficient risk aversion are positive and converge to the trivial zero-investment strategy as uncertainty aversion grows.
  • The CVaR-based uncertainty-aware strategy has a zero-investment region: it stays out of the market unless the estimated signal exceeds a threshold proportional to $1/\sqrt{N}$, acting as a built-in significance filter.
  • CVaR-SGD with memory capped at $m$ simultaneous subsamples behaves like vanilla SGD with roughly $m/(1-\alpha)$ subsamples, improving approximation accuracy for larger $\alpha$ at the same memory cost.
  • On the PEP–KO pairs-trading backtest, subsampling with CVaR matches the Kalman-filter Bayesian robustification; deep hedging robustified by CVaR over subsamples improves both the mean and the dispersion of the out-of-sample P&L objective across a test distribution of Heston models.

Reading between the lines

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

  • If the central claim holds, the same nesting trick transfers to any simulation-based optimization where a plug-in estimate is known to overfit, such as reinforcement learning, energy hedging, or generative model selection.
  • The CVaR zero-investment region suggests a practical diagnostic: the ratio of the estimated signal to its standard error determines both whether to trade and how aggressively, linking the method to classical significance testing.
  • A testable extension is to apply the subsampling uncertainty measure with other coherent risk measures, such as expected shortfall variants or distortion risk measures, and check whether the dominance over mixtures persists outside the Gaussian case.
  • The paper notes that the empirical-measure bootstrap underrepresents tails; a smoothed or tail-thickened bootstrap could be tested as a direct drop-in replacement for the subsample distribution.
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

3 major / 4 minor

Summary. The manuscript proposes a two-level decision framework in which a standard financial objective J(X(a), P_θ) is evaluated under an outer uncertainty measure U over a distribution P of candidate models, and develops an "ad hoc" subsampling approximation to P. The authors derive closed-form Gaussian comparisons (Section 3), a bootstrap-style subsampling scheme (Section 4), a memory-efficient CVaR-SGD algorithm (Section 5), and numerical experiments in pairs trading and deep hedging. The central claims are that uncertainty measures outperform mixture-of-measures robustification and match Bayesian methods, and that CVaR-SGD preserves convergence while reducing memory consumption.

Significance. If the central theoretical claims were established, the paper would make a useful contribution: it gives a model-agnostic way to robustify plug-in financial optimizers, with clean analytic normal benchmarks and a plausible route to deep-learning-scale implementation. The entropic/CVaR closed forms in Section 3 are internally consistent, and the subsampling bridge to bootstrapping is conceptually appealing. However, the advertised computational guarantee (Theorem 5.1) rests on an unsupported finite-sample unbiasedness assertion, and Algorithm 1 as written targets the wrong tail; these issues must be resolved before the memory-efficiency contribution can be accepted.

major comments (3)
  1. [Appendix C.1, Theorem 5.1] The proof of Theorem 5.1 asserts that the average of the gradients over the k smallest of m i.i.d. sample objectives satisfies E[b_g(phi)] = (1/alpha) E[grad J 1_{J <= zeta_alpha}]. For finite m this is not true: the event that sample i is among the k smallest depends on all m order statistics, so by symmetry E[b_g(phi)] is a conditional expectation of g_1 given J_1 <= J_(k), which is not the population CVaR_alpha gradient. The equality holds only asymptotically as m goes to infinity with k/m -> alpha, and the theorem is stated for finite m as used in Algorithm 1. Consequently the O(1/sqrt(t)) bound in Theorem 5.1 is not established.
  2. [Algorithm 1 versus Theorem 5.1] Algorithm 1 samples k with E[k/m] = 1 - alpha and then keeps the k smallest objectives, i.e. the worst (1-alpha)-fraction of the batch. Theorem 5.1 and the definition of CVaR_alpha require the worst alpha-fraction with E[k] = alpha m. For alpha in (0,1) these are complementary tails, so the implemented estimator estimates CVaR_{1-alpha}, not CVaR_alpha. As a result, the favorable precision results for CVaR-SGD in Section 5.1 do not demonstrate convergence to the intended uncertainty-aware objective.
  3. [Section 5 and Figures 4, 9, 10, 12] The empirical claims that uncertainty-aware subsampling outperforms mixture measures and matches Bayesian methods are not supported by error bars, confidence intervals, or significance tests: Figures 9, 10, and 12 report point estimates only, and Figure 12's standard deviations are of the objective, not of the estimated performance difference. In addition, Figure 4 selects the uncertainty-aversion parameters lambda-prime and alpha on the same test environments used to report performance, so the reported gains of a "reasonably calibrated" uncertainty-aware strategy are not shown to be out-of-sample predictive. These omissions are load-bearing for the paper's empirical conclusions.
minor comments (4)
  1. [Section B, proof of Lemma 3.1] The proof begins "By the same arguments as in Lemma 3.1" but Lemma 3.1 is the lemma being proved; this should refer to the preceding derivation or to Lemma B.1. The sentence "what results, under the assumption that ..." is also garbled and should be rewritten.
  2. [Related Work] The phrase "wore-case measure" should be "worst-case measure".
  3. [Section 4 and Figure 8] There are typos in "repitition" (Section 4, caption of Figure 6) and "eastimation" (caption of Figure 8), and the phrase "dStdestimates" in Section 5.2 should be "dSt estimates".
  4. [Section 5.1, Figure 10] The label "2-distance to analytic strategy" is ambiguous; the text says Euclidean distance, so the figure label should state "Euclidean distance" or "L2 distance" explicitly.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's analytic derivations are self-contained and do not reduce to their inputs.

full rationale

The central derivation chain is not circular. Lemmas 3.1 and 3.2 start from the stated objective (outer Entr or CVaR applied to an inner Entr under a Gaussian posterior for the drift) and solve the resulting one-dimensional optimization explicitly; the out-of-sample performance formulas are then computed under the stated Gaussian assumptions using true parameters, not recovered by assuming the conclusion. The mixture comparison, Remark 3.1, explicitly identifies the mixture strategy with the expectation-uncertainty case, which is a known coincidence rather than a hidden redefinition. The subsampling section derives the induced distribution of the mean and shows via CLT that it reproduces the analytic strategies in the Gaussian limit; this is an approximation result with stated assumptions. The CVaR-SGD content in Theorem 5.1 is the closest to a concern, but the issue there is soundness: the proof's claim that the k-smallest-gradient average is an unbiased estimator of the CVaR_alpha gradient is not established for finite m and Algorithm 1's k targets E[k/m]=1-alpha while the theorem's CVaR_alpha uses the alpha-tail. That is a correctness gap in a proof, not a circular reduction in which the target result is assumed by construction or a fitted parameter is renamed as a prediction. The empirical sections do tune uncertainty-aversion parameters on the evaluation environments (e.g., Figure 4 chooses lambda' and alpha to maximize out-of-sample performance, and the PEP-KO backtest reports the maximum around alpha=15%), which weakens the strength of the reported improvements and should be read as overfitting/tuning risk, but the analytic comparisons and the stated conditions for improvement (e.g., Lemma B.5) are derived independently of any fitted value. Self-citations occur for deep hedging and related robustness work, but none of these citations carries a load-bearing uniqueness or ansatz claim that the paper's own derivation depends on. Under the hard rule requiring an explicit reduction of an equation to its input or a fitted parameter renamed as a prediction, no circular step can be quoted; the appropriate finding is 'no significant circularity'.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The framework relies on the Klibanoff smooth-ambiguity setup as given, on a Gaussian sampling model for analytic tractability, and on bootstrap-style approximations. The CVaR-SGD convergence additionally assumes convexity and Lipschitzness. The empirical evaluation involves robustness parameters and a Heston test distribution selected by the authors, which are not independently grounded.

free parameters (3)
  • uncertainty aversion lambda' (entropic) or alpha (CVaR) = grid/lookup, e.g. PEP-KO example alpha ~ 0.15; Figure 4 plots optimal values
    Choice of robustness level is selected to maximize out-of-sample performance in the experiment, rather than fixed by risk preferences or a validation set.
  • test distribution of Heston parameters = Figure 11 densities; no explicit calibration code
    The evaluation distribution used for the deep hedging experiment is calibrated to typical one-day Heston parameter changes, but the calibration procedure is not detailed.
  • plug-in Heston model parameters = Table 1: reversion speed 1.0, reversion level 0.03, vol of vol 0.2, correlation -0.8, drift 0, initial vol 0.03
    These are presented as plug-in parameter estimations for the deep hedging experiment, but the estimation procedure and confidence are not specified.
assumptions (5)
  • domain assumption Klibanoff et al. (2005) smooth ambiguity framework
    The outer uncertainty-measure objective (Eq. 2) is adopted from [46] without derivation.
  • domain assumption Estimator normality: mu_hat ~ N(mu, sigma^2/N) and sigma_hat^2 assumed equal to sigma^2
    Equations (5)-(6) and assumptions in Lemmas 3.1 and 3.2; variance estimation error is ignored for analytic tractability.
  • standard math Subsampling asymptotic: empirical measure and central limit theorem approximate the population model distribution
    Section 4 uses the uniform subsample measure and CLT to equate subsampling with the analytic uncertainty strategies.
  • domain assumption Convexity and rho-Lipschitz continuity for CVaR-SGD
    Theorem 5.1 assumptions; deep hedging objectives are generally nonconvex, so the theorem does not apply to the main deep learning experiment.
  • ad hoc to paper Test distribution of Heston models calibrated to typical one-day parameter changes
    Section 5.2, Figure 11: used to evaluate hedges, calibration not detailed and no independent evidence for its realism.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Uncertainty-Aware Strategies: A Model-Agnostic Framework for Robust Financial Optimization through Subsampling." pith.science (2026). https://pith.science/paper/UGQ4TKLQ

@misc{pith2026250607299,
  author       = {Pith},
  title        = {Pith review of: Uncertainty-Aware Strategies: A Model-Agnostic Framework for Robust Financial Optimization through Subsampling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UGQ4TKLQ}},
  note         = {Machine review of arXiv:2506.07299}
}
read the original abstract

This paper addresses the challenge of model uncertainty in quantitative finance, where decisions in portfolio allocation, derivative pricing, and risk management rely on estimating stochastic models from limited data. In practice, the unavailability of the true probability measure forces reliance on an empirical approximation, and even small misestimations can lead to significant deviations in decision quality. Building on the framework of Klibanoff et al. (2005), we enhance the conventional objective - whether this is expected utility in an investing context or a hedging metric - by superimposing an outer "uncertainty measure", motivated by traditional monetary risk measures, on the space of models. In scenarios where a natural model distribution is lacking or Bayesian methods are impractical, we propose an ad hoc subsampling strategy, analogous to bootstrapping in statistical finance and related to mini-batch sampling in deep learning, to approximate model uncertainty. To address the quadratic memory demands of naive implementations, we also present an adapted stochastic gradient descent algorithm that enables efficient parallelization. Through analytical, simulated, and empirical studies - including multi-period, real data and high-dimensional examples - we demonstrate that uncertainty measures outperform traditional mixture of measures strategies and our model-agnostic subsampling-based approach not only enhances robustness against model risk but also achieves performance comparable to more elaborate Bayesian methods.

Figures

Figures reproduced from arXiv: 2506.07299 by the authors.

Figure 1
Figure 1. Mean estimations based on the observation of [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Strategies based on the mean estimate of 90 days of observed asset returns (Examples [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. The out of sample performance of the optimal, plug-in, and mixture distribution (as defined later in Section [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: The optimal choice of uncertainty aversion [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Performance of the plug-in strategy with a custom variance adjustment [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: On the left, one path is “bootstrapped” by sampling its increments (with repetition) to create new paths. On ˆ [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: Comparing out-of-sample performance of the analytic strategies (Lemmata [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]
Figure 8
Figure 8. Figure 8: ARCH(1)-GARCH(1,1) parameters calibrated to PEP-KO pair process with Kalman filter uncertainty [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: Out-of-sample performance of strategies on PEP-KO pair process. Here, “Subsampling + CVaR” uses a [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]
Figure 10
Figure 10. Figure 10: Approximation of the uncertainty-aware strategy in Lemma [PITH_FULL_IMAGE:figures/full_fig_p016_10.png]
Figure 11
Figure 11. Figure 11: Test distributions for true underlying parameters, given the estimation of parameters in Table [PITH_FULL_IMAGE:figures/full_fig_p016_11.png]
Figure 12
Figure 12. Figure 12: For every Heston parameter set, one can compute the objective, which is the standard deviation of the P&L ˜ [PITH_FULL_IMAGE:figures/full_fig_p017_12.png]

Discussion (0). Sign in 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. Robust Control under Stationary Ambiguity

    cs.LG 2026-08 conditional novelty 7.0 of 10

    Policies trained under stationary latent ambiguity, implemented by refreshing the latent parameter, preserve robustness to regime shifts better than policies trained under a fixed latent draw.

Reference graph

Works this paper leans on

70 extracted references · 57 canonical work pages · cited by 1 Pith paper

  1. [1]

    Coherent measures of risk.Mathematical Finance, 9(3):203–228, 1999

    Philippe Artzner, Freddy Delbaen, Jean-Marc Eber, and David Heath. Coherent measures of risk.Mathematical Finance, 9(3):203–228, 1999

    work page 1999

  2. [2]

    Bailey, Jonathan M

    David H. Bailey, Jonathan M. Borwein, Marcos López de Prado, and Qiji Jim Zhu. Pseudo-mathematics and financial charlatanism: the effects of backtest overfitting on out-of-sample performance.Notices Amer. Math. Soc., 61(5):458–471, 2014

    work page 2014

  3. [3]

    Daniel Bartl, Samuel Drapeau, Jan Obłó j, and Johannes Wiesel. Sensitivity analysis of wasserstein distributionally robust optimization problems.Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 477(2256), dec 2021

    work page 2021

  4. [4]

    Brown, and Constantine Caramanis

    Dimitris Bertsimas, David B. Brown, and Constantine Caramanis. Theory and applications of robust optimization. SIAM Review, 53(3):464–501, 2011

    work page 2011

  5. [5]

    Bielecki, Igor Cialenco, and Marek Rutkowski

    Tomasz R. Bielecki, Igor Cialenco, and Marek Rutkowski. Arbitrage-free pricing of derivatives in nonlinear market models.Probability, Uncertainty and Quantitative Risk, 3(1):2, 2018

    work page 2018

  6. [6]

    Global portfolio optimization.Financial Analysts Journal, 48(5):28–43, 1992

    Fischer Black and Robert Litterman. Global portfolio optimization.Financial Analysts Journal, 48(5):28–43, 1992

    1992

  7. [7]

    The pricing of options and corporate liabilities.J

    Fischer Black and Myron Scholes. The pricing of options and corporate liabilities.J. Polit. Econ., 81(3):637–654, 1973

    work page 1973

  8. [8]

    Large-scale machine learning with stochastic gradient descent

    Léon Bottou. Large-scale machine learning with stochastic gradient descent. InProceedings of COMPSTAT’2010: 19th International Conference on Computational StatisticsParis France, August 22-27, 2010 Keynote, Invited and Contributed Papers, pages 177–186. Springer, 2010

    work page 2010

Show all 70 references
  1. [9]

    Stochastic gradient descent tricks

    Léon Bottou. Stochastic gradient descent tricks. InNeural networks: tricks of the trade: second edition, pages 421–436. Springer, 2012

    2012

  2. [10]

    Buehler, L

    H. Buehler, L. Gonon, J. Teichmann, and B. Wood. Deep hedging.Quant. Finance, 19(8):1271–1291, 2019

    2019

  3. [11]

    Deep hedging.Quantitative Finance, 19(8):1271– 1291, 2019

    Hans Buehler, Lukas Gonon, Josef Teichmann, and Ben Wood. Deep hedging.Quantitative Finance, 19(8):1271– 1291, 2019

    2019

  4. [12]

    A data-driven market simulator for small data environments.arXiv preprint arXiv:2006.14498, 2020

    Hans Buehler, Blanka Horvath, Terry Lyons, Imanol Perez Arribas, and Ben Wood. A data-driven market simulator for small data environments.arXiv preprint arXiv:2006.14498, 2020

    2006 arXiv

  5. [13]

    A data-driven market simulator for small data environments.SSRN Electronic Journal, 2020

    Hans Buehler, Blanka Horvath, Terry Lyons, Imanol Perez Arribas, and Ben Wood. A data-driven market simulator for small data environments.SSRN Electronic Journal, 2020. 18 Uncertainty-Aware StrategiesA PREPRINT

    2020

  6. [14]

    Deep bellman hedging.arXiv preprint arXiv:2207.00932, 2022

    Hans Buehler, Phillip Murray, and Ben Wood. Deep bellman hedging.arXiv preprint arXiv:2207.00932, 2022

    2022 arXiv

  7. [15]

    Campbell, Andrew W

    John Y . Campbell, Andrew W. Lo, and A.Craig MacKinlay.The Econometrics of Financial Markets. Princeton University Press, 1997

    1997

  8. [16]

    Algorithms for cvar optimization in mdps

    Yinlam Chow and Mohammad Ghavamzadeh. Algorithms for cvar optimization in mdps. InAdvances in Neural Information Processing Systems 27 (NIPS 2014), pages 3509–3517, 2014

    2014

  9. [17]

    Risk-sensitive and robust decision-making: a CVaR optimization approach

    Yinlam Chow, Aviv Tamar, Shie Mannor, and Marco Pavone. Risk-sensitive and robust decision-making: a CVaR optimization approach. InAdvances in Neural Information Processing Systems 28 (NeurIPS 2015), pages 1522–1530, 2015

    2015

  10. [18]

    Springer, 2017

    Charles K Chui, Guanrong Chen, et al.Kalman filtering. Springer, 2017

    2017

  11. [19]

    Springer, 2001

    Stuart Coles.An Introduction to Statistical Modeling of Extreme Values. Springer, 2001

    2001

  12. [20]

    Model uncertainty and its impact on the pricing of derivative instruments.Math

    Rama Cont. Model uncertainty and its impact on the pricing of derivative instruments.Math. Finance, 16(3):519– 547, 2006

    2006

  13. [21]

    Risk measures under model uncertainty: a bayesian viewpoint, 2022

    Christa Cuchiero, Guido Gazzani, and Irene Klein. Risk measures under model uncertainty: a bayesian viewpoint, 2022

    2022

  14. [22]

    Istituto italiano degli attuari, 1940

    Bruno De Finetti.Il problema dei pieni. Istituto italiano degli attuari, 1940

    1940

  15. [23]

    Distributionally robust optimization under moment uncertainty with application to data-driven problems.Operations Research, 58(3):595–612, 2010

    Erick Delage and Yinyu Ye. Distributionally robust optimization under moment uncertainty with application to data-driven problems.Operations Research, 58(3):595–612, 2010

    2010

  16. [24]

    A theoretical framework for the pricing of contingent claims in the presence of model uncertainty.The Annals of Applied Probability, 16(2):827–852, 2006

    Laurent Denis and Claude Martini. A theoretical framework for the pricing of contingent claims in the presence of model uncertainty.The Annals of Applied Probability, 16(2):827–852, 2006

    2006

  17. [25]

    Estimate nothing.Quantitative Finance, 14(12):2065–2072, 2014

    Moritz Duembgen and LCG Rogers. Estimate nothing.Quantitative Finance, 14(12):2065–2072, 2014

    2014

  18. [26]

    Princeton University Press, 2010

    Darrell Duffie.Dynamic asset pricing theory. Princeton University Press, 2010

    2010

  19. [27]

    John Wiley & Sons, 2013

    Daniel J Duffy.Finite Difference methods in financial engineering: a Partial Differential Equation approach. John Wiley & Sons, 2013

    2013

  20. [28]

    Bootstrap methods: another look at the jackknife

    Bradley Efron. Bootstrap methods: another look at the jackknife. InBreakthroughs in statistics: Methodology and distribution, pages 569–593. Springer, 1992

    1992

  21. [29]

    Affine processes under parameter uncertainty.Probability, Uncertainty and Quantitative Risk, 4(1):5, 2019

    Tolulope Fadina, Ariel Neufeld, and Thorsten Schmidt. Affine processes under parameter uncertainty.Probability, Uncertainty and Quantitative Risk, 4(1):5, 2019

    2019

  22. [30]

    De Gruyter, Berlin, Boston, 2004

    Hans Föllmer and Alexander Schied.Stochastic Finance: An Introduction in Discrete Time. De Gruyter, Berlin, Boston, 2004

    2004

  23. [31]

    Portfolio selection with parameter and model uncertainty: A multi-prior approach.Review of Financial Studies, 20(1):41–81, 2007

    Lorenzo Garlappi, Raman Uppal, and Tan Wang. Portfolio selection with parameter and model uncertainty: A multi-prior approach.Review of Financial Studies, 20(1):41–81, 2007

    2007

  24. [32]

    Pairs trading: Performance of a relative-value arbitrage rule.The review of financial studies, 19(3):797–827, 2006

    Evan Gatev, William N Goetzmann, and K Geert Rouwenhorst. Pairs trading: Performance of a relative-value arbitrage rule.The review of financial studies, 19(3):797–827, 2006

    2006

  25. [33]

    John Wiley & Sons, 2011

    Jim Gatheral.The volatility surface: a practitioner’s guide. John Wiley & Sons, 2011

    2011

  26. [34]

    Robust risk measurement and model risk.Quantitative Finance, 14(1):29–58, 2014

    Paul Glasserman and Xingbo Xu. Robust risk measurement and model risk.Quantitative Finance, 14(1):29–58, 2014

    2014

  27. [35]

    Robust portfolio selection problems.Mathematics of Operations Research, 28(1):1–38, 2003

    Donald Goldfarb and Garud Iyengar. Robust portfolio selection problems.Mathematics of Operations Research, 28(1):1–38, 2003

    2003

  28. [36]

    MIT press, 2016

    Ian Goodfellow, Yoshua Bengio, and Aaron Courville.Deep learning. MIT press, 2016

    2016

  29. [37]

    Managing smile risk.The Best of Wilmott, 1:249–296, 2002

    Patrick S Hagan, Deep Kumar, Andrew S Lesniewski, and Diana E Woodward. Managing smile risk.The Best of Wilmott, 1:249–296, 2002

    2002

  30. [38]

    Robust control and model uncertainty.American Economic Review, 91(2):60–66, 2001

    Lars Peter Hansen and Thomas J Sargent. Robust control and model uncertainty.American Economic Review, 91(2):60–66, 2001

    2001

  31. [39]

    Sargent.Robustness

    Lars Peter Hansen and Thomas J. Sargent.Robustness. Princeton University Press, Princeton, NJ, 2008. 19 Uncertainty-Aware StrategiesA PREPRINT

    2008

  32. [40]

    Steven L. Heston. A closed-form solution for options with stochastic volatility with applications to bond and currency options.Rev. Financ. Stud., 6(2):327–343, 1993

    1993

  33. [41]

    Deep learning in finance and banking: A literature review and classification

    Jian Huang, Junyi Chai, and Stella Cho. Deep learning in finance and banking: A literature review and classification. Frontiers of Business Research in China, 14(1):13, 2020

    2020

  34. [42]

    Garud N. Iyengar. Robust dynamic programming.Mathematics of Operations Research, 30(2):257–280, 2005

    2005

  35. [43]

    A deep reinforcement learning framework for the financial portfolio management problem.arXiv preprint arXiv:1706.10059, 2017

    Zhengyao Jiang, Dixing Xu, and Jinjun Liang. A deep reinforcement learning framework for the financial portfolio management problem.arXiv preprint arXiv:1706.10059, 2017

    2017 arXiv

  36. [44]

    Bayes–stein estimation for portfolio analysis.Journal of Financial and Quantitative Analysis, 21(3):279–292, 1986

    Philippe Jorion. Bayes–stein estimation for portfolio analysis.Journal of Financial and Quantitative Analysis, 21(3):279–292, 1986

    1986

  37. [45]

    Jang Ho Kim, Woo Chang Kim, and Frank J. Fabozzi. Recent advancements in robust optimization for investment management.Annals of Operations Research, 266(1):183–198, 2018

    2018

  38. [46]

    A smooth model of decision making under ambiguity

    Peter Klibanoff, Massimo Marinacci, and Sujoy Mukerji. A smooth model of decision making under ambiguity. Econometrica, 73(6):1849–1892, 2005

    2005

  39. [47]

    Boston and New York, Houghton Mifflin Company, 1921

    Frank Hyneman Knight.Risk, uncertainty and profit, volume 31 ofHart, Schaffner & Marx Price Essays. Boston and New York, Houghton Mifflin Company, 1921

    1921

  40. [48]

    Robust hedging gans: Towards automated robustification of hedging strategies.Applied Mathematical Finance, 31(3):164–201, 2024

    Yannick Limmer and Blanka Horvath and. Robust hedging gans: Towards automated robustification of hedging strategies.Applied Mathematical Finance, 31(3):164–201, 2024

    2024

  41. [49]

    Robust deep hedging.Quantitative Finance, pages 1–16, 2021

    Eva Lütkebohmert, Thorsten Schmidt, and Julian Sester. Robust deep hedging.Quantitative Finance, pages 1–16, 2021

    2021

  42. [50]

    McNeil and Rüdiger Frey

    Alexander J. McNeil and Rüdiger Frey. Estimation of tail-related risk measures for heteroscedastic financial time series: an extreme value approach.Journal of Empirical Finance, 7(3-4):271–300, 2000

    2000

  43. [51]

    optimized

    Richard O. Michaud. The markowitz optimization enigma: Is “optimized” optimal?Financial Analysts Journal, 45(1):31–42, 1989

    1989

  44. [52]

    Data-driven distributionally robust optimization using the wasserstein metric: Performance guarantees and tractable reformulations.Mathematical Programming, 171(1– 2):115–166, 2018

    Peyman Mohajerin Esfahani and Daniel Kuhn. Data-driven distributionally robust optimization using the wasserstein metric: Performance guarantees and tractable reformulations.Mathematical Programming, 171(1– 2):115–166, 2018

    2018

  45. [53]

    A risk-neutral equilibrium leading to uncertain volatility pricing

    Johannes Muhle-Karbe and Marcel Nutz. A risk-neutral equilibrium leading to uncertain volatility pricing. Finance and Stochastics, 22(2):281–295, 2018

    2018

  46. [54]

    Pakkanen

    Phillip Murray, Ben Wood, Hans Buehler, Magnus Wiese, and Mikko S. Pakkanen. Deep hedging: Continuous reinforcement learning for hedging of general portfolios across multiple risk aversions. InProceedings of the 3rd ACM International Conference on AI in Finance, pages 361–368....

    2022

  47. [55]

    Sig-wasserstein gans for time series generation

    Hao Ni, Lukasz Szpruch, Marc Sabate-Vidales, Baoren Xiao, Magnus Wiese, and Shujian Liao. Sig-wasserstein gans for time series generation. InProceedings of the Second ACM International Conference on AI in Finance, pages 1–8, 2021

    2021

  48. [56]

    Conditional sig-wasserstein gans for time series generation.arXiv preprint arXiv:2006.05421, 2020

    Hao Ni, Lukasz Szpruch, Magnus Wiese, Shujian Liao, and Baoren Xiao. Conditional sig-wasserstein gans for time series generation.arXiv preprint arXiv:2006.05421, 2020

    2006 arXiv

  49. [57]

    Deep exploration via bootstrapped dqn

    Ian Osband, Charles Blundell, Alexander Pritzel, and Benjamin Van Roy. Deep exploration via bootstrapped dqn. InAdvances in Neural Information Processing Systems 29 (NIPS 2016), pages 4026–4034, 2016

    2016

  50. [58]

    Estimating and backtesting risk under heavy tails.Journal of Empirical Finance, 65:1–22, 2022

    Marcin Pitera and Thorsten Schmidt. Estimating and backtesting risk under heavy tails.Journal of Empirical Finance, 65:1–22, 2022

    2022

  51. [59]

    A novel scaling approach for unbiased adjustment of risk estimators.arXiv preprint arXiv:2312.05655, 2023

    Marcin Pitera, Thorsten Schmidt, and Łukasz Stettner. A novel scaling approach for unbiased adjustment of risk estimators.arXiv preprint arXiv:2312.05655, 2023

    2023 arXiv

  52. [60]

    A stochastic approximation method.The annals of mathematical statistics, pages 400–407, 1951

    Herbert Robbins and Sutton Monro. A stochastic approximation method.The annals of mathematical statistics, pages 400–407, 1951

    1951

  53. [61]

    Tyrrell Rockafellar, Stanislav Uryasev, et al

    R. Tyrrell Rockafellar, Stanislav Uryasev, et al. Optimization of conditional value-at-risk.Journal of risk, 2:21–42, 2000. 20 Uncertainty-Aware StrategiesA PREPRINT

    2000

  54. [62]

    Cam- bridge university press, 2014

    Shai Shalev-Shwartz and Shai Ben-David.Understanding machine learning: From theory to algorithms. Cam- bridge university press, 2014

    2014

  55. [63]

    Shreve.Stochastic calculus for finance

    Steven E. Shreve.Stochastic calculus for finance. I. Springer Finance. Springer-Verlag, New York, 2004. The binomial asset pricing model

    2004

  56. [64]

    Shreve.Stochastic calculus for finance

    Steven E. Shreve.Stochastic calculus for finance. II. Springer Finance. Springer-Verlag, New York, 2004. Continuous-time models

    2004

  57. [65]

    Policy gradient for coherent risk measures.arXiv preprint arXiv:1502.03919, 2015

    Aviv Tamar, Yinlam Chow, Mohammad Ghavamzadeh, and Shie Mannor. Policy gradient for coherent risk measures.arXiv preprint arXiv:1502.03919, 2015

    2015 arXiv

  58. [66]

    John Wiley & Sons, 2000

    Domingo Tavella and Curt Randall.Pricing financial instruments: The finite difference method, volume 13. John Wiley & Sons, 2000

    2000

  59. [67]

    Wand and M.C

    M.P. Wand and M.C. Jones. Kernel smoothing.Chapman and Hall/CRC Monographs on Statistics and Applied Probability, 60, 1995

    1995

  60. [68]

    Deep hedging: Learning to simulate equity option markets.SSRN Electronic Journal, 2019

    Magnus Wiese, Lianjun Bai, Ben Wood, and Hans Buehler. Deep hedging: Learning to simulate equity option markets.SSRN Electronic Journal, 2019

    2019

  61. [69]

    Quant GANs: deep generation of financial time series.Quant

    Magnus Wiese, Robert Knobloch, Ralf Korn, and Peter Kretschmer. Quant GANs: deep generation of financial time series.Quant. Finance, 20(9):1419–1440, 2020

    2020

  62. [70]

    out-of-sample

    David Wu and Sebastian Jaimungal. Robust risk-aware option hedging.Applied Mathematical Finance, 30(3):153– 174, 2023. 21 Uncertainty-Aware StrategiesA PREPRINT A Basics on Risk Measures Consider the probability space (Ω,F,P) and let X∈L 0(Ω,F,P) . Then, the entropy [30] for r...

    2023

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.