Pith. sign in

REVIEW 3 major objections 5 minor 36 references

A PPO agent that jointly chases return and reliability can rebalance portfolios under transaction costs and tail-risk measures as well as classical multi-objective optimizers across market regimes.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-11 01:53 UTC pith:P3CFOFEO

load-bearing objection Solid integration of reliability constraints + transaction costs into PPO portfolio RL, but the EVaR table is broken and the three-risk claim does not fully hold. the 3 major comments →

arxiv 2607.06610 v1 pith:P3CFOFEO submitted 2026-07-07 cs.LG cs.AImath.OC

Deep Reinforcement Learning for Reliability Based Bi-Objective Portfolio Optimization

Sounaq Das , Tanmay Sen , Raghu Nandan Sengupta , Aditya Gupta This is my paper
classification cs.LG cs.AImath.OC
keywords portfolio optimizationdeep reinforcement learningProximal Policy Optimizationreliability-based designCVaREVaRt-copulatransaction costs
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Portfolio managers need allocations that stay feasible with high probability when returns are uncertain, transaction costs bite, and crashes fatten the tails. Most reliability-based optimizers are static: they solve once and ignore sequential rebalancing. This paper turns the problem into a Markov decision process and trains a Proximal Policy Optimization agent whose reward rewards net return, risk-adjusted improvement, and empirical reliability of both return and risk constraints. Scenarios are built from GARCH volatility, extreme-value tails, a t-copula for dependence, and quasi-Monte Carlo sampling. The same agent is run under three risk measures—variance, CVaR, and EVaR—and compared with equal-weight and NSGA-II portfolios on ten global equity indices through pre-COVID, COVID, and post-COVID windows, plus a high-dimensional FTSE 100 test. The claim is that the learned policies deliver competitive risk–return trade-offs, keep downside risk in check under stress, and still scale when the asset universe grows large.

Core claim

MORP-DRL—a PPO actor–critic trained on reliability-shaped rewards and GARCH–EVT–t-copula scenarios—produces sequential portfolio policies that match or approach NSGA-II on return, volatility, Sharpe, and the chosen risk measure while satisfying transaction-cost and bound constraints across three market regimes and remaining usable on the FTSE 100 universe.

What carries the argument

The reliability-shaped PPO reward: net-return and risk-improvement terms plus a piecewise penalty Ψ(ρ, β) that rewards estimated return and risk reliability above prescribed thresholds β1 = β2 = 0.65, with scenarios generated by GARCH(1,1), EVT, t-copula, and quasi-Monte Carlo.

Load-bearing premise

That reliability probabilities estimated from the same scenario generator used in training are faithful enough proxies for true out-of-sample constraint satisfaction that shaping the reward with them produces genuinely reliable policies rather than scenario-overfit ones.

What would settle it

Hold out a later market window never used in scenario generation or training; recompute empirical return and risk reliability of the frozen PPO policy on that window and check whether both still clear the 0.65 thresholds while risk–return metrics remain competitive with NSGA-II.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes MORP-DRL, a PPO-based multi-objective reliability portfolio framework that jointly optimizes expected return and downside risk under variance, CVaR, and EVaR, with proportional transaction costs and probabilistic reliability constraints. Uncertainty is modeled via GARCH(1,1), EVT (GPD tails), a t-copula, and QMC scenarios; reliability estimates enter a shaped reward. The method is tested on ten global equity indices over pre-COVID, COVID, and post-COVID regimes and scaled to FTSE 100 constituents, with NSGA-II and equal-weight benchmarks. The authors claim competitive risk–return performance, better downside control in stress periods, and high-dimensional scalability.

Significance. If the three-risk-measure results hold under a consistent evaluation protocol, the paper would be a useful empirical contribution at the intersection of reliability-based design optimization and DRL portfolio management: it couples sequential rebalancing, transaction costs, and coherent tail-risk objectives with scenario-based reliability constraints, and provides a concrete FTSE 100 scaling path via time-decomposed QMC. The variance and CVaR legs (Tables 1–2) already show both optimizers beating equal weight with plausible magnitudes, and the allocation/concentration appendices usefully document diversification differences between PPO and NSGA-II. The EVaR leg is currently the load-bearing weak point for the multi-measure claim.

major comments (3)
  1. [§7 Table 3; also §3.1, Model C Eq. (8), §5.1] Table 3 (Model C, EVaR): reported EVaR levels are not on a common scale. Base and NSGA-II EVaR sit near 4.75 across all three regimes, while PPO reports 0.0013 (Pre), 0.0073 (COVID), and 0.0124 (Post)—orders of magnitude smaller—yet Sharpe/return are still presented as comparable. Section 3.1 defines EVaR via the standard exponential moment form, and Models B/C and §5.1 use the same risk object for reward and reliability, but the manuscript never states units, annualization, a different estimator, or a different evaluation protocol for PPO vs NSGA-II. Without a reconciled definition and recomputed Table 3 (and the corresponding Pareto frontiers in Figs. 6–7), the claim that MORP-DRL is competitive under all three risk measures, and that it reduces downside risk under EVaR in stress periods, is not supported by the reported numbers.
  2. [§5.1 reward/reliability; §6.2; §7] Reliability evaluation is largely in-sample relative to the training scenario generator. §5.1 estimates ρ_ret and ρ_risk from the same GARCH–EVT–t-copula–QMC scenarios used to train PPO, then shapes the reward with Ψ(ρ,β) at β1=β2=0.65 (§6.2). Tables 1–4 report return/volatility/Sharpe/risk but do not report realized reliability frequencies on held-out paths, alternative copulas, or bootstrap stress tests. For a reliability-based claim, the paper should show that the learned policies actually meet the prescribed β levels out of the training scenario measure; otherwise the reliability component may be scenario-overfit rather than a genuine feasibility guarantee.
  3. [§4 Models A–C; §5.1 Eq. (1); §7 Figs. 2–7] The bi-objective / multi-objective framing is only partially realized in the DRL path. Models A–C are written as bi-objective programs (max return, min risk) with reliability constraints, and NSGA-II produces Pareto frontiers, but PPO optimizes a single scalar reward r_t = ΔS_t + ΔR_t + R_rel_t (Eq. 1) with fixed hand-tuned scales in Ψ. The paper should clarify how the reported PPO "Pareto frontiers" (Figs. 3, 5, 7) are generated (e.g., reward-weight sweeps, threshold sweeps, or post-hoc filtering) and whether they are comparable to NSGA-II’s nondominated sets under the same constraint set. Without that protocol, claims of multi-objective competitiveness rest on single-policy snapshots rather than a controlled frontier comparison.
minor comments (5)
  1. [Title; Abstract; §1] Title and abstract say "bi-objective" while the framework name is multi-objective reliability based (MORP-DRL) and three risk measures are treated as alternative models; align terminology throughout.
  2. [§5.1 Eq. (1)] Eq. (1) writes S_t = R_t / R_t (typo); the intended risk-adjusted score should be stated explicitly (e.g., return/risk or Sharpe-like).
  3. [Algorithm 1; §5.2] Action sampling is described as a_t ~ Dirichlet(π_θ(s_t)) then projected onto bounds; specify how the policy network parameterizes the Dirichlet and how bound projection interacts with the PPO probability ratio.
  4. [Tables 1–3; §7] CPU times in Tables 1–3 show PPO ~10–15× slower than NSGA-II; a short note on wall-clock fairness (episode count, scenario budget, hardware) would help interpret the efficiency comparison in §7.
  5. [§1; References] Several repeated Markowitz citations and minor formatting issues (e.g., "capital asset Pricing model", missing spaces in headings) should be cleaned for production.

Circularity Check

1 steps flagged

Empirical RL portfolio method; reward-shaped reliability is design, not a by-construction prediction of the reported metrics.

specific steps
  1. other [§5.1 Reward Function and Reliability Constraints; Algorithm 1 lines 6, 15–16]
    "Generate market scenarios using GARCH(1,1), EVT, t-copula, and QMC ... Estimate reliability probabilities: ρ_ret_t , ρ_risk_t ... Compute reward r_t using Eq. (1) ... R_rel_t = Ψ(ρ_ret_t , β1) + Ψ(ρ_risk_t , β2)"

    Reliability probabilities that shape the PPO reward are estimated from the same GARCH–EVT–t-copula–QMC scenarios used to train the agent. This couples the training objective to the scenario model, so high in-sample ρ does not independently prove out-of-sample reliability. It is environment design, not a mathematical reduction of the paper’s reported historical return/Sharpe claims to their inputs; evaluation remains on separate historical regimes.

full rationale

MORP-DRL is a methods paper: it formulates bi-objective reliability-constrained portfolio models (variance/CVaR/EVaR with transaction costs), generates scenarios via GARCH–EVT–t-copula–QMC, trains PPO with a hand-designed reward that includes reliability penalties Ψ(ρ,β), and evaluates annualized return, volatility, Sharpe, and risk on historical pre/COVID/post regimes against equal-weight and NSGA-II. That chain is not a first-principles derivation that redefines its target as an input. The reliability terms in the reward use the same scenario generator used for training, which is standard RL environment design and can cause scenario-overfit, but evaluation metrics (Tables 1–4) are computed on realized historical series, not tautologically equal to the fitted reward components. Self-citations to prior reliability portfolio work ([32] Sengupta et al.) are used as baselines that the paper extends with transaction costs and DRL; they are not uniqueness theorems that force the present claims. No equation reduces a reported prediction to a fitted parameter by construction. Numerical inconsistencies in Table 3 EVaR scales (NSGA-II ~4.75 vs PPO ~0.001–0.012) are a correctness/reporting issue, not circularity of the derivation. Score 1 for mild shared-scenario design only.

Axiom & Free-Parameter Ledger

8 free parameters · 5 axioms · 1 invented entities

The central empirical claim rests on a stack of standard finance/ML modeling choices plus many hand-set knobs (reliability levels, bounds, reward scales, PPO hyperparameters) and fitted time-series models (GARCH, GPD tails, t-copula). No new physical entity is postulated; the ‘invented’ object is the MORP-DRL reward/constraint packaging. Load-bearing domain assumptions include scenario fidelity of GARCH–EVT–t-copula–QMC and that historical regime splits plus yfinance closes are adequate for policy evaluation.

free parameters (8)
  • reliability thresholds β1, β2 = 0.65
    Fixed at 0.65 for both return and risk reliability; directly shapes the reward via Ψ and defines feasibility. Not derived from investor utility or regulatory capital rules.
  • portfolio weight bounds wi,min, wi,max = 0.03 ≤ wi ≤ 0.35
    Hard box constraints on every asset; change the feasible set and concentration statistics.
  • proportional transaction cost ki = 2 bps
    2 basis points per rebalance on weight changes; enters net return and reliability indicators.
  • reliability reward scales in Ψ = +30 / −100 (and ×100)
    Hand coefficients +30(ρ−β) when feasible and −100(β−ρ) when violated, plus ×100 scalings on ΔS and ΔR; dominate learning signal relative to raw returns.
  • PPO training hyperparameters = AdamW 3e-4, γ=0.99, ε=0.2, 1000 episodes
    Learning rate, discount, clip ε, episode count control policy quality; not cross-validated in the text.
  • EVT exceedance threshold = 95th percentile
    95th percentile for GPD tail modeling; affects CVaR/EVaR scenario severity.
  • GARCH(1,1) and t-copula parameters = asset- and period-specific (not tabulated)
    Fitted per asset/period from historical returns; generate the training/evaluation scenarios that define both risk and reliability.
  • target thresholds r*_p, σ²*_p, CVaR*, EVaR* = not fully specified numerically in §6
    Reliability constraints compare scenario outcomes to these investor-specified caps/floors; values are not fully enumerated in the experimental section.
axioms (5)
  • domain assumption Asset returns are adequately described by GARCH(1,1) conditional volatility, GPD tails above a high threshold, and a Student-t copula dependence structure, with QMC scenarios representing future uncertainty.
    §5.1 transition dynamics and §6.1 preprocessing; all reliability probabilities and risk estimates inherit this generative model.
  • domain assumption Portfolio choice is an MDP with continuous weights on the simplex and unknown transition kernel learnable from historical/simulated paths.
    §5.1 MDP formulation; justifies PPO over static optimizers.
  • domain assumption CVaR and EVaR are coherent risk measures suitable as bi-objective risk legs alongside expected return under reliability constraints.
    §3.1 risk measures and Models B–C in §4; standard in risk literature but choice of α and formulation affects rankings.
  • standard math Clipped PPO surrogate with Dirichlet-sampled actions and projection onto weight bounds yields stable continuous portfolio policies.
    §5.2 and Algorithm 1; relies on standard PPO theory [37] without new convergence proof for this reward.
  • ad hoc to paper Pre-COVID / COVID / Post-COVID calendar splits (2018–2019 / 2020–2021 / 2022–2023) are meaningful distinct market regimes for evaluating adaptation.
    §6.1 period definitions; convenient but not uniquely justified relative to other crisis definitions.
invented entities (1)
  • MORP-DRL reward (ΔS + ΔR + R_rel with Ψ reliability shaping) no independent evidence
    purpose: Scalarizes multi-objective return, risk-adjusted improvement, and probabilistic reliability into a single PPO training signal under transaction costs.
    The specific combination and numerical shaping are paper-specific design choices without independent external validation beyond the reported backtests.

pith-pipeline@v1.1.0-grok45 · 22567 in / 4395 out tokens · 61240 ms · 2026-07-11T01:53:22.464966+00:00 · methodology

0 comments
read the original abstract

Portfolio optimization under uncertainty is inherently a multi-objective decision problem involving complex interactions among return, risk, market dynamics, and practical investment constraints. Existing reliability based portfolio optimization approaches primarily rely on static optimization frameworks and often fail to capture sequential decision making, tail risk, and market frictions such as transaction costs. To address these limitations, we propose a deep reinforcement learning framework for multi-objective reliability based portfolio optimization (MORP-DRL). The proposed framework jointly optimizes expected return and downside risk using three complementary risk measures: variance, Conditional Value-at-Risk (CVaR), and Entropic Value-at-Risk (EVaR). To model uncertainty and heavy-tailed market behavior, asset returns are represented using GARCH(1,1), Extreme Value Theory, and a t-copula dependence structure, while realistic scenarios are generated through quasi-Monte Carlo simulation. A Proximal Policy Optimization (PPO) based strategy is developed under practical constraints including transaction costs and portfolio bounds, and is benchmarked against NSGA-II. Experiments on ten global equity indices across pre-COVID, COVID, and post-COVID market regimes demonstrate that MORP-DRL achieves competitive risk-return performance, reduced downside risk during periods of market stress, and scalability to high-dimensional portfolio settings.

Figures

Figures reproduced from arXiv: 2607.06610 by Aditya Gupta, Raghu Nandan Sengupta, Sounaq Das, Tanmay Sen.

Figure 1
Figure 1. Figure 1: Workflow of the Proposed PPO-Based Portfolio Optimization Model [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: NSGA-II based Pareto frontiers under the Variance risk measure for (a) Pre COVID, (b) COVID, and (c) Post [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: PPO based Pareto frontiers under the Variance risk measure for (a) Pre COVID, (b) COVID, and (c) Post [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: NSGA-II based Pareto frontiers under the CVaR risk measure for (a) Pre COVID, (b) COVID, and (c) Post [PITH_FULL_IMAGE:figures/full_fig_p016_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Pareto frontiers obtained using PPO under the CVaR risk measure for (a) Pre COVID, (b) COVID, and (c) [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Pareto frontiers obtained using NSGA-II under the EVaR risk measure for (a) Pre COVID, (b) COVID, and [PITH_FULL_IMAGE:figures/full_fig_p018_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Pareto frontiers obtained using PPO under the EVaR risk measure for (a) Pre COVID, (b) COVID, and (c) [PITH_FULL_IMAGE:figures/full_fig_p019_7.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

36 extracted references · 36 canonical work pages · 4 internal anchors

  1. [2]

    Portfolio selection.Handbook of finance, 2:3–13, 2008

    Frank J Fabozzi, Harry M Markowitz, and Francis Gupta. Portfolio selection.Handbook of finance, 2:3–13, 2008

    work page 2008

  2. [3]

    The capital asset pricing model: Theory and evidence.Journal of economic perspectives, 18(3):25–46, 2004

    Eugene F Fama and Kenneth R French. The capital asset pricing model: Theory and evidence.Journal of economic perspectives, 18(3):25–46, 2004

    work page 2004

  3. [4]

    Value at risk.Financial analysts journal, 56(2):47–67, 2000

    Thomas J Linsmeier and Neil D Pearson. Value at risk.Financial analysts journal, 56(2):47–67, 2000

    work page 2000

  4. [5]

    Conditional value-at-risk for general loss distributions.Journal of banking & finance, 26(7):1443–1471, 2002

    R Tyrrell Rockafellar and Stanislav Uryasev. Conditional value-at-risk for general loss distributions.Journal of banking & finance, 26(7):1443–1471, 2002

    work page 2002

  5. [6]

    A comparison of risk measures for portfolio optimization with cardinality constraints.Expert Systems with Applications, 228:120412, 2023

    Henrique Pinto Ramos, Marcelo Brutti Righi, Pablo Cristini Guedes, and Fernanda Maria Müller. A comparison of risk measures for portfolio optimization with cardinality constraints.Expert Systems with Applications, 228:120412, 2023

    work page 2023

  6. [7]

    A simulation comparison of risk measures for portfolio optimization

    Marcelo Brutti Righi and Denis Borenstein. A simulation comparison of risk measures for portfolio optimization. Finance Research Letters, 24:105–112, 2018

    work page 2018

  7. [8]

    Portfolio optimisation problem: A taxonomic review of solution methodologies.IEEE Access, PP:1–1, 01 2023

    Zi Loke, Say Leng Goh, Graham Kendall, Salwani Abdullah, and Nasser Sabar. Portfolio optimisation problem: A taxonomic review of solution methodologies.IEEE Access, PP:1–1, 01 2023

    work page 2023

  8. [9]

    Fifty years of portfolio optimization

    Ahti Salo, Michalis Doumpos, Juuso Liesiö, and Constantin Zopounidis. Fifty years of portfolio optimization. European Journal of Operational Research, 318(1):1–18, 2024

    work page 2024

  9. [10]

    A survey of swarm intelligence for portfolio optimization: Algorithms and applications.Swarm and evolutionary computation, 39:36–52, 2018

    Okkes Ertenlice and Can B Kalayci. A survey of swarm intelligence for portfolio optimization: Algorithms and applications.Swarm and evolutionary computation, 39:36–52, 2018

    work page 2018

  10. [11]

    MIT press Cambridge, 2016

    Ian Goodfellow, Yoshua Bengio, Aaron Courville, and Yoshua Bengio.Deep learning, volume 1. MIT press Cambridge, 2016

    work page 2016

  11. [12]

    Deep learning with long short-term memory networks for financial market predictions.European journal of operational research, 270(2):654–669, 2018

    Thomas Fischer and Christopher Krauss. Deep learning with long short-term memory networks for financial market predictions.European journal of operational research, 270(2):654–669, 2018

    work page 2018

  12. [13]

    Prediction based mean-value- at-risk portfolio optimization using machine learning regression algorithms for multi-national stock markets

    Jyotirmayee Behera, Ajit Kumar Pasayat, Harekrushna Behera, and Pankaj Kumar. Prediction based mean-value- at-risk portfolio optimization using machine learning regression algorithms for multi-national stock markets. Engineering Applications of Artificial Intelligence, 120:105843, 2023

    work page 2023

  13. [14]

    Deep reinforcement learning: A brief survey.IEEE signal processing magazine, 34(6):26–38, 2017

    Kai Arulkumaran, Marc Peter Deisenroth, Miles Brundage, and Anil Anthony Bharath. Deep reinforcement learning: A brief survey.IEEE signal processing magazine, 34(6):26–38, 2017. 20 APREPRINT- JULY9, 2026

    work page 2017

  14. [15]

    Risk-adjusted deep reinforcement learning for portfolio optimization: A multi-reward approach.International Journal of Computational Intelligence Systems, 18(1):126, 2025

    Himanshu Choudhary, Arishi Orra, Kartik Sahoo, and Manoj Thakur. Risk-adjusted deep reinforcement learning for portfolio optimization: A multi-reward approach.International Journal of Computational Intelligence Systems, 18(1):126, 2025

    work page 2025

  15. [16]

    Empirical asset pricing via machine learning.The Review of Financial Studies, 33(5):2223–2273, 2020

    Shihao Gu, Bryan Kelly, and Dacheng Xiu. Empirical asset pricing via machine learning.The Review of Financial Studies, 33(5):2223–2273, 2020

    work page 2020

  16. [17]

    A cvar-constrained safe reinforcement learning framework with action repair for practical portfolio optimization.IEEE Transactions on Artificial Intelligence, 2026

    Himanshu Choudhary, Arishi Orra, Manoj Thakur, Xiao-Zhi Gao, and Prabhat Kumar Sahu. A cvar-constrained safe reinforcement learning framework with action repair for practical portfolio optimization.IEEE Transactions on Artificial Intelligence, 2026

    work page 2026

  17. [18]

    Portfolio selection.The Journal of Finance, 7(1):77–91, 1952

    Harry Markowitz. Portfolio selection.The Journal of Finance, 7(1):77–91, 1952

    work page 1952

  18. [19]

    60 years of portfolio optimization: Practical challenges and current trends.European Journal of Operational Research, 234(2):356–371, 2014

    Petter N Kolm, Reha Tütüncü, and Frank J Fabozzi. 60 years of portfolio optimization: Practical challenges and current trends.European Journal of Operational Research, 234(2):356–371, 2014

    work page 2014

  19. [20]

    Multi-objective heuristic algorithms for practical portfolio optimization and rebalancing with transaction cost.Applied Soft Computing, 67:865–894, 2018

    Suraj S Meghwani and Manoj Thakur. Multi-objective heuristic algorithms for practical portfolio optimization and rebalancing with transaction cost.Applied Soft Computing, 67:865–894, 2018

    work page 2018

  20. [21]

    Continuous-time optimal investment with portfolio constraints: a reinforcement learning approach.European Journal of Operational Research, 2025

    Huy Chau, Duy Nguyen, and Thai Nguyen. Continuous-time optimal investment with portfolio constraints: a reinforcement learning approach.European Journal of Operational Research, 2025

    work page 2025

  21. [22]

    Capturing Financial markets to apply Deep Reinforcement Learning

    Souradeep Chakraborty. Capturing financial markets to apply deep reinforcement learning.arXiv preprint arXiv:1907.04373, 2019

    work page internal anchor Pith review Pith/arXiv arXiv 1907

  22. [23]

    Application of deep reinforcement learning in stock trading strategies and stock forecasting.Computing, 2019

    Yuming Li, Pin Ni, and Victor Chang. Application of deep reinforcement learning in stock trading strategies and stock forecasting.Computing, 2019

    work page 2019

  23. [24]

    Application of deep q-network in portfolio management

    Ziming Gao, Yuan Gao, Yi Hu, Zhengyong Jiang, and Jionglong Su. Application of deep q-network in portfolio management. In2020 5th IEEE International Conference on Big Data Analytics (ICBDA), pages 268–275. IEEE, 2020

    work page 2020

  24. [25]

    A framework of hierarchical deep q-network for portfolio management

    Yuan Gao, Ziming Gao, Yi Hu, Sifan Song, Zhengyong Jiang, and Jionglong Su. A framework of hierarchical deep q-network for portfolio management. InICAART (2), pages 132–140, 2021

    work page 2021

  25. [26]

    Deep reinforcement learning for portfolio selection.Global Finance Journal, 62:101016, 2024

    Yifu Jiang, Jose Olmo, and Majed Atwi. Deep reinforcement learning for portfolio selection.Global Finance Journal, 62:101016, 2024

    work page 2024

  26. [27]

    Predictive multi- period multi-objective portfolio optimization based on higher order moments: Deep learning approach.Computers & industrial engineering, 183:109450, 2023

    Shaghayegh Abolmakarem, Farshid Abdi, Kaveh Khalili-Damghani, and Hosein Didehkhani. Predictive multi- period multi-objective portfolio optimization based on higher order moments: Deep learning approach.Computers & industrial engineering, 183:109450, 2023

    work page 2023

  27. [28]

    Advancing Investment Frontiers: Industry-grade Deep Reinforcement Learning for Portfolio Optimization

    Philip Ndikum and Serge Ndikum. Advancing investment frontiers: Industry-grade deep reinforcement learning for portfolio optimization.arXiv preprint arXiv:2403.07916, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  28. [29]

    Reinforcement learning for deep portfolio optimization.Electronic Research Archive, 32(9):5176, 2024

    Ruyu Yan, Jiafei Jin, and Kun Han. Reinforcement learning for deep portfolio optimization.Electronic Research Archive, 32(9):5176, 2024

    work page 2024

  29. [30]

    Deep reinforcement learning for stock portfolio optimization by connecting with modern portfolio theory.Expert Systems with Applications, 218:119556, 2023

    Junkyu Jang and NohYoon Seong. Deep reinforcement learning for stock portfolio optimization by connecting with modern portfolio theory.Expert Systems with Applications, 218:119556, 2023

    work page 2023

  30. [31]

    Model-based Deep Reinforcement Learning for Dynamic Portfolio Optimization

    Pengqian Yu, Joon Sern Lee, Ilya Kulyatin, Zekun Shi, and Sakyasingha Dasgupta. Model-based deep reinforce- ment learning for dynamic portfolio optimization.arXiv preprint arXiv:1901.08740, 2019

    work page internal anchor Pith review Pith/arXiv arXiv 1901

  31. [32]

    Bi-objective reliability based optimization: an application to investment analysis.Annals of Operations Research, 333(1):47–78, 2024

    Raghu Nandan Sengupta, Aditya Gupta, Subhankar Mukherjee, and Gregor Weiss. Bi-objective reliability based optimization: an application to investment analysis.Annals of Operations Research, 333(1):47–78, 2024

    work page 2024

  32. [33]

    Reliability-based design optimization: a state-of-the-art review of its methodologies, applications, and challenges.Structural and Multidisciplinary Optimization, 67(9):168, 2024

    Weifei Hu, Sichuang Cheng, Jiquan Yan, Jin Cheng, Xiang Peng, Hyunkyoo Cho, and Ikjin Lee. Reliability-based design optimization: a state-of-the-art review of its methodologies, applications, and challenges.Structural and Multidisciplinary Optimization, 67(9):168, 2024

    work page 2024

  33. [34]

    Reliability in portfolio optimization using uncertain estimates.Sankhya B, 85(Suppl 1):199–233, 2023

    Raghu Nandan Sengupta, Rachit Seth, and Peter Winker. Reliability in portfolio optimization using uncertain estimates.Sankhya B, 85(Suppl 1):199–233, 2023

    work page 2023

  34. [35]

    Multi-objective possibilistic model for portfolio selection with transaction cost.Journal of computational and applied mathematics, 228(1):188–196, 2009

    P Jana, TK Roy, and SK Mazumder. Multi-objective possibilistic model for portfolio selection with transaction cost.Journal of computational and applied mathematics, 228(1):188–196, 2009

    work page 2009

  35. [36]

    Artificial bee colony algorithm for constrained possibilistic portfolio optimization problem.Physica A: Statistical Mechanics and its Applications, 429:125–139, 2015

    Wei Chen. Artificial bee colony algorithm for constrained possibilistic portfolio optimization problem.Physica A: Statistical Mechanics and its Applications, 429:125–139, 2015

    work page 2015

  36. [37]

    Proximal Policy Optimization Algorithms

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms.ArXiv, abs/1707.06347, 2017. 21 APREPRINT- JULY9, 2026 A Appendix Table A1: Portfolio Weights Comparison Across Market Regimes – Variance Optimization Pre-COVID COVID Post-COVID Asset NSGA-II PPO NSGA-II PPO NSGA-II PPO SSE 3.0 6.7 25.7 7...

    work page internal anchor Pith review Pith/arXiv arXiv 2017