{"paper_id":"2307.09631","title":"Deep Reinforcement Learning for ESG financial portfolio management","authors":["Eduardo C. Garrido-Merchán","Sol Mora-Figueroa-Cruz-Guzmán","María Coronado-Vaca"],"paper_date":"2023-06-19","paper_type":"perspective-policy","paper_url":"https://arxiv.org/pdf/2307.09631v1","page_url":"https://thequant.space/flowcharts/deep-reinforcement-learning-for-esg-financial-portfolio-mana/","scores":{"empirical_rigor":6,"math_complexity":7.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a deep reinforcement learning framework that incorporates ESG information and market regulation based on ESG scores for financial portfolio management.","quote":"We leveraged an Advantage Actor-Critic (A2C) agent and conducted our experiments using environments encoded within the OpenAI Gym, adapted from the FinRL platform.","summary":"This paper investigates the application of Deep Reinforcement Learning (DRL) for Environment, Social, and Governance (ESG) financial portfolio management, with a specific focus on the potential benefits of ESG score-based market regulation. We leveraged an Advantage Actor-Critic (A2C) agent and…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Training data runs from 2009‑01‑01 to 2020‑07‑01 and trading data from 2020‑07‑01 to 2021‑10‑31.","type":"data","stated":"explicit","quote":"The training data period for all the DRL agents goes from 2009-01-01 to 2020-07-01 and the trading period goes from 2020-07-01 to 2021-10-31."},{"assumption":"Missing monthly Bloomberg ESG values are approximated from nearby dates, introducing bias.","type":"data","stated":"explicit","quote":"As the data is incomplete, we approximate missing ESG values from the existing ones that are closer to the date, incurring in bias that must be targeted in future work."},{"assumption":"The reward function is modified to include an ESG performance term based on weighted ESG values of the portfolio.","type":"model","stated":"explicit","quote":"We have considered to modify the reward signal r of the DRL agent to take into account the ESG performance of the portfolio and not only the financial returns."},{"assumption":"Grants are proportionally allotted to portfolios based on returns and ESG scores, taxes are assigned to portfolios below the mean ESG score of the index.","type":"market","stated":"explicit","quote":"In the ESG-regulated market, grants were proportionally allotted to portfolios based on their returns and ESG scores, while taxes were assigned to portfolios below the mean ESG score of the index."},{"assumption":"The ESG value of the index (ψ) is computed assuming equal weights (1/A) for all assets.","type":"model","stated":"explicit","quote":"Having computed φ, we now compute φ assuming that every value of the weights wti is 1/A, where A is the total number of assets and denote this quantity as ψ, that represents the ESG value of the index."},{"assumption":"Hyper‑parameters (e.g., learning rate η = 2e‑4, entropy coefficient c = 5e‑3) are fixed and not tuned in the reported experiments.","type":"execution","stated":"explicit","quote":"We use 50000 timesteps, a learning rate of η = 2e−4 and an entropy coefficient of c = 5e−3."}]},"datasets":{"status":"available","items":[{"name":"Monthly Bloomberg ESG information","provider":"Bloomberg","kind":"alternative","access":"commercial","period":null,"universe":null,"quote":"we have gained access to monthly Bloomberg ESG information about different assets.","known_id":"bloomberg"},{"name":"Daily financial data from Yahoo Finance","provider":"Yahoo Finance API","kind":"market-prices","access":"public","period":"2009-01-01 to 2021-10-31","universe":null,"quote":"we use the Yahoo Finance API to download the dialy financial data from 2009-01-01 to 2021-10-31.","known_id":"yahoo-finance"},{"name":"Dow Jones Industrial Average (DJIA) index","provider":null,"kind":"benchmark","access":"public","period":null,"universe":"DJIA constituents","quote":"... comparative analysis of DRL agent performance under standard Dow Jones Industrial Average (DJIA) market conditions"},{"name":"IBEX‑35 index","provider":null,"kind":"benchmark","access":"public","period":null,"universe":"IBEX‑35 constituents","quote":"We also analyze the behaviour of the DRL agent in IBEX 35 and NASDAQ-100 indexes."},{"name":"NASDAQ‑100 index","provider":null,"kind":"benchmark","access":"public","period":null,"universe":"NASDAQ‑100 constituents","quote":"We also analyze the behaviour of the DRL agent in IBEX 35 and NASDAQ-100 indexes."}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"daily"},"transaction_costs":{"status":"available","modeled":"no","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R1","candidates":{"front_matter":"no","evidence_card":"unclear","evidence_card_quoted":false},"meaning":"a trading or execution result is reported without costs"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"In the ESG‑regulated market the A2C agent achieves higher annual return (0.462587) and cumulative return (0.173427) than in the unregulated DJIA market (0.435209 and 0.164137).","kind":"performance","quote":"Metric name DRL A2C Agent ESG Regulated DRL A2C Agent DJIA"},{"finding":"The ESG‑regulated A2C agent attains a higher Sharpe ratio (2.441398) compared to the unregulated baseline (2.340178).","kind":"performance","quote":"Sharpe ratio 2.441398 2.340178 2.127074"},{"finding":"Including ESG variables in the state space does not degrade performance; agents with ESG information perform comparably to those without.","kind":"performance","quote":"Remarkably, we found that including ESG information into the state space of the agents did not adversely affect the performance of the portfolio."},{"finding":"Statistical tests across algorithms and indexes fail to reject the null hypothesis that ESG‑aware and ESG‑agnostic policies deliver the same results.","kind":"statistical","quote":"we can not reject the null hypothesis that both policies deliver the same results, in statistical terms."}],"reported_performance":null,"out_of_sample":{"method":"holdout","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":21,"citations":{"cited_by_in_corpus":2,"by_independent_authors":2,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":1,"unclassified":1},"independent_papers":[],"scope":"Citations between papers in this corpus only; citations from outside it are not counted."}},"implementation":{"status":"available","public":false,"code_url":null,"note":"No repository presented by the authors as their own was found in the paper. Absence here is not proof that no code exists."},"replication":{"status":"none_found","card":null,"note":"No replication of this paper has been run by thequant.space yet."},"extraction":{"text_source":"fulltext","model":"openai/gpt-oss-120b","version":1,"extracted_at":"2026-11-14T18:25:41Z"},"provenance":{"claims":"Everything here is what the authors state in the paper. Nothing was re-run or reproduced by the platform.","quotes":"Assumptions, datasets, cost detail and findings were extracted by a language model. Every quote is verbatim paper text. Where an item has `support`, \"full\" means every figure in it appears in the quoted sentences and \"partial\" means a figure is stated elsewhere in the paper; items with a figure found nowhere in the paper were discarded.","datasets_known":"datasets.known lists well-known vendors and public datasets by a stable id, found by a dictionary over the full text where the sentence says the data is used; it is independent of the model and covers only names in the dictionary.","flags":"transaction_costs.modeled, results.evaluation_type and results.out_of_sample.method are platform classifications (structured-decision models); see https://thequant.space/score-guide/#penalty-flags","comparisons":"supports / contradicts / qualifies are a language model's reading of two papers' quoted findings, restricted to closely related papers that share no author. They are pointers to read, not verified replications, and coverage is partial.","implementation":"code_url is a repository the paper itself presents as its own (rule-scored, ambiguous cases judged by a model), checked for liveness when it was recorded."},"docs":"https://thequant.space/agents/"}