{"paper_id":"2512.10913","title":"Reinforcement Learning in Financial Decision Making: A Systematic Review of Performance, Challenges, and Implementation Strategies","authors":["Mohammad Rezoanul Hoque","Md Meftahul Ferdaus","M. Kabir Hassan"],"paper_date":"2025-12-11","paper_type":"survey-review","paper_url":"https://arxiv.org/pdf/2512.10913v1","page_url":"https://thequant.space/flowcharts/reinforcement-learning-in-financial-decision-making-a-syste/","scores":{"empirical_rigor":3,"math_complexity":6,"quadrant":"Lab Rats"},"proposes":{"status":"available","statement":"The study proposes a unified framework to address common concerns such as explainability, robustness, and deployment feasibility in reinforcement learning for finance.","quote":"This study proposes a unified framework to address common concerns such as explainability, robustness, and deployment feasibility.","summary":"Reinforcement learning (RL) is an innovative approach to financial decision making, offering specialized solutions to complex investment problems where traditional methods fail. This review analyzes 167 articles from 2017–2025, focusing on market making, portfolio optimization, and algorithmic…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"The synthetic dataset reproduces key statistical patterns observed in the literature and preserves relationships among variables.","type":"data","stated":"explicit","quote":"Empirical Validation Framework To validate the meta-analysis findings, a synthetic dataset was developed that repro-duces key statistical patterns observed in the literature. The data generation process preserves the statistical relationships observed in the literature while enabling controlled analysis of performance drivers. 4.","support":"full"},{"assumption":"Transaction costs and market impact can be captured through explicit reward‑structure constraints in the RL formulation.","type":"execution","stated":"explicit","quote":"Recent portfolio op-timization research focuses on incorporating realistic limitations, such as transaction costs, into RL frameworks. The RL approach enables objective optimization by con-sidering transaction costs and market impact constraints through explicit reward structure constraints, along with legal and regulatory constraints using effectively designed reward functions.","support":"full"},{"assumption":"Continuous action spaces are appropriate for portfolio weight and market‑making decisions, enabling RL methods to adapt to market microstructure.","type":"model","stated":"explicit","quote":"Due to DDPG’s capability with continuous action spaces for portfolio weights, it is utilized in optimizing portfolio management. This improvement likely results from the adaptability and continuous nature of RL methods, which better capture the dy-namic market microstructure and efficiently solve complex multi-objective functions that are difficult for traditional market making methods.","support":"full"},{"assumption":"Performance gains reported for market making and cryptocurrency trading reflect typical ranges across multiple implementations rather than single‑study results.","type":"statistical","stated":"explicit","quote":"The performance metrics represent aggregated results from the meta-analysis of 167 studies, with Sharpe ratios reflecting typical performance ranges observed across mul-tiple implementations within each algorithm-application category rather than results from individual studies. Market making applications typically exhibit the highest performance gains, as demon-strated in Figure 5 where market making shows the","support":"full"}]},"datasets":{"status":"available","items":[{"name":"synthetic dataset","provider":null,"kind":"synthetic","access":"unknown","period":null,"universe":null,"quote":"Empirical Validation Framework To validate the meta-analysis findings, a synthetic dataset was developed that repro-duces key statistical patterns observed in the literature. The synthetic dataset includes 167 studies with variables representing feature dimensions, number of as-sets, training periods, algorithm types, application domains, and performance metrics.","support":"full"}],"known":[],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"unclear"},"transaction_costs":{"status":"not_applicable","modeled":"n/a","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R4a","candidates":{"front_matter":"n/a","evidence_card":"yes","evidence_card_quoted":true},"card_detail":"Transaction costs are incorporated via explicit reward‑structure constraints, including cost and market‑impact considerations.","card_quote":"Recent portfolio op-timization research focuses on incorporating realistic limitations, such as transaction costs, into RL frameworks. The RL approach enables objective optimization by con-sidering transaction costs and market impact constraints through explicit reward structure constraints, along with legal and regulatory constraints using effectively designed reward functions.","meaning":"the paper reports no trading or execution result; the cost the extractor found is a model parameter"},"results":{"status":"available","evaluation_type":"none","findings":[{"finding":"Market making applications exhibit the highest performance gains among RL‑enabled financial tasks, as shown by meta‑analysis of 167 studies.","kind":"performance","quote":"Market making applications typically exhibit the highest performance gains, as demon-strated in Figure 5 where market making shows the highest RL premium among all appli-cation domains.","support":"partial"},{"finding":"Synthetic validation indicates that implementation quality and domain expertise outweigh algorithmic complexity for successful RL in finance.","kind":"performance","quote":"Empirical evidence with synthetic data suggests that implementation quality and domain knowledge often outweigh algorithmic complexity. The meta-analysis indicates that successful RL in finance relies more on quality implementation, domain expertise, and data quality than complex algorithms.","support":"full"},{"finding":"RL methods outperform traditional approaches in market making by adapting to continuous control problems and optimizing bid‑ask spread capture.","kind":"performance","quote":"Market making involves managing inventory risk to maintain security levels within a target range while profiting from the bid-ask spread. In contrast, RL methods adapt to market conditions and manage multiple objectives like bid-ask spread capture, inventory, and risk.","support":"full"},{"finding":"Cryptocurrency trading shows the second‑highest performance, driven by high volatility and market inefficiencies that RL can exploit.","kind":"performance","quote":"Cryptocurrency trading applications follow as the second-highest performing domain, likely due to the greater re-active volatility and inefficiencies in these markets, which can be effectively exploited with RL algorithms.","support":"full"}],"reported_performance":null,"out_of_sample":{"method":"none","headline_out_of_sample":null}},"independent_evidence":{"status":"not_extracted","supports":[],"contradicts":[],"qualifies":[],"compared_with":0,"citations":{"cited_by_in_corpus":0,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":0},"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":2,"extracted_at":"2026-11-14T19:45:07Z"},"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/"}