{"paper_id":"2511.11481","title":"Risk-Aware Deep Reinforcement Learning for Dynamic Portfolio Optimization","authors":["Emmanuel Lwele","Sabuni Emmanuel","Sitali Gabriel Sitali"],"paper_date":"2025-11-14","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2511.11481v1","page_url":"https://thequant.space/flowcharts/risk-aware-deep-reinforcement-learning-for-dynamic-portfolio/","scores":{"empirical_rigor":6,"math_complexity":7.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a risk‑aware deep reinforcement learning model for dynamic portfolio optimization that uses a Sharpe‑ratio based reward and incorporates maximum drawdown and volatility constraints.","quote":"We introduce a risk-aware DRL model that integrates advanced reinforcement learning algorithms, a novel reward function based on the Sharpe ratio, and direct risk control strategies including maximum drawdown and portfolio volatility constraints.","summary":"This paper presents a deep reinforcement learning (DRL) framework for dynamic portfolio optimization under market uncertainty and risk. The proposed model integrates a Sharpe ratio-based reward function with direct risk control mechanisms, including maximum drawdown and volatility constraints.…","strategy":"Risk-Aware DRL Portfolio Allocation (PPO)"},"assumptions":{"status":"available","items":[{"assumption":"The model is trained and evaluated on historical financial time series data.","type":"data","stated":"explicit","quote":"Historical financial data was collected, including stock prices, economic indicators, and other relevant financial features."},{"assumption":"Portfolio rebalancing is performed on a monthly basis.","type":"execution","stated":"explicit","quote":"Historical data was utilised for model training and backtesting, with monthly portfolio rebalancing."},{"assumption":"Transaction costs are represented by a linear coefficient µ in the reward function.","type":"model","stated":"explicit","quote":"The reward at time t is based on the log of the total portfolio value, penalized by transaction costs. It is defined as: R(St−1, At−1) = ln At−1 · Yt−1 −µ"},{"assumption":"The agent’s actions are portfolio weight vectors that sum to one.","type":"model","stated":"implied","quote":"The action at time t, denoted as At, is the portfolio weight vector, derived from the weights at the previous time step At−1 = Wt−1."},{"assumption":"Proximal Policy Optimization (PPO) can learn a stable policy from the simulated market environment.","type":"model","stated":"explicit","quote":"This neural network-based policy is optimized using PPO (Proximal Policy Optimization), allowing the agent to gradually improve its decision-making over time by interacting with the market environment simulated through historical data."},{"assumption":"Regularization (L1/L2) helps mitigate overfitting of the neural network.","type":"model","stated":"explicit","quote":"We applied regularization techniques such as L1 (Lasso) and L2 (Ridge) to penalize large weights in the neural network and reduce the risk of overfitting."}]},"datasets":{"status":"available","items":[{"name":"Yahoo Finance 15‑asset dataset","provider":"Yahoo Finance","kind":"market-prices","access":"public","period":"2015-01-01 to 2024-01-01","universe":"AAPL, AMZN, GOOG, HD, INTC, JNJ, JPM, MA, META, NFLX, NVDA, PG, TSLA, UNH, V","quote":"For this experiment, 15 high-performing assets were selected from Yahoo Finance on the basis of their historical performance over a 10-year period spanning January 1, 2015, to January 1, 2024.","known_id":"yahoo-finance"}],"known":[],"sample":{"start":null,"end":null,"market":"high-performing equities","asset_class":"equities","frequency":"daily"},"data_resolution":"daily"},"transaction_costs":{"status":"available","modeled":"yes","headline_net_of_costs":null,"detail":"Transaction costs are included in the reward function via a coefficient µ that penalizes changes in portfolio weights.","quote":"The reward at time t is based on the log of the total portfolio value, penalized by transaction costs. ... µ","source":"evidence_card","rule":"R2","candidates":{"front_matter":"unclear","evidence_card":"yes","evidence_card_quoted":true},"card_detail":"Transaction costs are included in the reward function via a coefficient µ that penalizes changes in portfolio weights.","card_quote":"The reward at time t is based on the log of the total portfolio value, penalized by transaction costs. ... µ","meaning":"costs are part of the reported headline results"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"After training, the DRL agent’s annualized return fell to 2.10% with volatility reduced to 16.32% and Sharpe ratio dropping to 0.13, indicating a collapse in risk‑adjusted performance.","kind":"performance","quote":"After training, the DRL agent’s return declined sharply to 2.10%, with volatility reduced to 16.32%. The Sharpe ratio deteriorated to 0.13, suggesting a collapse in risk‑adjusted performance."},{"finding":"The trained model succeeded in stabilizing risk by cutting volatility roughly in half compared to the pre‑training baseline (34.9% → 16.32%).","kind":"performance","quote":"shift highlights a critical trade‑off: while the trained model successfully stabilized risk through volatility reduction, it failed to preserve profitability."}],"reported_performance":{"strategy":"Risk-Aware DRL Portfolio Allocation (PPO)","sharpe":0.13,"sharpe_benchmark":null,"annual_return_pct":2.1,"total_return_pct":null,"max_drawdown_pct":null},"out_of_sample":{"method":"holdout","headline_out_of_sample":true}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":11,"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":1,"extracted_at":"2026-11-14T17:21:48Z"},"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/"}